Merge branch 'main' into next

Merge pull request 'Garage v0.8.3' (#619 ) from next-0.8 into main
Reviewed-on: https://git.deuxfleurs.fr/Deuxfleurs/garage/pulls/619
2026-08-13 15:46:53 +00:00 · 2023-08-29 11:32:42 +02:00 · 2023-08-29 08:55:46 +00:00 · 2023-08-28 17:08:21 +02:00 · 2023-06-14 17:19:25 +02:00 · 2023-06-14 17:12:37 +02:00
84 changed files with 24852 additions and 1786 deletions
@@ -20,6 +20,7 @@ steps:
    image: nixpkgs/nix:nixos-22.05
    environment:
      GARAGE_TEST_INTEGRATION_EXE: result-bin/bin/garage
+      GARAGE_TEST_INTEGRATION_PATH: tmp-garage-integration
    commands:
      - nix-build --no-build-output --attr clippy.amd64 --argstr git_version ${DRONE_TAG:-$DRONE_COMMIT}
      - nix-build --no-build-output --attr test.amd64
@@ -31,8 +32,9 @@ steps:
      - ./result/bin/garage_util-*
      - ./result/bin/garage_web-*
      - ./result/bin/garage-*
-      - ./result/bin/integration-*
+      - ./result/bin/integration-* || (cat tmp-garage-integration/stderr.log; false)
      - rm result
+      - rm -rv tmp-garage-integration

  - name: integration tests
    image: nixpkgs/nix:nixos-22.05
@@ -283,6 +285,6 @@ trigger:

 ---
 kind: signature
-hmac: ac09a5a8c82502f67271f93afa1e1e21ce66383b8e24a6deb26b285cc1c378ba
+hmac: 24aebbcdba84fd0cdf963061d7bb72ae5b915bfdd0f50c7b019001126fb7fa56

 ...
@@ -1366,12 +1366,15 @@ dependencies = [
 "arc-swap",
 "async-trait",
 "bytes",
+ "bytesize",
 "err-derive",
 "futures",
 "futures-util",
+ "garage_db",
 "garage_util",
 "gethostname",
 "hex",
+ "itertools 0.10.5",
 "k8s-openapi",
 "kube",
 "kuska-sodiumoxide",
@@ -1434,7 +1437,7 @@ dependencies = [
 "netapp",
 "opentelemetry",
 "rand",
- "rmp-serde 0.15.5",
+ "rmp-serde",
 "rustc_version",
 "serde",
 "serde_json",
@@ -2283,7 +2286,7 @@ dependencies = [
 "opentelemetry-contrib",
 "pin-project",
 "rand",
- "rmp-serde 1.1.2",
+ "rmp-serde",
 "serde",
 "tokio",
 "tokio-stream",
@@ -3001,17 +3004,6 @@ dependencies = [
 "paste",
 ]

-[[package]]
-name = "rmp-serde"
-version = "0.15.5"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "723ecff9ad04f4ad92fe1c8ca6c20d2196d9286e9c60727c4cb5511629260e9d"
-dependencies = [
- "byteorder",
- "rmp",
- "serde",
-]
-
 [[package]]
 name = "rmp-serde"
 version = "1.1.2"
@@ -33,7 +33,7 @@ args@{
  ignoreLockHash,
 }:
 let
-  nixifiedLockHash = "4d7b9d8135ebc43da8a9353683fc99d626d0f55c895c6ac32bc6e11743088fdf";
+  nixifiedLockHash = "d4392b23d407f7ebc20d7f5db7583847e362665c1abb09f1c1d3305205e5996d";
  workspaceSrc = if args.workspaceSrc == null then ./. else args.workspaceSrc;
  currentLockHash = builtins.hashFile "sha256" (workspaceSrc + /Cargo.lock);
  lockHashIgnored = if ignoreLockHash
@@ -98,12 +98,12 @@ in
      (lib.optional (rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery") "std")
    ];
    dependencies = {
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery" then "cfg_if" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".cfg-if."1.0.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery" then "cfg_if" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".cfg-if."1.0.0" { inherit profileName; }).out;
      ${ if rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery" then "getrandom" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".getrandom."0.2.10" { inherit profileName; }).out;
-      ${ if (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery") && !((hostPlatform.parsed.cpu.name == "armv6l" || hostPlatform.parsed.cpu.name == "armv7l") && hostPlatform.parsed.kernel.name == "none") then "once_cell" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".once_cell."1.18.0" { inherit profileName; }).out;
+      ${ if (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery") && !((hostPlatform.parsed.cpu.name == "armv6l" || hostPlatform.parsed.cpu.name == "armv7l") && hostPlatform.parsed.kernel.name == "none") then "once_cell" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".once_cell."1.18.0" { inherit profileName; }).out;
    };
    buildDependencies = {
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery" then "version_check" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".version_check."0.9.4" { profileName = "__noProfile"; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery" then "version_check" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".version_check."0.9.4" { profileName = "__noProfile"; }).out;
    };
  });
  
@@ -128,7 +128,7 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "0942ffc6dcaadf03badf6e6a2d0228460359d5e34b57ccdc720b7382dfbd5ec5"; };
    features = builtins.concatLists [
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "alloc")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "alloc")
    ];
  });
  
@@ -824,7 +824,7 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "b1f45e9417d87227c7a56d22e471c6206462cba514c7590c09aff4cf6d1ddcad"; };
    dependencies = {
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "serde" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.188" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "serde" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.188" { inherit profileName; }).out;
    };
  });
  
@@ -888,8 +888,8 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "17febce684fd15d89027105661fec94afb475cb995fbc59d2865198446ba2eea"; };
    features = builtins.concatLists [
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "extern_crate_alloc")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "extern_crate_std")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "extern_crate_alloc")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "extern_crate_std")
    ];
  });
  
@@ -1182,13 +1182,13 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "d1cfb3ea8a53f37c40dea2c7bedcbd88bdfae54f5e2175d6ecaff1c988353add"; };
    features = builtins.concatLists [
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "alloc")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "default")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "std")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "alloc")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "default")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "std")
    ];
    dependencies = {
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "cfg_if" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".cfg-if."1.0.0" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "crossbeam_utils" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".crossbeam-utils."0.8.16" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "cfg_if" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".cfg-if."1.0.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "crossbeam_utils" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".crossbeam-utils."0.8.16" { inherit profileName; }).out;
    };
  });
  
@@ -1459,8 +1459,8 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "4443176a9f2c162692bd3d352d745ef9413eec5782a80d8fd6f8a1ac692a07f7"; };
    features = builtins.concatLists [
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "default")
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "std")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "default")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "std")
    ];
  });
  
@@ -1716,13 +1716,13 @@ in
      (lib.optional (rootFeatures' ? "garage/default") "default")
      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/k2v") "k2v")
      (lib.optional (rootFeatures' ? "garage/kubernetes-discovery") "kubernetes-discovery")
-      (lib.optional (rootFeatures' ? "garage/lmdb") "lmdb")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb") "lmdb")
      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/metrics") "metrics")
      (lib.optional (rootFeatures' ? "garage/opentelemetry-otlp" || rootFeatures' ? "garage/telemetry-otlp") "opentelemetry-otlp")
      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/metrics" || rootFeatures' ? "garage/opentelemetry-prometheus") "opentelemetry-prometheus")
      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/metrics" || rootFeatures' ? "garage/prometheus") "prometheus")
      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sled") "sled")
-      (lib.optional (rootFeatures' ? "garage/sqlite") "sqlite")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite") "sqlite")
      (lib.optional (rootFeatures' ? "garage/system-libs") "system-libs")
      (lib.optional (rootFeatures' ? "garage/telemetry-otlp") "telemetry-otlp")
    ];
@@ -1872,20 +1872,20 @@ in
      (lib.optional (rootFeatures' ? "garage_db/clap" || rootFeatures' ? "garage_db/cli") "clap")
      (lib.optional (rootFeatures' ? "garage_db/cli") "cli")
      (lib.optional (rootFeatures' ? "garage_db/default") "default")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "heed")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "lmdb")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "heed")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "lmdb")
      (lib.optional (rootFeatures' ? "garage_db/cli" || rootFeatures' ? "garage_db/pretty_env_logger") "pretty_env_logger")
-      (lib.optional (rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "rusqlite")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "rusqlite")
      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sled" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/sled" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sled") "sled")
-      (lib.optional (rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "sqlite")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "sqlite")
    ];
    dependencies = {
      ${ if rootFeatures' ? "garage_db/clap" || rootFeatures' ? "garage_db/cli" then "clap" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".clap."4.4.0" { inherit profileName; }).out;
      err_derive = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".err-derive."0.3.1" { profileName = "__noProfile"; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "heed" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".heed."0.11.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "heed" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".heed."0.11.0" { inherit profileName; }).out;
      hexdump = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hexdump."0.1.1" { inherit profileName; }).out;
      ${ if rootFeatures' ? "garage_db/cli" || rootFeatures' ? "garage_db/pretty_env_logger" then "pretty_env_logger" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".pretty_env_logger."0.5.0" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "rusqlite" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rusqlite."0.29.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "rusqlite" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rusqlite."0.29.0" { inherit profileName; }).out;
      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sled" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/sled" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sled" then "sled" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".sled."0.34.7" { inherit profileName; }).out;
      tracing = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".tracing."0.1.37" { inherit profileName; }).out;
    };
@@ -1902,9 +1902,9 @@ in
    features = builtins.concatLists [
      (lib.optional (rootFeatures' ? "garage_model/default") "default")
      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/k2v" || rootFeatures' ? "garage_api/k2v" || rootFeatures' ? "garage_model/k2v") "k2v")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_model/lmdb") "lmdb")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "lmdb")
      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sled" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sled") "sled")
-      (lib.optional (rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_model/sqlite") "sqlite")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "sqlite")
    ];
    dependencies = {
      arc_swap = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".arc-swap."1.6.0" { inherit profileName; }).out;
@@ -1950,12 +1950,15 @@ in
      arc_swap = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".arc-swap."1.6.0" { inherit profileName; }).out;
      async_trait = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".async-trait."0.1.73" { profileName = "__noProfile"; }).out;
      bytes = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bytes."1.4.0" { inherit profileName; }).out;
+      bytesize = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bytesize."1.3.0" { inherit profileName; }).out;
      ${ if rootFeatures' ? "garage/consul-discovery" || rootFeatures' ? "garage_rpc/consul-discovery" || rootFeatures' ? "garage_rpc/err-derive" then "err_derive" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".err-derive."0.3.1" { profileName = "__noProfile"; }).out;
      futures = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures."0.3.28" { inherit profileName; }).out;
      futures_util = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".futures-util."0.3.28" { inherit profileName; }).out;
+      garage_db = (rustPackages."unknown".garage_db."0.8.3" { inherit profileName; }).out;
      garage_util = (rustPackages."unknown".garage_util."0.8.3" { inherit profileName; }).out;
      gethostname = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".gethostname."0.4.3" { inherit profileName; }).out;
      hex = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hex."0.4.3" { inherit profileName; }).out;
+      itertools = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".itertools."0.10.5" { inherit profileName; }).out;
      ${ if rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage_rpc/k8s-openapi" || rootFeatures' ? "garage_rpc/kubernetes-discovery" then "k8s_openapi" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".k8s-openapi."0.16.0" { inherit profileName; }).out;
      ${ if rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery" then "kube" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".kube."0.75.0" { inherit profileName; }).out;
      sodiumoxide = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".kuska-sodiumoxide."0.2.5-0" { inherit profileName; }).out;
@@ -2026,7 +2029,7 @@ in
      netapp = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".netapp."0.5.3" { inherit profileName; }).out;
      opentelemetry = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".opentelemetry."0.17.0" { inherit profileName; }).out;
      rand = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rand."0.8.5" { inherit profileName; }).out;
-      rmp_serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rmp-serde."0.15.5" { inherit profileName; }).out;
+      rmp_serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rmp-serde."1.1.2" { inherit profileName; }).out;
      serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.188" { inherit profileName; }).out;
      serde_json = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde_json."1.0.105" { inherit profileName; }).out;
      sha2 = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".sha2."0.10.7" { inherit profileName; }).out;
@@ -2176,15 +2179,15 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "2c6201b9ff9fd90a5a3bac2e56a830d0caa509576f0e503818ee82c181b3437a"; };
    features = builtins.concatLists [
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "ahash")
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "allocator-api2")
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "default")
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "inline-more")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "ahash")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "allocator-api2")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "default")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "inline-more")
      (lib.optional (rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage/opentelemetry-otlp" || rootFeatures' ? "garage/telemetry-otlp" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery") "raw")
    ];
    dependencies = {
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "ahash" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".ahash."0.8.3" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "allocator_api2" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".allocator-api2."0.2.16" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "ahash" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".ahash."0.8.3" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "allocator_api2" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".allocator-api2."0.2.16" { inherit profileName; }).out;
    };
  });
  
@@ -2194,7 +2197,7 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "312f66718a2d7789ffef4f4b7b213138ed9f1eb3aa1d0d82fc99f88fb3ffd26f"; };
    dependencies = {
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "hashbrown" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hashbrown."0.14.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "hashbrown" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hashbrown."0.14.0" { inherit profileName; }).out;
    };
  });
  
@@ -2224,20 +2227,20 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "269c7486ed6def5d7b59a427cec3e87b4d4dd4381d01e21c8c9f2d3985688392"; };
    features = builtins.concatLists [
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "lmdb")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "lmdb-rkv-sys")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "lmdb")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "lmdb-rkv-sys")
    ];
    dependencies = {
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "bytemuck" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bytemuck."1.13.1" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "byteorder" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".byteorder."1.4.3" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "heed_traits" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".heed-traits."0.8.0" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "heed_types" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".heed-types."0.8.0" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "libc" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".libc."0.2.147" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "lmdb_sys" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".lmdb-rkv-sys."0.11.2" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "once_cell" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".once_cell."1.18.0" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "page_size" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".page_size."0.4.2" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "synchronoise" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".synchronoise."1.0.1" { inherit profileName; }).out;
-      ${ if (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") && hostPlatform.isWindows then "url" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".url."2.4.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "bytemuck" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bytemuck."1.13.1" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "byteorder" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".byteorder."1.4.3" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "heed_traits" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".heed-traits."0.8.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "heed_types" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".heed-types."0.8.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "libc" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".libc."0.2.147" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "lmdb_sys" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".lmdb-rkv-sys."0.11.2" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "once_cell" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".once_cell."1.18.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "page_size" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".page_size."0.4.2" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "synchronoise" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".synchronoise."1.0.1" { inherit profileName; }).out;
+      ${ if (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") && hostPlatform.isWindows then "url" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".url."2.4.0" { inherit profileName; }).out;
    };
  });
  
@@ -2254,20 +2257,20 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "9a6cf0a6952fcedc992602d5cddd1e3fff091fbe87d38636e3ec23a31f32acbd"; };
    features = builtins.concatLists [
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "bincode")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "default")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "serde")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "serde-bincode")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "serde-json")
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "serde_json")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "bincode")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "default")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "serde")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "serde-bincode")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "serde-json")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "serde_json")
    ];
    dependencies = {
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "bincode" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bincode."1.3.3" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "bytemuck" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bytemuck."1.13.1" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "byteorder" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".byteorder."1.4.3" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "heed_traits" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".heed-traits."0.8.0" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "serde" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.188" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "serde_json" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde_json."1.0.105" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "bincode" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bincode."1.3.3" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "bytemuck" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bytemuck."1.13.1" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "byteorder" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".byteorder."1.4.3" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "heed_traits" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".heed-traits."0.8.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "serde" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.188" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "serde_json" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde_json."1.0.105" { inherit profileName; }).out;
    };
  });
  
@@ -2626,12 +2629,12 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "b0fd2260e829bddf4cb6ea802289de2f86d6a7a690192fbe91b3f46e0f2c8473"; };
    features = builtins.concatLists [
-      (lib.optional (rootFeatures' ? "garage/opentelemetry-otlp" || rootFeatures' ? "garage/telemetry-otlp") "default")
-      (lib.optional (rootFeatures' ? "garage/opentelemetry-otlp" || rootFeatures' ? "garage/telemetry-otlp") "use_alloc")
-      (lib.optional (rootFeatures' ? "garage/opentelemetry-otlp" || rootFeatures' ? "garage/telemetry-otlp") "use_std")
+      [ "default" ]
+      [ "use_alloc" ]
+      [ "use_std" ]
    ];
    dependencies = {
-      ${ if rootFeatures' ? "garage/opentelemetry-otlp" || rootFeatures' ? "garage/telemetry-otlp" then "either" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".either."1.9.0" { inherit profileName; }).out;
+      either = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".either."1.9.0" { inherit profileName; }).out;
    };
  });
  
@@ -2987,15 +2990,15 @@ in
      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage_db/bundled-libs") "bundled")
      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage_db/bundled-libs") "bundled_bindings")
      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage_db/bundled-libs") "cc")
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "default")
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "min_sqlite_version_3_14_0")
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "pkg-config")
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite") "vcpkg")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "default")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "min_sqlite_version_3_14_0")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "pkg-config")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite") "vcpkg")
    ];
    buildDependencies = {
      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage_db/bundled-libs" then "cc" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".cc."1.0.83" { profileName = "__noProfile"; }).out;
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "pkg_config" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".pkg-config."0.3.27" { profileName = "__noProfile"; }).out;
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "vcpkg" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".vcpkg."0.2.15" { profileName = "__noProfile"; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "pkg_config" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".pkg-config."0.3.27" { profileName = "__noProfile"; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "vcpkg" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".vcpkg."0.2.15" { profileName = "__noProfile"; }).out;
    };
  });
  
@@ -3025,14 +3028,14 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "61b9ce6b3be08acefa3003c57b7565377432a89ec24476bbe72e11d101f852fe"; };
    features = builtins.concatLists [
-      (lib.optional (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") "default")
+      (lib.optional (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") "default")
    ];
    dependencies = {
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "libc" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".libc."0.2.147" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "libc" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".libc."0.2.147" { inherit profileName; }).out;
    };
    buildDependencies = {
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "cc" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".cc."1.0.83" { profileName = "__noProfile"; }).out;
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "pkg_config" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".pkg-config."0.3.27" { profileName = "__noProfile"; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "cc" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".cc."1.0.83" { profileName = "__noProfile"; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "pkg_config" else null } = (buildRustPackages."registry+https://github.com/rust-lang/crates.io-index".pkg-config."0.3.27" { profileName = "__noProfile"; }).out;
    };
  });
  
@@ -3442,7 +3445,7 @@ in
      [ "default" ]
      [ "race" ]
      [ "std" ]
-      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery") "unstable")
+      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/kubernetes-discovery" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" || rootFeatures' ? "garage_rpc/kube" || rootFeatures' ? "garage_rpc/kubernetes-discovery") "unstable")
    ];
  });
  
@@ -3580,8 +3583,8 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "eebde548fbbf1ea81a99b128872779c437752fb99f217c45245e1a61dcd9edcd"; };
    dependencies = {
-      ${ if (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") && hostPlatform.isUnix then "libc" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".libc."0.2.147" { inherit profileName; }).out;
-      ${ if (rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb") && hostPlatform.isWindows then "winapi" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".winapi."0.3.9" { inherit profileName; }).out;
+      ${ if (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") && hostPlatform.isUnix then "libc" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".libc."0.2.147" { inherit profileName; }).out;
+      ${ if (rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb") && hostPlatform.isWindows then "winapi" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".winapi."0.3.9" { inherit profileName; }).out;
    };
  });
  
@@ -4272,18 +4275,6 @@ in
    };
  });
  
-  "registry+https://github.com/rust-lang/crates.io-index".rmp-serde."0.15.5" = overridableMkRustCrate (profileName: rec {
-    name = "rmp-serde";
-    version = "0.15.5";
-    registry = "registry+https://github.com/rust-lang/crates.io-index";
-    src = fetchCratesIo { inherit name version; sha256 = "723ecff9ad04f4ad92fe1c8ca6c20d2196d9286e9c60727c4cb5511629260e9d"; };
-    dependencies = {
-      byteorder = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".byteorder."1.4.3" { inherit profileName; }).out;
-      rmp = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".rmp."0.8.12" { inherit profileName; }).out;
-      serde = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".serde."1.0.188" { inherit profileName; }).out;
-    };
-  });
-  
  "registry+https://github.com/rust-lang/crates.io-index".rmp-serde."1.1.2" = overridableMkRustCrate (profileName: rec {
    name = "rmp-serde";
    version = "1.1.2";
@@ -4321,12 +4312,12 @@ in
      (lib.optional (rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage_db/bundled-libs") "modern_sqlite")
    ];
    dependencies = {
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "bitflags" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bitflags."2.4.0" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "fallible_iterator" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".fallible-iterator."0.2.0" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "fallible_streaming_iterator" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".fallible-streaming-iterator."0.1.9" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "hashlink" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hashlink."0.8.3" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "libsqlite3_sys" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".libsqlite3-sys."0.26.0" { inherit profileName; }).out;
-      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/sqlite" then "smallvec" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".smallvec."1.11.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "bitflags" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".bitflags."2.4.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "fallible_iterator" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".fallible-iterator."0.2.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "fallible_streaming_iterator" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".fallible-streaming-iterator."0.1.9" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "hashlink" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".hashlink."0.8.3" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "libsqlite3_sys" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".libsqlite3-sys."0.26.0" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/bundled-libs" || rootFeatures' ? "garage/default" || rootFeatures' ? "garage/sqlite" || rootFeatures' ? "garage_db/bundled-libs" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/rusqlite" || rootFeatures' ? "garage_db/sqlite" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/sqlite" then "smallvec" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".smallvec."1.11.0" { inherit profileName; }).out;
    };
  });
  
@@ -4990,7 +4981,7 @@ in
    registry = "registry+https://github.com/rust-lang/crates.io-index";
    src = fetchCratesIo { inherit name version; sha256 = "3dbc01390fc626ce8d1cffe3376ded2b72a11bb70e1c75f404a210e4daa4def2"; };
    dependencies = {
-      ${ if rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/lmdb" then "crossbeam_queue" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".crossbeam-queue."0.3.8" { inherit profileName; }).out;
+      ${ if rootFeatures' ? "garage/default" || rootFeatures' ? "garage/lmdb" || rootFeatures' ? "garage_db/default" || rootFeatures' ? "garage_db/heed" || rootFeatures' ? "garage_db/lmdb" || rootFeatures' ? "garage_model/default" || rootFeatures' ? "garage_model/lmdb" then "crossbeam_queue" else null } = (rustPackages."registry+https://github.com/rust-lang/crates.io-index".crossbeam-queue."0.3.8" { inherit profileName; }).out;
    };
  });
  
@@ -37,7 +37,7 @@ Second, we suppose you have created a key and a bucket.
 As a reminder, you can create a key for your nextcloud instance as follow:

 ```bash
-garage key new --name nextcloud-key
+garage key create nextcloud-key
 ```

 Keep the Key ID and the Secret key in a pad, they will be needed later.  
@@ -139,7 +139,7 @@ a reasonable trade-off for some instances.
 Create a key for Peertube:

 ```bash
-garage key new --name peertube-key
+garage key create peertube-key
 ```

 Keep the Key ID and the Secret key in a pad, they will be needed later.  
@@ -253,7 +253,7 @@ As such, your Garage cluster should be configured appropriately for good perform
 This is the usual Garage setup:

 ```bash
-garage key new --name mastodon-key
+garage key create mastodon-key
 garage bucket create mastodon-data
 garage bucket allow mastodon-data --read --write --key mastodon-key
 ```
@@ -379,7 +379,7 @@ Supposing you have a working synapse installation, you can add the module with p
 Now create a bucket and a key for your matrix instance (note your Key ID and Secret Key somewhere, they will be needed later):

 ```bash
-garage key new --name matrix-key
+garage key create matrix-key
 garage bucket create matrix
 garage bucket allow matrix --read --write --key matrix-key
 ```
@@ -54,7 +54,7 @@ how to configure this.
 Create your key and bucket:

 ```bash
-garage key new my-key
+garage key create my-key
 garage bucket create backup
 garage bucket allow backup --read --write --key my-key
 ```
@@ -23,7 +23,7 @@ You can configure a different target for each data type (check `[lfs]` and `[att
 Let's start by creating a key and a bucket (your key id and secret will be needed later, keep them somewhere):

 ```bash
-garage key new --name gitea-key
+garage key create gitea-key
 garage bucket create gitea
 garage bucket allow gitea --read --write --key gitea-key
 ```
@@ -118,7 +118,7 @@ through another support, like a git repository.
 As a first step, we will need to create a bucket on Garage and enabling website access on it:

 ```bash
-garage key new --name nix-key
+garage key create nix-key
 garage bucket create nix.example.com
 garage bucket allow nix.example.com --read --write --key nix-key
 garage bucket website nix.example.com --allow
@@ -209,7 +209,7 @@ one key can access multiple buckets, multiple keys can access one bucket.
 Create an API key using the following command:

 ```
-garage key new --name nextcloud-app-key
+garage key create nextcloud-app-key
 ```

 The output should look as follows:
@@ -10,6 +10,8 @@ Here is an example `garage.toml` configuration file that illustrates all of the
 ```toml
 metadata_dir = "/var/lib/garage/meta"
 data_dir = "/var/lib/garage/data"
+metadata_fsync = true
+data_fsync = false

 db_engine = "lmdb"

@@ -130,6 +132,49 @@ convert-db -a <input db engine> -i <input db path> \
 Make sure to specify the full database path as presented in the table above,
 and not just the path to the metadata directory.

+### `metadata_fsync`
+
+Whether to enable synchronous mode for the database engine or not.
+This is disabled (`false`) by default.
+
+This reduces the risk of metadata corruption in case of power failures,
+at the cost of a significant drop in write performance,
+as Garage will have to pause to sync data to disk much more often
+(several times for API calls such as PutObject).
+
+Using this option reduces the risk of simultaneous metadata corruption on several
+cluster nodes, which could lead to data loss.
+
+If multi-site replication is used, this option is most likely not necessary, as
+it is extremely unlikely that two nodes in different locations will have a 
+power failure at the exact same time.
+
+(Metadata corruption on a single node is not an issue, the corrupted data file
+can always be deleted and reconstructed from the other nodes in the cluster.)
+
+Here is how this option impacts the different database engines:
+
+| Database | `metadata_fsync = false` (default) | `metadata_fsync = true`       |
+|----------|------------------------------------|-------------------------------|
+| Sled     | default options                    | *unsupported*                 |
+| Sqlite   | `PRAGMA synchronous = OFF`         | `PRAGMA synchronous = NORMAL` |
+| LMDB     | `MDB_NOMETASYNC` + `MDB_NOSYNC`    | `MDB_NOMETASYNC`              |
+
+Note that the Sqlite database is always ran in `WAL` mode (`PRAGMA journal_mode = WAL`).
+
+### `data_fsync`
+
+Whether to `fsync` data blocks and their containing directory after they are
+saved to disk.
+This is disabled (`false`) by default.
+
+This might reduce the risk that a data block is lost in rare
+situations such as simultaneous node losing power,
+at the cost of a moderate drop in write performance.
+
+Similarly to `metatada_fsync`, this is likely not necessary
+if geographical replication is used.
+
 ### `block_size`

 Garage splits stored objects in consecutive chunks of size `block_size`
@@ -52,11 +52,11 @@ Returns an HTTP status 200 if the node is ready to answer user's requests,
 and an HTTP status 503 (Service Unavailable) if there are some partitions
 for which a quorum of nodes is not available.
 A simple textual message is also returned in a body with content-type `text/plain`.
-See `/v0/health` for an API that also returns JSON output.
+See `/v1/health` for an API that also returns JSON output.

 ### Cluster operations

-#### GetClusterStatus `GET /v0/status`
+#### GetClusterStatus `GET /v1/status`

 Returns the cluster's current status in JSON, including:

@@ -70,86 +70,112 @@ Example response body:
 ```json
 {
  "node": "ec79480e0ce52ae26fd00c9da684e4fa56658d9c64cdcecb094e936de0bfe71f",
-  "garage_version": "git:v0.8.0",
-  "knownNodes": {
-    "ec79480e0ce52ae26fd00c9da684e4fa56658d9c64cdcecb094e936de0bfe71f": {
+  "garageVersion": "git:v0.9.0-dev",
+  "garageFeatures": [
+    "k2v",
+    "sled",
+    "lmdb",
+    "sqlite",
+    "metrics",
+    "bundled-libs"
+  ],
+  "rustVersion": "1.68.0",
+  "dbEngine": "LMDB (using Heed crate)",
+  "knownNodes": [
+    {
+      "id": "ec79480e0ce52ae26fd00c9da684e4fa56658d9c64cdcecb094e936de0bfe71f",
      "addr": "10.0.0.11:3901",
-      "is_up": true,
-      "last_seen_secs_ago": 9,
+      "isUp": true,
+      "lastSeenSecsAgo": 9,
      "hostname": "node1"
    },
-    "4a6ae5a1d0d33bf895f5bb4f0a418b7dc94c47c0dd2eb108d1158f3c8f60b0ff": {
+    {
+      "id": "4a6ae5a1d0d33bf895f5bb4f0a418b7dc94c47c0dd2eb108d1158f3c8f60b0ff",
      "addr": "10.0.0.12:3901",
-      "is_up": true,
-      "last_seen_secs_ago": 1,
+      "isUp": true,
+      "lastSeenSecsAgo": 1,
      "hostname": "node2"
    },
-    "23ffd0cdd375ebff573b20cc5cef38996b51c1a7d6dbcf2c6e619876e507cf27": {
+    {
+      "id": "23ffd0cdd375ebff573b20cc5cef38996b51c1a7d6dbcf2c6e619876e507cf27",
      "addr": "10.0.0.21:3901",
-      "is_up": true,
-      "last_seen_secs_ago": 7,
+      "isUp": true,
+      "lastSeenSecsAgo": 7,
      "hostname": "node3"
    },
-    "e2ee7984ee65b260682086ec70026165903c86e601a4a5a501c1900afe28d84b": {
+    {
+      "id": "e2ee7984ee65b260682086ec70026165903c86e601a4a5a501c1900afe28d84b",
      "addr": "10.0.0.22:3901",
-      "is_up": true,
-      "last_seen_secs_ago": 1,
+      "isUp": true,
+      "lastSeenSecsAgo": 1,
      "hostname": "node4"
    }
-  },
+  ],
  "layout": {
    "version": 12,
-    "roles": {
-      "ec79480e0ce52ae26fd00c9da684e4fa56658d9c64cdcecb094e936de0bfe71f": {
+    "roles": [
+      {
+        "id": "ec79480e0ce52ae26fd00c9da684e4fa56658d9c64cdcecb094e936de0bfe71f",
        "zone": "dc1",
-        "capacity": 4,
+        "capacity": 10737418240,
        "tags": [
          "node1"
        ]
      },
-      "4a6ae5a1d0d33bf895f5bb4f0a418b7dc94c47c0dd2eb108d1158f3c8f60b0ff": {
+      {
+        "id": "4a6ae5a1d0d33bf895f5bb4f0a418b7dc94c47c0dd2eb108d1158f3c8f60b0ff",
        "zone": "dc1",
-        "capacity": 6,
+        "capacity": 10737418240,
        "tags": [
          "node2"
        ]
      },
-      "23ffd0cdd375ebff573b20cc5cef38996b51c1a7d6dbcf2c6e619876e507cf27": {
+      {
+        "id": "23ffd0cdd375ebff573b20cc5cef38996b51c1a7d6dbcf2c6e619876e507cf27",
        "zone": "dc2",
-        "capacity": 10,
+        "capacity": 10737418240,
        "tags": [
          "node3"
        ]
      }
-    },
-    "stagedRoleChanges": {
-      "e2ee7984ee65b260682086ec70026165903c86e601a4a5a501c1900afe28d84b": {
+    ],
+    "stagedRoleChanges": [
+      {
+        "id": "e2ee7984ee65b260682086ec70026165903c86e601a4a5a501c1900afe28d84b",
+        "remove": false,
        "zone": "dc2",
-        "capacity": 5,
+        "capacity": 10737418240,
        "tags": [
          "node4"
        ]
      }
-    }
+      {
+        "id": "23ffd0cdd375ebff573b20cc5cef38996b51c1a7d6dbcf2c6e619876e507cf27",
+        "remove": true,
+        "zone": null,
+        "capacity": null,
+        "tags": null,
+      }
+    ]
  }
 }
 ```

-#### GetClusterHealth `GET /v0/health`
+#### GetClusterHealth `GET /v1/health`

 Returns the cluster's current health in JSON format, with the following variables:

- `status`: one of `Healthy`, `Degraded` or `Unavailable`:
-  - Healthy: Garage node is connected to all storage nodes
-  - Degraded: Garage node is not connected to all storage nodes, but a quorum of write nodes is available for all partitions
-  - Unavailable: a quorum of write nodes is not available for some partitions
- `known_nodes`: the number of nodes this Garage node has had a TCP connection to since the daemon started
- `connected_nodes`: the nubmer of nodes this Garage node currently has an open connection to
- `storage_nodes`: the number of storage nodes currently registered in the cluster layout
- `storage_nodes_ok`: the number of storage nodes to which a connection is currently open
+- `status`: one of `healthy`, `degraded` or `unavailable`:
+  - healthy: Garage node is connected to all storage nodes
+  - degraded: Garage node is not connected to all storage nodes, but a quorum of write nodes is available for all partitions
+  - unavailable: a quorum of write nodes is not available for some partitions
+- `knownNodes`: the number of nodes this Garage node has had a TCP connection to since the daemon started
+- `connectedNodes`: the nubmer of nodes this Garage node currently has an open connection to
+- `storageNodes`: the number of storage nodes currently registered in the cluster layout
+- `storageNodesOk`: the number of storage nodes to which a connection is currently open
 - `partitions`: the total number of partitions of the data (currently always 256)
- `partitions_quorum`: the number of partitions for which a quorum of write nodes is available
- `partitions_all_ok`: the number of partitions for which we are connected to all storage nodes responsible of storing it
+- `partitionsQuorum`: the number of partitions for which a quorum of write nodes is available
+- `partitionsAllOk`: the number of partitions for which we are connected to all storage nodes responsible of storing it

 Contrarily to `GET /health`, this endpoint always returns a 200 OK HTTP response code.

@@ -157,18 +183,18 @@ Example response body:

 ```json
 {
-    "status": "Degraded",
-    "known_nodes": 3,
-    "connected_nodes": 2,
-    "storage_nodes": 3,
-    "storage_nodes_ok": 2,
-    "partitions": 256,
-    "partitions_quorum": 256,
-    "partitions_all_ok": 0
+  "status": "degraded",
+  "knownNodes": 3,
+  "connectedNodes": 3,
+  "storageNodes": 4,
+  "storageNodesOk": 3,
+  "partitions": 256,
+  "partitionsQuorum": 256,
+  "partitionsAllOk": 64
 }
 ```

-#### ConnectClusterNodes `POST /v0/connect`
+#### ConnectClusterNodes `POST /v1/connect`

 Instructs this Garage node to connect to other Garage nodes at specified addresses.

@@ -198,7 +224,7 @@ Example response:
 ]
 ```

-#### GetClusterLayout `GET /v0/layout`
+#### GetClusterLayout `GET /v1/layout`

 Returns the cluster's current layout in JSON, including:

@@ -212,42 +238,54 @@ Example response body:
 ```json
 {
  "version": 12,
-  "roles": {
-    "ec79480e0ce52ae26fd00c9da684e4fa56658d9c64cdcecb094e936de0bfe71f": {
+  "roles": [
+    {
+      "id": "ec79480e0ce52ae26fd00c9da684e4fa56658d9c64cdcecb094e936de0bfe71f",
      "zone": "dc1",
-      "capacity": 4,
+      "capacity": 10737418240,
      "tags": [
        "node1"
      ]
    },
-    "4a6ae5a1d0d33bf895f5bb4f0a418b7dc94c47c0dd2eb108d1158f3c8f60b0ff": {
+    {
+      "id": "4a6ae5a1d0d33bf895f5bb4f0a418b7dc94c47c0dd2eb108d1158f3c8f60b0ff",
      "zone": "dc1",
-      "capacity": 6,
+      "capacity": 10737418240,
      "tags": [
        "node2"
      ]
    },
-    "23ffd0cdd375ebff573b20cc5cef38996b51c1a7d6dbcf2c6e619876e507cf27": {
+    {
+      "id": "23ffd0cdd375ebff573b20cc5cef38996b51c1a7d6dbcf2c6e619876e507cf27",
      "zone": "dc2",
-      "capacity": 10,
+      "capacity": 10737418240,
      "tags": [
        "node3"
      ]
    }
-  },
-  "stagedRoleChanges": {
-    "e2ee7984ee65b260682086ec70026165903c86e601a4a5a501c1900afe28d84b": {
+  ],
+  "stagedRoleChanges": [
+    {
+      "id": "e2ee7984ee65b260682086ec70026165903c86e601a4a5a501c1900afe28d84b",
+      "remove": false,
      "zone": "dc2",
-      "capacity": 5,
+      "capacity": 10737418240,
      "tags": [
        "node4"
      ]
    }
-  }
+    {
+      "id": "23ffd0cdd375ebff573b20cc5cef38996b51c1a7d6dbcf2c6e619876e507cf27",
+      "remove": true,
+      "zone": null,
+      "capacity": null,
+      "tags": null,
+    }
+  ]
 }
 ```

-#### UpdateClusterLayout `POST /v0/layout`
+#### UpdateClusterLayout `POST /v1/layout`

 Send modifications to the cluster layout. These modifications will
 be included in the staged role changes, visible in subsequent calls
@@ -259,8 +297,9 @@ the layout.
 Request body format:

 ```json
-{
-  <node_id>: {
+[
+  {
+    "id": <node_id>,
    "capacity": <new_capacity>,
    "zone": <new_zone>,
    "tags": [
@@ -268,17 +307,22 @@ Request body format:
      ...
    ]
  },
-  <node_id_to_remove>: null,
-  ...
-}
+  {
+    "id": <node_id_to_remove>,
+    "remove": true
+  }
+]
 ```

 Contrary to the CLI that may update only a subset of the fields
 `capacity`, `zone` and `tags`, when calling this API all of these
 values must be specified.

+This returns the new cluster layout with the proposed staged changes,
+as returned by GetClusterLayout.

-#### ApplyClusterLayout `POST /v0/layout/apply`
+
+#### ApplyClusterLayout `POST /v1/layout/apply`

 Applies to the cluster the layout changes currently registered as
 staged layout changes.
@@ -295,7 +339,10 @@ Similarly to the CLI, the body must include the version of the new layout
 that will be created, which MUST be 1 + the value of the currently
 existing layout in the cluster.

-#### RevertClusterLayout `POST /v0/layout/revert`
+This returns the message describing all the calculations done to compute the new
+layout, as well as the description of the layout as returned by GetClusterLayout.
+
+#### RevertClusterLayout `POST /v1/layout/revert`

 Clears all of the staged layout changes.

@@ -313,10 +360,13 @@ Similarly to the CLI, the body must include the incremented
 version number, which MUST be 1 + the value of the currently
 existing layout in the cluster.

+This returns the new cluster layout with all changes reverted,
+as returned by GetClusterLayout.
+

 ### Access key operations

-#### ListKeys `GET /v0/key`
+#### ListKeys `GET /v1/key`

 Returns all API access keys in the cluster.

@@ -335,34 +385,8 @@ Example response:
 ]
 ```

-#### CreateKey `POST /v0/key`
-
-Creates a new API access key.
-
-Request body format:
-
-```json
-{
-    "name": "NameOfMyKey"
-}
-```
-
-#### ImportKey `POST /v0/key/import`
-
-Imports an existing API key.
-
-Request body format:
-
-```json
-{
-    "accessKeyId": "GK31c2f218a2e44f485b94239e",
-    "secretAccessKey": "b892c0665f0ada8a4755dae98baa3b133590e11dae3bcc1f9d769d67f16c3835",
-    "name": "NameOfMyKey"
-}
-```
-
-#### GetKeyInfo `GET /v0/key?id=<acces key id>`
-#### GetKeyInfo `GET /v0/key?search=<pattern>`
+#### GetKeyInfo `GET /v1/key?id=<acces key id>`
+#### GetKeyInfo `GET /v1/key?search=<pattern>`

 Returns information about the requested API access key.

@@ -370,6 +394,9 @@ If `id` is set, the key is looked up using its exact identifier (faster).
 If `search` is set, the key is looked up using its name or prefix
 of identifier (slower, all keys are enumerated to do this).

+Optionnally, the query parameter `showSecretKey=true` can be set to reveal the
+associated secret access key.
+
 Example response:

 ```json
@@ -433,11 +460,40 @@ Example response:
 }
 ```

-#### DeleteKey `DELETE /v0/key?id=<acces key id>`
+#### CreateKey `POST /v1/key`

-Deletes an API access key.
+Creates a new API access key.

-#### UpdateKey `POST /v0/key?id=<acces key id>`
+Request body format:
+
+```json
+{
+    "name": "NameOfMyKey"
+}
+```
+
+This returns the key info, including the created secret key,
+in the same format as the result of GetKeyInfo.
+
+#### ImportKey `POST /v1/key/import`
+
+Imports an existing API key.
+This will check that the imported key is in the valid format, i.e.
+is a key that could have been generated by Garage.
+
+Request body format:
+
+```json
+{
+    "accessKeyId": "GK31c2f218a2e44f485b94239e",
+    "secretAccessKey": "b892c0665f0ada8a4755dae98baa3b133590e11dae3bcc1f9d769d67f16c3835",
+    "name": "NameOfMyKey"
+}
+```
+
+This returns the key info in the same format as the result of GetKeyInfo.
+
+#### UpdateKey `POST /v1/key?id=<acces key id>`

 Updates information about the specified API access key.

@@ -457,10 +513,16 @@ All fields (`name`, `allow` and `deny`) are optional.
 If they are present, the corresponding modifications are applied to the key, otherwise nothing is changed.
 The possible flags in `allow` and `deny` are: `createBucket`.

+This returns the key info in the same format as the result of GetKeyInfo.
+
+#### DeleteKey `DELETE /v1/key?id=<acces key id>`
+
+Deletes an API access key.
+

 ### Bucket operations

-#### ListBuckets `GET /v0/bucket`
+#### ListBuckets `GET /v1/bucket`

 Returns all storage buckets in the cluster.

@@ -502,8 +564,8 @@ Example response:
 ]
 ```

-#### GetBucketInfo `GET /v0/bucket?id=<bucket id>`
-#### GetBucketInfo `GET /v0/bucket?globalAlias=<alias>`
+#### GetBucketInfo `GET /v1/bucket?id=<bucket id>`
+#### GetBucketInfo `GET /v1/bucket?globalAlias=<alias>`

 Returns information about the requested storage bucket.

@@ -535,7 +597,10 @@ Example response:
        ],
        "objects": 14827,
        "bytes": 13189855625,
-        "unfinshedUploads": 0,
+        "unfinishedUploads": 1,
+        "unfinishedMultipartUploads": 1,
+        "unfinishedMultipartUploadParts": 11,
+        "unfinishedMultipartUploadBytes": 41943040,
        "quotas": {
            "maxSize": null,
            "maxObjects": null
@@ -543,7 +608,7 @@ Example response:
 }
 ```

-#### CreateBucket `POST /v0/bucket`
+#### CreateBucket `POST /v1/bucket`

 Creates a new storage bucket.

@@ -583,13 +648,7 @@ or no alias at all.
 Technically, you can also specify both `globalAlias` and `localAlias` and that would create
 two aliases, but I don't see why you would want to do that.

-#### DeleteBucket `DELETE /v0/bucket?id=<bucket id>`
-
-Deletes a storage bucket. A bucket cannot be deleted if it is not empty.
-
-Warning: this will delete all aliases associated with the bucket!
-
-#### UpdateBucket `PUT /v0/bucket?id=<bucket id>`
+#### UpdateBucket `PUT /v1/bucket?id=<bucket id>`

 Updates configuration of the given bucket.

@@ -621,9 +680,16 @@ In `quotas`: new values of `maxSize` and `maxObjects` must both be specified, or
 to remove the quotas. An absent value will be considered the same as a `null`. It is not possible
 to change only one of the two quotas.

+#### DeleteBucket `DELETE /v1/bucket?id=<bucket id>`
+
+Deletes a storage bucket. A bucket cannot be deleted if it is not empty.
+
+Warning: this will delete all aliases associated with the bucket!
+
+
 ### Operations on permissions for keys on buckets

-#### BucketAllowKey `POST /v0/bucket/allow`
+#### BucketAllowKey `POST /v1/bucket/allow`

 Allows a key to do read/write/owner operations on a bucket.

@@ -644,7 +710,7 @@ Request body format:
 Flags in `permissions` which have the value `true` will be activated.
 Other flags will remain unchanged.

-#### BucketDenyKey `POST /v0/bucket/deny`
+#### BucketDenyKey `POST /v1/bucket/deny`

 Denies a key from doing read/write/owner operations on a bucket.

@@ -668,19 +734,19 @@ Other flags will remain unchanged.

 ### Operations on bucket aliases

-#### GlobalAliasBucket `PUT /v0/bucket/alias/global?id=<bucket id>&alias=<global alias>`
+#### GlobalAliasBucket `PUT /v1/bucket/alias/global?id=<bucket id>&alias=<global alias>`

 Empty body. Creates a global alias for a bucket.

-#### GlobalUnaliasBucket `DELETE /v0/bucket/alias/global?id=<bucket id>&alias=<global alias>`
+#### GlobalUnaliasBucket `DELETE /v1/bucket/alias/global?id=<bucket id>&alias=<global alias>`

 Removes a global alias for a bucket.

-#### LocalAliasBucket `PUT /v0/bucket/alias/local?id=<bucket id>&accessKeyId=<access key ID>&alias=<local alias>`
+#### LocalAliasBucket `PUT /v1/bucket/alias/local?id=<bucket id>&accessKeyId=<access key ID>&alias=<local alias>`

 Empty body. Creates a local alias for a bucket in the namespace of a specific access key.

-#### LocalUnaliasBucket `DELETE /v0/bucket/alias/local?id=<bucket id>&accessKeyId<access key ID>&alias=<local alias>`
+#### LocalUnaliasBucket `DELETE /v1/bucket/alias/local?id=<bucket id>&accessKeyId<access key ID>&alias=<local alias>`

 Removes a local alias for a bucket in the namespace of a specific access key.

@@ -0,0 +1,13 @@
+optimal_layout.aux
+optimal_layout.log
+optimal_layout.synctex.gz
+optimal_layout.bbl
+optimal_layout.blg
+
+geodistrib.aux
+geodistrib.bbl
+geodistrib.blg
+geodistrib.log
+geodistrib.out
+geodistrib.synctex.gz
+
@@ -0,0 +1,317 @@
+\documentclass[]{article}
+
+\usepackage{amsmath,amssymb}
+\usepackage{amsthm}
+
+\usepackage{stmaryrd}
+
+\usepackage{graphicx,xcolor}
+\usepackage{hyperref}
+
+\usepackage{algorithm,algpseudocode,float} 
+
+\renewcommand\thesubsubsection{\Alph{subsubsection})}
+
+\newtheorem{proposition}{Proposition}
+
+%opening
+\title{An algorithm for geo-distributed and redundant storage in Garage}
+\author{Mendes Oulamara \\ \emph{mendes@deuxfleurs.fr}}
+\date{}
+
+\begin{document}
+
+\maketitle
+
+\begin{abstract}
+	Garage
+\end{abstract}
+
+\section{Introduction}
+
+Garage\footnote{\url{https://garagehq.deuxfleurs.fr/}} is an open-source distributed object storage service tailored for self-hosting. It was designed by the Deuxfleurs association\footnote{\url{https://deuxfleurs.fr/}} to enable small structures (associations, collectives, small companies) to share storage resources to reliably self-host their data, possibly with old and non-reliable machines.
+
+To achieve these reliability and availability goals, the data is broken into \emph{partitions} and every partition is replicated over 3 different machines (that we call \emph{nodes}). When the data is queried, a consensus algorithm allows to fetch it from one of the nodes. A \emph{replication factor} of 3 ensures the best guarantees in the consensus algorithm \cite{ADD RREF}, but this parameter can be different.
+
+Moreover, if the nodes are spread over different \emph{zones} (different houses, offices, cities\dots), we can ask the data to be replicated over nodes belonging to different zones, to improve the storage robustness against zone failure (such as power outage). To do so, we set a \emph{redundancy parameter}, that is no more than the replication factor, and we ask that any partition is replicated over this number of zones at least.
+
+In this work, we propose a repartition algorithm that, given the nodes specifications and the replication and redundancy parameters, computes an optimal assignation of partitions to nodes. We say that the assignation is optimal in the sense that it maximizes the size of the partitions, and hence the effective storage capacity of the system.
+
+Moreover, when a former assignation exists, which is not optimal anymore due to nodes or zones updates, our algorithm computes a new optimal assignation that minimizes the amount of data to be transferred during the assignation update (the \emph{transfer load}).
+
+We call the set of nodes cooperating to store the data a \emph{cluster}, and a description of the nodes, zones and the assignation of partitions to nodes a \emph{cluster layout}
+
+\subsection{Notations}
+
+Let $k$ be some fixed parameter value, typically 8, that we call the ``partition bits''.
+Every object to be stored in the system is split into data blocks of fixed size. We compute a hash $h(\mathbf{b})$ of every such block $\mathbf{b}$, and we define the $k$ last bits of this hash to be the partition number $p(\mathbf{b})$ of the block. This label can take $P=2^k$ different values, and hence there are $P$ different partitions. We denote $\mathbf{P}$ the set of partition labels (i.e. $\mathbf{P}=\llbracket1,P\rrbracket$).
+
+We are given a set $\mathbf{N}$ of $N$ nodes  and a set  $\mathbf{Z}$ of $Z$ zones. Every node $n$ has a non-negative storage capacity $c_n\ge 0$ and belongs to a zone $z_n\in \mathbf{Z}$. We are also given a replication parameter $\rho_\mathbf{N}$ and a redundancy parameter $\rho_\mathbf{Z}$ such that  $1\le \rho_\mathbf{Z} \le \rho_\mathbf{N}$ (typical values would be $\rho_N=3$ and $\rho_Z=2$).
+
+Our goal is to compute an assignment $\alpha = (\alpha_p^1, \ldots, \alpha_p^{\rho_\mathbf{N}})_{p\in \mathbf{P}}$ such that every partition $p$ is associated to $\rho_\mathbf{N}$ distinct nodes $\alpha_p^1, \ldots, \alpha_p^{\rho_\mathbf{N}} \in \mathbf{N}$ and these nodes belong to at least $\rho_\mathbf{Z}$ distinct zones. Among the possible assignations, we choose one that \emph{maximizes} the effective storage capacity of the cluster. If the layout contained a previous assignment $\alpha'$, we \emph{minimize} the amount of data to transfer during the layout update by making $\alpha$ as close as possible to $\alpha'$. These maximization and minimization are described more formally in the following section.
+
+\subsection{Optimization parameters}
+
+To link the effective storage capacity of the cluster to partition assignment, we make the following assumption:
+\begin{equation}
+	\tag{H1}
+	\text{\emph{All partitions have the same size $s$.}}
+\end{equation}
+This assumption is justified by the dispersion of the hashing function, when the number of partitions is small relative to the number of stored blocks.
+
+Every node $n$ wille store some number $p_n$ of partitions (it is the number of partitions $p$ such that $n$ appears in the $\alpha_p$). Hence the partitions stored by $n$ (and hence all partitions by our assumption) have there size bounded by $c_n/p_n$. This remark leads us to define the optimal size that we will want to maximize:
+
+\begin{equation}
+	\label{eq:optimal}
+	\tag{OPT}
+s^* = \min_{n \in N} \frac{c_n}{p_n}.
+\end{equation}
+
+When the capacities of the nodes are updated (this includes adding or removing a node), we want to update the assignment as well. However, transferring the data between nodes has a cost and we would like to limit the number of changes in the assignment. We make the following assumption:
+\begin{equation}
+	\tag{H2}
+	\text{\emph{Nodes updates happen rarely relatively to block operations.}}
+\end{equation}
+This assumption justifies that when we compute the new assignment $\alpha$, it is worth to optimize the partition size \eqref{eq:optimal} first, and then, among the possible optimal solution, to try to minimize the number of partition transfers. More formally, we minimize the distance between two assignments defined by
+\begin{equation}
+	d(\alpha, \alpha') := \#\{ (n,p) \in \mathbf{N}\times\mathbf{P} ~|~ n\in \alpha_p \triangle \alpha'_p \}
+\end{equation}
+where the symmetric difference $\alpha_p \triangle \alpha'_p$ denotes the nodes appearing in one of the assignations but not in both.
+
+\section{Computation of an optimal assignment}
+
+The algorithm that we propose takes as inputs the cluster layout parameters $\mathbf{N}$, $\mathbf{Z}$, $\mathbf{P}$, $(c_n)_{n\in \mathbf{N}}$, $\rho_\mathbf{N}$, $\rho_\mathbf{Z}$, that we defined in the introduction, together with the former assignation $\alpha'$ (if any). The computation of the new optimal assignation $\alpha^*$ is done in three successive steps that will be detailed in the following sections. The first step computes the largest partition size $s^*$ that an assignation can achieve. The second step computes an optimal candidate assignment $\alpha$ that achieves $s^*$ and a heuristic is used in the computation to make it hopefully close to $\alpha'$. The third steps modifies $\alpha$ iteratively to reduces $d(\alpha, \alpha')$ and yields an assignation $\alpha^*$ achieving $s^*$, and minimizing $d(\cdot, \alpha')$ among such assignations.
+
+We will explain in the next section how to represent an assignment $\alpha$ by a flow $f$ on a weighted graph $G$ to enable the use of flow and graph algorithms. The main function of the algorithm can be written as follows.
+
+\subsubsection*{Algorithm}
+
+\begin{algorithmic}[1]
+	\Function{Compute Layout}{$\mathbf{N}$, $\mathbf{Z}$, $\mathbf{P}$, $(c_n)_{n\in \mathbf{N}}$, $\rho_\mathbf{N}$, $\rho_\mathbf{Z}$, $\alpha'$}
+	\State $s^* \leftarrow$ \Call{Compute Partition Size}{$\mathbf{N}$, $\mathbf{Z}$, $\mathbf{P}$, $(c_n)_{n\in \mathbf{N}}$, $\rho_\mathbf{N}$, $\rho_\mathbf{Z}$}
+	\State $G \leftarrow G(s^*)$
+	\State $f \leftarrow$ 	\Call{Compute Candidate Assignment}{$G$, $\alpha'$}
+	\State $f^* \leftarrow$ 	\Call{Minimize transfer load}{$G$, $f$, $\alpha'$}
+	\State Build $\alpha^*$ from $f^*$
+	\State \Return $\alpha^*$
+	\EndFunction
+\end{algorithmic}
+
+\subsubsection*{Complexity}
+As we will see in the next sections, the worst case complexity of this algorithm is $O(P^2 N^2)$. The minimization of transfer load is the most expensive step, and it can run with a timeout since it is only an optimization step. Without this step (or with a smart timeout), the worst cas complexity can be $O((PN)^{3/2}\log C)$ where $C$ is the total storage capacity of the cluster. 
+
+\subsection{Determination of the partition size $s^*$}
+
+We will represent an assignment $\alpha$ as a flow in a specific graph $G$. We will not compute the optimal partition size $s^*$ a priori, but we will determine it by dichotomy, as the largest size $s$ such that the maximal flow achievable on $G=G(s)$ has value $\rho_\mathbf{N}P$. We will assume that the capacities are given in a small enough unit (say, Megabytes), and we will determine $s^*$ at the precision of the given unit.
+
+Given some candidate size value $s$, we describe the oriented weighted graph $G=(V,E)$ with vertex set $V$ arc set $E$ (see Figure \ref{fig:flowgraph}).
+
+The set of vertices $V$ contains the source $\mathbf{s}$, the sink $\mathbf{t}$, vertices 
+$\mathbf{p^+, p^-}$ for every partition $p$, vertices $\mathbf{x}_{p,z}$ for every partition $p$ and zone $z$, and vertices $\mathbf{n}$ for every node $n$. 
+
+The set of arcs $E$ contains:
+\begin{itemize}
+	\item ($\mathbf{s}$,$\mathbf{p}^+$, $\rho_\mathbf{Z}$) for every partition $p$;
+	\item ($\mathbf{s}$,$\mathbf{p}^-$, $\rho_\mathbf{N}-\rho_\mathbf{Z}$) for every partition $p$;
+	\item ($\mathbf{p}^+$,$\mathbf{x}_{p,z}$, 1) for every partition $p$ and zone $z$;
+	\item ($\mathbf{p}^-$,$\mathbf{x}_{p,z}$, $\rho_\mathbf{N}-\rho_\mathbf{Z}$) for every partition $p$ and zone $z$;
+	\item ($\mathbf{x}_{p,z}$,$\mathbf{n}$, 1) for every partition $p$, zone $z$ and node $n\in z$;
+	\item ($\mathbf{n}$, $\mathbf{t}$, $\lfloor c_n/s \rfloor$) for every node $n$.
+\end{itemize}
+
+\begin{figure}
+	\centering
+	\includegraphics[width=\linewidth]{figures/flow_graph_param}
+	\caption{An example of graph $G(s)$. Arcs are oriented from left to right, and unlabeled arcs have capacity 1. In this example, nodes $n_1,n_2,n_3$ belong to zone $z_1$, and nodes $n_4,n_5$ belong to zone $z_2$.}
+	\label{fig:flowgraph}
+\end{figure}
+
+In the following complexity calculations, we will use the number of vertices and edges of $G$. Remark from now that $\# V = O(PZ)$ and $\# E = O(PN)$.
+
+\begin{proposition}
+	An assignment $\alpha$ is realizable with partition size $s$ and the redundancy constraints $(\rho_\mathbf{N},\rho_\mathbf{Z})$ if and only if there exists a maximal flow function $f$ in $G$ with total flow $\rho_\mathbf{N}P$, such that the arcs ($\mathbf{x}_{p,z}$,$\mathbf{n}$, 1) used are exactly those for which $p$ is associated to $n$ in $\alpha$.
+\end{proposition}
+\begin{proof}
+	Given such flow $f$, we can reconstruct a candidate $\alpha$. In $f$, the flow passing through $\mathbf{p^+}$ and $\mathbf{p^-}$ is $\rho_\mathbf{N}$, and since the outgoing capacity of every $\mathbf{x}_{p,z}$ is 1, every partition is associated to $\rho_\mathbf{N}$ distinct nodes. The fraction $\rho_\mathbf{Z}$ of the flow passing through every $\mathbf{p^+}$ must be spread over as many distinct zones as every arc outgoing from $\mathbf{p^+}$ has capacity 1. So the reconstructed $\alpha$ verifies the redundancy constraints. For every node $n$, the flow between $\mathbf{n}$ and $\mathbf{t}$ corresponds to the number of partitions associated to $n$. By construction of $f$, this does not exceed $\lfloor c_n/s \rfloor$. We assumed that the partition size is $s$, hence this association does not exceed the storage capacity of the nodes.
+	
+	In the other direction, given an assignment $\alpha$, one can similarly check that the facts that $\alpha$ respects the redundancy constraints, and the storage capacities of the nodes, are necessary condition to construct a maximal flow function $f$.
+\end{proof}
+
+\textbf{Implementation remark:} In the flow algorithm, while exploring the graph, we explore the neighbours of every vertex in a random order to heuristically spread the associations between nodes and partitions.
+
+\subsubsection*{Algorithm}
+With this result mind, we can describe the first step of our algorithm. All divisions are supposed to be integer divisions.
+\begin{algorithmic}[1]
+	\Function{Compute Partition Size}{$\mathbf{N}$, $\mathbf{Z}$, $\mathbf{P}$, $(c_n)_{n\in \mathbf{N}}$, $\rho_\mathbf{N}$, $\rho_\mathbf{Z}$}
+	
+	\State Build the graph $G=G(s=1)$
+	\State $ f \leftarrow$ \Call{Maximal flow}{$G$}
+	\If{$f.\mathrm{total flow} < \rho_\mathbf{N}P$}
+	
+	\State \Return Error: capacities too small or constraints too strong.
+	\EndIf
+	
+	\State $s^- \leftarrow 1$
+	\State $s^+ \leftarrow 1+\frac{1}{\rho_\mathbf{N}}\sum_{n \in \mathbf{N}} c_n$
+	
+	\While{$s^-+1 < s^+$}
+	\State Build the graph $G=G(s=(s^-+s^+)/2)$
+	\State $ f \leftarrow$ \Call{Maximal flow}{$G$}
+	\If{$f.\mathrm{total flow} < \rho_\mathbf{N}P$}
+	\State $s^+ \leftarrow (s^- + s^+)/2$
+	\Else
+	\State $s^- \leftarrow (s^- + s^+)/2$
+	\EndIf
+	\EndWhile
+	
+	\State \Return $s^-$
+	\EndFunction
+\end{algorithmic}
+
+\subsubsection*{Complexity}
+
+To compute the maximal flow, we use Dinic's algorithm. Its complexity on general graphs is $O(\#V^2 \#E)$, but on graphs with edge capacity bounded by a constant, it turns out to be $O(\#E^{3/2})$. The graph $G$ does not fall in this case since the capacities of the arcs incoming to $\mathbf{t}$ are far from bounded. However, the proof of this complexity function works readily for graphs where we only ask the edges \emph{not} incoming to the sink $\mathbf{t}$ to have their capacities bounded by a constant. One can find the proof of this claim in \cite[Section 2]{even1975network}.
+The dichotomy adds a logarithmic factor $\log (C)$ where $C=\sum_{n \in \mathbf{N}} c_n$ is the total capacity of the cluster. The total complexity of this first function is hence 
+$O(\#E^{3/2}\log C ) = O\big((PN)^{3/2} \log C\big)$.
+
+\subsubsection*{Metrics}
+We can display the discrepancy between the computed $s^*$ and the best size we could have hoped for the given total capacity, that is $C/\rho_\mathbf{N}$.
+
+\subsection{Computation of a candidate assignment}
+
+Now that we have the optimal partition size $s^*$, to compute a candidate assignment it would be enough to compute a maximal flow function $f$ on $G(s^*)$. This is what we do if there is no former assignation $\alpha'$.
+
+If there is some $\alpha'$, we add a step that will heuristically help to obtain a candidate $\alpha$ closer to $\alpha'$. We fist compute a flow function $\tilde{f}$ that uses only the partition-to-node associations appearing in $\alpha'$. Most likely, $\tilde{f}$ will not be a maximal flow of $G(s^*)$. In Dinic's algorithm, we can start from a non maximal flow function and then discover improving paths. This is what we do by starting from $\tilde{f}$. The hope\footnote{This is only a hope, because one can find examples where the construction of $f$ from $\tilde{f}$ produces an assignment $\alpha$ that is not as close as possible to $\alpha'$.} is that the final flow function $f$ will tend to keep the associations appearing in $\tilde{f}$.
+
+More formally, we construct the graph $G_{|\alpha'}$ from $G$ by removing all the arcs $(\mathbf{x}_{p,z},\mathbf{n}, 1)$ where $p$ is not associated to $n$ in $\alpha'$. We compute a maximal flow function $\tilde{f}$ in $G_{|\alpha'}$. The flow $\tilde{f}$ is also a valid (most likely non maximal) flow function on $G$. We compute a maximal flow function $f$ on $G$ by starting Dinic's algorithm on $\tilde{f}$.
+
+\subsubsection*{Algorithm}
+\begin{algorithmic}[1]
+	\Function{Compute Candidate Assignment}{$G$, $\alpha'$}
+	\State Build the graph $G_{|\alpha'}$
+	\State $ \tilde{f} \leftarrow$ \Call{Maximal flow}{$G_{|\alpha'}$}
+	\State $ f \leftarrow$ \Call{Maximal flow from flow}{$G$, $\tilde{f}$}
+	\State \Return $f$
+	\EndFunction
+\end{algorithmic}
+
+~ 
+
+\textbf{Remark:} The function ``Maximal flow'' can be just seen as the function ``Maximal flow from flow'' called with the zero flow function as starting flow.
+
+\subsubsection*{Complexity}
+With the considerations of the last section, we have the complexity of the Dinic's algorithm $O(\#E^{3/2}) = O((PN)^{3/2})$.
+
+\subsubsection*{Metrics}
+
+We can display the flow value of $\tilde{f}$, which is an upper bound of the distance between $\alpha$ and $\alpha'$. It might be more a Debug level display than Info.
+
+\subsection{Minimization of the transfer load}
+
+Now that we have a candidate flow function $f$, we want to modify it to make its corresponding assignation $\alpha$ as close as possible to $\alpha'$. Denote by $f'$ the maximal flow corresponding to $\alpha'$, and let $d(f, \alpha')=d(f, f'):=d(\alpha,\alpha')$\footnote{It is the number of arcs of type $(\mathbf{x}_{p,z},\mathbf{n})$ saturated in one flow and not in the other.}. 
+We want to build a sequence $f=f_0, f_1, f_2 \dots$ of maximal flows such that $d(f_i, \alpha')$ decreases as $i$ increases. The distance being a non-negative integer, this sequence of flow functions must be finite. We now explain how to find some improving $f_{i+1}$ from $f_i$.
+
+For any maximal flow $f$ in $G$, we define the oriented weighted graph $G_f=(V, E_f)$ as follows. The vertices of $G_f$ are the same as the vertices of $G$. $E_f$ contains the arc $(v_1,v_2, w)$ between vertices $v_1,v_2\in V$ with weight $w$ if and only if the arc $(v_1,v_2)$ is not saturated in $f$ (i.e. $c(v_1,v_2)-f(v_1,v_2) \ge 1$, we also consider reversed arcs). The weight $w$ is: 
+\begin{itemize}
+	\item  $-1$  if $(v_1,v_2)$ is of type $(\mathbf{x}_{p,z},\mathbf{n})$ or $(\mathbf{x}_{p,z},\mathbf{n})$ and is saturated in only one of the two flows $f,f'$;
+	\item $+1$  if $(v_1,v_2)$ is of type $(\mathbf{x}_{p,z},\mathbf{n})$ or $(\mathbf{x}_{p,z},\mathbf{n})$ and is saturated in either both or none of the two flows $f,f'$;
+	\item $0$ otherwise.
+\end{itemize}
+
+If $\gamma$ is a simple cycle of arcs in $G_f$, we define its weight $w(\gamma)$ as the sum of the weights of its arcs. We can add $+1$ to the value of $f$ on the arcs of $\gamma$, and by construction of $G_f$ and the fact that $\gamma$ is a cycle, the function that we get is still a valid flow function on $G$, it is maximal as it has the same flow value as $f$. We denote this new function $f+\gamma$.
+
+\begin{proposition}
+	Given a maximal flow $f$ and a simple cycle $\gamma$ in $G_f$, we have $d(f+\gamma, f') - d(f,f') = w(\gamma)$.
+\end{proposition}
+\begin{proof}
+	Let $X$ be the set of arcs of type $(\mathbf{x}_{p,z},\mathbf{n})$. Then we can express $d(f,f')$ as
+	\begin{align*}
+		d(f,f') & = \#\{e\in X ~|~ f(e)\neq f'(e)\}
+		= \sum_{e\in X} 1_{f(e)\neq f'(e)} \\
+		& = \frac{1}{2}\big( \#X + \sum_{e\in X} 1_{f(e)\neq f'(e)} - 1_{f(e)= f'(e)} \big).
+	\end{align*}
+	We can express the cycle weight as
+	\begin{align*}
+		w(\gamma) &  = \sum_{e\in X, e\in \gamma} - 1_{f(e)\neq f'(e)} + 1_{f(e)= f'(e)}.
+	\end{align*}
+	Remark that since we passed on unit of flow in $\gamma$ to construct $f+\gamma$, we have for any $e\in X$, $f(e)=f'(e)$ if and only if $(f+\gamma)(e) \neq f'(e)$.
+	Hence
+	\begin{align*}
+		w(\gamma) &  = \frac{1}{2}(w(\gamma) + w(\gamma)) \\
+		&= \frac{1}{2} \Big( 
+		\sum_{e\in X, e\in \gamma} - 1_{f(e)\neq f'(e)} +  1_{f(e)= f'(e)} \\
+		& \qquad +
+		\sum_{e\in X, e\in \gamma} 1_{(f+\gamma)(e)\neq f'(e)} +  1_{(f+\gamma)(e)= f'(e)}
+		\Big).
+	\end{align*}
+	Plugging this in the previous equation, we find that 
+	$$d(f,f')+w(\gamma) = d(f+\gamma, f').$$
+\end{proof}
+
+This result suggests that given some flow $f_i$, we just need to find a negative cycle $\gamma$ in $G_{f_i}$ to construct $f_{i+1}$ as $f_i+\gamma$. The following proposition ensures that this greedy strategy reaches an optimal flow. 
+
+\begin{proposition}
+	For any maximal flow $f$, $G_f$ contains a negative cycle if and only if there exists a maximal flow $f^*$ in $G$ such that $d(f^*, f') < d(f, f')$.
+\end{proposition}
+\begin{proof}
+	Suppose that there is such flow $f^*$. Define the oriented multigraph $M_{f,f^*}=(V,E_M)$ with the same vertex set $V$ as in $G$, and for every $v_1,v_2 \in V$, $E_M$ contains $(f^*(v_1,v_2) - f(v_1,v_2))_+$  copies of the arc $(v_1,v_2)$. For every vertex $v$, its total degree (meaning its outer degree minus its inner degree) is equal to 
+	\begin{align*}
+		\deg v & = \sum_{u\in V} (f^*(v,u) - f(v,u))_+ - \sum_{u\in V} (f^*(u,v) - f(u,v))_+  \\
+		& = \sum_{u\in V} f^*(v,u) - f(v,u) = \sum_{u\in V} f^*(v,u) - \sum_{u\in V}  f(v,u).
+	\end{align*}
+	The last two sums are zero for any inner vertex since $f,f^*$ are flows, and they are equal on the source and sink since the two flows are both maximal and have hence the same value. Thus, $\deg v = 0$ for every vertex $v$.
+	
+	This implies that the multigraph $M_{f,f^*}$ is the union of disjoint simple cycles. $f$ can be transformed into $f^*$ by pushing a mass 1 along all these cycles in any order. Since $d(f^*, f')<d(f,f')$, there must exists one of these simple cycles $\gamma$ with $d(f+\gamma, f') < d(f, f')$. Finally, since we can push a mass in $f$ along $\gamma$, it must appear in $G_f$. Hence $\gamma$ is a cycle of $G_f$ with negative weight.
+\end{proof}
+
+In the next section we describe the corresponding algorithm. Instead of discovering only one cycle, we are allowed to discover a set $\Gamma$ of disjoint negative cycles. 
+
+\subsubsection*{Algorithm}
+\begin{algorithmic}[1]
+	\Function{Minimize transfer load}{$G$, $f$, $\alpha'$}
+	\State Build the graph $G_f$
+	\State $\Gamma \leftarrow$ \Call{Detect Negative Cycles}{$G_f$}
+	\While{$\Gamma \neq \emptyset$}
+	\ForAll{$\gamma \in \Gamma$}
+	\State $f \leftarrow f+\gamma$
+	\EndFor
+	\State Update $G_f$
+	\State $\Gamma \leftarrow$ \Call{Detect Negative Cycles}{$G_f$}
+	\EndWhile
+	\State \Return $f$
+	\EndFunction
+\end{algorithmic}
+
+\subsubsection*{Complexity}
+The distance $d(f,f')$ is bounded by the maximal number of differences in the associated assignment. If these assignment are totally disjoint, this distance is $2\rho_N P$. At every iteration of the While loop, the distance decreases, so there is at most $O(\rho_N P) = O(P)$ iterations.
+
+The detection of negative cycle is done with the Bellman-Ford algorithm, whose complexity should normally be $O(\#E\#V)$. In our case, it amounts to $O(P^2ZN)$. Multiplied by the complexity of the outer loop, it amounts to $O(P^3ZN)$ which is a lot when the number of partitions and nodes starts to be large. To avoid that, we adapt the Bellman-Ford algorithm.
+
+The Bellman-Ford algorithm runs $\#V$ iterations of an outer loop, and an inner loop over $E$. The idea is to compute the shortest paths from a source vertex $v$ to all other vertices. After $k$ iterations of the outer loop, the algorithm has computed all shortest path of length at most $k$. All simple paths have length at most $\#V-1$, so if there is an update in the last iteration of the loop, it means that there is a negative cycle in the graph. The observation that will enable us to improve the complexity is the following:
+
+\begin{proposition}
+	In the graph $G_f$ (and $G$), all simple paths have a length at most $4N$.
+\end{proposition}
+\begin{proof}
+	Since $f$ is a maximal flow, there is no outgoing edge from $\mathbf{s}$ in $G_f$. One can thus check than any simple path of length 4 must contain at least two node of type $\mathbf{n}$. Hence on a path, at most 4 arcs separate two successive nodes of type $\mathbf{n}$. 
+\end{proof}
+
+Thus, in the absence of negative cycles, shortest paths in $G_f$ have length at most $4N$. So we can do only $4N+1$ iterations of the outer loop in the Bellman-Ford algorithm. This makes the complexity of the detection of one set of cycle to be $O(N\#E) = O(N^2 P)$.
+
+With this improvement, the complexity of the whole algorithm is, in the worst case, $O(N^2P^2)$. However, since we detect several cycles at once and we start with a flow that might be close to the previous one, the number of iterations of the outer loop might be smaller in practice.
+
+
+
+\subsubsection*{Metrics}
+We can display the node and zone utilization ratio, by dividing the flow passing through them divided by their outgoing capacity. In particular, we can pinpoint saturated nodes and zones (i.e. used at their full potential).
+
+We can display the distance to the previous assignment, and the number of partition transfers.
+
+
+\bibliography{optimal_layout} 
+\bibliographystyle{ieeetr}
+
+\end{document}
+
+
+
@@ -0,0 +1,11 @@
+
+@article{even1975network,
+	title={Network flow and testing graph connectivity},
+	author={Even, Shimon and Tarjan, R Endre},
+	journal={SIAM journal on computing},
+	volume={4},
+	number={4},
+	pages={507--518},
+	year={1975},
+	publisher={SIAM}
+}
@@ -0,0 +1,709 @@
+\documentclass[]{article}
+
+\usepackage{amsmath,amssymb}
+\usepackage{amsthm}
+
+\usepackage{graphicx,xcolor}
+
+\usepackage{algorithm,algpseudocode,float} 
+
+\renewcommand\thesubsubsection{\Alph{subsubsection})}
+
+\newtheorem{proposition}{Proposition}
+
+%opening
+\title{Optimal partition assignment in Garage}
+\author{Mendes}
+
+\begin{document}
+
+\maketitle
+
+\section{Introduction}
+
+\subsection{Context}
+
+Garage is an open-source distributed storage service blablabla$\dots$
+
+Every object to be stored in the system falls in a partition given by the last $k$ bits of its hash. There are $P=2^k$ partitions. Every partition will be stored on distinct nodes of the system. The goal of the assignment of partitions to nodes is to ensure (nodes and zone) redundancy and to be as efficient as possible.
+
+\subsection{Formal description of the problem}
+
+We are given a set of nodes $\mathbf{N}$ and a set of zones $\mathbf{Z}$. Every node $n$ has a non-negative storage capacity $c_n\ge 0$ and belongs to a zone $z\in \mathbf{Z}$. We are also given a number of partition $P>0$ (typically $P=256$).
+
+We would like to compute an assignment of nodes to partitions. We will impose some redundancy constraints to this assignment, and under these constraints, we want our system to have the largest storage capacity possible. To link storage capacity to partition assignment, we make the following assumption:
+\begin{equation}
+	\tag{H1}
+	\text{\emph{All partitions have the same size $s$.}}
+\end{equation}
+This assumption is justified by the dispersion of the hashing function, when the number of partitions is small relative to the number of stored large objects.
+
+Every node $n$ wille store some number $k_n$ of partitions. Hence the partitions stored by $n$ (and hence all partitions by our assumption) have there size bounded by $c_n/k_n$. This remark leads us to define the optimal size that we will want to maximize:
+
+\begin{equation}
+	\label{eq:optimal}
+	\tag{OPT}
+s^* = \min_{n \in N} \frac{c_n}{k_n}.
+\end{equation}
+
+When the capacities of the nodes are updated (this includes adding or removing a node), we want to update the assignment as well. However, transferring the data between nodes has a cost and we would like to limit the number of changes in the assignment. We make the following assumption:
+\begin{equation}
+	\tag{H2}
+	\text{\emph{Updates of capacity happens rarely relatively to object storing.}}
+\end{equation}
+This assumption justifies that when we compute the new assignment, it is worth to optimize the partition size \eqref{eq:optimal} first, and then, among the possible optimal solution, to try to minimize the number of partition transfers.
+
+For now, in the following, we ask the following redundancy constraint:
+
+\textbf{Parametric node and zone redundancy:} Given two integer parameters $1\le \rho_\mathbf{Z} \le \rho_\mathbf{N}$, we ask every partition to be stored on $\rho_\mathbf{N}$ distinct nodes, and these nodes must belong to at least $\rho_\mathbf{Z}$ distinct zones.
+
+
+\textbf{Mode 3-strict:}  every partition needs to be assignated to three nodes belonging to three different zones.
+
+\textbf{Mode 3:} every partition needs to be assignated to three nodes. We try to spread the three nodes over different zones as much as possible.
+
+\textbf{Warning:} This is a working document written incrementaly. The last version of the algorithm is the \textbf{parametric assignment} described in the next section.
+
+
+\section{Computation of a parametric assignment}
+\textbf{Attention : }We change notations in this section.
+
+Notations : let $P$ be the number of partitions, $N$ the number of nodes, $Z$ the number of zones. Let $\mathbf{P,N,Z}$ be the label sets of, respectively, partitions, nodes and zones.
+Let $s^*$ be the largest partition size achievable with the redundancy constraints. Let $(c_n)_{n\in \mathbf{N}}$ be the storage capacity of every node. 
+
+In this section, we propose a third specification of the problem. The user inputs two redundancy parameters $1\le \rho_\mathbf{Z} \le \rho_\mathbf{N}$. We compute an assignment $\alpha = (\alpha_p^1, \ldots, \alpha_p^{\rho_\mathbf{N}})_{p\in \mathbf{P}}$ such that every partition $p$ is associated to $\rho_\mathbf{N}$ distinct nodes $\alpha_p^1, \ldots, \alpha_p^{\rho_\mathbf{N}}$ and these nodes belong to at least $\rho_\mathbf{Z}$ distinct zones.
+
+If the layout contained a previous assignment $\alpha'$, we try to minimize the amount of data to transfer during the layout update by making $\alpha$ as close as possible to $\alpha'$.
+
+In the following subsections, we describe the successive steps of the algorithm we propose to compute $\alpha$.
+
+\subsubsection*{Algorithm}
+
+\begin{algorithmic}[1]
+	\Function{Compute Layout}{$\mathbf{N}$, $\mathbf{Z}$, $\mathbf{P}$, $(c_n)_{n\in \mathbf{N}}$, $\rho_\mathbf{N}$, $\rho_\mathbf{Z}$, $\alpha'$}
+	\State $s^* \leftarrow$ \Call{Compute Partition Size}{$\mathbf{N}$, $\mathbf{Z}$, $\mathbf{P}$, $(c_n)_{n\in \mathbf{N}}$, $\rho_\mathbf{N}$, $\rho_\mathbf{Z}$}
+	\State $G \leftarrow G(s^*)$
+	\State $f \leftarrow$ 	\Call{Compute Candidate Assignment}{$G$, $\alpha'$}
+	\State $f^* \leftarrow$ 	\Call{Minimize transfer load}{$G$, $f$, $\alpha'$}
+	\State Build $\alpha^*$ from $f^*$
+	\State \Return $\alpha^*$
+	\EndFunction
+\end{algorithmic}
+
+\subsubsection*{Complexity}
+As we will see in the next sections, the worst case complexity of this algorithm is $O(P^2 N^2)$. The minimization of transfer load is the most expensive step, and it can run with a timeout since it is only an optimization step. Without this step (or with a smart timeout), the worst cas complexity can be $O((PN)^{3/2}\log C)$ where $C$ is the total storage capacity of the cluster. 
+
+\subsection{Determination of the partition size $s^*$}
+
+Again, we will represent an assignment $\alpha$ as a flow in a specific graph $G$. We will not compute the optimal partition size $s^*$ a priori, but we will determine it by dichotomy, as the largest size $s$ such that the maximal flow achievable on $G=G(s)$ has value $\rho_\mathbf{N}P$. We will assume that the capacities are given in a small enough unit (say, Megabytes), and we will determine $s^*$ at the precision of the given unit.
+
+Given some candidate size value $s$, we describe the oriented weighted graph $G=(V,E)$ with vertex set $V$ arc set $E$.
+
+The set of vertices $V$ contains the source $\mathbf{s}$, the sink $\mathbf{t}$, vertices 
+$\mathbf{p^+, p^-}$ for every partition $p$, vertices $\mathbf{x}_{p,z}$ for every partition $p$ and zone $z$, and vertices $\mathbf{n}$ for every node $n$. 
+
+The set of arcs $E$ contains:
+\begin{itemize}
+	\item ($\mathbf{s}$,$\mathbf{p}^+$, $\rho_\mathbf{Z}$) for every partition $p$;
+	\item ($\mathbf{s}$,$\mathbf{p}^-$, $\rho_\mathbf{N}-\rho_\mathbf{Z}$) for every partition $p$;
+	\item ($\mathbf{p}^+$,$\mathbf{x}_{p,z}$, 1) for every partition $p$ and zone $z$;
+	\item ($\mathbf{p}^-$,$\mathbf{x}_{p,z}$, $\rho_\mathbf{N}-\rho_\mathbf{Z}$) for every partition $p$ and zone $z$;
+	\item ($\mathbf{x}_{p,z}$,$\mathbf{n}$, 1) for every partition $p$, zone $z$ and node $n\in z$;
+	\item ($\mathbf{n}$, $\mathbf{t}$, $\lfloor c_n/s \rfloor$) for every node $n$.
+\end{itemize}
+
+In the following complexity calculations, we will use the number of vertices and edges of $G$. Remark from now that $\# V = O(PZ)$ and $\# E = O(PN)$.
+
+\begin{proposition}
+	An assignment $\alpha$ is realizable with partition size $s$ and the redundancy constraints $(\rho_\mathbf{N},\rho_\mathbf{Z})$ if and only if there exists a maximal flow function $f$ in $G$ with total flow $\rho_\mathbf{N}P$, such that the arcs ($\mathbf{x}_{p,z}$,$\mathbf{n}$, 1) used are exactly those for which $p$ is associated to $n$ in $\alpha$.
+\end{proposition}
+\begin{proof}
+	Given such flow $f$, we can reconstruct a candidate $\alpha$. In $f$, the flow passing through $\mathbf{p^+}$ and $\mathbf{p^-}$ is $\rho_\mathbf{N}$, and since the outgoing capacity of every $\mathbf{x}_{p,z}$ is 1, every partition is associated to $\rho_\mathbf{N}$ distinct nodes. The fraction $\rho_\mathbf{Z}$ of the flow passing through every $\mathbf{p^+}$ must be spread over as many distinct zones as every arc outgoing from $\mathbf{p^+}$ has capacity 1. So the reconstructed $\alpha$ verifies the redundancy constraints. For every node $n$, the flow between $\mathbf{n}$ and $\mathbf{t}$ corresponds to the number of partitions associated to $n$. By construction of $f$, this does not exceed $\lfloor c_n/s \rfloor$. We assumed that the partition size is $s$, hence this association does not exceed the storage capacity of the nodes.
+	
+	In the other direction, given an assignment $\alpha$, one can similarly check that the facts that $\alpha$ respects the redundancy constraints, and the storage capacities of the nodes, are necessary condition to construct a maximal flow function $f$.
+\end{proof}
+
+\textbf{Implementation remark:} In the flow algorithm, while exploring the graph, we explore the neighbours of every vertex in a random order to heuristically spread the association between nodes and partitions.
+
+\subsubsection*{Algorithm}
+With this result mind, we can describe the first step of our algorithm. All divisions are supposed to be integer division.
+\begin{algorithmic}[1]
+	\Function{Compute Partition Size}{$\mathbf{N}$, $\mathbf{Z}$, $\mathbf{P}$, $(c_n)_{n\in \mathbf{N}}$, $\rho_\mathbf{N}$, $\rho_\mathbf{Z}$}
+	
+	\State Build the graph $G=G(s=1)$
+	\State $ f \leftarrow$ \Call{Maximal flow}{$G$}
+	\If{$f.\mathrm{total flow} < \rho_\mathbf{N}P$}
+	
+	\State \Return Error: capacities too small or constraints too strong.
+	\EndIf
+	
+	\State $s^- \leftarrow 1$
+	\State $s^+ \leftarrow 1+\frac{1}{\rho_\mathbf{N}}\sum_{n \in \mathbf{N}} c_n$
+	
+	\While{$s^-+1 < s^+$}
+	\State Build the graph $G=G(s=(s^-+s^+)/2)$
+	\State $ f \leftarrow$ \Call{Maximal flow}{$G$}
+	\If{$f.\mathrm{total flow} < \rho_\mathbf{N}P$}
+	\State $s^+ \leftarrow (s^- + s^+)/2$
+	\Else
+	\State $s^- \leftarrow (s^- + s^+)/2$
+	\EndIf
+	\EndWhile
+	
+	\State \Return $s^-$
+	\EndFunction
+\end{algorithmic}
+
+\subsubsection*{Complexity}
+
+To compute the maximal flow, we use Dinic's algorithm. Its complexity on general graphs is $O(\#V^2 \#E)$, but on graphs with edge capacity bounded by a constant, it turns out to be $O(\#E^{3/2})$. The graph $G$ does not fall in this case since the capacities of the arcs incoming to $\mathbf{t}$ are far from bounded. However, the proof of this complexity works readily for graph where we only ask the edges \emph{not} incoming to the sink $\mathbf{t}$ to have their capacities bounded by a constant. One can find the proof of this claim in \cite[Section 2]{even1975network}.
+The dichotomy adds a logarithmic factor $\log (C)$ where $C=\sum_{n \in \mathbf{N}} c_n$ is the total capacity of the cluster. The total complexity of this first function is hence 
+$O(\#E^{3/2}\log C ) = O\big((PN)^{3/2} \log C\big)$.
+
+\subsubsection*{Metrics}
+We can display the discrepancy between the computed $s^*$ and the best size we could hope for a given total capacity, that is $C/\rho_\mathbf{N}$.
+
+\subsection{Computation of a candidate assignment}
+
+Now that we have the optimal partition size $s^*$, to compute a candidate assignment, it would be enough to compute a maximal flow function $f$ on $G(s^*)$. This is what we do if there was no previous assignment $\alpha'$.
+
+If there was some $\alpha'$, we add a step that will heuristically help to obtain a candidate $\alpha$ closer to $\alpha'$. to do so, we fist compute a flow function $\tilde{f}$ that uses only the partition-to-node association appearing in $\alpha'$. Most likely, $\tilde{f}$ will not be a maximal flow of $G(s^*)$. In Dinic's algorithm, we can start from a non maximal flow function and then discover improving paths. This is what we do in starting from $\tilde{f}$. The hope\footnote{This is only a hope, because one can find examples where the construction of $f$ from $\tilde{f}$ produces an assignment $\alpha$ that is not as close as possible to $\alpha'$.} is that the final flow function $f$ will tend to keep the associations appearing in $\tilde{f}$.
+
+More formally, we construct the graph $G_{|\alpha'}$ from $G$ by removing all the arcs $(\mathbf{x}_{p,z},\mathbf{n}, 1)$ where $p$ is not associated to $n$ in $\alpha'$. We compute a maximal flow function $\tilde{f}$ in $G_{|\alpha'}$. $\tilde{f}$ is also a valid (most likely non maximal) flow function in $G$. We compute a maximal flow function $f$ on $G$ by starting Dinic's algorithm on $\tilde{f}$.
+
+\subsubsection*{Algorithm}
+\begin{algorithmic}[1]
+	\Function{Compute Candidate Assignment}{$G$, $\alpha'$}
+	\State Build the graph $G_{|\alpha'}$
+	\State $ \tilde{f} \leftarrow$ \Call{Maximal flow}{$G_{|\alpha'}$}
+	\State $ f \leftarrow$ \Call{Maximal flow from flow}{$G$, $\tilde{f}$}
+	\State \Return $f$
+	\EndFunction
+\end{algorithmic}
+
+\textbf{Remark:} The function ``Maximal flow'' can be just seen as the function ``Maximal flow from flow'' called with the zero flow function as starting flow.
+
+\subsubsection*{Complexity}
+From the consideration of the last section, we have the complexity of the Dinic's algorithm $O(\#E^{3/2}) = O((PN)^{3/2})$.
+
+\subsubsection*{Metrics}
+
+We can display the flow value of $\tilde{f}$, which is an upper bound of the distance between $\alpha$ and $\alpha'$. It might be more a Debug level display than Info.
+
+\subsection{Minimization of the transfer load}
+
+Now that we have a candidate flow function $f$, we want to modify it to make its associated assignment as close as possible to $\alpha'$. Denote by $f'$ the maximal flow associated to $\alpha'$, and let $d(f, f')$ be distance between the associated assignments\footnote{It is the number of arcs of type $(\mathbf{x}_{p,z},\mathbf{n})$ saturated in one flow and not in the other.}. 
+We want to build a sequence $f=f_0, f_1, f_2 \dots$ of maximal flows such that $d(f_i, \alpha')$ decreases as $i$ increases. The distance being a non-negative integer, this sequence of flow functions must be finite. We now explain how to find some improving $f_{i+1}$ from $f_i$.
+
+For any maximal flow $f$ in $G$, we define the oriented weighted graph $G_f=(V, E_f)$ as follows. The vertices of $G_f$ are the same as the vertices of $G$. $E_f$ contains the arc $(v_1,v_2, w)$ between vertices $v_1,v_2\in V$ with weight $w$ if and only if the arc $(v_1,v_2)$ is not saturated in $f$ (i.e. $c(v_1,v_2)-f(v_1,v_2) \ge 1$, we also consider reversed arcs). The weight $w$ is: 
+\begin{itemize}
+	\item  $-1$  if $(v_1,v_2)$ is of type $(\mathbf{x}_{p,z},\mathbf{n})$ or $(\mathbf{x}_{p,z},\mathbf{n})$ and is saturated in only one of the two flows $f,f'$;
+	\item $+1$  if $(v_1,v_2)$ is of type $(\mathbf{x}_{p,z},\mathbf{n})$ or $(\mathbf{x}_{p,z},\mathbf{n})$ and is saturated in either both or none of the two flows $f,f'$;
+	\item $0$ otherwise.
+\end{itemize}
+
+If $\gamma$ is a simple cycle of arcs in $G_f$, we define its weight $w(\gamma)$ as the sum of the weights of its arcs. We can add $+1$ to the value of $f$ on the arcs of $\gamma$, and by construction of $G_f$ and the fact that $\gamma$ is a cycle, the function that we get is still a valid flow function on $G$, it is maximal as it has the same flow value as $f$. We denote this new function $f+\gamma$.
+
+\begin{proposition}
+	Given a maximal flow $f$ and a simple cycle $\gamma$ in $G_f$, we have $d(f+\gamma, f') - d(f,f') = w(\gamma)$.
+\end{proposition}
+\begin{proof}
+	Let $X$ be the set of arcs of type $(\mathbf{x}_{p,z},\mathbf{n})$. Then we can express $d(f,f')$ as
+	\begin{align*}
+		d(f,f') & = \#\{e\in X ~|~ f(e)\neq f'(e)\}
+		= \sum_{e\in X} 1_{f(e)\neq f'(e)} \\
+		& = \frac{1}{2}\big( \#X + \sum_{e\in X} 1_{f(e)\neq f'(e)} - 1_{f(e)= f'(e)} \big).
+	\end{align*}
+	We can express the cycle weight as
+	\begin{align*}
+		w(\gamma) &  = \sum_{e\in X, e\in \gamma} - 1_{f(e)\neq f'(e)} + 1_{f(e)= f'(e)}.
+	\end{align*}
+	Remark that since we passed on unit of flow in $\gamma$ to construct $f+\gamma$, we have for any $e\in X$, $f(e)=f'(e)$ if and only if $(f+\gamma)(e) \neq f'(e)$.
+	Hence
+	\begin{align*}
+		w(\gamma) &  = \frac{1}{2}(w(\gamma) + w(\gamma)) \\
+		&= \frac{1}{2} \Big( 
+		\sum_{e\in X, e\in \gamma} - 1_{f(e)\neq f'(e)} +  1_{f(e)= f'(e)} \\
+		& \qquad +
+		\sum_{e\in X, e\in \gamma} 1_{(f+\gamma)(e)\neq f'(e)} +  1_{(f+\gamma)(e)= f'(e)}
+		\Big).
+	\end{align*}
+	Plugging this in the previous equation, we find that 
+	$$d(f,f')+w(\gamma) = d(f+\gamma, f').$$
+\end{proof}
+
+This result suggests that given some flow $f_i$, we just need to find a negative cycle $\gamma$ in $G_{f_i}$ to construct $f_{i+1}$ as $f_i+\gamma$. The following proposition ensures that this greedy strategy reaches an optimal flow. 
+
+\begin{proposition}
+	For any maximal flow $f$, $G_f$ contains a negative cycle if and only if there exists a maximal flow $f^*$ in $G$ such that $d(f^*, f') < d(f, f')$.
+\end{proposition}
+\begin{proof}
+	Suppose that there is such flow $f^*$. Define the oriented multigraph $M_{f,f^*}=(V,E_M)$ with the same vertex set $V$ as in $G$, and for every $v_1,v_2 \in V$, $E_M$ contains $(f^*(v_1,v_2) - f(v_1,v_2))_+$  copies of the arc $(v_1,v_2)$. For every vertex $v$, its total degree (meaning its outer degree minus its inner degree) is equal to 
+	\begin{align*}
+		\deg v & = \sum_{u\in V} (f^*(v,u) - f(v,u))_+ - \sum_{u\in V} (f^*(u,v) - f(u,v))_+  \\
+		& = \sum_{u\in V} f^*(v,u) - f(v,u) = \sum_{u\in V} f^*(v,u) - \sum_{u\in V}  f(v,u).
+	\end{align*}
+	The last two sums are zero for any inner vertex since $f,f^*$ are flows, and they are equal on the source and sink since the two flows are both maximal and have hence the same value. Thus, $\deg v = 0$ for every vertex $v$.
+	
+	This implies that the multigraph $M_{f,f^*}$ is the union of disjoint simple cycles. $f$ can be transformed into $f^*$ by pushing a mass 1 along all these cycles in any order. Since $d(f^*, f')<d(f,f')$, there must exists one of these simple cycles $\gamma$ with $d(f+\gamma, f') < d(f, f')$. Finally, since we can push a mass in $f$ along $\gamma$, it must appear in $G_f$. Hence $\gamma$ is a cycle of $G_f$ with negative weight.
+\end{proof}
+
+In the next section we describe the corresponding algorithm. Instead of discovering only one cycle, we are allowed to discover a set $\Gamma$ of disjoint negative cycles. 
+
+\subsubsection*{Algorithm}
+\begin{algorithmic}[1]
+	\Function{Minimize transfer load}{$G$, $f$, $\alpha'$}
+	\State Build the graph $G_f$
+	\State $\Gamma \leftarrow$ \Call{Detect Negative Cycles}{$G_f$}
+	\While{$\Gamma \neq \emptyset$}
+	\ForAll{$\gamma \in \Gamma$}
+	\State $f \leftarrow f+\gamma$
+	\EndFor
+	\State Update $G_f$
+	\State $\Gamma \leftarrow$ \Call{Detect Negative Cycles}{$G_f$}
+	\EndWhile
+	\State \Return $f$
+	\EndFunction
+\end{algorithmic}
+
+\subsubsection*{Complexity}
+The distance $d(f,f')$ is bounded by the maximal number of differences in the associated assignment. If these assignment are totally disjoint, this distance is $2\rho_N P$. At every iteration of the While loop, the distance decreases, so there is at most $O(\rho_N P) = O(P)$ iterations.
+
+The detection of negative cycle is done with the Bellman-Ford algorithm, whose complexity should normally be $O(\#E\#V)$. In our case, it amounts to $O(P^2ZN)$. Multiplied by the complexity of the outer loop, it amounts to $O(P^3ZN)$ which is a lot when the number of partitions and nodes starts to be large. To avoid that, we adapt the Bellman-Ford algorithm.
+
+The Bellman-Ford algorithm runs $\#V$ iterations of an outer loop, and an inner loop over $E$. The idea is to compute the shortest paths from a source vertex $v$ to all other vertices. After $k$ iterations of the outer loop, the algorithm has computed all shortest path of length at most $k$. All simple paths have length at most $\#V-1$, so if there is an update in the last iteration of the loop, it means that there is a negative cycle in the graph. The observation that will enable us to improve the complexity is the following:
+
+\begin{proposition}
+	In the graph $G_f$ (and $G$), all simple paths have a length at most $4N$.
+\end{proposition}
+\begin{proof}
+	Since $f$ is a maximal flow, there is no outgoing edge from $\mathbf{s}$ in $G_f$. One can thus check than any simple path of length 4 must contain at least two node of type $\mathbf{n}$. Hence on a path, at most 4 arcs separate two successive nodes of type $\mathbf{n}$. 
+\end{proof}
+
+Thus, in the absence of negative cycles, shortest paths in $G_f$ have length at most $4N$. So we can do only $4N+1$ iterations of the outer loop in Bellman-Ford algorithm. This makes the complexity of the detection of one set of cycle to be $O(N\#E) = O(N^2 P)$.
+
+With this improvement, the complexity of the whole algorithm is, in the worst case, $O(N^2P^2)$. However, since we detect several cycles at once and we start with a flow that might be close to the previous one, the number of iterations of the outer loop might be smaller in practice.
+
+
+
+\subsubsection*{Metrics}
+We can display the node and zone utilization ratio, by dividing the flow passing through them divided by their outgoing capacity. In particular, we can pinpoint saturated nodes and zones (i.e. used at their full potential).
+
+We can display the distance to the previous assignment, and the number of partition transfers.
+
+
+
+
+
+\section{Properties of an optimal 3-strict assignment}
+
+\subsection{Optimal assignment}
+\label{sec:opt_assign}
+
+For every zone $z\in Z$, define the zone capacity $c_z = \sum_{v, z_v=z} c_v$ and define $C = \sum_v c_v = \sum_z c_z$.
+
+One can check that the best we could be doing to maximize $s^*$ would be to use the nodes proportionally to their capacity. This would yield $s^*=C/(3N)$. This is not possible because of (i) redundancy constraints and (ii) integer rounding but it gives and upper bound. 
+
+\subsubsection*{Optimal utilization}
+
+We call an \emph{utilization} a collection of non-negative integers $(n_v)_{v\in V}$ such that $\sum_v n_v = 3N$ and for every zone $z$, $\sum_{v\in z} n_v \le N$. We call such utilization \emph{optimal} if it maximizes $s^*$.
+
+We start by computing a node sub-utilization $(\hat{n}_v)_{v\in V}$ such that for every zone $z$, $\sum_{v\in z} \hat{n}_v \le N$ and we show that there is an optimal utilization respecting the constraints and such that $\hat{n}_v \le n_v$ for every node.
+
+Assume that there is a zone $z_0$ such that $c_{z_0}/C \ge 1/3$. Then for any $v\in z_0$, we define 
+$$\hat{n}_v = \left\lfloor\frac{c_v}{c_{z_0}}N\right\rfloor.$$ 
+This choice ensures for any such $v$ that
+$$
+\frac{c_v}{\hat{n}_v} \ge \frac{c_{z_0}}{N} \ge \frac{C}{3N}
+$$
+which is the universal upper bound on $s^*$. Hence any optimal utilization $(n_v)$ can be modified to another optimal utilization such that $n_v\ge \hat{n}_v$
+
+Because $z_0$ cannot store more than $N$ partition occurences, in any assignment, at least $2N$ partitions must be assignated to the zones $Z\setminus\{z_0\}$. Let $C_0 = C-c_{z_0}$. Suppose that there exists a zone $z_1\neq z_0$ such that $c_{z_1}/C_0 \ge 1/2$. Then, with the same argument as for $z_0$, we can define 
+$$\hat{n}_v = \left\lfloor\frac{c_v}{c_{z_1}}N\right\rfloor$$
+for every $v\in z_1$.
+
+Now we can assign the remaining partitions. Let $(\hat{N}, \hat{C})$ to be
+\begin{itemize}
+	\item $(3N,C)$ if we did not find any $z_0$;
+	\item $(2N,C-c_{z_0})$ if there was a $z_0$ but no $z_1$;
+	\item $(N,C-c_{z_0}-c_{z_1})$ if there was a $z_0$ and a $z_1$.
+\end{itemize}
+Then at least $\hat{N}$ partitions must be spread among the remaining zones. Hence $s^*$ is upper bounded by $\hat{C}/\hat{N}$ and without loss of generality, we can define, for every node that is not in $z_0$ nor $z_1$, 
+$$\hat{n}_v = \left\lfloor\frac{c_v}{\hat{C}}\hat{N}\right\rfloor.$$ 
+
+We constructed a sub-utilization $\hat{n}_v$. Now notice that $3N-\sum_v \hat{n}_v \le \# V$ where $\# V$ denotes the number of nodes. We can iteratively pick a node $v^*$ such that 
+\begin{itemize}
+	\item $\sum_{v\in z_{v^*}} \hat{n}_v < N$ where $z_{v^*}$ is the zone of $v^*$;
+	\item $v^*$ maximizes the quantity $c_v/(\hat{n}_v+1)$ among the vertices satisfying the first condition (i.e. not in a saturated zone).
+\end{itemize}
+We iterate these instructions until $\sum_v \hat{n}_v= 3N$, and at this stage we define $(n_v) = (\hat{n}_v)$. It is easy to prove by induction that at every step, there is an optimal utilization that is pointwise larger than $\hat{n}_v$, and in particular, that $(n_v)$ is optimal.
+
+\subsubsection*{Existence of an optimal assignment}
+
+As for now, the \emph{optimal utilization} that we obtained is just a vector of numbers and it is not clear that it can be realized as the utilization of some concrete assignment. Here is a way to get a concrete assignment.
+
+Define $3N$ tokens $t_1,\ldots, t_{3N}\in V$ as follows:
+\begin{itemize}
+	\item Enumerate the zones $z$ of $Z$ in any order;
+	\item enumerate the nodes $v$ of $z$ in any order;
+	\item repeat $n_v$ times the token $v$.
+\end{itemize} 
+Then for $1\le i \le N$, define the triplet $T_i$ to be 
+$(t_i, t_{i+N}, t_{i+2N})$. Since the same nodes of a zone appear contiguously, the three nodes of a triplet must belong to three distinct zones.
+
+However simple, this solution to go from an utilization to an assignment has the drawback of not spreading the triplets: a node will tend to be associated to the same two other nodes for many partitions. Hence, during data transfer, it will tend to use only two link, instead of spreading the bandwith use over many other links to other nodes. To achieve this goal, we will reframe the search of an assignment as a flow problem. and in the flow algorithm, we will introduce randomness in the order of exploration. This will be sufficient to obtain a good dispersion of the triplets.
+
+\begin{figure}
+	\centering
+	\includegraphics[width=0.9\linewidth]{figures/naive}
+	\caption{On the left, the creation of a concrete assignment with the naive approach of repeating tokens. On the right, the zones containing the nodes.}
+\end{figure}
+
+\subsubsection*{Assignment as a maximum flow problem}
+
+We describe the flow problem via its graph $(X,E)$ where $X$ is a set of vertices, and $E$ are directed weighted edges between the vertices. For every zone $z$, define $n_z=\sum_{v\in z} n_v$.
+
+The set of vertices $X$ contains the source $\mathbf{s}$ and the sink $\mathbf{t}$; a vertex $\mathbf{x}_z$ for every zone $z\in Z$, and a vertex $\mathbf{y}_i$ for every partition index $1\le i\le N$. 
+
+The set of edges $E$ contains
+\begin{itemize}
+	\item the edge $(\mathbf{s}, \mathbf{x}_z, n_z)$ for every zone $z\in Z$;
+	\item the edge $(\mathbf{x}_z, \mathbf{y}_i, 1)$ for every zone $z\in Z$ and partition $1\le i\le N$;
+	\item the edge $(\mathbf{y}_i, \mathbf{t}, 3)$ for every partition $1\le i\le N$.
+\end{itemize}
+
+\begin{figure}[b]
+	\centering
+	\includegraphics[width=0.6\linewidth]{figures/flow}
+	\caption{Flow problem to compute and optimal assignment.}
+\end{figure}
+
+We first show the equivalence between this problem and and the construction of an assignment. Given some optimal assignment $(n_v)$, define the flow $f:E\to \mathbb{N}$ that saturates every edge from $\mathbf{s}$ or to $\mathbf{t}$, takes value $1$ on the edge between $\mathbf{x}_z$ and $\mathbf{y}_i$ if partition $i$ is stored in some node of the zone $z$, and $0$ otherwise. One can easily check that $f$ thus defined is indeed a flow and is maximum.
+
+Reciprocally, by the existence of maximum flows constructed from optimal assignments, any maximum flow must saturate the edges linked to the source or the sink. It can only take value 0 or 1 on the other edge, and every partition vertex is associated to exactly three distinct zone vertices. Every zone is associated to exactly $n_z$ partitions.
+
+A maximum flow can be constructed using, for instance, Dinic's algorithm. This algorithm works by discovering augmenting path to iteratively increase the flow. During the exploration of the graph to find augmenting path, we can shuffle the order of enumeration of the neighbours to spread the associations between zones and partitions.
+
+Once we have such association, we can randomly distribute the $n_z$ edges picked for every zone $z$ to its nodes $v\in z$ such that every such $v$ gets $n_z$ edges. This defines an optimal assignment of partitions to nodes.
+
+
+\subsection{Minimal transfer}
+
+Assume that there was a previous assignment $(T'_i)_{1\le i\le N}$ corresponding to utilizations $(n'_v)_{v\in V}$. We would like the new computed assignment $(T_i)_{1\le i\le N}$ from some $(n_v)_{v\in V}$ to minimize the number of partitions that need to be transferred. We can imagine two different objectives corresponding to different hypotheses: 
+\begin{equation}
+	\tag{H3A}
+	\label{hyp:A}
+	\text{\emph{Transfers between different zones cost much more than inside a zone.}}
+\end{equation}
+\begin{equation}
+	\tag{H3B}
+	\label{hyp:B}
+	\text{\emph{Changing zone is not the largest cost when transferring a partition.}}
+\end{equation}
+
+In case $A$, our goal will be to minimize the number of changes of zone in the assignment of partitions to zone. More formally, we will maximize the quantity
+$$
+Q_Z := 
+\sum_{1\le i\le N} 
+\#\{z\in Z ~|~ z\cap T_i \neq \emptyset, z\cap T'_i \neq \emptyset \}
+.$$
+
+In case $B$, our goal will be to minimize the number of changes of nodes in the assignment of partitions to nodes. We will maximize the quantity
+$$
+Q_V :=
+\sum_{1\le i\le N} \#(T_i \cap T'_i).
+$$
+
+It is tempting to hope that there is a way to maximize both quantity, that having the least discrepancy in terms of nodes will lead to the least discrepancy in terms of zones. But this is actually wrong! We propose the following counter-example to convince the reader:
+
+We consider eight nodes $a, a', b, c, d, d', e, e'$ belonging to five different zones $\{a,a'\}, \{b\}, \{c\}, \{d,d'\}, \{e, e'\}$. We take three partitions ($N=3$), that are originally assigned with some utilization $(n'_v)_{v\in V}$ as follows:
+$$
+T'_1=(a,b,c) \qquad
+T'_2=(a',b,d) \qquad 
+T'_3=(b,c,e).
+$$
+This assignment, with updated utilizations $(n_v)_{v\in V}$ minimizes the number of zone changes:
+$$
+T_1=(d,b,c) \qquad
+T_2=(a,b,d) \qquad 
+T_3=(b,c,e').
+$$
+This one, with the same utilization, minimizes the number of node changes:
+$$
+T_1=(a,b,c) \qquad
+T_2=(e',b,d) \qquad 
+T_3=(b,c,d').
+$$
+One can check that in this case, it is impossible to minimize both the number of zone and node changes.
+
+Because of the redundancy constraint, we cannot use a greedy algorithm to just replace nodes in the triplets to try to get the new utilization rate: this could lead to blocking situation where there is still a hole to fill in a triplet but no available node satisfies the zone separation constraint. To circumvent this issue, we propose an algorithm based on finding cycles in a graph encoding of the assignment. As in section \ref{sec:opt_assign}, we can explore the neigbours in a random order in the graph algorithms, to spread the triplets distribution.
+
+
+\subsubsection{Minimizing the zone discrepancy}
+
+
+First, notice that, given an assignment of partitions to \emph{zones}, it is easy to deduce an assignment to \emph{nodes} that minimizes the number of transfers for this zone assignment: For every zone $z$ and every node $v\in z$, pick in any way a set $P_v$ of partitions that where assigned to $v$ in $T'$, to $z_v$ in $T$, with the cardinality of $P_v$ smaller than $n_v$. Once all these sets are chosen, complement the assignment to reach the right utilization for every node. If $\#P_v > n_v$, it means that all the partitions that could stay in $v$ (i.e. that were already in $v$ and are still assigned to its zone) do stay in $v$. If $\#P_v = n_v$, then $n_v$ partitions stay in $v$, which is the number of partitions that need to be in $v$ in the end. In both cases, we could not hope for better given the partition to zone assignment.
+
+Our goal now is to find a assignment of partitions to zones that minimizes the number of zone transfers. To do so we are going to represent an assignment as a graph.
+
+Let $G_T=(X,E_T)$ be the directed weighted graph with vertices $(\mathbf{x}_i)_{1\le i\le N}$ and $(\mathbf{y}_z)_{z\in Z}$. For any $1\le i\le N$ and $z\in Z$, $E_T$ contains  the arc:
+\begin{itemize}
+	\item $(\mathbf{x}_i, \mathbf{y}_z, +1)$, if $z$ appears in $T_i'$ and $T_i$;
+	\item $(\mathbf{x}_i, \mathbf{y}_z, -1)$, if $z$ appears in $T_i$ but not in $T'_i$;
+	\item $(\mathbf{y}_z, \mathbf{x}_i, -1)$, if $z$ appears in $T'_i$ but not in $T_i$;
+	\item $(\mathbf{y}_z, \mathbf{x}_i, +1)$, if $z$ does not appear in $T'_i$ nor in $T_i$.
+\end{itemize}
+In other words, the orientation of the arc encodes whether partition $i$ is stored in zone $z$ in the assignment $T$ and the weight $\pm 1$ encodes whether this corresponds to what happens in the assignment $T'$.
+
+\begin{figure}[t]
+	\centering
+	\begin{minipage}{.40\linewidth}
+		\centering
+		\includegraphics[width=.8\linewidth]{figures/mini_zone}
+	\end{minipage}
+	\begin{minipage}{.55\linewidth}
+		\centering
+		\includegraphics[width=.8\linewidth]{figures/mini_node}
+	\end{minipage}
+	\caption{On the left: the graph $G_T$ encoding an assignment to minimize the zone discrepancy. On the right: the graph $G_T$ encoding an assignment to minimize the node discrepancy.}
+\end{figure}
+
+
+Notice that at every partition, there are three outgoing arcs, and at every zone, there are $n_z$ incoming arcs. Moreover, if $w(e)$ is the weight of an arc $e$, define the weight of $G_T$ by
+\begin{align*}
+w(G_T) := \sum_{e\in E} w(e) &= \#Z \times N - 4 \sum_{1\le i\le N} \#\{z\in Z ~|~ z\cap T_i = \emptyset, z\cap T'_i \neq \emptyset\}	\\
+&=\#Z \times N - 4 \sum_{1\le i\le N} 3- \#\{z\in Z ~|~ z\cap T_i \neq \emptyset, z\cap T'_i \neq \emptyset\} \\
+&= (\#Z-12)N + 4 Q_Z.
+\end{align*}
+Hence maximizing $Q_Z$ is equivalent to maximizing $w(G_T)$.
+
+Assume that their exist some assignment $T^*$ with the same utilization $(n_v)_{v\in V}$. Define $G_{T^*}$ similarly and consider the set $E_\mathrm{Diff} = E_T \setminus E_{T^*}$ of arcs that appear only in $G_T$. Since all vertices have the same number of incoming arcs in $G_T$ and $G_{T^*}$, the vertices of the graph $(X, E_\mathrm{Diff})$ must all have the same number number of incoming and outgoing arrows. So $E_\mathrm{Diff}$ can be expressed as a union of disjoint cycles. Moreover, the edges of $E_\mathrm{Diff}$ must appear in $E_{T^*}$ with reversed orientation and opposite weight. Hence, we have
+$$
+ w(G_T) - w(G_{T^*}) = 2 \sum_{e\in E_\mathrm{Diff}} w(e). 
+$$
+Hence, if $T$ is not optimal, there exists some $T^*$ with $w(G_T) < w(G_{T^*})$, and by the considerations above, there must exist a cycle in $E_\mathrm{Diff}$, and hence in $G_T$, with negative weight. If we reverse the edges and weights along this cycle, we obtain some graph. Since we did not change the incoming degree of any vertex, this is the graph encoding of some valid assignment $T^+$ such that $w(G_{T^+}) > w(G_T)$. We can iterate this operation until there is no other assignment $T^*$ with larger weight, that is until we obtain an optimal assignment.
+
+
+
+\subsubsection{Minimizing the node discrepancy}
+
+We will follow an approach similar to the one where we minimize the zone discrepancy. Here we will directly obtain a node assignment from a graph encoding.
+
+Let $G_T=(X,E_T)$ be the directed weighted graph with vertices $(\mathbf{x}_i)_{1\le i\le N}$, $(\mathbf{y}_{z,i})_{z\in Z, 1\le i\le N}$ and $(\mathbf{u}_v)_{v\in V}$. For any $1\le i\le N$ and $z\in Z$, $E_T$ contains  the arc:
+\begin{itemize}
+	\item $(\mathbf{x}_i, \mathbf{y}_{z,i}, 0)$, if $z$ appears in $T_i$;
+	\item $(\mathbf{y}_{z,i}, \mathbf{x}_i, 0)$, if $z$ does not appear in $T_i$.
+\end{itemize}
+For any $1\le i\le N$ and $v\in V$, $E_T$ contains  the arc:
+\begin{itemize}
+	\item $(\mathbf{y}_{z_v,i}, \mathbf{u}_v, +1)$, if $v$ appears in $T_i'$ and $T_i$;
+	\item $(\mathbf{y}_{z_v,i}, \mathbf{u}_v, -1)$, if $v$ appears in $T_i$ but not in $T'_i$;
+	\item $(\mathbf{u}_v, \mathbf{y}_{z_v,i}, -1)$, if $v$ appears in $T'_i$ but not in $T_i$;
+	\item $(\mathbf{u}_v, \mathbf{y}_{z_v,i}, +1)$, if $v$ does not appear in $T'_i$ nor in $T_i$.
+\end{itemize}
+Every vertex $\mathbb{x}_i$ has outgoing degree 3, every vertex $\mathbf{y}_{z,v}$ has outgoing degree 1, and every vertex $\mathbf{u}_v$ has incoming degree $n_v$. 
+Remark that any graph respecting these degree constraints is the encoding of a valid assignment with utilizations $(n_v)_{v\in V}$, in particular no partition is stored in two nodes of the same zone.
+
+We define $w(G_T)$ similarly:
+\begin{align*}
+	w(G_T) := \sum_{e\in E_T} w(e) &= \#V \times N - 4\sum_{1\le i\le N} 3-\#(T_i\cap T'_i) \\
+	&= (\#V-12)N + 4Q_V.
+\end{align*}
+
+Exactly like in the previous section, the existence of an assignment with larger weight implies the existence of a negatively weighted cycle in $G_T$. Reversing this cycle gives us the encoding of a valid assignment with a larger weight. Iterating this operation yields an optimal assignment.
+
+
+\subsubsection{Linear combination of both criteria}
+
+In the graph $G_T$ defined in the previous section, instead of having weights $0$ and $\pm 1$, we could be having weights $\pm\alpha$ between $\mathbf{x}$ and  $\mathbf{y}$ vertices, and weights $\pm\beta$ between $\mathbf{y}$ and  $\mathbf{u}$ vertices, for some $\alpha,\beta>0$ (we have positive weight if the assignment corresponds to $T'$ and negative otherwise). Then
+\begin{align*}
+	w(G_T) &= \sum_{e\in E_T} w(e) = 
+	\alpha \big( (\#Z-12)N + 4 Q_Z\big) + 
+	\beta \big( (\#V-12)N + 4 Q_V\big) \\
+	&= \mathrm{const}+ 4(\alpha Q_Z + \beta Q_V).
+\end{align*}
+So maximizing the weight of such graph encoding would be equivalent to maximizing a linear combination of $Q_Z$ and $Q_V$.
+
+
+\subsection{Algorithm}
+We give a high level description of the algorithm to compute an optimal 3-strict assignment. The operations appearing at lines 1,2,4 are respectively described by Algorithms \ref{alg:util},\ref{alg:opt} and \ref{alg:mini}.
+
+
+
+\begin{algorithm}[H]
+	\caption{Optimal 3-strict assignment}
+	\label{alg:total}
+	\begin{algorithmic}[1]
+	\Function{Optimal 3-strict assignment}{$N$, $(c_v)_{v\in V}$, $T'$}
+		\State $(n_v)_{v\in V} \leftarrow$ \Call{Compute optimal utilization}{$N$, $(c_v)_{v\in V}$}
+		\State $(T_i)_{1\le i\le N} \leftarrow$ \Call{Compute candidate assignment}{$N$, $(n_v)_{v\in V}$}
+		\If {there was a previous assignment $T'$}
+			\State $T \leftarrow$ \Call{Minimization of transfers}{$(T_i)_{1\le i\le N}$, $(T'_i)_{1\le i\le N}$}
+		\EndIf
+		\State \Return $T$.
+	\EndFunction
+	\end{algorithmic}
+\end{algorithm}
+
+We give some considerations of worst case complexity for these algorithms. In the following, we assume $N>\#V>\#Z$. The complexity of Algorithm \ref{alg:total} is $O(N^3\# Z)$ if we assume \eqref{hyp:A} and $O(N^3 \#Z \#V)$ if we assume \eqref{hyp:B}.
+
+Algorithm \ref{alg:util} can be implemented with complexity $O(\#V^2)$. The complexity of the function call at line \ref{lin:subutil} is $O(\#V)$. The difference between the sum of the subutilizations and $3N$ is at most the sum of the rounding errors when computing the $\hat{n}_v$. Hence it is bounded by $\#V$ and the loop at line \ref{lin:loopsub} is iterated at most $\#V$ times. Finding the minimizing $v$ at line \ref{lin:findmin} takes $O(\#V)$ operations (naively, we could also use a heap).
+
+Algorithm \ref{alg:opt} can be implemented with complexity $O(N^3\times \#Z)$. The flow graph has $O(N+\#Z)$ vertices and $O(N\times \#Z)$ edges. Dinic's algorithm has complexity $O(\#\mathrm{Vertices}^2\#\mathrm{Edges})$ hence in our case it is $O(N^3\times \#Z)$.
+ 
+Algorithm \ref{alg:mini} can be implented with complexity $O(N^3\# Z)$ under \eqref{hyp:A} and $O(N^3 \#Z \#V)$ under \eqref{hyp:B}.
+The graph $G_T$ has $O(N)$ vertices and $O(N\times \#Z)$ edges under assumption \eqref{hyp:A} and respectively $O(N\times \#Z)$ vertices and $O(N\times \#V)$ edges under assumption \eqref{hyp:B}. The loop at line \ref{lin:repeat} is iterated at most $N$ times since the distance between $T$ and $T'$ decreases at every iteration. Bellman-Ford algorithm has complexity $O(\#\mathrm{Vertices}\#\mathrm{Edges})$, which in our case amounts to $O(N^2\# Z)$ under \eqref{hyp:A} and $O(N^2 \#Z \#V)$ under \eqref{hyp:B}.
+ 
+\begin{algorithm}
+	\caption{Computation of the optimal utilization}
+	\label{alg:util}
+	\begin{algorithmic}[1]
+\Function{Compute optimal utilization}{$N$, $(c_v)_{v\in V}$}
+	\State $(\hat{n}_v)_{v\in V} \leftarrow $ \Call{Compute subutilization}{$N$, $(c_v)_{v\in V}$} \label{lin:subutil}
+	\While{$\sum_{v\in V} \hat{n}_v < 3N$} \label{lin:loopsub}
+	\State Pick $v\in V$ minimizing $\frac{c_v}{\hat{n}_v+1}$ and such that
+		$\sum_{v'\in z_v} \hat{n}_{v'} < N$ \label{lin:findmin}
+		\State $\hat{n}_v \leftarrow \hat{n}_v+1$
+	\EndWhile
+	\State \Return $(\hat{n}_v)_{v\in V}$
+\EndFunction
+\State
+
+\Function{Compute subutilization}{$N$, $(c_v)_{v\in V}$}
+		\State $R \leftarrow 3$
+\For{$v\in V$}
+\State $\hat{n}_v \leftarrow \mathrm{unset}$
+\EndFor
+\For{$z\in Z$}
+\State $c_z \leftarrow \sum_{v\in z} c_v$
+\EndFor
+\State $C \leftarrow \sum_{z\in Z} c_z$
+\While{$\exists z \in Z$ such that $R\times c_{z} > C$}
+\For{$v\in z$}
+\State $\hat{n}_v \leftarrow \left\lfloor \frac{c_v}{c_z} N \right\rfloor$
+\EndFor
+\State $C \leftarrow C-c_z$
+\State $R\leftarrow R-1$
+\EndWhile
+\For{$v\in V$}
+\If{$\hat{n}_v = \mathrm{unset}$}
+\State $\hat{n}_v \leftarrow \left\lfloor \frac{Rc_v}{C} N \right\rfloor$
+\EndIf
+\EndFor
+\State \Return $(\hat{n}_v)_{v\in V}$
+\EndFunction
+	\end{algorithmic}
+\end{algorithm}
+
+\begin{algorithm}
+	\caption{Computation of a candidate assignment}
+	\label{alg:opt}
+	\begin{algorithmic}[1]
+		\Function{Compute candidate assignment}{$N$, $(n_v)_{v\in V}$}
+		\State Compute the flow graph $G$
+		\State Compute the maximal flow $f$ using Dinic's algorithm with randomized neighbours enumeration
+		\State Construct the assignment $(T_i)_{1\le i\le N}$ from $f$
+		\State \Return $(T_i)_{1\le i\le N}$
+		\EndFunction
+	\end{algorithmic}
+\end{algorithm}
+
+
+\begin{algorithm}
+	\caption{Minimization of the number of transfers}
+	\label{alg:mini}
+	\begin{algorithmic}[1]
+		\Function{Minimization of transfers}{$(T_i)_{1\le i\le N}$, $(T'_i)_{1\le i\le N}$}
+		\State Construct the graph encoding $G_T$
+		\Repeat \label{lin:repeat}
+		\State Find a negative cycle $\gamma$ using Bellman-Ford algorithm on $G_T$
+		\State Reverse the orientations and weights of edges in $\gamma$
+		\Until{no negative cycle is found}
+		\State Update $(T_i)_{1\le i\le N}$ from $G_T$
+		\State \Return $(T_i)_{1\le i\le N}$
+		\EndFunction
+	\end{algorithmic}
+\end{algorithm}
+
+\newpage
+
+\section{Computation of a 3-non-strict assignment}
+
+\subsection{Choices of optimality}
+
+In this mode, we primarily want to store every partition on three nodes, and only secondarily try to spread the nodes among different zone. So we make the choice of not taking the zone repartition in the criterion of optimality.
+
+We try to maximize $s^*$ defined in \eqref{eq:optimal}. So we can compute the optimal utilizations $(n_v)_{v\in V}$ with the only constraint that $n_v \le N$ for every node $v$. As in the previous section, we start with a sub-utilization proportional to $c_v$ (and capped at $N$), and we iteratively increase the $\hat{n}_v$ that is less than $N$ and maximizes the quantity $c_v/(\hat{n}_v+1)$, until the total sum is $3N$.
+
+\subsection{Computation of a candidate assignment}
+
+To compute a candidate assignment (that does not optimize zone spreading nor distance to a previous assignment yet), we can use the folowing flow problem.
+
+Define the oriented weighted graph $(X,E)$. The set of vertices $X$ contains the source $\mathbf{s}$, the sink $\mathbf{t}$, vertices 
+$\mathbf{x}_p, \mathbf{u}^+_p, \mathbf{u}^-_p$ for every partition $p$, vertices $\mathbf{y}_{p,z}$ for every partition $p$ and zone $z$, and vertices $\mathbf{z}_v$ for every node $v$. 
+
+The set of edges is composed of the following arcs:
+\begin{itemize}
+	\item ($\mathbf{s}$,$\mathbf{x}_p$, 3) for every partition $p$;
+	\item ($\mathbf{x}_p$,$\mathbf{u}^+_p$, 3) for every partition $p$;
+	\item ($\mathbf{x}_p$,$\mathbf{u}^-_p$, 2) for every partition $p$;
+	\item ($\mathbf{u}^+_p$,$\mathbf{y}_{p,z}$, 1) for every partition $p$ and zone $z$;
+	\item ($\mathbf{u}^-_p$,$\mathbf{y}_{p,z}$, 2) for every partition $p$ and zone $z$;
+	\item ($\mathbf{y}_{p,z}$,$\mathbf{z}_v$, 1) for every partition $p$, zone $z$ and node $v\in z$;
+	\item ($\mathbf{z}_v$, $\mathbf{t}$, $n_v$) for every node $v$;
+\end{itemize}
+
+One can check that any maximal flow in this graph corresponds to an assignment of partitions to nodes. In such a flow, all the arcs from $\mathbf{s}$ and to $\mathbf{t}$ are saturated. The arc from $\mathbf{y}_{p,z}$ to $\mathbf{z}_v$ is saturated if and only if $p$ is associated to~$v$. 
+Finally the flow from $\mathbf{x}_p$ to $\mathbf{y}_{p,z}$ can go either through $\mathbf{u}^+_p$ or $\mathbf{u}^-_p$.
+
+
+
+\subsection{Maximal spread and minimal transfers}
+Notice that if the arc $\mathbf{u}_p^+\mathbf{y}_{p,z}$ is not saturated but there is some flow in $\mathbf{u}_p^-\mathbf{y}_{p,z}$, then it is possible to transfer a unit of flow from the path  $\mathbf{x}_p\mathbf{u}_p^-\mathbf{y}_{p,z}$ to the path $\mathbf{x}_p\mathbf{u}_p^+\mathbf{y}_{p,z}$. So we can always find an equivalent maximal flow $f^*$ that uses the path through $\mathbf{u}_p^-$ only if the path through $\mathbf{u}_p^+$ is saturated.
+
+We will use this fact to consider the amount of flow going through the vertices $\mathbf{u}^+$ as a measure of how well the partitions are spread over nodes belonging to different zones. If the partition $p$ is associated to 3 different zones, then a flow of 3 will cross $\mathbf{u}_p^+$ in $f^*$ (i.e. a flow of 0 will cross $\mathbf{u}_p^+$). If $p$ is associated to two zones, a flow of $2$ will cross $\mathbf{u}_p^+$. If $p$ is associated to a single zone, a flow of $1$ will cross $\mathbf{u}_p^+$.
+
+Let $N_1, N_2, N_3$ be the number of partitions associated to respectively 1,2 and 3 distinct zones. We will optimize a linear combination of these variables using the discovery of positively weighted circuits in a graph.
+
+At the same step, we will also optimize the distance to a previous assignment $T'$. Let $\alpha> \beta> \gamma \ge 0$ be three parameters.
+
+Given the flow $f$, let $G_f=(X',E_f)$ be the multi-graph where $X' = X\setminus\{\mathbf{s},\mathbf{t}\}$. The set $E_f$ is composed of the arcs:
+\begin{itemize}
+\item As many arcs from $(\mathbf{x}_p, \mathbf{u}^+_p,\alpha), (\mathbf{x}_p, \mathbf{u}^+_p,\beta), (\mathbf{x}_p, \mathbf{u}^+_p,\gamma)$ (selected in this order) as there is flow crossing $\mathbf{u}^+_p$ in $f$;
+\item As many arcs from $(\mathbf{u}^+_p, \mathbf{x}_p,-\gamma), (\mathbf{u}^+_p, \mathbf{x}_p,-\beta), (\mathbf{u}^+_p, \mathbf{x}_p,-\alpha)$ (selected in this order) as there is flow crossing $\mathbf{u}^-_p$ in $f$;
+\item  As many copies of $(\mathbf{x}_p, \mathbf{u}^-_p,0)$ as there is flow through $\mathbf{u}^-_p$;
+\item  As many copies of $(\mathbf{u}^-_p,\mathbf{x}_p,0)$ so that the number of arcs between these two vertices is 2;
+\item $(\mathbf{u}^+_p,\mathbf{y}_{p,z}, 0)$ if the flow between these vertices is 1, and the opposite arc otherwise;
+\item as many copies of $(\mathbf{u}^-_p,\mathbf{y}_{p,z}, 0)$ as the flow between these vertices, and as many copies of the opposite arc as 2~$-$~the flow;
+\item $(\mathbf{y}_{p,z},\mathbf{z}_v, \pm1)$ if it is saturated in $f$, with $+1$ if $v\in T'_p$ and $-1$ otherwise;
+\item $(\mathbf{z}_v,\mathbf{y}_{p,z}, \pm1)$ if it is not saturated in $f$, with $+1$ if $v\notin T'_p$ and $-1$ otherwise.
+\end{itemize}
+To summarize, arcs are oriented left to right if they correspond to a presence of flow in $f$, and right to left if they correspond to an absence of flow. They are positively weighted if we want them to stay at their current state, and negatively if we want them to switch. Let us compute the weight of such graph.
+
+\begin{multline*}
+	w(G_f)  = \sum_{e\in E_f} w(e_f)  \\
+	= 
+	(\alpha - \beta -\gamma) N_1 + (\alpha +\beta - \gamma) N_2 + (\alpha+\beta+\gamma) N_3 
+	\\ + 
+	\#V\times N - 4 \sum_p 3-\#(T_p\cap T'_p) \\ 
+	=(\#V-12+\alpha-\beta-\gamma)\times N + 4Q_V + 2\beta N_2 + 2(\beta+\gamma) N_3 \\
+\end{multline*}
+
+As for the mode 3-strict, one can check that the difference of two such graphs corresponding to the same $(n_v)$ is always eulerian. Hence we can navigate in this class with the same greedy algorithm that discovers positive cycles and flips them.
+
+The function that we optimize is 
+$$
+2Q_V +  \beta N_2 + (\beta+\gamma) N_3.
+$$
+The choice of parameters $\beta$ and $\gamma$ should be lead by the following question: For $\beta$, where to put the tradeoff between zone dispersion and distance to the previous configuration? For $\gamma$, do we prefer to have more partitions spread between 2 zones, or have less between at least 2 zones but more between 3 zones.
+
+The quantity $Q_V$ varies between $0$ and $3N$, it should be of order $N$. The quantity $N_2+N_3$ should also be of order $N$ (it is exactly $N$ in the strict mode). So the two terms of the function are comparable. 
+
+
+\bibliography{optimal_layout} 
+\bibliographystyle{ieeetr}
+
+\end{document}
+
+
+
@@ -189,7 +189,7 @@ let
  rootFeatures = if features != null then
    features
  else
-    ([ "garage/bundled-libs" "garage/sled" "garage/k2v" ] ++ (if release then [
+    ([ "garage/bundled-libs" "garage/sled" "garage/lmdb" "garage/k2v" ] ++ (if release then [
      "garage/consul-discovery"
      "garage/kubernetes-discovery"
      "garage/metrics"
@@ -10,7 +10,7 @@ NIX_RELEASE="${REPO_FOLDER}/result/bin/"
 PATH="${GARAGE_DEBUG}:${GARAGE_RELEASE}:${NIX_RELEASE}:$PATH"

 garage -c /tmp/config.1.toml bucket create eprouvette
-KEY_INFO=$(garage -c /tmp/config.1.toml key new --name opérateur)
+KEY_INFO=$(garage -c /tmp/config.1.toml key create opérateur)
 ACCESS_KEY=`echo $KEY_INFO|grep -Po 'GK[a-f0-9]+'`
 SECRET_KEY=`echo $KEY_INFO|grep -Po 'Secret key: [a-f0-9]+'|grep -Po '[a-f0-9]+$'`
 garage -c /tmp/config.1.toml bucket allow eprouvette --read --write --owner --key $ACCESS_KEY
@@ -11,7 +11,7 @@ PATH="${GARAGE_DEBUG}:${GARAGE_RELEASE}:${NIX_RELEASE}:$PATH"
 FANCYCOLORS=("41m" "42m" "44m" "45m" "100m" "104m")

 export RUST_BACKTRACE=1 
-export RUST_LOG=garage=info,garage_api=debug,netapp=trace
+export RUST_LOG=garage=info,garage_api=debug
 MAIN_LABEL="\e[${FANCYCOLORS[0]}[main]\e[49m"

 WHICH_GARAGE=$(which garage || exit 1)
@@ -25,7 +25,8 @@ garage -c /tmp/config.1.toml status \
 	| grep 'NO ROLE' \
 	| grep -Po '^[0-9a-f]+' \
 	| while read id; do 
-	  garage -c /tmp/config.1.toml layout assign $id -z dc1 -c 1
+	  garage -c /tmp/config.1.toml layout assign $id -z dc1 -c 1G
 	done

+garage -c /tmp/config.1.toml layout config -r 1
 garage -c /tmp/config.1.toml layout apply --version 1
@@ -0,0 +1,138 @@
+#!/usr/bin/env bash
+
+: '
+	This script tests part renumbering on an S3 remote (here configured for Minio).
+
+	On Minio:
+
+	The results confirm that if I upload parts with number 1, 4, 5 and 6,
+	they are renumbered to 1, 2, 3 and 4 after CompleteMultipartUpload.
+	Thus, specifying partNumber=4 on a GetObject/HeadObject should return
+	information on the part I originally uploaded with part number
+
+	On S3: not tested
+
+	Sample output (on Minio):
+
+		f07e1404cc527d494242824ded3a616b  part1
+		78974cd4d0f622eb3426ea7cd22f5a1c  part4
+		f9cc379f8baa61645558d9ba7e6351fa  part5
+		1bd2383eebbac1f8e7143575ba5b1f4a  part6
+		Upload ID: 6838b813-d0ca-400b-9d28-ec8b2b5cd004
+		PART 1 ETag: "f07e1404cc527d494242824ded3a616b"
+		PART 4 ETag: "78974cd4d0f622eb3426ea7cd22f5a1c"
+		PART 5 ETag: "f9cc379f8baa61645558d9ba7e6351fa"
+		PART 6 ETag: "1bd2383eebbac1f8e7143575ba5b1f4a"
+		======================================== LIST ====
+		{
+		  "Parts": [
+			{
+			  "PartNumber": 1,
+			  "LastModified": "2023-04-25T10:21:54.350000+00:00",
+			  "ETag": "\"f07e1404cc527d494242824ded3a616b\"",
+			  "Size": 20971520
+			},
+			{
+			  "PartNumber": 4,
+			  "LastModified": "2023-04-25T10:21:54.350000+00:00",
+			  "ETag": "\"78974cd4d0f622eb3426ea7cd22f5a1c\"",
+			  "Size": 20971520
+			},
+			{
+			  "PartNumber": 5,
+			  "LastModified": "2023-04-25T10:21:54.350000+00:00",
+			  "ETag": "\"f9cc379f8baa61645558d9ba7e6351fa\"",
+			  "Size": 20971520
+			},
+			{
+			  "PartNumber": 6,
+			  "LastModified": "2023-04-25T10:21:54.350000+00:00",
+			  "ETag": "\"1bd2383eebbac1f8e7143575ba5b1f4a\"",
+			  "Size": 20971520
+			}
+		  ],
+		  "ChecksumAlgorithm": "",
+		  "Initiator": {
+			"ID": "02d6176db174dc93cb1b899f7c6078f08654445fe8cf1b6ce98d8855f66bdbf4",
+			"DisplayName": "02d6176db174dc93cb1b899f7c6078f08654445fe8cf1b6ce98d8855f66bdbf4"
+		  },
+		  "Owner": {
+			"DisplayName": "02d6176db174dc93cb1b899f7c6078f08654445fe8cf1b6ce98d8855f66bdbf4",
+			"ID": "02d6176db174dc93cb1b899f7c6078f08654445fe8cf1b6ce98d8855f66bdbf4"
+		  },
+		  "StorageClass": "STANDARD"
+		}
+		======================================== COMPLETE ====
+		{
+			"Location": "http://localhost:9000/test/upload",
+			"Bucket": "test",
+			"Key": "upload",
+			"ETag": "\"8e817c8ccd442f9a79c77b58fe808c43-4\""
+		}
+		======================================== LIST ====
+
+		An error occurred (NoSuchUpload) when calling the ListParts operation: The specified multipart upload does not exist. The upload ID may be invalid, or the upload may have been aborted or completed.
+		======================================== GET PART 4 ====
+		{
+			"AcceptRanges": "bytes",
+			"LastModified": "2023-04-25T10:21:59+00:00",
+			"ContentLength": 20971520,
+			"ETag": "\"8e817c8ccd442f9a79c77b58fe808c43-4\"",
+			"ContentRange": "bytes 62914560-83886079/83886080",
+			"ContentType": "binary/octet-stream",
+			"Metadata": {},
+			"PartsCount": 4
+		}
+		1bd2383eebbac1f8e7143575ba5b1f4a  get-part4
+
+	
+	Conclusions:
+
+	- Parts are indeed renumbered with consecutive numbers
+	- ListParts only applies to multipart uploads in progress,
+	  it cannot be used once the multipart upload has been completed
+'
+
+export AWS_ACCESS_KEY_ID=1D8Pk2k4oQSoh1BU
+export AWS_SECRET_ACCESS_KEY=4B46SR8U7FUgY0raB8Zuxg1NLyLTvbNV
+
+function aws { command aws --endpoint-url http://localhost:9000 $@ ; }
+
+aws --version
+
+aws s3 mb s3://test
+
+for NUM in 1 4 5 6; do
+	dd if=/dev/urandom of=part$NUM bs=1M count=10
+done
+md5sum part*
+
+UPLOAD=$(aws s3api create-multipart-upload --bucket test --key 'upload' | jq -r ".UploadId")
+echo "Upload ID: $UPLOAD"
+
+PARTS=""
+
+for NUM in 1 4 5 6; do
+	ETAG=$(aws s3api upload-part --bucket test --key 'upload' --part-number $NUM \
+		--body "part$NUM" --upload-id "$UPLOAD" | jq -r ".ETag")
+	echo "PART $NUM ETag: $ETAG"
+	if [ -n "$PARTS" ]; then
+		PARTS="$PARTS,"
+	fi
+	PARTS="$PARTS {\"ETag\":$ETAG,\"PartNumber\":$NUM}"
+done
+
+echo "======================================== LIST ===="
+aws s3api list-parts --bucket test --key upload --upload-id "$UPLOAD" | jq
+
+echo "======================================== COMPLETE ===="
+echo "{\"Parts\":[$PARTS]}" > mpu
+aws s3api complete-multipart-upload --multipart-upload file://mpu \
+		--bucket test --key 'upload' --upload-id "$UPLOAD"
+
+echo "======================================== LIST ===="
+aws s3api list-parts --bucket test --key upload --upload-id "$UPLOAD" | jq
+
+echo "======================================== GET PART 4 ===="
+aws s3api get-object --bucket test --key upload --part-number 4 get-part4
+md5sum get-part4
@@ -0,0 +1,103 @@
+#!/usr/bin/env bash
+
+: '
+	This script tests whether uploaded parts can be skipped in a
+	CompleteMultipartUpoad
+
+	On Minio: yes, parts can be skipped
+
+	On S3: not tested
+
+	Sample output (on Minio):
+
+		f23911bcd1230f5ebe8887cbf5bc396e  part1
+		a2657143167eaf647c40473e78a091dc  part4
+		72f72c02c5163bc81024b28ac818c5e0  part5
+		e29cf500d20498218904b8df8806caa2  part6
+		Upload ID: e8fe7b83-9800-46fb-ae90-9d7ccd42fe76
+		PART 1 ETag: "f23911bcd1230f5ebe8887cbf5bc396e"
+		PART 4 ETag: "a2657143167eaf647c40473e78a091dc"
+		PART 5 ETag: "72f72c02c5163bc81024b28ac818c5e0"
+		PART 6 ETag: "e29cf500d20498218904b8df8806caa2"
+		======================================== COMPLETE ====
+		{
+			"Location": "http://localhost:9000/test/upload",
+			"Bucket": "test",
+			"Key": "upload",
+			"ETag": "\"48246e44d4b38bdc2f3c10ee25b1af17-3\""
+		}
+		======================================== GET FULL ====
+		{
+			"AcceptRanges": "bytes",
+			"LastModified": "2023-04-25T10:54:35+00:00",
+			"ContentLength": 31457280,
+			"ETag": "\"48246e44d4b38bdc2f3c10ee25b1af17-3\"",
+			"ContentType": "binary/octet-stream",
+			"Metadata": {}
+		}
+		97fb904da7ad310699a6afab0eb6e061  get-full
+		97fb904da7ad310699a6afab0eb6e061  -
+		======================================== GET PART 3 ====
+		{
+			"AcceptRanges": "bytes",
+			"LastModified": "2023-04-25T10:54:35+00:00",
+			"ContentLength": 10485760,
+			"ETag": "\"48246e44d4b38bdc2f3c10ee25b1af17-3\"",
+			"ContentRange": "bytes 20971520-31457279/31457280",
+			"ContentType": "binary/octet-stream",
+			"Metadata": {},
+			"PartsCount": 3
+		}
+		e29cf500d20498218904b8df8806caa2  get-part3
+
+	Conclusions:
+
+	- Skipping a part in a CompleteMultipartUpoad call is OK
+	- The part is simply not included in the stored object
+	- Sequential part renumbering counts only non-skipped parts
+'
+
+export AWS_ACCESS_KEY_ID=1D8Pk2k4oQSoh1BU
+export AWS_SECRET_ACCESS_KEY=4B46SR8U7FUgY0raB8Zuxg1NLyLTvbNV
+
+function aws { command aws --endpoint-url http://localhost:9000 $@ ; }
+
+aws --version
+
+aws s3 mb s3://test
+
+for NUM in 1 4 5 6; do
+	dd if=/dev/urandom of=part$NUM bs=1M count=10
+done
+md5sum part*
+
+UPLOAD=$(aws s3api create-multipart-upload --bucket test --key 'upload' | jq -r ".UploadId")
+echo "Upload ID: $UPLOAD"
+
+PARTS=""
+
+for NUM in 1 4 5 6; do
+	ETAG=$(aws s3api upload-part --bucket test --key 'upload' --part-number $NUM \
+		--body "part$NUM" --upload-id "$UPLOAD" | jq -r ".ETag")
+	echo "PART $NUM ETag: $ETAG"
+	if [ "$NUM" != "5" ]; then
+		if [ -n "$PARTS" ]; then
+			PARTS="$PARTS,"
+		fi
+		PARTS="$PARTS {\"ETag\":$ETAG,\"PartNumber\":$NUM}"
+	fi
+done
+
+echo "======================================== COMPLETE ===="
+echo "{\"Parts\":[$PARTS]}" > mpu
+aws s3api complete-multipart-upload --multipart-upload file://mpu \
+		--bucket test --key 'upload' --upload-id "$UPLOAD"
+
+echo "======================================== GET FULL ===="
+aws s3api get-object --bucket test --key upload get-full
+md5sum get-full
+cat part1 part4 part6 | md5sum
+
+echo "======================================== GET PART 3 ===="
+aws s3api get-object --bucket test --key upload --part-number 3 get-part3
+md5sum get-part3
@@ -31,6 +31,11 @@ dd if=/dev/urandom of=/tmp/garage.1.rnd bs=1k count=2 # No multipart, inline sto
 dd if=/dev/urandom of=/tmp/garage.2.rnd bs=1M count=5 # No multipart but file will be chunked
 dd if=/dev/urandom of=/tmp/garage.3.rnd bs=1M count=10 # by default, AWS starts using multipart at 8MB

+dd if=/dev/urandom of=/tmp/garage.part1.rnd bs=1M count=5
+dd if=/dev/urandom of=/tmp/garage.part2.rnd bs=1M count=5
+dd if=/dev/urandom of=/tmp/garage.part3.rnd bs=1M count=5
+dd if=/dev/urandom of=/tmp/garage.part4.rnd bs=1M count=5
+
 # data of lower entropy, to test compression
 dd if=/dev/urandom bs=1k count=2  | base64 -w0 > /tmp/garage.1.b64
 dd if=/dev/urandom bs=1M count=5  | base64 -w0 > /tmp/garage.2.b64
@@ -40,7 +45,7 @@ echo "🧪 S3 API testing..."

 # AWS
 if [ -z "$SKIP_AWS" ]; then
-  echo "🛠️ Testing with awscli"
+  echo "🛠️ Testing with awscli (aws s3)"
  source ${SCRIPT_FOLDER}/dev-env-aws.sh
  aws s3 ls
  for idx in {1..3}.{rnd,b64}; do
@@ -51,8 +56,36 @@ if [ -z "$SKIP_AWS" ]; then
    rm /tmp/garage.$idx.dl
    aws s3 rm "s3://eprouvette/&+-é\"/garage.$idx.aws"
  done
+
+  echo "🛠️ Testing multipart uploads with awscli (aws s3api)"
+  UPLOAD=$(aws s3api create-multipart-upload --bucket eprouvette --key 'upload' | jq -r ".UploadId")
+  echo "Upload ID: $UPLOAD"
+  ETAG3=$(aws s3api upload-part --bucket eprouvette --key 'upload' \
+      --part-number 3 --body "/tmp/garage.part1.rnd" --upload-id "$UPLOAD" \
+      | jq -r ".ETag")
+  ETAG2=$(aws s3api upload-part --bucket eprouvette --key 'upload' \
+      --part-number 2 --body "/tmp/garage.part2.rnd" --upload-id "$UPLOAD" \
+      | jq -r ".ETag")
+  ETAG3=$(aws s3api upload-part --bucket eprouvette --key 'upload' \
+      --part-number 3 --body "/tmp/garage.part3.rnd" --upload-id "$UPLOAD" \
+      | jq -r ".ETag")
+  ETAG6=$(aws s3api upload-part --bucket eprouvette --key 'upload' \
+      --part-number 6 --body "/tmp/garage.part4.rnd" --upload-id "$UPLOAD" \
+      | jq -r ".ETag")
+  MPU="{\"Parts\":[{\"PartNumber\":2,\"ETag\":$ETAG2}, {\"PartNumber\":3,\"ETag\":$ETAG3}, {\"PartNumber\":6,\"ETag\":$ETAG6}]}"
+  echo $MPU > /tmp/garage.mpu.json
+  aws s3api complete-multipart-upload --multipart-upload file:///tmp/garage.mpu.json \
+      --bucket eprouvette --key 'upload' --upload-id "$UPLOAD"
+  aws s3api get-object --bucket eprouvette --key upload /tmp/garage.mpu.get
+  if [ "$(md5sum /tmp/garage.mpu.get | cut -d ' ' -f 1)" != "$(cat /tmp/garage.part{2,3,4}.rnd | md5sum | cut -d ' ' -f 1)" ]; then
+    echo "Invalid multipart upload"
+    exit 1
+  fi
 fi

+echo "OK!!"
+exit 0
+
 # S3CMD
 if [ -z "$SKIP_S3CMD" ]; then
  echo "🛠️ Testing with s3cmd"
@@ -141,6 +174,7 @@ rm eprouvette/winscp
 EOF
 fi

+rm /tmp/garage.part{1..4}.rnd
 rm /tmp/garage.{1..3}.{rnd,b64}

 echo "🏁 Teardown"
@@ -276,8 +276,13 @@ impl ApiHandler for AdminApiServer {
 			Endpoint::RevertClusterLayout => handle_revert_cluster_layout(&self.garage, req).await,
 			// Keys
 			Endpoint::ListKeys => handle_list_keys(&self.garage).await,
-			Endpoint::GetKeyInfo { id, search } => {
-				handle_get_key_info(&self.garage, id, search).await
+			Endpoint::GetKeyInfo {
+				id,
+				search,
+				show_secret_key,
+			} => {
+				let show_secret_key = show_secret_key.map(|x| x == "true").unwrap_or(false);
+				handle_get_key_info(&self.garage, id, search, show_secret_key).await
 			}
 			Endpoint::CreateKey => handle_create_key(&self.garage, req).await,
 			Endpoint::ImportKey => handle_import_key(&self.garage, req).await,
@@ -14,6 +14,7 @@ use garage_model::bucket_alias_table::*;
 use garage_model::bucket_table::*;
 use garage_model::garage::Garage;
 use garage_model::permission::*;
+use garage_model::s3::mpu_table;
 use garage_model::s3::object_table::*;

 use crate::admin::error::*;
@@ -124,6 +125,14 @@ async fn bucket_info_results(
 		.map(|x| x.filtered_values(&garage.system.ring.borrow()))
 		.unwrap_or_default();

+	let mpu_counters = garage
+		.mpu_counter_table
+		.table
+		.get(&bucket_id, &EmptyKey)
+		.await?
+		.map(|x| x.filtered_values(&garage.system.ring.borrow()))
+		.unwrap_or_default();
+
 	let mut relevant_keys = HashMap::new();
 	for (k, _) in bucket
 		.state
@@ -208,12 +217,12 @@ async fn bucket_info_results(
 					}
 				})
 				.collect::<Vec<_>>(),
-			objects: counters.get(OBJECTS).cloned().unwrap_or_default(),
-			bytes: counters.get(BYTES).cloned().unwrap_or_default(),
-			unfinished_uploads: counters
-				.get(UNFINISHED_UPLOADS)
-				.cloned()
-				.unwrap_or_default(),
+			objects: *counters.get(OBJECTS).unwrap_or(&0),
+			bytes: *counters.get(BYTES).unwrap_or(&0),
+			unfinished_uploads: *counters.get(UNFINISHED_UPLOADS).unwrap_or(&0),
+			unfinished_multipart_uploads: *mpu_counters.get(mpu_table::UPLOADS).unwrap_or(&0),
+			unfinished_multipart_upload_parts: *mpu_counters.get(mpu_table::PARTS).unwrap_or(&0),
+			unfinished_multipart_upload_bytes: *mpu_counters.get(mpu_table::BYTES).unwrap_or(&0),
 			quotas: ApiBucketQuotas {
 				max_size: quotas.max_size,
 				max_objects: quotas.max_objects,
@@ -235,6 +244,9 @@ struct GetBucketInfoResult {
 	objects: i64,
 	bytes: i64,
 	unfinished_uploads: i64,
+	unfinished_multipart_uploads: i64,
+	unfinished_multipart_upload_parts: i64,
+	unfinished_multipart_upload_bytes: i64,
 	quotas: ApiBucketQuotas,
 }

@@ -1,14 +1,13 @@
-use std::collections::HashMap;
 use std::net::SocketAddr;
 use std::sync::Arc;

-use hyper::{Body, Request, Response, StatusCode};
+use hyper::{Body, Request, Response};
 use serde::{Deserialize, Serialize};

 use garage_util::crdt::*;
 use garage_util::data::*;

-use garage_rpc::layout::*;
+use garage_rpc::layout;

 use garage_model::garage::Garage;

@@ -26,26 +25,37 @@ pub async fn handle_get_cluster_status(garage: &Arc<Garage>) -> Result<Response<
 			.system
 			.get_known_nodes()
 			.into_iter()
-			.map(|i| {
-				(
-					hex::encode(i.id),
-					KnownNodeResp {
-						addr: i.addr,
-						is_up: i.is_up,
-						last_seen_secs_ago: i.last_seen_secs_ago,
-						hostname: i.status.hostname,
-					},
-				)
+			.map(|i| KnownNodeResp {
+				id: hex::encode(i.id),
+				addr: i.addr,
+				is_up: i.is_up,
+				last_seen_secs_ago: i.last_seen_secs_ago,
+				hostname: i.status.hostname,
 			})
 			.collect(),
-		layout: get_cluster_layout(garage),
+		layout: format_cluster_layout(&garage.system.get_cluster_layout()),
 	};

 	Ok(json_ok_response(&res)?)
 }

 pub async fn handle_get_cluster_health(garage: &Arc<Garage>) -> Result<Response<Body>, Error> {
+	use garage_rpc::system::ClusterHealthStatus;
 	let health = garage.system.health();
+	let health = ClusterHealth {
+		status: match health.status {
+			ClusterHealthStatus::Healthy => "healthy",
+			ClusterHealthStatus::Degraded => "degraded",
+			ClusterHealthStatus::Unavailable => "unavailable",
+		},
+		known_nodes: health.known_nodes,
+		connected_nodes: health.connected_nodes,
+		storage_nodes: health.storage_nodes,
+		storage_nodes_ok: health.storage_nodes_ok,
+		partitions: health.partitions,
+		partitions_quorum: health.partitions_quorum,
+		partitions_all_ok: health.partitions_all_ok,
+	};
 	Ok(json_ok_response(&health)?)
 }

@@ -74,33 +84,68 @@ pub async fn handle_connect_cluster_nodes(
 }

 pub async fn handle_get_cluster_layout(garage: &Arc<Garage>) -> Result<Response<Body>, Error> {
-	let res = get_cluster_layout(garage);
+	let res = format_cluster_layout(&garage.system.get_cluster_layout());

 	Ok(json_ok_response(&res)?)
 }

-fn get_cluster_layout(garage: &Arc<Garage>) -> GetClusterLayoutResponse {
-	let layout = garage.system.get_cluster_layout();
+fn format_cluster_layout(layout: &layout::ClusterLayout) -> GetClusterLayoutResponse {
+	let roles = layout
+		.roles
+		.items()
+		.iter()
+		.filter_map(|(k, _, v)| v.0.clone().map(|x| (k, x)))
+		.map(|(k, v)| NodeRoleResp {
+			id: hex::encode(k),
+			zone: v.zone.clone(),
+			capacity: v.capacity,
+			tags: v.tags.clone(),
+		})
+		.collect::<Vec<_>>();
+
+	let staged_role_changes = layout
+		.staging_roles
+		.items()
+		.iter()
+		.filter(|(k, _, v)| layout.roles.get(k) != Some(v))
+		.map(|(k, _, v)| match &v.0 {
+			None => NodeRoleChange {
+				id: hex::encode(k),
+				action: NodeRoleChangeEnum::Remove { remove: true },
+			},
+			Some(r) => NodeRoleChange {
+				id: hex::encode(k),
+				action: NodeRoleChangeEnum::Update {
+					zone: r.zone.clone(),
+					capacity: r.capacity,
+					tags: r.tags.clone(),
+				},
+			},
+		})
+		.collect::<Vec<_>>();

 	GetClusterLayoutResponse {
 		version: layout.version,
-		roles: layout
-			.roles
-			.items()
-			.iter()
-			.filter(|(_, _, v)| v.0.is_some())
-			.map(|(k, _, v)| (hex::encode(k), v.0.clone()))
-			.collect(),
-		staged_role_changes: layout
-			.staging
-			.items()
-			.iter()
-			.filter(|(k, _, v)| layout.roles.get(k) != Some(v))
-			.map(|(k, _, v)| (hex::encode(k), v.0.clone()))
-			.collect(),
+		roles,
+		staged_role_changes,
 	}
 }

+// ----
+
+#[derive(Debug, Clone, Copy, Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct ClusterHealth {
+	status: &'static str,
+	known_nodes: usize,
+	connected_nodes: usize,
+	storage_nodes: usize,
+	storage_nodes_ok: usize,
+	partitions: usize,
+	partitions_quorum: usize,
+	partitions_all_ok: usize,
+}
+
 #[derive(Serialize)]
 #[serde(rename_all = "camelCase")]
 struct GetClusterStatusResponse {
@@ -109,11 +154,19 @@ struct GetClusterStatusResponse {
 	garage_features: Option<&'static [&'static str]>,
 	rust_version: &'static str,
 	db_engine: String,
-	known_nodes: HashMap<String, KnownNodeResp>,
+	known_nodes: Vec<KnownNodeResp>,
 	layout: GetClusterLayoutResponse,
 }

 #[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+struct ApplyClusterLayoutResponse {
+	message: Vec<String>,
+	layout: GetClusterLayoutResponse,
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
 struct ConnectClusterNodesResponse {
 	success: bool,
 	error: Option<String>,
@@ -123,18 +176,31 @@ struct ConnectClusterNodesResponse {
 #[serde(rename_all = "camelCase")]
 struct GetClusterLayoutResponse {
 	version: u64,
-	roles: HashMap<String, Option<NodeRole>>,
-	staged_role_changes: HashMap<String, Option<NodeRole>>,
+	roles: Vec<NodeRoleResp>,
+	staged_role_changes: Vec<NodeRoleChange>,
 }

 #[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+struct NodeRoleResp {
+	id: String,
+	zone: String,
+	capacity: Option<u64>,
+	tags: Vec<String>,
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
 struct KnownNodeResp {
+	id: String,
 	addr: SocketAddr,
 	is_up: bool,
 	last_seen_secs_ago: Option<u64>,
 	hostname: String,
 }

+// ---- update functions ----
+
 pub async fn handle_update_cluster_layout(
 	garage: &Arc<Garage>,
 	req: Request<Body>,
@@ -144,22 +210,35 @@ pub async fn handle_update_cluster_layout(
 	let mut layout = garage.system.get_cluster_layout();

 	let mut roles = layout.roles.clone();
-	roles.merge(&layout.staging);
+	roles.merge(&layout.staging_roles);

-	for (node, role) in updates {
-		let node = hex::decode(node).ok_or_bad_request("Invalid node identifier")?;
+	for change in updates {
+		let node = hex::decode(&change.id).ok_or_bad_request("Invalid node identifier")?;
 		let node = Uuid::try_from(&node).ok_or_bad_request("Invalid node identifier")?;

+		let new_role = match change.action {
+			NodeRoleChangeEnum::Remove { remove: true } => None,
+			NodeRoleChangeEnum::Update {
+				zone,
+				capacity,
+				tags,
+			} => Some(layout::NodeRole {
+				zone,
+				capacity,
+				tags,
+			}),
+			_ => return Err(Error::bad_request("Invalid layout change")),
+		};
+
 		layout
-			.staging
-			.merge(&roles.update_mutator(node, NodeRoleV(role)));
+			.staging_roles
+			.merge(&roles.update_mutator(node, layout::NodeRoleV(new_role)));
 	}

 	garage.system.update_cluster_layout(&layout).await?;

-	Ok(Response::builder()
-		.status(StatusCode::NO_CONTENT)
-		.body(Body::empty())?)
+	let res = format_cluster_layout(&layout);
+	Ok(json_ok_response(&res)?)
 }

 pub async fn handle_apply_cluster_layout(
@@ -169,12 +248,15 @@ pub async fn handle_apply_cluster_layout(
 	let param = parse_json_body::<ApplyRevertLayoutRequest>(req).await?;

 	let layout = garage.system.get_cluster_layout();
-	let layout = layout.apply_staged_changes(Some(param.version))?;
+	let (layout, msg) = layout.apply_staged_changes(Some(param.version))?;
+
 	garage.system.update_cluster_layout(&layout).await?;

-	Ok(Response::builder()
-		.status(StatusCode::NO_CONTENT)
-		.body(Body::empty())?)
+	let res = ApplyClusterLayoutResponse {
+		message: msg,
+		layout: format_cluster_layout(&layout),
+	};
+	Ok(json_ok_response(&res)?)
 }

 pub async fn handle_revert_cluster_layout(
@@ -187,14 +269,39 @@ pub async fn handle_revert_cluster_layout(
 	let layout = layout.revert_staged_changes(Some(param.version))?;
 	garage.system.update_cluster_layout(&layout).await?;

-	Ok(Response::builder()
-		.status(StatusCode::NO_CONTENT)
-		.body(Body::empty())?)
+	let res = format_cluster_layout(&layout);
+	Ok(json_ok_response(&res)?)
 }

-type UpdateClusterLayoutRequest = HashMap<String, Option<NodeRole>>;
+// ----
+
+type UpdateClusterLayoutRequest = Vec<NodeRoleChange>;

 #[derive(Deserialize)]
+#[serde(rename_all = "camelCase")]
 struct ApplyRevertLayoutRequest {
 	version: u64,
 }
+
+// ----
+
+#[derive(Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+struct NodeRoleChange {
+	id: String,
+	#[serde(flatten)]
+	action: NodeRoleChangeEnum,
+}
+
+#[derive(Serialize, Deserialize)]
+#[serde(untagged)]
+enum NodeRoleChangeEnum {
+	#[serde(rename_all = "camelCase")]
+	Remove { remove: bool },
+	#[serde(rename_all = "camelCase")]
+	Update {
+		zone: String,
+		capacity: Option<u64>,
+		tags: Vec<String>,
+	},
+}
@@ -10,7 +10,7 @@ use garage_model::garage::Garage;
 use garage_model::key_table::*;

 use crate::admin::error::*;
-use crate::helpers::{json_ok_response, parse_json_body};
+use crate::helpers::{is_default, json_ok_response, parse_json_body};

 pub async fn handle_list_keys(garage: &Arc<Garage>) -> Result<Response<Body>, Error> {
 	let res = garage
@@ -34,6 +34,7 @@ pub async fn handle_list_keys(garage: &Arc<Garage>) -> Result<Response<Body>, Er
 }

 #[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
 struct ListKeyResultItem {
 	id: String,
 	name: String,
@@ -43,6 +44,7 @@ pub async fn handle_get_key_info(
 	garage: &Arc<Garage>,
 	id: Option<String>,
 	search: Option<String>,
+	show_secret_key: bool,
 ) -> Result<Response<Body>, Error> {
 	let key = if let Some(id) = id {
 		garage.key_helper().get_existing_key(&id).await?
@@ -55,7 +57,7 @@ pub async fn handle_get_key_info(
 		unreachable!();
 	};

-	key_info_results(garage, key).await
+	key_info_results(garage, key, show_secret_key).await
 }

 pub async fn handle_create_key(
@@ -64,15 +66,16 @@ pub async fn handle_create_key(
 ) -> Result<Response<Body>, Error> {
 	let req = parse_json_body::<CreateKeyRequest>(req).await?;

-	let key = Key::new(&req.name);
+	let key = Key::new(req.name.as_deref().unwrap_or("Unnamed key"));
 	garage.key_table.insert(&key).await?;

-	key_info_results(garage, key).await
+	key_info_results(garage, key, true).await
 }

 #[derive(Deserialize)]
+#[serde(rename_all = "camelCase")]
 struct CreateKeyRequest {
-	name: String,
+	name: Option<String>,
 }

 pub async fn handle_import_key(
@@ -86,10 +89,15 @@ pub async fn handle_import_key(
 		return Err(Error::KeyAlreadyExists(req.access_key_id.to_string()));
 	}

-	let imported_key = Key::import(&req.access_key_id, &req.secret_access_key, &req.name);
+	let imported_key = Key::import(
+		&req.access_key_id,
+		&req.secret_access_key,
+		req.name.as_deref().unwrap_or("Imported key"),
+	)
+	.ok_or_bad_request("Invalid key format")?;
 	garage.key_table.insert(&imported_key).await?;

-	key_info_results(garage, imported_key).await
+	key_info_results(garage, imported_key, false).await
 }

 #[derive(Deserialize)]
@@ -97,7 +105,7 @@ pub async fn handle_import_key(
 struct ImportKeyRequest {
 	access_key_id: String,
 	secret_access_key: String,
-	name: String,
+	name: Option<String>,
 }

 pub async fn handle_update_key(
@@ -127,10 +135,11 @@ pub async fn handle_update_key(

 	garage.key_table.insert(&key).await?;

-	key_info_results(garage, key).await
+	key_info_results(garage, key, false).await
 }

 #[derive(Deserialize)]
+#[serde(rename_all = "camelCase")]
 struct UpdateKeyRequest {
 	name: Option<String>,
 	allow: Option<KeyPerm>,
@@ -149,7 +158,11 @@ pub async fn handle_delete_key(garage: &Arc<Garage>, id: String) -> Result<Respo
 		.body(Body::empty())?)
 }

-async fn key_info_results(garage: &Arc<Garage>, key: Key) -> Result<Response<Body>, Error> {
+async fn key_info_results(
+	garage: &Arc<Garage>,
+	key: Key,
+	show_secret: bool,
+) -> Result<Response<Body>, Error> {
 	let mut relevant_buckets = HashMap::new();

 	let key_state = key.state.as_option().unwrap();
@@ -178,7 +191,11 @@ async fn key_info_results(garage: &Arc<Garage>, key: Key) -> Result<Response<Bod
 	let res = GetKeyInfoResult {
 		name: key_state.name.get().clone(),
 		access_key_id: key.key_id.clone(),
-		secret_access_key: key_state.secret_key.clone(),
+		secret_access_key: if show_secret {
+			Some(key_state.secret_key.clone())
+		} else {
+			None
+		},
 		permissions: KeyPerm {
 			create_bucket: *key_state.allow_create_bucket.get(),
 		},
@@ -224,7 +241,8 @@ async fn key_info_results(garage: &Arc<Garage>, key: Key) -> Result<Response<Bod
 struct GetKeyInfoResult {
 	name: String,
 	access_key_id: String,
-	secret_access_key: String,
+	#[serde(skip_serializing_if = "is_default")]
+	secret_access_key: Option<String>,
 	permissions: KeyPerm,
 	buckets: Vec<KeyInfoBucketResult>,
 }
@@ -246,6 +264,7 @@ struct KeyInfoBucketResult {
 }

 #[derive(Serialize, Deserialize, Default)]
+#[serde(rename_all = "camelCase")]
 pub(crate) struct ApiBucketKeyPerm {
 	#[serde(default)]
 	pub(crate) read: bool,
@@ -35,6 +35,7 @@ pub enum Endpoint {
 	GetKeyInfo {
 		id: Option<String>,
 		search: Option<String>,
+		show_secret_key: Option<String>,
 	},
 	DeleteKey {
 		id: String,
@@ -95,37 +96,37 @@ impl Endpoint {
 			GET "/check" => CheckDomain,
 			GET "/health" => Health,
 			GET "/metrics" => Metrics,
-			GET "/v0/status" => GetClusterStatus,
-			GET "/v0/health" => GetClusterHealth,
-			POST "/v0/connect" => ConnectClusterNodes,
+			GET "/v1/status" => GetClusterStatus,
+			GET "/v1/health" => GetClusterHealth,
+			POST ("/v0/connect" | "/v1/connect") => ConnectClusterNodes,
 			// Layout endpoints
-			GET "/v0/layout" => GetClusterLayout,
-			POST "/v0/layout" => UpdateClusterLayout,
-			POST "/v0/layout/apply" => ApplyClusterLayout,
-			POST "/v0/layout/revert" => RevertClusterLayout,
+			GET "/v1/layout" => GetClusterLayout,
+			POST "/v1/layout" => UpdateClusterLayout,
+			POST "/v1/layout/apply" => ApplyClusterLayout,
+			POST ("/v0/layout/revert" | "/v1/layout/revert") => RevertClusterLayout,
 			// API key endpoints
-			GET "/v0/key" if id => GetKeyInfo (query_opt::id, query_opt::search),
-			GET "/v0/key" if search => GetKeyInfo (query_opt::id, query_opt::search),
-			POST "/v0/key" if id => UpdateKey (query::id),
-			POST "/v0/key" => CreateKey,
-			POST "/v0/key/import" => ImportKey,
-			DELETE "/v0/key" if id => DeleteKey (query::id),
-			GET "/v0/key" => ListKeys,
+			GET "/v1/key" if id => GetKeyInfo (query_opt::id, query_opt::search, query_opt::show_secret_key),
+			GET "/v1/key" if search => GetKeyInfo (query_opt::id, query_opt::search, query_opt::show_secret_key),
+			POST "/v1/key" if id => UpdateKey (query::id),
+			POST "/v1/key" => CreateKey,
+			POST "/v1/key/import" => ImportKey,
+			DELETE ("/v0/key" | "/v1/key") if id => DeleteKey (query::id),
+			GET ("/v0/key" | "/v1/key") => ListKeys,
 			// Bucket endpoints
-			GET "/v0/bucket" if id => GetBucketInfo (query_opt::id, query_opt::global_alias),
-			GET "/v0/bucket" if global_alias => GetBucketInfo (query_opt::id, query_opt::global_alias),
-			GET "/v0/bucket" => ListBuckets,
-			POST "/v0/bucket" => CreateBucket,
-			DELETE "/v0/bucket" if id => DeleteBucket (query::id),
-			PUT "/v0/bucket" if id => UpdateBucket (query::id),
+			GET ("/v0/bucket" | "/v1/bucket") if id => GetBucketInfo (query_opt::id, query_opt::global_alias),
+			GET ("/v0/bucket" | "/v1/bucket") if global_alias => GetBucketInfo (query_opt::id, query_opt::global_alias),
+			GET ("/v0/bucket" | "/v1/bucket") => ListBuckets,
+			POST ("/v0/bucket" | "/v1/bucket") => CreateBucket,
+			DELETE ("/v0/bucket" | "/v1/bucket") if id => DeleteBucket (query::id),
+			PUT ("/v0/bucket" | "/v1/bucket") if id => UpdateBucket (query::id),
 			// Bucket-key permissions
-			POST "/v0/bucket/allow" => BucketAllowKey,
-			POST "/v0/bucket/deny" => BucketDenyKey,
+			POST ("/v0/bucket/allow" | "/v1/bucket/allow") => BucketAllowKey,
+			POST ("/v0/bucket/deny" | "/v1/bucket/deny") => BucketDenyKey,
 			// Bucket aliases
-			PUT "/v0/bucket/alias/global" => GlobalAliasBucket (query::id, query::alias),
-			DELETE "/v0/bucket/alias/global" => GlobalUnaliasBucket (query::id, query::alias),
-			PUT "/v0/bucket/alias/local" => LocalAliasBucket (query::id, query::access_key_id, query::alias),
-			DELETE "/v0/bucket/alias/local" => LocalUnaliasBucket (query::id, query::access_key_id, query::alias),
+			PUT ("/v0/bucket/alias/global" | "/v1/bucket/alias/global") => GlobalAliasBucket (query::id, query::alias),
+			DELETE ("/v0/bucket/alias/global" | "/v1/bucket/alias/global") => GlobalUnaliasBucket (query::id, query::alias),
+			PUT ("/v0/bucket/alias/local" | "/v1/bucket/alias/local") => LocalAliasBucket (query::id, query::access_key_id, query::alias),
+			DELETE ("/v0/bucket/alias/local" | "/v1/bucket/alias/local") => LocalUnaliasBucket (query::id, query::access_key_id, query::alias),
 		]);

 		if let Some(message) = query.nonempty_message() {
@@ -153,6 +154,7 @@ generateQueryParameters! {
 		"search" => search,
 		"globalAlias" => global_alias,
 		"alias" => alias,
-		"accessKeyId" => access_key_id
+		"accessKeyId" => access_key_id,
+		"showSecretKey" => show_secret_key
 	]
 }
@@ -152,6 +152,10 @@ pub fn json_ok_response<T: Serialize>(res: &T) -> Result<Response<Body>, Error>
 		.body(Body::from(resp_json))?)
 }

+pub fn is_default<T: Default + PartialEq>(v: &T) -> bool {
+	*v == T::default()
+}
+
 #[cfg(test)]
 mod tests {
 	use super::*;
@@ -26,6 +26,7 @@ macro_rules! router_match {
     $($meth:ident $path:pat $(if $required:ident)? => $api:ident $(($($conv:ident :: $param:ident),*))?,)*
     ]) => {{
        {
+            #[allow(unused_parens)]
            match ($method, $reqpath) {
                $(
                    (&Method::$meth, $path) if true $(&& $query.$required.is_some())? => Endpoint::$api {
@@ -128,12 +129,6 @@ macro_rules! router_match {
            }
        }
    };
-    (@if ($($cond:tt)+) then ($($then:tt)*) else ($($else:tt)*)) => {
-        $($then)*
-    };
-    (@if () then ($($then:tt)*) else ($($else:tt)*)) => {
-        $($else)*
-    };
 }

 /// This macro is used to generate part of the code in this module. It must be called only one, and
@@ -27,6 +27,7 @@ use crate::s3::cors::*;
 use crate::s3::delete::*;
 use crate::s3::get::*;
 use crate::s3::list::*;
+use crate::s3::multipart::*;
 use crate::s3::post_object::handle_post_object;
 use crate::s3::put::*;
 use crate::s3::router::Endpoint;
@@ -256,7 +257,7 @@ impl ApiHandler for S3ApiServer {
 							bucket_name,
 							bucket_id,
 							delimiter: delimiter.map(|d| d.to_string()),
-							page_size: max_keys.map(|p| p.clamp(1, 1000)).unwrap_or(1000),
+							page_size: max_keys.unwrap_or(1000).clamp(1, 1000),
 							prefix: prefix.unwrap_or_default(),
 							urlencode_resp: encoding_type.map(|e| e == "url").unwrap_or(false),
 						},
@@ -286,7 +287,7 @@ impl ApiHandler for S3ApiServer {
 								bucket_name,
 								bucket_id,
 								delimiter: delimiter.map(|d| d.to_string()),
-								page_size: max_keys.map(|p| p.clamp(1, 1000)).unwrap_or(1000),
+								page_size: max_keys.unwrap_or(1000).clamp(1, 1000),
 								urlencode_resp: encoding_type.map(|e| e == "url").unwrap_or(false),
 								prefix: prefix.unwrap_or_default(),
 							},
@@ -319,7 +320,7 @@ impl ApiHandler for S3ApiServer {
 							bucket_name,
 							bucket_id,
 							delimiter: delimiter.map(|d| d.to_string()),
-							page_size: max_uploads.map(|p| p.clamp(1, 1000)).unwrap_or(1000),
+							page_size: max_uploads.unwrap_or(1000).clamp(1, 1000),
 							prefix: prefix.unwrap_or_default(),
 							urlencode_resp: encoding_type.map(|e| e == "url").unwrap_or(false),
 						},
@@ -343,7 +344,7 @@ impl ApiHandler for S3ApiServer {
 						key,
 						upload_id,
 						part_number_marker: part_number_marker.map(|p| p.clamp(1, 10000)),
-						max_parts: max_parts.map(|p| p.clamp(1, 1000)).unwrap_or(1000),
+						max_parts: max_parts.unwrap_or(1000).clamp(1, 1000),
 					},
 				)
 				.await
@@ -2,7 +2,7 @@ use std::pin::Pin;
 use std::sync::Arc;
 use std::time::{Duration, SystemTime, UNIX_EPOCH};

-use futures::{stream, stream::Stream, StreamExt, TryFutureExt};
+use futures::{stream, stream::Stream, StreamExt};
 use md5::{Digest as Md5Digest, Md5};

 use bytes::Bytes;
@@ -18,12 +18,14 @@ use garage_util::time::*;
 use garage_model::garage::Garage;
 use garage_model::key_table::Key;
 use garage_model::s3::block_ref_table::*;
+use garage_model::s3::mpu_table::*;
 use garage_model::s3::object_table::*;
 use garage_model::s3::version_table::*;

 use crate::helpers::parse_bucket_key;
 use crate::s3::error::*;
-use crate::s3::put::{decode_upload_id, get_headers};
+use crate::s3::multipart;
+use crate::s3::put::get_headers;
 use crate::s3::xml::{self as s3_xml, xmlns_tag};

 pub async fn handle_copy(
@@ -92,7 +94,10 @@ pub async fn handle_copy(
 			let tmp_dest_object_version = ObjectVersion {
 				uuid: new_uuid,
 				timestamp: new_timestamp,
-				state: ObjectVersionState::Uploading(new_meta.headers.clone()),
+				state: ObjectVersionState::Uploading {
+					headers: new_meta.headers.clone(),
+					multipart: false,
+				},
 			};
 			let tmp_dest_object = Object::new(
 				dest_bucket_id,
@@ -105,8 +110,14 @@ pub async fn handle_copy(
 			// this means that the BlockRef entries linked to this version cannot be
 			// marked as deleted (they are marked as deleted only if the Version
 			// doesn't exist or is marked as deleted).
-			let mut dest_version =
-				Version::new(new_uuid, dest_bucket_id, dest_key.to_string(), false);
+			let mut dest_version = Version::new(
+				new_uuid,
+				VersionBacklink::Object {
+					bucket_id: dest_bucket_id,
+					key: dest_key.to_string(),
+				},
+				false,
+			);
 			garage.version_table.insert(&dest_version).await?;

 			// Fill in block list for version and insert block refs
@@ -179,17 +190,13 @@ pub async fn handle_upload_part_copy(
 ) -> Result<Response<Body>, Error> {
 	let copy_precondition = CopyPreconditionHeaders::parse(req)?;

-	let dest_version_uuid = decode_upload_id(upload_id)?;
+	let dest_upload_id = multipart::decode_upload_id(upload_id)?;

 	let dest_key = dest_key.to_string();
-	let (source_object, dest_object) = futures::try_join!(
+	let (source_object, (_, _, mut dest_mpu)) = futures::try_join!(
 		get_copy_source(&garage, api_key, req),
-		garage
-			.object_table
-			.get(&dest_bucket_id, &dest_key)
-			.map_err(Error::from),
+		multipart::get_upload(&garage, &dest_bucket_id, &dest_key, &dest_upload_id)
 	)?;
-	let dest_object = dest_object.ok_or(Error::NoSuchKey)?;

 	let (source_object_version, source_version_data, source_version_meta) =
 		extract_source_info(&source_object)?;
@@ -217,15 +224,6 @@ pub async fn handle_upload_part_copy(
 		},
 	};

-	// Check destination version is indeed in uploading state
-	if !dest_object
-		.versions()
-		.iter()
-		.any(|v| v.uuid == dest_version_uuid && v.is_uploading())
-	{
-		return Err(Error::NoSuchUpload);
-	}
-
 	// Check source version is not inlined
 	match source_version_data {
 		ObjectVersionData::DeleteMarker => unreachable!(),
@@ -242,23 +240,11 @@ pub async fn handle_upload_part_copy(

 	// Fetch source versin with its block list,
 	// and destination version to check part hasn't yet been uploaded
-	let (source_version, dest_version) = futures::try_join!(
-		garage
-			.version_table
-			.get(&source_object_version.uuid, &EmptyKey),
-		garage.version_table.get(&dest_version_uuid, &EmptyKey),
-	)?;
-	let source_version = source_version.ok_or(Error::NoSuchKey)?;
-
-	// Check this part number hasn't yet been uploaded
-	if let Some(dv) = dest_version {
-		if dv.has_part_number(part_number) {
-			return Err(Error::bad_request(format!(
-				"Part number {} has already been uploaded",
-				part_number
-			)));
-		}
-	}
+	let source_version = garage
+		.version_table
+		.get(&source_object_version.uuid, &EmptyKey)
+		.await?
+		.ok_or(Error::NoSuchKey)?;

 	// We want to reuse blocks from the source version as much as possible.
 	// However, we still need to get the data from these blocks
@@ -299,6 +285,33 @@ pub async fn handle_upload_part_copy(
 		current_offset = block_end;
 	}

+	// Calculate the identity of destination part: timestamp, version id
+	let dest_version_id = gen_uuid();
+	let dest_mpu_part_key = MpuPartKey {
+		part_number,
+		timestamp: dest_mpu.next_timestamp(part_number),
+	};
+
+	// Create the uploaded part
+	dest_mpu.parts.clear();
+	dest_mpu.parts.put(
+		dest_mpu_part_key,
+		MpuPart {
+			version: dest_version_id,
+			etag: None,
+			size: None,
+		},
+	);
+	garage.mpu_table.insert(&dest_mpu).await?;
+
+	let mut dest_version = Version::new(
+		dest_version_id,
+		VersionBacklink::MultipartUpload {
+			upload_id: dest_upload_id,
+		},
+		false,
+	);
+
 	// Now, actually copy the blocks
 	let mut md5hasher = Md5::new();

@@ -348,8 +361,8 @@ pub async fn handle_upload_part_copy(
 		let must_upload = existing_block_hash.is_none();
 		let final_hash = existing_block_hash.unwrap_or_else(|| blake2sum(&data[..]));

-		let mut version = Version::new(dest_version_uuid, dest_bucket_id, dest_key.clone(), false);
-		version.blocks.put(
+		dest_version.blocks.clear();
+		dest_version.blocks.put(
 			VersionBlockKey {
 				part_number,
 				offset: current_offset,
@@ -363,7 +376,7 @@ pub async fn handle_upload_part_copy(

 		let block_ref = BlockRef {
 			block: final_hash,
-			version: dest_version_uuid,
+			version: dest_version_id,
 			deleted: false.into(),
 		};

@@ -378,23 +391,33 @@ pub async fn handle_upload_part_copy(
 					Ok(())
 				}
 			},
-			// Thing 2: we need to insert the block in the version
-			garage.version_table.insert(&version),
-			// Thing 3: we need to add a block reference
-			garage.block_ref_table.insert(&block_ref),
+			async {
+				// Thing 2: we need to insert the block in the version
+				garage.version_table.insert(&dest_version).await?;
+				// Thing 3: we need to add a block reference
+				garage.block_ref_table.insert(&block_ref).await
+			},
 			// Thing 4: we need to prefetch the next block
 			defragmenter.next(),
 		)?;
-		next_block = res.3;
+		next_block = res.2;
 	}

+	assert_eq!(current_offset, source_range.length);
+
 	let data_md5sum = md5hasher.finalize();
 	let etag = hex::encode(data_md5sum);

 	// Put the part's ETag in the Versiontable
-	let mut version = Version::new(dest_version_uuid, dest_bucket_id, dest_key.clone(), false);
-	version.parts_etags.put(part_number, etag.clone());
-	garage.version_table.insert(&version).await?;
+	dest_mpu.parts.put(
+		dest_mpu_part_key,
+		MpuPart {
+			version: dest_version_id,
+			etag: Some(etag.clone()),
+			size: Some(current_offset),
+		},
+	);
+	garage.mpu_table.insert(&dest_mpu).await?;

 	// LGTM
 	let resp_xml = s3_xml::to_xml_with_header(&CopyPartResult {
@@ -149,7 +149,6 @@ pub async fn handle_head(

 				let (part_offset, part_end) =
 					calculate_part_bounds(&version, pn).ok_or(Error::InvalidPart)?;
-				let n_parts = version.parts_etags.items().len();

 				Ok(object_headers(object_version, version_meta)
 					.header(CONTENT_LENGTH, format!("{}", part_end - part_offset))
@@ -162,7 +161,7 @@ pub async fn handle_head(
 							version_meta.size
 						),
 					)
-					.header(X_AMZ_MP_PARTS_COUNT, format!("{}", n_parts))
+					.header(X_AMZ_MP_PARTS_COUNT, format!("{}", version.n_parts()?))
 					.status(StatusCode::PARTIAL_CONTENT)
 					.body(Body::empty())?)
 			}
@@ -376,7 +375,6 @@ async fn handle_get_part(

 			let (begin, end) =
 				calculate_part_bounds(&version, part_number).ok_or(Error::InvalidPart)?;
-			let n_parts = version.parts_etags.items().len();

 			let body = body_from_blocks_range(garage, version.blocks.items(), begin, end);

@@ -386,7 +384,7 @@ async fn handle_get_part(
 					CONTENT_RANGE,
 					format!("bytes {}-{}/{}", begin, end - 1, version_meta.size),
 				)
-				.header(X_AMZ_MP_PARTS_COUNT, format!("{}", n_parts))
+				.header(X_AMZ_MP_PARTS_COUNT, format!("{}", version.n_parts()?))
 				.body(body)?)
 		}
 		_ => unreachable!(),
@@ -1,4 +1,3 @@
-use std::cmp::Ordering;
 use std::collections::{BTreeMap, BTreeSet};
 use std::iter::{Iterator, Peekable};
 use std::sync::Arc;
@@ -11,15 +10,15 @@ use garage_util::error::Error as GarageError;
 use garage_util::time::*;

 use garage_model::garage::Garage;
+use garage_model::s3::mpu_table::*;
 use garage_model::s3::object_table::*;
-use garage_model::s3::version_table::Version;

-use garage_table::{EmptyKey, EnumerationOrder};
+use garage_table::EnumerationOrder;

 use crate::encoding::*;
 use crate::helpers::key_after_prefix;
 use crate::s3::error::*;
-use crate::s3::put as s3_put;
+use crate::s3::multipart as s3_multipart;
 use crate::s3::xml as s3_xml;

 const DUMMY_NAME: &str = "Dummy Key";
@@ -176,7 +175,9 @@ pub async fn handle_list_multipart_upload(
 			t.get_range(
 				&bucket,
 				key,
-				Some(ObjectFilter::IsUploading),
+				Some(ObjectFilter::IsUploading {
+					check_multipart: Some(true),
+				}),
 				count,
 				EnumerationOrder::Forward,
 			)
@@ -272,24 +273,26 @@ pub async fn handle_list_parts(
 ) -> Result<Response<Body>, Error> {
 	debug!("ListParts {:?}", query);

-	let upload_id = s3_put::decode_upload_id(&query.upload_id)?;
+	let upload_id = s3_multipart::decode_upload_id(&query.upload_id)?;

-	let (object, version) = futures::try_join!(
-		garage.object_table.get(&query.bucket_id, &query.key),
-		garage.version_table.get(&upload_id, &EmptyKey),
-	)?;
+	let (_, _, mpu) =
+		s3_multipart::get_upload(&garage, &query.bucket_id, &query.key, &upload_id).await?;

-	let (info, next) = fetch_part_info(query, object, version, upload_id)?;
+	let (info, next) = fetch_part_info(query, &mpu)?;

 	let result = s3_xml::ListPartsResult {
 		xmlns: (),
+
+		// Query parameters
 		bucket: s3_xml::Value(query.bucket_name.to_string()),
 		key: s3_xml::Value(query.key.to_string()),
 		upload_id: s3_xml::Value(query.upload_id.to_string()),
 		part_number_marker: query.part_number_marker.map(|e| s3_xml::IntValue(e as i64)),
-		next_part_number_marker: next.map(|e| s3_xml::IntValue(e as i64)),
 		max_parts: s3_xml::IntValue(query.max_parts as i64),
-		is_truncated: s3_xml::Value(next.map(|_| "true").unwrap_or("false").to_string()),
+
+		// Result values
+		next_part_number_marker: next.map(|e| s3_xml::IntValue(e as i64)),
+		is_truncated: s3_xml::Value(format!("{}", next.is_some())),
 		parts: info
 			.iter()
 			.map(|part| s3_xml::PartItem {
@@ -299,6 +302,8 @@ pub async fn handle_list_parts(
 				size: s3_xml::IntValue(part.size as i64),
 			})
 			.collect(),
+
+		// Dummy result values (unsupported features)
 		initiator: s3_xml::Initiator {
 			display_name: s3_xml::Value(DUMMY_NAME.to_string()),
 			id: s3_xml::Value(DUMMY_KEY.to_string()),
@@ -335,8 +340,8 @@ struct UploadInfo {
 }

 #[derive(Debug, PartialEq)]
-struct PartInfo {
-	etag: String,
+struct PartInfo<'a> {
+	etag: &'a str,
 	timestamp: u64,
 	part_number: u64,
 	size: u64,
@@ -456,107 +461,51 @@ where
 	}
 }

-fn fetch_part_info(
+fn fetch_part_info<'a>(
 	query: &ListPartsQuery,
-	object: Option<Object>,
-	version: Option<Version>,
-	upload_id: Uuid,
-) -> Result<(Vec<PartInfo>, Option<u64>), Error> {
-	// Check results
-	let object = object.ok_or(Error::NoSuchKey)?;
+	mpu: &'a MultipartUpload,
+) -> Result<(Vec<PartInfo<'a>>, Option<u64>), Error> {
+	assert!((1..=1000).contains(&query.max_parts)); // see s3/api_server.rs

-	let obj_version = object
-		.versions()
-		.iter()
-		.find(|v| v.uuid == upload_id && v.is_uploading())
-		.ok_or(Error::NoSuchUpload)?;
-
-	let version = version.ok_or(Error::NoSuchKey)?;
-
-	// Cut the beginning of our 2 vectors if required
-	let (etags, blocks) = match &query.part_number_marker {
-		Some(marker) => {
-			let next = marker + 1;
-
-			let part_idx = into_ok_or_err(
-				version
-					.parts_etags
-					.items()
-					.binary_search_by(|(part_num, _)| part_num.cmp(&next)),
-			);
-			let parts = &version.parts_etags.items()[part_idx..];
-
-			let block_idx = into_ok_or_err(
-				version
-					.blocks
-					.items()
-					.binary_search_by(|(vkey, _)| vkey.part_number.cmp(&next)),
-			);
-			let blocks = &version.blocks.items()[block_idx..];
-
-			(parts, blocks)
-		}
-		None => (version.parts_etags.items(), version.blocks.items()),
-	};
-
-	// Use the block vector to compute a (part_number, size) vector
-	let mut size = Vec::<(u64, u64)>::new();
-	blocks.iter().for_each(|(key, val)| {
-		let mut new_size = val.size;
-		match size.pop() {
-			Some((part_number, size)) if part_number == key.part_number => new_size += size,
-			Some(v) => size.push(v),
-			None => (),
-		}
-		size.push((key.part_number, new_size))
-	});
-
-	// Merge the etag vector and size vector to build a PartInfo vector
-	let max_parts = query.max_parts as usize;
-	let (mut etag_iter, mut size_iter) = (etags.iter().peekable(), size.iter().peekable());
-
-	let mut info = Vec::<PartInfo>::with_capacity(max_parts);
-
-	while info.len() < max_parts {
-		match (etag_iter.peek(), size_iter.peek()) {
-			(Some((ep, etag)), Some((sp, size))) => match ep.cmp(sp) {
-				Ordering::Less => {
-					debug!("ETag information ignored due to missing corresponding block information. Query: {:?}", query);
-					etag_iter.next();
+	// Parse multipart upload part list, removing parts not yet finished
+	// and failed part uploads that were overwritten
+	let mut parts: Vec<PartInfo<'a>> = Vec::with_capacity(mpu.parts.items().len());
+	for (pk, p) in mpu.parts.items().iter() {
+		if let (Some(etag), Some(size)) = (&p.etag, p.size) {
+			let part_info = PartInfo {
+				part_number: pk.part_number,
+				timestamp: pk.timestamp,
+				etag,
+				size,
+			};
+			match parts.last_mut() {
+				Some(lastpart) if lastpart.part_number == pk.part_number => {
+					*lastpart = part_info;
 				}
-				Ordering::Equal => {
-					info.push(PartInfo {
-						etag: etag.to_string(),
-						timestamp: obj_version.timestamp,
-						part_number: *ep,
-						size: *size,
-					});
-					etag_iter.next();
-					size_iter.next();
+				_ => {
+					parts.push(part_info);
 				}
-				Ordering::Greater => {
-					debug!("Block information ignored due to missing corresponding ETag information. Query: {:?}", query);
-					size_iter.next();
-				}
-			},
-			(None, None) => return Ok((info, None)),
-			_ => {
-				debug!(
-					"Additional block or ETag information ignored. Query: {:?}",
-					query
-				);
-				return Ok((info, None));
 			}
 		}
 	}

-	match info.last() {
-		Some(part_info) => {
-			let pagination = Some(part_info.part_number);
-			Ok((info, pagination))
-		}
-		None => Ok((info, None)),
+	// Cut the beginning if we have a marker
+	if let Some(marker) = &query.part_number_marker {
+		let next = marker + 1;
+		let part_idx = parts
+			.binary_search_by(|part| part.part_number.cmp(&next))
+			.unwrap_or_else(|x| x);
+		parts = parts.split_off(part_idx);
 	}
+
+	// Cut the end if we have too many parts
+	if parts.len() > query.max_parts as usize {
+		parts.truncate(query.max_parts as usize);
+		let pagination = Some(parts.last().unwrap().part_number);
+		return Ok((parts, pagination));
+	}
+
+	Ok((parts, None))
 }

 /*
@@ -651,7 +600,7 @@ impl ListMultipartUploadsQuery {
 				}),
 				uuid => Ok(RangeBegin::AfterUpload {
 					key: key_marker.to_string(),
-					upload: s3_put::decode_upload_id(uuid)?,
+					upload: s3_multipart::decode_upload_id(uuid)?,
 				}),
 			},

@@ -843,7 +792,7 @@ impl ExtractAccumulator for UploadAccumulator {
 		let mut uploads_for_key = object
 			.versions()
 			.iter()
-			.filter(|x| x.is_uploading())
+			.filter(|x| x.is_uploading(Some(true)))
 			.collect::<Vec<&ObjectVersion>>();

 		// S3 logic requires lexicographically sorted upload ids.
@@ -918,14 +867,6 @@ impl ExtractAccumulator for UploadAccumulator {
 * Utility functions
 */

-/// This is a stub for Result::into_ok_or_err that is not yet in Rust stable
-fn into_ok_or_err<T>(r: Result<T, T>) -> T {
-	match r {
-		Ok(r) => r,
-		Err(r) => r,
-	}
-}
-
 /// Returns the common prefix of the object given the query prefix and delimiter
 fn common_prefix<'a>(object: &'a Object, query: &ListQueryCommon) -> Option<&'a str> {
 	match &query.delimiter {
@@ -951,7 +892,6 @@ fn uriencode_maybe(s: &str, yes: bool) -> s3_xml::Value {
 #[cfg(test)]
 mod tests {
 	use super::*;
-	use garage_model::s3::version_table::*;
 	use garage_util::*;
 	use std::iter::FromIterator;

@@ -991,10 +931,13 @@ mod tests {
 		ObjectVersion {
 			uuid: Uuid::from(uuid),
 			timestamp: TS,
-			state: ObjectVersionState::Uploading(ObjectVersionHeaders {
-				content_type: "text/plain".to_string(),
-				other: BTreeMap::<String, String>::new(),
-			}),
+			state: ObjectVersionState::Uploading {
+				multipart: true,
+				headers: ObjectVersionHeaders {
+					content_type: "text/plain".to_string(),
+					other: BTreeMap::<String, String>::new(),
+				},
+			},
 		}
 	}

@@ -1169,83 +1112,77 @@ mod tests {
 		Ok(())
 	}

-	fn version() -> Version {
+	fn mpu() -> MultipartUpload {
 		let uuid = Uuid::from([0x08; 32]);

-		let blocks = vec![
+		let parts = vec![
 			(
-				VersionBlockKey {
+				MpuPartKey {
 					part_number: 1,
-					offset: 1,
+					timestamp: TS,
 				},
-				VersionBlock {
-					hash: uuid,
-					size: 3,
+				MpuPart {
+					version: uuid,
+					size: Some(3),
+					etag: Some("etag1".into()),
 				},
 			),
 			(
-				VersionBlockKey {
-					part_number: 1,
-					offset: 2,
-				},
-				VersionBlock {
-					hash: uuid,
-					size: 2,
-				},
-			),
-			(
-				VersionBlockKey {
+				MpuPartKey {
 					part_number: 2,
-					offset: 1,
+					timestamp: TS,
 				},
-				VersionBlock {
-					hash: uuid,
-					size: 8,
+				MpuPart {
+					version: uuid,
+					size: None,
+					etag: None,
 				},
 			),
 			(
-				VersionBlockKey {
+				MpuPartKey {
+					part_number: 3,
+					timestamp: TS,
+				},
+				MpuPart {
+					version: uuid,
+					size: Some(10),
+					etag: Some("etag2".into()),
+				},
+			),
+			(
+				MpuPartKey {
 					part_number: 5,
-					offset: 1,
+					timestamp: TS,
 				},
-				VersionBlock {
-					hash: uuid,
-					size: 7,
+				MpuPart {
+					version: uuid,
+					size: Some(7),
+					etag: Some("etag3".into()),
 				},
 			),
 			(
-				VersionBlockKey {
+				MpuPartKey {
 					part_number: 8,
-					offset: 1,
+					timestamp: TS,
 				},
-				VersionBlock {
-					hash: uuid,
-					size: 5,
+				MpuPart {
+					version: uuid,
+					size: Some(5),
+					etag: Some("etag4".into()),
 				},
 			),
 		];
-		let etags = vec![
-			(1, "etag1".to_string()),
-			(3, "etag2".to_string()),
-			(5, "etag3".to_string()),
-			(8, "etag4".to_string()),
-			(9, "etag5".to_string()),
-		];

-		Version {
-			bucket_id: uuid,
-			key: "a".to_string(),
-			uuid,
+		MultipartUpload {
+			upload_id: uuid,
+			timestamp: TS,
 			deleted: false.into(),
-			blocks: crdt::Map::<VersionBlockKey, VersionBlock>::from_iter(blocks),
-			parts_etags: crdt::Map::<u64, String>::from_iter(etags),
+			parts: crdt::Map::<MpuPartKey, MpuPart>::from_iter(parts),
+			bucket_id: uuid,
+			key: "a".into(),
 		}
 	}

-	fn obj() -> Object {
-		Object::new(bucket(), "d".to_string(), vec![objup_version([0x08; 32])])
-	}
-
 	#[test]
 	fn test_fetch_part_info() -> Result<(), Error> {
 		let uuid = Uuid::from([0x08; 32]);
@@ -1258,82 +1195,85 @@ mod tests {
 			max_parts: 2,
 		};

-		assert!(
-			fetch_part_info(&query, None, None, uuid).is_err(),
-			"No object and version should fail"
-		);
-		assert!(
-			fetch_part_info(&query, Some(obj()), None, uuid).is_err(),
-			"No version should faild"
-		);
-		assert!(
-			fetch_part_info(&query, None, Some(version()), uuid).is_err(),
-			"No object should fail"
-		);
+		let mpu = mpu();

 		// Start from the beginning but with limited size to trigger pagination
-		let (info, pagination) = fetch_part_info(&query, Some(obj()), Some(version()), uuid)?;
-		assert_eq!(pagination.unwrap(), 5);
+		let (info, pagination) = fetch_part_info(&query, &mpu)?;
+		assert_eq!(pagination.unwrap(), 3);
 		assert_eq!(
 			info,
 			vec![
 				PartInfo {
-					etag: "etag1".to_string(),
+					etag: "etag1",
 					timestamp: TS,
 					part_number: 1,
-					size: 5
+					size: 3
 				},
 				PartInfo {
-					etag: "etag3".to_string(),
+					etag: "etag2",
 					timestamp: TS,
-					part_number: 5,
-					size: 7
+					part_number: 3,
+					size: 10
 				},
 			]
 		);

 		// Use previous pagination to make a new request
 		query.part_number_marker = Some(pagination.unwrap());
-		let (info, pagination) = fetch_part_info(&query, Some(obj()), Some(version()), uuid)?;
+		let (info, pagination) = fetch_part_info(&query, &mpu)?;
 		assert!(pagination.is_none());
 		assert_eq!(
 			info,
-			vec![PartInfo {
-				etag: "etag4".to_string(),
-				timestamp: TS,
-				part_number: 8,
-				size: 5
-			},]
+			vec![
+				PartInfo {
+					etag: "etag3",
+					timestamp: TS,
+					part_number: 5,
+					size: 7
+				},
+				PartInfo {
+					etag: "etag4",
+					timestamp: TS,
+					part_number: 8,
+					size: 5
+				},
+			]
 		);

 		// Trying to access a part that is way larger than registered ones
 		query.part_number_marker = Some(9999);
-		let (info, pagination) = fetch_part_info(&query, Some(obj()), Some(version()), uuid)?;
+		let (info, pagination) = fetch_part_info(&query, &mpu)?;
 		assert!(pagination.is_none());
 		assert_eq!(info, vec![]);

 		// Try without any limitation
 		query.max_parts = 1000;
 		query.part_number_marker = None;
-		let (info, pagination) = fetch_part_info(&query, Some(obj()), Some(version()), uuid)?;
+		let (info, pagination) = fetch_part_info(&query, &mpu)?;
 		assert!(pagination.is_none());
 		assert_eq!(
 			info,
 			vec![
 				PartInfo {
-					etag: "etag1".to_string(),
+					etag: "etag1",
 					timestamp: TS,
 					part_number: 1,
-					size: 5
+					size: 3
 				},
 				PartInfo {
-					etag: "etag3".to_string(),
+					etag: "etag2",
+					timestamp: TS,
+					part_number: 3,
+					size: 10
+				},
+				PartInfo {
+					etag: "etag3",
 					timestamp: TS,
 					part_number: 5,
 					size: 7
 				},
 				PartInfo {
-					etag: "etag4".to_string(),
+					etag: "etag4",
 					timestamp: TS,
 					part_number: 8,
 					size: 5
@@ -7,6 +7,7 @@ pub mod cors;
 mod delete;
 pub mod get;
 mod list;
+mod multipart;
 mod post_object;
 mod put;
 mod website;
@@ -0,0 +1,465 @@
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use futures::prelude::*;
+use hyper::body::Body;
+use hyper::{Request, Response};
+use md5::{Digest as Md5Digest, Md5};
+
+use garage_table::*;
+use garage_util::async_hash::*;
+use garage_util::data::*;
+use garage_util::time::*;
+
+use garage_model::bucket_table::Bucket;
+use garage_model::garage::Garage;
+use garage_model::s3::block_ref_table::*;
+use garage_model::s3::mpu_table::*;
+use garage_model::s3::object_table::*;
+use garage_model::s3::version_table::*;
+
+use crate::s3::error::*;
+use crate::s3::put::*;
+use crate::s3::xml as s3_xml;
+use crate::signature::verify_signed_content;
+
+// ----
+
+pub async fn handle_create_multipart_upload(
+	garage: Arc<Garage>,
+	req: &Request<Body>,
+	bucket_name: &str,
+	bucket_id: Uuid,
+	key: &str,
+) -> Result<Response<Body>, Error> {
+	let upload_id = gen_uuid();
+	let timestamp = now_msec();
+	let headers = get_headers(req.headers())?;
+
+	// Create object in object table
+	let object_version = ObjectVersion {
+		uuid: upload_id,
+		timestamp,
+		state: ObjectVersionState::Uploading {
+			multipart: true,
+			headers,
+		},
+	};
+	let object = Object::new(bucket_id, key.to_string(), vec![object_version]);
+	garage.object_table.insert(&object).await?;
+
+	// Create multipart upload in mpu table
+	// This multipart upload will hold references to uploaded parts
+	// (which are entries in the Version table)
+	let mpu = MultipartUpload::new(upload_id, timestamp, bucket_id, key.into(), false);
+	garage.mpu_table.insert(&mpu).await?;
+
+	// Send success response
+	let result = s3_xml::InitiateMultipartUploadResult {
+		xmlns: (),
+		bucket: s3_xml::Value(bucket_name.to_string()),
+		key: s3_xml::Value(key.to_string()),
+		upload_id: s3_xml::Value(hex::encode(upload_id)),
+	};
+	let xml = s3_xml::to_xml_with_header(&result)?;
+
+	Ok(Response::new(Body::from(xml.into_bytes())))
+}
+
+pub async fn handle_put_part(
+	garage: Arc<Garage>,
+	req: Request<Body>,
+	bucket_id: Uuid,
+	key: &str,
+	part_number: u64,
+	upload_id: &str,
+	content_sha256: Option<Hash>,
+) -> Result<Response<Body>, Error> {
+	let upload_id = decode_upload_id(upload_id)?;
+
+	let content_md5 = match req.headers().get("content-md5") {
+		Some(x) => Some(x.to_str()?.to_string()),
+		None => None,
+	};
+
+	// Read first chuck, and at the same time try to get object to see if it exists
+	let key = key.to_string();
+
+	let body = req.into_body().map_err(Error::from);
+	let mut chunker = StreamChunker::new(body, garage.config.block_size);
+
+	let ((_, _, mut mpu), first_block) = futures::try_join!(
+		get_upload(&garage, &bucket_id, &key, &upload_id),
+		chunker.next(),
+	)?;
+
+	// Check object is valid and part can be accepted
+	let first_block = first_block.ok_or_bad_request("Empty body")?;
+
+	// Calculate part identity: timestamp, version id
+	let version_uuid = gen_uuid();
+	let mpu_part_key = MpuPartKey {
+		part_number,
+		timestamp: mpu.next_timestamp(part_number),
+	};
+
+	// The following consists in many steps that can each fail.
+	// Keep track that some cleanup will be needed if things fail
+	// before everything is finished (cleanup is done using the Drop trait).
+	let mut interrupted_cleanup = InterruptedCleanup(Some(InterruptedCleanupInner {
+		garage: garage.clone(),
+		upload_id,
+		version_uuid,
+	}));
+
+	// Create version and link version from MPU
+	mpu.parts.clear();
+	mpu.parts.put(
+		mpu_part_key,
+		MpuPart {
+			version: version_uuid,
+			etag: None,
+			size: None,
+		},
+	);
+	garage.mpu_table.insert(&mpu).await?;
+
+	let version = Version::new(
+		version_uuid,
+		VersionBacklink::MultipartUpload { upload_id },
+		false,
+	);
+	garage.version_table.insert(&version).await?;
+
+	// Copy data to version
+	let first_block_hash = async_blake2sum(first_block.clone()).await;
+
+	let (total_size, data_md5sum, data_sha256sum) = read_and_put_blocks(
+		&garage,
+		&version,
+		part_number,
+		first_block,
+		first_block_hash,
+		&mut chunker,
+	)
+	.await?;
+
+	// Verify that checksums map
+	ensure_checksum_matches(
+		data_md5sum.as_slice(),
+		data_sha256sum,
+		content_md5.as_deref(),
+		content_sha256,
+	)?;
+
+	// Store part etag in version
+	let data_md5sum_hex = hex::encode(data_md5sum);
+	mpu.parts.put(
+		mpu_part_key,
+		MpuPart {
+			version: version_uuid,
+			etag: Some(data_md5sum_hex.clone()),
+			size: Some(total_size),
+		},
+	);
+	garage.mpu_table.insert(&mpu).await?;
+
+	// We were not interrupted, everything went fine.
+	// We won't have to clean up on drop.
+	interrupted_cleanup.cancel();
+
+	let response = Response::builder()
+		.header("ETag", format!("\"{}\"", data_md5sum_hex))
+		.body(Body::empty())
+		.unwrap();
+	Ok(response)
+}
+
+struct InterruptedCleanup(Option<InterruptedCleanupInner>);
+struct InterruptedCleanupInner {
+	garage: Arc<Garage>,
+	upload_id: Uuid,
+	version_uuid: Uuid,
+}
+
+impl InterruptedCleanup {
+	fn cancel(&mut self) {
+		drop(self.0.take());
+	}
+}
+impl Drop for InterruptedCleanup {
+	fn drop(&mut self) {
+		if let Some(info) = self.0.take() {
+			tokio::spawn(async move {
+				let version = Version::new(
+					info.version_uuid,
+					VersionBacklink::MultipartUpload {
+						upload_id: info.upload_id,
+					},
+					true,
+				);
+				if let Err(e) = info.garage.version_table.insert(&version).await {
+					warn!("Cannot cleanup after aborted UploadPart: {}", e);
+				}
+			});
+		}
+	}
+}
+
+pub async fn handle_complete_multipart_upload(
+	garage: Arc<Garage>,
+	req: Request<Body>,
+	bucket_name: &str,
+	bucket: &Bucket,
+	key: &str,
+	upload_id: &str,
+	content_sha256: Option<Hash>,
+) -> Result<Response<Body>, Error> {
+	let body = hyper::body::to_bytes(req.into_body()).await?;
+
+	if let Some(content_sha256) = content_sha256 {
+		verify_signed_content(content_sha256, &body[..])?;
+	}
+
+	let body_xml = roxmltree::Document::parse(std::str::from_utf8(&body)?)?;
+	let body_list_of_parts = parse_complete_multipart_upload_body(&body_xml)
+		.ok_or_bad_request("Invalid CompleteMultipartUpload XML")?;
+	debug!(
+		"CompleteMultipartUpload list of parts: {:?}",
+		body_list_of_parts
+	);
+
+	let upload_id = decode_upload_id(upload_id)?;
+
+	// Get object and multipart upload
+	let key = key.to_string();
+	let (_, mut object_version, mpu) = get_upload(&garage, &bucket.id, &key, &upload_id).await?;
+
+	if mpu.parts.is_empty() {
+		return Err(Error::bad_request("No data was uploaded"));
+	}
+
+	let headers = match object_version.state {
+		ObjectVersionState::Uploading { headers, .. } => headers,
+		_ => unreachable!(),
+	};
+
+	// Check that part numbers are an increasing sequence.
+	// (it doesn't need to start at 1 nor to be a continuous sequence,
+	// see discussion in #192)
+	if body_list_of_parts.is_empty() {
+		return Err(Error::EntityTooSmall);
+	}
+	if !body_list_of_parts
+		.iter()
+		.zip(body_list_of_parts.iter().skip(1))
+		.all(|(p1, p2)| p1.part_number < p2.part_number)
+	{
+		return Err(Error::InvalidPartOrder);
+	}
+
+	// Check that the list of parts they gave us corresponds to parts we have here
+	debug!("Parts stored in multipart upload: {:?}", mpu.parts.items());
+	let mut have_parts = HashMap::new();
+	for (pk, pv) in mpu.parts.items().iter() {
+		have_parts.insert(pk.part_number, pv);
+	}
+	let mut parts = vec![];
+	for req_part in body_list_of_parts.iter() {
+		match have_parts.get(&req_part.part_number) {
+			Some(part) if part.etag.as_ref() == Some(&req_part.etag) && part.size.is_some() => {
+				parts.push(*part)
+			}
+			_ => return Err(Error::InvalidPart),
+		}
+	}
+
+	let grg = &garage;
+	let parts_versions = futures::future::try_join_all(parts.iter().map(|p| async move {
+		grg.version_table
+			.get(&p.version, &EmptyKey)
+			.await?
+			.ok_or_internal_error("Part version missing from version table")
+	}))
+	.await?;
+
+	// Create final version and block refs
+	let mut final_version = Version::new(
+		upload_id,
+		VersionBacklink::Object {
+			bucket_id: bucket.id,
+			key: key.to_string(),
+		},
+		false,
+	);
+	for (part_number, part_version) in parts_versions.iter().enumerate() {
+		if part_version.deleted.get() {
+			return Err(Error::InvalidPart);
+		}
+		for (vbk, vb) in part_version.blocks.items().iter() {
+			final_version.blocks.put(
+				VersionBlockKey {
+					part_number: (part_number + 1) as u64,
+					offset: vbk.offset,
+				},
+				*vb,
+			);
+		}
+	}
+	garage.version_table.insert(&final_version).await?;
+
+	let block_refs = final_version.blocks.items().iter().map(|(_, b)| BlockRef {
+		block: b.hash,
+		version: upload_id,
+		deleted: false.into(),
+	});
+	garage.block_ref_table.insert_many(block_refs).await?;
+
+	// Calculate etag of final object
+	// To understand how etags are calculated, read more here:
+	// https://teppen.io/2018/06/23/aws_s3_etags/
+	let mut etag_md5_hasher = Md5::new();
+	for part in parts.iter() {
+		etag_md5_hasher.update(part.etag.as_ref().unwrap().as_bytes());
+	}
+	let etag = format!(
+		"{}-{}",
+		hex::encode(etag_md5_hasher.finalize()),
+		parts.len()
+	);
+
+	// Calculate total size of final object
+	let total_size = parts.iter().map(|x| x.size.unwrap()).sum();
+
+	if let Err(e) = check_quotas(&garage, bucket, &key, total_size).await {
+		object_version.state = ObjectVersionState::Aborted;
+		let final_object = Object::new(bucket.id, key.clone(), vec![object_version]);
+		garage.object_table.insert(&final_object).await?;
+
+		return Err(e);
+	}
+
+	// Write final object version
+	object_version.state = ObjectVersionState::Complete(ObjectVersionData::FirstBlock(
+		ObjectVersionMeta {
+			headers,
+			size: total_size,
+			etag: etag.clone(),
+		},
+		final_version.blocks.items()[0].1.hash,
+	));
+
+	let final_object = Object::new(bucket.id, key.clone(), vec![object_version]);
+	garage.object_table.insert(&final_object).await?;
+
+	// Send response saying ok we're done
+	let result = s3_xml::CompleteMultipartUploadResult {
+		xmlns: (),
+		location: None,
+		bucket: s3_xml::Value(bucket_name.to_string()),
+		key: s3_xml::Value(key),
+		etag: s3_xml::Value(format!("\"{}\"", etag)),
+	};
+	let xml = s3_xml::to_xml_with_header(&result)?;
+
+	Ok(Response::new(Body::from(xml.into_bytes())))
+}
+
+pub async fn handle_abort_multipart_upload(
+	garage: Arc<Garage>,
+	bucket_id: Uuid,
+	key: &str,
+	upload_id: &str,
+) -> Result<Response<Body>, Error> {
+	let upload_id = decode_upload_id(upload_id)?;
+
+	let (_, mut object_version, _) =
+		get_upload(&garage, &bucket_id, &key.to_string(), &upload_id).await?;
+
+	object_version.state = ObjectVersionState::Aborted;
+	let final_object = Object::new(bucket_id, key.to_string(), vec![object_version]);
+	garage.object_table.insert(&final_object).await?;
+
+	Ok(Response::new(Body::from(vec![])))
+}
+
+// ======== helpers ============
+
+#[allow(clippy::ptr_arg)]
+pub(crate) async fn get_upload(
+	garage: &Garage,
+	bucket_id: &Uuid,
+	key: &String,
+	upload_id: &Uuid,
+) -> Result<(Object, ObjectVersion, MultipartUpload), Error> {
+	let (object, mpu) = futures::try_join!(
+		garage.object_table.get(bucket_id, key).map_err(Error::from),
+		garage
+			.mpu_table
+			.get(upload_id, &EmptyKey)
+			.map_err(Error::from),
+	)?;
+
+	let object = object.ok_or(Error::NoSuchUpload)?;
+	let mpu = mpu.ok_or(Error::NoSuchUpload)?;
+
+	let object_version = object
+		.versions()
+		.iter()
+		.find(|v| v.uuid == *upload_id && v.is_uploading(Some(true)))
+		.ok_or(Error::NoSuchUpload)?
+		.clone();
+
+	Ok((object, object_version, mpu))
+}
+
+pub fn decode_upload_id(id: &str) -> Result<Uuid, Error> {
+	let id_bin = hex::decode(id).map_err(|_| Error::NoSuchUpload)?;
+	if id_bin.len() != 32 {
+		return Err(Error::NoSuchUpload);
+	}
+	let mut uuid = [0u8; 32];
+	uuid.copy_from_slice(&id_bin[..]);
+	Ok(Uuid::from(uuid))
+}
+
+#[derive(Debug)]
+struct CompleteMultipartUploadPart {
+	etag: String,
+	part_number: u64,
+}
+
+fn parse_complete_multipart_upload_body(
+	xml: &roxmltree::Document,
+) -> Option<Vec<CompleteMultipartUploadPart>> {
+	let mut parts = vec![];
+
+	let root = xml.root();
+	let cmu = root.first_child()?;
+	if !cmu.has_tag_name("CompleteMultipartUpload") {
+		return None;
+	}
+
+	for item in cmu.children() {
+		// Only parse <Part> nodes
+		if !item.is_element() {
+			continue;
+		}
+
+		if item.has_tag_name("Part") {
+			let etag = item.children().find(|e| e.has_tag_name("ETag"))?.text()?;
+			let part_number = item
+				.children()
+				.find(|e| e.has_tag_name("PartNumber"))?
+				.text()?;
+			parts.push(CompleteMultipartUploadPart {
+				etag: etag.trim_matches('"').to_string(),
+				part_number: part_number.parse().ok()?,
+			});
+		} else {
+			return None;
+		}
+	}
+
+	Some(parts)
+}
@@ -1,4 +1,4 @@
-use std::collections::{BTreeMap, BTreeSet, HashMap};
+use std::collections::{BTreeMap, HashMap};
 use std::sync::Arc;

 use base64::prelude::*;
@@ -30,8 +30,6 @@ use garage_model::s3::object_table::*;
 use garage_model::s3::version_table::*;

 use crate::s3::error::*;
-use crate::s3::xml as s3_xml;
-use crate::signature::verify_signed_content;

 pub async fn handle_put(
 	garage: Arc<Garage>,
@@ -123,20 +121,23 @@ pub(crate) async fn save_stream<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
 	// The following consists in many steps that can each fail.
 	// Keep track that some cleanup will be needed if things fail
 	// before everything is finished (cleanup is done using the Drop trait).
-	let mut interrupted_cleanup = InterruptedCleanup(Some((
-		garage.clone(),
-		bucket.id,
-		key.into(),
+	let mut interrupted_cleanup = InterruptedCleanup(Some(InterruptedCleanupInner {
+		garage: garage.clone(),
+		bucket_id: bucket.id,
+		key: key.into(),
 		version_uuid,
 		version_timestamp,
-	)));
+	}));

 	// Write version identifier in object table so that we have a trace
 	// that we are uploading something
 	let mut object_version = ObjectVersion {
 		uuid: version_uuid,
 		timestamp: version_timestamp,
-		state: ObjectVersionState::Uploading(headers.clone()),
+		state: ObjectVersionState::Uploading {
+			headers: headers.clone(),
+			multipart: false,
+		},
 	};
 	let object = Object::new(bucket.id, key.into(), vec![object_version.clone()]);
 	garage.object_table.insert(&object).await?;
@@ -145,7 +146,14 @@ pub(crate) async fn save_stream<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
 	// Write this entry now, even with empty block list,
 	// to prevent block_ref entries from being deleted (they can be deleted
 	// if the reference a version that isn't found in the version table)
-	let version = Version::new(version_uuid, bucket.id, key.into(), false);
+	let version = Version::new(
+		version_uuid,
+		VersionBacklink::Object {
+			bucket_id: bucket.id,
+			key: key.into(),
+		},
+		false,
+	);
 	garage.version_table.insert(&version).await?;

 	// Transfer data and verify checksum
@@ -192,7 +200,7 @@ pub(crate) async fn save_stream<S: Stream<Item = Result<Bytes, Error>> + Unpin>(

 /// Validate MD5 sum against content-md5 header
 /// and sha256sum against signed content-sha256
-fn ensure_checksum_matches(
+pub(crate) fn ensure_checksum_matches(
 	data_md5sum: &[u8],
 	data_sha256sum: garage_util::data::FixedBytes32,
 	content_md5: Option<&str>,
@@ -218,7 +226,7 @@ fn ensure_checksum_matches(
 }

 /// Check that inserting this object with this size doesn't exceed bucket quotas
-async fn check_quotas(
+pub(crate) async fn check_quotas(
 	garage: &Arc<Garage>,
 	bucket: &Bucket,
 	key: &str,
@@ -275,7 +283,7 @@ async fn check_quotas(
 	Ok(())
 }

-async fn read_and_put_blocks<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
+pub(crate) async fn read_and_put_blocks<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
 	garage: &Garage,
 	version: &Version,
 	part_number: u64,
@@ -381,7 +389,7 @@ async fn put_block_meta(
 	Ok(())
 }

-struct StreamChunker<S: Stream<Item = Result<Bytes, Error>>> {
+pub(crate) struct StreamChunker<S: Stream<Item = Result<Bytes, Error>>> {
 	stream: S,
 	read_all: bool,
 	block_size: usize,
@@ -389,7 +397,7 @@ struct StreamChunker<S: Stream<Item = Result<Bytes, Error>>> {
 }

 impl<S: Stream<Item = Result<Bytes, Error>> + Unpin> StreamChunker<S> {
-	fn new(stream: S, block_size: usize) -> Self {
+	pub(crate) fn new(stream: S, block_size: usize) -> Self {
 		Self {
 			stream,
 			read_all: false,
@@ -398,7 +406,7 @@ impl<S: Stream<Item = Result<Bytes, Error>> + Unpin> StreamChunker<S> {
 		}
 	}

-	async fn next(&mut self) -> Result<Option<Bytes>, Error> {
+	pub(crate) async fn next(&mut self) -> Result<Option<Bytes>, Error> {
 		while !self.read_all && self.buf.len() < self.block_size {
 			if let Some(block) = self.stream.next().await {
 				let bytes = block?;
@@ -425,7 +433,14 @@ pub fn put_response(version_uuid: Uuid, md5sum_hex: String) -> Response<Body> {
 		.unwrap()
 }

-struct InterruptedCleanup(Option<(Arc<Garage>, Uuid, String, Uuid, u64)>);
+struct InterruptedCleanup(Option<InterruptedCleanupInner>);
+struct InterruptedCleanupInner {
+	garage: Arc<Garage>,
+	bucket_id: Uuid,
+	key: String,
+	version_uuid: Uuid,
+	version_timestamp: u64,
+}

 impl InterruptedCleanup {
 	fn cancel(&mut self) {
@@ -434,15 +449,15 @@ impl InterruptedCleanup {
 }
 impl Drop for InterruptedCleanup {
 	fn drop(&mut self) {
-		if let Some((garage, bucket_id, key, version_uuid, version_ts)) = self.0.take() {
+		if let Some(info) = self.0.take() {
 			tokio::spawn(async move {
 				let object_version = ObjectVersion {
-					uuid: version_uuid,
-					timestamp: version_ts,
+					uuid: info.version_uuid,
+					timestamp: info.version_timestamp,
 					state: ObjectVersionState::Aborted,
 				};
-				let object = Object::new(bucket_id, key, vec![object_version]);
-				if let Err(e) = garage.object_table.insert(&object).await {
+				let object = Object::new(info.bucket_id, info.key, vec![object_version]);
+				if let Err(e) = info.garage.object_table.insert(&object).await {
 					warn!("Cannot cleanup after aborted PutObject: {}", e);
 				}
 			});
@@ -450,326 +465,9 @@ impl Drop for InterruptedCleanup {
 	}
 }

-// ----
+// ============ helpers ============

-pub async fn handle_create_multipart_upload(
-	garage: Arc<Garage>,
-	req: &Request<Body>,
-	bucket_name: &str,
-	bucket_id: Uuid,
-	key: &str,
-) -> Result<Response<Body>, Error> {
-	let version_uuid = gen_uuid();
-	let headers = get_headers(req.headers())?;
-
-	// Create object in object table
-	let object_version = ObjectVersion {
-		uuid: version_uuid,
-		timestamp: now_msec(),
-		state: ObjectVersionState::Uploading(headers),
-	};
-	let object = Object::new(bucket_id, key.to_string(), vec![object_version]);
-	garage.object_table.insert(&object).await?;
-
-	// Insert empty version so that block_ref entries refer to something
-	// (they are inserted concurrently with blocks in the version table, so
-	// there is the possibility that they are inserted before the version table
-	// is created, in which case it is allowed to delete them, e.g. in repair_*)
-	let version = Version::new(version_uuid, bucket_id, key.into(), false);
-	garage.version_table.insert(&version).await?;
-
-	// Send success response
-	let result = s3_xml::InitiateMultipartUploadResult {
-		xmlns: (),
-		bucket: s3_xml::Value(bucket_name.to_string()),
-		key: s3_xml::Value(key.to_string()),
-		upload_id: s3_xml::Value(hex::encode(version_uuid)),
-	};
-	let xml = s3_xml::to_xml_with_header(&result)?;
-
-	Ok(Response::new(Body::from(xml.into_bytes())))
-}
-
-pub async fn handle_put_part(
-	garage: Arc<Garage>,
-	req: Request<Body>,
-	bucket_id: Uuid,
-	key: &str,
-	part_number: u64,
-	upload_id: &str,
-	content_sha256: Option<Hash>,
-) -> Result<Response<Body>, Error> {
-	let version_uuid = decode_upload_id(upload_id)?;
-
-	let content_md5 = match req.headers().get("content-md5") {
-		Some(x) => Some(x.to_str()?.to_string()),
-		None => None,
-	};
-
-	// Read first chuck, and at the same time try to get object to see if it exists
-	let key = key.to_string();
-
-	let body = req.into_body().map_err(Error::from);
-	let mut chunker = StreamChunker::new(body, garage.config.block_size);
-
-	let (object, version, first_block) = futures::try_join!(
-		garage
-			.object_table
-			.get(&bucket_id, &key)
-			.map_err(Error::from),
-		garage
-			.version_table
-			.get(&version_uuid, &EmptyKey)
-			.map_err(Error::from),
-		chunker.next(),
-	)?;
-
-	// Check object is valid and multipart block can be accepted
-	let first_block = first_block.ok_or_bad_request("Empty body")?;
-	let object = object.ok_or_bad_request("Object not found")?;
-
-	if !object
-		.versions()
-		.iter()
-		.any(|v| v.uuid == version_uuid && v.is_uploading())
-	{
-		return Err(Error::NoSuchUpload);
-	}
-
-	// Check part hasn't already been uploaded
-	if let Some(v) = version {
-		if v.has_part_number(part_number) {
-			return Err(Error::bad_request(format!(
-				"Part number {} has already been uploaded",
-				part_number
-			)));
-		}
-	}
-
-	// Copy block to store
-	let version = Version::new(version_uuid, bucket_id, key, false);
-
-	let first_block_hash = async_blake2sum(first_block.clone()).await;
-
-	let (_, data_md5sum, data_sha256sum) = read_and_put_blocks(
-		&garage,
-		&version,
-		part_number,
-		first_block,
-		first_block_hash,
-		&mut chunker,
-	)
-	.await?;
-
-	// Verify that checksums map
-	ensure_checksum_matches(
-		data_md5sum.as_slice(),
-		data_sha256sum,
-		content_md5.as_deref(),
-		content_sha256,
-	)?;
-
-	// Store part etag in version
-	let data_md5sum_hex = hex::encode(data_md5sum);
-	let mut version = version;
-	version
-		.parts_etags
-		.put(part_number, data_md5sum_hex.clone());
-	garage.version_table.insert(&version).await?;
-
-	let response = Response::builder()
-		.header("ETag", format!("\"{}\"", data_md5sum_hex))
-		.body(Body::empty())
-		.unwrap();
-	Ok(response)
-}
-
-pub async fn handle_complete_multipart_upload(
-	garage: Arc<Garage>,
-	req: Request<Body>,
-	bucket_name: &str,
-	bucket: &Bucket,
-	key: &str,
-	upload_id: &str,
-	content_sha256: Option<Hash>,
-) -> Result<Response<Body>, Error> {
-	let body = hyper::body::to_bytes(req.into_body()).await?;
-
-	if let Some(content_sha256) = content_sha256 {
-		verify_signed_content(content_sha256, &body[..])?;
-	}
-
-	let body_xml = roxmltree::Document::parse(std::str::from_utf8(&body)?)?;
-	let body_list_of_parts = parse_complete_multipart_upload_body(&body_xml)
-		.ok_or_bad_request("Invalid CompleteMultipartUpload XML")?;
-	debug!(
-		"CompleteMultipartUpload list of parts: {:?}",
-		body_list_of_parts
-	);
-
-	let version_uuid = decode_upload_id(upload_id)?;
-
-	// Get object and version
-	let key = key.to_string();
-	let (object, version) = futures::try_join!(
-		garage.object_table.get(&bucket.id, &key),
-		garage.version_table.get(&version_uuid, &EmptyKey),
-	)?;
-
-	let object = object.ok_or(Error::NoSuchKey)?;
-	let mut object_version = object
-		.versions()
-		.iter()
-		.find(|v| v.uuid == version_uuid && v.is_uploading())
-		.cloned()
-		.ok_or(Error::NoSuchUpload)?;
-
-	let version = version.ok_or(Error::NoSuchKey)?;
-	if version.blocks.is_empty() {
-		return Err(Error::bad_request("No data was uploaded"));
-	}
-
-	let headers = match object_version.state {
-		ObjectVersionState::Uploading(headers) => headers,
-		_ => unreachable!(),
-	};
-
-	// Check that part numbers are an increasing sequence.
-	// (it doesn't need to start at 1 nor to be a continuous sequence,
-	// see discussion in #192)
-	if body_list_of_parts.is_empty() {
-		return Err(Error::EntityTooSmall);
-	}
-	if !body_list_of_parts
-		.iter()
-		.zip(body_list_of_parts.iter().skip(1))
-		.all(|(p1, p2)| p1.part_number < p2.part_number)
-	{
-		return Err(Error::InvalidPartOrder);
-	}
-
-	// Garage-specific restriction, see #204: part numbers must be
-	// consecutive starting at 1
-	if body_list_of_parts[0].part_number != 1
-		|| !body_list_of_parts
-			.iter()
-			.zip(body_list_of_parts.iter().skip(1))
-			.all(|(p1, p2)| p1.part_number + 1 == p2.part_number)
-	{
-		return Err(Error::NotImplemented("Garage does not support completing a Multipart upload with non-consecutive part numbers. This is a restriction of Garage's data model, which might be fixed in a future release. See issue #204 for more information on this topic.".into()));
-	}
-
-	// Check that the list of parts they gave us corresponds to the parts we have here
-	debug!("Expected parts from request: {:?}", body_list_of_parts);
-	debug!("Parts stored in version: {:?}", version.parts_etags.items());
-	let parts = version
-		.parts_etags
-		.items()
-		.iter()
-		.map(|pair| (&pair.0, &pair.1));
-	let same_parts = body_list_of_parts
-		.iter()
-		.map(|x| (&x.part_number, &x.etag))
-		.eq(parts);
-	if !same_parts {
-		return Err(Error::InvalidPart);
-	}
-
-	// Check that all blocks belong to one of the parts
-	let block_parts = version
-		.blocks
-		.items()
-		.iter()
-		.map(|(bk, _)| bk.part_number)
-		.collect::<BTreeSet<_>>();
-	let same_parts = body_list_of_parts
-		.iter()
-		.map(|x| x.part_number)
-		.eq(block_parts.into_iter());
-	if !same_parts {
-		return Err(Error::bad_request(
-			"Part numbers in block list and part list do not match. This can happen if a part was partially uploaded. Please abort the multipart upload and try again."
-		));
-	}
-
-	// Calculate etag of final object
-	// To understand how etags are calculated, read more here:
-	// https://teppen.io/2018/06/23/aws_s3_etags/
-	let num_parts = body_list_of_parts.len();
-	let mut etag_md5_hasher = Md5::new();
-	for (_, etag) in version.parts_etags.items().iter() {
-		etag_md5_hasher.update(etag.as_bytes());
-	}
-	let etag = format!("{}-{}", hex::encode(etag_md5_hasher.finalize()), num_parts);
-
-	// Calculate total size of final object
-	let total_size = version.blocks.items().iter().map(|x| x.1.size).sum();
-
-	if let Err(e) = check_quotas(&garage, bucket, &key, total_size).await {
-		object_version.state = ObjectVersionState::Aborted;
-		let final_object = Object::new(bucket.id, key.clone(), vec![object_version]);
-		garage.object_table.insert(&final_object).await?;
-
-		return Err(e);
-	}
-
-	// Write final object version
-	object_version.state = ObjectVersionState::Complete(ObjectVersionData::FirstBlock(
-		ObjectVersionMeta {
-			headers,
-			size: total_size,
-			etag: etag.clone(),
-		},
-		version.blocks.items()[0].1.hash,
-	));
-
-	let final_object = Object::new(bucket.id, key.clone(), vec![object_version]);
-	garage.object_table.insert(&final_object).await?;
-
-	// Send response saying ok we're done
-	let result = s3_xml::CompleteMultipartUploadResult {
-		xmlns: (),
-		location: None,
-		bucket: s3_xml::Value(bucket_name.to_string()),
-		key: s3_xml::Value(key),
-		etag: s3_xml::Value(format!("\"{}\"", etag)),
-	};
-	let xml = s3_xml::to_xml_with_header(&result)?;
-
-	Ok(Response::new(Body::from(xml.into_bytes())))
-}
-
-pub async fn handle_abort_multipart_upload(
-	garage: Arc<Garage>,
-	bucket_id: Uuid,
-	key: &str,
-	upload_id: &str,
-) -> Result<Response<Body>, Error> {
-	let version_uuid = decode_upload_id(upload_id)?;
-
-	let object = garage
-		.object_table
-		.get(&bucket_id, &key.to_string())
-		.await?;
-	let object = object.ok_or(Error::NoSuchKey)?;
-
-	let object_version = object
-		.versions()
-		.iter()
-		.find(|v| v.uuid == version_uuid && v.is_uploading());
-	let mut object_version = match object_version {
-		None => return Err(Error::NoSuchUpload),
-		Some(x) => x.clone(),
-	};
-
-	object_version.state = ObjectVersionState::Aborted;
-	let final_object = Object::new(bucket_id, key.to_string(), vec![object_version]);
-	garage.object_table.insert(&final_object).await?;
-
-	Ok(Response::new(Body::from(vec![])))
-}
-
-fn get_mime_type(headers: &HeaderMap<HeaderValue>) -> Result<String, Error> {
+pub(crate) fn get_mime_type(headers: &HeaderMap<HeaderValue>) -> Result<String, Error> {
 	Ok(headers
 		.get(hyper::header::CONTENT_TYPE)
 		.map(|x| x.to_str())
@@ -821,54 +519,3 @@ pub(crate) fn get_headers(headers: &HeaderMap<HeaderValue>) -> Result<ObjectVers
 		other,
 	})
 }
-
-pub fn decode_upload_id(id: &str) -> Result<Uuid, Error> {
-	let id_bin = hex::decode(id).map_err(|_| Error::NoSuchUpload)?;
-	if id_bin.len() != 32 {
-		return Err(Error::NoSuchUpload);
-	}
-	let mut uuid = [0u8; 32];
-	uuid.copy_from_slice(&id_bin[..]);
-	Ok(Uuid::from(uuid))
-}
-
-#[derive(Debug)]
-struct CompleteMultipartUploadPart {
-	etag: String,
-	part_number: u64,
-}
-
-fn parse_complete_multipart_upload_body(
-	xml: &roxmltree::Document,
-) -> Option<Vec<CompleteMultipartUploadPart>> {
-	let mut parts = vec![];
-
-	let root = xml.root();
-	let cmu = root.first_child()?;
-	if !cmu.has_tag_name("CompleteMultipartUpload") {
-		return None;
-	}
-
-	for item in cmu.children() {
-		// Only parse <Part> nodes
-		if !item.is_element() {
-			continue;
-		}
-
-		if item.has_tag_name("Part") {
-			let etag = item.children().find(|e| e.has_tag_name("ETag"))?.text()?;
-			let part_number = item
-				.children()
-				.find(|e| e.has_tag_name("PartNumber"))?
-				.text()?;
-			parts.push(CompleteMultipartUploadPart {
-				etag: etag.trim_matches('"').to_string(),
-				part_number: part_number.parse().ok()?,
-			});
-		} else {
-			return None;
-		}
-	}
-
-	Some(parts)
-}
@@ -80,6 +80,7 @@ pub struct BlockManager {
 	/// Directory in which block are stored
 	pub data_dir: PathBuf,

+	data_fsync: bool,
 	compression_level: Option<i32>,

 	mutation_lock: [Mutex<BlockManagerLocked>; 256],
@@ -114,6 +115,7 @@ impl BlockManager {
 	pub fn new(
 		db: &db::Db,
 		data_dir: PathBuf,
+		data_fsync: bool,
 		compression_level: Option<i32>,
 		replication: TableShardedReplication,
 		system: Arc<System>,
@@ -141,6 +143,7 @@ impl BlockManager {
 		let block_manager = Arc::new(Self {
 			replication,
 			data_dir,
+			data_fsync,
 			compression_level,
 			mutation_lock: [(); 256].map(|_| Mutex::new(BlockManagerLocked())),
 			rc,
@@ -713,7 +716,11 @@ impl BlockManagerLocked {

 		let mut f = fs::File::create(&path_tmp).await?;
 		f.write_all(data).await?;
-		f.sync_all().await?;
+
+		if mgr.data_fsync {
+			f.sync_all().await?;
+		}
+
 		drop(f);

 		fs::rename(path_tmp, path).await?;
@@ -724,18 +731,20 @@ impl BlockManagerLocked {
 			fs::remove_file(to_delete).await?;
 		}

-		// We want to ensure that when this function returns, data is properly persisted
-		// to disk. The first step is the sync_all above that does an fsync on the data file.
-		// Now, we do an fsync on the containing directory, to ensure that the rename
-		// is persisted properly. See:
-		// http://thedjbway.b0llix.net/qmail/syncdir.html
-		let dir = fs::OpenOptions::new()
-			.read(true)
-			.mode(0)
-			.open(directory)
-			.await?;
-		dir.sync_all().await?;
-		drop(dir);
+		if mgr.data_fsync {
+			// We want to ensure that when this function returns, data is properly persisted
+			// to disk. The first step is the sync_all above that does an fsync on the data file.
+			// Now, we do an fsync on the containing directory, to ensure that the rename
+			// is persisted properly. See:
+			// http://thedjbway.b0llix.net/qmail/syncdir.html
+			let dir = fs::OpenOptions::new()
+				.read(true)
+				.mode(0)
+				.open(directory)
+				.await?;
+			dir.sync_all().await?;
+			drop(dir);
+		}

 		Ok(())
 	}
@@ -33,7 +33,7 @@ pretty_env_logger = { version = "0.5", optional = true }
 mktemp = "0.5"

 [features]
-default = [ "sled" ]
+default = [ "sled", "lmdb", "sqlite" ]
 bundled-libs = [ "rusqlite?/bundled" ]
 cli = ["clap", "pretty_env_logger"]
 lmdb = [ "heed" ]
@@ -48,6 +48,8 @@ fn open_db(path: PathBuf, engine: String) -> Result<Db> {
 		}
 		"sqlite" | "sqlite3" | "rusqlite" => {
 			let db = sqlite_adapter::rusqlite::Connection::open(&path)?;
+			db.pragma_update(None, "journal_mode", &"WAL")?;
+			db.pragma_update(None, "synchronous", &"NORMAL")?;
 			Ok(sqlite_adapter::SqliteDb::init(db))
 		}
 		"lmdb" | "heed" => {
@@ -57,11 +59,13 @@ fn open_db(path: PathBuf, engine: String) -> Result<Db> {

 			let map_size = lmdb_adapter::recommended_map_size();

-			let db = lmdb_adapter::heed::EnvOpenOptions::new()
-				.max_dbs(100)
-				.map_size(map_size)
-				.open(&path)
-				.unwrap();
+			let mut env_builder = lmdb_adapter::heed::EnvOpenOptions::new();
+			env_builder.max_dbs(100);
+			env_builder.map_size(map_size);
+			unsafe {
+				env_builder.flag(heed::flags::Flags::MdbNoMetaSync);
+			}
+			let db = env_builder.open(&path)?;
 			Ok(lmdb_adapter::LmdbDb::init(db))
 		}
 		e => Err(Error(format!("Invalid DB engine: {}", e).into())),
@@ -2,9 +2,6 @@
 #[cfg(feature = "sqlite")]
 extern crate tracing;

-#[cfg(not(any(feature = "lmdb", feature = "sled", feature = "sqlite")))]
-compile_error!("Must activate the Cargo feature for at least one DB engine: lmdb, sled or sqlite.");
-
 #[cfg(feature = "lmdb")]
 pub mod lmdb_adapter;
 #[cfg(feature = "sled")]
@@ -38,7 +38,15 @@ pub struct SledDb {
 }

 impl SledDb {
+	#[deprecated(
+		since = "0.9.0",
+		note = "The Sled database is now deprecated and will be removed in Garage v1.0. Please migrate to LMDB or Sqlite as soon as possible."
+	)]
 	pub fn init(db: sled::Db) -> Db {
+		tracing::warn!("--------------------    IMPORTANT WARNING !!!    ----------------------");
+		tracing::warn!("The Sled database is now deprecated and will be removed in Garage v1.0.");
+		tracing::warn!("Please migrate to LMDB or Sqlite as soon as possible.");
+		tracing::warn!("-----------------------------------------------------------------------");
 		let s = Self {
 			db,
 			trees: RwLock::new((Vec::new(), HashMap::new())),
@@ -78,7 +78,7 @@ k2v-client.workspace = true


 [features]
-default = [ "bundled-libs", "metrics", "sled", "k2v" ]
+default = [ "bundled-libs", "metrics", "sled", "lmdb", "sqlite", "k2v" ]

 k2v = [ "garage_util/k2v", "garage_api/k2v" ]

@@ -34,6 +34,7 @@ impl AdminRpcHandler {
 			.get_range(&hash, None, None, 10000, Default::default())
 			.await?;
 		let mut versions = vec![];
+		let mut uploads = vec![];
 		for br in block_refs {
 			if let Some(v) = self
 				.garage
@@ -41,6 +42,11 @@ impl AdminRpcHandler {
 				.get(&br.version, &EmptyKey)
 				.await?
 			{
+				if let VersionBacklink::MultipartUpload { upload_id } = &v.backlink {
+					if let Some(u) = self.garage.mpu_table.get(upload_id, &EmptyKey).await? {
+						uploads.push(u);
+					}
+				}
 				versions.push(Ok(v));
 			} else {
 				versions.push(Err(br.version));
@@ -50,6 +56,7 @@ impl AdminRpcHandler {
 			hash,
 			refcount,
 			versions,
+			uploads,
 		})
 	}

@@ -93,6 +100,7 @@ impl AdminRpcHandler {
 		}

 		let mut obj_dels = 0;
+		let mut mpu_dels = 0;
 		let mut ver_dels = 0;

 		for hash in blocks {
@@ -105,56 +113,80 @@ impl AdminRpcHandler {
 				.await?;

 			for br in block_refs {
-				let version = match self
+				if let Some(version) = self
 					.garage
 					.version_table
 					.get(&br.version, &EmptyKey)
 					.await?
 				{
-					Some(v) => v,
-					None => continue,
-				};
+					self.handle_block_purge_version_backlink(
+						&version,
+						&mut obj_dels,
+						&mut mpu_dels,
+					)
+					.await?;

-				if let Some(object) = self
-					.garage
-					.object_table
-					.get(&version.bucket_id, &version.key)
-					.await?
-				{
-					let ov = object.versions().iter().rev().find(|v| v.is_complete());
-					if let Some(ov) = ov {
-						if ov.uuid == br.version {
-							let del_uuid = gen_uuid();
-							let deleted_object = Object::new(
-								version.bucket_id,
-								version.key.clone(),
-								vec![ObjectVersion {
-									uuid: del_uuid,
-									timestamp: ov.timestamp + 1,
-									state: ObjectVersionState::Complete(
-										ObjectVersionData::DeleteMarker,
-									),
-								}],
-							);
-							self.garage.object_table.insert(&deleted_object).await?;
-							obj_dels += 1;
-						}
+					if !version.deleted.get() {
+						let deleted_version = Version::new(version.uuid, version.backlink, true);
+						self.garage.version_table.insert(&deleted_version).await?;
+						ver_dels += 1;
 					}
 				}
-
-				if !version.deleted.get() {
-					let deleted_version =
-						Version::new(version.uuid, version.bucket_id, version.key.clone(), true);
-					self.garage.version_table.insert(&deleted_version).await?;
-					ver_dels += 1;
-				}
 			}
 		}
+
 		Ok(AdminRpc::Ok(format!(
-			"{} blocks were purged: {} object deletion markers added, {} versions marked deleted",
+			"Purged {} blocks, {} versions, {} objects, {} multipart uploads",
 			blocks.len(),
+			ver_dels,
 			obj_dels,
-			ver_dels
+			mpu_dels,
 		)))
 	}
+
+	async fn handle_block_purge_version_backlink(
+		&self,
+		version: &Version,
+		obj_dels: &mut usize,
+		mpu_dels: &mut usize,
+	) -> Result<(), Error> {
+		let (bucket_id, key, ov_id) = match &version.backlink {
+			VersionBacklink::Object { bucket_id, key } => (*bucket_id, key.clone(), version.uuid),
+			VersionBacklink::MultipartUpload { upload_id } => {
+				if let Some(mut mpu) = self.garage.mpu_table.get(upload_id, &EmptyKey).await? {
+					if !mpu.deleted.get() {
+						mpu.parts.clear();
+						mpu.deleted.set();
+						self.garage.mpu_table.insert(&mpu).await?;
+						*mpu_dels += 1;
+					}
+					(mpu.bucket_id, mpu.key.clone(), *upload_id)
+				} else {
+					return Ok(());
+				}
+			}
+		};
+
+		if let Some(object) = self.garage.object_table.get(&bucket_id, &key).await? {
+			let ov = object.versions().iter().rev().find(|v| v.is_complete());
+			if let Some(ov) = ov {
+				if ov.uuid == ov_id {
+					let del_uuid = gen_uuid();
+					let deleted_object = Object::new(
+						bucket_id,
+						key,
+						vec![ObjectVersion {
+							uuid: del_uuid,
+							timestamp: ov.timestamp + 1,
+							state: ObjectVersionState::Complete(ObjectVersionData::DeleteMarker),
+						}],
+					);
+					self.garage.object_table.insert(&deleted_object).await?;
+					*obj_dels += 1;
+				}
+			}
+		}
+
+		Ok(())
+	}
 }
@@ -73,6 +73,15 @@ impl AdminRpcHandler {
 			.map(|x| x.filtered_values(&self.garage.system.ring.borrow()))
 			.unwrap_or_default();

+		let mpu_counters = self
+			.garage
+			.mpu_counter_table
+			.table
+			.get(&bucket_id, &EmptyKey)
+			.await?
+			.map(|x| x.filtered_values(&self.garage.system.ring.borrow()))
+			.unwrap_or_default();
+
 		let mut relevant_keys = HashMap::new();
 		for (k, _) in bucket
 			.state
@@ -112,6 +121,7 @@ impl AdminRpcHandler {
 			bucket,
 			relevant_keys,
 			counters,
+			mpu_counters,
 		})
 	}

@@ -2,7 +2,7 @@ use std::collections::HashMap;

 use garage_table::*;

-use garage_model::helper::error::Error;
+use garage_model::helper::error::*;
 use garage_model::key_table::*;

 use crate::cli::*;
@@ -14,7 +14,7 @@ impl AdminRpcHandler {
 		match cmd {
 			KeyOperation::List => self.handle_list_keys().await,
 			KeyOperation::Info(query) => self.handle_key_info(query).await,
-			KeyOperation::New(query) => self.handle_create_key(query).await,
+			KeyOperation::Create(query) => self.handle_create_key(query).await,
 			KeyOperation::Rename(query) => self.handle_rename_key(query).await,
 			KeyOperation::Delete(query) => self.handle_delete_key(query).await,
 			KeyOperation::Allow(query) => self.handle_allow_key(query).await,
@@ -41,12 +41,17 @@ impl AdminRpcHandler {
 		Ok(AdminRpc::KeyList(key_ids))
 	}

-	async fn handle_key_info(&self, query: &KeyOpt) -> Result<AdminRpc, Error> {
-		let key = self
+	async fn handle_key_info(&self, query: &KeyInfoOpt) -> Result<AdminRpc, Error> {
+		let mut key = self
 			.garage
 			.key_helper()
 			.get_existing_matching_key(&query.key_pattern)
 			.await?;
+
+		if !query.show_secret {
+			key.state.as_option_mut().unwrap().secret_key = "(redacted)".into();
+		}
+
 		self.key_info_result(key).await
 	}

@@ -118,11 +123,17 @@ impl AdminRpcHandler {
 	}

 	async fn handle_import_key(&self, query: &KeyImportOpt) -> Result<AdminRpc, Error> {
+		if !query.yes {
+			return Err(Error::BadRequest("This command is intended to re-import keys that were previously generated by Garage. If you want to create a new key, use `garage key new` instead. Add the --yes flag if you really want to re-import a key.".to_string()));
+		}
+
 		let prev_key = self.garage.key_table.get(&EmptyKey, &query.key_id).await?;
 		if prev_key.is_some() {
 			return Err(Error::BadRequest(format!("Key {} already exists in data store. Even if it is deleted, we can't let you create a new key with the same ID. Sorry.", query.key_id)));
 		}
-		let imported_key = Key::import(&query.key_id, &query.secret_key, &query.name);
+
+		let imported_key = Key::import(&query.key_id, &query.secret_key, &query.name)
+			.ok_or_bad_request("Invalid key format")?;
 		self.garage.key_table.insert(&imported_key).await?;

 		self.key_info_result(imported_key).await
@@ -28,6 +28,7 @@ use garage_model::garage::Garage;
 use garage_model::helper::error::{Error, OkOrBadRequest};
 use garage_model::key_table::*;
 use garage_model::migrate::Migrate;
+use garage_model::s3::mpu_table::MultipartUpload;
 use garage_model::s3::version_table::Version;

 use crate::cli::*;
@@ -53,6 +54,7 @@ pub enum AdminRpc {
 		bucket: Bucket,
 		relevant_keys: HashMap<String, Key>,
 		counters: HashMap<String, i64>,
+		mpu_counters: HashMap<String, i64>,
 	},
 	KeyList(Vec<(String, String)>),
 	KeyInfo(Key, HashMap<Uuid, Bucket>),
@@ -67,6 +69,7 @@ pub enum AdminRpc {
 		hash: Hash,
 		refcount: u64,
 		versions: Vec<Result<Version, Uuid>>,
+		uploads: Vec<MultipartUpload>,
 	},
 }

@@ -274,7 +277,7 @@ impl AdminRpcHandler {
 		// Gather storage node and free space statistics
 		let layout = &self.garage.system.ring.borrow().layout;
 		let mut node_partition_count = HashMap::<Uuid, u64>::new();
-		for short_id in layout.ring_assignation_data.iter() {
+		for short_id in layout.ring_assignment_data.iter() {
 			let id = layout.node_id_vec[*short_id as usize];
 			*node_partition_count.entry(id).or_default() += 1;
 		}
@@ -85,7 +85,7 @@ pub async fn cmd_status(rpc_cli: &Endpoint<SystemRpc, ()>, rpc_host: NodeID) ->
 				));
 			}
 			_ => {
-				let new_role = match layout.staging.get(&adv.id) {
+				let new_role = match layout.staging_roles.get(&adv.id) {
 					Some(NodeRoleV(Some(_))) => "(pending)",
 					_ => "NO ROLE ASSIGNED",
 				};
@@ -190,8 +190,9 @@ pub async fn cmd_admin(
 			bucket,
 			relevant_keys,
 			counters,
+			mpu_counters,
 		} => {
-			print_bucket_info(&bucket, &relevant_keys, &counters);
+			print_bucket_info(&bucket, &relevant_keys, &counters, &mpu_counters);
 		}
 		AdminRpc::KeyList(kl) => {
 			print_key_list(kl);
@@ -215,8 +216,9 @@ pub async fn cmd_admin(
 			hash,
 			refcount,
 			versions,
+			uploads,
 		} => {
-			print_block_info(hash, refcount, versions);
+			print_block_info(hash, refcount, versions, uploads);
 		}
 		r => {
 			error!("Unexpected response: {:?}", r);
@@ -1,3 +1,5 @@
+use bytesize::ByteSize;
+
 use format_table::format_table;
 use garage_util::crdt::Crdt;
 use garage_util::error::*;
@@ -14,8 +16,8 @@ pub async fn cli_layout_command_dispatch(
 	rpc_host: NodeID,
 ) -> Result<(), Error> {
 	match cmd {
-		LayoutOperation::Assign(configure_opt) => {
-			cmd_assign_role(system_rpc_endpoint, rpc_host, configure_opt).await
+		LayoutOperation::Assign(assign_opt) => {
+			cmd_assign_role(system_rpc_endpoint, rpc_host, assign_opt).await
 		}
 		LayoutOperation::Remove(remove_opt) => {
 			cmd_remove_role(system_rpc_endpoint, rpc_host, remove_opt).await
@@ -27,6 +29,9 @@ pub async fn cli_layout_command_dispatch(
 		LayoutOperation::Revert(revert_opt) => {
 			cmd_revert_layout(system_rpc_endpoint, rpc_host, revert_opt).await
 		}
+		LayoutOperation::Config(config_opt) => {
+			cmd_config_layout(system_rpc_endpoint, rpc_host, config_opt).await
+		}
 	}
 }

@@ -60,14 +65,14 @@ pub async fn cmd_assign_role(
 		.collect::<Result<Vec<_>, _>>()?;

 	let mut roles = layout.roles.clone();
-	roles.merge(&layout.staging);
+	roles.merge(&layout.staging_roles);

 	for replaced in args.replace.iter() {
 		let replaced_node = find_matching_node(layout.node_ids().iter().cloned(), replaced)?;
 		match roles.get(&replaced_node) {
 			Some(NodeRoleV(Some(_))) => {
 				layout
-					.staging
+					.staging_roles
 					.merge(&roles.update_mutator(replaced_node, NodeRoleV(None)));
 			}
 			_ => {
@@ -83,7 +88,7 @@ pub async fn cmd_assign_role(
 		return Err(Error::Message(
 				"-c and -g are mutually exclusive, please configure node either with c>0 to act as a storage node or with -g to act as a gateway node".into()));
 	}
-	if args.capacity == Some(0) {
+	if args.capacity == Some(ByteSize::b(0)) {
 		return Err(Error::Message("Invalid capacity value: 0".into()));
 	}

@@ -91,7 +96,7 @@ pub async fn cmd_assign_role(
 		let new_entry = match roles.get(&added_node) {
 			Some(NodeRoleV(Some(old))) => {
 				let capacity = match args.capacity {
-					Some(c) => Some(c),
+					Some(c) => Some(c.as_u64()),
 					None if args.gateway => None,
 					None => old.capacity,
 				};
@@ -108,7 +113,7 @@ pub async fn cmd_assign_role(
 			}
 			_ => {
 				let capacity = match args.capacity {
-					Some(c) => Some(c),
+					Some(c) => Some(c.as_u64()),
 					None if args.gateway => None,
 					None => return Err(Error::Message(
 							"Please specify a capacity with the -c flag, or set node explicitly as gateway with -g".into())),
@@ -125,7 +130,7 @@ pub async fn cmd_assign_role(
 		};

 		layout
-			.staging
+			.staging_roles
 			.merge(&roles.update_mutator(added_node, NodeRoleV(Some(new_entry))));
 	}

@@ -145,13 +150,13 @@ pub async fn cmd_remove_role(
 	let mut layout = fetch_layout(rpc_cli, rpc_host).await?;

 	let mut roles = layout.roles.clone();
-	roles.merge(&layout.staging);
+	roles.merge(&layout.staging_roles);

 	let deleted_node =
 		find_matching_node(roles.items().iter().map(|(id, _, _)| *id), &args.node_id)?;

 	layout
-		.staging
+		.staging_roles
 		.merge(&roles.update_mutator(deleted_node, NodeRoleV(None)));

 	send_layout(rpc_cli, rpc_host, layout).await?;
@@ -166,7 +171,7 @@ pub async fn cmd_show_layout(
 	rpc_cli: &Endpoint<SystemRpc, ()>,
 	rpc_host: NodeID,
 ) -> Result<(), Error> {
-	let mut layout = fetch_layout(rpc_cli, rpc_host).await?;
+	let layout = fetch_layout(rpc_cli, rpc_host).await?;

 	println!("==== CURRENT CLUSTER LAYOUT ====");
 	if !print_cluster_layout(&layout) {
@@ -176,30 +181,41 @@ pub async fn cmd_show_layout(
 	println!();
 	println!("Current cluster layout version: {}", layout.version);

-	if print_staging_role_changes(&layout) {
-		layout.roles.merge(&layout.staging);
-
-		println!();
-		println!("==== NEW CLUSTER LAYOUT AFTER APPLYING CHANGES ====");
-		if !print_cluster_layout(&layout) {
-			println!("No nodes have a role in the new layout.");
-		}
-		println!();
+	let has_role_changes = print_staging_role_changes(&layout);
+	let has_param_changes = print_staging_parameters_changes(&layout);
+	if has_role_changes || has_param_changes {
+		let v = layout.version;
+		let res_apply = layout.apply_staged_changes(Some(v + 1));

 		// this will print the stats of what partitions
 		// will move around when we apply
-		if layout.calculate_partition_assignation() {
-			println!("To enact the staged role changes, type:");
-			println!();
-			println!("    garage layout apply --version {}", layout.version + 1);
-			println!();
-			println!(
-				"You can also revert all proposed changes with: garage layout revert --version {}",
-				layout.version + 1
-			);
-		} else {
-			println!("Not enough nodes have an assigned role to maintain enough copies of data.");
-			println!("This new layout cannot yet be applied.");
+		match res_apply {
+			Ok((layout, msg)) => {
+				println!();
+				println!("==== NEW CLUSTER LAYOUT AFTER APPLYING CHANGES ====");
+				if !print_cluster_layout(&layout) {
+					println!("No nodes have a role in the new layout.");
+				}
+				println!();
+
+				for line in msg.iter() {
+					println!("{}", line);
+				}
+				println!("To enact the staged role changes, type:");
+				println!();
+				println!("    garage layout apply --version {}", v + 1);
+				println!();
+				println!(
+                    "You can also revert all proposed changes with: garage layout revert --version {}",
+                    v + 1)
+			}
+			Err(e) => {
+				println!("Error while trying to compute the assignment: {}", e);
+				println!("This new layout cannot yet be applied.");
+				println!(
+                    "You can also revert all proposed changes with: garage layout revert --version {}",
+                    v + 1)
+			}
 		}
 	}

@@ -213,11 +229,14 @@ pub async fn cmd_apply_layout(
 ) -> Result<(), Error> {
 	let layout = fetch_layout(rpc_cli, rpc_host).await?;

-	let layout = layout.apply_staged_changes(apply_opt.version)?;
+	let (layout, msg) = layout.apply_staged_changes(apply_opt.version)?;
+	for line in msg.iter() {
+		println!("{}", line);
+	}

 	send_layout(rpc_cli, rpc_host, layout).await?;

-	println!("New cluster layout with updated role assignation has been applied in cluster.");
+	println!("New cluster layout with updated role assignment has been applied in cluster.");
 	println!("Data will now be moved around between nodes accordingly.");

 	Ok(())
@@ -238,6 +257,45 @@ pub async fn cmd_revert_layout(
 	Ok(())
 }

+pub async fn cmd_config_layout(
+	rpc_cli: &Endpoint<SystemRpc, ()>,
+	rpc_host: NodeID,
+	config_opt: ConfigLayoutOpt,
+) -> Result<(), Error> {
+	let mut layout = fetch_layout(rpc_cli, rpc_host).await?;
+
+	let mut did_something = false;
+	match config_opt.redundancy {
+		None => (),
+		Some(r) => {
+			if r > layout.replication_factor {
+				println!(
+					"The zone redundancy must be smaller or equal to the \
+                replication factor ({}).",
+					layout.replication_factor
+				);
+			} else if r < 1 {
+				println!("The zone redundancy must be at least 1.");
+			} else {
+				layout
+					.staging_parameters
+					.update(LayoutParameters { zone_redundancy: r });
+				println!("The new zone redundancy has been saved ({}).", r);
+			}
+			did_something = true;
+		}
+	}
+
+	if !did_something {
+		return Err(Error::Message(
+			"Please specify an action for `garage layout config` to do".into(),
+		));
+	}
+
+	send_layout(rpc_cli, rpc_host, layout).await?;
+	Ok(())
+}
+
 // --- utility ---

 pub async fn fetch_layout(
@@ -269,21 +327,39 @@ pub async fn send_layout(
 }

 pub fn print_cluster_layout(layout: &ClusterLayout) -> bool {
-	let mut table = vec!["ID\tTags\tZone\tCapacity".to_string()];
+	let mut table = vec!["ID\tTags\tZone\tCapacity\tUsable capacity".to_string()];
 	for (id, _, role) in layout.roles.items().iter() {
 		let role = match &role.0 {
 			Some(r) => r,
 			_ => continue,
 		};
 		let tags = role.tags.join(",");
-		table.push(format!(
-			"{:?}\t{}\t{}\t{}",
-			id,
-			tags,
-			role.zone,
-			role.capacity_string()
-		));
+		let usage = layout.get_node_usage(id).unwrap_or(0);
+		let capacity = layout.get_node_capacity(id).unwrap_or(0);
+		if capacity > 0 {
+			table.push(format!(
+				"{:?}\t{}\t{}\t{}\t{} ({:.1}%)",
+				id,
+				tags,
+				role.zone,
+				role.capacity_string(),
+				ByteSize::b(usage as u64 * layout.partition_size).to_string_as(false),
+				(100.0 * usage as f32 * layout.partition_size as f32) / (capacity as f32)
+			));
+		} else {
+			table.push(format!(
+				"{:?}\t{}\t{}\t{}",
+				id,
+				tags,
+				role.zone,
+				role.capacity_string()
+			));
+		};
 	}
+	println!();
+	println!("Parameters of the layout computation:");
+	println!("Zone redundancy: {}", layout.parameters.zone_redundancy);
+	println!();
 	if table.len() == 1 {
 		false
 	} else {
@@ -292,9 +368,23 @@ pub fn print_cluster_layout(layout: &ClusterLayout) -> bool {
 	}
 }

+pub fn print_staging_parameters_changes(layout: &ClusterLayout) -> bool {
+	let has_changes = *layout.staging_parameters.get() != layout.parameters;
+	if has_changes {
+		println!();
+		println!("==== NEW LAYOUT PARAMETERS ====");
+		println!(
+			"Zone redundancy: {}",
+			layout.staging_parameters.get().zone_redundancy
+		);
+		println!();
+	}
+	has_changes
+}
+
 pub fn print_staging_role_changes(layout: &ClusterLayout) -> bool {
 	let has_changes = layout
-		.staging
+		.staging_roles
 		.items()
 		.iter()
 		.any(|(k, _, v)| layout.roles.get(k) != Some(v));
@@ -303,7 +393,7 @@ pub fn print_staging_role_changes(layout: &ClusterLayout) -> bool {
 		println!();
 		println!("==== STAGED ROLE CHANGES ====");
 		let mut table = vec!["ID\tTags\tZone\tCapacity".to_string()];
-		for (id, _, role) in layout.staging.items().iter() {
+		for (id, _, role) in layout.staging_roles.items().iter() {
 			if layout.roles.get(id) == Some(role) {
 				continue;
 			}
@@ -17,7 +17,7 @@ pub enum Command {
 	#[structopt(name = "node", version = garage_version())]
 	Node(NodeOperation),

-	/// Operations on the assignation of node roles in the cluster layout
+	/// Operations on the assignment of node roles in the cluster layout
 	#[structopt(name = "layout", version = garage_version())]
 	Layout(LayoutOperation),

@@ -91,6 +91,10 @@ pub enum LayoutOperation {
 	#[structopt(name = "remove", version = garage_version())]
 	Remove(RemoveRoleOpt),

+	/// Configure parameters value for the layout computation
+	#[structopt(name = "config", version = garage_version())]
+	Config(ConfigLayoutOpt),
+
 	/// Show roles currently assigned to nodes and changes staged for commit
 	#[structopt(name = "show", version = garage_version())]
 	Show,
@@ -114,9 +118,9 @@ pub struct AssignRoleOpt {
 	#[structopt(short = "z", long = "zone")]
 	pub(crate) zone: Option<String>,

-	/// Capacity (in relative terms, use 1 to represent your smallest server)
+	/// Storage capacity, in bytes (supported suffixes: B, KB, MB, GB, TB, PB)
 	#[structopt(short = "c", long = "capacity")]
-	pub(crate) capacity: Option<u32>,
+	pub(crate) capacity: Option<bytesize::ByteSize>,

 	/// Gateway-only node
 	#[structopt(short = "g", long = "gateway")]
@@ -137,6 +141,13 @@ pub struct RemoveRoleOpt {
 	pub(crate) node_id: String,
 }

+#[derive(StructOpt, Debug)]
+pub struct ConfigLayoutOpt {
+	/// Zone redundancy parameter
+	#[structopt(short = "r", long = "redundancy")]
+	pub(crate) redundancy: Option<usize>,
+}
+
 #[derive(StructOpt, Debug)]
 pub struct ApplyLayoutOpt {
 	/// Version number of new configuration: this command will fail if
@@ -317,11 +328,11 @@ pub enum KeyOperation {

 	/// Get key info
 	#[structopt(name = "info", version = garage_version())]
-	Info(KeyOpt),
+	Info(KeyInfoOpt),

 	/// Create new key
-	#[structopt(name = "new", version = garage_version())]
-	New(KeyNewOpt),
+	#[structopt(name = "create", version = garage_version())]
+	Create(KeyNewOpt),

 	/// Rename key
 	#[structopt(name = "rename", version = garage_version())]
@@ -345,15 +356,18 @@ pub enum KeyOperation {
 }

 #[derive(Serialize, Deserialize, StructOpt, Debug)]
-pub struct KeyOpt {
+pub struct KeyInfoOpt {
 	/// ID or name of the key
 	pub key_pattern: String,
+	/// Whether to display the secret key
+	#[structopt(long = "show-secret")]
+	pub show_secret: bool,
 }

 #[derive(Serialize, Deserialize, StructOpt, Debug)]
 pub struct KeyNewOpt {
 	/// Name of the key
-	#[structopt(long = "name", default_value = "Unnamed key")]
+	#[structopt(default_value = "Unnamed key")]
 	pub name: String,
 }

@@ -397,6 +411,10 @@ pub struct KeyImportOpt {
 	/// Key name
 	#[structopt(short = "n", default_value = "Imported key")]
 	pub name: String,
+
+	/// Confirm key import
+	#[structopt(long = "yes")]
+	pub yes: bool,
 }

 #[derive(Serialize, Deserialize, StructOpt, Debug, Clone)]
@@ -432,19 +450,22 @@ pub struct RepairOpt {

 #[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)]
 pub enum RepairWhat {
-	/// Only do a full sync of metadata tables
+	/// Do a full sync of metadata tables
 	#[structopt(name = "tables", version = garage_version())]
 	Tables,
-	/// Only repair (resync/rebalance) the set of stored blocks
+	/// Repair (resync/rebalance) the set of stored blocks
 	#[structopt(name = "blocks", version = garage_version())]
 	Blocks,
-	/// Only redo the propagation of object deletions to the version table (slow)
+	/// Repropagate object deletions to the version table
 	#[structopt(name = "versions", version = garage_version())]
 	Versions,
-	/// Only redo the propagation of version deletions to the block ref table (extremely slow)
+	/// Repropagate object deletions to the multipart upload table
+	#[structopt(name = "mpu", version = garage_version())]
+	MultipartUploads,
+	/// Repropagate version deletions to the block ref table
 	#[structopt(name = "block_refs", version = garage_version())]
 	BlockRefs,
-	/// Verify integrity of all blocks on disc (extremely slow, i/o intensive)
+	/// Verify integrity of all blocks on disc
 	#[structopt(name = "scrub", version = garage_version())]
 	Scrub {
 		#[structopt(subcommand)]
@@ -12,8 +12,9 @@ use garage_block::manager::BlockResyncErrorInfo;

 use garage_model::bucket_table::*;
 use garage_model::key_table::*;
-use garage_model::s3::object_table::{BYTES, OBJECTS, UNFINISHED_UPLOADS};
-use garage_model::s3::version_table::Version;
+use garage_model::s3::mpu_table::{self, MultipartUpload};
+use garage_model::s3::object_table;
+use garage_model::s3::version_table::*;

 use crate::cli::structs::WorkerListOpt;

@@ -135,6 +136,7 @@ pub fn print_bucket_info(
 	bucket: &Bucket,
 	relevant_keys: &HashMap<String, Key>,
 	counters: &HashMap<String, i64>,
+	mpu_counters: &HashMap<String, i64>,
 ) {
 	let key_name = |k| {
 		relevant_keys
@@ -148,7 +150,7 @@ pub fn print_bucket_info(
 		Deletable::Deleted => println!("Bucket is deleted."),
 		Deletable::Present(p) => {
 			let size =
-				bytesize::ByteSize::b(counters.get(BYTES).cloned().unwrap_or_default() as u64);
+				bytesize::ByteSize::b(*counters.get(object_table::BYTES).unwrap_or(&0) as u64);
 			println!(
 				"\nSize: {} ({})",
 				size.to_string_as(true),
@@ -156,14 +158,22 @@ pub fn print_bucket_info(
 			);
 			println!(
 				"Objects: {}",
-				counters.get(OBJECTS).cloned().unwrap_or_default()
+				*counters.get(object_table::OBJECTS).unwrap_or(&0)
+			);
+			println!(
+				"Unfinished uploads (multipart and non-multipart): {}",
+				*counters.get(object_table::UNFINISHED_UPLOADS).unwrap_or(&0)
 			);
 			println!(
 				"Unfinished multipart uploads: {}",
-				counters
-					.get(UNFINISHED_UPLOADS)
-					.cloned()
-					.unwrap_or_default()
+				*mpu_counters.get(mpu_table::UPLOADS).unwrap_or(&0)
+			);
+			let mpu_size =
+				bytesize::ByteSize::b(*mpu_counters.get(mpu_table::BYTES).unwrap_or(&0) as u64);
+			println!(
+				"Size of unfinished multipart uploads: {} ({})",
+				mpu_size.to_string_as(true),
+				mpu_size.to_string_as(false),
 			);

 			println!("\nWebsite access: {}", p.website_config.get().is_some());
@@ -390,29 +400,49 @@ pub fn print_block_error_list(el: Vec<BlockResyncErrorInfo>) {
 	format_table(table);
 }

-pub fn print_block_info(hash: Hash, refcount: u64, versions: Vec<Result<Version, Uuid>>) {
+pub fn print_block_info(
+	hash: Hash,
+	refcount: u64,
+	versions: Vec<Result<Version, Uuid>>,
+	uploads: Vec<MultipartUpload>,
+) {
 	println!("Block hash: {}", hex::encode(hash.as_slice()));
 	println!("Refcount: {}", refcount);
 	println!();

-	let mut table = vec!["Version\tBucket\tKey\tDeleted".into()];
+	let mut table = vec!["Version\tBucket\tKey\tMPU\tDeleted".into()];
 	let mut nondeleted_count = 0;
 	for v in versions.iter() {
 		match v {
 			Ok(ver) => {
-				table.push(format!(
-					"{:?}\t{:?}\t{}\t{:?}",
-					ver.uuid,
-					ver.bucket_id,
-					ver.key,
-					ver.deleted.get()
-				));
+				match &ver.backlink {
+					VersionBacklink::Object { bucket_id, key } => {
+						table.push(format!(
+							"{:?}\t{:?}\t{}\t\t{:?}",
+							ver.uuid,
+							bucket_id,
+							key,
+							ver.deleted.get()
+						));
+					}
+					VersionBacklink::MultipartUpload { upload_id } => {
+						let upload = uploads.iter().find(|x| x.upload_id == *upload_id);
+						table.push(format!(
+							"{:?}\t{:?}\t{}\t{:?}\t{:?}",
+							ver.uuid,
+							upload.map(|u| u.bucket_id).unwrap_or_default(),
+							upload.map(|u| u.key.as_str()).unwrap_or_default(),
+							upload_id,
+							ver.deleted.get()
+						));
+					}
+				}
 				if !ver.deleted.get() {
 					nondeleted_count += 1;
 				}
 			}
 			Err(vh) => {
-				table.push(format!("{:?}\t\t\tyes", vh));
+				table.push(format!("{:?}\t\t\t\tyes", vh));
 			}
 		}
 	}
@@ -17,6 +17,9 @@ compile_error!("Either bundled-libs or system-libs Cargo feature must be enabled
 #[cfg(all(feature = "bundled-libs", feature = "system-libs"))]
 compile_error!("Only one of bundled-libs and system-libs Cargo features must be enabled");

+#[cfg(not(any(feature = "lmdb", feature = "sled", feature = "sqlite")))]
+compile_error!("Must activate the Cargo feature for at least one DB engine: lmdb, sled or sqlite.");
+
 use std::net::SocketAddr;
 use std::path::PathBuf;

@@ -5,11 +5,16 @@ use async_trait::async_trait;
 use tokio::sync::watch;

 use garage_block::repair::ScrubWorkerCommand;
+
 use garage_model::garage::Garage;
 use garage_model::s3::block_ref_table::*;
+use garage_model::s3::mpu_table::*;
 use garage_model::s3::object_table::*;
 use garage_model::s3::version_table::*;
+
+use garage_table::replication::*;
 use garage_table::*;
+
 use garage_util::background::*;
 use garage_util::error::Error;
 use garage_util::migrate::Migrate;
@@ -32,11 +37,15 @@ pub async fn launch_online_repair(
 		}
 		RepairWhat::Versions => {
 			info!("Repairing the versions table");
-			bg.spawn_worker(RepairVersionsWorker::new(garage.clone()));
+			bg.spawn_worker(TableRepairWorker::new(garage.clone(), RepairVersions));
+		}
+		RepairWhat::MultipartUploads => {
+			info!("Repairing the multipart uploads table");
+			bg.spawn_worker(TableRepairWorker::new(garage.clone(), RepairMpu));
 		}
 		RepairWhat::BlockRefs => {
 			info!("Repairing the block refs table");
-			bg.spawn_worker(RepairBlockrefsWorker::new(garage.clone()));
+			bg.spawn_worker(TableRepairWorker::new(garage.clone(), RepairBlockRefs));
 		}
 		RepairWhat::Blocks => {
 			info!("Repairing the stored blocks");
@@ -67,70 +76,70 @@ pub async fn launch_online_repair(

 // ----

-struct RepairVersionsWorker {
+#[async_trait]
+trait TableRepair: Send + Sync + 'static {
+	type T: TableSchema;
+
+	fn table(garage: &Garage) -> &Table<Self::T, TableShardedReplication>;
+
+	async fn process(
+		&mut self,
+		garage: &Garage,
+		entry: <<Self as TableRepair>::T as TableSchema>::E,
+	) -> Result<bool, Error>;
+}
+
+struct TableRepairWorker<T: TableRepair> {
 	garage: Arc<Garage>,
 	pos: Vec<u8>,
 	counter: usize,
+	repairs: usize,
+	inner: T,
 }

-impl RepairVersionsWorker {
-	fn new(garage: Arc<Garage>) -> Self {
+impl<R: TableRepair> TableRepairWorker<R> {
+	fn new(garage: Arc<Garage>, inner: R) -> Self {
 		Self {
 			garage,
+			inner,
 			pos: vec![],
 			counter: 0,
+			repairs: 0,
 		}
 	}
 }

 #[async_trait]
-impl Worker for RepairVersionsWorker {
+impl<R: TableRepair> Worker for TableRepairWorker<R> {
 	fn name(&self) -> String {
-		"Version repair worker".into()
+		format!("{} repair worker", R::T::TABLE_NAME)
 	}

 	fn status(&self) -> WorkerStatus {
 		WorkerStatus {
-			progress: Some(self.counter.to_string()),
+			progress: Some(format!("{} ({})", self.counter, self.repairs)),
 			..Default::default()
 		}
 	}

 	async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> {
-		let (item_bytes, next_pos) = match self.garage.version_table.data.store.get_gt(&self.pos)? {
+		let (item_bytes, next_pos) = match R::table(&self.garage).data.store.get_gt(&self.pos)? {
 			Some((k, v)) => (v, k),
 			None => {
-				info!("repair_versions: finished, done {}", self.counter);
+				info!(
+					"{}: finished, done {}, fixed {}",
+					self.name(),
+					self.counter,
+					self.repairs
+				);
 				return Ok(WorkerState::Done);
 			}
 		};

-		let version = Version::decode(&item_bytes).ok_or_message("Cannot decode Version")?;
-		if !version.deleted.get() {
-			let object = self
-				.garage
-				.object_table
-				.get(&version.bucket_id, &version.key)
-				.await?;
-			let version_exists = match object {
-				Some(o) => o
-					.versions()
-					.iter()
-					.any(|x| x.uuid == version.uuid && x.state != ObjectVersionState::Aborted),
-				None => false,
-			};
-			if !version_exists {
-				info!("Repair versions: marking version as deleted: {:?}", version);
-				self.garage
-					.version_table
-					.insert(&Version::new(
-						version.uuid,
-						version.bucket_id,
-						version.key,
-						true,
-					))
-					.await?;
-			}
+		let entry = <R::T as TableSchema>::E::decode(&item_bytes)
+			.ok_or_message("Cannot decode table entry")?;
+		if self.inner.process(&self.garage, entry).await? {
+			self.repairs += 1;
 		}

 		self.counter += 1;
@@ -146,77 +155,124 @@ impl Worker for RepairVersionsWorker {

 // ----

-struct RepairBlockrefsWorker {
-	garage: Arc<Garage>,
-	pos: Vec<u8>,
-	counter: usize,
-}
-
-impl RepairBlockrefsWorker {
-	fn new(garage: Arc<Garage>) -> Self {
-		Self {
-			garage,
-			pos: vec![],
-			counter: 0,
-		}
-	}
-}
+struct RepairVersions;

 #[async_trait]
-impl Worker for RepairBlockrefsWorker {
-	fn name(&self) -> String {
-		"Block refs repair worker".into()
+impl TableRepair for RepairVersions {
+	type T = VersionTable;
+
+	fn table(garage: &Garage) -> &Table<Self::T, TableShardedReplication> {
+		&garage.version_table
 	}

-	fn status(&self) -> WorkerStatus {
-		WorkerStatus {
-			progress: Some(self.counter.to_string()),
-			..Default::default()
-		}
-	}
-
-	async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> {
-		let (item_bytes, next_pos) =
-			match self.garage.block_ref_table.data.store.get_gt(&self.pos)? {
-				Some((k, v)) => (v, k),
-				None => {
-					info!("repair_block_ref: finished, done {}", self.counter);
-					return Ok(WorkerState::Done);
-				}
+	async fn process(&mut self, garage: &Garage, version: Version) -> Result<bool, Error> {
+		if !version.deleted.get() {
+			let ref_exists = match &version.backlink {
+				VersionBacklink::Object { bucket_id, key } => garage
+					.object_table
+					.get(bucket_id, key)
+					.await?
+					.map(|o| {
+						o.versions().iter().any(|x| {
+							x.uuid == version.uuid && x.state != ObjectVersionState::Aborted
+						})
+					})
+					.unwrap_or(false),
+				VersionBacklink::MultipartUpload { upload_id } => garage
+					.mpu_table
+					.get(upload_id, &EmptyKey)
+					.await?
+					.map(|u| !u.deleted.get())
+					.unwrap_or(false),
 			};

-		let block_ref = BlockRef::decode(&item_bytes).ok_or_message("Cannot decode BlockRef")?;
+			if !ref_exists {
+				info!("Repair versions: marking version as deleted: {:?}", version);
+				garage
+					.version_table
+					.insert(&Version::new(version.uuid, version.backlink, true))
+					.await?;
+				return Ok(true);
+			}
+		}
+
+		Ok(false)
+	}
+}
+
+// ----
+
+struct RepairBlockRefs;
+
+#[async_trait]
+impl TableRepair for RepairBlockRefs {
+	type T = BlockRefTable;
+
+	fn table(garage: &Garage) -> &Table<Self::T, TableShardedReplication> {
+		&garage.block_ref_table
+	}
+
+	async fn process(&mut self, garage: &Garage, mut block_ref: BlockRef) -> Result<bool, Error> {
 		if !block_ref.deleted.get() {
-			let version = self
-				.garage
+			let ref_exists = garage
 				.version_table
 				.get(&block_ref.version, &EmptyKey)
-				.await?;
-			// The version might not exist if it has been GC'ed
-			let ref_exists = version.map(|v| !v.deleted.get()).unwrap_or(false);
+				.await?
+				.map(|v| !v.deleted.get())
+				.unwrap_or(false);
+
 			if !ref_exists {
 				info!(
 					"Repair block ref: marking block_ref as deleted: {:?}",
 					block_ref
 				);
-				self.garage
-					.block_ref_table
-					.insert(&BlockRef {
-						block: block_ref.block,
-						version: block_ref.version,
-						deleted: true.into(),
-					})
-					.await?;
+				block_ref.deleted.set();
+				garage.block_ref_table.insert(&block_ref).await?;
+				return Ok(true);
 			}
 		}

-		self.counter += 1;
-		self.pos = next_pos;
-
-		Ok(WorkerState::Busy)
-	}
-
-	async fn wait_for_work(&mut self) -> WorkerState {
-		unreachable!()
+		Ok(false)
+	}
+}
+
+// ----
+
+struct RepairMpu;
+
+#[async_trait]
+impl TableRepair for RepairMpu {
+	type T = MultipartUploadTable;
+
+	fn table(garage: &Garage) -> &Table<Self::T, TableShardedReplication> {
+		&garage.mpu_table
+	}
+
+	async fn process(&mut self, garage: &Garage, mut mpu: MultipartUpload) -> Result<bool, Error> {
+		if !mpu.deleted.get() {
+			let ref_exists = garage
+				.object_table
+				.get(&mpu.bucket_id, &mpu.key)
+				.await?
+				.map(|o| {
+					o.versions()
+						.iter()
+						.any(|x| x.uuid == mpu.upload_id && x.is_uploading(Some(true)))
+				})
+				.unwrap_or(false);
+
+			if !ref_exists {
+				info!(
+					"Repair multipart uploads: marking mpu as deleted: {:?}",
+					mpu
+				);
+				mpu.parts.clear();
+				mpu.deleted.set();
+				garage.mpu_table.insert(&mpu).await?;
+				return Ok(true);
+			}
+		}
+
+		Ok(false)
 	}
 }
@@ -52,6 +52,7 @@ impl Instance {
 			r#"
 metadata_dir = "{path}/meta"
 data_dir = "{path}/data"
+db_engine = "sled"

 replication_mode = "1"

@@ -141,7 +142,7 @@ api_bind_addr = "127.0.0.1:{admin_port}"
 		self.command()
 			.args(["layout", "assign"])
 			.arg(node_short_id)
-			.args(["-c", "1", "-z", "unzonned"])
+			.args(["-c", "1G", "-z", "unzonned"])
 			.quiet()
 			.expect_success_status("Could not assign garage node layout");
 		self.command()
@@ -186,9 +187,9 @@ api_bind_addr = "127.0.0.1:{admin_port}"
 		let mut key = Key::default();

 		let mut cmd = self.command();
-		let base = cmd.args(["key", "new"]);
+		let base = cmd.args(["key", "create"]);
 		let with_name = match maybe_name {
-			Some(name) => base.args(["--name", name]),
+			Some(name) => base.args([name]),
 			None => base,
 		};

@@ -5,6 +5,190 @@ use aws_sdk_s3::types::{CompletedMultipartUpload, CompletedPart};
 const SZ_5MB: usize = 5 * 1024 * 1024;
 const SZ_10MB: usize = 10 * 1024 * 1024;

+#[tokio::test]
+async fn test_multipart_upload() {
+	let ctx = common::context();
+	let bucket = ctx.create_bucket("testmpu");
+
+	let u1 = vec![0x11; SZ_5MB];
+	let u2 = vec![0x22; SZ_5MB];
+	let u3 = vec![0x33; SZ_5MB];
+	let u4 = vec![0x44; SZ_5MB];
+	let u5 = vec![0x55; SZ_5MB];
+
+	let up = ctx
+		.client
+		.create_multipart_upload()
+		.bucket(&bucket)
+		.key("a")
+		.send()
+		.await
+		.unwrap();
+	assert!(up.upload_id.is_some());
+
+	let uid = up.upload_id.as_ref().unwrap();
+
+	let p3 = ctx
+		.client
+		.upload_part()
+		.bucket(&bucket)
+		.key("a")
+		.upload_id(uid)
+		.part_number(3)
+		.body(ByteStream::from(u3.clone()))
+		.send()
+		.await
+		.unwrap();
+
+	let _p1 = ctx
+		.client
+		.upload_part()
+		.bucket(&bucket)
+		.key("a")
+		.upload_id(uid)
+		.part_number(1)
+		.body(ByteStream::from(u1))
+		.send()
+		.await
+		.unwrap();
+
+	let _p4 = ctx
+		.client
+		.upload_part()
+		.bucket(&bucket)
+		.key("a")
+		.upload_id(uid)
+		.part_number(4)
+		.body(ByteStream::from(u4))
+		.send()
+		.await
+		.unwrap();
+
+	let p1bis = ctx
+		.client
+		.upload_part()
+		.bucket(&bucket)
+		.key("a")
+		.upload_id(uid)
+		.part_number(1)
+		.body(ByteStream::from(u2.clone()))
+		.send()
+		.await
+		.unwrap();
+
+	let p6 = ctx
+		.client
+		.upload_part()
+		.bucket(&bucket)
+		.key("a")
+		.upload_id(uid)
+		.part_number(6)
+		.body(ByteStream::from(u5.clone()))
+		.send()
+		.await
+		.unwrap();
+
+	{
+		let r = ctx
+			.client
+			.list_parts()
+			.bucket(&bucket)
+			.key("a")
+			.upload_id(uid)
+			.send()
+			.await
+			.unwrap();
+		assert_eq!(r.parts.unwrap().len(), 4);
+	}
+
+	let cmp = CompletedMultipartUpload::builder()
+		.parts(
+			CompletedPart::builder()
+				.part_number(1)
+				.e_tag(p1bis.e_tag.unwrap())
+				.build(),
+		)
+		.parts(
+			CompletedPart::builder()
+				.part_number(3)
+				.e_tag(p3.e_tag.unwrap())
+				.build(),
+		)
+		.parts(
+			CompletedPart::builder()
+				.part_number(6)
+				.e_tag(p6.e_tag.unwrap())
+				.build(),
+		)
+		.build();
+
+	ctx.client
+		.complete_multipart_upload()
+		.bucket(&bucket)
+		.key("a")
+		.upload_id(uid)
+		.multipart_upload(cmp)
+		.send()
+		.await
+		.unwrap();
+
+	// The multipart upload must not appear anymore
+	assert!(ctx
+		.client
+		.list_parts()
+		.bucket(&bucket)
+		.key("a")
+		.upload_id(uid)
+		.send()
+		.await
+		.is_err());
+
+	{
+		// The object must appear as a regular object
+		let r = ctx
+			.client
+			.head_object()
+			.bucket(&bucket)
+			.key("a")
+			.send()
+			.await
+			.unwrap();
+
+		assert_eq!(r.content_length, (SZ_5MB * 3) as i64);
+	}
+
+	{
+		let o = ctx
+			.client
+			.get_object()
+			.bucket(&bucket)
+			.key("a")
+			.send()
+			.await
+			.unwrap();
+
+		assert_bytes_eq!(o.body, &[&u2[..], &u3[..], &u5[..]].concat());
+	}
+
+	{
+		for (part_number, data) in [(1, &u2), (2, &u3), (3, &u5)] {
+			let o = ctx
+				.client
+				.get_object()
+				.bucket(&bucket)
+				.key("a")
+				.part_number(part_number)
+				.send()
+				.await
+				.unwrap();
+
+			eprintln!("get_object with part_number = {}", part_number);
+			assert_eq!(o.content_length, SZ_5MB as i64);
+			assert_bytes_eq!(o.body, data);
+		}
+	}
+}
+
 #[tokio::test]
 async fn test_uploadlistpart() {
 	let ctx = common::context();
@@ -65,7 +249,8 @@ async fn test_uploadlistpart() {

 		let ps = r.parts.unwrap();
 		assert_eq!(ps.len(), 1);
-		let fp = ps.iter().find(|x| x.part_number == 2).unwrap();
+		assert_eq!(ps[0].part_number, 2);
+		let fp = &ps[0];
 		assert!(fp.last_modified.is_some());
 		assert_eq!(
 			fp.e_tag.as_ref().unwrap(),
@@ -100,13 +285,24 @@ async fn test_uploadlistpart() {

 		let ps = r.parts.unwrap();
 		assert_eq!(ps.len(), 2);
-		let fp = ps.iter().find(|x| x.part_number == 1).unwrap();
+
+		assert_eq!(ps[0].part_number, 1);
+		let fp = &ps[0];
 		assert!(fp.last_modified.is_some());
 		assert_eq!(
 			fp.e_tag.as_ref().unwrap(),
 			"\"3c484266f9315485694556e6c693bfa2\""
 		);
 		assert_eq!(fp.size, SZ_5MB as i64);
+
+		assert_eq!(ps[1].part_number, 2);
+		let sp = &ps[1];
+		assert!(sp.last_modified.is_some());
+		assert_eq!(
+			sp.e_tag.as_ref().unwrap(),
+			"\"3366bb9dcf710d6801b5926467d02e19\""
+		);
+		assert_eq!(sp.size, SZ_5MB as i64);
 	}

 	{
@@ -123,12 +319,19 @@ async fn test_uploadlistpart() {
 			.unwrap();

 		assert!(r.part_number_marker.is_none());
-		assert!(r.next_part_number_marker.is_some());
+		assert_eq!(r.next_part_number_marker.as_deref(), Some("1"));
 		assert_eq!(r.max_parts, 1_i32);
 		assert!(r.is_truncated);
 		assert_eq!(r.key.unwrap(), "a");
 		assert_eq!(r.upload_id.unwrap().as_str(), uid.as_str());
-		assert_eq!(r.parts.unwrap().len(), 1);
+		let parts = r.parts.unwrap();
+		assert_eq!(parts.len(), 1);
+		let fp = &parts[0];
+		assert_eq!(fp.part_number, 1);
+		assert_eq!(
+			fp.e_tag.as_ref().unwrap(),
+			"\"3c484266f9315485694556e6c693bfa2\""
+		);

 		let r2 = ctx
 			.client
@@ -147,10 +350,18 @@ async fn test_uploadlistpart() {
 			r.next_part_number_marker.as_ref().unwrap()
 		);
 		assert_eq!(r2.max_parts, 1_i32);
-		assert!(r2.is_truncated);
 		assert_eq!(r2.key.unwrap(), "a");
 		assert_eq!(r2.upload_id.unwrap().as_str(), uid.as_str());
-		assert_eq!(r2.parts.unwrap().len(), 1);
+		let parts = r2.parts.unwrap();
+		assert_eq!(parts.len(), 1);
+		let fp = &parts[0];
+		assert_eq!(fp.part_number, 2);
+		assert_eq!(
+			fp.e_tag.as_ref().unwrap(),
+			"\"3366bb9dcf710d6801b5926467d02e19\""
+		);
+		//assert!(r2.is_truncated);   // WHY? (this was the test before)
+		assert!(!r2.is_truncated);
 	}

 	let cmp = CompletedMultipartUpload::builder()
@@ -41,7 +41,7 @@ opentelemetry = "0.17"
 netapp = "0.5"

 [features]
-default = [ "sled" ]
+default = [ "sled", "lmdb", "sqlite" ]
 k2v = [ "garage_util/k2v" ]
 lmdb = [ "garage_db/lmdb" ]
 sled = [ "garage_db/sled" ]
@@ -17,6 +17,7 @@ use garage_table::replication::TableShardedReplication;
 use garage_table::*;

 use crate::s3::block_ref_table::*;
+use crate::s3::mpu_table::*;
 use crate::s3::object_table::*;
 use crate::s3::version_table::*;

@@ -57,6 +58,10 @@ pub struct Garage {
 	pub object_table: Arc<Table<ObjectTable, TableShardedReplication>>,
 	/// Counting table containing object counters
 	pub object_counter_table: Arc<IndexCounter<Object>>,
+	/// Table containing S3 multipart uploads
+	pub mpu_table: Arc<Table<MultipartUploadTable, TableShardedReplication>>,
+	/// Counting table containing multipart object counters
+	pub mpu_counter_table: Arc<IndexCounter<MultipartUpload>>,
 	/// Table containing S3 object versions
 	pub version_table: Arc<Table<VersionTable, TableShardedReplication>>,
 	/// Table containing S3 block references (not blocks themselves)
@@ -91,6 +96,11 @@ impl Garage {
 			// ---- Sled DB ----
 			#[cfg(feature = "sled")]
 			"sled" => {
+				if config.metadata_fsync {
+					return Err(Error::Message(format!(
+						"`metadata_fsync = true` is not supported with the Sled database engine"
+					)));
+				}
 				db_path.push("db");
 				info!("Opening Sled database at: {}", db_path.display());
 				let db = db::sled_adapter::sled::Config::default()
@@ -109,6 +119,15 @@ impl Garage {
 				db_path.push("db.sqlite");
 				info!("Opening Sqlite database at: {}", db_path.display());
 				let db = db::sqlite_adapter::rusqlite::Connection::open(db_path)
+					.and_then(|db| {
+						db.pragma_update(None, "journal_mode", &"WAL")?;
+						if config.metadata_fsync {
+							db.pragma_update(None, "synchronous", &"NORMAL")?;
+						} else {
+							db.pragma_update(None, "synchronous", &"OFF")?;
+						}
+						Ok(db)
+					})
 					.ok_or_message("Unable to open sqlite DB")?;
 				db::sqlite_adapter::SqliteDb::init(db)
 			}
@@ -133,8 +152,10 @@ impl Garage {
 				env_builder.max_readers(500);
 				env_builder.map_size(map_size);
 				unsafe {
-					env_builder.flag(heed::flags::Flags::MdbNoSync);
 					env_builder.flag(heed::flags::Flags::MdbNoMetaSync);
+					if !config.metadata_fsync {
+						env_builder.flag(heed::flags::Flags::MdbNoSync);
+					}
 				}
 				let db = match env_builder.open(&db_path) {
 					Err(heed::Error::Io(e)) if e.kind() == std::io::ErrorKind::OutOfMemory => {
@@ -204,6 +225,7 @@ impl Garage {
 		let block_manager = BlockManager::new(
 			&db,
 			config.data_dir.clone(),
+			config.data_fsync,
 			config.compression_level,
 			data_rep_param,
 			system.clone(),
@@ -244,6 +266,20 @@ impl Garage {
 			&db,
 		);

+		info!("Initialize multipart upload counter table...");
+		let mpu_counter_table = IndexCounter::new(system.clone(), meta_rep_param.clone(), &db);
+
+		info!("Initialize multipart upload table...");
+		let mpu_table = Table::new(
+			MultipartUploadTable {
+				version_table: version_table.clone(),
+				mpu_counter_table: mpu_counter_table.clone(),
+			},
+			meta_rep_param.clone(),
+			system.clone(),
+			&db,
+		);
+
 		info!("Initialize object counter table...");
 		let object_counter_table = IndexCounter::new(system.clone(), meta_rep_param.clone(), &db);

@@ -252,6 +288,7 @@ impl Garage {
 		let object_table = Table::new(
 			ObjectTable {
 				version_table: version_table.clone(),
+				mpu_table: mpu_table.clone(),
 				object_counter_table: object_counter_table.clone(),
 			},
 			meta_rep_param.clone(),
@@ -280,6 +317,8 @@ impl Garage {
 			key_table,
 			object_table,
 			object_counter_table,
+			mpu_table,
+			mpu_counter_table,
 			version_table,
 			block_ref_table,
 			#[cfg(feature = "k2v")]
@@ -296,6 +335,8 @@ impl Garage {

 		self.object_table.spawn_workers(bg);
 		self.object_counter_table.spawn_workers(bg);
+		self.mpu_table.spawn_workers(bg);
+		self.mpu_counter_table.spawn_workers(bg);
 		self.version_table.spawn_workers(bg);
 		self.block_ref_table.spawn_workers(bg);

@@ -478,7 +478,9 @@ impl<'a> BucketHelper<'a> {
 	// ----

 	/// Deletes all incomplete multipart uploads that are older than a certain time.
-	/// Returns the number of uploads aborted
+	/// Returns the number of uploads aborted.
+	/// This will also include non-multipart uploads, which may be lingering
+	/// after a node crash
 	pub async fn cleanup_incomplete_uploads(
 		&self,
 		bucket_id: &Uuid,
@@ -496,7 +498,9 @@ impl<'a> BucketHelper<'a> {
 				.get_range(
 					bucket_id,
 					start,
-					Some(ObjectFilter::IsUploading),
+					Some(ObjectFilter::IsUploading {
+						check_multipart: None,
+					}),
 					1000,
 					EnumerationOrder::Forward,
 				)
@@ -508,7 +512,7 @@ impl<'a> BucketHelper<'a> {
 					let aborted_versions = object
 						.versions()
 						.iter()
-						.filter(|v| v.is_uploading() && v.timestamp < older_than)
+						.filter(|v| v.is_uploading(None) && v.timestamp < older_than)
 						.map(|v| ObjectVersion {
 							state: ObjectVersionState::Aborted,
 							uuid: v.uuid,
@@ -149,11 +149,19 @@ impl Key {
 	}

 	/// Import a key from it's parts
-	pub fn import(key_id: &str, secret_key: &str, name: &str) -> Self {
-		Self {
+	pub fn import(key_id: &str, secret_key: &str, name: &str) -> Result<Self, &'static str> {
+		if key_id.len() != 26 || &key_id[..2] != "GK" || hex::decode(&key_id[2..]).is_err() {
+			return Err("The specified key ID is not a valid Garage key ID (starts with `GK`, followed by 12 hex-encoded bytes)");
+		}
+
+		if secret_key.len() != 64 || hex::decode(&secret_key).is_err() {
+			return Err("The specified secret key is not a valid Garage secret key (composed of 32 hex-encoded bytes)");
+		}
+
+		Ok(Self {
 			key_id: key_id.to_string(),
 			state: crdt::Deletable::present(KeyParams::new(secret_key, name)),
-		}
+		})
 	}

 	/// Create a new Key which can me merged to mark an existing key deleted
@@ -1,3 +1,4 @@
 pub mod block_ref_table;
+pub mod mpu_table;
 pub mod object_table;
 pub mod version_table;
@@ -0,0 +1,254 @@
+use std::sync::Arc;
+
+use garage_db as db;
+
+use garage_util::crdt::Crdt;
+use garage_util::data::*;
+use garage_util::time::*;
+
+use garage_table::replication::TableShardedReplication;
+use garage_table::*;
+
+use crate::index_counter::*;
+use crate::s3::version_table::*;
+
+pub const UPLOADS: &str = "uploads";
+pub const PARTS: &str = "parts";
+pub const BYTES: &str = "bytes";
+
+mod v09 {
+	use garage_util::crdt;
+	use garage_util::data::Uuid;
+	use serde::{Deserialize, Serialize};
+
+	/// A part of a multipart upload
+	#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+	pub struct MultipartUpload {
+		/// Partition key = Upload id = UUID of the object version
+		pub upload_id: Uuid,
+
+		/// The timestamp at which the multipart upload was created
+		pub timestamp: u64,
+		/// Is this multipart upload deleted
+		/// The MultipartUpload is marked as deleted as soon as the
+		/// multipart upload is either completed or aborted
+		pub deleted: crdt::Bool,
+		/// List of uploaded parts, key = (part number, timestamp)
+		/// In case of retries, all versions for each part are kept
+		/// Everything is cleaned up only once the MultipartUpload is marked deleted
+		pub parts: crdt::Map<MpuPartKey, MpuPart>,
+
+		// Back link to bucket+key so that we can find the object this mpu
+		// belongs to and check whether it is still valid
+		/// Bucket in which the related object is stored
+		pub bucket_id: Uuid,
+		/// Key in which the related object is stored
+		pub key: String,
+	}
+
+	#[derive(PartialEq, Eq, Clone, Copy, Debug, Serialize, Deserialize)]
+	pub struct MpuPartKey {
+		/// Number of the part
+		pub part_number: u64,
+		/// Timestamp of part upload
+		pub timestamp: u64,
+	}
+
+	/// The version of an uploaded part
+	#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+	pub struct MpuPart {
+		/// Links to a Version in VersionTable
+		pub version: Uuid,
+		/// ETag of the content of this part (known only once done uploading)
+		pub etag: Option<String>,
+		/// Size of this part (known only once done uploading)
+		pub size: Option<u64>,
+	}
+
+	impl garage_util::migrate::InitialFormat for MultipartUpload {
+		const VERSION_MARKER: &'static [u8] = b"G09s3mpu";
+	}
+}
+
+pub use v09::*;
+
+impl Ord for MpuPartKey {
+	fn cmp(&self, other: &Self) -> std::cmp::Ordering {
+		self.part_number
+			.cmp(&other.part_number)
+			.then(self.timestamp.cmp(&other.timestamp))
+	}
+}
+
+impl PartialOrd for MpuPartKey {
+	fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
+		Some(self.cmp(other))
+	}
+}
+
+impl MultipartUpload {
+	pub fn new(
+		upload_id: Uuid,
+		timestamp: u64,
+		bucket_id: Uuid,
+		key: String,
+		deleted: bool,
+	) -> Self {
+		Self {
+			upload_id,
+			timestamp,
+			deleted: crdt::Bool::new(deleted),
+			parts: crdt::Map::new(),
+			bucket_id,
+			key,
+		}
+	}
+
+	pub fn next_timestamp(&self, part_number: u64) -> u64 {
+		std::cmp::max(
+			now_msec(),
+			1 + self
+				.parts
+				.items()
+				.iter()
+				.filter(|(x, _)| x.part_number == part_number)
+				.map(|(x, _)| x.timestamp)
+				.max()
+				.unwrap_or(0),
+		)
+	}
+}
+
+impl Entry<Uuid, EmptyKey> for MultipartUpload {
+	fn partition_key(&self) -> &Uuid {
+		&self.upload_id
+	}
+	fn sort_key(&self) -> &EmptyKey {
+		&EmptyKey
+	}
+	fn is_tombstone(&self) -> bool {
+		self.deleted.get()
+	}
+}
+
+impl Crdt for MultipartUpload {
+	fn merge(&mut self, other: &Self) {
+		self.deleted.merge(&other.deleted);
+
+		if self.deleted.get() {
+			self.parts.clear();
+		} else {
+			self.parts.merge(&other.parts);
+		}
+	}
+}
+
+impl Crdt for MpuPart {
+	fn merge(&mut self, other: &Self) {
+		self.etag = match (self.etag.take(), &other.etag) {
+			(None, Some(_)) => other.etag.clone(),
+			(Some(x), Some(y)) if x < *y => other.etag.clone(),
+			(x, _) => x,
+		};
+		self.size = match (self.size, other.size) {
+			(None, Some(_)) => other.size,
+			(Some(x), Some(y)) if x < y => other.size,
+			(x, _) => x,
+		};
+	}
+}
+
+pub struct MultipartUploadTable {
+	pub version_table: Arc<Table<VersionTable, TableShardedReplication>>,
+	pub mpu_counter_table: Arc<IndexCounter<MultipartUpload>>,
+}
+
+impl TableSchema for MultipartUploadTable {
+	const TABLE_NAME: &'static str = "multipart_upload";
+
+	type P = Uuid;
+	type S = EmptyKey;
+	type E = MultipartUpload;
+	type Filter = DeletedFilter;
+
+	fn updated(
+		&self,
+		tx: &mut db::Transaction,
+		old: Option<&Self::E>,
+		new: Option<&Self::E>,
+	) -> db::TxOpResult<()> {
+		// 1. Count
+		let counter_res = self.mpu_counter_table.count(tx, old, new);
+		if let Err(e) = db::unabort(counter_res)? {
+			error!(
+				"Unable to update multipart object part counter: {}. Index values will be wrong!",
+				e
+			);
+		}
+
+		// 2. Propagate deletions to version table
+		if let (Some(old_mpu), Some(new_mpu)) = (old, new) {
+			if new_mpu.deleted.get() && !old_mpu.deleted.get() {
+				let deleted_versions = old_mpu.parts.items().iter().map(|(_k, p)| {
+					Version::new(
+						p.version,
+						VersionBacklink::MultipartUpload {
+							upload_id: old_mpu.upload_id,
+						},
+						true,
+					)
+				});
+				for version in deleted_versions {
+					let res = self.version_table.queue_insert(tx, &version);
+					if let Err(e) = db::unabort(res)? {
+						error!("Unable to enqueue version deletion propagation: {}. A repair will be needed.", e);
+					}
+				}
+			}
+		}
+
+		Ok(())
+	}
+
+	fn matches_filter(entry: &Self::E, filter: &Self::Filter) -> bool {
+		filter.apply(entry.is_tombstone())
+	}
+}
+
+impl CountedItem for MultipartUpload {
+	const COUNTER_TABLE_NAME: &'static str = "bucket_mpu_counter";
+
+	// Partition key = bucket id
+	type CP = Uuid;
+	// Sort key = nothing
+	type CS = EmptyKey;
+
+	fn counter_partition_key(&self) -> &Uuid {
+		&self.bucket_id
+	}
+	fn counter_sort_key(&self) -> &EmptyKey {
+		&EmptyKey
+	}
+
+	fn counts(&self) -> Vec<(&'static str, i64)> {
+		let uploads = if self.deleted.get() { 0 } else { 1 };
+		let mut parts = self
+			.parts
+			.items()
+			.iter()
+			.map(|(k, _)| k.part_number)
+			.collect::<Vec<_>>();
+		parts.dedup();
+		let bytes = self
+			.parts
+			.items()
+			.iter()
+			.map(|(_, p)| p.size.unwrap_or(0))
+			.sum::<u64>();
+		vec![
+			(UPLOADS, uploads),
+			(PARTS, parts.len() as i64),
+			(BYTES, bytes as i64),
+		]
+	}
+}
@@ -10,6 +10,7 @@ use garage_table::replication::TableShardedReplication;
 use garage_table::*;

 use crate::index_counter::*;
+use crate::s3::mpu_table::*;
 use crate::s3::version_table::*;

 pub const OBJECTS: &str = "objects";
@@ -130,7 +131,86 @@ mod v08 {
 	}
 }

-pub use v08::*;
+mod v09 {
+	use garage_util::data::Uuid;
+	use serde::{Deserialize, Serialize};
+
+	use super::v08;
+
+	pub use v08::{ObjectVersionData, ObjectVersionHeaders, ObjectVersionMeta};
+
+	/// An object
+	#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+	pub struct Object {
+		/// The bucket in which the object is stored, used as partition key
+		pub bucket_id: Uuid,
+
+		/// The key at which the object is stored in its bucket, used as sorting key
+		pub key: String,
+
+		/// The list of currenty stored versions of the object
+		pub(super) versions: Vec<ObjectVersion>,
+	}
+
+	/// Informations about a version of an object
+	#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+	pub struct ObjectVersion {
+		/// Id of the version
+		pub uuid: Uuid,
+		/// Timestamp of when the object was created
+		pub timestamp: u64,
+		/// State of the version
+		pub state: ObjectVersionState,
+	}
+
+	/// State of an object version
+	#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+	pub enum ObjectVersionState {
+		/// The version is being received
+		Uploading {
+			/// Indicates whether this is a multipart upload
+			multipart: bool,
+			/// Headers to be included in the final object
+			headers: ObjectVersionHeaders,
+		},
+		/// The version is fully received
+		Complete(ObjectVersionData),
+		/// The version uploaded containded errors or the upload was explicitly aborted
+		Aborted,
+	}
+
+	impl garage_util::migrate::Migrate for Object {
+		const VERSION_MARKER: &'static [u8] = b"G09s3o";
+
+		type Previous = v08::Object;
+
+		fn migrate(old: v08::Object) -> Object {
+			let versions = old
+				.versions
+				.into_iter()
+				.map(|x| ObjectVersion {
+					uuid: x.uuid,
+					timestamp: x.timestamp,
+					state: match x.state {
+						v08::ObjectVersionState::Uploading(h) => ObjectVersionState::Uploading {
+							multipart: false,
+							headers: h,
+						},
+						v08::ObjectVersionState::Complete(d) => ObjectVersionState::Complete(d),
+						v08::ObjectVersionState::Aborted => ObjectVersionState::Aborted,
+					},
+				})
+				.collect();
+			Object {
+				bucket_id: old.bucket_id,
+				key: old.key,
+				versions,
+			}
+		}
+	}
+}
+
+pub use v09::*;

 impl Object {
 	/// Initialize an Object struct from parts
@@ -180,11 +260,11 @@ impl Crdt for ObjectVersionState {
 				Complete(a) => {
 					a.merge(b);
 				}
-				Uploading(_) => {
+				Uploading { .. } => {
 					*self = Complete(b.clone());
 				}
 			},
-			Uploading(_) => {}
+			Uploading { .. } => {}
 		}
 	}
 }
@@ -199,8 +279,17 @@ impl ObjectVersion {
 	}

 	/// Is the object version currently being uploaded
-	pub fn is_uploading(&self) -> bool {
-		matches!(self.state, ObjectVersionState::Uploading(_))
+	///
+	/// matches only multipart uploads if check_multipart is Some(true)
+	/// matches only non-multipart uploads if check_multipart is Some(false)
+	/// matches both if check_multipart is None
+	pub fn is_uploading(&self, check_multipart: Option<bool>) -> bool {
+		match &self.state {
+			ObjectVersionState::Uploading { multipart, .. } => {
+				check_multipart.map(|x| x == *multipart).unwrap_or(true)
+			}
+			_ => false,
+		}
 	}

 	/// Is the object version completely received
@@ -267,13 +356,20 @@ impl Crdt for Object {

 pub struct ObjectTable {
 	pub version_table: Arc<Table<VersionTable, TableShardedReplication>>,
+	pub mpu_table: Arc<Table<MultipartUploadTable, TableShardedReplication>>,
 	pub object_counter_table: Arc<IndexCounter<Object>>,
 }

 #[derive(Clone, Copy, Debug, Serialize, Deserialize)]
 pub enum ObjectFilter {
+	/// Is the object version available (received and not a tombstone)
 	IsData,
-	IsUploading,
+	/// Is the object version currently being uploaded
+	///
+	/// matches only multipart uploads if check_multipart is Some(true)
+	/// matches only non-multipart uploads if check_multipart is Some(false)
+	/// matches both if check_multipart is None
+	IsUploading { check_multipart: Option<bool> },
 }

 impl TableSchema for ObjectTable {
@@ -301,21 +397,28 @@ impl TableSchema for ObjectTable {

 		// 2. Enqueue propagation deletions to version table
 		if let (Some(old_v), Some(new_v)) = (old, new) {
-			// Propagate deletion of old versions
 			for v in old_v.versions.iter() {
-				let newly_deleted = match new_v
+				let new_v_id = new_v
 					.versions
-					.binary_search_by(|nv| nv.cmp_key().cmp(&v.cmp_key()))
-				{
+					.binary_search_by(|nv| nv.cmp_key().cmp(&v.cmp_key()));
+
+				// Propagate deletion of old versions to the Version table
+				let delete_version = match new_v_id {
 					Err(_) => true,
 					Ok(i) => {
 						new_v.versions[i].state == ObjectVersionState::Aborted
 							&& v.state != ObjectVersionState::Aborted
 					}
 				};
-				if newly_deleted {
-					let deleted_version =
-						Version::new(v.uuid, old_v.bucket_id, old_v.key.clone(), true);
+				if delete_version {
+					let deleted_version = Version::new(
+						v.uuid,
+						VersionBacklink::Object {
+							bucket_id: old_v.bucket_id,
+							key: old_v.key.clone(),
+						},
+						true,
+					);
 					let res = self.version_table.queue_insert(tx, &deleted_version);
 					if let Err(e) = db::unabort(res)? {
 						error!(
@@ -324,6 +427,39 @@ impl TableSchema for ObjectTable {
 						);
 					}
 				}
+
+				// After abortion or completion of multipart uploads, delete MPU table entry
+				if matches!(
+					v.state,
+					ObjectVersionState::Uploading {
+						multipart: true,
+						..
+					}
+				) {
+					let delete_mpu = match new_v_id {
+						Err(_) => true,
+						Ok(i) => !matches!(
+							new_v.versions[i].state,
+							ObjectVersionState::Uploading { .. }
+						),
+					};
+					if delete_mpu {
+						let deleted_mpu = MultipartUpload::new(
+							v.uuid,
+							v.timestamp,
+							old_v.bucket_id,
+							old_v.key.clone(),
+							true,
+						);
+						let res = self.mpu_table.queue_insert(tx, &deleted_mpu);
+						if let Err(e) = db::unabort(res)? {
+							error!(
+								"Unable to enqueue multipart upload deletion propagation: {}. A repair will be needed.",
+								e
+							);
+						}
+					}
+				}
 			}
 		}

@@ -333,7 +469,10 @@ impl TableSchema for ObjectTable {
 	fn matches_filter(entry: &Self::E, filter: &Self::Filter) -> bool {
 		match filter {
 			ObjectFilter::IsData => entry.versions.iter().any(|v| v.is_data()),
-			ObjectFilter::IsUploading => entry.versions.iter().any(|v| v.is_uploading()),
+			ObjectFilter::IsUploading { check_multipart } => entry
+				.versions
+				.iter()
+				.any(|v| v.is_uploading(*check_multipart)),
 		}
 	}
 }
@@ -360,10 +499,7 @@ impl CountedItem for Object {
 		} else {
 			0
 		};
-		let n_unfinished_uploads = versions
-			.iter()
-			.filter(|v| matches!(v.state, ObjectVersionState::Uploading(_)))
-			.count();
+		let n_unfinished_uploads = versions.iter().filter(|v| v.is_uploading(None)).count();
 		let n_bytes = versions
 			.iter()
 			.map(|v| match &v.state {
@@ -3,6 +3,7 @@ use std::sync::Arc;
 use garage_db as db;

 use garage_util::data::*;
+use garage_util::error::*;

 use garage_table::crdt::*;
 use garage_table::replication::TableShardedReplication;
@@ -66,6 +67,8 @@ mod v08 {

 	use super::v05;

+	pub use v05::{VersionBlock, VersionBlockKey};
+
 	/// A version of an object
 	#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 	pub struct Version {
@@ -90,8 +93,6 @@ mod v08 {
 		pub key: String,
 	}

-	pub use v05::{VersionBlock, VersionBlockKey};
-
 	impl garage_util::migrate::Migrate for Version {
 		type Previous = v05::Version;

@@ -110,32 +111,94 @@ mod v08 {
 	}
 }

-pub use v08::*;
+pub(crate) mod v09 {
+	use garage_util::crdt;
+	use garage_util::data::Uuid;
+	use serde::{Deserialize, Serialize};
+
+	use super::v08;
+
+	pub use v08::{VersionBlock, VersionBlockKey};
+
+	/// A version of an object
+	#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+	pub struct Version {
+		/// UUID of the version, used as partition key
+		pub uuid: Uuid,
+
+		// Actual data: the blocks for this version
+		// In the case of a multipart upload, also store the etags
+		// of individual parts and check them when doing CompleteMultipartUpload
+		/// Is this version deleted
+		pub deleted: crdt::Bool,
+		/// list of blocks of data composing the version
+		pub blocks: crdt::Map<VersionBlockKey, VersionBlock>,
+
+		// Back link to owner of this version (either an object or a multipart
+		// upload), used to find whether it has been deleted and this version
+		// should in turn be deleted (see versions repair procedure)
+		pub backlink: VersionBacklink,
+	}
+
+	#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+	pub enum VersionBacklink {
+		Object {
+			/// Bucket in which the related object is stored
+			bucket_id: Uuid,
+			/// Key in which the related object is stored
+			key: String,
+		},
+		MultipartUpload {
+			upload_id: Uuid,
+		},
+	}
+
+	impl garage_util::migrate::Migrate for Version {
+		const VERSION_MARKER: &'static [u8] = b"G09s3v";
+
+		type Previous = v08::Version;
+
+		fn migrate(old: v08::Version) -> Version {
+			Version {
+				uuid: old.uuid,
+				deleted: old.deleted,
+				blocks: old.blocks,
+				backlink: VersionBacklink::Object {
+					bucket_id: old.bucket_id,
+					key: old.key,
+				},
+			}
+		}
+	}
+}
+
+pub use v09::*;

 impl Version {
-	pub fn new(uuid: Uuid, bucket_id: Uuid, key: String, deleted: bool) -> Self {
+	pub fn new(uuid: Uuid, backlink: VersionBacklink, deleted: bool) -> Self {
 		Self {
 			uuid,
 			deleted: deleted.into(),
 			blocks: crdt::Map::new(),
-			parts_etags: crdt::Map::new(),
-			bucket_id,
-			key,
+			backlink,
 		}
 	}

 	pub fn has_part_number(&self, part_number: u64) -> bool {
-		let case1 = self
-			.parts_etags
-			.items()
-			.binary_search_by(|(k, _)| k.cmp(&part_number))
-			.is_ok();
-		let case2 = self
-			.blocks
+		self.blocks
 			.items()
 			.binary_search_by(|(k, _)| k.part_number.cmp(&part_number))
-			.is_ok();
-		case1 || case2
+			.is_ok()
+	}
+
+	pub fn n_parts(&self) -> Result<u64, Error> {
+		Ok(self
+			.blocks
+			.items()
+			.last()
+			.ok_or_message("version has no parts")?
+			.0
+			.part_number)
 	}
 }

@@ -175,10 +238,8 @@ impl Crdt for Version {

 		if self.deleted.get() {
 			self.blocks.clear();
-			self.parts_etags.clear();
 		} else {
 			self.blocks.merge(&other.blocks);
-			self.parts_etags.merge(&other.parts_etags);
 		}
 	}
 }
@@ -14,14 +14,17 @@ path = "lib.rs"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html

 [dependencies]
+garage_db.workspace = true
 garage_util.workspace = true

 arc-swap = "1.0"
 bytes = "1.0"
+bytesize = "1.1"
 gethostname = "0.4"
 hex = "0.4"
 tracing = "0.1"
 rand = "0.8"
+itertools="0.10"
 sodiumoxide = { version = "0.2.5-0", package = "kuska-sodiumoxide" }
 systemstat = "0.2.3"

@@ -45,7 +48,7 @@ tokio = { version = "1.0", default-features = false, features = ["rt", "rt-multi
 tokio-stream = { version = "0.1", features = ["net"] }
 opentelemetry = "0.17"

-netapp = { version = "0.5.2", features = ["telemetry"] }
+netapp = { version = "0.5.3", features = ["telemetry"] }

 [features]
 kubernetes-discovery = [ "kube", "k8s-openapi", "schemars" ]
@@ -0,0 +1,411 @@
+//! This module deals with graph algorithms.
+//! It is used in layout.rs to build the partition to node assignment.
+
+use rand::prelude::SliceRandom;
+use std::cmp::{max, min};
+use std::collections::HashMap;
+use std::collections::VecDeque;
+
+/// Vertex data structures used in all the graphs used in layout.rs.
+/// usize parameters correspond to node/zone/partitions ids.
+/// To understand the vertex roles below, please refer to the formal description
+/// of the layout computation algorithm.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub enum Vertex {
+	Source,
+	Pup(usize),       // The vertex p+ of partition p
+	Pdown(usize),     // The vertex p- of partition p
+	PZ(usize, usize), // The vertex corresponding to x_(partition p, zone z)
+	N(usize),         // The vertex corresponding to node n
+	Sink,
+}
+
+/// Edge data structure for the flow algorithm.
+#[derive(Clone, Copy, Debug)]
+pub struct FlowEdge {
+	cap: u64,    // flow maximal capacity of the edge
+	flow: i64,   // flow value on the edge
+	dest: usize, // destination vertex id
+	rev: usize,  // index of the reversed edge (v, self) in the edge list of vertex v
+}
+
+/// Edge data structure for the detection of negative cycles.
+#[derive(Clone, Copy, Debug)]
+pub struct WeightedEdge {
+	w: i64, // weight of the edge
+	dest: usize,
+}
+
+pub trait Edge: Clone + Copy {}
+impl Edge for FlowEdge {}
+impl Edge for WeightedEdge {}
+
+/// Struct for the graph structure. We do encapsulation here to be able to both
+/// provide user friendly Vertex enum to address vertices, and to use internally usize
+/// indices and Vec instead of HashMap in the graph algorithm to optimize execution speed.
+pub struct Graph<E: Edge> {
+	vertex_to_id: HashMap<Vertex, usize>,
+	id_to_vertex: Vec<Vertex>,
+
+	// The graph is stored as an adjacency list
+	graph: Vec<Vec<E>>,
+}
+
+pub type CostFunction = HashMap<(Vertex, Vertex), i64>;
+
+impl<E: Edge> Graph<E> {
+	pub fn new(vertices: &[Vertex]) -> Self {
+		let mut map = HashMap::<Vertex, usize>::new();
+		for (i, vert) in vertices.iter().enumerate() {
+			map.insert(*vert, i);
+		}
+		Graph::<E> {
+			vertex_to_id: map,
+			id_to_vertex: vertices.to_vec(),
+			graph: vec![Vec::<E>::new(); vertices.len()],
+		}
+	}
+
+	fn get_vertex_id(&self, v: &Vertex) -> Result<usize, String> {
+		self.vertex_to_id
+			.get(v)
+			.cloned()
+			.ok_or_else(|| format!("The graph does not contain vertex {:?}", v))
+	}
+}
+
+impl Graph<FlowEdge> {
+	/// This function adds a directed edge to the graph with capacity c, and the
+	/// corresponding reversed edge with capacity 0.
+	pub fn add_edge(&mut self, u: Vertex, v: Vertex, c: u64) -> Result<(), String> {
+		let idu = self.get_vertex_id(&u)?;
+		let idv = self.get_vertex_id(&v)?;
+		if idu == idv {
+			return Err("Cannot add edge from vertex to itself in flow graph".into());
+		}
+
+		let rev_u = self.graph[idu].len();
+		let rev_v = self.graph[idv].len();
+		self.graph[idu].push(FlowEdge {
+			cap: c,
+			dest: idv,
+			flow: 0,
+			rev: rev_v,
+		});
+		self.graph[idv].push(FlowEdge {
+			cap: 0,
+			dest: idu,
+			flow: 0,
+			rev: rev_u,
+		});
+		Ok(())
+	}
+
+	/// This function returns the list of vertices that receive a positive flow from
+	/// vertex v.
+	pub fn get_positive_flow_from(&self, v: Vertex) -> Result<Vec<Vertex>, String> {
+		let idv = self.get_vertex_id(&v)?;
+		let mut result = Vec::<Vertex>::new();
+		for edge in self.graph[idv].iter() {
+			if edge.flow > 0 {
+				result.push(self.id_to_vertex[edge.dest]);
+			}
+		}
+		Ok(result)
+	}
+
+	/// This function returns the value of the flow incoming to v.
+	pub fn get_inflow(&self, v: Vertex) -> Result<i64, String> {
+		let idv = self.get_vertex_id(&v)?;
+		let mut result = 0;
+		for edge in self.graph[idv].iter() {
+			result += max(0, self.graph[edge.dest][edge.rev].flow);
+		}
+		Ok(result)
+	}
+
+	/// This function returns the value of the flow outgoing from v.
+	pub fn get_outflow(&self, v: Vertex) -> Result<i64, String> {
+		let idv = self.get_vertex_id(&v)?;
+		let mut result = 0;
+		for edge in self.graph[idv].iter() {
+			result += max(0, edge.flow);
+		}
+		Ok(result)
+	}
+
+	/// This function computes the flow total value by computing the outgoing flow
+	/// from the source.
+	pub fn get_flow_value(&mut self) -> Result<i64, String> {
+		self.get_outflow(Vertex::Source)
+	}
+
+	/// This function shuffles the order of the edge lists. It keeps the ids of the
+	/// reversed edges consistent.
+	fn shuffle_edges(&mut self) {
+		let mut rng = rand::thread_rng();
+		for i in 0..self.graph.len() {
+			self.graph[i].shuffle(&mut rng);
+			// We need to update the ids of the reverse edges.
+			for j in 0..self.graph[i].len() {
+				let target_v = self.graph[i][j].dest;
+				let target_rev = self.graph[i][j].rev;
+				self.graph[target_v][target_rev].rev = j;
+			}
+		}
+	}
+
+	/// Computes an upper bound of the flow on the graph
+	pub fn flow_upper_bound(&self) -> Result<u64, String> {
+		let idsource = self.get_vertex_id(&Vertex::Source)?;
+		let mut flow_upper_bound = 0;
+		for edge in self.graph[idsource].iter() {
+			flow_upper_bound += edge.cap;
+		}
+		Ok(flow_upper_bound)
+	}
+
+	/// This function computes the maximal flow using Dinic's algorithm. It starts with
+	/// the flow values already present in the graph. So it is possible to add some edge to
+	/// the graph, compute a flow, add other edges, update the flow.
+	pub fn compute_maximal_flow(&mut self) -> Result<(), String> {
+		let idsource = self.get_vertex_id(&Vertex::Source)?;
+		let idsink = self.get_vertex_id(&Vertex::Sink)?;
+
+		let nb_vertices = self.graph.len();
+
+		let flow_upper_bound = self.flow_upper_bound()?;
+
+		// To ensure the dispersion of the associations generated by the
+		// assignment, we shuffle the neighbours of the nodes. Hence,
+		// the vertices do not consider their neighbours in the same order.
+		self.shuffle_edges();
+
+		// We run Dinic's max flow algorithm
+		loop {
+			// We build the level array from Dinic's algorithm.
+			let mut level = vec![None; nb_vertices];
+
+			let mut fifo = VecDeque::new();
+			fifo.push_back((idsource, 0));
+			while let Some((id, lvl)) = fifo.pop_front() {
+				if level[id] == None {
+					// it means id has not yet been reached
+					level[id] = Some(lvl);
+					for edge in self.graph[id].iter() {
+						if edge.cap as i64 - edge.flow > 0 {
+							fifo.push_back((edge.dest, lvl + 1));
+						}
+					}
+				}
+			}
+			if level[idsink] == None {
+				// There is no residual flow
+				break;
+			}
+			// Now we run DFS respecting the level array
+			let mut next_nbd = vec![0; nb_vertices];
+			let mut lifo = Vec::new();
+
+			lifo.push((idsource, flow_upper_bound));
+
+			while let Some((id, f)) = lifo.last().cloned() {
+				if id == idsink {
+					// The DFS reached the sink, we can add a
+					// residual flow.
+					lifo.pop();
+					while let Some((id, _)) = lifo.pop() {
+						let nbd = next_nbd[id];
+						self.graph[id][nbd].flow += f as i64;
+						let id_rev = self.graph[id][nbd].dest;
+						let nbd_rev = self.graph[id][nbd].rev;
+						self.graph[id_rev][nbd_rev].flow -= f as i64;
+					}
+					lifo.push((idsource, flow_upper_bound));
+					continue;
+				}
+				// else we did not reach the sink
+				let nbd = next_nbd[id];
+				if nbd >= self.graph[id].len() {
+					// There is nothing to explore from id anymore
+					lifo.pop();
+					if let Some((parent, _)) = lifo.last() {
+						next_nbd[*parent] += 1;
+					}
+					continue;
+				}
+				// else we can try to send flow from id to its nbd
+				let new_flow = min(
+					f as i64,
+					self.graph[id][nbd].cap as i64 - self.graph[id][nbd].flow,
+				) as u64;
+				if new_flow == 0 {
+					next_nbd[id] += 1;
+					continue;
+				}
+				if let (Some(lvldest), Some(lvlid)) = (level[self.graph[id][nbd].dest], level[id]) {
+					if lvldest <= lvlid {
+						// We cannot send flow to nbd.
+						next_nbd[id] += 1;
+						continue;
+					}
+				}
+				// otherwise, we send flow to nbd.
+				lifo.push((self.graph[id][nbd].dest, new_flow));
+			}
+		}
+		Ok(())
+	}
+
+	/// This function takes a flow, and a cost function on the edges, and tries to find an
+	/// equivalent flow with a better cost, by finding improving overflow cycles. It uses
+	/// as subroutine the Bellman Ford algorithm run up to path_length.
+	/// We assume that the cost of edge (u,v) is the opposite of the cost of (v,u), and
+	/// only one needs to be present in the cost function.
+	pub fn optimize_flow_with_cost(
+		&mut self,
+		cost: &CostFunction,
+		path_length: usize,
+	) -> Result<(), String> {
+		// We build the weighted graph g where we will look for negative cycle
+		let mut gf = self.build_cost_graph(cost)?;
+		let mut cycles = gf.list_negative_cycles(path_length);
+		while !cycles.is_empty() {
+			// we enumerate negative cycles
+			for c in cycles.iter() {
+				for i in 0..c.len() {
+					// We add one flow unit to the edge (u,v) of cycle c
+					let idu = self.vertex_to_id[&c[i]];
+					let idv = self.vertex_to_id[&c[(i + 1) % c.len()]];
+					for j in 0..self.graph[idu].len() {
+						// since idu appears at most once in the cycles, we enumerate every
+						// edge at most once.
+						let edge = self.graph[idu][j];
+						if edge.dest == idv {
+							self.graph[idu][j].flow += 1;
+							self.graph[idv][edge.rev].flow -= 1;
+							break;
+						}
+					}
+				}
+			}
+
+			gf = self.build_cost_graph(cost)?;
+			cycles = gf.list_negative_cycles(path_length);
+		}
+		Ok(())
+	}
+
+	/// Construct the weighted graph G_f from the flow and the cost function
+	fn build_cost_graph(&self, cost: &CostFunction) -> Result<Graph<WeightedEdge>, String> {
+		let mut g = Graph::<WeightedEdge>::new(&self.id_to_vertex);
+		let nb_vertices = self.id_to_vertex.len();
+		for i in 0..nb_vertices {
+			for edge in self.graph[i].iter() {
+				if edge.cap as i64 - edge.flow > 0 {
+					// It is possible to send overflow through this edge
+					let u = self.id_to_vertex[i];
+					let v = self.id_to_vertex[edge.dest];
+					if cost.contains_key(&(u, v)) {
+						g.add_edge(u, v, cost[&(u, v)])?;
+					} else if cost.contains_key(&(v, u)) {
+						g.add_edge(u, v, -cost[&(v, u)])?;
+					} else {
+						g.add_edge(u, v, 0)?;
+					}
+				}
+			}
+		}
+		Ok(g)
+	}
+}
+
+impl Graph<WeightedEdge> {
+	/// This function adds a single directed weighted edge to the graph.
+	pub fn add_edge(&mut self, u: Vertex, v: Vertex, w: i64) -> Result<(), String> {
+		let idu = self.get_vertex_id(&u)?;
+		let idv = self.get_vertex_id(&v)?;
+		self.graph[idu].push(WeightedEdge { w, dest: idv });
+		Ok(())
+	}
+
+	/// This function lists the negative cycles it manages to find after path_length
+	/// iterations of the main loop of the Bellman-Ford algorithm. For the classical
+	/// algorithm, path_length needs to be equal to the number of vertices. However,
+	/// for particular graph structures like in our case, the algorithm is still correct
+	/// when path_length is the length of the longest possible simple path.
+	/// See the formal description of the algorithm for more details.
+	fn list_negative_cycles(&self, path_length: usize) -> Vec<Vec<Vertex>> {
+		let nb_vertices = self.graph.len();
+
+		// We start with every vertex at distance 0 of some imaginary extra -1 vertex.
+		let mut distance = vec![0; nb_vertices];
+		// The prev vector collects for every vertex from where does the shortest path come
+		let mut prev = vec![None; nb_vertices];
+
+		for _ in 0..path_length + 1 {
+			for id in 0..nb_vertices {
+				for e in self.graph[id].iter() {
+					if distance[id] + e.w < distance[e.dest] {
+						distance[e.dest] = distance[id] + e.w;
+						prev[e.dest] = Some(id);
+					}
+				}
+			}
+		}
+
+		// If self.graph contains a negative cycle, then at this point the graph described
+		// by prev (which is a directed 1-forest/functional graph)
+		// must contain a cycle. We list the cycles of prev.
+		let cycles_prev = cycles_of_1_forest(&prev);
+
+		// Remark that the cycle in prev is in the reverse order compared to the cycle
+		// in the graph. Thus the .rev().
+		return cycles_prev
+			.iter()
+			.map(|cycle| {
+				cycle
+					.iter()
+					.rev()
+					.map(|id| self.id_to_vertex[*id])
+					.collect()
+			})
+			.collect();
+	}
+}
+
+/// This function returns the list of cycles of a directed 1 forest. It does not
+/// check for the consistency of the input.
+fn cycles_of_1_forest(forest: &[Option<usize>]) -> Vec<Vec<usize>> {
+	let mut cycles = Vec::<Vec<usize>>::new();
+	let mut time_of_discovery = vec![None; forest.len()];
+
+	for t in 0..forest.len() {
+		let mut id = t;
+		// while we are on a valid undiscovered node
+		while time_of_discovery[id] == None {
+			time_of_discovery[id] = Some(t);
+			if let Some(i) = forest[id] {
+				id = i;
+			} else {
+				break;
+			}
+		}
+		if forest[id] != None && time_of_discovery[id] == Some(t) {
+			// We discovered an id that we explored at this iteration t.
+			// It means we are on a cycle
+			let mut cy = vec![id; 1];
+			let mut id2 = id;
+			while let Some(id_next) = forest[id2] {
+				id2 = id_next;
+				if id2 != id {
+					cy.push(id2);
+				} else {
+					break;
+				}
+			}
+			cycles.push(cy);
+		}
+	}
+	cycles
+}
@@ -11,6 +11,7 @@ mod consul;
 #[cfg(feature = "kubernetes-discovery")]
 mod kubernetes;

+pub mod graph_algo;
 pub mod layout;
 pub mod replication_mode;
 pub mod ring;
@@ -40,6 +40,7 @@ pub struct Ring {
 // Type to store compactly the id of a node in the system
 // Change this to u16 the day we want to have more than 256 nodes in a cluster
 pub type CompactNodeType = u8;
+pub const MAX_NODE_NUMBER: usize = 256;

 // The maximum number of times an object might get replicated
 // This must be at least 3 because Garage supports 3-way replication
@@ -62,12 +63,12 @@ struct RingEntry {
 impl Ring {
 	pub(crate) fn new(layout: ClusterLayout, replication_factor: usize) -> Self {
 		if replication_factor != layout.replication_factor {
-			warn!("Could not build ring: replication factor does not match between local configuration and network role assignation.");
+			warn!("Could not build ring: replication factor does not match between local configuration and network role assignment.");
 			return Self::empty(layout, replication_factor);
 		}

-		if layout.ring_assignation_data.len() != replication_factor * (1 << PARTITION_BITS) {
-			warn!("Could not build ring: network role assignation data has invalid length");
+		if layout.ring_assignment_data.len() != replication_factor * (1 << PARTITION_BITS) {
+			warn!("Could not build ring: network role assignment data has invalid length");
 			return Self::empty(layout, replication_factor);
 		}

@@ -77,7 +78,7 @@ impl Ring {
 				let top = (i as u16) << (16 - PARTITION_BITS);
 				let mut nodes_buf = [0u8; MAX_REPLICATION];
 				nodes_buf[..replication_factor].copy_from_slice(
-					&layout.ring_assignation_data
+					&layout.ring_assignment_data
 						[replication_factor * i..replication_factor * (i + 1)],
 				);
 				RingEntry {
@@ -151,7 +151,7 @@ pub struct KnownNodeInfo {
 	pub status: NodeStatus,
 }

-#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
+#[derive(Debug, Clone, Copy)]
 pub struct ClusterHealth {
 	/// The current health status of the cluster (see below)
 	pub status: ClusterHealthStatus,
@@ -171,7 +171,7 @@ pub struct ClusterHealth {
 	pub partitions_all_ok: usize,
 }

-#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
+#[derive(Debug, Clone, Copy)]
 pub enum ClusterHealthStatus {
 	/// All nodes are available
 	Healthy,
@@ -666,9 +666,9 @@ impl System {
 		let update_ring = self.update_ring.lock().await;
 		let mut layout: ClusterLayout = self.ring.borrow().layout.clone();

-		let prev_layout_check = layout.check();
+		let prev_layout_check = layout.check().is_ok();
 		if layout.merge(adv) {
-			if prev_layout_check && !layout.check() {
+			if prev_layout_check && !layout.check().is_ok() {
 				error!("New cluster layout is invalid, discarding.");
 				return Err(Error::Message(
 					"New cluster layout is invalid, discarding.".into(),
@@ -724,7 +724,7 @@ impl System {

 	async fn discovery_loop(self: &Arc<Self>, mut stop_signal: watch::Receiver<bool>) {
 		while !*stop_signal.borrow() {
-			let not_configured = !self.ring.borrow().layout.check();
+			let not_configured = !self.ring.borrow().layout.check().is_ok();
 			let no_peers = self.fullmesh.get_peer_list().len() < self.replication_factor;
 			let expected_n_nodes = self.ring.borrow().layout.num_nodes();
 			let bad_peers = self
@@ -347,9 +347,7 @@ impl<F: TableSchema, R: TableReplication> TableData<F, R> {
 	// ---- Utility functions ----

 	pub fn tree_key(&self, p: &F::P, s: &F::S) -> Vec<u8> {
-		let mut ret = p.hash().to_vec();
-		ret.extend(s.sort_key());
-		ret
+		[p.hash().as_slice(), s.sort_key()].concat()
 	}

 	pub fn decode_entry(&self, bytes: &[u8]) -> Result<F::E, Error> {
@@ -6,6 +6,8 @@ use garage_util::migrate::Migrate;

 use crate::crdt::Crdt;

+// =================================== PARTITION KEYS
+
 /// Trait for field used to partition data
 pub trait PartitionKey:
 	Clone + PartialEq + Serialize + for<'de> Deserialize<'de> + Send + Sync + 'static
@@ -29,6 +31,8 @@ impl PartitionKey for FixedBytes32 {
 	}
 }

+// =================================== SORT KEYS
+
 /// Trait for field used to sort data
 pub trait SortKey: Clone + Serialize + for<'de> Deserialize<'de> + Send + Sync + 'static {
 	/// Get the key used to sort
@@ -47,6 +51,8 @@ impl SortKey for FixedBytes32 {
 	}
 }

+// =================================== SCHEMA
+
 /// Trait for an entry in a table. It must be sortable and partitionnable.
 pub trait Entry<P: PartitionKey, S: SortKey>:
 	Crdt + PartialEq + Clone + Migrate + Send + Sync + 'static
@@ -31,7 +31,7 @@ rand = "0.8"
 sha2 = "0.10"

 chrono = "0.4"
-rmp-serde = "0.15"
+rmp-serde = "1.1.2"
 serde = { version = "1.0", default-features = false, features = ["derive", "rc"] }
 serde_json = "1.0"
 toml = "0.6"
@@ -15,6 +15,13 @@ pub struct Config {
 	/// Path where to store data. Can be slower, but need higher volume
 	pub data_dir: PathBuf,

+	/// Whether to fsync after all metadata transactions (disabled by default)
+	#[serde(default)]
+	pub metadata_fsync: bool,
+	/// Whether to fsync after all data block writes (disabled by default)
+	#[serde(default)]
+	pub data_fsync: bool,
+
 	/// Size of data blocks to save to disk
 	#[serde(default = "default_block_size")]
 	pub block_size: usize,
@@ -183,7 +190,7 @@ pub struct KubernetesDiscoveryConfig {
 }

 fn default_db_engine() -> String {
-	"sled".into()
+	"lmdb".into()
 }

 fn default_sled_cache_capacity() -> u64 {
@@ -8,9 +8,7 @@ where
 	T: Serialize + ?Sized,
 {
 	let mut wr = Vec::with_capacity(128);
-	let mut se = rmp_serde::Serializer::new(&mut wr)
-		.with_struct_map()
-		.with_string_variants();
+	let mut se = rmp_serde::Serializer::new(&mut wr).with_struct_map();
 	val.serialize(&mut se)?;
 	Ok(wr)
 }
@@ -22,7 +20,7 @@ pub fn nonversioned_decode<T>(bytes: &[u8]) -> Result<T, rmp_serde::decode::Erro
 where
 	T: for<'de> Deserialize<'de> + ?Sized,
 {
-	rmp_serde::decode::from_read_ref::<_, T>(bytes)
+	rmp_serde::decode::from_slice::<_>(bytes)
 }

 /// Serialize to JSON, truncating long result
@@ -19,7 +19,7 @@ pub trait Migrate: Serialize + for<'de> Deserialize<'de> + 'static {
 	fn decode(bytes: &[u8]) -> Option<Self> {
 		let marker_len = Self::VERSION_MARKER.len();
 		if bytes.get(..marker_len) == Some(Self::VERSION_MARKER) {
-			if let Ok(value) = rmp_serde::decode::from_read_ref::<_, Self>(&bytes[marker_len..]) {
+			if let Ok(value) = rmp_serde::decode::from_slice::<_>(&bytes[marker_len..]) {
 				return Some(value);
 			}
 		}
@@ -31,9 +31,7 @@ pub trait Migrate: Serialize + for<'de> Deserialize<'de> + 'static {
 	fn encode(&self) -> Result<Vec<u8>, rmp_serde::encode::Error> {
 		let mut wr = Vec::with_capacity(128);
 		wr.extend_from_slice(Self::VERSION_MARKER);
-		let mut se = rmp_serde::Serializer::new(&mut wr)
-			.with_struct_map()
-			.with_string_variants();
+		let mut se = rmp_serde::Serializer::new(&mut wr).with_struct_map();
 		self.serialize(&mut se)?;
 		Ok(wr)
 	}
Author	SHA1	Message	Date
Alex Auvolat	2e90e1c124	Merge branch 'main' into next	2023-08-29 11:32:42 +02:00
Alex	32e5686ad8	Merge pull request 'Garage v0.8.3' (#619 ) from next-0.8 into main Reviewed-on: https://git.deuxfleurs.fr/Deuxfleurs/garage/pulls/619	2023-08-29 08:55:46 +00:00
Alex Auvolat	06369c8f4a	add garage_db dependency in garage_rpc	2023-08-28 17:08:21 +02:00
Alex Auvolat	8ef42c9609	admin docs: reformatting, key admin: add check	2023-06-14 17:19:25 +02:00
Alex Auvolat	a83a092c03	admin: uniformize layout api and improve code	2023-06-14 17:12:37 +02:00
Alex Auvolat	7895f99d3a	admin and cli: hide secret keys unless asked	2023-06-14 16:56:15 +02:00
Alex Auvolat	4a82f6380e	admin api: move all endpoints to v1/ by default (v0/ still supported)	2023-06-14 14:15:51 +02:00
Alex Auvolat	28cc9f178a	admin api: make name optionnal for CreateKey	2023-06-14 13:56:37 +02:00
Alex Auvolat	2c83006608	admin api: fix doc in drafts	2023-06-14 13:54:34 +02:00
Alex Auvolat	35c108b85d	admin api: switch GetClusterHealth to camelcase (fix #381 again)	2023-06-14 13:53:19 +02:00
Alex Auvolat	52376d47ca	admin api: change cluster status/layout to use lists and not maps (fix #377 )	2023-06-14 13:45:27 +02:00
Alex Auvolat	187240e539	Merge branch 'main' into next	2023-06-14 13:02:46 +02:00
Alex Auvolat	5670367126	multipartupload in test: add forgotten timestamp	2023-06-13 23:10:46 +02:00
Alex Auvolat	cda957b4b1	update netapp's rmp-serde dependency to v1.1	2023-06-13 17:34:49 +02:00
Alex Auvolat	90b2d43eb4	Merge branch 'main' into next	2023-06-13 17:14:11 +02:00
Alex Auvolat	bf19a44fd9	admin API: add missing camelCase conversions (fix #381 )	2023-06-13 16:15:50 +02:00
Alex Auvolat	7126f3e1d1	garage key import: add checks and `--yes` CLI flag (fix #278 )	2023-06-13 15:56:48 +02:00
Alex Auvolat	942c1f1bfe	multipart uploads: save timestamp	2023-06-13 10:48:22 +02:00
Alex	0a06fda0da	Merge pull request 'Fix #204 (full Multipart Uploads semantics)' (#553 ) from nlnet-task1 into next Reviewed-on: https://git.deuxfleurs.fr/Deuxfleurs/garage/pulls/553	2023-06-09 15:34:09 +00:00
Alex Auvolat	3d477906d4	properly delete multipart uploads after completion	2023-06-09 17:13:27 +02:00
Alex Auvolat	e645bbd3ce	smoke test: add multipart upload test with part re-upload	2023-06-09 16:23:37 +02:00
Alex Auvolat	58563ed700	Add multipart upload using aws s3api	2023-06-09 16:23:37 +02:00
Alex Auvolat	a6cc563bdd	UploadPart: automatic cleanup of version (and reference blocked) when interrupted	2023-06-09 16:23:37 +02:00
Alex Auvolat	c14d3735e5	Add test for multipart uploads and fix part renumbering	2023-06-09 16:23:37 +02:00
Alex Auvolat	53bf2f070c	undo sort_key() returning Cow	2023-06-09 16:23:37 +02:00
Alex Auvolat	412ab77b08	comments and clippy lint fixes	2023-06-09 16:23:37 +02:00
Alex Auvolat	511e07ecd4	fix mpu counter (add missing workers) and report info at appropriate places	2023-06-09 16:23:37 +02:00
Alex Auvolat	4ea53dc759	Add multipart upload repair	2023-06-09 16:23:37 +02:00
Alex Auvolat	058518c22b	refactor repair workers with a trait	2023-06-09 16:23:37 +02:00
Alex Auvolat	8644376ac2	fix test; simplify code	2023-06-09 16:23:37 +02:00
Alex Auvolat	7ad7dae5d4	fix s3 list test	2023-06-09 16:23:37 +02:00
Alex Auvolat	75a0e01372	fix online repair	2023-06-09 16:23:37 +02:00
Alex Auvolat	bb176ebcb8	cargo fmt	2023-06-09 16:23:37 +02:00
Alex Auvolat	c1e1764f17	move git-version dependency to main crate to reduce rebuilds	2023-06-09 16:23:37 +02:00
Alex Auvolat	87be8eeb93	updaet block admin for new multipartupload models	2023-06-09 16:23:37 +02:00
Alex Auvolat	82e75c0e29	Adapt S3 API code to use new multipart upload models - Create and PutPart - completemultipartupload - upload part copy - list_parts	2023-06-09 16:23:37 +02:00
Alex Auvolat	38d6ac4295	New multipart upload table layout	2023-06-09 16:23:37 +02:00
Alex Auvolat	6005491cd8	Use Cow<[u8]> for sort keys	2023-06-09 16:23:37 +02:00
Alex Auvolat	ea3bfd2ab1	Minio tests for multipart upload behaviour: - upload part renumbering test - part skipping test	2023-06-09 16:23:37 +02:00
Alex Auvolat	e7e164a280	Make fsync an option for meta and data	2023-06-09 16:23:21 +02:00
Alex Auvolat	1e466b11eb	Revert integration tests to using Sled as LMDB causes failures	2023-06-09 13:23:08 +02:00
Alex Auvolat	865f0c7d0c	Add LMDB to debug builds	2023-06-09 12:04:28 +02:00
Alex Auvolat	906fe78b24	Integration tests: print logs when fails	2023-06-09 12:03:44 +02:00
Alex Auvolat	8a74e1c2bd	Split garage/admin.rs into smaller files	2023-06-06 15:39:15 +02:00
Alex Auvolat	19639705e6	Mark sled as deprecated, make lmdb default, and improve sqlite and lmdb defaults	2023-05-17 14:30:53 +02:00
Alex Auvolat	351d734e6c	Merge branch 'main' into next	2023-05-09 12:40:08 +02:00
Alex Auvolat	a1fcf1b175	Merge branch 'main' into next	2023-04-25 16:58:57 +02:00
Alex Auvolat	fa78d806e3	Merge branch 'main' into next	2023-04-25 12:34:26 +02:00
Alex Auvolat	654999e254	Update Cargo.nix	2023-01-26 15:50:54 +01:00
Jonathan Davies	db56d4658f	util/Cargo.toml: Updated rmp-serde from 0.15 to 1.1.	2023-01-26 11:03:43 +00:00
Alex Auvolat	12a4e1f303	Merge branch 'optimal-layout' into next	2023-01-11 17:50:42 +01:00
Alex Auvolat	84b4a868e3	Migration of cluster layout from v0.8 to v0.9	2023-01-11 17:47:46 +01:00
Alex	4f409f73dc	Merge pull request 'Changed all instances of assignation to assignment' (#465 ) from jpds/garage:assignments-correction into next Reviewed-on: https://git.deuxfleurs.fr/Deuxfleurs/garage/pulls/465	2023-01-11 16:05:27 +00:00
Mendes	597d64b31a	change in gitignore	2023-01-09 16:06:47 +01:00
Mendes	e3cc7a89b0	First draft of t a preprint describing the layout computation algorithm	2023-01-09 16:05:20 +01:00
Jonathan Davies	cb07e6145c	Changed all instances of assignation to assignment.	2023-01-05 11:09:25 +00:00
Alex	80e4abb98d	Merge pull request 'Changed all instances of 'key new' to 'key create' to make it the same as the bucket commands.' (#459 ) from jpds/garage:key-create-standardize into next Reviewed-on: https://git.deuxfleurs.fr/Deuxfleurs/garage/pulls/459	2023-01-04 10:35:49 +00:00
Alex Auvolat	570e5e5bbb	Merge branch 'main' into next	2023-01-04 11:34:43 +01:00
Jonathan Davies	8be862aa19	Changed all instances of 'key new' to 'key create' to make it consistent as bucket commands issued normally around the same time.	2023-01-03 11:11:12 +00:00
Alex	6e44369cbc	Merge pull request 'Optimal layout assignation algorithm' (#296 ) from optimal-layout into next Reviewed-on: https://git.deuxfleurs.fr/Deuxfleurs/garage/pulls/296	2022-12-11 17:41:53 +00:00
Alex Auvolat	2c2e65ad8b	Merge commit 'ec12d6c' into next	2022-12-11 18:41:15 +01:00
Alex Auvolat	9d83364ad9	itertools .unique() doesn't require sorted items	2022-12-11 18:30:02 +01:00
Alex Auvolat	ec12d6c8dd	Slightly simplify code at places	2022-11-08 16:15:45 +01:00
Alex Auvolat	217abdca18	Fix HTTP return code	2022-11-08 15:38:53 +01:00
Alex Auvolat	fc2729cd81	Fix integration test	2022-11-08 15:19:46 +01:00
Alex Auvolat	d75b37b018	Return more info when layout's .check() fails, fix compilation, fix test	2022-11-08 14:58:39 +01:00
Alex Auvolat	73a4ca8b15	Use bytes as capacity units	2022-11-07 21:12:11 +01:00
Alex Auvolat	fd5bc142b5	Ensure .sort() is called before counting unique items	2022-11-07 20:29:25 +01:00
Alex Auvolat	ea5afc2511	Style improvements	2022-11-07 20:11:30 +01:00
Alex Auvolat	28d7a49f63	Merge branch 'main' into optimal-layout	2022-11-07 12:20:59 +01:00
Alex Auvolat	3039bb5d43	rm .gitattributes	2022-10-13 12:40:42 +02:00
Mendes	bcdd1e0c33	Added some comment	2022-10-11 18:29:21 +02:00
Mendes	e5664c9822	Improved the statistics displayed in layout show corrected a few bugs	2022-10-11 17:17:13 +02:00
Mendes	4abab246f1	cargo fmt	2022-10-10 17:21:13 +02:00
Mendes	fcf9ac674a	Tests written in layout.rs added staged_parameters to ClusterLayout removed the serde(default) -> will need a migration function	2022-10-10 17:19:25 +02:00
Mendes	911eb17bd9	corrected warnings of cargo clippy	2022-10-06 14:53:57 +02:00
Mendes	9407df60cc	Corrected two bugs: - self.node_id_vec was not properly updated when the previous ring was empty - ClusterLayout::merge was not considering changes in the layout parameters	2022-10-06 12:54:51 +02:00
Mendes	a951b6c452	Added a CLI command to update the parameters for the layout computation (for now, only the zone redundancy)	2022-10-05 16:04:19 +02:00
Mendes	ceac3713d6	modifications in several files to : - have consistent error return types - store the zone redundancy in a Lww - print the error and message in the CLI (TODO: for the server Api, should msg be returned in the body response?)	2022-10-05 15:29:48 +02:00
Mendes	829f815a89	Merge remote-tracking branch 'origin/main' into optimal-layout	2022-10-04 18:14:49 +02:00
Mendes	99f96b9564	deleted zone_redundancy from System struct	2022-10-04 18:09:24 +02:00
Mendes	bd842e1388	Correction of a few bugs in the tests, modification of ClusterLayout::check	2022-09-22 19:30:01 +02:00
Mendes	7f3249a237	New version of the algorithm that calculate the layout. It takes as paramters the replication factor and the zone redundancy, computes the largest partition size reachable with these constraints, and among the possible assignation with this partition size, it computes the one that moves the least number of partitions compared to the previous assignation. This computation uses graph algorithms defined in graph_algo.rs	2022-09-21 14:39:59 +02:00
Mendes	c4adbeed51	Added the section with description proofs of the parametric assignment computation in the optimal layout report	2022-09-10 13:51:12 +02:00
Mendes	d38fb6c250	ignore log files in commit	2022-09-08 12:43:33 +02:00
Mendes	81083dd415	Added a first draft version of the algorithm and analysis for the non-strict mode.	2022-08-19 21:21:41 +02:00
Mendes	7b2c065c82	Merge branch 'optimal-layout' of https://git.deuxfleurs.fr/Deuxfleurs/garage into optimal-layout	2022-07-19 13:30:49 +02:00
Mendes	03e3a1bd15	Added the latex report on the optimal layout algorithm	2022-07-18 22:35:29 +02:00
Alex Auvolat	617f28bfa4	Correct small formatting issue	2022-05-05 14:21:57 +02:00
Mendes	948ff93cf1	Corrected the warnings and errors issued by cargo clippy	2022-05-01 16:05:39 +02:00
Alex Auvolat	3ba2c5b424	updated cargo.lock	2022-05-01 10:11:43 +02:00
Alex Auvolat	2aeaddd5e2	Apply cargo fmt	2022-05-01 09:57:05 +02:00
Alex Auvolat	c1d1646c4d	Change the way new layout assignations are computed. The function now computes an optimal assignation (with respect to partition size) that minimizes the distance to the former assignation, using flow algorithms. This commit was written by Mendes Oulamara <mendes.oulamara@pm.me>	2022-05-01 09:54:19 +02:00