Merge pull request 'RPC performance changes' (#387) from configurable-timeouts into main

Reviewed-on: https://git.deuxfleurs.fr/Deuxfleurs/garage/pulls/387

.drone.yml

@@ -2,68 +2,26 @@
 kind: pipeline
 name: default
 
-workspace:
-  base: /drone/garage
-
-volumes:
-- name: nix_store
-  host:
-    path: /var/lib/drone/nix
-- name: nix_config
-  temp: {}
-
-environment:
-  HOME: /drone/garage
+node:
+  nix-daemon: 1
 
 steps:
-  - name: setup nix
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+  - name: check formatting
+    image: nixpkgs/nix:nixos-22.05
     commands:
-      - cp nix/nix.conf /etc/nix/nix.conf
-      - nix-build --no-build-output --no-out-link shell.nix --arg release false  -A inputDerivation
-
-  - name: code quality
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
-    commands:
-      - nix-shell --arg release false --run "cargo fmt -- --check"
-      - nix-shell --arg release false --run "cargo clippy -- --deny warnings"
+      - nix-shell --attr rust --run "cargo fmt -- --check"
 
   - name: build
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     commands:
-      - nix-build --no-build-output --option log-lines 100 --argstr target x86_64-unknown-linux-musl --arg release false --argstr git_version $DRONE_COMMIT
+      - nix-build --no-build-output --attr clippy.amd64 --argstr git_version ${DRONE_TAG:-$DRONE_COMMIT}
 
   - name: unit + func tests
-    image: nixpkgs/nix:nixos-21.05
+    image: nixpkgs/nix:nixos-22.05
     environment:
       GARAGE_TEST_INTEGRATION_EXE: result/bin/garage
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
     commands:
-      - |
-        nix-build \
-          --no-build-output \
-          --option log-lines 100 \
-          --argstr target x86_64-unknown-linux-musl \
-          --argstr compileMode test
+      - nix-build --no-build-output --attr test.amd64
       - ./result/bin/garage_api-*
       - ./result/bin/garage_model-*
       - ./result/bin/garage_rpc-*
@@ -73,16 +31,11 @@ steps:
       - ./result/bin/garage-*
       - ./result/bin/integration-*
 
-  - name: smoke-test
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+  - name: integration tests
+    image: nixpkgs/nix:nixos-22.05
     commands:
-      - nix-build --no-build-output --argstr target x86_64-unknown-linux-musl --arg release false --argstr git_version $DRONE_COMMIT
-      - nix-shell --arg release false --run ./script/test-smoke.sh || (cat /tmp/garage.log; false)
+      - nix-build --no-build-output --attr clippy.amd64 --argstr git_version ${DRONE_TAG:-$DRONE_COMMIT}
+      - nix-shell --attr integration --run ./script/test-smoke.sh || (cat /tmp/garage.log; false)
 
 trigger:
   event:
@@ -92,78 +45,39 @@ trigger:
   - tag
   - cron
 
-node:
-  nix: 1
-
 ---
 kind: pipeline
 type: docker
-name: release-linux-x86_64
+name: release-linux-amd64
 
-volumes:
-- name: nix_store
-  host:
-    path: /var/lib/drone/nix
-- name: nix_config
-  temp: {}
-
-environment:
-  TARGET: x86_64-unknown-linux-musl
+node:
+  nix-daemon: 1
 
 steps:
-  - name: setup nix
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
-    commands:
-      - cp nix/nix.conf /etc/nix/nix.conf
-      - nix-build --no-build-output --no-out-link shell.nix -A inputDerivation
-
   - name: build
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     commands:
-      - nix-build --no-build-output --argstr target $TARGET --arg release true --argstr git_version $DRONE_COMMIT
+      - nix-build --no-build-output --attr pkgs.amd64.release --argstr git_version ${DRONE_TAG:-$DRONE_COMMIT}
+      - nix-shell --attr rust --run "./script/not-dynamic.sh result/bin/garage"
 
   - name: integration
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     commands:
-      - nix-shell --run ./script/test-smoke.sh || (cat /tmp/garage.log; false)
+      - nix-shell --attr integration --run ./script/test-smoke.sh || (cat /tmp/garage.log; false)
 
   - name: push static binary
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     environment:
       AWS_ACCESS_KEY_ID:
         from_secret: garagehq_aws_access_key_id
       AWS_SECRET_ACCESS_KEY:
         from_secret: garagehq_aws_secret_access_key
+      TARGET: "x86_64-unknown-linux-musl"
     commands:
-      - nix-shell --arg rust false --arg integration false --run "to_s3"
+      - nix-shell --attr release --run "to_s3"
 
   - name: docker build and publish
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     environment:
       DOCKER_AUTH:
         from_secret: docker_auth
@@ -174,7 +88,7 @@ steps:
       - mkdir -p /kaniko/.docker
       - echo $DOCKER_AUTH > /kaniko/.docker/config.json
       - export CONTAINER_TAG=${DRONE_TAG:-$DRONE_COMMIT}
-      - nix-shell --arg rust false --arg integration false --run "to_docker"
+      - nix-shell --attr release --run "to_docker"
 
 
 trigger:
@@ -182,78 +96,39 @@ trigger:
   - promote
   - cron
 
-node:
-  nix: 1
-
 ---
 kind: pipeline
 type: docker
-name: release-linux-i686
+name: release-linux-i386
 
-volumes:
-- name: nix_store
-  host:
-    path: /var/lib/drone/nix
-- name: nix_config
-  temp: {}
-
-environment:
-  TARGET: i686-unknown-linux-musl
+node:
+  nix-daemon: 1
 
 steps:
-  - name: setup nix
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
-    commands:
-      - cp nix/nix.conf /etc/nix/nix.conf
-      - nix-build --no-build-output --no-out-link shell.nix -A inputDerivation
-
   - name: build
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     commands:
-      - nix-build --no-build-output --argstr target $TARGET --arg release true --argstr git_version $DRONE_COMMIT
+      - nix-build --no-build-output --attr pkgs.i386.release --argstr git_version ${DRONE_TAG:-$DRONE_COMMIT}
+      - nix-shell --attr rust --run "./script/not-dynamic.sh result/bin/garage"
 
   - name: integration
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     commands:
-      - nix-shell --run ./script/test-smoke.sh || (cat /tmp/garage.log; false)
+      - nix-shell --attr integration --run ./script/test-smoke.sh || (cat /tmp/garage.log; false)
 
   - name: push static binary
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     environment:
       AWS_ACCESS_KEY_ID:
         from_secret: garagehq_aws_access_key_id
       AWS_SECRET_ACCESS_KEY:
         from_secret: garagehq_aws_secret_access_key
+      TARGET: "i686-unknown-linux-musl"
     commands:
-      - nix-shell --arg rust false --arg integration false --run "to_s3"
+      - nix-shell --attr release --run "to_s3"
 
   - name: docker build and publish
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     environment:
       DOCKER_AUTH:
         from_secret: docker_auth
@@ -264,75 +139,41 @@ steps:
       - mkdir -p /kaniko/.docker
       - echo $DOCKER_AUTH > /kaniko/.docker/config.json
       - export CONTAINER_TAG=${DRONE_TAG:-$DRONE_COMMIT}
-      - nix-shell --arg rust false --arg integration false --run "to_docker"
+      - nix-shell --attr release --run "to_docker"
 
 trigger:
   event:
   - promote
   - cron
 
-node:
-  nix: 1
-
 ---
 kind: pipeline
 type: docker
-name: release-linux-aarch64
+name: release-linux-arm64
 
-volumes:
-- name: nix_store
-  host:
-    path: /var/lib/drone/nix
-- name: nix_config
-  temp: {}
-
-environment:
-  TARGET: aarch64-unknown-linux-musl
+node:
+  nix-daemon: 1
 
 steps:
-  - name: setup nix
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
-    commands:
-      - cp nix/nix.conf /etc/nix/nix.conf
-      - nix-build --no-build-output --no-out-link ./shell.nix --arg rust false --arg integration false -A inputDerivation
-
   - name: build
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     commands:
-      - nix-build --no-build-output --argstr target $TARGET --arg release true --argstr git_version $DRONE_COMMIT
+      - nix-build --no-build-output --attr pkgs.arm64.release --argstr git_version ${DRONE_TAG:-$DRONE_COMMIT}
+      - nix-shell --attr rust --run "./script/not-dynamic.sh result/bin/garage"
 
   - name: push static binary
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     environment:
       AWS_ACCESS_KEY_ID:
         from_secret: garagehq_aws_access_key_id
       AWS_SECRET_ACCESS_KEY:
         from_secret: garagehq_aws_secret_access_key
+      TARGET: "aarch64-unknown-linux-musl"
     commands:
-      - nix-shell --arg rust false --arg integration false --run "to_s3"
+      - nix-shell --attr release --run "to_s3"
 
   - name: docker build and publish
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     environment:
       DOCKER_AUTH:
         from_secret: docker_auth
@@ -343,75 +184,41 @@ steps:
       - mkdir -p /kaniko/.docker
       - echo $DOCKER_AUTH > /kaniko/.docker/config.json
       - export CONTAINER_TAG=${DRONE_TAG:-$DRONE_COMMIT}
-      - nix-shell --arg rust false --arg integration false --run "to_docker"
+      - nix-shell --attr release --run "to_docker"
 
 trigger:
   event:
   - promote
   - cron
 
-node:
-  nix: 1
-
 ---
 kind: pipeline
 type: docker
-name: release-linux-armv6l
+name: release-linux-arm
 
-volumes:
-- name: nix_store
-  host:
-    path: /var/lib/drone/nix
-- name: nix_config
-  temp: {}
-
-environment:
-  TARGET: armv6l-unknown-linux-musleabihf
+node:
+  nix-daemon: 1
 
 steps:
-  - name: setup nix
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
-    commands:
-      - cp nix/nix.conf /etc/nix/nix.conf
-      - nix-build --no-build-output --no-out-link --arg rust false --arg integration false -A inputDerivation
-
   - name: build
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     commands:
-      - nix-build --no-build-output --argstr target $TARGET --arg release true --argstr git_version $DRONE_COMMIT
+      - nix-build --no-build-output --attr pkgs.arm.release --argstr git_version ${DRONE_TAG:-$DRONE_COMMIT}
+      - nix-shell --attr rust --run "./script/not-dynamic.sh result/bin/garage"
 
   - name: push static binary
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     environment:
       AWS_ACCESS_KEY_ID:
         from_secret: garagehq_aws_access_key_id
       AWS_SECRET_ACCESS_KEY:
         from_secret: garagehq_aws_secret_access_key
+      TARGET: "armv6l-unknown-linux-musleabihf"
     commands:
-      - nix-shell --arg integration false --arg rust false --run "to_s3"
+      - nix-shell --attr release --run "to_s3"
 
   - name: docker build and publish
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
-    - name: nix_config
-      path: /etc/nix
+    image: nixpkgs/nix:nixos-22.05
     environment:
       DOCKER_AUTH:
         from_secret: docker_auth
@@ -422,32 +229,24 @@ steps:
       - mkdir -p /kaniko/.docker
       - echo $DOCKER_AUTH > /kaniko/.docker/config.json
       - export CONTAINER_TAG=${DRONE_TAG:-$DRONE_COMMIT}
-      - nix-shell --arg rust false --arg integration false --run "to_docker"
+      - nix-shell --attr release --run "to_docker"
 
 trigger:
   event:
   - promote
   - cron
 
-node:
-  nix: 1
-
 ---
 kind: pipeline
 type: docker
 name: refresh-release-page
 
-volumes:
-- name: nix_store
-  host:
-    path: /var/lib/drone/nix
+node:
+  nix-daemon: 1
 
 steps:
   - name: refresh-index
-    image: nixpkgs/nix:nixos-21.05
-    volumes:
-    - name: nix_store
-      path: /nix
+    image: nixpkgs/nix:nixos-22.05
     environment:
       AWS_ACCESS_KEY_ID:
         from_secret: garagehq_aws_access_key_id
@@ -455,24 +254,21 @@ steps:
         from_secret: garagehq_aws_secret_access_key
     commands:
       - mkdir -p /etc/nix && cp nix/nix.conf /etc/nix/nix.conf
-      - nix-shell --arg integration false --arg rust false --run "refresh_index"
+      - nix-shell --attr release --run "refresh_index"
 
 depends_on:
-  - release-linux-x86_64
-  - release-linux-i686
-  - release-linux-aarch64
-  - release-linux-armv6l
+  - release-linux-amd64
+  - release-linux-i386
+  - release-linux-arm64
+  - release-linux-arm
 
 trigger:
   event:
   - promote
   - cron
 
-node:
-  nix: 1
-
 ---
 kind: signature
-hmac: 3fc19d6f9a3555519c8405e3281b2e74289bb802f644740d5481d53df3a01fa4
+hmac: 362639b4c9541ad9bd06ff7f72b5235b2b0216bcb16eececd25285b6fe94ba6f
 
 ...

Cargo.toml

@@ -1,4 +1,5 @@
 [workspace]
+resolver = "2"
 members = [
 	"src/db",
 	"src/util",

Makefile

@@ -1,7 +1,7 @@
 .PHONY: doc all release shell run1 run2 run3
 
 all:
-	clear; cargo build --all-features
+	clear; cargo build
 
 release:
 	nix-build --arg release true

default.nix

@@ -1,147 +1,33 @@
 {
   system ? builtins.currentSystem,
-  release ? false,
-  target ? "x86_64-unknown-linux-musl",
-  compileMode ? null,
   git_version ? null,
 }:
 
 with import ./nix/common.nix;
 
-let
-  crossSystem = { config = target; };
-in let
-  log = v: builtins.trace v v;
+let 
+  pkgs = import pkgsSrc { };
+  compile = import ./nix/compile.nix;
+  build_debug_and_release = (target: {
+    debug = (compile { inherit target git_version; release = false; }).workspace.garage { compileMode = "build"; }; 
+    release = (compile { inherit target git_version; release = true; }).workspace.garage { compileMode = "build"; }; 
+  });
+  test = (rustPkgs: pkgs.symlinkJoin {
+    name ="garage-tests";
+    paths = builtins.map (key: rustPkgs.workspace.${key} { compileMode = "test"; }) (builtins.attrNames rustPkgs.workspace);
+  });
 
-  pkgs = import pkgsSrc {
-    inherit system crossSystem;
-    overlays = [ cargo2nixOverlay ];
+in {
+  pkgs = {
+    amd64 = build_debug_and_release "x86_64-unknown-linux-musl";
+    i386 = build_debug_and_release "i686-unknown-linux-musl";
+    arm64 = build_debug_and_release "aarch64-unknown-linux-musl";
+    arm = build_debug_and_release "armv6l-unknown-linux-musleabihf";
   };
-
-
-  /*
-   Rust and Nix triples are not the same. Cargo2nix has a dedicated library
-   to convert Nix triples to Rust ones. We need this conversion as we want to
-   set later options linked to our (rust) target in a generic way. Not only
-   the triple terminology is different, but also the "roles" are named differently.
-   Nix uses a build/host/target terminology where Nix's "host" maps to Cargo's "target".
-  */
-  rustTarget = log (pkgs.rustBuilder.rustLib.rustTriple pkgs.stdenv.hostPlatform);
-
-  /*
-   Cargo2nix is built for rustOverlay which installs Rust from Mozilla releases.
-   We want our own Rust to avoid incompatibilities, like we had with musl 1.2.0.
-   rustc was built with musl < 1.2.0 and nix shipped musl >= 1.2.0 which lead to compilation breakage.
-   So we want a Rust release that is bound to our Nix repository to avoid these problems.
-   See here for more info: https://musl.libc.org/time64.html
-   Because Cargo2nix does not support the Rust environment shipped by NixOS,
-   we emulate the structure of the Rust object created by rustOverlay.
-   In practise, rustOverlay ships rustc+cargo in a single derivation while
-   NixOS ships them in separate ones. We reunite them with symlinkJoin.
-   */
-  rustChannel = pkgs.symlinkJoin {
-    name ="rust-channel";
-    paths = [ 
-      pkgs.rustPlatform.rust.rustc
-      pkgs.rustPlatform.rust.cargo
-    ];
+  test = {
+    amd64 = test (compile { inherit git_version; target = "x86_64-unknown-linux-musl"; });
   };
-
-  /*
-   Cargo2nix provides many overrides by default, you can take inspiration from them:
-   https://github.com/cargo2nix/cargo2nix/blob/master/overlay/overrides.nix
-
-   You can have a complete list of the available options by looking at the overriden object, mkcrate:
-   https://github.com/cargo2nix/cargo2nix/blob/master/overlay/mkcrate.nix
-  */
-  overrides = pkgs.rustBuilder.overrides.all ++ [
-    /*
-     [1] We need to alter Nix hardening to be able to statically compile: PIE,
-     Position Independent Executables seems to be supported only on amd64. Having
-     this flags set either make our executables crash or compile as dynamic on many platforms.
-     In the following section codegenOpts, we reactive it for the supported targets
-     (only amd64 curently) through the `-static-pie` flag. PIE is a feature used
-     by ASLR, which helps mitigate security issues.
-     Learn more about Nix Hardening: https://github.com/NixOS/nixpkgs/blob/master/pkgs/build-support/cc-wrapper/add-hardening.sh
-
-     [2] We want to inject the git version while keeping the build deterministic.
-     As we do not want to consider the .git folder as part of the input source,
-     we ask the user (the CI often) to pass the value to Nix.
-    */
-    (pkgs.rustBuilder.rustLib.makeOverride {
-      name = "garage_rpc";
-      overrideAttrs = drv:
-        /* [1] */ { hardeningDisable = [ "pie" ]; }
-        //
-        /* [2] */ (if git_version != null then {
-          preConfigure = ''
-            ${drv.preConfigure or ""}
-            export GIT_VERSION="${git_version}"
-          '';
-        } else {});
-    })
-
-    /*
-     We ship some parts of the code disabled by default by putting them behind a flag.
-     It speeds up the compilation (when the feature is not required) and released crates have less dependency by default (less attack surface, disk space, etc.).
-     But we want to ship these additional features when we release Garage.
-     In the end, we chose to exclude all features from debug builds while putting (all of) them in the release builds.
-     Currently, the only feature of Garage is kubernetes-discovery from the garage_rpc crate.
-    */
-    (pkgs.rustBuilder.rustLib.makeOverride {
-      name = "garage_rpc";
-      overrideArgs = old:
-        {
-          features = if release then [ "kubernetes-discovery" ] else [];
-        };
-    })
-  ];
-
-  packageFun = import ./Cargo.nix;
-
-  /*
-    We compile fully static binaries with musl to simplify deployment on most systems.
-    When possible, we reactivate PIE hardening (see above).
-
-    Also, if you set the RUSTFLAGS environment variable, the following parameters will
-    be ignored.
-
-    For more information on static builds, please refer to Rust's RFC 1721.
-    https://rust-lang.github.io/rfcs/1721-crt-static.html#specifying-dynamicstatic-c-runtime-linkage
-  */
-
-  codegenOpts = {
-   "armv6l-unknown-linux-musleabihf" = [ "target-feature=+crt-static" "link-arg=-static" ]; /* compile as dynamic with static-pie */
-   "aarch64-unknown-linux-musl" = [ "target-feature=+crt-static" "link-arg=-static" ]; /* segfault with static-pie */
-   "i686-unknown-linux-musl" = [ "target-feature=+crt-static" "link-arg=-static" ]; /* segfault with static-pie */
-   "x86_64-unknown-linux-musl" = [ "target-feature=+crt-static" "link-arg=-static-pie" ];
+  clippy = {
+    amd64 = (compile { inherit git_version; compiler = "clippy"; }).workspace.garage { compileMode = "build"; } ;
   };
-
-  /*
-   The following definition is not elegant as we use a low level function of Cargo2nix
-   that enables us to pass our custom rustChannel object. We need this low level definition
-   to pass Nix's Rust toolchains instead of Mozilla's one.
-
-   target is mandatory but must be kept to null to allow cargo2nix to set it to the appropriate value
-   for each crate.
-  */ 
-  rustPkgs = pkgs.rustBuilder.makePackageSet {
-    inherit packageFun rustChannel release codegenOpts;
-    packageOverrides = overrides;
-    target = null;
-
-    buildRustPackages = pkgs.buildPackages.rustBuilder.makePackageSet {
-      inherit rustChannel packageFun codegenOpts;
-      packageOverrides = overrides;
-      target = null;
-    };
-  };
-
-
-in
-  if compileMode == "test"
-    then pkgs.symlinkJoin {
-      name ="garage-tests";
-      paths = builtins.map (key: rustPkgs.workspace.${key} { inherit compileMode; }) (builtins.attrNames rustPkgs.workspace);
-    }
-    else rustPkgs.workspace.garage { inherit compileMode; }
+}

doc/book/cookbook/exposing-websites.md

@@ -5,12 +5,14 @@ weight = 25
 
 ## Configuring a bucket for website access
 
-There are two methods to expose buckets as website:
+There are three methods to expose buckets as website:
 
 1. using the PutBucketWebsite S3 API call, which is allowed for access keys that have the owner permission bit set
 
 2. from the Garage CLI, by an adminstrator of the cluster
 
+3. using the Garage administration API
+
 The `PutBucketWebsite` API endpoint [is documented](https://docs.aws.amazon.com/AmazonS3/latest/API/API_PutBucketWebsite.html) in the official AWS docs.
 This endpoint can also be called [using `aws s3api`](https://docs.aws.amazon.com/cli/latest/reference/s3api/put-bucket-website.html) on the command line.
 The website configuration supported by Garage is only a subset of the possibilities on Amazon S3: redirections are not supported, only the index document and error document can be specified.

doc/book/cookbook/from-source.md

@@ -20,40 +20,76 @@ sudo apt-get update
 sudo apt-get install build-essential
 ```
 
-## Using source from `crates.io`
-
-Garage's source code is published on `crates.io`, Rust's official package repository.
-This means you can simply ask `cargo` to download and build this source code for you:
-
-```bash
-cargo install garage
-```
-
-That's all, `garage` should be in `$HOME/.cargo/bin`.
-
-You can add this folder to your `$PATH` or copy the binary somewhere else on your system.
-For instance:
-
-```bash
-sudo cp $HOME/.cargo/bin/garage /usr/local/bin/garage
-```
-
-
-## Using source from the Gitea repository
+## Building from source from the Gitea repository
 
 The primary location for Garage's source code is the
-[Gitea repository](https://git.deuxfleurs.fr/Deuxfleurs/garage).
+[Gitea repository](https://git.deuxfleurs.fr/Deuxfleurs/garage),
+which contains all of the released versions as well as the code
+for the developpement of the next version.
 
-Clone the repository and build Garage with the following commands:
+Clone the repository and enter it as follows:
 
 ```bash
 git clone https://git.deuxfleurs.fr/Deuxfleurs/garage.git
 cd garage
-cargo build
 ```
 
-Be careful, as this will make a debug build of Garage, which will be extremely slow!
-To make a release build, invoke `cargo build --release` (this takes much longer).
+If you wish to build a specific version of Garage, check out the corresponding tag. For instance:
 
-The binaries built this way are found in `target/{debug,release}/garage`.
+```bash
+git tag  				# List available tags
+git checkout v0.8.0		# Change v0.8.0 with the version you wish to build
+```
 
+Otherwise you will be building a developpement build from the `main` branch
+that includes all of the changes to be released in the next version.
+Be careful that such a build might be unstable or contain bugs,
+and could be incompatible with nodes that run stable versions of Garage.
+
+Finally, build Garage with the following command:
+
+```bash
+cargo build --release
+```
+
+The binary built this way can now be found in `target/release/garage`.
+You may simply copy this binary to somewhere in your `$PATH` in order to
+have the `garage` command available in your shell, for instance:
+
+```bash
+sudo cp target/release/garage /usr/local/bin/garage
+```
+
+If you are planning to develop Garage,
+you might be interested in producing debug builds, which compile faster but run slower:
+this can be done by removing the `--release` flag, and the resulting build can then
+be found in `target/debug/garage`.
+
+## List of available Cargo feature flags
+
+Garage supports a number of compilation options in the form of Cargo feature flags,
+which can be used to provide builds adapted to your system and your use case.
+To produce a build with a given set of features, invoke the `cargo build` command
+as follows:
+
+```bash
+# This will build the default feature set plus feature1, feature2 and feature3
+cargo build --release --features feature1,feature2,feature3
+# This will build ONLY feature1, feature2 and feature3
+cargo build --release --no-default-features \
+            --features feature1,feature2,feature3
+```
+
+The following feature flags are available in v0.8.0:
+
+| Feature flag | Enabled | Description |
+| ------------ | ------- | ----------- |
+| `bundled-libs` | *by default* | Use bundled version of sqlite3, zstd, lmdb and libsodium |
+| `system-libs` | optional | Use system version of sqlite3, zstd, lmdb and libsodium<br>if available (exclusive with `bundled-libs`, build using<br>`cargo build --no-default-features --features system-libs`) |
+| `k2v` | optional | Enable the experimental K2V API (if used, all nodes on your<br>Garage cluster must have it enabled as well) |
+| `kubernetes-discovery` | optional | Enable automatic registration and discovery<br>of cluster nodes through the Kubernetes API |
+| `metrics` | *by default* | Enable collection of metrics in Prometheus format on the admin API |
+| `telemetry-otlp` | optional | Enable collection of execution traces using OpenTelemetry |
+| `sled` | *by default* | Enable using Sled to store Garage's metadata |
+| `lmdb` | optional | Enable using LMDB to store Garage's metadata |
+| `sqlite` | optional | Enable using Sqlite3 to store Garage's metadata |

doc/book/cookbook/reverse-proxy.md

@@ -280,3 +280,25 @@ Traefik's caching middleware is only available on [entreprise version](https://d
 [http.middlewares]
   [http.middlewares.gzip_compress.compress]
 ```
+
+## Caddy
+
+Your Caddy configuration can be as simple as:
+
+```caddy
+s3.garage.tld, *.s3.garage.tld {
+	reverse_proxy localhost:3900 192.168.1.2:3900 example.tld:3900
+}
+
+*.web.garage.tld {
+	reverse_proxy localhost:3902 192.168.1.2:3900 example.tld:3900
+}
+
+admin.garage.tld {
+	reverse_proxy localhost:3903
+}
+```
+
+But at the same time, the `reverse_proxy` is very flexible.
+For a production deployment, you should [read its documentation](https://caddyserver.com/docs/caddyfile/directives/reverse_proxy) as it supports features like DNS discovery of upstreams, load balancing with checks, streaming parameters, etc.
+

doc/book/design/benchmarks/index.md

@@ -1,6 +1,6 @@
 +++
 title = "Benchmarks"
-weight = 10
+weight = 40
 +++
 
 With Garage, we wanted to build a software defined storage service that follow the [KISS principle](https://en.wikipedia.org/wiki/KISS_principle),

doc/book/design/goals.md

@@ -1,13 +1,13 @@
 +++
 title = "Goals and use cases"
-weight = 5
+weight = 10
 +++
 
 ## Goals and non-goals
 
 Garage is a lightweight geo-distributed data store that implements the
 [Amazon S3](https://docs.aws.amazon.com/AmazonS3/latest/API/Welcome.html)
-object storage protocole. It enables applications to store large blobs such
+object storage protocol. It enables applications to store large blobs such
 as pictures, video, images, documents, etc., in a redundant multi-node
 setting. S3 is versatile enough to also be used to publish a static
 website.

doc/book/design/internals.md

@@ -20,6 +20,49 @@ In the meantime, you can find some information at the following links:
 - [an old design draft](@/documentation/working-documents/design-draft.md)
 
 
+## Request routing logic
+
+Data retrieval requests to Garage endpoints (S3 API and websites) are resolved 
+to an individual object in a bucket. Since objects are replicated to multiple nodes 
+Garage must ensure consistency before answering the request.
+
+### Using quorum to ensure consistency
+
+Garage ensures consistency by attempting to establish a quorum with the
+data nodes responsible for the object. When a majority of the data nodes
+have provided metadata on a object Garage can then answer the request.
+
+When a request arrives Garage will, assuming the recommended 3 replicas, perform the following actions:
+
+- Make a request to the two preferred nodes for object metadata
+- Try the third node if one of the two initial requests fail
+- Check that the metadata from at least 2 nodes match
+- Check that the object hasn't been marked deleted
+- Answer the request with inline data from metadata if object is small enough
+- Or get data blocks from the preferred nodes and answer using the assembled object
+
+Garage dynamically determines which nodes to query based on health, preference, and 
+which nodes actually host a given data. Garage has no concept of "primary" so any 
+healthy node with the data can be used as long as a quorum is reached for the metadata.
+
+### Node health
+
+Garage keeps a TCP session open to each node in the cluster and periodically pings them. If a connection
+cannot be established, or a node fails to answer a number of pings, the target node is marked as failed.
+Failed nodes are not used for quorum or other internal requests.
+
+### Node preference
+
+Garage prioritizes which nodes to query according to a few criteria:
+
+- A node always prefers itself if it can answer the request
+- Then the node prioritizes nodes in the same zone
+- Finally the nodes with the lowest latency are prioritized 
+
+
+For further reading on the cluster structure look at the [gateway](@/documentation/cookbook/gateways.md) 
+and [cluster layout management](@/documentation/reference-manual/layout.md) pages.
+
 ## Garbage collection
 
 A faulty garbage collection procedure has been the cause of

doc/book/design/related-work.md

@@ -1,6 +1,6 @@
 +++
 title = "Related work"
-weight = 15
+weight = 50
 +++
 
 ## Context

doc/book/quick-start/_index.md

@@ -9,6 +9,15 @@ Let's start your Garage journey!
 In this chapter, we explain how to deploy Garage as a single-node server
 and how to interact with it.
 
+## What is Garage?
+
+Before jumping in, you might be interested in reading the following pages:
+
+- [Goals and use cases](@/documentation/design/goals.md)
+- [List of features](@/documentation/reference-manual/features.md)
+
+## Scope of this tutorial
+
 Our goal is to introduce you to Garage's workflows.
 Following this guide is recommended before moving on to
 [configuring a multi-node cluster](@/documentation/cookbook/real-world.md).

doc/book/reference-manual/admin-api.md

@@ -1,6 +1,6 @@
 +++
 title = "Administration API"
-weight = 16
+weight = 60
 +++
 
 The Garage administration API is accessible through a dedicated server whose

doc/book/reference-manual/cli.md

@@ -1,6 +1,6 @@
 +++
 title = "Garage CLI"
-weight = 15
+weight = 30
 +++
 
 The Garage CLI is mostly self-documented. Make use of the `help` subcommand

doc/book/reference-manual/configuration.md

@@ -1,6 +1,6 @@
 +++
 title = "Configuration file format"
-weight = 5
+weight = 20
 +++
 
 Here is an example `garage.toml` configuration file that illustrates all of the possible options:
@@ -10,7 +10,6 @@ metadata_dir = "/var/lib/garage/meta"
 data_dir = "/var/lib/garage/data"
 
 block_size = 1048576
-block_manager_background_tranquility = 2
 
 replication_mode = "3"
 
@@ -87,17 +86,6 @@ files will remain available. This however means that chunks from existing files
 will not be deduplicated with chunks from newly uploaded files, meaning you
 might use more storage space that is optimally possible.
 
-### `block_manager_background_tranquility`
-
-This parameter tunes the activity of the background worker responsible for
-resyncing data blocks between nodes. The higher the tranquility value is set,
-the more the background worker will wait between iterations, meaning the load
-on the system (including network usage between nodes) will be reduced. The
-minimal value for this parameter is `0`, where the background worker will
-allways work at maximal throughput to resynchronize blocks. The default value
-is `2`, where the background worker will try to spend at most 1/3 of its time
-working, and 2/3 sleeping in order to reduce system load.
-
 ### `replication_mode`
 
 Garage supports the following replication modes:

doc/book/reference-manual/features.md

@@ -0,0 +1,125 @@
++++
+title = "List of Garage features"
+weight = 10
++++
+
+
+### S3 API
+
+The main goal of Garage is to provide an object storage service that is compatible with the
+[S3 API](https://docs.aws.amazon.com/AmazonS3/latest/API/Welcome.html) from Amazon Web Services.
+We try to adhere as strictly as possible to the semantics of the API as implemented by Amazon
+and other vendors such as Minio or CEPH.
+
+Of course Garage does not implement the full span of API endpoints that AWS S3 does;
+the exact list of S3 features implemented by Garage can be found [on our S3 compatibility page](@/documentation/reference-manual/s3-compatibility.md).
+
+### Geo-distribution
+
+Garage allows you to store copies of your data in multiple geographical locations in order to maximize resilience
+to adverse events, such as network/power outages or hardware failures.
+This allows Garage to run very well even at home, using consumer-grade Internet connectivity
+(such as FTTH) and power, as long as cluster nodes can be spawned at several physical locations.
+Garage exploits knowledge of the capacity and physical location of each storage node to design
+a storage plan that best exploits the available storage capacity while satisfying the geo-distributed replication constraint.
+
+To learn more about geo-distributed Garage clusters,
+read our documentation on [setting up a real-world deployment](@/documentation/cookbook/real-world.md).
+
+### Standalone/self-contained
+
+Garage is extremely simple to deploy, and does not depend on any external service to run.
+This makes setting up and administering storage clusters, we hope, as easy as it could be.
+
+### Flexible topology
+
+A Garage cluster can very easily evolve over time, as storage nodes are added or removed.
+Garage will automatically rebalance data between nodes as needed to ensure the desired number of copies.
+Read about cluster layout management [here](@/documentation/reference-manual/layout.md).
+
+### No RAFT slowing you down
+
+It might seem strange to tout the absence of something as a desirable feature,
+but this is in fact a very important point! Garage does not use RAFT or another
+consensus algorithm internally to order incoming requests: this means that all requests
+directed to a Garage cluster can be handled independently of one another instead
+of going through a central bottleneck (the leader node).
+As a consequence, requests can be handled much faster, even in cases where latency
+between cluster nodes is important (see our [benchmarks](@/documentation/design/benchmarks/index.md) for data on this).
+This is particularly usefull when nodes are far from one another and talk to one other through standard Internet connections.
+
+### Several replication modes
+
+Garage supports a variety of replication modes, with 1 copy, 2 copies or 3 copies of your data,
+and with various levels of consistency, in order to adapt to a variety of usage scenarios.
+Read our reference page on [supported replication modes](@/documentation/reference-manual/configuration.md#replication-mode)
+to select the replication mode best suited to your use case (hint: in most cases, `replication_mode = "3"` is what you want).
+
+### Web server for static websites
+
+A storage bucket can easily be configured to be served directly by Garage as a static web site.
+Domain names for multiple websites directly map to bucket names, making it easy to build
+a platform for your users to autonomously build and host their websites over Garage.
+Surprisingly, none of the other alternative S3 implementations we surveyed (such as Minio
+or CEPH) support publishing static websites from S3 buckets, a feature that is however
+directly inherited from S3 on AWS.
+Read more on our [dedicated documentation page](@/documentation/cookbook/exposing-websites.md).
+
+### Bucket names as aliases
+
+In Garage, a bucket may have several names, known as aliases.
+Aliases can easily be added and removed on demand:
+this allows to easily rename buckets if needed
+without having to copy all of their content, something that cannot be done on AWS.
+For buckets served as static websites, having multiple aliases for a bucket can allow
+exposing the same content under different domain names.
+
+Garage also supports bucket aliases which are local to a single user:
+this allows different users to have different buckets with the same name, thus avoiding naming collisions.
+This can be helpfull for instance if you want to write an application that creates per-user buckets with always the same name.
+
+This feature is totally invisible to S3 clients and does not break compatibility with AWS.
+
+### Cluster administration API
+
+Garage provides a fully-fledged REST API to administer your cluster programatically.
+Functionnality included in the admin API include: setting up and monitoring
+cluster nodes, managing access credentials, and managing storage buckets and bucket aliases.
+A full reference of the administration API is available [here](@/documentation/reference-manual/admin-api.md).
+
+### Metrics and traces
+
+Garage makes some internal metrics available in the Prometheus data format,
+which allows you to build interactive dashboards to visualize the load and internal state of your storage cluster.
+
+For developpers and performance-savvy administrators,
+Garage also supports exporting traces of what it does internally in OpenTelemetry format.
+This allows to monitor the time spent at various steps of the processing of requests,
+in order to detect potential performance bottlenecks.
+
+### Kubernetes and Nomad integrations
+
+Garage can automatically discover other nodes in the cluster thanks to integration
+with orchestrators such as Kubernetes and Nomad (when used with Consul).
+This eases the configuration of your cluster as it removes one step where nodes need
+to be manually connected to one another.
+
+### Support for changing IP addresses
+
+As long as all of your nodes don't thange their IP address at the same time,
+Garage should be able to tolerate nodes with changing/dynamic IP addresses,
+as nodes will regularly exchange the IP addresses of their peers and try to
+reconnect using newer addresses when existing connections are broken.
+
+### K2V API (experimental)
+
+As part of an ongoing research project, Garage can expose an experimental key/value storage API called K2V.
+K2V is made for the storage and retrieval of many small key/value pairs that need to be processed in bulk.
+This completes the S3 API with an alternative that can be used to easily store and access metadata
+related to objects stored in an S3 bucket.
+
+In the context of our research project, [Aérogramme](https://aerogramme.deuxfleurs.fr),
+K2V is used to provide metadata and log storage for operations on encrypted e-mail storage.
+
+Learn more on the specification of K2V [here](https://git.deuxfleurs.fr/Deuxfleurs/garage/src/branch/k2v/doc/drafts/k2v-spec.md)
+and on how to enable it in Garage [here](@/documentation/reference-manual/k2v.md).

doc/book/reference-manual/k2v.md

@@ -1,6 +1,6 @@
 +++
 title = "K2V"
-weight = 30
+weight = 70
 +++
 
 Starting with version 0.7.2, Garage introduces an optionnal feature, K2V,

doc/book/reference-manual/layout.md

@@ -1,6 +1,6 @@
 +++
 title = "Cluster layout management"
-weight = 10
+weight = 50
 +++
 
 The cluster layout in Garage is a table that assigns to each node a role in

doc/book/reference-manual/routing.md

@@ -1,45 +0,0 @@
-+++
-title = "Request routing logic"
-weight = 10
-+++
-
-Data retrieval requests to Garage endpoints (S3 API and websites) are resolved 
-to an individual object in a bucket. Since objects are replicated to multiple nodes 
-Garage must ensure consistency before answering the request.
-
-## Using quorum to ensure consistency
-
-Garage ensures consistency by attempting to establish a quorum with the
-data nodes responsible for the object. When a majority of the data nodes
-have provided metadata on a object Garage can then answer the request.
-
-When a request arrives Garage will, assuming the recommended 3 replicas, perform the following actions:
-
-- Make a request to the two preferred nodes for object metadata
-- Try the third node if one of the two initial requests fail
-- Check that the metadata from at least 2 nodes match
-- Check that the object hasn't been marked deleted
-- Answer the request with inline data from metadata if object is small enough
-- Or get data blocks from the preferred nodes and answer using the assembled object
-
-Garage dynamically determines which nodes to query based on health, preference, and 
-which nodes actually host a given data. Garage has no concept of "primary" so any 
-healthy node with the data can be used as long as a quorum is reached for the metadata.
-
-## Node health
-
-Garage keeps a TCP session open to each node in the cluster and periodically pings them. If a connection
-cannot be established, or a node fails to answer a number of pings, the target node is marked as failed.
-Failed nodes are not used for quorum or other internal requests.
-
-## Node preference
-
-Garage prioritizes which nodes to query according to a few criteria:
-
-- A node always prefers itself if it can answer the request
-- Then the node prioritizes nodes in the same zone
-- Finally the nodes with the lowest latency are prioritized 
-
-
-For further reading on the cluster structure look at the [gateway](@/documentation/cookbook/gateways.md) 
-and [cluster layout management](@/documentation/reference-manual/layout.md) pages.

doc/book/reference-manual/s3-compatibility.md

@@ -1,6 +1,6 @@
 +++
 title = "S3 Compatibility status"
-weight = 20
+weight = 40
 +++
 
 ## DISCLAIMER

doc/book/working-documents/design-draft.md

@@ -1,6 +1,6 @@
 +++
-title = "Design draft"
-weight = 25
+title = "Design draft (obsolete)"
+weight = 50
 +++
 
 **WARNING: this documentation is a design draft which was written before Garage's actual implementation.

doc/book/working-documents/load-balancing.md

@@ -1,6 +1,6 @@
 +++
-title = "Load balancing data"
-weight = 10
+title = "Load balancing data (obsolete)"
+weight = 60
 +++
 
 **This is being yet improved in release 0.5. The working document has not been updated yet, it still only applies to Garage 0.2 through 0.4.**

nix/common.nix

@@ -4,18 +4,16 @@ rec {
    */
   pkgsSrc = fetchTarball {
     # As of 2021-10-04
-    url ="https://github.com/NixOS/nixpkgs/archive/b27d18a412b071f5d7991d1648cfe78ee7afe68a.tar.gz";
+    url = "https://github.com/NixOS/nixpkgs/archive/b27d18a412b071f5d7991d1648cfe78ee7afe68a.tar.gz";
     sha256 = "1xy9zpypqfxs5gcq5dcla4bfkhxmh5nzn9dyqkr03lqycm9wg5cr";
   };
   cargo2nixSrc = fetchGit {
-    # As of 2022-03-17
-    url = "https://github.com/superboum/cargo2nix";
-    ref = "main";
-    rev = "bcbf3ba99e9e01a61eb83a24624419c2dd9dec64";
+    # As of 2022-08-29, stacking two patches: superboum@dedup_propagate and Alexis211@fix_fetchcrategit
+    url = "https://github.com/Alexis211/cargo2nix";
+    ref = "fix_fetchcrategit";
+    rev = "4b31c0cc05b6394916d46e9289f51263d81973b9";
   };
 
-
-
   /*
    * Shared objects
    */

nix/compile.nix

@@ -0,0 +1,244 @@
+{
+  system ? builtins.currentSystem,
+  target ? null,
+  compiler ? "rustc",
+  release ? false,
+  git_version ? null,
+}:
+
+with import ./common.nix;
+
+let
+  log = v: builtins.trace v v;
+
+  pkgs = import pkgsSrc {
+    inherit system; 
+    ${ if target == null then null else "crossSystem" } = { config = target; };
+    overlays = [ cargo2nixOverlay ];
+  };
+
+  /*
+   Rust and Nix triples are not the same. Cargo2nix has a dedicated library
+   to convert Nix triples to Rust ones. We need this conversion as we want to
+   set later options linked to our (rust) target in a generic way. Not only
+   the triple terminology is different, but also the "roles" are named differently.
+   Nix uses a build/host/target terminology where Nix's "host" maps to Cargo's "target".
+  */
+  rustTarget = log (pkgs.rustBuilder.rustLib.rustTriple pkgs.stdenv.hostPlatform);
+
+  /*
+   Cargo2nix is built for rustOverlay which installs Rust from Mozilla releases.
+   We want our own Rust to avoid incompatibilities, like we had with musl 1.2.0.
+   rustc was built with musl < 1.2.0 and nix shipped musl >= 1.2.0 which lead to compilation breakage.
+   So we want a Rust release that is bound to our Nix repository to avoid these problems.
+   See here for more info: https://musl.libc.org/time64.html
+   Because Cargo2nix does not support the Rust environment shipped by NixOS,
+   we emulate the structure of the Rust object created by rustOverlay.
+   In practise, rustOverlay ships rustc+cargo in a single derivation while
+   NixOS ships them in separate ones. We reunite them with symlinkJoin.
+   */
+  rustChannel = {
+    rustc = pkgs.symlinkJoin {
+      name = "rust-channel";
+      paths = [ 
+        pkgs.rustPlatform.rust.cargo
+        pkgs.rustPlatform.rust.rustc
+      ];
+    };
+    clippy = pkgs.symlinkJoin {
+      name = "clippy-channel";
+      paths = [ 
+        pkgs.rustPlatform.rust.cargo
+        pkgs.rustPlatform.rust.rustc
+        pkgs.clippy
+      ];
+    };
+  }.${compiler};
+
+  clippyBuilder = pkgs.writeScriptBin "clippy" ''
+    #!${pkgs.stdenv.shell}
+    . ${cargo2nixSrc + "/overlay/utils.sh"}
+    isBuildScript=
+    args=("$@")
+    for i in "''${!args[@]}"; do
+      if [ "xmetadata=" = "x''${args[$i]::9}" ]; then
+        args[$i]=metadata=$NIX_RUST_METADATA
+      elif [ "x--crate-name" = "x''${args[$i]}" ] && [ "xbuild_script_" = "x''${args[$i+1]::13}" ]; then
+        isBuildScript=1
+      fi
+    done
+    if [ "$isBuildScript" ]; then
+      args+=($NIX_RUST_BUILD_LINK_FLAGS)
+    else
+      args+=($NIX_RUST_LINK_FLAGS)
+    fi
+    touch invoke.log
+    echo "''${args[@]}" >>invoke.log
+
+    exec ${rustChannel}/bin/clippy-driver --deny warnings "''${args[@]}"
+  '';
+
+  buildEnv = (drv: {
+    rustc = drv.setBuildEnv;
+    clippy = ''
+      ${drv.setBuildEnv or "" }
+      echo
+      echo --- BUILDING WITH CLIPPY ---
+      echo 
+
+      export RUSTC=${clippyBuilder}/bin/clippy
+    '';
+  }.${compiler});
+
+  /*
+   Cargo2nix provides many overrides by default, you can take inspiration from them:
+   https://github.com/cargo2nix/cargo2nix/blob/master/overlay/overrides.nix
+
+   You can have a complete list of the available options by looking at the overriden object, mkcrate:
+   https://github.com/cargo2nix/cargo2nix/blob/master/overlay/mkcrate.nix
+  */
+  overrides = pkgs.rustBuilder.overrides.all ++ [
+    /*
+     [1] We add some logic to compile our crates with clippy, it provides us many additional lints
+
+     [2] We need to alter Nix hardening to make static binaries: PIE,
+     Position Independent Executables seems to be supported only on amd64. Having
+     this flag set either 1. make our executables crash or 2. compile as dynamic on some platforms.
+     Here, we deactivate it. Later (find `codegenOpts`), we reactivate it for supported targets
+     (only amd64 curently) through the `-static-pie` flag.
+     PIE is a feature used by ASLR, which helps mitigate security issues.
+     Learn more about Nix Hardening at: https://github.com/NixOS/nixpkgs/blob/master/pkgs/build-support/cc-wrapper/add-hardening.sh
+
+     [3] We want to inject the git version while keeping the build deterministic.
+     As we do not want to consider the .git folder as part of the input source,
+     we ask the user (the CI often) to pass the value to Nix.
+
+     [4] We ship some parts of the code disabled by default by putting them behind a flag.
+     It speeds up the compilation (when the feature is not required) and released crates have less dependency by default (less attack surface, disk space, etc.).
+     But we want to ship these additional features when we release Garage.
+     In the end, we chose to exclude all features from debug builds while putting (all of) them in the release builds.
+
+     [5] We don't want libsodium-sys and zstd-sys to try to use pkgconfig to build against a system library.
+     However the features to do so get activated for some reason (due to a bug in cargo2nix?),
+     so disable them manually here.
+    */
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "garage";
+      overrideAttrs = drv:
+        (if git_version != null then {
+          /* [3] */ preConfigure = ''
+            ${drv.preConfigure or ""}
+            export GIT_VERSION="${git_version}"
+          '';
+        } else {})
+        //
+      {
+        /* [1] */ setBuildEnv = (buildEnv drv);
+        /* [2] */ hardeningDisable = [ "pie" ];
+      };
+      overrideArgs = old: {
+        /* [4] */ features = [ "bundled-libs" "sled" ]
+          ++ (if release then [ "kubernetes-discovery" "telemetry-otlp" "metrics" "lmdb" "sqlite" ] else []);
+      };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "garage_rpc";
+      overrideAttrs = drv: { /* [1] */ setBuildEnv = (buildEnv drv); };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "garage_db";
+      overrideAttrs = drv: { /* [1] */ setBuildEnv = (buildEnv drv); };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "garage_util";
+      overrideAttrs = drv: { /* [1] */ setBuildEnv = (buildEnv drv); };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "garage_table";
+      overrideAttrs = drv: { /* [1] */ setBuildEnv = (buildEnv drv); };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "garage_block";
+      overrideAttrs = drv: { /* [1] */ setBuildEnv = (buildEnv drv); };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "garage_model";
+      overrideAttrs = drv: { /* [1] */ setBuildEnv = (buildEnv drv); };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "garage_api";
+      overrideAttrs = drv: { /* [1] */ setBuildEnv = (buildEnv drv); };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "garage_web";
+      overrideAttrs = drv: { /* [1] */ setBuildEnv = (buildEnv drv); };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "k2v-client";
+      overrideAttrs = drv: { /* [1] */ setBuildEnv = (buildEnv drv); };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "libsodium-sys";
+      overrideArgs = old: {
+        features = [ ]; /* [5] */
+      };
+    })
+
+    (pkgs.rustBuilder.rustLib.makeOverride {
+      name = "zstd-sys";
+      overrideArgs = old: {
+        features = [ ]; /* [5] */
+      };
+    })
+  ];
+
+  packageFun = import ../Cargo.nix;
+
+  /*
+    We compile fully static binaries with musl to simplify deployment on most systems.
+    When possible, we reactivate PIE hardening (see above).
+
+    Also, if you set the RUSTFLAGS environment variable, the following parameters will
+    be ignored.
+
+    For more information on static builds, please refer to Rust's RFC 1721.
+    https://rust-lang.github.io/rfcs/1721-crt-static.html#specifying-dynamicstatic-c-runtime-linkage
+  */
+
+  codegenOpts = {
+   "armv6l-unknown-linux-musleabihf" = [ "target-feature=+crt-static" "link-arg=-static" ]; /* compile as dynamic with static-pie */
+   "aarch64-unknown-linux-musl" = [ "target-feature=+crt-static" "link-arg=-static" ]; /* segfault with static-pie */
+   "i686-unknown-linux-musl" = [ "target-feature=+crt-static" "link-arg=-static" ]; /* segfault with static-pie */
+   "x86_64-unknown-linux-musl" = [ "target-feature=+crt-static" "link-arg=-static-pie" ];
+  };
+
+in
+  /*
+   The following definition is not elegant as we use a low level function of Cargo2nix
+   that enables us to pass our custom rustChannel object. We need this low level definition
+   to pass Nix's Rust toolchains instead of Mozilla's one.
+
+   target is mandatory but must be kept to null to allow cargo2nix to set it to the appropriate value
+   for each crate.
+  */ 
+  pkgs.rustBuilder.makePackageSet {
+    inherit packageFun rustChannel release codegenOpts;
+    packageOverrides = overrides;
+    target = null;
+
+    buildRustPackages = pkgs.buildPackages.rustBuilder.makePackageSet {
+      inherit rustChannel packageFun codegenOpts;
+      packageOverrides = overrides;
+      target = null;
+    };
+  }

nix/nix.conf

@@ -1,4 +1,9 @@
 substituters = https://cache.nixos.org https://nix.web.deuxfleurs.fr
 trusted-public-keys = cache.nixos.org-1:6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY= nix.web.deuxfleurs.fr:eTGL6kvaQn6cDR/F9lDYUIP9nCVR/kkshYfLDJf1yKs=
 max-jobs = auto
-cores = 4
+cores = 0
+log-lines = 200
+filter-syscalls = false
+sandbox = false
+keep-outputs = true
+keep-derivations = true

script/dev-cluster.sh

@@ -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
+export RUST_LOG=garage=info,garage_api=debug,netapp=trace
 MAIN_LABEL="\e[${FANCYCOLORS[0]}[main]\e[49m"
 
 WHICH_GARAGE=$(which garage || exit 1)

script/not-dynamic.sh

@@ -0,0 +1,14 @@
+#!/usr/bin/env bash
+
+set -e
+
+if [ "$#" -ne 1 ]; then
+    echo "[fail] usage: $0 binary"
+    exit 2
+fi
+
+if file $1 | grep 'dynamically linked' 2>&1; then
+    echo "[fail] $1 is dynamic"
+    exit 1
+fi
+echo "[ok] $1 is probably static"

shell.nix

@@ -1,8 +1,5 @@
 {
   system ? builtins.currentSystem,
-  rust ? true,
-  integration ? true,
-  release ? true,
 }:
 
 with import ./nix/common.nix;
@@ -16,9 +13,59 @@ let
   winscp = (import ./nix/winscp.nix) pkgs;
 
 in
+  {
 
-pkgs.mkShell {
-  shellHook = ''
+    /* --- Rust Shell ---
+     * Use it to compile Garage
+     */
+    rust = pkgs.mkShell {
+      shellHook = ''
+function refresh_toolchain {
+  nix copy \
+    --to 's3://nix?endpoint=garage.deuxfleurs.fr&region=garage&secret-key=/etc/nix/signing-key.sec' \
+    $(nix-store -qR \
+      $(nix-build --quiet --no-build-output --no-out-link nix/toolchain.nix))
+}
+      '';
+
+      nativeBuildInputs = [
+        #pkgs.rustPlatform.rust.rustc
+        pkgs.rustPlatform.rust.cargo
+        #pkgs.clippy
+        pkgs.rustfmt
+        #pkgs.perl
+        #pkgs.protobuf
+        #pkgs.pkg-config
+        #pkgs.openssl
+        pkgs.file
+        #cargo2nix.packages.x86_64-linux.cargo2nix
+      ];
+    };
+
+    /* --- Integration shell ---
+     * Use it to test Garage with common S3 clients
+     */
+    integration = pkgs.mkShell {
+      nativeBuildInputs = [
+        winscp
+        pkgs.s3cmd
+        pkgs.awscli2
+        pkgs.minio-client
+        pkgs.rclone
+        pkgs.socat
+        pkgs.psmisc
+        pkgs.which
+        pkgs.openssl
+        pkgs.curl
+        pkgs.jq
+      ];
+    };
+
+    /* --- Release shell ---
+     * A shell built to make releasing easier
+     */
+    release = pkgs.mkShell {
+      shellHook = ''
 function to_s3 {
   aws \
       --endpoint-url https://garage.deuxfleurs.fr \
@@ -62,45 +109,12 @@ function refresh_index {
       result/share/_releases.html \
       s3://garagehq.deuxfleurs.fr/
 }
+        '';
+      nativeBuildInputs = [
+        pkgs.awscli2
+        kaniko 
+      ];
+    };
+  }
 
-function refresh_toolchain {
-  nix copy \
-    --to 's3://nix?endpoint=garage.deuxfleurs.fr&region=garage&secret-key=/etc/nix/signing-key.sec' \
-    $(nix-store -qR \
-      $(nix-build --quiet --no-build-output --no-out-link nix/toolchain.nix))
-}
-  '';
 
-  nativeBuildInputs = 
-   (if rust then [
-     pkgs.rustPlatform.rust.rustc
-     pkgs.rustPlatform.rust.cargo
-     pkgs.clippy
-     pkgs.rustfmt
-     pkgs.perl
-     pkgs.protobuf
-     pkgs.pkg-config
-     pkgs.openssl
-     cargo2nix.packages.x86_64-linux.cargo2nix
-    ] else [])
-   ++
-   (if integration then [
-     winscp
-     pkgs.s3cmd
-     pkgs.awscli2
-     pkgs.minio-client
-     pkgs.rclone
-     pkgs.socat
-     pkgs.psmisc
-     pkgs.which
-     pkgs.openssl
-     pkgs.curl
-     pkgs.jq
-    ] else [])
-   ++
-   (if release then [ 
-     pkgs.awscli2
-     kaniko 
-    ] else [])
-   ;
-}

src/api/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "garage_api"
-version = "0.7.0"
+version = "0.8.0"
 authors = ["Alex Auvolat <alex@adnab.me>"]
 edition = "2018"
 license = "AGPL-3.0"
@@ -14,30 +14,31 @@ path = "lib.rs"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-garage_model = { version = "0.7.0", path = "../model" }
-garage_table = { version = "0.7.0", path = "../table" }
-garage_block = { version = "0.7.0", path = "../block" }
-garage_util = { version = "0.7.0", path = "../util" }
-garage_rpc = { version = "0.7.0", path = "../rpc" }
+garage_model = { version = "0.8.0", path = "../model" }
+garage_table = { version = "0.8.0", path = "../table" }
+garage_block = { version = "0.8.0", path = "../block" }
+garage_util = { version = "0.8.0", path = "../util" }
+garage_rpc = { version = "0.8.0", path = "../rpc" }
 
 async-trait = "0.1.7"
 base64 = "0.13"
 bytes = "1.0"
 chrono = "0.4"
-crypto-mac = "0.10"
+crypto-common = "0.1"
 err-derive = "0.3"
 hex = "0.4"
-hmac = "0.10"
+hmac = "0.12"
 idna = "0.2"
 tracing = "0.1.30"
-md-5 = "0.9"
+md-5 = "0.10"
 nom = "7.1"
-sha2 = "0.9"
+sha2 = "0.10"
 
 futures = "0.3"
 futures-util = "0.3"
 pin-project = "1.0"
 tokio = { version = "1.0", default-features = false, features = ["rt", "rt-multi-thread", "io-util", "net", "time", "macros", "sync", "signal", "fs"] }
+tokio-stream = "0.1"
 
 form_urlencoded = "1.0.0"
 http = "0.2"
@@ -54,9 +55,9 @@ quick-xml = { version = "0.21", features = [ "serialize" ] }
 url = "2.1"
 
 opentelemetry = "0.17"
-opentelemetry-prometheus = "0.10"
-opentelemetry-otlp = "0.10"
-prometheus = "0.13"
+opentelemetry-prometheus = { version = "0.10", optional = true }
+prometheus = { version = "0.13", optional = true }
 
 [features]
 k2v = [ "garage_util/k2v", "garage_model/k2v" ]
+metrics = [ "opentelemetry-prometheus", "prometheus" ]

src/api/admin/api_server.rs

@@ -1,15 +1,17 @@
+use std::net::SocketAddr;
 use std::sync::Arc;
 
 use async_trait::async_trait;
 
 use futures::future::Future;
-use http::header::{
-	ACCESS_CONTROL_ALLOW_METHODS, ACCESS_CONTROL_ALLOW_ORIGIN, ALLOW, CONTENT_TYPE,
-};
+use http::header::{ACCESS_CONTROL_ALLOW_METHODS, ACCESS_CONTROL_ALLOW_ORIGIN, ALLOW};
 use hyper::{Body, Request, Response};
 
-use opentelemetry::trace::{SpanRef, Tracer};
+use opentelemetry::trace::SpanRef;
+
+#[cfg(feature = "metrics")]
 use opentelemetry_prometheus::PrometheusExporter;
+#[cfg(feature = "metrics")]
 use prometheus::{Encoder, TextEncoder};
 
 use garage_model::garage::Garage;
@@ -25,6 +27,7 @@ use crate::admin::router::{Authorization, Endpoint};
 
 pub struct AdminApiServer {
 	garage: Arc<Garage>,
+	#[cfg(feature = "metrics")]
 	exporter: PrometheusExporter,
 	metrics_token: Option<String>,
 	admin_token: Option<String>,
@@ -32,7 +35,6 @@ pub struct AdminApiServer {
 
 impl AdminApiServer {
 	pub fn new(garage: Arc<Garage>) -> Self {
-		let exporter = opentelemetry_prometheus::exporter().init();
 		let cfg = &garage.config.admin;
 		let metrics_token = cfg
 			.metrics_token
@@ -44,21 +46,22 @@ impl AdminApiServer {
 			.map(|tok| format!("Bearer {}", tok));
 		Self {
 			garage,
-			exporter,
+			#[cfg(feature = "metrics")]
+			exporter: opentelemetry_prometheus::exporter().init(),
 			metrics_token,
 			admin_token,
 		}
 	}
 
-	pub async fn run(self, shutdown_signal: impl Future<Output = ()>) -> Result<(), GarageError> {
-		if let Some(bind_addr) = self.garage.config.admin.api_bind_addr {
-			let region = self.garage.config.s3_api.s3_region.clone();
-			ApiServer::new(region, self)
-				.run_server(bind_addr, shutdown_signal)
-				.await
-		} else {
-			Ok(())
-		}
+	pub async fn run(
+		self,
+		bind_addr: SocketAddr,
+		shutdown_signal: impl Future<Output = ()>,
+	) -> Result<(), GarageError> {
+		let region = self.garage.config.s3_api.s3_region.clone();
+		ApiServer::new(region, self)
+			.run_server(bind_addr, shutdown_signal)
+			.await
 	}
 
 	fn handle_options(&self, _req: &Request<Body>) -> Result<Response<Body>, Error> {
@@ -71,22 +74,31 @@ impl AdminApiServer {
 	}
 
 	fn handle_metrics(&self) -> Result<Response<Body>, Error> {
-		let mut buffer = vec![];
-		let encoder = TextEncoder::new();
+		#[cfg(feature = "metrics")]
+		{
+			use opentelemetry::trace::Tracer;
 
-		let tracer = opentelemetry::global::tracer("garage");
-		let metric_families = tracer.in_span("admin/gather_metrics", |_| {
-			self.exporter.registry().gather()
-		});
+			let mut buffer = vec![];
+			let encoder = TextEncoder::new();
 
-		encoder
-			.encode(&metric_families, &mut buffer)
-			.ok_or_internal_error("Could not serialize metrics")?;
+			let tracer = opentelemetry::global::tracer("garage");
+			let metric_families = tracer.in_span("admin/gather_metrics", |_| {
+				self.exporter.registry().gather()
+			});
 
-		Ok(Response::builder()
-			.status(200)
-			.header(CONTENT_TYPE, encoder.format_type())
-			.body(Body::from(buffer))?)
+			encoder
+				.encode(&metric_families, &mut buffer)
+				.ok_or_internal_error("Could not serialize metrics")?;
+
+			Ok(Response::builder()
+				.status(200)
+				.header(http::header::CONTENT_TYPE, encoder.format_type())
+				.body(Body::from(buffer))?)
+		}
+		#[cfg(not(feature = "metrics"))]
+		Err(Error::bad_request(
+			"Garage was built without the metrics feature".to_string(),
+		))
 	}
 }
 

src/api/admin/cluster.rs

@@ -18,7 +18,8 @@ use crate::helpers::{json_ok_response, parse_json_body};
 pub async fn handle_get_cluster_status(garage: &Arc<Garage>) -> Result<Response<Body>, Error> {
 	let res = GetClusterStatusResponse {
 		node: hex::encode(garage.system.id),
-		garage_version: garage.system.garage_version(),
+		garage_version: garage_util::version::garage_version(),
+		garage_features: garage_util::version::garage_features(),
 		db_engine: garage.db.engine(),
 		known_nodes: garage
 			.system
@@ -99,6 +100,7 @@ fn get_cluster_layout(garage: &Arc<Garage>) -> GetClusterLayoutResponse {
 struct GetClusterStatusResponse {
 	node: String,
 	garage_version: &'static str,
+	garage_features: Option<&'static [&'static str]>,
 	db_engine: String,
 	known_nodes: HashMap<String, KnownNodeResp>,
 	layout: GetClusterLayoutResponse,

src/api/k2v/api_server.rs

@@ -1,3 +1,4 @@
+use std::net::SocketAddr;
 use std::sync::Arc;
 
 use async_trait::async_trait;
@@ -36,20 +37,13 @@ pub(crate) struct K2VApiEndpoint {
 impl K2VApiServer {
 	pub async fn run(
 		garage: Arc<Garage>,
+		bind_addr: SocketAddr,
+		s3_region: String,
 		shutdown_signal: impl Future<Output = ()>,
 	) -> Result<(), GarageError> {
-		if let Some(cfg) = &garage.config.k2v_api {
-			let bind_addr = cfg.api_bind_addr;
-
-			ApiServer::new(
-				garage.config.s3_api.s3_region.clone(),
-				K2VApiServer { garage },
-			)
+		ApiServer::new(s3_region, K2VApiServer { garage })
 			.run_server(bind_addr, shutdown_signal)
 			.await
-		} else {
-			Ok(())
-		}
 	}
 }
 

src/api/s3/api_server.rs

@@ -1,3 +1,4 @@
+use std::net::SocketAddr;
 use std::sync::Arc;
 
 use async_trait::async_trait;
@@ -43,16 +44,13 @@ pub(crate) struct S3ApiEndpoint {
 impl S3ApiServer {
 	pub async fn run(
 		garage: Arc<Garage>,
+		addr: SocketAddr,
+		s3_region: String,
 		shutdown_signal: impl Future<Output = ()>,
 	) -> Result<(), GarageError> {
-		let addr = garage.config.s3_api.api_bind_addr;
-
-		ApiServer::new(
-			garage.config.s3_api.s3_region.clone(),
-			S3ApiServer { garage },
-		)
-		.run_server(addr, shutdown_signal)
-		.await
+		ApiServer::new(s3_region, S3ApiServer { garage })
+			.run_server(addr, shutdown_signal)
+			.await
 	}
 
 	async fn handle_request_without_bucket(

src/api/s3/bucket.rs

@@ -295,7 +295,6 @@ fn parse_create_bucket_xml(xml_bytes: &[u8]) -> Option<Option<String>> {
 
 	let mut ret = None;
 	for item in cbc.children() {
-		println!("{:?}", item);
 		if item.has_tag_name("LocationConstraint") {
 			if ret != None {
 				return None;

src/api/s3/copy.rs

@@ -5,9 +5,12 @@ use std::time::{Duration, SystemTime, UNIX_EPOCH};
 use futures::{stream, stream::Stream, StreamExt, TryFutureExt};
 use md5::{Digest as Md5Digest, Md5};
 
+use bytes::Bytes;
 use hyper::{Body, Request, Response};
 use serde::Serialize;
 
+use garage_rpc::netapp::bytes_buf::BytesBuf;
+use garage_rpc::rpc_helper::OrderTag;
 use garage_table::*;
 use garage_util::data::*;
 use garage_util::time::*;
@@ -305,13 +308,18 @@ pub async fn handle_upload_part_copy(
 	// if and only if the block returned is a block that already existed
 	// in the Garage data store (thus we don't need to save it again).
 	let garage2 = garage.clone();
+	let order_stream = OrderTag::stream();
 	let source_blocks = stream::iter(blocks_to_copy)
-		.flat_map(|(block_hash, range_to_copy)| {
+		.enumerate()
+		.flat_map(|(i, (block_hash, range_to_copy))| {
 			let garage3 = garage2.clone();
 			stream::once(async move {
-				let data = garage3.block_manager.rpc_get_block(&block_hash).await?;
+				let data = garage3
+					.block_manager
+					.rpc_get_block(&block_hash, Some(order_stream.order(i as u64)))
+					.await?;
 				match range_to_copy {
-					Some(r) => Ok((data[r].to_vec(), None)),
+					Some(r) => Ok((data.slice(r), None)),
 					None => Ok((data, Some(block_hash))),
 				}
 			})
@@ -553,13 +561,13 @@ impl CopyPreconditionHeaders {
 	}
 }
 
-type BlockStreamItemOk = (Vec<u8>, Option<Hash>);
+type BlockStreamItemOk = (Bytes, Option<Hash>);
 type BlockStreamItem = Result<BlockStreamItemOk, garage_util::error::Error>;
 
 struct Defragmenter<S: Stream<Item = BlockStreamItem>> {
 	block_size: usize,
 	block_stream: Pin<Box<stream::Peekable<S>>>,
-	buffer: Vec<u8>,
+	buffer: BytesBuf,
 	hash: Option<Hash>,
 }
 
@@ -568,7 +576,7 @@ impl<S: Stream<Item = BlockStreamItem>> Defragmenter<S> {
 		Self {
 			block_size,
 			block_stream,
-			buffer: vec![],
+			buffer: BytesBuf::new(),
 			hash: None,
 		}
 	}
@@ -586,7 +594,7 @@ impl<S: Stream<Item = BlockStreamItem>> Defragmenter<S> {
 
 			if self.buffer.is_empty() {
 				let (next_block, next_block_hash) = self.block_stream.next().await.unwrap()?;
-				self.buffer = next_block;
+				self.buffer.extend(next_block);
 				self.hash = next_block_hash;
 			} else if self.buffer.len() + peeked_next_block.len() > self.block_size {
 				break;
@@ -597,11 +605,11 @@ impl<S: Stream<Item = BlockStreamItem>> Defragmenter<S> {
 			}
 		}
 
-		Ok((std::mem::take(&mut self.buffer), self.hash.take()))
+		Ok((self.buffer.take_all(), self.hash.take()))
 	}
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct CopyObjectResult {
 	#[serde(rename = "LastModified")]
 	pub last_modified: s3_xml::Value,
@@ -609,7 +617,7 @@ pub struct CopyObjectResult {
 	pub etag: s3_xml::Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct CopyPartResult {
 	#[serde(serialize_with = "xmlns_tag")]
 	pub xmlns: (),
@@ -654,7 +662,6 @@ mod tests {
 			last_modified: s3_xml::Value("2011-04-11T20:34:56.000Z".into()),
 			etag: s3_xml::Value("\"9b2cf535f27731c974343645a3985328\"".into()),
 		};
-		println!("{}", to_xml_with_header(&v)?);
 
 		assert_eq!(to_xml_with_header(&v)?, expected_retval);
 

src/api/s3/delete.rs

@@ -64,14 +64,13 @@ pub async fn handle_delete(
 	bucket_id: Uuid,
 	key: &str,
 ) -> Result<Response<Body>, Error> {
-	let (_deleted_version, delete_marker_version) =
-		handle_delete_internal(&garage, bucket_id, key).await?;
-
-	Ok(Response::builder()
-		.header("x-amz-version-id", hex::encode(delete_marker_version))
-		.status(StatusCode::NO_CONTENT)
-		.body(Body::from(vec![]))
-		.unwrap())
+	match handle_delete_internal(&garage, bucket_id, key).await {
+		Ok(_) | Err(Error::NoSuchKey) => Ok(Response::builder()
+			.status(StatusCode::NO_CONTENT)
+			.body(Body::from(vec![]))
+			.unwrap()),
+		Err(e) => Err(e),
+	}
 }
 
 pub async fn handle_delete_objects(

src/api/s3/get.rs

@@ -2,16 +2,19 @@
 use std::sync::Arc;
 use std::time::{Duration, UNIX_EPOCH};
 
-use futures::stream::*;
+use futures::future;
+use futures::stream::{self, StreamExt};
 use http::header::{
 	ACCEPT_RANGES, CONTENT_LENGTH, CONTENT_RANGE, CONTENT_TYPE, ETAG, IF_MODIFIED_SINCE,
 	IF_NONE_MATCH, LAST_MODIFIED, RANGE,
 };
-use hyper::body::Bytes;
 use hyper::{Body, Request, Response, StatusCode};
+use tokio::sync::mpsc;
 
+use garage_rpc::rpc_helper::{netapp::stream::ByteStream, OrderTag};
 use garage_table::EmptyKey;
 use garage_util::data::*;
+use garage_util::error::OkOrMessage;
 
 use garage_model::garage::Garage;
 use garage_model::s3::object_table::*;
@@ -242,36 +245,56 @@ pub async fn handle_get(
 			Ok(resp_builder.body(body)?)
 		}
 		ObjectVersionData::FirstBlock(_, first_block_hash) => {
-			let read_first_block = garage.block_manager.rpc_get_block(first_block_hash);
-			let get_next_blocks = garage.version_table.get(&last_v.uuid, &EmptyKey);
+			let (tx, rx) = mpsc::channel(2);
 
-			let (first_block, version) = futures::try_join!(read_first_block, get_next_blocks)?;
-			let version = version.ok_or(Error::NoSuchKey)?;
+			let order_stream = OrderTag::stream();
+			let first_block_hash = *first_block_hash;
+			let version_uuid = last_v.uuid;
 
-			let mut blocks = version
-				.blocks
-				.items()
-				.iter()
-				.map(|(_, vb)| (vb.hash, None))
-				.collect::<Vec<_>>();
-			blocks[0].1 = Some(first_block);
+			tokio::spawn(async move {
+				match async {
+					let garage2 = garage.clone();
+					let version_fut = tokio::spawn(async move {
+						garage2.version_table.get(&version_uuid, &EmptyKey).await
+					});
 
-			let body_stream = futures::stream::iter(blocks)
-				.map(move |(hash, data_opt)| {
-					let garage = garage.clone();
-					async move {
-						if let Some(data) = data_opt {
-							Ok(Bytes::from(data))
-						} else {
-							garage
-								.block_manager
-								.rpc_get_block(&hash)
-								.await
-								.map(Bytes::from)
-						}
+					let stream_block_0 = garage
+						.block_manager
+						.rpc_get_block_streaming(&first_block_hash, Some(order_stream.order(0)))
+						.await?;
+					tx.send(stream_block_0)
+						.await
+						.ok_or_message("channel closed")?;
+
+					let version = version_fut.await.unwrap()?.ok_or(Error::NoSuchKey)?;
+					for (i, (_, vb)) in version.blocks.items().iter().enumerate().skip(1) {
+						let stream_block_i = garage
+							.block_manager
+							.rpc_get_block_streaming(&vb.hash, Some(order_stream.order(i as u64)))
+							.await?;
+						tx.send(stream_block_i)
+							.await
+							.ok_or_message("channel closed")?;
 					}
-				})
-				.buffered(2);
+
+					Ok::<(), Error>(())
+				}
+				.await
+				{
+					Ok(()) => (),
+					Err(e) => {
+						let err = std::io::Error::new(
+							std::io::ErrorKind::Other,
+							format!("Error while getting object data: {}", e),
+						);
+						let _ = tx
+							.send(Box::pin(stream::once(future::ready(Err(err)))))
+							.await;
+					}
+				}
+			});
+
+			let body_stream = tokio_stream::wrappers::ReceiverStream::new(rx).flatten();
 
 			let body = hyper::body::Body::wrap_stream(body_stream);
 			Ok(resp_builder.body(body)?)
@@ -422,40 +445,79 @@ fn body_from_blocks_range(
 		all_blocks.len(),
 		4 + ((end - begin) / std::cmp::max(all_blocks[0].1.size as u64, 1024)) as usize,
 	));
-	let mut true_offset = 0;
+	let mut block_offset: u64 = 0;
 	for (_, b) in all_blocks.iter() {
-		if true_offset >= end {
+		if block_offset >= end {
 			break;
 		}
 		// Keep only blocks that have an intersection with the requested range
-		if true_offset < end && true_offset + b.size > begin {
-			blocks.push((*b, true_offset));
+		if block_offset < end && block_offset + b.size > begin {
+			blocks.push((*b, block_offset));
 		}
-		true_offset += b.size;
+		block_offset += b.size as u64;
 	}
 
+	let order_stream = OrderTag::stream();
 	let body_stream = futures::stream::iter(blocks)
-		.map(move |(block, true_offset)| {
+		.enumerate()
+		.map(move |(i, (block, block_offset))| {
 			let garage = garage.clone();
 			async move {
-				let data = garage.block_manager.rpc_get_block(&block.hash).await?;
-				let data = Bytes::from(data);
-				let start_in_block = if true_offset > begin {
-					0
-				} else {
-					begin - true_offset
-				};
-				let end_in_block = if true_offset + block.size < end {
-					block.size
-				} else {
-					end - true_offset
-				};
-				Result::<Bytes, Error>::Ok(
-					data.slice(start_in_block as usize..end_in_block as usize),
-				)
+				garage
+					.block_manager
+					.rpc_get_block_streaming(&block.hash, Some(order_stream.order(i as u64)))
+					.await
+					.unwrap_or_else(|e| error_stream(i, e))
+					.scan(block_offset, move |chunk_offset, chunk| {
+						let r = match chunk {
+							Ok(chunk_bytes) => {
+								let chunk_len = chunk_bytes.len() as u64;
+								let r = if *chunk_offset >= end {
+									// The current chunk is after the part we want to read.
+									// Returning None here will stop the scan, the rest of the
+									// stream will be ignored
+									None
+								} else if *chunk_offset + chunk_len <= begin {
+									// The current chunk is before the part we want to read.
+									// We return a None that will be removed by the filter_map
+									// below.
+									Some(None)
+								} else {
+									// The chunk has an intersection with the requested range
+									let start_in_chunk = if *chunk_offset > begin {
+										0
+									} else {
+										begin - *chunk_offset
+									};
+									let end_in_chunk = if *chunk_offset + chunk_len < end {
+										chunk_len
+									} else {
+										end - *chunk_offset
+									};
+									Some(Some(Ok(chunk_bytes
+										.slice(start_in_chunk as usize..end_in_chunk as usize))))
+								};
+								*chunk_offset += chunk_bytes.len() as u64;
+								r
+							}
+							Err(e) => Some(Some(Err(e))),
+						};
+						futures::future::ready(r)
+					})
+					.filter_map(futures::future::ready)
 			}
 		})
-		.buffered(2);
+		.buffered(2)
+		.flatten();
 
 	hyper::body::Body::wrap_stream(body_stream)
 }
+
+fn error_stream(i: usize, e: garage_util::error::Error) -> ByteStream {
+	Box::pin(futures::stream::once(async move {
+		Err(std::io::Error::new(
+			std::io::ErrorKind::Other,
+			format!("Could not get block {}: {}", i, e),
+		))
+	}))
+}

src/api/s3/put.rs

@@ -1,4 +1,4 @@
-use std::collections::{BTreeMap, BTreeSet, HashMap, VecDeque};
+use std::collections::{BTreeMap, BTreeSet, HashMap};
 use std::sync::Arc;
 
 use futures::prelude::*;
@@ -8,7 +8,14 @@ use hyper::{Request, Response};
 use md5::{digest::generic_array::*, Digest as Md5Digest, Md5};
 use sha2::Sha256;
 
+use opentelemetry::{
+	trace::{FutureExt as OtelFutureExt, TraceContextExt, Tracer},
+	Context,
+};
+
+use garage_rpc::netapp::bytes_buf::BytesBuf;
 use garage_table::*;
+use garage_util::async_hash::*;
 use garage_util::data::*;
 use garage_util::error::Error as GarageError;
 use garage_util::time::*;
@@ -102,7 +109,7 @@ pub(crate) async fn save_stream<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
 					size,
 					etag: data_md5sum_hex.clone(),
 				},
-				first_block,
+				first_block.to_vec(),
 			)),
 		};
 
@@ -130,7 +137,7 @@ pub(crate) async fn save_stream<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
 	garage.version_table.insert(&version).await?;
 
 	// Transfer data and verify checksum
-	let first_block_hash = blake2sum(&first_block[..]);
+	let first_block_hash = async_blake2sum(first_block.clone()).await;
 
 	let tx_result = (|| async {
 		let (total_size, data_md5sum, data_sha256sum) = read_and_put_blocks(
@@ -273,14 +280,23 @@ async fn read_and_put_blocks<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
 	garage: &Garage,
 	version: &Version,
 	part_number: u64,
-	first_block: Vec<u8>,
+	first_block: Bytes,
 	first_block_hash: Hash,
 	chunker: &mut StreamChunker<S>,
 ) -> Result<(u64, GenericArray<u8, typenum::U16>, Hash), Error> {
-	let mut md5hasher = Md5::new();
-	let mut sha256hasher = Sha256::new();
-	md5hasher.update(&first_block[..]);
-	sha256hasher.update(&first_block[..]);
+	let tracer = opentelemetry::global::tracer("garage");
+
+	let md5hasher = AsyncHasher::<Md5>::new();
+	let sha256hasher = AsyncHasher::<Sha256>::new();
+
+	futures::future::join(
+		md5hasher.update(first_block.clone()),
+		sha256hasher.update(first_block.clone()),
+	)
+	.with_context(Context::current_with_span(
+		tracer.start("Hash first block (md5, sha256)"),
+	))
+	.await;
 
 	let mut next_offset = first_block.len();
 	let mut put_curr_version_block = put_block_meta(
@@ -302,9 +318,15 @@ async fn read_and_put_blocks<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
 			chunker.next(),
 		)?;
 		if let Some(block) = next_block {
-			md5hasher.update(&block[..]);
-			sha256hasher.update(&block[..]);
-			let block_hash = blake2sum(&block[..]);
+			let (_, _, block_hash) = futures::future::join3(
+				md5hasher.update(block.clone()),
+				sha256hasher.update(block.clone()),
+				async_blake2sum(block.clone()),
+			)
+			.with_context(Context::current_with_span(
+				tracer.start("Hash block (md5, sha256, blake2)"),
+			))
+			.await;
 			let block_len = block.len();
 			put_curr_version_block = put_block_meta(
 				garage,
@@ -322,9 +344,9 @@ async fn read_and_put_blocks<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
 	}
 
 	let total_size = next_offset as u64;
-	let data_md5sum = md5hasher.finalize();
+	let data_md5sum = md5hasher.finalize().await;
 
-	let data_sha256sum = sha256hasher.finalize();
+	let data_sha256sum = sha256hasher.finalize().await;
 	let data_sha256sum = Hash::try_from(&data_sha256sum[..]).unwrap();
 
 	Ok((total_size, data_md5sum, data_sha256sum))
@@ -364,7 +386,7 @@ struct StreamChunker<S: Stream<Item = Result<Bytes, Error>>> {
 	stream: S,
 	read_all: bool,
 	block_size: usize,
-	buf: VecDeque<u8>,
+	buf: BytesBuf,
 }
 
 impl<S: Stream<Item = Result<Bytes, Error>> + Unpin> StreamChunker<S> {
@@ -373,11 +395,11 @@ impl<S: Stream<Item = Result<Bytes, Error>> + Unpin> StreamChunker<S> {
 			stream,
 			read_all: false,
 			block_size,
-			buf: VecDeque::with_capacity(2 * block_size),
+			buf: BytesBuf::new(),
 		}
 	}
 
-	async fn next(&mut self) -> Result<Option<Vec<u8>>, Error> {
+	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?;
@@ -390,12 +412,8 @@ impl<S: Stream<Item = Result<Bytes, Error>> + Unpin> StreamChunker<S> {
 
 		if self.buf.is_empty() {
 			Ok(None)
-		} else if self.buf.len() <= self.block_size {
-			let block = self.buf.drain(..).collect::<Vec<u8>>();
-			Ok(Some(block))
 		} else {
-			let block = self.buf.drain(..self.block_size).collect::<Vec<u8>>();
-			Ok(Some(block))
+			Ok(Some(self.buf.take_max(self.block_size)))
 		}
 	}
 }
@@ -504,7 +522,9 @@ pub async fn handle_put_part(
 
 	// Copy block to store
 	let version = Version::new(version_uuid, bucket_id, key, false);
-	let first_block_hash = blake2sum(&first_block[..]);
+
+	let first_block_hash = async_blake2sum(first_block.clone()).await;
+
 	let (_, data_md5sum, data_sha256sum) = read_and_put_blocks(
 		&garage,
 		&version,

src/api/s3/xml.rs

@@ -25,7 +25,7 @@ impl From<&str> for Value {
 #[derive(Debug, Serialize, Deserialize, PartialEq, Eq, PartialOrd, Ord)]
 pub struct IntValue(#[serde(rename = "$value")] pub i64);
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct Bucket {
 	#[serde(rename = "CreationDate")]
 	pub creation_date: Value,
@@ -33,7 +33,7 @@ pub struct Bucket {
 	pub name: Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct Owner {
 	#[serde(rename = "DisplayName")]
 	pub display_name: Value,
@@ -41,13 +41,13 @@ pub struct Owner {
 	pub id: Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct BucketList {
 	#[serde(rename = "Bucket")]
 	pub entries: Vec<Bucket>,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct ListAllMyBucketsResult {
 	#[serde(rename = "Buckets")]
 	pub buckets: BucketList,
@@ -55,7 +55,7 @@ pub struct ListAllMyBucketsResult {
 	pub owner: Owner,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct LocationConstraint {
 	#[serde(serialize_with = "xmlns_tag")]
 	pub xmlns: (),
@@ -63,7 +63,7 @@ pub struct LocationConstraint {
 	pub region: String,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct Deleted {
 	#[serde(rename = "Key")]
 	pub key: Value,
@@ -73,7 +73,7 @@ pub struct Deleted {
 	pub delete_marker_version_id: Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct Error {
 	#[serde(rename = "Code")]
 	pub code: Value,
@@ -85,7 +85,7 @@ pub struct Error {
 	pub region: Option<Value>,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct DeleteError {
 	#[serde(rename = "Code")]
 	pub code: Value,
@@ -97,7 +97,7 @@ pub struct DeleteError {
 	pub version_id: Option<Value>,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct DeleteResult {
 	#[serde(serialize_with = "xmlns_tag")]
 	pub xmlns: (),
@@ -107,7 +107,7 @@ pub struct DeleteResult {
 	pub errors: Vec<DeleteError>,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct InitiateMultipartUploadResult {
 	#[serde(serialize_with = "xmlns_tag")]
 	pub xmlns: (),
@@ -119,7 +119,7 @@ pub struct InitiateMultipartUploadResult {
 	pub upload_id: Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct CompleteMultipartUploadResult {
 	#[serde(serialize_with = "xmlns_tag")]
 	pub xmlns: (),
@@ -133,7 +133,7 @@ pub struct CompleteMultipartUploadResult {
 	pub etag: Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct Initiator {
 	#[serde(rename = "DisplayName")]
 	pub display_name: Value,
@@ -141,7 +141,7 @@ pub struct Initiator {
 	pub id: Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct ListMultipartItem {
 	#[serde(rename = "Initiated")]
 	pub initiated: Value,
@@ -157,7 +157,7 @@ pub struct ListMultipartItem {
 	pub storage_class: Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct ListMultipartUploadsResult {
 	#[serde(serialize_with = "xmlns_tag")]
 	pub xmlns: (),
@@ -187,7 +187,7 @@ pub struct ListMultipartUploadsResult {
 	pub encoding_type: Option<Value>,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct PartItem {
 	#[serde(rename = "ETag")]
 	pub etag: Value,
@@ -199,7 +199,7 @@ pub struct PartItem {
 	pub size: IntValue,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct ListPartsResult {
 	#[serde(serialize_with = "xmlns_tag")]
 	pub xmlns: (),
@@ -227,7 +227,7 @@ pub struct ListPartsResult {
 	pub storage_class: Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct ListBucketItem {
 	#[serde(rename = "Key")]
 	pub key: Value,
@@ -241,13 +241,13 @@ pub struct ListBucketItem {
 	pub storage_class: Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct CommonPrefix {
 	#[serde(rename = "Prefix")]
 	pub prefix: Value,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct ListBucketResult {
 	#[serde(serialize_with = "xmlns_tag")]
 	pub xmlns: (),
@@ -281,7 +281,7 @@ pub struct ListBucketResult {
 	pub common_prefixes: Vec<CommonPrefix>,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct VersioningConfiguration {
 	#[serde(serialize_with = "xmlns_tag")]
 	pub xmlns: (),
@@ -289,7 +289,7 @@ pub struct VersioningConfiguration {
 	pub status: Option<Value>,
 }
 
-#[derive(Debug, Serialize, PartialEq)]
+#[derive(Debug, Serialize, PartialEq, Eq)]
 pub struct PostObject {
 	#[serde(serialize_with = "xmlns_tag")]
 	pub xmlns: (),

src/api/signature/mod.rs

@@ -1,5 +1,5 @@
 use chrono::{DateTime, Utc};
-use hmac::{Hmac, Mac, NewMac};
+use hmac::{Hmac, Mac};
 use sha2::Sha256;
 
 use garage_util::data::{sha256sum, Hash};
@@ -29,17 +29,17 @@ pub fn signing_hmac(
 	secret_key: &str,
 	region: &str,
 	service: &str,
-) -> Result<HmacSha256, crypto_mac::InvalidKeyLength> {
+) -> Result<HmacSha256, crypto_common::InvalidLength> {
 	let secret = String::from("AWS4") + secret_key;
-	let mut date_hmac = HmacSha256::new_varkey(secret.as_bytes())?;
+	let mut date_hmac = HmacSha256::new_from_slice(secret.as_bytes())?;
 	date_hmac.update(datetime.format(SHORT_DATE).to_string().as_bytes());
-	let mut region_hmac = HmacSha256::new_varkey(&date_hmac.finalize().into_bytes())?;
+	let mut region_hmac = HmacSha256::new_from_slice(&date_hmac.finalize().into_bytes())?;
 	region_hmac.update(region.as_bytes());
-	let mut service_hmac = HmacSha256::new_varkey(&region_hmac.finalize().into_bytes())?;
+	let mut service_hmac = HmacSha256::new_from_slice(&region_hmac.finalize().into_bytes())?;
 	service_hmac.update(service.as_bytes());
-	let mut signing_hmac = HmacSha256::new_varkey(&service_hmac.finalize().into_bytes())?;
+	let mut signing_hmac = HmacSha256::new_from_slice(&service_hmac.finalize().into_bytes())?;
 	signing_hmac.update(b"aws4_request");
-	let hmac = HmacSha256::new_varkey(&signing_hmac.finalize().into_bytes())?;
+	let hmac = HmacSha256::new_from_slice(&signing_hmac.finalize().into_bytes())?;
 	Ok(hmac)
 }
 

src/block/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "garage_block"
-version = "0.7.0"
+version = "0.8.0"
 authors = ["Alex Auvolat <alex@adnab.me>"]
 edition = "2018"
 license = "AGPL-3.0"
@@ -15,9 +15,9 @@ path = "lib.rs"
 
 [dependencies]
 garage_db = { version = "0.8.0", path = "../db" }
-garage_rpc = { version = "0.7.0", path = "../rpc" }
-garage_util = { version = "0.7.0", path = "../util" }
-garage_table = { version = "0.7.0", path = "../table" }
+garage_rpc = { version = "0.8.0", path = "../rpc" }
+garage_util = { version = "0.8.0", path = "../util" }
+garage_table = { version = "0.8.0", path = "../table" }
 
 opentelemetry = "0.17"
 
@@ -27,6 +27,8 @@ bytes = "1.0"
 hex = "0.4"
 tracing = "0.1.30"
 rand = "0.8"
+
+async-compression = { version = "0.3", features = ["tokio", "zstd"] }
 zstd = { version = "0.9", default-features = false }
 
 rmp-serde = "0.15"
@@ -36,3 +38,7 @@ serde_bytes = "0.11"
 futures = "0.3"
 futures-util = "0.3"
 tokio = { version = "1.0", default-features = false, features = ["rt", "rt-multi-thread", "io-util", "net", "time", "macros", "sync", "signal", "fs"] }
+tokio-util = { version = "0.6", features = ["io"] }
+
+[features]
+system-libs = [ "zstd/pkg-config" ]

src/block/block.rs

@@ -1,16 +1,22 @@
+use bytes::Bytes;
 use serde::{Deserialize, Serialize};
 use zstd::stream::{decode_all as zstd_decode, Encoder};
 
 use garage_util::data::*;
 use garage_util::error::*;
 
+#[derive(Debug, Serialize, Deserialize, Copy, Clone)]
+pub enum DataBlockHeader {
+	Plain,
+	Compressed,
+}
+
 /// A possibly compressed block of data
-#[derive(Debug, Serialize, Deserialize)]
 pub enum DataBlock {
 	/// Uncompressed data
-	Plain(#[serde(with = "serde_bytes")] Vec<u8>),
+	Plain(Bytes),
 	/// Data compressed with zstd
-	Compressed(#[serde(with = "serde_bytes")] Vec<u8>),
+	Compressed(Bytes),
 }
 
 impl DataBlock {
@@ -30,7 +36,7 @@ impl DataBlock {
 	/// Get the buffer, possibly decompressing it, and verify it's integrity.
 	/// For Plain block, data is compared to hash, for Compressed block, zstd checksumming system
 	/// is used instead.
-	pub fn verify_get(self, hash: Hash) -> Result<Vec<u8>, Error> {
+	pub fn verify_get(self, hash: Hash) -> Result<Bytes, Error> {
 		match self {
 			DataBlock::Plain(data) => {
 				if blake2sum(&data) == hash {
@@ -39,9 +45,9 @@ impl DataBlock {
 					Err(Error::CorruptData(hash))
 				}
 			}
-			DataBlock::Compressed(data) => {
-				zstd_decode(&data[..]).map_err(|_| Error::CorruptData(hash))
-			}
+			DataBlock::Compressed(data) => zstd_decode(&data[..])
+				.map_err(|_| Error::CorruptData(hash))
+				.map(Bytes::from),
 		}
 	}
 
@@ -61,13 +67,31 @@ impl DataBlock {
 		}
 	}
 
-	pub fn from_buffer(data: Vec<u8>, level: Option<i32>) -> DataBlock {
-		if let Some(level) = level {
-			if let Ok(data) = zstd_encode(&data[..], level) {
-				return DataBlock::Compressed(data);
+	pub async fn from_buffer(data: Bytes, level: Option<i32>) -> DataBlock {
+		tokio::task::spawn_blocking(move || {
+			if let Some(level) = level {
+				if let Ok(data) = zstd_encode(&data[..], level) {
+					return DataBlock::Compressed(data.into());
+				}
 			}
+			DataBlock::Plain(data)
+		})
+		.await
+		.unwrap()
+	}
+
+	pub fn into_parts(self) -> (DataBlockHeader, Bytes) {
+		match self {
+			DataBlock::Plain(data) => (DataBlockHeader::Plain, data),
+			DataBlock::Compressed(data) => (DataBlockHeader::Compressed, data),
+		}
+	}
+
+	pub fn from_parts(h: DataBlockHeader, bytes: Bytes) -> Self {
+		match h {
+			DataBlockHeader::Plain => DataBlock::Plain(bytes),
+			DataBlockHeader::Compressed => DataBlock::Compressed(bytes),
 		}
-		DataBlock::Plain(data)
 	}
 }
 

src/block/lib.rs

@@ -3,6 +3,7 @@ extern crate tracing;
 
 pub mod manager;
 pub mod repair;
+pub mod resync;
 
 mod block;
 mod metrics;

src/block/repair.rs

@@ -19,7 +19,17 @@ use garage_util::tranquilizer::Tranquilizer;
 
 use crate::manager::*;
 
-const SCRUB_INTERVAL: Duration = Duration::from_secs(3600 * 24 * 30); // full scrub every 30 days
+// Full scrub every 30 days
+const SCRUB_INTERVAL: Duration = Duration::from_secs(3600 * 24 * 30);
+// Scrub tranquility is initially set to 4, but can be changed in the CLI
+// and the updated version is persisted over Garage restarts
+const INITIAL_SCRUB_TRANQUILITY: u32 = 4;
+
+// ---- ---- ----
+// FIRST KIND OF REPAIR: FINDING MISSING BLOCKS/USELESS BLOCKS
+// This is a one-shot repair operation that can be launched,
+// checks everything, and then exits.
+// ---- ---- ----
 
 pub struct RepairWorker {
 	manager: Arc<BlockManager>,
@@ -102,7 +112,9 @@ impl Worker for RepairWorker {
 				}
 
 				for hash in batch_of_hashes.into_iter() {
-					self.manager.put_to_resync(&hash, Duration::from_secs(0))?;
+					self.manager
+						.resync
+						.put_to_resync(&hash, Duration::from_secs(0))?;
 					self.next_start = Some(hash)
 				}
 
@@ -114,7 +126,9 @@ impl Worker for RepairWorker {
 				// This allows us to find blocks we are storing but don't actually need,
 				// so that we can offload them if necessary and then delete them locally.
 				if let Some(hash) = bi.next().await? {
-					self.manager.put_to_resync(&hash, Duration::from_secs(0))?;
+					self.manager
+						.resync
+						.put_to_resync(&hash, Duration::from_secs(0))?;
 					Ok(WorkerState::Busy)
 				} else {
 					Ok(WorkerState::Done)
@@ -128,7 +142,13 @@ impl Worker for RepairWorker {
 	}
 }
 
-// ----
+// ---- ---- ----
+// SECOND KIND OF REPAIR: SCRUBBING THE DATASTORE
+// This is significantly more complex than the process above,
+// as it is a continuously-running task that triggers automatically
+// every SCRUB_INTERVAL, but can also be triggered manually
+// and whose parameter (esp. speed) can be controlled at runtime.
+// ---- ---- ----
 
 pub struct ScrubWorker {
 	manager: Arc<BlockManager>,
@@ -176,7 +196,7 @@ impl ScrubWorker {
 			Ok(v) => v,
 			Err(_) => ScrubWorkerPersisted {
 				time_last_complete_scrub: 0,
-				tranquility: 4,
+				tranquility: INITIAL_SCRUB_TRANQUILITY,
 				corruptions_detected: 0,
 			},
 		};
@@ -343,7 +363,9 @@ impl Worker for ScrubWorker {
 	}
 }
 
-// ----
+// ---- ---- ----
+// UTILITY FOR ENUMERATING THE BLOCK STORE
+// ---- ---- ----
 
 struct BlockStoreIterator {
 	path: Vec<ReadingDir>,

src/block/resync.rs

@@ -0,0 +1,589 @@
+use std::collections::HashSet;
+use std::convert::TryInto;
+use std::sync::{Arc, Mutex};
+use std::time::Duration;
+
+use arc_swap::ArcSwap;
+use async_trait::async_trait;
+use serde::{Deserialize, Serialize};
+
+use tokio::select;
+use tokio::sync::{watch, Notify};
+
+use opentelemetry::{
+	trace::{FutureExt as OtelFutureExt, TraceContextExt, Tracer},
+	Context, KeyValue,
+};
+
+use garage_db as db;
+use garage_db::counted_tree_hack::CountedTree;
+
+use garage_util::background::*;
+use garage_util::data::*;
+use garage_util::error::*;
+use garage_util::metrics::RecordDuration;
+use garage_util::persister::Persister;
+use garage_util::time::*;
+use garage_util::tranquilizer::Tranquilizer;
+
+use garage_rpc::system::System;
+use garage_rpc::*;
+
+use garage_table::replication::TableReplication;
+
+use crate::manager::*;
+
+// The delay between the time where a resync operation fails
+// and the time when it is retried, with exponential backoff
+// (multiplied by 2, 4, 8, 16, etc. for every consecutive failure).
+pub(crate) const RESYNC_RETRY_DELAY: Duration = Duration::from_secs(60);
+// The minimum retry delay is 60 seconds = 1 minute
+// The maximum retry delay is 60 seconds * 2^6 = 60 seconds << 6 = 64 minutes (~1 hour)
+pub(crate) const RESYNC_RETRY_DELAY_MAX_BACKOFF_POWER: u64 = 6;
+
+// No more than 4 resync workers can be running in the system
+pub(crate) const MAX_RESYNC_WORKERS: usize = 4;
+// Resync tranquility is initially set to 2, but can be changed in the CLI
+// and the updated version is persisted over Garage restarts
+const INITIAL_RESYNC_TRANQUILITY: u32 = 2;
+
+pub struct BlockResyncManager {
+	pub(crate) queue: CountedTree,
+	pub(crate) notify: Notify,
+	pub(crate) errors: CountedTree,
+
+	busy_set: BusySet,
+
+	persister: Persister<ResyncPersistedConfig>,
+	persisted: ArcSwap<ResyncPersistedConfig>,
+}
+
+#[derive(Serialize, Deserialize, Clone, Copy)]
+struct ResyncPersistedConfig {
+	n_workers: usize,
+	tranquility: u32,
+}
+
+enum ResyncIterResult {
+	BusyDidSomething,
+	BusyDidNothing,
+	IdleFor(Duration),
+}
+
+type BusySet = Arc<Mutex<HashSet<Vec<u8>>>>;
+
+struct BusyBlock {
+	time_bytes: Vec<u8>,
+	hash_bytes: Vec<u8>,
+	busy_set: BusySet,
+}
+
+impl BlockResyncManager {
+	pub(crate) fn new(db: &db::Db, system: &System) -> Self {
+		let queue = db
+			.open_tree("block_local_resync_queue")
+			.expect("Unable to open block_local_resync_queue tree");
+		let queue = CountedTree::new(queue).expect("Could not count block_local_resync_queue");
+
+		let errors = db
+			.open_tree("block_local_resync_errors")
+			.expect("Unable to open block_local_resync_errors tree");
+		let errors = CountedTree::new(errors).expect("Could not count block_local_resync_errors");
+
+		let persister = Persister::new(&system.metadata_dir, "resync_cfg");
+		let persisted = match persister.load() {
+			Ok(v) => v,
+			Err(_) => ResyncPersistedConfig {
+				n_workers: 1,
+				tranquility: INITIAL_RESYNC_TRANQUILITY,
+			},
+		};
+
+		Self {
+			queue,
+			notify: Notify::new(),
+			errors,
+			busy_set: Arc::new(Mutex::new(HashSet::new())),
+			persister,
+			persisted: ArcSwap::new(Arc::new(persisted)),
+		}
+	}
+
+	/// Get lenght of resync queue
+	pub fn queue_len(&self) -> Result<usize, Error> {
+		// This currently can't return an error because the CountedTree hack
+		// doesn't error on .len(), but this will change when we remove the hack
+		// (hopefully someday!)
+		Ok(self.queue.len())
+	}
+
+	/// Get number of blocks that have an error
+	pub fn errors_len(&self) -> Result<usize, Error> {
+		// (see queue_len comment)
+		Ok(self.errors.len())
+	}
+
+	// ---- Resync loop ----
+
+	// This part manages a queue of blocks that need to be
+	// "resynchronized", i.e. that need to have a check that
+	// they are at present if we need them, or that they are
+	// deleted once the garbage collection delay has passed.
+	//
+	// Here are some explanations on how the resync queue works.
+	// There are two Sled trees that are used to have information
+	// about the status of blocks that need to be resynchronized:
+	//
+	// - resync.queue: a tree that is ordered first by a timestamp
+	//   (in milliseconds since Unix epoch) that is the time at which
+	//   the resync must be done, and second by block hash.
+	//   The key in this tree is just:
+	//       concat(timestamp (8 bytes), hash (32 bytes))
+	//   The value is the same 32-byte hash.
+	//
+	// - resync.errors: a tree that indicates for each block
+	//   if the last resync resulted in an error, and if so,
+	//   the following two informations (see the ErrorCounter struct):
+	//   - how many consecutive resync errors for this block?
+	//   - when was the last try?
+	//   These two informations are used to implement an
+	//   exponential backoff retry strategy.
+	//   The key in this tree is the 32-byte hash of the block,
+	//   and the value is the encoded ErrorCounter value.
+	//
+	// We need to have these two trees, because the resync queue
+	// is not just a queue of items to process, but a set of items
+	// that are waiting a specific delay until we can process them
+	// (the delay being necessary both internally for the exponential
+	// backoff strategy, and exposed as a parameter when adding items
+	// to the queue, e.g. to wait until the GC delay has passed).
+	// This is why we need one tree ordered by time, and one
+	// ordered by identifier of item to be processed (block hash).
+	//
+	// When the worker wants to process an item it takes from
+	// resync.queue, it checks in resync.errors that if there is an
+	// exponential back-off delay to await, it has passed before we
+	// process the item. If not, the item in the queue is skipped
+	// (but added back for later processing after the time of the
+	// delay).
+	//
+	// An alternative that would have seemed natural is to
+	// only add items to resync.queue with a processing time that is
+	// after the delay, but there are several issues with this:
+	// - This requires to synchronize updates to resync.queue and
+	//   resync.errors (with the current model, there is only one thread,
+	//   the worker thread, that accesses resync.errors,
+	//   so no need to synchronize) by putting them both in a lock.
+	//   This would mean that block_incref might need to take a lock
+	//   before doing its thing, meaning it has much more chances of
+	//   not completing successfully if something bad happens to Garage.
+	//   Currently Garage is not able to recover from block_incref that
+	//   doesn't complete successfully, because it is necessary to ensure
+	//   the consistency between the state of the block manager and
+	//   information in the BlockRef table.
+	// - If a resync fails, we put that block in the resync.errors table,
+	//   and also add it back to resync.queue to be processed after
+	//   the exponential back-off delay,
+	//   but maybe the block is already scheduled to be resynced again
+	//   at another time that is before the exponential back-off delay,
+	//   and we have no way to check that easily. This means that
+	//   in all cases, we need to check the resync.errors table
+	//   in the resync loop at the time when a block is popped from
+	//   the resync.queue.
+	// Overall, the current design is therefore simpler and more robust
+	// because it tolerates inconsistencies between the resync.queue
+	// and resync.errors table (items being scheduled in resync.queue
+	// for times that are earlier than the exponential back-off delay
+	// is a natural condition that is handled properly).
+
+	pub(crate) fn put_to_resync(&self, hash: &Hash, delay: Duration) -> db::Result<()> {
+		let when = now_msec() + delay.as_millis() as u64;
+		self.put_to_resync_at(hash, when)
+	}
+
+	pub(crate) fn put_to_resync_at(&self, hash: &Hash, when: u64) -> db::Result<()> {
+		trace!("Put resync_queue: {} {:?}", when, hash);
+		let mut key = u64::to_be_bytes(when).to_vec();
+		key.extend(hash.as_ref());
+		self.queue.insert(key, hash.as_ref())?;
+		self.notify.notify_waiters();
+		Ok(())
+	}
+
+	async fn resync_iter(&self, manager: &BlockManager) -> Result<ResyncIterResult, db::Error> {
+		if let Some(block) = self.get_block_to_resync()? {
+			let time_msec = u64::from_be_bytes(block.time_bytes[0..8].try_into().unwrap());
+			let now = now_msec();
+
+			if now >= time_msec {
+				let hash = Hash::try_from(&block.hash_bytes[..]).unwrap();
+
+				if let Some(ec) = self.errors.get(hash.as_slice())? {
+					let ec = ErrorCounter::decode(&ec);
+					if now < ec.next_try() {
+						// if next retry after an error is not yet,
+						// don't do resync and return early, but still
+						// make sure the item is still in queue at expected time
+						self.put_to_resync_at(&hash, ec.next_try())?;
+						// ec.next_try() > now >= time_msec, so this remove
+						// is not removing the one we added just above
+						// (we want to do the remove after the insert to ensure
+						// that the item is not lost if we crash in-between)
+						self.queue.remove(&block.time_bytes)?;
+						return Ok(ResyncIterResult::BusyDidNothing);
+					}
+				}
+
+				let tracer = opentelemetry::global::tracer("garage");
+				let trace_id = gen_uuid();
+				let span = tracer
+					.span_builder("Resync block")
+					.with_trace_id(
+						opentelemetry::trace::TraceId::from_hex(&hex::encode(
+							&trace_id.as_slice()[..16],
+						))
+						.unwrap(),
+					)
+					.with_attributes(vec![KeyValue::new("block", format!("{:?}", hash))])
+					.start(&tracer);
+
+				let res = self
+					.resync_block(manager, &hash)
+					.with_context(Context::current_with_span(span))
+					.bound_record_duration(&manager.metrics.resync_duration)
+					.await;
+
+				manager.metrics.resync_counter.add(1);
+
+				if let Err(e) = &res {
+					manager.metrics.resync_error_counter.add(1);
+					warn!("Error when resyncing {:?}: {}", hash, e);
+
+					let err_counter = match self.errors.get(hash.as_slice())? {
+						Some(ec) => ErrorCounter::decode(&ec).add1(now + 1),
+						None => ErrorCounter::new(now + 1),
+					};
+
+					self.errors.insert(hash.as_slice(), err_counter.encode())?;
+
+					self.put_to_resync_at(&hash, err_counter.next_try())?;
+					// err_counter.next_try() >= now + 1 > now,
+					// the entry we remove from the queue is not
+					// the entry we inserted with put_to_resync_at
+					self.queue.remove(&block.time_bytes)?;
+				} else {
+					self.errors.remove(hash.as_slice())?;
+					self.queue.remove(&block.time_bytes)?;
+				}
+
+				Ok(ResyncIterResult::BusyDidSomething)
+			} else {
+				Ok(ResyncIterResult::IdleFor(Duration::from_millis(
+					time_msec - now,
+				)))
+			}
+		} else {
+			// Here we wait either for a notification that an item has been
+			// added to the queue, or for a constant delay of 10 secs to expire.
+			// The delay avoids a race condition where the notification happens
+			// between the time we checked the queue and the first poll
+			// to resync_notify.notified(): if that happens, we'll just loop
+			// back 10 seconds later, which is fine.
+			Ok(ResyncIterResult::IdleFor(Duration::from_secs(10)))
+		}
+	}
+
+	fn get_block_to_resync(&self) -> Result<Option<BusyBlock>, db::Error> {
+		let mut busy = self.busy_set.lock().unwrap();
+		for it in self.queue.iter()? {
+			let (time_bytes, hash_bytes) = it?;
+			if !busy.contains(&time_bytes) {
+				busy.insert(time_bytes.clone());
+				return Ok(Some(BusyBlock {
+					time_bytes,
+					hash_bytes,
+					busy_set: self.busy_set.clone(),
+				}));
+			}
+		}
+		Ok(None)
+	}
+
+	async fn resync_block(&self, manager: &BlockManager, hash: &Hash) -> Result<(), Error> {
+		let BlockStatus { exists, needed } = manager.check_block_status(hash).await?;
+
+		if exists != needed.is_needed() || exists != needed.is_nonzero() {
+			debug!(
+				"Resync block {:?}: exists {}, nonzero rc {}, deletable {}",
+				hash,
+				exists,
+				needed.is_nonzero(),
+				needed.is_deletable(),
+			);
+		}
+
+		if exists && needed.is_deletable() {
+			info!("Resync block {:?}: offloading and deleting", hash);
+
+			let mut who = manager.replication.write_nodes(hash);
+			if who.len() < manager.replication.write_quorum() {
+				return Err(Error::Message("Not trying to offload block because we don't have a quorum of nodes to write to".to_string()));
+			}
+			who.retain(|id| *id != manager.system.id);
+
+			let who_needs_resps = manager
+				.system
+				.rpc
+				.call_many(
+					&manager.endpoint,
+					&who,
+					BlockRpc::NeedBlockQuery(*hash),
+					RequestStrategy::with_priority(PRIO_BACKGROUND),
+				)
+				.await?;
+
+			let mut need_nodes = vec![];
+			for (node, needed) in who_needs_resps {
+				match needed.err_context("NeedBlockQuery RPC")? {
+					BlockRpc::NeedBlockReply(needed) => {
+						if needed {
+							need_nodes.push(node);
+						}
+					}
+					m => {
+						return Err(Error::unexpected_rpc_message(m));
+					}
+				}
+			}
+
+			if !need_nodes.is_empty() {
+				trace!(
+					"Block {:?} needed by {} nodes, sending",
+					hash,
+					need_nodes.len()
+				);
+
+				for node in need_nodes.iter() {
+					manager
+						.metrics
+						.resync_send_counter
+						.add(1, &[KeyValue::new("to", format!("{:?}", node))]);
+				}
+
+				let block = manager.read_block(hash).await?;
+				let (header, bytes) = block.into_parts();
+				let put_block_message = Req::new(BlockRpc::PutBlock {
+					hash: *hash,
+					header,
+				})?
+				.with_stream_from_buffer(bytes);
+				manager
+					.system
+					.rpc
+					.try_call_many(
+						&manager.endpoint,
+						&need_nodes[..],
+						put_block_message,
+						RequestStrategy::with_priority(PRIO_BACKGROUND)
+							.with_quorum(need_nodes.len()),
+					)
+					.await
+					.err_context("PutBlock RPC")?;
+			}
+			info!(
+				"Deleting unneeded block {:?}, offload finished ({} / {})",
+				hash,
+				need_nodes.len(),
+				who.len()
+			);
+
+			manager.delete_if_unneeded(hash).await?;
+
+			manager.rc.clear_deleted_block_rc(hash)?;
+		}
+
+		if needed.is_nonzero() && !exists {
+			info!(
+				"Resync block {:?}: fetching absent but needed block (refcount > 0)",
+				hash
+			);
+
+			let block_data = manager.rpc_get_raw_block(hash, None).await?;
+
+			manager.metrics.resync_recv_counter.add(1);
+
+			manager.write_block(hash, &block_data).await?;
+		}
+
+		Ok(())
+	}
+
+	async fn update_persisted(
+		&self,
+		update: impl Fn(&mut ResyncPersistedConfig),
+	) -> Result<(), Error> {
+		let mut cfg: ResyncPersistedConfig = *self.persisted.load().as_ref();
+		update(&mut cfg);
+		self.persister.save_async(&cfg).await?;
+		self.persisted.store(Arc::new(cfg));
+		self.notify.notify_waiters();
+		Ok(())
+	}
+
+	pub async fn set_n_workers(&self, n_workers: usize) -> Result<(), Error> {
+		if !(1..=MAX_RESYNC_WORKERS).contains(&n_workers) {
+			return Err(Error::Message(format!(
+				"Invalid number of resync workers, must be between 1 and {}",
+				MAX_RESYNC_WORKERS
+			)));
+		}
+		self.update_persisted(|cfg| cfg.n_workers = n_workers).await
+	}
+
+	pub async fn set_tranquility(&self, tranquility: u32) -> Result<(), Error> {
+		self.update_persisted(|cfg| cfg.tranquility = tranquility)
+			.await
+	}
+}
+
+impl Drop for BusyBlock {
+	fn drop(&mut self) {
+		let mut busy = self.busy_set.lock().unwrap();
+		busy.remove(&self.time_bytes);
+	}
+}
+
+pub(crate) struct ResyncWorker {
+	index: usize,
+	manager: Arc<BlockManager>,
+	tranquilizer: Tranquilizer,
+	next_delay: Duration,
+}
+
+impl ResyncWorker {
+	pub(crate) fn new(index: usize, manager: Arc<BlockManager>) -> Self {
+		Self {
+			index,
+			manager,
+			tranquilizer: Tranquilizer::new(30),
+			next_delay: Duration::from_secs(10),
+		}
+	}
+}
+
+#[async_trait]
+impl Worker for ResyncWorker {
+	fn name(&self) -> String {
+		format!("Block resync worker #{}", self.index + 1)
+	}
+
+	fn info(&self) -> Option<String> {
+		let persisted = self.manager.resync.persisted.load();
+
+		if self.index >= persisted.n_workers {
+			return Some("(unused)".into());
+		}
+
+		let mut ret = vec![];
+		ret.push(format!("tranquility = {}", persisted.tranquility));
+
+		let qlen = self.manager.resync.queue_len().unwrap_or(0);
+		if qlen > 0 {
+			ret.push(format!("{} blocks in queue", qlen));
+		}
+
+		let elen = self.manager.resync.errors_len().unwrap_or(0);
+		if elen > 0 {
+			ret.push(format!("{} blocks in error state", elen));
+		}
+
+		Some(ret.join(", "))
+	}
+
+	async fn work(&mut self, _must_exit: &mut watch::Receiver<bool>) -> Result<WorkerState, Error> {
+		if self.index >= self.manager.resync.persisted.load().n_workers {
+			return Ok(WorkerState::Idle);
+		}
+
+		self.tranquilizer.reset();
+		match self.manager.resync.resync_iter(&self.manager).await {
+			Ok(ResyncIterResult::BusyDidSomething) => Ok(self
+				.tranquilizer
+				.tranquilize_worker(self.manager.resync.persisted.load().tranquility)),
+			Ok(ResyncIterResult::BusyDidNothing) => Ok(WorkerState::Busy),
+			Ok(ResyncIterResult::IdleFor(delay)) => {
+				self.next_delay = delay;
+				Ok(WorkerState::Idle)
+			}
+			Err(e) => {
+				// The errors that we have here are only Sled errors
+				// We don't really know how to handle them so just ¯\_(ツ)_/¯
+				// (there is kind of an assumption that Sled won't error on us,
+				// if it does there is not much we can do -- TODO should we just panic?)
+				// Here we just give the error to the worker manager,
+				// it will print it to the logs and increment a counter
+				Err(e.into())
+			}
+		}
+	}
+
+	async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerState {
+		while self.index >= self.manager.resync.persisted.load().n_workers {
+			self.manager.resync.notify.notified().await
+		}
+
+		select! {
+			_ = tokio::time::sleep(self.next_delay) => (),
+			_ = self.manager.resync.notify.notified() => (),
+		};
+
+		WorkerState::Busy
+	}
+}
+
+/// Counts the number of errors when resyncing a block,
+/// and the time of the last try.
+/// Used to implement exponential backoff.
+#[derive(Clone, Copy, Debug)]
+struct ErrorCounter {
+	errors: u64,
+	last_try: u64,
+}
+
+impl ErrorCounter {
+	fn new(now: u64) -> Self {
+		Self {
+			errors: 1,
+			last_try: now,
+		}
+	}
+
+	fn decode(data: &[u8]) -> Self {
+		Self {
+			errors: u64::from_be_bytes(data[0..8].try_into().unwrap()),
+			last_try: u64::from_be_bytes(data[8..16].try_into().unwrap()),
+		}
+	}
+	fn encode(&self) -> Vec<u8> {
+		[
+			u64::to_be_bytes(self.errors),
+			u64::to_be_bytes(self.last_try),
+		]
+		.concat()
+	}
+
+	fn add1(self, now: u64) -> Self {
+		Self {
+			errors: self.errors + 1,
+			last_try: now,
+		}
+	}
+
+	fn delay_msec(&self) -> u64 {
+		(RESYNC_RETRY_DELAY.as_millis() as u64)
+			<< std::cmp::min(self.errors - 1, RESYNC_RETRY_DELAY_MAX_BACKOFF_POWER)
+	}
+	fn next_try(&self) -> u64 {
+		self.last_try + self.delay_msec()
+	}
+}

src/db/Cargo.toml

@@ -21,9 +21,9 @@ err-derive = "0.3"
 hexdump = "0.1"
 tracing = "0.1.30"
 
-heed = "0.11"
-rusqlite = { version = "0.27", features = ["bundled"] }
-sled = "0.34"
+heed = { version = "0.11", default-features = false, features = ["lmdb"], optional = true }
+rusqlite = { version = "0.27", optional = true }
+sled = { version = "0.34", optional = true }
 
 # cli deps
 clap = { version = "3.1.18", optional = true, features = ["derive", "env"] }
@@ -33,4 +33,7 @@ pretty_env_logger = { version = "0.4", optional = true }
 mktemp = "0.4"
 
 [features]
+bundled-libs = [ "rusqlite/bundled" ]
 cli = ["clap", "pretty_env_logger"]
+lmdb = [ "heed" ]
+sqlite = [ "rusqlite" ]

src/db/lib.rs

@@ -1,5 +1,15 @@
+#[macro_use]
+#[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")]
 pub mod sled_adapter;
+#[cfg(feature = "sqlite")]
 pub mod sqlite_adapter;
 
 pub mod counted_tree_hack;

src/db/lmdb_adapter.rs

@@ -345,7 +345,6 @@ pub fn recommended_map_size() -> usize {
 
 #[cfg(target_pointer_width = "32")]
 pub fn recommended_map_size() -> usize {
-	use log::warn;
 	warn!("LMDB is not recommended on 32-bit systems, database size will be limited");
 	1usize << 30
 }

src/db/sqlite_adapter.rs

@@ -6,8 +6,6 @@ use std::pin::Pin;
 use std::ptr::NonNull;
 use std::sync::{Arc, Mutex, MutexGuard};
 
-use tracing::trace;
-
 use rusqlite::{params, Connection, Rows, Statement, Transaction};
 
 use crate::{

src/garage/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "garage"
-version = "0.7.0"
+version = "0.8.0"
 authors = ["Alex Auvolat <alex@adnab.me>"]
 edition = "2018"
 license = "AGPL-3.0"
@@ -22,20 +22,20 @@ path = "tests/lib.rs"
 
 [dependencies]
 garage_db = { version = "0.8.0", path = "../db" }
-garage_api = { version = "0.7.0", path = "../api" }
-garage_block = { version = "0.7.0", path = "../block" }
-garage_model = { version = "0.7.0", path = "../model" }
-garage_rpc = { version = "0.7.0", path = "../rpc" }
-garage_table = { version = "0.7.0", path = "../table" }
-garage_util = { version = "0.7.0", path = "../util" }
-garage_web = { version = "0.7.0", path = "../web" }
+garage_api = { version = "0.8.0", path = "../api" }
+garage_block = { version = "0.8.0", path = "../block" }
+garage_model = { version = "0.8.0", path = "../model" }
+garage_rpc = { version = "0.8.0", path = "../rpc" }
+garage_table = { version = "0.8.0", path = "../table" }
+garage_util = { version = "0.8.0", path = "../util" }
+garage_web = { version = "0.8.0", path = "../web" }
 
 bytes = "1.0"
 bytesize = "1.1"
 timeago = "0.3"
 hex = "0.4"
 tracing = { version = "0.1.30", features = ["log-always"] }
-pretty_env_logger = "0.4"
+tracing-subscriber = { version = "0.3", features = ["env-filter"] }
 rand = "0.8"
 async-trait = "0.1.7"
 sodiumoxide = { version = "0.2.5-0", package = "kuska-sodiumoxide" }
@@ -50,22 +50,20 @@ futures = "0.3"
 futures-util = "0.3"
 tokio = { version = "1.0", default-features = false, features = ["rt", "rt-multi-thread", "io-util", "net", "time", "macros", "sync", "signal", "fs"] }
 
-#netapp = { version = "0.3.2", git = "https://git.deuxfleurs.fr/lx/netapp" }
-#netapp = { version = "0.4", path = "../../../netapp" }
-netapp = "0.4"
+netapp = "0.5"
 
 opentelemetry = { version = "0.17", features = [ "rt-tokio" ] }
-opentelemetry-prometheus = "0.10"
-opentelemetry-otlp = "0.10"
-prometheus = "0.13"
+opentelemetry-prometheus = { version = "0.10", optional = true }
+opentelemetry-otlp = { version = "0.10", optional = true }
+prometheus = { version = "0.13", optional = true }
 
 [dev-dependencies]
 aws-sdk-s3 = "0.8"
 chrono = "0.4"
 http = "0.2"
-hmac = "0.10"
+hmac = "0.12"
 hyper = { version = "0.14", features = ["client", "http1", "runtime"] }
-sha2 = "0.9"
+sha2 = "0.10"
 
 static_init = "1.0"
 assert-json-diff = "2.0"
@@ -74,5 +72,26 @@ base64 = "0.13"
 
 
 [features]
-kubernetes-discovery = [ "garage_rpc/kubernetes-discovery" ]
+default = [ "bundled-libs", "metrics", "sled" ]
+
 k2v = [ "garage_util/k2v", "garage_api/k2v" ]
+
+# Database engines, Sled is still our default even though we don't like it
+sled = [ "garage_model/sled" ]
+lmdb = [ "garage_model/lmdb" ]
+sqlite = [ "garage_model/sqlite" ]
+
+# Automatic registration and discovery via Kubernetes API
+kubernetes-discovery = [ "garage_rpc/kubernetes-discovery" ]
+# Prometheus exporter (/metrics endpoint).
+metrics = [ "garage_api/metrics", "opentelemetry-prometheus", "prometheus" ]
+# Exporter for the OpenTelemetry Collector.
+telemetry-otlp = [ "opentelemetry-otlp" ]
+
+# NOTE: bundled-libs and system-libs should be treat as mutually exclusive;
+# exactly one of them should be enabled.
+
+# Use bundled libsqlite instead of linking against system-provided.
+bundled-libs = [ "garage_db/bundled-libs" ]
+# Link against system-provided libsodium and libzstd.
+system-libs = [ "garage_block/system-libs", "garage_rpc/system-libs", "sodiumoxide/use-pkg-config" ]

src/garage/admin.rs

@@ -15,6 +15,8 @@ use garage_table::*;
 
 use garage_rpc::*;
 
+use garage_block::repair::ScrubWorkerCommand;
+
 use garage_model::bucket_alias_table::*;
 use garage_model::bucket_table::*;
 use garage_model::garage::Garage;
@@ -681,7 +683,7 @@ impl AdminRpcHandler {
 					.endpoint
 					.call(
 						&node,
-						&AdminRpc::LaunchRepair(opt_to_send.clone()),
+						AdminRpc::LaunchRepair(opt_to_send.clone()),
 						PRIO_NORMAL,
 					)
 					.await;
@@ -721,7 +723,7 @@ impl AdminRpcHandler {
 				let node_id = (*node).into();
 				match self
 					.endpoint
-					.call(&node_id, &AdminRpc::Stats(opt), PRIO_NORMAL)
+					.call(&node_id, AdminRpc::Stats(opt), PRIO_NORMAL)
 					.await?
 				{
 					Ok(AdminRpc::Ok(s)) => writeln!(&mut ret, "{}", s).unwrap(),
@@ -739,8 +741,11 @@ impl AdminRpcHandler {
 		let mut ret = String::new();
 		writeln!(
 			&mut ret,
-			"\nGarage version: {}",
-			self.garage.system.garage_version(),
+			"\nGarage version: {} [features: {}]",
+			garage_util::version::garage_version(),
+			garage_util::version::garage_features()
+				.map(|list| list.join(", "))
+				.unwrap_or_else(|| "(unknown)".into()),
 		)
 		.unwrap();
 		writeln!(&mut ret, "\nDatabase engine: {}", self.garage.db.engine()).unwrap();
@@ -779,13 +784,13 @@ impl AdminRpcHandler {
 		writeln!(
 			&mut ret,
 			"  resync queue length: {}",
-			self.garage.block_manager.resync_queue_len()?
+			self.garage.block_manager.resync.queue_len()?
 		)
 		.unwrap();
 		writeln!(
 			&mut ret,
 			"  blocks with resync errors: {}",
-			self.garage.block_manager.resync_errors_len()?
+			self.garage.block_manager.resync.errors_len()?
 		)
 		.unwrap();
 
@@ -836,6 +841,32 @@ impl AdminRpcHandler {
 				let workers = self.garage.background.get_worker_info();
 				Ok(AdminRpc::WorkerList(workers, opt))
 			}
+			WorkerCmd::Set { opt } => match opt {
+				WorkerSetCmd::ScrubTranquility { tranquility } => {
+					let scrub_command = ScrubWorkerCommand::SetTranquility(tranquility);
+					self.garage
+						.block_manager
+						.send_scrub_command(scrub_command)
+						.await;
+					Ok(AdminRpc::Ok("Scrub tranquility updated".into()))
+				}
+				WorkerSetCmd::ResyncNWorkers { n_workers } => {
+					self.garage
+						.block_manager
+						.resync
+						.set_n_workers(n_workers)
+						.await?;
+					Ok(AdminRpc::Ok("Number of resync workers updated".into()))
+				}
+				WorkerSetCmd::ResyncTranquility { tranquility } => {
+					self.garage
+						.block_manager
+						.resync
+						.set_tranquility(tranquility)
+						.await?;
+					Ok(AdminRpc::Ok("Resync tranquility updated".into()))
+				}
+			},
 		}
 	}
 }

src/garage/cli/cmd.rs

@@ -47,7 +47,7 @@ pub async fn cli_command_dispatch(
 
 pub async fn cmd_status(rpc_cli: &Endpoint<SystemRpc, ()>, rpc_host: NodeID) -> Result<(), Error> {
 	let status = match rpc_cli
-		.call(&rpc_host, &SystemRpc::GetKnownNodes, PRIO_NORMAL)
+		.call(&rpc_host, SystemRpc::GetKnownNodes, PRIO_NORMAL)
 		.await??
 	{
 		SystemRpc::ReturnKnownNodes(nodes) => nodes,
@@ -149,7 +149,7 @@ pub async fn cmd_connect(
 	args: ConnectNodeOpt,
 ) -> Result<(), Error> {
 	match rpc_cli
-		.call(&rpc_host, &SystemRpc::Connect(args.node), PRIO_NORMAL)
+		.call(&rpc_host, SystemRpc::Connect(args.node), PRIO_NORMAL)
 		.await??
 	{
 		SystemRpc::Ok => {
@@ -165,7 +165,7 @@ pub async fn cmd_admin(
 	rpc_host: NodeID,
 	args: AdminRpc,
 ) -> Result<(), HelperError> {
-	match rpc_cli.call(&rpc_host, &args, PRIO_NORMAL).await?? {
+	match rpc_cli.call(&rpc_host, args, PRIO_NORMAL).await?? {
 		AdminRpc::Ok(msg) => {
 			println!("{}", msg);
 		}

src/garage/cli/layout.rs

@@ -36,7 +36,7 @@ pub async fn cmd_assign_role(
 	args: AssignRoleOpt,
 ) -> Result<(), Error> {
 	let status = match rpc_cli
-		.call(&rpc_host, &SystemRpc::GetKnownNodes, PRIO_NORMAL)
+		.call(&rpc_host, SystemRpc::GetKnownNodes, PRIO_NORMAL)
 		.await??
 	{
 		SystemRpc::ReturnKnownNodes(nodes) => nodes,
@@ -245,7 +245,7 @@ pub async fn fetch_layout(
 	rpc_host: NodeID,
 ) -> Result<ClusterLayout, Error> {
 	match rpc_cli
-		.call(&rpc_host, &SystemRpc::PullClusterLayout, PRIO_NORMAL)
+		.call(&rpc_host, SystemRpc::PullClusterLayout, PRIO_NORMAL)
 		.await??
 	{
 		SystemRpc::AdvertiseClusterLayout(t) => Ok(t),
@@ -261,7 +261,7 @@ pub async fn send_layout(
 	rpc_cli
 		.call(
 			&rpc_host,
-			&SystemRpc::AdvertiseClusterLayout(layout),
+			SystemRpc::AdvertiseClusterLayout(layout),
 			PRIO_NORMAL,
 		)
 		.await??;

src/garage/cli/structs.rs

@@ -1,64 +1,65 @@
 use serde::{Deserialize, Serialize};
-
 use structopt::StructOpt;
 
+use garage_util::version::garage_version;
+
 #[derive(StructOpt, Debug)]
 pub enum Command {
 	/// Run Garage server
-	#[structopt(name = "server")]
+	#[structopt(name = "server", version = garage_version())]
 	Server,
 
 	/// Get network status
-	#[structopt(name = "status")]
+	#[structopt(name = "status", version = garage_version())]
 	Status,
 
 	/// Operations on individual Garage nodes
-	#[structopt(name = "node")]
+	#[structopt(name = "node", version = garage_version())]
 	Node(NodeOperation),
 
 	/// Operations on the assignation of node roles in the cluster layout
-	#[structopt(name = "layout")]
+	#[structopt(name = "layout", version = garage_version())]
 	Layout(LayoutOperation),
 
 	/// Operations on buckets
-	#[structopt(name = "bucket")]
+	#[structopt(name = "bucket", version = garage_version())]
 	Bucket(BucketOperation),
 
 	/// Operations on S3 access keys
-	#[structopt(name = "key")]
+	#[structopt(name = "key", version = garage_version())]
 	Key(KeyOperation),
 
 	/// Run migrations from previous Garage version
 	/// (DO NOT USE WITHOUT READING FULL DOCUMENTATION)
-	#[structopt(name = "migrate")]
+	#[structopt(name = "migrate", version = garage_version())]
 	Migrate(MigrateOpt),
 
 	/// Start repair of node data on remote node
-	#[structopt(name = "repair")]
+	#[structopt(name = "repair", version = garage_version())]
 	Repair(RepairOpt),
 
 	/// Offline reparation of node data (these repairs must be run offline
 	/// directly on the server node)
-	#[structopt(name = "offline-repair")]
+	#[structopt(name = "offline-repair", version = garage_version())]
 	OfflineRepair(OfflineRepairOpt),
 
 	/// Gather node statistics
-	#[structopt(name = "stats")]
+	#[structopt(name = "stats", version = garage_version())]
 	Stats(StatsOpt),
 
 	/// Manage background workers
-	#[structopt(name = "worker")]
+	#[structopt(name = "worker", version = garage_version())]
 	Worker(WorkerOpt),
 }
 
 #[derive(StructOpt, Debug)]
 pub enum NodeOperation {
 	/// Print identifier (public key) of this Garage node
-	#[structopt(name = "id")]
+	#[structopt(name = "id", version = garage_version())]
 	NodeId(NodeIdOpt),
 
 	/// Connect to Garage node that is currently isolated from the system
-	#[structopt(name = "connect")]
+	#[structopt(name = "connect", version = garage_version())]
 	Connect(ConnectNodeOpt),
 }
 
@@ -79,23 +80,23 @@ pub struct ConnectNodeOpt {
 #[derive(StructOpt, Debug)]
 pub enum LayoutOperation {
 	/// Assign role to Garage node
-	#[structopt(name = "assign")]
+	#[structopt(name = "assign", version = garage_version())]
 	Assign(AssignRoleOpt),
 
 	/// Remove role from Garage cluster node
-	#[structopt(name = "remove")]
+	#[structopt(name = "remove", version = garage_version())]
 	Remove(RemoveRoleOpt),
 
 	/// Show roles currently assigned to nodes and changes staged for commit
-	#[structopt(name = "show")]
+	#[structopt(name = "show", version = garage_version())]
 	Show,
 
 	/// Apply staged changes to cluster layout
-	#[structopt(name = "apply")]
+	#[structopt(name = "apply", version = garage_version())]
 	Apply(ApplyLayoutOpt),
 
 	/// Revert staged changes to cluster layout
-	#[structopt(name = "revert")]
+	#[structopt(name = "revert", version = garage_version())]
 	Revert(RevertLayoutOpt),
 }
 
@@ -150,43 +151,43 @@ pub struct RevertLayoutOpt {
 #[derive(Serialize, Deserialize, StructOpt, Debug)]
 pub enum BucketOperation {
 	/// List buckets
-	#[structopt(name = "list")]
+	#[structopt(name = "list", version = garage_version())]
 	List,
 
 	/// Get bucket info
-	#[structopt(name = "info")]
+	#[structopt(name = "info", version = garage_version())]
 	Info(BucketOpt),
 
 	/// Create bucket
-	#[structopt(name = "create")]
+	#[structopt(name = "create", version = garage_version())]
 	Create(BucketOpt),
 
 	/// Delete bucket
-	#[structopt(name = "delete")]
+	#[structopt(name = "delete", version = garage_version())]
 	Delete(DeleteBucketOpt),
 
 	/// Alias bucket under new name
-	#[structopt(name = "alias")]
+	#[structopt(name = "alias", version = garage_version())]
 	Alias(AliasBucketOpt),
 
 	/// Remove bucket alias
-	#[structopt(name = "unalias")]
+	#[structopt(name = "unalias", version = garage_version())]
 	Unalias(UnaliasBucketOpt),
 
 	/// Allow key to read or write to bucket
-	#[structopt(name = "allow")]
+	#[structopt(name = "allow", version = garage_version())]
 	Allow(PermBucketOpt),
 
 	/// Deny key from reading or writing to bucket
-	#[structopt(name = "deny")]
+	#[structopt(name = "deny", version = garage_version())]
 	Deny(PermBucketOpt),
 
 	/// Expose as website or not
-	#[structopt(name = "website")]
+	#[structopt(name = "website", version = garage_version())]
 	Website(WebsiteOpt),
 
 	/// Set the quotas for this bucket
-	#[structopt(name = "set-quotas")]
+	#[structopt(name = "set-quotas", version = garage_version())]
 	SetQuotas(SetQuotasOpt),
 }
 
@@ -292,35 +293,35 @@ pub struct SetQuotasOpt {
 #[derive(Serialize, Deserialize, StructOpt, Debug)]
 pub enum KeyOperation {
 	/// List keys
-	#[structopt(name = "list")]
+	#[structopt(name = "list", version = garage_version())]
 	List,
 
 	/// Get key info
-	#[structopt(name = "info")]
+	#[structopt(name = "info", version = garage_version())]
 	Info(KeyOpt),
 
 	/// Create new key
-	#[structopt(name = "new")]
+	#[structopt(name = "new", version = garage_version())]
 	New(KeyNewOpt),
 
 	/// Rename key
-	#[structopt(name = "rename")]
+	#[structopt(name = "rename", version = garage_version())]
 	Rename(KeyRenameOpt),
 
 	/// Delete key
-	#[structopt(name = "delete")]
+	#[structopt(name = "delete", version = garage_version())]
 	Delete(KeyDeleteOpt),
 
 	/// Set permission flags for key
-	#[structopt(name = "allow")]
+	#[structopt(name = "allow", version = garage_version())]
 	Allow(KeyPermOpt),
 
 	/// Unset permission flags for key
-	#[structopt(name = "deny")]
+	#[structopt(name = "deny", version = garage_version())]
 	Deny(KeyPermOpt),
 
 	/// Import key
-	#[structopt(name = "import")]
+	#[structopt(name = "import", version = garage_version())]
 	Import(KeyImportOpt),
 }
 
@@ -392,7 +393,7 @@ pub struct MigrateOpt {
 #[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)]
 pub enum MigrateWhat {
 	/// Migrate buckets and permissions from v0.5.0
-	#[structopt(name = "buckets050")]
+	#[structopt(name = "buckets050", version = garage_version())]
 	Buckets050,
 }
 
@@ -413,19 +414,19 @@ pub struct RepairOpt {
 #[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)]
 pub enum RepairWhat {
 	/// Only do a full sync of metadata tables
-	#[structopt(name = "tables")]
+	#[structopt(name = "tables", version = garage_version())]
 	Tables,
 	/// Only repair (resync/rebalance) the set of stored blocks
-	#[structopt(name = "blocks")]
+	#[structopt(name = "blocks", version = garage_version())]
 	Blocks,
 	/// Only redo the propagation of object deletions to the version table (slow)
-	#[structopt(name = "versions")]
+	#[structopt(name = "versions", version = garage_version())]
 	Versions,
 	/// Only redo the propagation of version deletions to the block ref table (extremely slow)
-	#[structopt(name = "block_refs")]
+	#[structopt(name = "block_refs", version = garage_version())]
 	BlockRefs,
 	/// Verify integrity of all blocks on disc (extremely slow, i/o intensive)
-	#[structopt(name = "scrub")]
+	#[structopt(name = "scrub", version = garage_version())]
 	Scrub {
 		#[structopt(subcommand)]
 		cmd: ScrubCmd,
@@ -435,19 +436,19 @@ pub enum RepairWhat {
 #[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)]
 pub enum ScrubCmd {
 	/// Start scrub
-	#[structopt(name = "start")]
+	#[structopt(name = "start", version = garage_version())]
 	Start,
 	/// Pause scrub (it will resume automatically after 24 hours)
-	#[structopt(name = "pause")]
+	#[structopt(name = "pause", version = garage_version())]
 	Pause,
 	/// Resume paused scrub
-	#[structopt(name = "resume")]
+	#[structopt(name = "resume", version = garage_version())]
 	Resume,
 	/// Cancel scrub in progress
-	#[structopt(name = "cancel")]
+	#[structopt(name = "cancel", version = garage_version())]
 	Cancel,
 	/// Set tranquility level for in-progress and future scrubs
-	#[structopt(name = "set-tranquility")]
+	#[structopt(name = "set-tranquility", version = garage_version())]
 	SetTranquility {
 		#[structopt()]
 		tranquility: u32,
@@ -468,10 +469,10 @@ pub struct OfflineRepairOpt {
 pub enum OfflineRepairWhat {
 	/// Repair K2V item counters
 	#[cfg(feature = "k2v")]
-	#[structopt(name = "k2v_item_counters")]
+	#[structopt(name = "k2v_item_counters", version = garage_version())]
 	K2VItemCounters,
 	/// Repair object counters
-	#[structopt(name = "object_counters")]
+	#[structopt(name = "object_counters", version = garage_version())]
 	ObjectCounters,
 }
 
@@ -495,11 +496,17 @@ pub struct WorkerOpt {
 #[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)]
 pub enum WorkerCmd {
 	/// List all workers on Garage node
-	#[structopt(name = "list")]
+	#[structopt(name = "list", version = garage_version())]
 	List {
 		#[structopt(flatten)]
 		opt: WorkerListOpt,
 	},
+	/// Set worker parameter
+	#[structopt(name = "set", version = garage_version())]
+	Set {
+		#[structopt(subcommand)]
+		opt: WorkerSetCmd,
+	},
 }
 
 #[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone, Copy)]
@@ -511,3 +518,16 @@ pub struct WorkerListOpt {
 	#[structopt(short = "e", long = "errors")]
 	pub errors: bool,
 }
+
+#[derive(Serialize, Deserialize, StructOpt, Debug, Eq, PartialEq, Clone)]
+pub enum WorkerSetCmd {
+	/// Set tranquility of scrub operations
+	#[structopt(name = "scrub-tranquility", version = garage_version())]
+	ScrubTranquility { tranquility: u32 },
+	/// Set number of concurrent block resync workers
+	#[structopt(name = "resync-n-workers", version = garage_version())]
+	ResyncNWorkers { n_workers: usize },
+	/// Set tranquility of block resync operations
+	#[structopt(name = "resync-tranquility", version = garage_version())]
+	ResyncTranquility { tranquility: u32 },
+}

src/garage/main.rs

@@ -8,8 +8,15 @@ mod admin;
 mod cli;
 mod repair;
 mod server;
+#[cfg(feature = "telemetry-otlp")]
 mod tracing_setup;
 
+#[cfg(not(any(feature = "bundled-libs", feature = "system-libs")))]
+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");
+
 use std::net::SocketAddr;
 use std::path::PathBuf;
 
@@ -29,7 +36,10 @@ use admin::*;
 use cli::*;
 
 #[derive(StructOpt, Debug)]
-#[structopt(name = "garage")]
+#[structopt(
+	name = "garage",
+	about = "S3-compatible object store for self-hosted geo-distributed deployments"
+)]
 struct Opt {
 	/// Host to connect to for admin operations, in the format:
 	/// <public-key>@<ip>:<port>
@@ -58,7 +68,10 @@ async fn main() {
 	if std::env::var("RUST_LOG").is_err() {
 		std::env::set_var("RUST_LOG", "netapp=info,garage=info")
 	}
-	pretty_env_logger::init();
+	tracing_subscriber::fmt()
+		.with_writer(std::io::stderr)
+		.with_env_filter(tracing_subscriber::filter::EnvFilter::from_default_env())
+		.init();
 	sodiumoxide::init().expect("Unable to init sodiumoxide");
 
 	// Abort on panic (same behavior as in Go)
@@ -67,7 +80,40 @@ async fn main() {
 		std::process::abort();
 	}));
 
-	let opt = Opt::from_args();
+	// Initialize version and features info
+	let features = &[
+		#[cfg(feature = "k2v")]
+		"k2v",
+		#[cfg(feature = "sled")]
+		"sled",
+		#[cfg(feature = "lmdb")]
+		"lmdb",
+		#[cfg(feature = "sqlite")]
+		"sqlite",
+		#[cfg(feature = "kubernetes-discovery")]
+		"kubernetes-discovery",
+		#[cfg(feature = "metrics")]
+		"metrics",
+		#[cfg(feature = "telemetry-otlp")]
+		"telemetry-otlp",
+		#[cfg(feature = "bundled-libs")]
+		"bundled-libs",
+		#[cfg(feature = "system-libs")]
+		"system-libs",
+	][..];
+	if let Some(git_version) = option_env!("GIT_VERSION") {
+		garage_util::version::init_version(git_version);
+	}
+	garage_util::version::init_features(features);
+
+	// Parse arguments
+	let version = format!(
+		"{} [features: {}]",
+		garage_util::version::garage_version(),
+		features.join(", ")
+	);
+	let opt = Opt::from_clap(&Opt::clap().version(version.as_str()).get_matches());
+
 	let res = match opt.cmd {
 		Command::Server => server::run_server(opt.config_file).await,
 		Command::OfflineRepair(repair_opt) => {
@@ -116,7 +162,13 @@ async fn cli_command(opt: Opt) -> Result<(), Error> {
 	} else {
 		let node_id = garage_rpc::system::read_node_id(&config.as_ref().unwrap().metadata_dir)
 			.err_context(READ_KEY_ERROR)?;
-		if let Some(a) = config.as_ref().and_then(|c| c.rpc_public_addr) {
+		if let Some(a) = config.as_ref().and_then(|c| c.rpc_public_addr.as_ref()) {
+			use std::net::ToSocketAddrs;
+			let a = a
+				.to_socket_addrs()
+				.ok_or_message("unable to resolve rpc_public_addr specified in config file")?
+				.next()
+				.ok_or_message("unable to resolve rpc_public_addr specified in config file")?;
 			(node_id, a)
 		} else {
 			let default_addr = SocketAddr::new(

src/garage/server.rs

@@ -9,12 +9,13 @@ use garage_util::error::Error;
 use garage_api::admin::api_server::AdminApiServer;
 use garage_api::s3::api_server::S3ApiServer;
 use garage_model::garage::Garage;
-use garage_web::run_web_server;
+use garage_web::WebServer;
 
 #[cfg(feature = "k2v")]
 use garage_api::k2v::api_server::K2VApiServer;
 
 use crate::admin::*;
+#[cfg(feature = "telemetry-otlp")]
 use crate::tracing_setup::*;
 
 async fn wait_from(mut chan: watch::Receiver<bool>) {
@@ -29,6 +30,8 @@ pub async fn run_server(config_file: PathBuf) -> Result<(), Error> {
 	info!("Loading configuration...");
 	let config = read_config(config_file)?;
 
+	// ---- Initialize Garage internals ----
+
 	info!("Initializing background runner...");
 	let watch_cancel = netapp::util::watch_ctrl_c();
 	let (background, await_background_done) = BackgroundRunner::new(16, watch_cancel.clone());
@@ -36,9 +39,14 @@ pub async fn run_server(config_file: PathBuf) -> Result<(), Error> {
 	info!("Initializing Garage main data store...");
 	let garage = Garage::new(config.clone(), background)?;
 
-	info!("Initialize tracing...");
-	if let Some(export_to) = config.admin.trace_sink {
-		init_tracing(&export_to, garage.system.id)?;
+	if config.admin.trace_sink.is_some() {
+		info!("Initialize tracing...");
+
+		#[cfg(feature = "telemetry-otlp")]
+		init_tracing(config.admin.trace_sink.as_ref().unwrap(), garage.system.id)?;
+
+		#[cfg(not(feature = "telemetry-otlp"))]
+		error!("Garage was built without OTLP exporter, admin.trace_sink is ignored.");
 	}
 
 	info!("Initialize Admin API server and metrics collector...");
@@ -50,53 +58,78 @@ pub async fn run_server(config_file: PathBuf) -> Result<(), Error> {
 	info!("Create admin RPC handler...");
 	AdminRpcHandler::new(garage.clone());
 
-	info!("Initializing S3 API server...");
-	let s3_api_server = tokio::spawn(S3ApiServer::run(
-		garage.clone(),
-		wait_from(watch_cancel.clone()),
-	));
+	// ---- Launch public-facing API servers ----
 
-	#[cfg(feature = "k2v")]
-	let k2v_api_server = {
-		info!("Initializing K2V API server...");
-		tokio::spawn(K2VApiServer::run(
-			garage.clone(),
-			wait_from(watch_cancel.clone()),
-		))
-	};
+	let mut servers = vec![];
 
-	info!("Initializing web server...");
-	let web_server = tokio::spawn(run_web_server(
-		garage.clone(),
-		wait_from(watch_cancel.clone()),
-	));
+	if let Some(s3_bind_addr) = &config.s3_api.api_bind_addr {
+		info!("Initializing S3 API server...");
+		servers.push((
+			"S3 API",
+			tokio::spawn(S3ApiServer::run(
+				garage.clone(),
+				*s3_bind_addr,
+				config.s3_api.s3_region.clone(),
+				wait_from(watch_cancel.clone()),
+			)),
+		));
+	}
 
-	info!("Launching Admin API server...");
-	let admin_server = tokio::spawn(admin_server.run(wait_from(watch_cancel.clone())));
+	if config.k2v_api.is_some() {
+		#[cfg(feature = "k2v")]
+		{
+			info!("Initializing K2V API server...");
+			servers.push((
+				"K2V API",
+				tokio::spawn(K2VApiServer::run(
+					garage.clone(),
+					config.k2v_api.as_ref().unwrap().api_bind_addr,
+					config.s3_api.s3_region.clone(),
+					wait_from(watch_cancel.clone()),
+				)),
+			));
+		}
+		#[cfg(not(feature = "k2v"))]
+		error!("K2V is not enabled in this build, cannot start K2V API server");
+	}
+
+	if let Some(web_config) = &config.s3_web {
+		info!("Initializing web server...");
+		servers.push((
+			"Web",
+			tokio::spawn(WebServer::run(
+				garage.clone(),
+				web_config.bind_addr,
+				web_config.root_domain.clone(),
+				wait_from(watch_cancel.clone()),
+			)),
+		));
+	}
+
+	if let Some(admin_bind_addr) = &config.admin.api_bind_addr {
+		info!("Launching Admin API server...");
+		servers.push((
+			"Admin",
+			tokio::spawn(admin_server.run(*admin_bind_addr, wait_from(watch_cancel.clone()))),
+		));
+	}
+
+	#[cfg(not(feature = "metrics"))]
+	if config.admin.metrics_token.is_some() {
+		warn!("This Garage version is built without the metrics feature");
+	}
 
 	// Stuff runs
 
 	// When a cancel signal is sent, stuff stops
-	if let Err(e) = s3_api_server.await? {
-		warn!("S3 API server exited with error: {}", e);
-	} else {
-		info!("S3 API server exited without error.");
-	}
-	#[cfg(feature = "k2v")]
-	if let Err(e) = k2v_api_server.await? {
-		warn!("K2V API server exited with error: {}", e);
-	} else {
-		info!("K2V API server exited without error.");
-	}
-	if let Err(e) = web_server.await? {
-		warn!("Web server exited with error: {}", e);
-	} else {
-		info!("Web server exited without error.");
-	}
-	if let Err(e) = admin_server.await? {
-		warn!("Admin web server exited with error: {}", e);
-	} else {
-		info!("Admin API server exited without error.");
+
+	// Collect stuff
+	for (desc, join_handle) in servers {
+		if let Err(e) = join_handle.await? {
+			error!("{} server exited with error: {}", desc, e);
+		} else {
+			info!("{} server exited without error.", desc);
+		}
 	}
 
 	// Remove RPC handlers for system to break reference cycles

src/garage/tests/lib.rs

@@ -3,5 +3,8 @@ mod common;
 
 mod admin;
 mod bucket;
-mod k2v;
+
 mod s3;
+
+#[cfg(feature = "k2v")]
+mod k2v;

src/garage/tests/s3/objects.rs

@@ -263,4 +263,13 @@ async fn test_deleteobject() {
 		.unwrap();
 
 	assert!(l.contents.is_none());
+
+	// Deleting a non-existing object shouldn't be a problem
+	ctx.client
+		.delete_object()
+		.bucket(&bucket)
+		.key("l-0")
+		.send()
+		.await
+		.unwrap();
 }

src/k2v-client/Cargo.toml

@@ -22,7 +22,7 @@ tokio = "1.17.0"
 
 # cli deps
 clap = { version = "3.1.18", optional = true, features = ["derive", "env"] }
-garage_util = { version = "0.7.0", path = "../util", optional = true }
+garage_util = { version = "0.8.0", path = "../util", optional = true }
 
 
 [features]

src/model/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "garage_model"
-version = "0.7.0"
+version = "0.8.0"
 authors = ["Alex Auvolat <alex@adnab.me>"]
 edition = "2018"
 license = "AGPL-3.0"
@@ -15,11 +15,10 @@ path = "lib.rs"
 
 [dependencies]
 garage_db = { version = "0.8.0", path = "../db" }
-garage_rpc = { version = "0.7.0", path = "../rpc" }
-garage_table = { version = "0.7.0", path = "../table" }
-garage_block = { version = "0.7.0", path = "../block" }
-garage_util = { version = "0.7.0", path = "../util" }
-garage_model_050 = { package = "garage_model", version = "0.5.1" }
+garage_rpc = { version = "0.8.0", path = "../rpc" }
+garage_table = { version = "0.8.0", path = "../table" }
+garage_block = { version = "0.8.0", path = "../block" }
+garage_util = { version = "0.8.0", path = "../util" }
 
 async-trait = "0.1.7"
 arc-swap = "1.0"
@@ -40,9 +39,10 @@ futures-util = "0.3"
 tokio = { version = "1.0", default-features = false, features = ["rt", "rt-multi-thread", "io-util", "net", "time", "macros", "sync", "signal", "fs"] }
 opentelemetry = "0.17"
 
-#netapp = { version = "0.3.0", git = "https://git.deuxfleurs.fr/lx/netapp" }
-#netapp = { version = "0.4", path = "../../../netapp" }
-netapp = "0.4"
+netapp = "0.5"
 
 [features]
 k2v = [ "garage_util/k2v" ]
+lmdb = [ "garage_db/lmdb" ]
+sled = [ "garage_db/sled" ]
+sqlite = [ "garage_db/sqlite" ]

src/model/bucket_alias_table.rs

@@ -7,7 +7,7 @@ use garage_table::*;
 
 /// The bucket alias table holds the names given to buckets
 /// in the global namespace.
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub struct BucketAlias {
 	name: String,
 	pub state: crdt::Lww<Option<Uuid>>,

src/model/bucket_table.rs

@@ -12,7 +12,7 @@ use crate::permission::BucketKeyPerm;
 /// Its parameters are not directly accessible as:
 ///  - It must be possible to merge paramaters, hence the use of a LWW CRDT.
 ///  - A bucket has 2 states, Present or Deleted and parameters make sense only if present.
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub struct Bucket {
 	/// ID of the bucket
 	pub id: Uuid,
@@ -21,7 +21,7 @@ pub struct Bucket {
 }
 
 /// Configuration for a bucket
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub struct BucketParams {
 	/// Bucket's creation date
 	pub creation_date: u64,

src/model/garage.rs

@@ -6,7 +6,7 @@ use garage_db as db;
 
 use garage_util::background::*;
 use garage_util::config::*;
-use garage_util::error::Error;
+use garage_util::error::*;
 
 use garage_rpc::system::System;
 
@@ -76,10 +76,17 @@ pub struct GarageK2V {
 impl Garage {
 	/// Create and run garage
 	pub fn new(config: Config, background: Arc<BackgroundRunner>) -> Result<Arc<Self>, Error> {
+		// Create meta dir and data dir if they don't exist already
+		std::fs::create_dir_all(&config.metadata_dir)
+			.ok_or_message("Unable to create Garage metadata directory")?;
+		std::fs::create_dir_all(&config.data_dir)
+			.ok_or_message("Unable to create Garage data directory")?;
+
 		info!("Opening database...");
 		let mut db_path = config.metadata_dir.clone();
-		std::fs::create_dir_all(&db_path).expect("Unable to create Garage meta data directory");
 		let db = match config.db_engine.as_str() {
+			// ---- Sled DB ----
+			#[cfg(feature = "sled")]
 			"sled" => {
 				db_path.push("db");
 				info!("Opening Sled database at: {}", db_path.display());
@@ -91,6 +98,10 @@ impl Garage {
 					.expect("Unable to open sled DB");
 				db::sled_adapter::SledDb::init(db)
 			}
+			#[cfg(not(feature = "sled"))]
+			"sled" => return Err(Error::Message("sled db not available in this build".into())),
+			// ---- Sqlite DB ----
+			#[cfg(feature = "sqlite")]
 			"sqlite" | "sqlite3" | "rusqlite" => {
 				db_path.push("db.sqlite");
 				info!("Opening Sqlite database at: {}", db_path.display());
@@ -98,23 +109,48 @@ impl Garage {
 					.expect("Unable to open sqlite DB");
 				db::sqlite_adapter::SqliteDb::init(db)
 			}
+			#[cfg(not(feature = "sqlite"))]
+			"sqlite" | "sqlite3" | "rusqlite" => {
+				return Err(Error::Message(
+					"sqlite db not available in this build".into(),
+				))
+			}
+			// ---- LMDB DB ----
+			#[cfg(feature = "lmdb")]
 			"lmdb" | "heed" => {
 				db_path.push("db.lmdb");
 				info!("Opening LMDB database at: {}", db_path.display());
 				std::fs::create_dir_all(&db_path).expect("Unable to create LMDB data directory");
 				let map_size = garage_db::lmdb_adapter::recommended_map_size();
 
-				let db = db::lmdb_adapter::heed::EnvOpenOptions::new()
-					.max_dbs(100)
-					.map_size(map_size)
-					.open(&db_path)
-					.expect("Unable to open LMDB DB");
+				use db::lmdb_adapter::heed;
+				let mut env_builder = heed::EnvOpenOptions::new();
+				env_builder.max_dbs(100);
+				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);
+				}
+				let db = env_builder.open(&db_path).expect("Unable to open LMDB DB");
 				db::lmdb_adapter::LmdbDb::init(db)
 			}
+			#[cfg(not(feature = "lmdb"))]
+			"lmdb" | "heed" => return Err(Error::Message("lmdb db not available in this build".into())),
+			// ---- Unavailable DB engine ----
 			e => {
 				return Err(Error::Message(format!(
-					"Unsupported DB engine: {} (options: sled, sqlite, lmdb)",
-					e
+					"Unsupported DB engine: {} (options: {})",
+					e,
+					vec![
+						#[cfg(feature = "sled")]
+						"sled",
+						#[cfg(feature = "sqlite")]
+						"sqlite",
+						#[cfg(feature = "lmdb")]
+						"lmdb",
+					]
+					.join(", ")
 				)));
 			}
 		};
@@ -133,7 +169,7 @@ impl Garage {
 			background.clone(),
 			replication_mode.replication_factor(),
 			&config,
-		);
+		)?;
 
 		let data_rep_param = TableShardedReplication {
 			system: system.clone(),
@@ -159,7 +195,6 @@ impl Garage {
 			&db,
 			config.data_dir.clone(),
 			config.compression_level,
-			config.block_manager_background_tranquility,
 			data_rep_param,
 			system.clone(),
 		);

src/model/index_counter.rs

@@ -81,7 +81,7 @@ impl<T: CountedItem> CounterEntry<T> {
 }
 
 /// A counter entry in the global table
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub struct CounterValue {
 	pub node_values: BTreeMap<Uuid, (u64, i64)>,
 }

src/model/k2v/causality.rs

@@ -15,7 +15,7 @@ pub fn make_node_id(node_id: Uuid) -> K2VNodeId {
 	u64::from_be_bytes(tmp)
 }
 
-#[derive(PartialEq, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Debug, Serialize, Deserialize)]
 pub struct CausalContext {
 	pub vector_clock: BTreeMap<K2VNodeId, u64>,
 }

src/model/k2v/item_table.rs

@@ -17,7 +17,7 @@ pub const CONFLICTS: &str = "conflicts";
 pub const VALUES: &str = "values";
 pub const BYTES: &str = "bytes";
 
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub struct K2VItem {
 	pub partition: K2VItemPartition,
 	pub sort_key: String,
@@ -25,19 +25,19 @@ pub struct K2VItem {
 	items: BTreeMap<K2VNodeId, DvvsEntry>,
 }
 
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize, Hash, Eq)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize, Hash)]
 pub struct K2VItemPartition {
 	pub bucket_id: Uuid,
 	pub partition_key: String,
 }
 
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 struct DvvsEntry {
 	t_discard: u64,
 	values: Vec<(u64, DvvsValue)>,
 }
 
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub enum DvvsValue {
 	Value(#[serde(with = "serde_bytes")] Vec<u8>),
 	Deleted,

src/model/k2v/rpc.rs

@@ -23,7 +23,6 @@ use garage_rpc::system::System;
 use garage_rpc::*;
 
 use garage_table::replication::{TableReplication, TableShardedReplication};
-use garage_table::table::TABLE_RPC_TIMEOUT;
 use garage_table::{PartitionKey, Table};
 
 use crate::k2v::causality::*;
@@ -117,7 +116,6 @@ impl K2VRpcHandler {
 				}),
 				RequestStrategy::with_priority(PRIO_NORMAL)
 					.with_quorum(1)
-					.with_timeout(TABLE_RPC_TIMEOUT)
 					.interrupt_after_quorum(true),
 			)
 			.await?;
@@ -169,7 +167,6 @@ impl K2VRpcHandler {
 					K2VRpc::InsertManyItems(items),
 					RequestStrategy::with_priority(PRIO_NORMAL)
 						.with_quorum(1)
-						.with_timeout(TABLE_RPC_TIMEOUT)
 						.interrupt_after_quorum(true),
 				)
 				.await?;
@@ -205,22 +202,23 @@ impl K2VRpcHandler {
 			.replication
 			.write_nodes(&poll_key.partition.hash());
 
-		let resps = self
-			.system
-			.rpc
-			.try_call_many(
-				&self.endpoint,
-				&nodes[..],
-				K2VRpc::PollItem {
-					key: poll_key,
-					causal_context,
-					timeout_msec,
-				},
-				RequestStrategy::with_priority(PRIO_NORMAL)
-					.with_quorum(self.item_table.data.replication.read_quorum())
-					.with_timeout(Duration::from_millis(timeout_msec) + TABLE_RPC_TIMEOUT),
-			)
-			.await?;
+		let rpc = self.system.rpc.try_call_many(
+			&self.endpoint,
+			&nodes[..],
+			K2VRpc::PollItem {
+				key: poll_key,
+				causal_context,
+				timeout_msec,
+			},
+			RequestStrategy::with_priority(PRIO_NORMAL)
+				.with_quorum(self.item_table.data.replication.read_quorum())
+				.without_timeout(),
+		);
+		let timeout_duration = Duration::from_millis(timeout_msec) + self.system.rpc.rpc_timeout();
+		let resps = select! {
+			r = rpc => r?,
+			_ = tokio::time::sleep(timeout_duration) => return Ok(None),
+		};
 
 		let mut resp: Option<K2VItem> = None;
 		for v in resps {

src/model/key_table.rs

@@ -6,10 +6,10 @@ use garage_util::data::*;
 
 use crate::permission::BucketKeyPerm;
 
-use garage_model_050::key_table as old;
+use crate::prev::v051::key_table as old;
 
 /// An api key
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub struct Key {
 	/// The id of the key (immutable), used as partition key
 	pub key_id: String,
@@ -19,7 +19,7 @@ pub struct Key {
 }
 
 /// Configuration for a key
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub struct KeyParams {
 	/// The secret_key associated (immutable)
 	pub secret_key: String,

src/model/lib.rs

@@ -1,6 +1,9 @@
 #[macro_use]
 extern crate tracing;
 
+// For migration from previous versions
+pub(crate) mod prev;
+
 pub mod permission;
 
 pub mod index_counter;

src/model/migrate.rs

@@ -5,7 +5,7 @@ use garage_util::data::*;
 use garage_util::error::Error as GarageError;
 use garage_util::time::*;
 
-use garage_model_050::bucket_table as old_bucket;
+use crate::prev::v051::bucket_table as old_bucket;
 
 use crate::bucket_alias_table::*;
 use crate::bucket_table::*;

src/model/prev/mod.rs

@@ -0,0 +1 @@
+pub(crate) mod v051;

src/model/prev/v051/bucket_table.rs

@@ -0,0 +1,63 @@
+use serde::{Deserialize, Serialize};
+
+use garage_table::crdt::Crdt;
+use garage_table::*;
+
+use super::key_table::PermissionSet;
+
+/// A bucket is a collection of objects
+///
+/// Its parameters are not directly accessible as:
+///  - It must be possible to merge paramaters, hence the use of a LWW CRDT.
+///  - A bucket has 2 states, Present or Deleted and parameters make sense only if present.
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+pub struct Bucket {
+	/// Name of the bucket
+	pub name: String,
+	/// State, and configuration if not deleted, of the bucket
+	pub state: crdt::Lww<BucketState>,
+}
+
+/// State of a bucket
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+pub enum BucketState {
+	/// The bucket is deleted
+	Deleted,
+	/// The bucket exists
+	Present(BucketParams),
+}
+
+impl Crdt for BucketState {
+	fn merge(&mut self, o: &Self) {
+		match o {
+			BucketState::Deleted => *self = BucketState::Deleted,
+			BucketState::Present(other_params) => {
+				if let BucketState::Present(params) = self {
+					params.merge(other_params);
+				}
+			}
+		}
+	}
+}
+
+/// Configuration for a bucket
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+pub struct BucketParams {
+	/// Map of key with access to the bucket, and what kind of access they give
+	pub authorized_keys: crdt::LwwMap<String, PermissionSet>,
+	/// Is the bucket served as http
+	pub website: crdt::Lww<bool>,
+}
+
+impl Crdt for BucketParams {
+	fn merge(&mut self, o: &Self) {
+		self.authorized_keys.merge(&o.authorized_keys);
+		self.website.merge(&o.website);
+	}
+}
+
+impl Crdt for Bucket {
+	fn merge(&mut self, other: &Self) {
+		self.state.merge(&other.state);
+	}
+}

src/model/prev/v051/key_table.rs

@@ -0,0 +1,50 @@
+use serde::{Deserialize, Serialize};
+
+use garage_table::crdt::*;
+use garage_table::*;
+
+/// An api key
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+pub struct Key {
+	/// The id of the key (immutable), used as partition key
+	pub key_id: String,
+
+	/// The secret_key associated
+	pub secret_key: String,
+
+	/// Name for the key
+	pub name: crdt::Lww<String>,
+
+	/// Is the key deleted
+	pub deleted: crdt::Bool,
+
+	/// Buckets in which the key is authorized. Empty if `Key` is deleted
+	// CRDT interaction: deleted implies authorized_buckets is empty
+	pub authorized_buckets: crdt::LwwMap<String, PermissionSet>,
+}
+
+/// Permission given to a key in a bucket
+#[derive(PartialOrd, Ord, PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
+pub struct PermissionSet {
+	/// The key can be used to read the bucket
+	pub allow_read: bool,
+	/// The key can be used to write in the bucket
+	pub allow_write: bool,
+}
+
+impl AutoCrdt for PermissionSet {
+	const WARN_IF_DIFFERENT: bool = true;
+}
+
+impl Crdt for Key {
+	fn merge(&mut self, other: &Self) {
+		self.name.merge(&other.name);
+		self.deleted.merge(&other.deleted);
+
+		if self.deleted.get() {
+			self.authorized_buckets.clear();
+		} else {
+			self.authorized_buckets.merge(&other.authorized_buckets);
+		}
+	}
+}

src/model/prev/v051/mod.rs

@@ -0,0 +1,4 @@
+pub(crate) mod bucket_table;
+pub(crate) mod key_table;
+pub(crate) mod object_table;
+pub(crate) mod version_table;

src/model/prev/v051/object_table.rs

@@ -0,0 +1,149 @@
+use serde::{Deserialize, Serialize};
+use std::collections::BTreeMap;
+
+use garage_util::data::*;
+
+use garage_table::crdt::*;
+
+/// 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: String,
+
+	/// 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
+	versions: Vec<ObjectVersion>,
+}
+
+impl Object {
+	/// Get a list of currently stored versions of `Object`
+	pub fn versions(&self) -> &[ObjectVersion] {
+		&self.versions[..]
+	}
+}
+
+/// 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(ObjectVersionHeaders),
+	/// The version is fully received
+	Complete(ObjectVersionData),
+	/// The version uploaded containded errors or the upload was explicitly aborted
+	Aborted,
+}
+
+impl Crdt for ObjectVersionState {
+	fn merge(&mut self, other: &Self) {
+		use ObjectVersionState::*;
+		match other {
+			Aborted => {
+				*self = Aborted;
+			}
+			Complete(b) => match self {
+				Aborted => {}
+				Complete(a) => {
+					a.merge(b);
+				}
+				Uploading(_) => {
+					*self = Complete(b.clone());
+				}
+			},
+			Uploading(_) => {}
+		}
+	}
+}
+
+/// Data stored in object version
+#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
+pub enum ObjectVersionData {
+	/// The object was deleted, this Version is a tombstone to mark it as such
+	DeleteMarker,
+	/// The object is short, it's stored inlined
+	Inline(ObjectVersionMeta, #[serde(with = "serde_bytes")] Vec<u8>),
+	/// The object is not short, Hash of first block is stored here, next segments hashes are
+	/// stored in the version table
+	FirstBlock(ObjectVersionMeta, Hash),
+}
+
+impl AutoCrdt for ObjectVersionData {
+	const WARN_IF_DIFFERENT: bool = true;
+}
+
+/// Metadata about the object version
+#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
+pub struct ObjectVersionMeta {
+	/// Headers to send to the client
+	pub headers: ObjectVersionHeaders,
+	/// Size of the object
+	pub size: u64,
+	/// etag of the object
+	pub etag: String,
+}
+
+/// Additional headers for an object
+#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Debug, Serialize, Deserialize)]
+pub struct ObjectVersionHeaders {
+	/// Content type of the object
+	pub content_type: String,
+	/// Any other http headers to send
+	pub other: BTreeMap<String, String>,
+}
+
+impl ObjectVersion {
+	fn cmp_key(&self) -> (u64, Uuid) {
+		(self.timestamp, self.uuid)
+	}
+
+	/// Is the object version completely received
+	pub fn is_complete(&self) -> bool {
+		matches!(self.state, ObjectVersionState::Complete(_))
+	}
+}
+
+impl Crdt for Object {
+	fn merge(&mut self, other: &Self) {
+		// Merge versions from other into here
+		for other_v in other.versions.iter() {
+			match self
+				.versions
+				.binary_search_by(|v| v.cmp_key().cmp(&other_v.cmp_key()))
+			{
+				Ok(i) => {
+					self.versions[i].state.merge(&other_v.state);
+				}
+				Err(i) => {
+					self.versions.insert(i, other_v.clone());
+				}
+			}
+		}
+
+		// Remove versions which are obsolete, i.e. those that come
+		// before the last version which .is_complete().
+		let last_complete = self
+			.versions
+			.iter()
+			.enumerate()
+			.rev()
+			.find(|(_, v)| v.is_complete())
+			.map(|(vi, _)| vi);
+
+		if let Some(last_vi) = last_complete {
+			self.versions = self.versions.drain(last_vi..).collect::<Vec<_>>();
+		}
+	}
+}

src/model/prev/v051/version_table.rs

@@ -0,0 +1,79 @@
+use serde::{Deserialize, Serialize};
+
+use garage_util::data::*;
+
+use garage_table::crdt::*;
+use garage_table::*;
+
+/// 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>,
+	/// Etag of each part in case of a multipart upload, empty otherwise
+	pub parts_etags: crdt::Map<u64, String>,
+
+	// Back link to bucket+key so that we can figure if
+	// this was deleted later on
+	/// Bucket in which the related object is stored
+	pub bucket: String,
+	/// Key in which the related object is stored
+	pub key: String,
+}
+
+#[derive(PartialEq, Eq, Clone, Copy, Debug, Serialize, Deserialize)]
+pub struct VersionBlockKey {
+	/// Number of the part
+	pub part_number: u64,
+	/// Offset of this sub-segment in its part
+	pub offset: u64,
+}
+
+impl Ord for VersionBlockKey {
+	fn cmp(&self, other: &Self) -> std::cmp::Ordering {
+		self.part_number
+			.cmp(&other.part_number)
+			.then(self.offset.cmp(&other.offset))
+	}
+}
+
+impl PartialOrd for VersionBlockKey {
+	fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
+		Some(self.cmp(other))
+	}
+}
+
+/// Informations about a single block
+#[derive(PartialEq, Eq, Ord, PartialOrd, Clone, Copy, Debug, Serialize, Deserialize)]
+pub struct VersionBlock {
+	/// Blake2 sum of the block
+	pub hash: Hash,
+	/// Size of the block
+	pub size: u64,
+}
+
+impl AutoCrdt for VersionBlock {
+	const WARN_IF_DIFFERENT: bool = true;
+}
+
+impl Crdt for Version {
+	fn merge(&mut self, other: &Self) {
+		self.deleted.merge(&other.deleted);
+
+		if self.deleted.get() {
+			self.blocks.clear();
+			self.parts_etags.clear();
+		} else {
+			self.blocks.merge(&other.blocks);
+			self.parts_etags.merge(&other.parts_etags);
+		}
+	}
+}

src/model/s3/block_ref_table.rs

@@ -10,7 +10,7 @@ use garage_table::*;
 
 use garage_block::manager::*;
 
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub struct BlockRef {
 	/// Hash (blake2 sum) of the block, used as partition key
 	pub block: Hash,

src/model/s3/object_table.rs

@@ -14,14 +14,14 @@ use garage_table::*;
 use crate::index_counter::*;
 use crate::s3::version_table::*;
 
-use garage_model_050::object_table as old;
+use crate::prev::v051::object_table as old;
 
 pub const OBJECTS: &str = "objects";
 pub const UNFINISHED_UPLOADS: &str = "unfinished_uploads";
 pub const BYTES: &str = "bytes";
 
 /// An object
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[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,
@@ -70,7 +70,7 @@ impl Object {
 }
 
 /// Informations about a version of an object
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub struct ObjectVersion {
 	/// Id of the version
 	pub uuid: Uuid,
@@ -81,7 +81,7 @@ pub struct ObjectVersion {
 }
 
 /// State of an object version
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub enum ObjectVersionState {
 	/// The version is being received
 	Uploading(ObjectVersionHeaders),

src/model/s3/version_table.rs

@@ -12,10 +12,10 @@ use garage_table::*;
 
 use crate::s3::block_ref_table::*;
 
-use garage_model_050::version_table as old;
+use crate::prev::v051::version_table as old;
 
 /// A version of an object
-#[derive(PartialEq, Clone, Debug, Serialize, Deserialize)]
+#[derive(PartialEq, Eq, Clone, Debug, Serialize, Deserialize)]
 pub struct Version {
 	/// UUID of the version, used as partition key
 	pub uuid: Uuid,

src/rpc/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "garage_rpc"
-version = "0.7.0"
+version = "0.8.0"
 authors = ["Alex Auvolat <alex@adnab.me>"]
 edition = "2018"
 license = "AGPL-3.0"
@@ -14,12 +14,11 @@ path = "lib.rs"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-garage_util = { version = "0.7.0", path = "../util" }
+garage_util = { version = "0.8.0", path = "../util" }
 
 arc-swap = "1.0"
 bytes = "1.0"
 gethostname = "0.2"
-git-version = "0.3.4"
 hex = "0.4"
 tracing = "0.1.30"
 rand = "0.8"
@@ -46,12 +45,11 @@ tokio = { version = "1.0", default-features = false, features = ["rt", "rt-multi
 tokio-stream = { version = "0.1", features = ["net"] }
 opentelemetry = "0.17"
 
-#netapp = { version = "0.3.0", git = "https://git.deuxfleurs.fr/lx/netapp" }
-#netapp = { version = "0.4", path = "../../../netapp", features = ["telemetry"] }
-netapp = { version = "0.4.4", features = ["telemetry"] }
+netapp = { version = "0.5.2", features = ["telemetry"] }
 
 hyper = { version = "0.14", features = ["client", "http1", "runtime", "tcp"] }
 
 
 [features]
 kubernetes-discovery = [ "kube", "k8s-openapi", "openssl", "schemars" ]
+system-libs = [ "sodiumoxide/use-pkg-config" ]

src/rpc/kubernetes.rs

@@ -56,7 +56,7 @@ pub async fn get_kubernetes_nodes(
 	let mut ret = Vec::with_capacity(nodes.items.len());
 
 	for node in nodes {
-		println!("Found Pod: {:?}", node.metadata.name);
+		info!("Found Pod: {:?}", node.metadata.name);
 
 		let pubkey = &node
 			.metadata

src/rpc/metrics.rs

@@ -1,31 +1,18 @@
-use std::sync::Arc;
-
 use opentelemetry::{global, metrics::*};
-use tokio::sync::Semaphore;
 
 /// TableMetrics reference all counter used for metrics
 pub struct RpcMetrics {
-	pub(crate) _rpc_available_permits: ValueObserver<u64>,
-
 	pub(crate) rpc_counter: Counter<u64>,
 	pub(crate) rpc_timeout_counter: Counter<u64>,
 	pub(crate) rpc_netapp_error_counter: Counter<u64>,
 	pub(crate) rpc_garage_error_counter: Counter<u64>,
 
 	pub(crate) rpc_duration: ValueRecorder<f64>,
-	pub(crate) rpc_queueing_time: ValueRecorder<f64>,
 }
 impl RpcMetrics {
-	pub fn new(sem: Arc<Semaphore>) -> Self {
+	pub fn new() -> Self {
 		let meter = global::meter("garage_rpc");
 		RpcMetrics {
-			_rpc_available_permits: meter
-				.u64_value_observer("rpc.available_permits", move |observer| {
-					observer.observe(sem.available_permits() as u64, &[])
-				})
-				.with_description("Number of available RPC permits")
-				.init(),
-
 			rpc_counter: meter
 				.u64_counter("rpc.request_counter")
 				.with_description("Number of RPC requests emitted")
@@ -46,10 +33,6 @@ impl RpcMetrics {
 				.f64_value_recorder("rpc.duration")
 				.with_description("Duration of RPCs")
 				.init(),
-			rpc_queueing_time: meter
-				.f64_value_recorder("rpc.queueing_time")
-				.with_description("Time RPC requests were queued for before being sent")
-				.init(),
 		}
 	}
 }

src/rpc/rpc_helper.rs

@@ -7,7 +7,7 @@ use futures::stream::futures_unordered::FuturesUnordered;
 use futures::stream::StreamExt;
 use futures_util::future::FutureExt;
 use tokio::select;
-use tokio::sync::{watch, Semaphore};
+use tokio::sync::watch;
 
 use opentelemetry::KeyValue;
 use opentelemetry::{
@@ -15,10 +15,14 @@ use opentelemetry::{
 	Context,
 };
 
-pub use netapp::endpoint::{Endpoint, EndpointHandler, Message as Rpc};
+pub use netapp::endpoint::{Endpoint, EndpointHandler, StreamingEndpointHandler};
+use netapp::message::IntoReq;
+pub use netapp::message::{
+	Message as Rpc, OrderTag, Req, RequestPriority, Resp, PRIO_BACKGROUND, PRIO_HIGH, PRIO_NORMAL,
+	PRIO_SECONDARY,
+};
 use netapp::peering::fullmesh::FullMeshPeeringStrategy;
-pub use netapp::proto::*;
-pub use netapp::{NetApp, NodeID};
+pub use netapp::{self, NetApp, NodeID};
 
 use garage_util::background::BackgroundRunner;
 use garage_util::data::*;
@@ -28,34 +32,37 @@ use garage_util::metrics::RecordDuration;
 use crate::metrics::RpcMetrics;
 use crate::ring::Ring;
 
-const DEFAULT_TIMEOUT: Duration = Duration::from_secs(10);
-
-// Try to never have more than 200MB of outgoing requests
-// buffered at the same time. Other requests are queued until
-// space is freed.
-const REQUEST_BUFFER_SIZE: usize = 200 * 1024 * 1024;
+// Default RPC timeout = 5 minutes
+const DEFAULT_TIMEOUT: Duration = Duration::from_secs(300);
 
 /// Strategy to apply when making RPC
 #[derive(Copy, Clone)]
 pub struct RequestStrategy {
-	/// Max time to wait for reponse
-	pub rs_timeout: Duration,
 	/// Min number of response to consider the request successful
 	pub rs_quorum: Option<usize>,
 	/// Should requests be dropped after enough response are received
 	pub rs_interrupt_after_quorum: bool,
 	/// Request priority
 	pub rs_priority: RequestPriority,
+	/// Custom timeout for this request
+	rs_timeout: Timeout,
+}
+
+#[derive(Copy, Clone)]
+enum Timeout {
+	None,
+	Default,
+	Custom(Duration),
 }
 
 impl RequestStrategy {
 	/// Create a RequestStrategy with default timeout and not interrupting when quorum reached
 	pub fn with_priority(prio: RequestPriority) -> Self {
 		RequestStrategy {
-			rs_timeout: DEFAULT_TIMEOUT,
 			rs_quorum: None,
 			rs_interrupt_after_quorum: false,
 			rs_priority: prio,
+			rs_timeout: Timeout::Default,
 		}
 	}
 	/// Set quorum to be reached for request
@@ -63,17 +70,22 @@ impl RequestStrategy {
 		self.rs_quorum = Some(quorum);
 		self
 	}
-	/// Set timeout of the strategy
-	pub fn with_timeout(mut self, timeout: Duration) -> Self {
-		self.rs_timeout = timeout;
-		self
-	}
 	/// Set if requests can be dropped after quorum has been reached
 	/// In general true for read requests, and false for write
 	pub fn interrupt_after_quorum(mut self, interrupt: bool) -> Self {
 		self.rs_interrupt_after_quorum = interrupt;
 		self
 	}
+	/// Deactivate timeout for this request
+	pub fn without_timeout(mut self) -> Self {
+		self.rs_timeout = Timeout::None;
+		self
+	}
+	/// Set custom timeout for this request
+	pub fn with_custom_timeout(mut self, timeout: Duration) -> Self {
+		self.rs_timeout = Timeout::Custom(timeout);
+		self
+	}
 }
 
 #[derive(Clone)]
@@ -84,8 +96,8 @@ struct RpcHelperInner {
 	fullmesh: Arc<FullMeshPeeringStrategy>,
 	background: Arc<BackgroundRunner>,
 	ring: watch::Receiver<Arc<Ring>>,
-	request_buffer_semaphore: Arc<Semaphore>,
 	metrics: RpcMetrics,
+	rpc_timeout: Duration,
 }
 
 impl RpcHelper {
@@ -94,45 +106,35 @@ impl RpcHelper {
 		fullmesh: Arc<FullMeshPeeringStrategy>,
 		background: Arc<BackgroundRunner>,
 		ring: watch::Receiver<Arc<Ring>>,
+		rpc_timeout: Option<Duration>,
 	) -> Self {
-		let sem = Arc::new(Semaphore::new(REQUEST_BUFFER_SIZE));
-
-		let metrics = RpcMetrics::new(sem.clone());
+		let metrics = RpcMetrics::new();
 
 		Self(Arc::new(RpcHelperInner {
 			our_node_id,
 			fullmesh,
 			background,
 			ring,
-			request_buffer_semaphore: sem,
 			metrics,
+			rpc_timeout: rpc_timeout.unwrap_or(DEFAULT_TIMEOUT),
 		}))
 	}
 
-	pub async fn call<M, H, S>(
-		&self,
-		endpoint: &Endpoint<M, H>,
-		to: Uuid,
-		msg: M,
-		strat: RequestStrategy,
-	) -> Result<S, Error>
-	where
-		M: Rpc<Response = Result<S, Error>>,
-		H: EndpointHandler<M>,
-	{
-		self.call_arc(endpoint, to, Arc::new(msg), strat).await
+	pub fn rpc_timeout(&self) -> Duration {
+		self.0.rpc_timeout
 	}
 
-	pub async fn call_arc<M, H, S>(
+	pub async fn call<M, N, H, S>(
 		&self,
 		endpoint: &Endpoint<M, H>,
 		to: Uuid,
-		msg: Arc<M>,
+		msg: N,
 		strat: RequestStrategy,
 	) -> Result<S, Error>
 	where
 		M: Rpc<Response = Result<S, Error>>,
-		H: EndpointHandler<M>,
+		N: IntoReq<M> + Send,
+		H: StreamingEndpointHandler<M>,
 	{
 		let metric_tags = [
 			KeyValue::new("rpc_endpoint", endpoint.path().to_string()),
@@ -140,29 +142,27 @@ impl RpcHelper {
 			KeyValue::new("to", format!("{:?}", to)),
 		];
 
-		let msg_size = rmp_to_vec_all_named(&msg)?.len() as u32;
-		let permit = self
-			.0
-			.request_buffer_semaphore
-			.acquire_many(msg_size)
-			.record_duration(&self.0.metrics.rpc_queueing_time, &metric_tags)
-			.await?;
-
 		self.0.metrics.rpc_counter.add(1, &metric_tags);
 
 		let node_id = to.into();
 		let rpc_call = endpoint
-			.call(&node_id, msg, strat.rs_priority)
+			.call_streaming(&node_id, msg, strat.rs_priority)
 			.record_duration(&self.0.metrics.rpc_duration, &metric_tags);
 
+		let timeout = async {
+			match strat.rs_timeout {
+				Timeout::None => futures::future::pending().await,
+				Timeout::Default => tokio::time::sleep(self.0.rpc_timeout).await,
+				Timeout::Custom(t) => tokio::time::sleep(t).await,
+			}
+		};
+
 		select! {
 			res = rpc_call => {
-				drop(permit);
-
 				if res.is_err() {
 					self.0.metrics.rpc_netapp_error_counter.add(1, &metric_tags);
 				}
-				let res = res?;
+				let res = res?.into_msg();
 
 				if res.is_err() {
 					self.0.metrics.rpc_garage_error_counter.add(1, &metric_tags);
@@ -170,46 +170,49 @@ impl RpcHelper {
 
 				Ok(res?)
 			}
-			_ = tokio::time::sleep(strat.rs_timeout) => {
-				drop(permit);
+			() = timeout => {
 				self.0.metrics.rpc_timeout_counter.add(1, &metric_tags);
 				Err(Error::Timeout)
 			}
 		}
 	}
 
-	pub async fn call_many<M, H, S>(
+	pub async fn call_many<M, N, H, S>(
 		&self,
 		endpoint: &Endpoint<M, H>,
 		to: &[Uuid],
-		msg: M,
+		msg: N,
 		strat: RequestStrategy,
-	) -> Vec<(Uuid, Result<S, Error>)>
+	) -> Result<Vec<(Uuid, Result<S, Error>)>, Error>
 	where
 		M: Rpc<Response = Result<S, Error>>,
-		H: EndpointHandler<M>,
+		N: IntoReq<M>,
+		H: StreamingEndpointHandler<M>,
 	{
-		let msg = Arc::new(msg);
+		let msg = msg.into_req().map_err(netapp::error::Error::from)?;
+
 		let resps = join_all(
 			to.iter()
-				.map(|to| self.call_arc(endpoint, *to, msg.clone(), strat)),
+				.map(|to| self.call(endpoint, *to, msg.clone(), strat)),
 		)
 		.await;
-		to.iter()
+		Ok(to
+			.iter()
 			.cloned()
 			.zip(resps.into_iter())
-			.collect::<Vec<_>>()
+			.collect::<Vec<_>>())
 	}
 
-	pub async fn broadcast<M, H, S>(
+	pub async fn broadcast<M, N, H, S>(
 		&self,
 		endpoint: &Endpoint<M, H>,
-		msg: M,
+		msg: N,
 		strat: RequestStrategy,
-	) -> Vec<(Uuid, Result<S, Error>)>
+	) -> Result<Vec<(Uuid, Result<S, Error>)>, Error>
 	where
 		M: Rpc<Response = Result<S, Error>>,
-		H: EndpointHandler<M>,
+		N: IntoReq<M>,
+		H: StreamingEndpointHandler<M>,
 	{
 		let to = self
 			.0
@@ -223,16 +226,17 @@ impl RpcHelper {
 
 	/// Make a RPC call to multiple servers, returning either a Vec of responses,
 	/// or an error if quorum could not be reached due to too many errors
-	pub async fn try_call_many<M, H, S>(
+	pub async fn try_call_many<M, N, H, S>(
 		&self,
 		endpoint: &Arc<Endpoint<M, H>>,
 		to: &[Uuid],
-		msg: M,
+		msg: N,
 		strategy: RequestStrategy,
 	) -> Result<Vec<S>, Error>
 	where
 		M: Rpc<Response = Result<S, Error>> + 'static,
-		H: EndpointHandler<M> + 'static,
+		N: IntoReq<M>,
+		H: StreamingEndpointHandler<M> + 'static,
 		S: Send + 'static,
 	{
 		let quorum = strategy.rs_quorum.unwrap_or(to.len());
@@ -262,20 +266,21 @@ impl RpcHelper {
 			.await
 	}
 
-	async fn try_call_many_internal<M, H, S>(
+	async fn try_call_many_internal<M, N, H, S>(
 		&self,
 		endpoint: &Arc<Endpoint<M, H>>,
 		to: &[Uuid],
-		msg: M,
+		msg: N,
 		strategy: RequestStrategy,
 		quorum: usize,
 	) -> Result<Vec<S>, Error>
 	where
 		M: Rpc<Response = Result<S, Error>> + 'static,
-		H: EndpointHandler<M> + 'static,
+		N: IntoReq<M>,
+		H: StreamingEndpointHandler<M> + 'static,
 		S: Send + 'static,
 	{
-		let msg = Arc::new(msg);
+		let msg = msg.into_req().map_err(netapp::error::Error::from)?;
 
 		// Build future for each request
 		// They are not started now: they are added below in a FuturesUnordered
@@ -285,7 +290,7 @@ impl RpcHelper {
 			let msg = msg.clone();
 			let endpoint2 = endpoint.clone();
 			(to, async move {
-				self2.call_arc(&endpoint2, to, msg, strategy).await
+				self2.call(&endpoint2, to, msg, strategy).await
 			})
 		});
 
@@ -299,47 +304,19 @@ impl RpcHelper {
 			// to reach a quorum, priorizing nodes with the lowest latency.
 			// When there are errors, we start new requests to compensate.
 
-			// Retrieve some status variables that we will use to sort requests
-			let peer_list = self.0.fullmesh.get_peer_list();
-			let ring: Arc<Ring> = self.0.ring.borrow().clone();
-			let our_zone = match ring.layout.node_role(&self.0.our_node_id) {
-				Some(pc) => &pc.zone,
-				None => "",
-			};
-
-			// Augment requests with some information used to sort them.
-			// The tuples are as follows:
-			//         (is another node?, is another zone?, latency, node ID, request future)
-			// We store all of these tuples in a vec that we can sort.
-			// By sorting this vec, we priorize ourself, then nodes in the same zone,
-			// and within a same zone we priorize nodes with the lowest latency.
-			let mut requests = requests
-				.map(|(to, fut)| {
-					let peer_zone = match ring.layout.node_role(&to) {
-						Some(pc) => &pc.zone,
-						None => "",
-					};
-					let peer_avg_ping = peer_list
-						.iter()
-						.find(|x| x.id.as_ref() == to.as_slice())
-						.and_then(|pi| pi.avg_ping)
-						.unwrap_or_else(|| Duration::from_secs(1));
-					(
-						to != self.0.our_node_id,
-						peer_zone != our_zone,
-						peer_avg_ping,
-						to,
-						fut,
-					)
-				})
+			// Reorder requests to priorize closeness / low latency
+			let request_order = self.request_order(to);
+			let mut ord_requests = vec![(); request_order.len()]
+				.into_iter()
+				.map(|_| None)
 				.collect::<Vec<_>>();
-
-			// Sort requests by (priorize ourself, priorize same zone, priorize low latency)
-			requests
-				.sort_by_key(|(diffnode, diffzone, ping, _to, _fut)| (*diffnode, *diffzone, *ping));
+			for (to, fut) in requests {
+				let i = request_order.iter().position(|x| *x == to).unwrap();
+				ord_requests[i] = Some((to, fut));
+			}
 
 			// Make an iterator to take requests in their sorted order
-			let mut requests = requests.into_iter();
+			let mut requests = ord_requests.into_iter().map(Option::unwrap);
 
 			// resp_stream will contain all of the requests that are currently in flight.
 			// (for the moment none, they will be added in the loop below)
@@ -350,7 +327,7 @@ impl RpcHelper {
 				// If the current set of requests that are running is not enough to possibly
 				// reach quorum, start some new requests.
 				while successes.len() + resp_stream.len() < quorum {
-					if let Some((_, _, _, req_to, fut)) = requests.next() {
+					if let Some((req_to, fut)) = requests.next() {
 						let tracer = opentelemetry::global::tracer("garage");
 						let span = tracer.start(format!("RPC to {:?}", req_to));
 						resp_stream.push(tokio::spawn(
@@ -420,4 +397,49 @@ impl RpcHelper {
 			Err(Error::Quorum(quorum, successes.len(), to.len(), errors))
 		}
 	}
+
+	pub fn request_order(&self, nodes: &[Uuid]) -> Vec<Uuid> {
+		// Retrieve some status variables that we will use to sort requests
+		let peer_list = self.0.fullmesh.get_peer_list();
+		let ring: Arc<Ring> = self.0.ring.borrow().clone();
+		let our_zone = match ring.layout.node_role(&self.0.our_node_id) {
+			Some(pc) => &pc.zone,
+			None => "",
+		};
+
+		// Augment requests with some information used to sort them.
+		// The tuples are as follows:
+		//         (is another node?, is another zone?, latency, node ID, request future)
+		// We store all of these tuples in a vec that we can sort.
+		// By sorting this vec, we priorize ourself, then nodes in the same zone,
+		// and within a same zone we priorize nodes with the lowest latency.
+		let mut nodes = nodes
+			.iter()
+			.map(|to| {
+				let peer_zone = match ring.layout.node_role(to) {
+					Some(pc) => &pc.zone,
+					None => "",
+				};
+				let peer_avg_ping = peer_list
+					.iter()
+					.find(|x| x.id.as_ref() == to.as_slice())
+					.and_then(|pi| pi.avg_ping)
+					.unwrap_or_else(|| Duration::from_secs(10));
+				(
+					*to != self.0.our_node_id,
+					peer_zone != our_zone,
+					peer_avg_ping,
+					*to,
+				)
+			})
+			.collect::<Vec<_>>();
+
+		// Sort requests by (priorize ourself, priorize same zone, priorize low latency)
+		nodes.sort_by_key(|(diffnode, diffzone, ping, _to)| (*diffnode, *diffzone, *ping));
+
+		nodes
+			.into_iter()
+			.map(|(_, _, _, to)| to)
+			.collect::<Vec<_>>()
+	}
 }

src/rpc/system.rs

@@ -16,9 +16,9 @@ use tokio::sync::watch;
 use tokio::sync::Mutex;
 
 use netapp::endpoint::{Endpoint, EndpointHandler};
+use netapp::message::*;
 use netapp::peering::fullmesh::FullMeshPeeringStrategy;
-use netapp::proto::*;
-use netapp::util::parse_and_resolve_peer_addr;
+use netapp::util::parse_and_resolve_peer_addr_async;
 use netapp::{NetApp, NetworkKey, NodeID, NodeKey};
 
 use garage_util::background::BackgroundRunner;
@@ -37,10 +37,11 @@ use crate::rpc_helper::*;
 
 const DISCOVERY_INTERVAL: Duration = Duration::from_secs(60);
 const STATUS_EXCHANGE_INTERVAL: Duration = Duration::from_secs(10);
-const PING_TIMEOUT: Duration = Duration::from_secs(2);
 
-/// Version tag used for version check upon Netapp connection
-pub const GARAGE_VERSION_TAG: u64 = 0x6761726167650007; // garage 0x0007
+/// Version tag used for version check upon Netapp connection.
+/// Cluster nodes with different version tags are deemed
+/// incompatible and will refuse to connect.
+pub const GARAGE_VERSION_TAG: u64 = 0x6761726167650008; // garage 0x0008
 
 /// RPC endpoint used for calls related to membership
 pub const SYSTEM_RPC_PATH: &str = "garage_rpc/membership.rs/SystemRpc";
@@ -90,7 +91,7 @@ pub struct System {
 
 	rpc_listen_addr: SocketAddr,
 	rpc_public_addr: Option<SocketAddr>,
-	bootstrap_peers: Vec<(NodeID, SocketAddr)>,
+	bootstrap_peers: Vec<String>,
 
 	consul_discovery: Option<ConsulDiscoveryParam>,
 	#[cfg(feature = "kubernetes-discovery")]
@@ -196,7 +197,7 @@ impl System {
 		background: Arc<BackgroundRunner>,
 		replication_factor: usize,
 		config: &Config,
-	) -> Arc<Self> {
+	) -> Result<Arc<Self>, Error> {
 		let node_key =
 			gen_node_key(&config.metadata_dir).expect("Unable to read or generate node ID");
 		info!(
@@ -204,11 +205,21 @@ impl System {
 			hex::encode(&node_key.public_key()[..8])
 		);
 
-		let persist_cluster_layout = Persister::new(&config.metadata_dir, "cluster_layout");
+		let persist_cluster_layout: Persister<ClusterLayout> =
+			Persister::new(&config.metadata_dir, "cluster_layout");
 		let persist_peer_list = Persister::new(&config.metadata_dir, "peer_list");
 
 		let cluster_layout = match persist_cluster_layout.load() {
-			Ok(x) => x,
+			Ok(x) => {
+				if x.replication_factor != replication_factor {
+					return Err(Error::Message(format!(
+						"Prevous cluster layout has replication factor {}, which is different than the one specified in the config file ({}). The previous cluster layout can be purged, if you know what you are doing, simply by deleting the `cluster_layout` file in your metadata directory.",
+						x.replication_factor,
+						replication_factor
+					)));
+				}
+				x
+			}
 			Err(e) => {
 				info!(
 					"No valid previous cluster layout stored ({}), starting fresh.",
@@ -230,8 +241,29 @@ impl System {
 		let ring = Ring::new(cluster_layout, replication_factor);
 		let (update_ring, ring) = watch::channel(Arc::new(ring));
 
-		let rpc_public_addr = match config.rpc_public_addr {
-			Some(a) => Some(a),
+		let rpc_public_addr = match &config.rpc_public_addr {
+			Some(a_str) => {
+				use std::net::ToSocketAddrs;
+				match a_str.to_socket_addrs() {
+					Err(e) => {
+						error!(
+							"Cannot resolve rpc_public_addr {} from config file: {}.",
+							a_str, e
+						);
+						None
+					}
+					Ok(a) => {
+						let a = a.collect::<Vec<_>>();
+						if a.is_empty() {
+							error!("rpc_public_addr {} resolve to no known IP address", a_str);
+						}
+						if a.len() > 1 {
+							warn!("Multiple possible resolutions for rpc_public_addr: {:?}. Taking the first one.", a);
+						}
+						a.into_iter().next()
+					}
+				}
+			}
 			None => {
 				let addr =
 					get_default_ip().map(|ip| SocketAddr::new(ip, config.rpc_bind_addr.port()));
@@ -241,13 +273,15 @@ impl System {
 				addr
 			}
 		};
+		if rpc_public_addr.is_none() {
+			warn!("This Garage node does not know its publicly reachable RPC address, this might hamper intra-cluster communication.");
+		}
 
 		let netapp = NetApp::new(GARAGE_VERSION_TAG, network_key, node_key);
-		let fullmesh = FullMeshPeeringStrategy::new(
-			netapp.clone(),
-			config.bootstrap_peers.clone(),
-			rpc_public_addr,
-		);
+		let fullmesh = FullMeshPeeringStrategy::new(netapp.clone(), vec![], rpc_public_addr);
+		if let Some(ping_timeout) = config.rpc_ping_timeout_msec {
+			fullmesh.set_ping_timeout_millis(ping_timeout);
+		}
 
 		let system_endpoint = netapp.endpoint(SYSTEM_RPC_PATH.into());
 
@@ -285,7 +319,13 @@ impl System {
 			node_status: RwLock::new(HashMap::new()),
 			netapp: netapp.clone(),
 			fullmesh: fullmesh.clone(),
-			rpc: RpcHelper::new(netapp.id.into(), fullmesh, background.clone(), ring.clone()),
+			rpc: RpcHelper::new(
+				netapp.id.into(),
+				fullmesh,
+				background.clone(),
+				ring.clone(),
+				config.rpc_timeout_msec.map(Duration::from_millis),
+			),
 			system_endpoint,
 			replication_factor,
 			rpc_listen_addr: config.rpc_bind_addr,
@@ -301,7 +341,7 @@ impl System {
 			metadata_dir: config.metadata_dir.clone(),
 		});
 		sys.system_endpoint.set_handler(sys.clone());
-		sys
+		Ok(sys)
 	}
 
 	/// Perform bootstraping, starting the ping loop
@@ -319,14 +359,6 @@ impl System {
 	// ---- Administrative operations (directly available and
 	//      also available through RPC) ----
 
-	pub fn garage_version(&self) -> &'static str {
-		option_env!("GIT_VERSION").unwrap_or(git_version::git_version!(
-			prefix = "git:",
-			cargo_prefix = "cargo:",
-			fallback = "unknown"
-		))
-	}
-
 	pub fn get_known_nodes(&self) -> Vec<KnownNodeInfo> {
 		let node_status = self.node_status.read().unwrap();
 		let known_nodes = self
@@ -366,12 +398,14 @@ impl System {
 	}
 
 	pub async fn connect(&self, node: &str) -> Result<(), Error> {
-		let (pubkey, addrs) = parse_and_resolve_peer_addr(node).ok_or_else(|| {
-			Error::Message(format!(
-				"Unable to parse or resolve node specification: {}",
-				node
-			))
-		})?;
+		let (pubkey, addrs) = parse_and_resolve_peer_addr_async(node)
+			.await
+			.ok_or_else(|| {
+				Error::Message(format!(
+					"Unable to parse or resolve node specification: {}",
+					node
+				))
+			})?;
 		let mut errors = vec![];
 		for ip in addrs.iter() {
 			match self
@@ -491,7 +525,7 @@ impl System {
 		let local_info = self.local_status.load();
 
 		if local_info.replication_factor < info.replication_factor {
-			error!("Some node have a higher replication factor ({}) than this one ({}). This is not supported and might lead to bugs",
+			error!("Some node have a higher replication factor ({}) than this one ({}). This is not supported and will lead to data corruption. Shutting down for safety.",
 				info.replication_factor,
 				local_info.replication_factor);
 			std::process::exit(1);
@@ -519,6 +553,16 @@ impl System {
 		self: &Arc<Self>,
 		adv: &ClusterLayout,
 	) -> Result<SystemRpc, Error> {
+		if adv.replication_factor != self.replication_factor {
+			let msg = format!(
+				"Received a cluster layout from another node with replication factor {}, which is different from what we have in our configuration ({}). Discarding the cluster layout we received.",
+				adv.replication_factor,
+				self.replication_factor
+			);
+			error!("{}", msg);
+			return Err(Error::Message(msg));
+		}
+
 		let update_ring = self.update_ring.lock().await;
 		let mut layout: ClusterLayout = self.ring.borrow().layout.clone();
 
@@ -544,7 +588,7 @@ impl System {
 						SystemRpc::AdvertiseClusterLayout(layout),
 						RequestStrategy::with_priority(PRIO_HIGH),
 					)
-					.await;
+					.await?;
 				Ok(())
 			});
 			self.background.spawn(self.clone().save_cluster_layout());
@@ -559,11 +603,12 @@ impl System {
 
 			self.update_local_status();
 			let local_status: NodeStatus = self.local_status.load().as_ref().clone();
-			self.rpc
+			let _ = self
+				.rpc
 				.broadcast(
 					&self.system_endpoint,
 					SystemRpc::AdvertiseStatus(local_status),
-					RequestStrategy::with_priority(PRIO_HIGH).with_timeout(PING_TIMEOUT),
+					RequestStrategy::with_priority(PRIO_HIGH),
 				)
 				.await;
 
@@ -589,7 +634,7 @@ impl System {
 			if not_configured || no_peers || bad_peers {
 				info!("Doing a bootstrap/discovery step (not_configured: {}, no_peers: {}, bad_peers: {})", not_configured, no_peers, bad_peers);
 
-				let mut ping_list = self.bootstrap_peers.clone();
+				let mut ping_list = resolve_peers(&self.bootstrap_peers).await;
 
 				// Add peer list from list stored on disk
 				if let Ok(peers) = self.persist_peer_list.load_async().await {
@@ -687,7 +732,7 @@ impl System {
 				&self.system_endpoint,
 				peer,
 				SystemRpc::PullClusterLayout,
-				RequestStrategy::with_priority(PRIO_HIGH).with_timeout(PING_TIMEOUT),
+				RequestStrategy::with_priority(PRIO_HIGH),
 			)
 			.await;
 		if let Ok(SystemRpc::AdvertiseClusterLayout(layout)) = resp {
@@ -720,6 +765,25 @@ fn get_default_ip() -> Option<IpAddr> {
 		.map(|a| a.ip())
 }
 
+async fn resolve_peers(peers: &[String]) -> Vec<(NodeID, SocketAddr)> {
+	let mut ret = vec![];
+
+	for peer in peers.iter() {
+		match parse_and_resolve_peer_addr_async(peer).await {
+			Some((pubkey, addrs)) => {
+				for ip in addrs {
+					ret.push((pubkey, ip));
+				}
+			}
+			None => {
+				warn!("Unable to parse and/or resolve peer hostname {}", peer);
+			}
+		}
+	}
+
+	ret
+}
+
 struct ConsulDiscoveryParam {
 	consul_host: String,
 	service_name: String,

src/table/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "garage_table"
-version = "0.7.0"
+version = "0.8.0"
 authors = ["Alex Auvolat <alex@adnab.me>"]
 edition = "2018"
 license = "AGPL-3.0"
@@ -15,13 +15,14 @@ path = "lib.rs"
 
 [dependencies]
 garage_db = { version = "0.8.0", path = "../db" }
-garage_rpc = { version = "0.7.0", path = "../rpc" }
-garage_util = { version = "0.7.0", path = "../util" }
+garage_rpc = { version = "0.8.0", path = "../rpc" }
+garage_util = { version = "0.8.0", path = "../util" }
 
 opentelemetry = "0.17"
 
 async-trait = "0.1.7"
 bytes = "1.0"
+hex = "0.4"
 hexdump = "0.1"
 tracing = "0.1.30"
 rand = "0.8"

src/table/gc.rs

@@ -25,7 +25,6 @@ use crate::replication::*;
 use crate::schema::*;
 
 const TABLE_GC_BATCH_SIZE: usize = 1024;
-const TABLE_GC_RPC_TIMEOUT: Duration = Duration::from_secs(30);
 
 // GC delay for table entries: 1 day (24 hours)
 // (the delay before the entry is added in the GC todo list
@@ -236,9 +235,7 @@ where
 				&self.endpoint,
 				&nodes[..],
 				GcRpc::Update(updates),
-				RequestStrategy::with_priority(PRIO_BACKGROUND)
-					.with_quorum(nodes.len())
-					.with_timeout(TABLE_GC_RPC_TIMEOUT),
+				RequestStrategy::with_priority(PRIO_BACKGROUND).with_quorum(nodes.len()),
 			)
 			.await
 			.err_context("GC: send tombstones")?;
@@ -259,9 +256,7 @@ where
 				&self.endpoint,
 				&nodes[..],
 				GcRpc::DeleteIfEqualHash(deletes),
-				RequestStrategy::with_priority(PRIO_BACKGROUND)
-					.with_quorum(nodes.len())
-					.with_timeout(TABLE_GC_RPC_TIMEOUT),
+				RequestStrategy::with_priority(PRIO_BACKGROUND).with_quorum(nodes.len()),
 			)
 			.await
 			.err_context("GC: remote delete tombstones")?;

src/table/schema.rs

@@ -60,7 +60,7 @@ pub trait Entry<P: PartitionKey, S: SortKey>:
 }
 
 /// Trait for the schema used in a table
-pub trait TableSchema: Send + Sync {
+pub trait TableSchema: Send + Sync + 'static {
 	/// The name of the table in the database
 	const TABLE_NAME: &'static str;
 

src/table/sync.rs

@@ -24,8 +24,6 @@ use crate::merkle::*;
 use crate::replication::*;
 use crate::*;
 
-const TABLE_SYNC_RPC_TIMEOUT: Duration = Duration::from_secs(30);
-
 // Do anti-entropy every 10 minutes
 const ANTI_ENTROPY_INTERVAL: Duration = Duration::from_secs(10 * 60);
 
@@ -247,9 +245,7 @@ where
 				&self.endpoint,
 				nodes,
 				SyncRpc::Items(values),
-				RequestStrategy::with_priority(PRIO_BACKGROUND)
-					.with_quorum(nodes.len())
-					.with_timeout(TABLE_SYNC_RPC_TIMEOUT),
+				RequestStrategy::with_priority(PRIO_BACKGROUND).with_quorum(nodes.len()),
 			)
 			.await?;
 
@@ -310,8 +306,7 @@ where
 				&self.endpoint,
 				who,
 				SyncRpc::RootCkHash(partition.partition, root_ck_hash),
-				RequestStrategy::with_priority(PRIO_BACKGROUND)
-					.with_timeout(TABLE_SYNC_RPC_TIMEOUT),
+				RequestStrategy::with_priority(PRIO_BACKGROUND),
 			)
 			.await?;
 
@@ -350,11 +345,11 @@ where
 					// Just send that item directly
 					if let Some(val) = self.data.store.get(&ik[..])? {
 						if blake2sum(&val[..]) != ivhash {
-							warn!("({}) Hashes differ between stored value and Merkle tree, key: {:?} (if your server is very busy, don't worry, this happens when the Merkle tree can't be updated fast enough)", F::TABLE_NAME, ik);
+							debug!("({}) Hashes differ between stored value and Merkle tree, key: {} (if your server is very busy, don't worry, this happens when the Merkle tree can't be updated fast enough)", F::TABLE_NAME, hex::encode(ik));
 						}
 						todo_items.push(val.to_vec());
 					} else {
-						warn!("({}) Item from Merkle tree not found in store: {:?} (if your server is very busy, don't worry, this happens when the Merkle tree can't be updated fast enough)", F::TABLE_NAME, ik);
+						debug!("({}) Item from Merkle tree not found in store: {} (if your server is very busy, don't worry, this happens when the Merkle tree can't be updated fast enough)", F::TABLE_NAME, hex::encode(ik));
 					}
 				}
 				MerkleNode::Intermediate(l) => {
@@ -367,8 +362,7 @@ where
 							&self.endpoint,
 							who,
 							SyncRpc::GetNode(key.clone()),
-							RequestStrategy::with_priority(PRIO_BACKGROUND)
-								.with_timeout(TABLE_SYNC_RPC_TIMEOUT),
+							RequestStrategy::with_priority(PRIO_BACKGROUND),
 						)
 						.await?
 					{
@@ -444,8 +438,7 @@ where
 				&self.endpoint,
 				who,
 				SyncRpc::Items(values),
-				RequestStrategy::with_priority(PRIO_BACKGROUND)
-					.with_timeout(TABLE_SYNC_RPC_TIMEOUT),
+				RequestStrategy::with_priority(PRIO_BACKGROUND),
 			)
 			.await?;
 		if let SyncRpc::Ok = rpc_resp {

src/table/table.rs

@@ -1,7 +1,6 @@
 use std::borrow::Borrow;
 use std::collections::{BTreeMap, BTreeSet, HashMap};
 use std::sync::Arc;
-use std::time::Duration;
 
 use async_trait::async_trait;
 use futures::stream::*;
@@ -31,8 +30,6 @@ use crate::schema::*;
 use crate::sync::*;
 use crate::util::*;
 
-pub const TABLE_RPC_TIMEOUT: Duration = Duration::from_secs(10);
-
 pub struct Table<F: TableSchema + 'static, R: TableReplication + 'static> {
 	pub system: Arc<System>,
 	pub data: Arc<TableData<F, R>>,
@@ -113,7 +110,6 @@ where
 	async fn insert_internal(&self, e: &F::E) -> Result<(), Error> {
 		let hash = e.partition_key().hash();
 		let who = self.data.replication.write_nodes(&hash);
-		//eprintln!("insert who: {:?}", who);
 
 		let e_enc = Arc::new(ByteBuf::from(rmp_to_vec_all_named(e)?));
 		let rpc = TableRpc::<F>::Update(vec![e_enc]);
@@ -125,8 +121,7 @@ where
 				&who[..],
 				rpc,
 				RequestStrategy::with_priority(PRIO_NORMAL)
-					.with_quorum(self.data.replication.write_quorum())
-					.with_timeout(TABLE_RPC_TIMEOUT),
+					.with_quorum(self.data.replication.write_quorum()),
 			)
 			.await?;
 
@@ -178,7 +173,7 @@ where
 					&self.endpoint,
 					node,
 					rpc,
-					RequestStrategy::with_priority(PRIO_NORMAL).with_timeout(TABLE_RPC_TIMEOUT),
+					RequestStrategy::with_priority(PRIO_NORMAL),
 				)
 				.await?;
 			Ok::<_, Error>((node, resp))
@@ -235,7 +230,6 @@ where
 				rpc,
 				RequestStrategy::with_priority(PRIO_NORMAL)
 					.with_quorum(self.data.replication.read_quorum())
-					.with_timeout(TABLE_RPC_TIMEOUT)
 					.interrupt_after_quorum(true),
 			)
 			.await?;
@@ -330,7 +324,6 @@ where
 				rpc,
 				RequestStrategy::with_priority(PRIO_NORMAL)
 					.with_quorum(self.data.replication.read_quorum())
-					.with_timeout(TABLE_RPC_TIMEOUT)
 					.interrupt_after_quorum(true),
 			)
 			.await?;
@@ -407,9 +400,7 @@ where
 				&self.endpoint,
 				who,
 				TableRpc::<F>::Update(vec![what_enc]),
-				RequestStrategy::with_priority(PRIO_NORMAL)
-					.with_quorum(who.len())
-					.with_timeout(TABLE_RPC_TIMEOUT),
+				RequestStrategy::with_priority(PRIO_NORMAL).with_quorum(who.len()),
 			)
 			.await?;
 		Ok(())

src/util/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "garage_util"
-version = "0.7.0"
+version = "0.8.0"
 authors = ["Alex Auvolat <alex@adnab.me>"]
 edition = "2018"
 license = "AGPL-3.0"
@@ -16,14 +16,19 @@ path = "lib.rs"
 [dependencies]
 garage_db = { version = "0.8.0", path = "../db" }
 
+arc-swap = "1.0"
 async-trait = "0.1"
 blake2 = "0.9"
+bytes = "1.0"
+digest = "0.10"
 err-derive = "0.3"
+git-version = "0.3.4"
 xxhash-rust = { version = "0.8", default-features = false, features = ["xxh3"] }
 hex = "0.4"
+lazy_static = "1.4"
 tracing = "0.1.30"
 rand = "0.8"
-sha2 = "0.9"
+sha2 = "0.10"
 
 chrono = "0.4"
 rmp-serde = "0.15"
@@ -34,14 +39,13 @@ toml = "0.5"
 futures = "0.3"
 tokio = { version = "1.0", default-features = false, features = ["rt", "rt-multi-thread", "io-util", "net", "time", "macros", "sync", "signal", "fs"] }
 
-#netapp = { version = "0.3.0", git = "https://git.deuxfleurs.fr/lx/netapp" }
-#netapp = { version = "0.4", path = "../../../netapp" }
-netapp = "0.4"
+netapp = "0.5"
 
 http = "0.2"
 hyper = "0.14"
 
 opentelemetry = { version = "0.17", features = [ "rt-tokio", "metrics", "trace" ] }
 
+
 [features]
 k2v = []

src/util/async_hash.rs

@@ -0,0 +1,61 @@
+use bytes::Bytes;
+use digest::Digest;
+
+use tokio::sync::mpsc;
+use tokio::task::JoinHandle;
+
+use crate::data::*;
+
+/// Compute the sha256 of a slice,
+/// spawning on a tokio thread for CPU-intensive processing
+/// The argument has to be an owned Bytes, as it is moved out to a new thread.
+pub async fn async_sha256sum(data: Bytes) -> Hash {
+	tokio::task::spawn_blocking(move || sha256sum(&data))
+		.await
+		.unwrap()
+}
+
+/// Compute the blake2sum of a slice,
+/// spawning on a tokio thread for CPU-intensive processing.
+/// The argument has to be an owned Bytes, as it is moved out to a new thread.
+pub async fn async_blake2sum(data: Bytes) -> Hash {
+	tokio::task::spawn_blocking(move || blake2sum(&data))
+		.await
+		.unwrap()
+}
+
+// ----
+
+pub struct AsyncHasher<D: Digest> {
+	sendblk: mpsc::Sender<Bytes>,
+	task: JoinHandle<digest::Output<D>>,
+}
+
+impl<D: Digest> AsyncHasher<D> {
+	pub fn new() -> Self {
+		let (sendblk, mut recvblk) = mpsc::channel::<Bytes>(1);
+		let task = tokio::task::spawn_blocking(move || {
+			let mut digest = D::new();
+			while let Some(blk) = recvblk.blocking_recv() {
+				digest.update(&blk[..]);
+			}
+			digest.finalize()
+		});
+		Self { sendblk, task }
+	}
+
+	pub async fn update(&self, b: Bytes) {
+		self.sendblk.send(b).await.unwrap();
+	}
+
+	pub async fn finalize(self) -> digest::Output<D> {
+		drop(self.sendblk);
+		self.task.await.unwrap()
+	}
+}
+
+impl<D: Digest> Default for AsyncHasher<D> {
+	fn default() -> Self {
+		Self::new()
+	}
+}

src/util/background/worker.rs

@@ -9,7 +9,6 @@ use futures::StreamExt;
 use serde::{Deserialize, Serialize};
 use tokio::select;
 use tokio::sync::{mpsc, watch};
-use tracing::*;
 
 use crate::background::WorkerInfo;
 use crate::error::Error;

src/util/config.rs

@@ -3,12 +3,8 @@ use std::io::Read;
 use std::net::SocketAddr;
 use std::path::PathBuf;
 
-use serde::de::Error as SerdeError;
 use serde::{de, Deserialize};
 
-use netapp::util::parse_and_resolve_peer_addr;
-use netapp::NodeID;
-
 use crate::error::Error;
 
 /// Represent the whole configuration
@@ -23,10 +19,6 @@ pub struct Config {
 	#[serde(default = "default_block_size")]
 	pub block_size: usize,
 
-	/// Size of data blocks to save to disk
-	#[serde(default = "default_block_manager_background_tranquility")]
-	pub block_manager_background_tranquility: u32,
-
 	/// Replication mode. Supported values:
 	/// - none, 1 -> no replication
 	/// - 2 -> 2-way replication
@@ -47,11 +39,16 @@ pub struct Config {
 	/// Address to bind for RPC
 	pub rpc_bind_addr: SocketAddr,
 	/// Public IP address of this node
-	pub rpc_public_addr: Option<SocketAddr>,
+	pub rpc_public_addr: Option<String>,
+
+	/// Timeout for Netapp's ping messagess
+	pub rpc_ping_timeout_msec: Option<u64>,
+	/// Timeout for Netapp RPC calls
+	pub rpc_timeout_msec: Option<u64>,
 
 	/// Bootstrap peers RPC address
-	#[serde(deserialize_with = "deserialize_vec_addr", default)]
-	pub bootstrap_peers: Vec<(NodeID, SocketAddr)>,
+	#[serde(default)]
+	pub bootstrap_peers: Vec<String>,
 	/// Consul host to connect to to discover more peers
 	pub consul_host: Option<String>,
 	/// Consul service name to use
@@ -81,11 +78,10 @@ pub struct Config {
 	pub s3_api: S3ApiConfig,
 
 	/// Configuration for K2V api
-	#[cfg(feature = "k2v")]
 	pub k2v_api: Option<K2VApiConfig>,
 
 	/// Configuration for serving files as normal web server
-	pub s3_web: WebConfig,
+	pub s3_web: Option<WebConfig>,
 
 	/// Configuration for the admin API endpoint
 	#[serde(default = "Default::default")]
@@ -96,7 +92,7 @@ pub struct Config {
 #[derive(Deserialize, Debug, Clone)]
 pub struct S3ApiConfig {
 	/// Address and port to bind for api serving
-	pub api_bind_addr: SocketAddr,
+	pub api_bind_addr: Option<SocketAddr>,
 	/// S3 region to use
 	pub s3_region: String,
 	/// Suffix to remove from domain name to find bucket. If None,
@@ -105,7 +101,6 @@ pub struct S3ApiConfig {
 }
 
 /// Configuration for K2V api
-#[cfg(feature = "k2v")]
 #[derive(Deserialize, Debug, Clone)]
 pub struct K2VApiConfig {
 	/// Address and port to bind for api serving
@@ -147,9 +142,6 @@ fn default_sled_flush_every_ms() -> u64 {
 fn default_block_size() -> usize {
 	1048576
 }
-fn default_block_manager_background_tranquility() -> u32 {
-	2
-}
 
 /// Read and parse configuration
 pub fn read_config(config_file: PathBuf) -> Result<Config, Error> {
@@ -163,24 +155,6 @@ pub fn read_config(config_file: PathBuf) -> Result<Config, Error> {
 	Ok(toml::from_str(&config)?)
 }
 
-fn deserialize_vec_addr<'de, D>(deserializer: D) -> Result<Vec<(NodeID, SocketAddr)>, D::Error>
-where
-	D: de::Deserializer<'de>,
-{
-	let mut ret = vec![];
-
-	for peer in <Vec<&str>>::deserialize(deserializer)? {
-		let (pubkey, addrs) = parse_and_resolve_peer_addr(peer).ok_or_else(|| {
-			D::Error::custom(format!("Unable to parse or resolve peer: {}", peer))
-		})?;
-		for ip in addrs {
-			ret.push((pubkey, ip));
-		}
-	}
-
-	Ok(ret)
-}
-
 fn default_compression() -> Option<i32> {
 	Some(1)
 }

src/util/crdt/bool.rs

@@ -3,7 +3,7 @@ use serde::{Deserialize, Serialize};
 use crate::crdt::crdt::*;
 
 /// Boolean, where `true` is an absorbing state
-#[derive(Clone, Copy, Debug, Serialize, Deserialize, PartialEq)]
+#[derive(Clone, Copy, Debug, Serialize, Deserialize, PartialEq, Eq)]
 pub struct Bool(bool);
 
 impl Bool {

src/util/crdt/deletable.rs

@@ -3,7 +3,7 @@ use serde::{Deserialize, Serialize};
 use crate::crdt::crdt::*;
 
 /// Deletable object (once deleted, cannot go back)
-#[derive(Clone, Copy, Debug, Serialize, Deserialize, PartialEq)]
+#[derive(Clone, Copy, Debug, Serialize, Deserialize, PartialEq, Eq)]
 pub enum Deletable<T> {
 	Present(T),
 	Deleted,

src/util/crdt/lww.rs

@@ -37,7 +37,7 @@ use crate::crdt::crdt::*;
 ///
 /// This scheme is used by AWS S3 or Soundcloud and often without knowing
 /// in enterprise when reconciliating databases with ad-hoc scripts.
-#[derive(Clone, Debug, Serialize, Deserialize, PartialEq)]
+#[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq)]
 pub struct Lww<T> {
 	ts: u64,
 	v: T,

src/util/crdt/lww_map.rs

@@ -23,7 +23,7 @@ use crate::crdt::crdt::*;
 /// However, note that even if we were using a more efficient data structure such as a `BTreeMap`,
 /// the serialization cost `O(n)` would still have to be paid at each modification, so we are
 /// actually not losing anything here.
-#[derive(Clone, Debug, Serialize, Deserialize, PartialEq)]
+#[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq)]
 pub struct LwwMap<K, V> {
 	vals: Vec<(K, u64, V)>,
 }