Merge branch 'main' into fix/b5-t5-c2-preflight-lock

docs(testing): document CI gate matrix (#6412 )
feat(allocator): replace mimalloc/libmimalloc-sys with rustfs-mimalloc/rustfs-mimalloc-sys (#6404 )
2026-08-23 04:39:04 +00:00 · 2026-08-23 12:29:10 +08:00 · 2026-08-23 12:09:06 +08:00 · 2026-08-23 12:07:25 +08:00 · 2026-08-23 12:07:20 +08:00 · 2026-08-21 18:59:59 +08:00
14 changed files with 370 additions and 122 deletions
@@ -30,7 +30,8 @@ make build-docker BUILD_OS=ubuntu22.04
 - Crate membership: `Cargo.toml` `[workspace].members`
 - Architecture, layering, crate map: [ARCHITECTURE.md](ARCHITECTURE.md)
 - Migration guardrails & readiness contracts: [docs/architecture/](docs/architecture/README.md)
- CI gates: `.github/workflows/ci.yml` (source of truth; never copy its steps into docs)
+- CI workflow steps: `.github/workflows/`; event, timeout, and required-status
+  matrix: [docs/testing/ci-gates.md](docs/testing/ci-gates.md)
 - Test-layer taxonomy, per-layer entry commands, serial/nextest rules, flake
  policy: [docs/testing/README.md](docs/testing/README.md)
 - Tier/ILM transition debugging (xl.meta inspection, versionId tracing):
@@ -70,6 +70,8 @@ make pre-pr

 > For the full test-layer taxonomy (unit / ecstore black-box / e2e / s3s-e2e / S3 compatibility / chaos / fuzz / bench), each layer's entry command, the naming conventions the migration gate depends on, and the serial/nextest rules, see [docs/testing/README.md](docs/testing/README.md).

+> For the event, timeout, required-status, and local reproduction matrix, see [docs/testing/ci-gates.md](docs/testing/ci-gates.md).
+
 ### 🔒 Automated Pre-commit Hooks
 #### What `make pre-commit` and `make pre-pr` actually run

@@ -1858,9 +1858,9 @@ dependencies = [

 [[package]]
 name = "cc"
-version = "1.4.3"
+version = "1.4.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "509591b7bcd67f4ef775afad7662703b4935daaa6ec0e5605cfb1090b32a2b6d"
+checksum = "0ad534f4357a5264cce5019c989cf66a4f0dc4e0d1b1d15f8aacec0ff7360273"
 dependencies = [
 "find-msvc-tools",
 "jobserver",
@@ -2522,12 +2522,6 @@ dependencies = [
 "subtle",
 ]

-[[package]]
-name = "cty"
-version = "0.2.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b365fabc795046672053e29c954733ec3b05e4be654ab130fe8f1f94d7051f35"
-
 [[package]]
 name = "curve25519-dalek"
 version = "4.1.3"
@@ -5988,15 +5982,6 @@ version = "0.2.16"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b6d2cec3eae94f9f509c767b45932f1ada8350c4bdb85af2fcab4a3c14807981"

-[[package]]
-name = "libmimalloc-sys"
-version = "0.1.49"
-source = "git+https://github.com/xonatius/mimalloc_rust.git?rev=6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11#6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11"
-dependencies = [
- "cc",
- "cty",
-]
-
 [[package]]
 name = "libredox"
 version = "0.1.20"
@@ -6397,14 +6382,6 @@ dependencies = [
 "synstructure 0.13.2",
 ]

-[[package]]
-name = "mimalloc"
-version = "0.1.52"
-source = "git+https://github.com/xonatius/mimalloc_rust.git?rev=6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11#6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11"
-dependencies = [
- "libmimalloc-sys",
-]
-
 [[package]]
 name = "mime"
 version = "0.3.17"
@@ -9162,13 +9139,11 @@ dependencies = [
 "insta",
 "jiff",
 "libc",
- "libmimalloc-sys",
 "libsystemd",
 "matchit 0.9.2",
 "md-5 0.11.0",
 "metrics",
 "metrics-util",
- "mimalloc",
 "mime_guess",
 "opentelemetry",
 "opentelemetry_sdk",
@@ -9204,6 +9179,8 @@ dependencies = [
 "rustfs-lock",
 "rustfs-log-analyzer",
 "rustfs-madmin",
+ "rustfs-mimalloc",
+ "rustfs-mimalloc-sys",
 "rustfs-notify",
 "rustfs-object-capacity",
 "rustfs-object-data-cache",
@@ -9875,6 +9852,24 @@ dependencies = [
 "tokio",
 ]

+[[package]]
+name = "rustfs-mimalloc"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a406f4aa07084301d485beec873af6dccc8e3f8762da244743df92038b1db1a6"
+dependencies = [
+ "rustfs-mimalloc-sys",
+]
+
+[[package]]
+name = "rustfs-mimalloc-sys"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c3051b819175f58445d4c369a72f0ab88149f3885ba8bea2aff3be01f53fe7cd"
+dependencies = [
+ "cc",
+]
+
 [[package]]
 name = "rustfs-notify"
 version = "1.0.0-rc.3"
@@ -350,8 +350,8 @@ russh-sftp = "2.4.0"
 dav-server = "0.11.0"

 # Performance Analysis and Memory Profiling
-mimalloc = { version = "0.1.52", git = "https://github.com/xonatius/mimalloc_rust.git", rev = "6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11" }
-libmimalloc-sys = { version = "0.1.49", git = "https://github.com/xonatius/mimalloc_rust.git", rev = "6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11", features = ["extended"] }
+rustfs-mimalloc = { version = "0.5.0" }
+rustfs-mimalloc-sys = { version = "0.5.0" }
 hotpath = { version = "0.23.3", default-features = false }
 # Snapshot testing for output format regression detection
 insta = { version = "1.48" }
@@ -2124,26 +2124,13 @@ impl SetDisks {

            let put_object_size = known_put_object_storage_size(data.size());
            let shard_file_size_raw = erasure.shard_file_size(put_object_size);
-            let is_inline_buffer =
-                storage_class_config.should_inline(shard_file_size_raw, erasure.data_shards, opts.versioned);
+            let is_inline_buffer = storage_class_config.should_inline(shard_file_size_raw, erasure.data_shards, opts.versioned);

            let collect_stage_timing = rustfs_io_metrics::put_stage_metrics_enabled() || issue3031_diag_enabled();
            let shard_file_size = shard_file_size_raw;
            let shard_size = erasure.shard_size();
            let write_path = classify_put_write_path(is_inline_buffer, put_object_size, fi.erasure.block_size);
            let direct_inline_commit = matches!(write_path, SmallWritePath::Inline);
-            {
-                use std::io::Write;
-                let msg = format!(
-                    "INLINE_DEBUG: bucket={} obj={} size={} shard_fs={} ds={} bs={} inline={} direct={} path={} iblock={} ver={}\n",
-                    bucket, object, put_object_size, shard_file_size_raw, erasure.data_shards, fi.erasure.block_size,
-                    is_inline_buffer, direct_inline_commit, write_path.metric_label(), storage_class_config.inline_block(), opts.versioned
-                );
-                if let Ok(mut f) = std::fs::OpenOptions::new().create(true).append(true).open("/tmp/rustfs_inline_debug.log") {
-                    let _ = f.write_all(msg.as_bytes());
-                }
-                let _ = std::io::stderr().write_all(msg.as_bytes());
-            }
            rustfs_io_metrics::record_put_object_path(write_path.metric_label());
            let writer_setup_stage_start = collect_stage_timing.then(Instant::now);
            let (mut writers, errors) = if direct_inline_commit {
@@ -3194,7 +3194,7 @@ impl ECStore {

        // Default return value
        let mut del_objects = vec![DeletedObject::default(); objects.len()];
-        let mut accounting = vec![None; objects.len()];
+        let accounting = vec![None; objects.len()];

        let mut del_errs = Vec::with_capacity(objects.len());
        for _ in 0..objects.len() {
@@ -271,7 +271,7 @@ pub(super) fn resolve_latest_object_info_candidates(
            .filter(|candidate| latest_candidate_mod_time(candidate) == Some(latest_mod_time))
            .collect::<Vec<_>>();

-        latest_candidates.sort_by(|left, right| right.idx.cmp(&left.idx));
+        latest_candidates.sort_by_key(|candidate| std::cmp::Reverse(candidate.idx));

        let Some(winner) = latest_candidates.first() else {
            return Err(Error::ErasureReadQuorum);
@@ -43,9 +43,6 @@ allow-git = [
    # RustFS fork carrying presigned expiry and constant-time authentication fixes.
    # owner: rustfs-maintainers  review: 2026-10
    "https://github.com/rustfs/s3s.git",
-    # MiMalloc fork pinned for hotpath allocation counting support.
-    # owner: houseme  review: 2026-10
-    "https://github.com/xonatius/mimalloc_rust.git",
 ]

 [bans]
@@ -0,0 +1,149 @@
+# CI gate matrix
+
+This file is the source of truth for which validation runs on each event, its
+configured wall-clock budget, and whether it can block a merge. Test taxonomy,
+naming, and nextest serialization rules remain in [README.md](README.md); e2e
+membership and counts remain in
+[e2e-suite-inventory.md](e2e-suite-inventory.md).
+
+The distinction between **required** and **report-only** is load-bearing:
+a failing job blocks a merge only when its exact check name is present in the
+live `main` ruleset. A workflow name, a `merge_group` trigger, or a red PR check
+does not make a job required by itself.
+
+## Required merge checks
+
+The live `main` ruleset (`6436880`) currently requires exactly these contexts:
+
+| Required context | Producer | Validation |
+|---|---|---|
+| `CLA Check` | `.github/workflows/cla.yml` | Contributor agreement |
+| `Quick Checks` | `.github/workflows/ci.yml` | Formatting and repository guard scripts |
+| `Test and Lint` | `.github/workflows/ci.yml` | Clippy, workspace nextest excluding `e2e_test`, doctests, and migration proofs |
+
+For pull requests limited to the paths excluded by the main CI workflow,
+`.github/workflows/ci-docs-only.yml` reports `Quick Checks` and
+`Test and Lint` under the same names. It runs the real quick checks and the
+planning-document guard; it does not claim that Rust compilation or runtime
+tests ran. Despite the workflow name, these paths also include selected deploy,
+workflow, and lock files.
+
+Verify the live rule rather than trusting this snapshot before changing merge
+policy:
+
+```bash
+gh api repos/rustfs/rustfs/rulesets/6436880 \
+  --jq '.rules[] | select(.type == "required_status_checks") | .parameters'
+```
+
+The ruleset currently has `strict_required_status_checks_policy=false`.
+`Continuous Integration` accepts `merge_group` events and runs `e2e-full` for
+them, but `End-to-End Tests (full merge gate)` is not currently a required
+context. Therefore the repository is prepared to test a merge-queue SHA, but
+the workflow alone does not prove that every merge passed that lane.
+
+## Pull request and merge matrix
+
+Budgets below are job `timeout-minutes`, not typical runtimes. “Report-only”
+means the result is visible and actionable but is not in the live required
+context list.
+
+| Event | Validation | Budget | Merge status | Reproduction |
+|---|---|---:|---|---|
+| PR, non-doc change | `Quick Checks` | 10 min | Required | `make pre-commit` (broader local umbrella) |
+| PR, non-doc change | `Test and Lint` | 90 min | Required | `cargo nextest run --profile ci --all --exclude e2e_test` |
+| PR, non-doc change | `Typos` | 10 min | Report-only | `typos` |
+| PR, non-doc change | `ILM Integration (serial)` | 90 min | Report-only | Use the exact command in `.github/workflows/ci.yml` |
+| PR, non-doc change | rio-v2 / swift / sftp test-and-lint variants | 90 min each | Report-only | `cargo nextest run` with the workflow's feature set |
+| PR, non-doc change | `Build RustFS Debug Binary` | 30 min | Report-only; prerequisite for black-box lanes | `cargo build -p rustfs --bins` |
+| PR, non-doc change | `io_uring Integration (real)` | 30 min | Report-only | `cargo test -p rustfs-ecstore --lib uring_ -- --test-threads=1 --nocapture` |
+| PR, non-doc change | `End-to-End Tests` (`e2e-smoke` plus `s3s-e2e`) | 30 min | Report-only | `cargo nextest run --profile e2e-smoke -p e2e_test`; then `./scripts/e2e-run.sh ./target/debug/rustfs <data-dir>` |
+| PR, non-doc change | `S3 Implemented Tests` | 60 min | Report-only | Build `rustfs`, then run `scripts/s3-tests/run.sh` with `DEPLOY_MODE=binary`, `TEST_MODE=single`, and `MAXFAIL=0` |
+| PR, non-doc change | `S3 Lifecycle Behavior Tests` | 30 min | Report-only | Use the accelerated scanner environment in `.github/workflows/ci.yml` with `scripts/s3-tests/run.sh` |
+| PR touching dependency or workflow inputs | Cargo Deny / Workflow Pin Report / Dependency Review | 20 / 5 / 30 min | Report-only | `cargo deny check`; `scripts/security/check_workflow_pins.sh` |
+| PR touching architecture rules or architecture docs | `Architecture Migration Rules` | 10 min | Report-only | `scripts/check_architecture_migration_rules.sh` |
+| PR touching Nix or workspace manifests | `Nix Build & Check` | 60 min | Report-only | `nix flake check` |
+| PR limited to main-CI-excluded paths | companion `Quick Checks` and `Test and Lint` | 10 min each | Required | `git diff --check`; `make doc-paths-check` when documentation paths changed |
+| `merge_group` | Standard CI plus `e2e-full` | 55 min for `e2e-full` | Standard required contexts only; `e2e-full` report-only | `cargo nextest run --profile e2e-full -p e2e_test` |
+| Push to `main` | Standard CI plus `e2e-full` | 55 min for `e2e-full` | Post-merge detection | Same as `merge_group` |
+| PR touching fuzz inputs or harness paths | Build plus five 60-second fuzz smoke targets | 60 min build; 30 min per target | Report-only | `MAX_TOTAL_TIME=60 ./scripts/fuzz/run.sh` |
+| PR touching selected ecstore disk/format paths | `Rename Safety` on Windows | 60 min | Report-only | Run the four `cargo test -p rustfs-ecstore --lib <filter>` commands in `windows-filesystem.yml` on Windows |
+
+The authoritative e2e filters live in `.config/nextest.toml`; extend a profile
+instead of adding a second ad-hoc selector. Before a profile runs,
+`scripts/check_test_wiring.py` compares its exact membership to the committed
+digest so a silent test drop fails closed.
+
+## Scheduled and manual validation
+
+Scheduled lanes are independent fault domains. They do not block a pull
+request, but their workflow-local gate can fail the run and scheduled failures
+are routed to the shared failure-issue action. The scheduled-validation
+watchdog and freshness workflow separately detect incomplete runs and missing
+schedules.
+
+| Cadence (UTC unless noted) | Workflow / validation | Budget | Verdict and artifacts | Reproduction |
+|---|---|---:|---|---|
+| Daily 02:17 | Fuzz: five nightly corpus targets | 60 min build; 60 min per target | Gate; corpus/crash artifacts, scheduled failure alert | `MAX_TOTAL_TIME=<seconds> ./scripts/fuzz/run.sh` |
+| Daily 03:17 | MinIO interop (EC + SSE read parity) | 40 min | Gate; scheduled failure alert | Dispatch `minio-interop.yml` or follow its pinned Docker fixture steps |
+| Daily 04:29 | Replication / cluster-fault / protocol e2e | 45 / 90 / 90 min | Three independent gates; JUnit, membership, and server logs | `cargo nextest run --profile e2e-repl-nightly -p e2e_test`; `--profile e2e-nightly`; `-j 1 --profile e2e-protocols` |
+| Daily 06:31 | Warp performance A/B | 180 min | Regression budget gate; A/B summaries and server logs | `bash scripts/run_hotpath_warp_abba.sh --help` |
+| Daily 00:07 Asia/Shanghai (16:07 UTC previous day) | Nightly GNU build and Vault lanes | 150 / 90 / 60 min | Build, live Vault, and HA failover gates | Use the commands and pinned Vault images in `nightly-gnu.yml` |
+| Daily 03:23 | Security Audit | 20 / 5 min, plus 30 min on PR dependency review | Cargo Deny and workflow-pin gates; scheduled failure alert | `cargo deny check`; `scripts/security/check_workflow_pins.sh` |
+| Daily 23:47 | Scheduled Validation Freshness | 10 min | Fails when a critical schedule was never created or is stale | Dispatch `scheduled-validation-freshness.yml` |
+| Sunday 00:11 | Full `Continuous Integration` matrix | Per-job budgets above | Weekly variant coverage, including dormant rio-v2 binary/e2e lanes | Dispatch `ci.yml` |
+| Sunday 01:13 | Seven-platform build matrix | 150 min per platform | Build/package integrity; scheduled failure alert | Dispatch `build.yml` with an exact platform set |
+| Sunday 02:19 | Ceph s3-tests full sweep: single and real four-node, four shards each | 180 min per shard | Compatibility gate; report, JUnit, exact node IDs, and server logs | `scripts/s3-tests/run.sh` against an existing single or distributed target |
+| Sunday 06:41 | Mint | 120 min | **Report-only by design**; per-suite PASS/FAIL/NA and raw `log.json` | Reproduce the pinned Docker sequence in `mint.yml` or dispatch it |
+| Sunday 07:43 | Workspace line coverage | 120 min | Report-only trend; lcov and JSON retained 90 days | `make coverage` |
+| Monthly, day 1 06:37 | Runner Hygiene | 15 min | Validates runner ephemerality; scheduled failure alert | Dispatch `runner-hygiene.yml` |
+
+Manual `workflow_dispatch` exists for the scheduled workflows above. Manual
+runs are debugging evidence and intentionally do not open scheduled-failure
+issues. A manual performance run may explicitly allow a known regression; that
+override must not be treated as an ordinary passing baseline.
+
+## Release validation
+
+Release validation is post-merge and tag-driven; it does not substitute for a
+pull-request gate.
+
+| Event | Validation | Budget | Result |
+|---|---|---:|---|
+| Push to `main` or weekly schedule | `Build and Release` platform matrix | 150 min per platform | Build artifacts for all selected targets; no release publication on a main push |
+| Valid release or preview tag | `Build and Release` plus asset checks | 150 min per platform | Draft release, checksummed assets, and publish step |
+| Successful non-preview release-tag build | Docker image build and image scan | 60 min build; 30 min scan | Multi-architecture images plus vulnerability report |
+| Successful release-tag build | DEB/RPM packaging | 30 min per architecture | Packages and checksum files uploaded to the release |
+| Successful non-preview release-tag build | Helm template test and package | 30 min build; 30 min publish | Versioned chart and repository index |
+
+Use an exact preview tag for end-to-end release rehearsal. Manual dispatches
+are backfill/debug paths and do not prove the automatic `workflow_run` chain.
+
+## Evidence requirements
+
+A green check is useful only when it proves the intended behavior ran:
+
+- Record the exact commit SHA and run URL.
+- Separate product failure from runner prerequisites, service readiness, and
+  cancellation. Repair the precondition, then rerun the exact workload.
+- Preserve membership manifests, JUnit, raw compatibility logs, seeds, and
+  server logs where the workflow provides them.
+- For a bug fix or a new fault checker, provide sensitivity evidence: the old
+  behavior or an intentional mutation must fail the new oracle, and the fixed
+  behavior must pass it.
+- Never promote a report-only lane to required from one green run. Require at
+  least 14 days and 30 representative pull requests with at least 99% complete
+  execution, then update the ruleset and this table together.
+
+## Change checklist
+
+Update this file in the same pull request when any of these change:
+
+- workflow triggers, job names, timeouts, or nextest profile ownership;
+- required status contexts or strict/merge-queue policy;
+- scheduled cadence, alert routing, artifact contract, or local reproduction;
+- report-only versus gating semantics.
+
+Do not copy per-module test counts here. Update
+[e2e-suite-inventory.md](e2e-suite-inventory.md) and its enforced membership
+digest instead.
@@ -336,13 +336,13 @@ opentelemetry = { workspace = true }
 tracing-opentelemetry = { workspace = true }
 # Data structures
 hashbrown = { workspace = true, features = ["serde", "rayon"] }
-mimalloc = { workspace = true }
+rustfs-mimalloc = { workspace = true }

 [target.'cfg(target_os = "linux")'.dependencies]
 libsystemd.workspace = true

 [target.'cfg(not(target_os = "windows"))'.dependencies]
-libmimalloc-sys.workspace = true
+rustfs-mimalloc-sys.workspace = true

 [dev-dependencies]
 uuid = { workspace = true, features = ["v4", "v5", "fast-rng", "macro-diagnostics"] }
@@ -138,6 +138,11 @@ const SITE_REPL_RESYNC_DEFAULT_PAGE_SIZE: usize = 100;
 const SITE_REPL_RESYNC_MAX_PAGE_SIZE: usize = 1000;
 const SITE_REPLICATION_PEER_REQUEST_TIMEOUT: Duration = Duration::from_secs(10);
 const SITE_REPLICATION_PEER_CONNECT_TIMEOUT: Duration = Duration::from_secs(3);
+/// Bound on waiting for the lifecycle lock (below). 3x the peer request
+/// timeout: outlives one full peer round of a healthy concurrent lifecycle
+/// operation, while converting a holder wedged on unreachable peers into a
+/// retryable 503 for the waiter instead of an unbounded hang.
+const SITE_REPLICATION_LIFECYCLE_LOCK_TIMEOUT: Duration = Duration::from_secs(30);
 const SITE_REPLICATION_PEER_ERROR_DETAIL_LIMIT: usize = 256;
 const SITE_REPLICATION_INITIAL_SYNC_ERROR_LIMIT: usize = 32;
 const MAX_PEER_CA_CERT_PEM_SIZE: usize = 256 * 1024;
@@ -388,9 +393,17 @@ struct SiteReplicationLifecycleGuard {
 }

 impl SiteReplicationLifecycleGuard {
-    async fn acquire() -> Self {
-        Self {
-            _guard: SITE_REPLICATION_LIFECYCLE_LOCK.lock().await,
+    /// Bounded acquire: a holder wedged on unreachable peers (each probe
+    /// costs up to [`SITE_REPLICATION_PEER_REQUEST_TIMEOUT`]) must not hang
+    /// every other lifecycle operation indefinitely, so waiters get a
+    /// retryable 503 after [`SITE_REPLICATION_LIFECYCLE_LOCK_TIMEOUT`].
+    async fn acquire() -> S3Result<Self> {
+        match tokio::time::timeout(SITE_REPLICATION_LIFECYCLE_LOCK_TIMEOUT, SITE_REPLICATION_LIFECYCLE_LOCK.lock()).await {
+            Ok(guard) => Ok(Self { _guard: guard }),
+            Err(_) => Err(S3Error::with_message(
+                S3ErrorCode::ServiceUnavailable,
+                "another site replication lifecycle operation is in progress; retry later".to_string(),
+            )),
        }
    }

@@ -2128,6 +2141,27 @@ async fn remote_add_preflight_info(site: &PeerSite) -> S3Result<SiteReplicationA
    add_preflight_info_from_sr_info(site, info, idp_settings)
 }

+/// Preflight every site in an add request while the lifecycle lock is held.
+/// Probes run concurrently (matching the other peer fan-outs in this file):
+/// k unreachable sites cost roughly one peer request timeout, not k of them.
+/// Results (and the first error, if any) are reported in request order.
+async fn add_preflight_infos(
+    sites: &[PeerSite],
+    current_state: &SiteReplicationState,
+    local_peer: &PeerInfo,
+) -> S3Result<Vec<SiteReplicationAddPreflightInfo>> {
+    futures::future::join_all(sites.iter().map(|site| async move {
+        if same_identity_endpoint(&site.endpoint, &local_peer.endpoint) {
+            local_add_preflight_info(current_state, local_peer, site).await
+        } else {
+            remote_add_preflight_info(site).await
+        }
+    }))
+    .await
+    .into_iter()
+    .collect()
+}
+
 fn validate_add_preflight_topology(infos: &[SiteReplicationAddPreflightInfo], local_peer: &PeerInfo) -> S3Result<()> {
    let mut deployment_ids = HashSet::new();
    let mut local_seen = false;
@@ -9872,7 +9906,7 @@ impl Operation for SiteReplicationAddHandler {
        let cred = validate_site_replication_admin_request(&req, AdminAction::SiteReplicationAddAction).await?;
        reject_site_replicator_on_public_admin(&cred)?;
        let replicate_ilm_expiry = sr_add_replicate_ilm_expiry(&req.uri);
-        let lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await;
+        let lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await?;
        // Everything up to the commit below is preflight: peer probes, IAM
        // work and the join fan-out all talk to the network, so none of it may
        // run inside the state transaction. The snapshot read here is what the
@@ -9887,14 +9921,7 @@ impl Operation for SiteReplicationAddHandler {
        // inject it so the add preflight (which requires the local deployment) succeeds. No-op for `mc`.
        ensure_local_site_present(&mut sites, &local_peer);
        validate_add_sites(&sites, &local_peer)?;
-        let mut preflight_infos = Vec::with_capacity(sites.len());
-        for site in &sites {
-            if same_identity_endpoint(&site.endpoint, &local_peer.endpoint) {
-                preflight_infos.push(local_add_preflight_info(&current_state, &local_peer, site).await?);
-            } else {
-                preflight_infos.push(remote_add_preflight_info(site).await?);
-            }
-        }
+        let preflight_infos = add_preflight_infos(&sites, &current_state, &local_peer).await?;
        validate_add_preflight_topology(&preflight_infos, &local_peer)?;
        let expected_updated_at = current_state.updated_at;
        require_add_peer_tls_capability(&sites, &local_peer).await?;
@@ -10102,7 +10129,7 @@ impl Operation for SiteReplicationRemoveHandler {
    async fn call(&self, req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
        let cred = validate_site_replication_admin_request(&req, AdminAction::SiteReplicationRemoveAction).await?;
        reject_site_replicator_on_public_admin(&cred)?;
-        let _lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await;
+        let _lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await?;
        // The request body is read before the bucket-op guard and the state
        // transaction: a client that stalls mid-body must hold neither the
        // state-object lock nor the write half of the bucket-op RwLock (which
@@ -10312,7 +10339,7 @@ where
    F: FnOnce(SRPeerJoinReq) -> Fut + Send + 'static,
    Fut: std::future::Future<Output = S3Result<()>> + Send + 'static,
 {
-    let _lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await;
+    let _lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await?;
    admit_peer_join_across_nodes(local_endpoint, join_req, defer_sync_state_enable, apply_iam).await
 }

@@ -11142,7 +11169,7 @@ impl Operation for SRPeerRemoveHandler {
    async fn call(&self, req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
        validate_site_replication_admin_request(&req, AdminAction::SiteReplicationRemoveAction).await?;
        let remove_req: SRRemoveReq = read_site_replication_json(req, "", false).await?;
-        let _lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await;
+        let _lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await?;
        let _bucket_op_guard = SITE_REPLICATION_BUCKET_OP_LOCK.write().await;
        let removed_deployment_ids = update_site_replication_state(move |state| {
            if pending_endpoint_refresh(state).is_some() {
@@ -11186,7 +11213,7 @@ impl Operation for SiteReplicationResyncOpHandler {
        let operation = query.get("operation").cloned().unwrap_or_default();
        let resolved_store = object_store_from_req(&req);
        let requested_peer: PeerInfo = read_site_replication_json(req, "", false).await?;
-        let _lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await;
+        let _lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await?;
        let (peer, existing_status) = {
            let state = load_site_replication_state().await?;
            let local_peer = current_local_runtime_peer(&state);
@@ -11478,7 +11505,7 @@ impl Operation for SRRotateServiceAccountHandler {
        // mid-repair and race its own IAM write against the reconciler's
        // stale one. (The removed process mutex used to provide this
        // exclusion as a side effect.)
-        let _lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await;
+        let _lifecycle_guard = SiteReplicationLifecycleGuard::acquire().await?;
        let local_endpoint = site_replication_local_endpoint(&req.uri, &req.headers);
        let rotation_parent = cred.access_key.clone();
        let (pending_rotation, local_peer, previous_access_key) = update_site_replication_state_when_changed(move |state| {
@@ -14019,7 +14046,9 @@ mod tests {
    async fn test_add_bootstrap_scope_only_allows_expected_bucket_setup_until_guard_drops() {
        let token;
        {
-            let lifecycle = SiteReplicationLifecycleGuard::acquire().await;
+            let lifecycle = SiteReplicationLifecycleGuard::acquire()
+                .await
+                .expect("acquire lifecycle guard");
            let guard = SiteReplicationAddInProgressGuard::start(lifecycle, HashSet::from(["legacy-bucket".to_string()]))
                .expect("start site replication add guard");
            token = guard.token.to_string();
@@ -14075,14 +14104,18 @@ mod tests {
    #[tokio::test]
    #[serial]
    async fn test_add_lifecycle_allows_callback_before_remove_writer() {
-        let lifecycle = SiteReplicationLifecycleGuard::acquire().await;
+        let lifecycle = SiteReplicationLifecycleGuard::acquire()
+            .await
+            .expect("acquire lifecycle guard");
        let add_guard =
            SiteReplicationAddInProgressGuard::start(lifecycle, HashSet::new()).expect("start site replication add guard");
        let (started_tx, started_rx) = tokio::sync::oneshot::channel();
        let (entered_tx, mut entered_rx) = tokio::sync::oneshot::channel();
        let remove = tokio::spawn(async move {
            let _ = started_tx.send(());
-            let _lifecycle = SiteReplicationLifecycleGuard::acquire().await;
+            let _lifecycle = SiteReplicationLifecycleGuard::acquire()
+                .await
+                .expect("acquire lifecycle guard");
            let _bucket_op = SITE_REPLICATION_BUCKET_OP_LOCK.write().await;
            let _ = entered_tx.send(());
        });
@@ -14102,6 +14135,111 @@ mod tests {
        entered_rx.await.expect("remove entered lifecycle");
    }

+    /// Deleting either constant (or "simplifying" the client builders to
+    /// inline values) removes the only bound on how long a lifecycle
+    /// operation can be wedged per unreachable peer (#1889 C1 / #1952 C2).
+    #[test]
+    fn test_peer_timeout_constants_bound_unreachable_peer_probes() {
+        assert_eq!(SITE_REPLICATION_PEER_REQUEST_TIMEOUT, Duration::from_secs(10));
+        assert_eq!(SITE_REPLICATION_PEER_CONNECT_TIMEOUT, Duration::from_secs(3));
+        assert!(
+            SITE_REPLICATION_LIFECYCLE_LOCK_TIMEOUT >= SITE_REPLICATION_PEER_REQUEST_TIMEOUT,
+            "a waiter must not give up before the holder's single wedged peer probe can finish"
+        );
+    }
+
+    #[tokio::test(start_paused = true)]
+    #[serial]
+    async fn test_lifecycle_guard_acquire_times_out_with_retryable_503() {
+        let holder = SiteReplicationLifecycleGuard::acquire().await.expect("first acquire");
+        let err =
+            match tokio::time::timeout(SITE_REPLICATION_LIFECYCLE_LOCK_TIMEOUT * 2, SiteReplicationLifecycleGuard::acquire())
+                .await
+                .expect("bounded acquire must not hang while the lock is held")
+            {
+                Ok(_) => panic!("acquire while the lock is held should time out"),
+                Err(err) => err,
+            };
+        assert_eq!(err.code(), &S3ErrorCode::ServiceUnavailable);
+
+        drop(holder);
+        tokio::time::timeout(Duration::from_secs(1), SiteReplicationLifecycleGuard::acquire())
+            .await
+            .expect("acquire after release must not wait")
+            .expect("acquire after release");
+    }
+
+    #[derive(Clone)]
+    struct PreflightFanoutTestState {
+        metainfo_barrier: Arc<tokio::sync::Barrier>,
+    }
+
+    async fn preflight_fanout_test_handler(State(state): State<PreflightFanoutTestState>, uri: Uri) -> (StatusCode, String) {
+        if uri.path().ends_with("/site-replication/metainfo") {
+            state.metainfo_barrier.wait().await;
+        }
+        (StatusCode::OK, "{}".to_string())
+    }
+
+    #[tokio::test]
+    #[serial]
+    async fn test_add_preflight_probes_sites_concurrently() {
+        temp_env::async_with_vars(
+            [(ALLOW_LOOPBACK_REPLICATION_TARGET_ENV, Some("true"))],
+            add_preflight_probes_sites_concurrently_inner(),
+        )
+        .await;
+    }
+
+    async fn add_preflight_probes_sites_concurrently_inner() {
+        const REMOTE_SITES: usize = 3;
+        let listener = match TcpListener::bind("127.0.0.1:0").await {
+            Ok(listener) => listener,
+            Err(err) if err.kind() == std::io::ErrorKind::PermissionDenied => return,
+            Err(err) => panic!("bind preflight test server: {err}"),
+        };
+        let endpoint = format!("http://{}", listener.local_addr().expect("preflight test address"));
+        let state = PreflightFanoutTestState {
+            metainfo_barrier: Arc::new(tokio::sync::Barrier::new(REMOTE_SITES)),
+        };
+        let server = tokio::spawn(async move {
+            axum::serve(listener, Router::new().fallback(any(preflight_fanout_test_handler)).with_state(state))
+                .await
+                .expect("serve preflight test requests");
+        });
+
+        let sites: Vec<PeerSite> = (0..REMOTE_SITES)
+            .map(|index| PeerSite {
+                name: format!("site-{index}"),
+                endpoint: endpoint.clone(),
+                access_key: "test-access".to_string(),
+                secret_key: "test-secret".to_string(),
+                ..Default::default()
+            })
+            .collect();
+        let local_peer = PeerInfo {
+            deployment_id: "local".to_string(),
+            endpoint: "http://192.0.2.1:9000".to_string(),
+            ..Default::default()
+        };
+        let current_state = SiteReplicationState::default();
+
+        // Each site's metainfo request parks on a barrier that only releases
+        // once every site's request has arrived: serial probing never sends
+        // the second request and dies on the peer request timeout, so
+        // finishing well inside that timeout proves the probes overlap —
+        // which is what caps k unreachable sites at one timeout, not k.
+        let infos = tokio::time::timeout(
+            SITE_REPLICATION_PEER_REQUEST_TIMEOUT / 2,
+            add_preflight_infos(&sites, &current_state, &local_peer),
+        )
+        .await
+        .expect("preflight probes must fan out concurrently, not serially")
+        .expect("preflight infos");
+        assert_eq!(infos.len(), REMOTE_SITES);
+        server.abort();
+    }
+
    #[test]
    fn test_merge_add_sites_propagates_replicate_ilm_expiry() {
        let state = merge_add_sites(
@@ -369,14 +369,8 @@ pub fn allocator_reclaim_controller_snapshot(ctx: &CancellationToken) -> Allocat
 }

 #[cfg(not(target_os = "windows"))]
-#[allow(unsafe_code)]
 fn collect_allocator_memory(force: bool) -> Result<(), String> {
-    // SAFETY: `mi_collect` is provided by the active global allocator backend
-    // on this target family. It is explicitly intended to reclaim retained
-    // pages/segments and does not require additional invariants from the caller.
-    unsafe {
-        libmimalloc_sys::mi_collect(force);
-    }
+    rustfs_mimalloc::MiMalloc::collect(force);
    Ok(())
 }

@@ -26,22 +26,22 @@ struct MiMallocAllocator;
 unsafe impl GlobalAlloc for MiMallocAllocator {
    unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
        // SAFETY: the caller upholds GlobalAlloc's contract for layout.
-        unsafe { mimalloc::MiMalloc.alloc(layout) }
+        unsafe { rustfs_mimalloc::MiMalloc.alloc(layout) }
    }

    unsafe fn alloc_zeroed(&self, layout: Layout) -> *mut u8 {
        // SAFETY: the caller upholds GlobalAlloc's contract for layout.
-        unsafe { mimalloc::MiMalloc.alloc_zeroed(layout) }
+        unsafe { rustfs_mimalloc::MiMalloc.alloc_zeroed(layout) }
    }

    unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
        // SAFETY: ptr and layout came from this allocator and are forwarded unchanged.
-        unsafe { mimalloc::MiMalloc.dealloc(ptr, layout) }
+        unsafe { rustfs_mimalloc::MiMalloc.dealloc(ptr, layout) }
    }

    unsafe fn realloc(&self, ptr: *mut u8, layout: Layout, new_size: usize) -> *mut u8 {
        // SAFETY: ptr and layout came from this allocator and are forwarded unchanged.
-        unsafe { mimalloc::MiMalloc.realloc(ptr, layout, new_size) }
+        unsafe { rustfs_mimalloc::MiMalloc.realloc(ptr, layout, new_size) }
    }
 }

@@ -51,7 +51,7 @@ static GLOBAL: hotpath::CountingAllocator<MiMallocAllocator> = hotpath::Counting

 #[cfg(not(all(feature = "hotpath", feature = "hotpath-alloc")))]
 #[global_allocator]
-static GLOBAL: mimalloc::MiMalloc = mimalloc::MiMalloc;
+static GLOBAL: rustfs_mimalloc::MiMalloc = rustfs_mimalloc::MiMalloc;

 fn main() {
    let _hotpath_guard = hotpath::HotpathGuardBuilder::new("main").build();
@@ -71,8 +71,9 @@ mod tests {
        allocation.extend_from_slice(&[7_u8; 64]);

        assert_eq!(allocation.len(), 64);
+        let heap = rustfs_mimalloc::heap::Heap::main();
        // SAFETY: the live Vec pointer is valid to inspect for heap ownership.
-        assert!(unsafe { libmimalloc_sys::mi_is_in_heap_region(allocation.as_ptr().cast()) });
+        assert!(unsafe { heap.contains(allocation.as_ptr()) });
    }

    #[test]
@@ -85,12 +86,13 @@ mod tests {
        let layout = Layout::from_size_align(32, 8).expect("valid test allocation layout");
        let grown_layout = Layout::from_size_align(64, 8).expect("valid grown test allocation layout");
        let allocator = super::MiMallocAllocator;
+        let heap = rustfs_mimalloc::heap::Heap::main();

        // SAFETY: The pointer is checked for null before use and later released
        // through the same allocator with the corresponding layout.
        let ptr = unsafe { allocator.alloc_zeroed(layout) };
        assert!(!ptr.is_null());
-        assert!(unsafe { libmimalloc_sys::mi_is_in_heap_region(ptr.cast()) });
+        assert!(unsafe { heap.contains(ptr) });
        assert!(unsafe { std::slice::from_raw_parts(ptr, 32).iter().all(|byte| *byte == 0) });

        // SAFETY: `ptr` was allocated by `allocator` with `layout`; on failure
@@ -102,7 +104,7 @@ mod tests {
            panic!("mimalloc realloc failed in allocator smoke test");
        }

-        assert!(unsafe { libmimalloc_sys::mi_is_in_heap_region(grown_ptr.cast()) });
+        assert!(unsafe { heap.contains(grown_ptr) });
        // SAFETY: `grown_ptr` was reallocated by `allocator` and is released
        // with the matching grown layout.
        unsafe { allocator.dealloc(grown_ptr, grown_layout) };
@@ -17,10 +17,7 @@ use rustfs_io_metrics::{
    record_cpu_usage, record_memory_usage, record_process_memory_split,
 };
 use serde::Serialize;
-#[cfg(any(test, not(target_os = "windows")))]
 use serde_json::Value;
-#[cfg(not(target_os = "windows"))]
-use std::ffi::CStr;
 use std::path::Path;
 use std::sync::{Arc, Mutex, OnceLock};
 use std::time::Duration;
@@ -231,7 +228,18 @@ fn read_cgroup_memory_snapshot() -> Option<CgroupMemorySnapshot> {
    read_cgroup_v2().or_else(read_cgroup_v1)
 }

-#[cfg(any(test, not(target_os = "windows")))]
+fn read_allocator_memory_snapshot() -> Option<AllocatorMemorySnapshot> {
+    let json = rustfs_mimalloc::MiMalloc::stats_json();
+    if json.is_empty() {
+        return None;
+    }
+    let observation = parse_mimalloc_stats_json(&json)?;
+    Some(AllocatorMemorySnapshot {
+        backend: crate::allocator_reclaim::allocator_backend(),
+        observation,
+    })
+}
+
 fn numeric_json_value(value: &Value) -> Option<u64> {
    match value {
        Value::Number(number) => number
@@ -242,7 +250,6 @@ fn numeric_json_value(value: &Value) -> Option<u64> {
    }
 }

-#[cfg(any(test, not(target_os = "windows")))]
 fn numeric_json_field(value: &Value, field: &str) -> Option<u64> {
    match value {
        Value::Object(fields) => fields
@@ -254,7 +261,6 @@ fn numeric_json_field(value: &Value, field: &str) -> Option<u64> {
    }
 }

-#[cfg(any(test, not(target_os = "windows")))]
 fn mimalloc_stat_field(value: &Value, metric: &str, field: &str) -> Option<u64> {
    match value {
        Value::Object(fields) => {
@@ -271,12 +277,10 @@ fn mimalloc_stat_field(value: &Value, metric: &str, field: &str) -> Option<u64>
    }
 }

-#[cfg(any(test, not(target_os = "windows")))]
 fn mimalloc_stat_current(value: &Value, metric: &str) -> Option<u64> {
    mimalloc_stat_field(value, metric, "current")
 }

-#[cfg(any(test, not(target_os = "windows")))]
 fn mimalloc_stat_sum(value: &Value, metrics: &[&str], field: &str) -> Option<u64> {
    metrics
        .iter()
@@ -285,7 +289,6 @@ fn mimalloc_stat_sum(value: &Value, metrics: &[&str], field: &str) -> Option<u64
        .filter(|value| *value > 0)
 }

-#[cfg(any(test, not(target_os = "windows")))]
 fn parse_mimalloc_stats_json(stats_json: &str) -> Option<AllocatorMemoryObservation> {
    let value = serde_json::from_str::<Value>(stats_json).ok()?;
    let malloc_metrics = ["malloc_normal", "malloc_huge"];
@@ -312,33 +315,6 @@ fn parse_mimalloc_stats_json(stats_json: &str) -> Option<AllocatorMemoryObservat
    }
 }

-#[cfg(not(target_os = "windows"))]
-#[allow(unsafe_code)]
-fn read_allocator_memory_snapshot() -> Option<AllocatorMemorySnapshot> {
-    // SAFETY: `mi_stats_get_json` returns a null-terminated JSON buffer owned by
-    // mimalloc when called with a null input buffer. The mimalloc API requires
-    // freeing that buffer with `mi_free`; parsing finishes before the buffer is freed.
-    let observation = unsafe {
-        let stats_ptr = libmimalloc_sys::mi_stats_get_json(0, std::ptr::null_mut());
-        if stats_ptr.is_null() {
-            return None;
-        }
-
-        let observation = CStr::from_ptr(stats_ptr).to_str().ok().and_then(parse_mimalloc_stats_json);
-        libmimalloc_sys::mi_free(stats_ptr.cast());
-        observation?
-    };
-    Some(AllocatorMemorySnapshot {
-        backend: crate::allocator_reclaim::allocator_backend(),
-        observation,
-    })
-}
-
-#[cfg(target_os = "windows")]
-fn read_allocator_memory_snapshot() -> Option<AllocatorMemorySnapshot> {
-    None
-}
-
 fn configured_memory_observability_interval_secs() -> u64 {
    rustfs_utils::get_env_u64(ENV_MEMORY_OBSERVABILITY_INTERVAL_SECS, DEFAULT_MEMORY_OBSERVABILITY_INTERVAL_SECS).max(1)
 }
@@ -566,6 +542,13 @@ mod tests {
        assert_eq!(parse_mimalloc_stats_json(r#"{ "allocator": "unknown" }"#), None);
    }

+    #[test]
+    fn read_allocator_memory_snapshot_uses_mimalloc_stats_json() {
+        let snapshot = super::read_allocator_memory_snapshot();
+        #[cfg(not(target_os = "windows"))]
+        assert!(snapshot.is_some(), "allocator snapshot should be available on non-Windows");
+    }
+
    #[test]
    fn memory_observability_snapshot_reports_disabled_when_metrics_are_disabled() {
        let snapshot = build_memory_observability_status_snapshot(false, 15, false);
Author	SHA1	Message	Date
houseme	48f742644c	Merge branch 'main' into fix/b5-t5-c2-preflight-lock	2026-08-23 12:29:10 +08:00
Zhengchao An	b6ba89d9e4	docs(testing): document CI gate matrix (#6412 )	2026-08-23 12:09:06 +08:00
houseme	648d5166e2	feat(allocator): replace mimalloc/libmimalloc-sys with rustfs-mimalloc/rustfs-mimalloc-sys (#6404 ) Replace the upstream xonatius/mimalloc_rust.git fork (mimalloc + libmimalloc-sys) with the published rustfs-mimalloc (v0.5.0) and rustfs-mimalloc-sys (v0.5.0) crates from crates.io. The new crates are based on mimalloc V3 (v3.5.0) and provide: - MiMalloc global allocator with safe API (collect, stats_json, process_info) - Heap management and arena operations (heap module) - Full FFI bindings to mimalloc V3 Changes: - Workspace deps: mimalloc + libmimalloc-sys (git) → rustfs-mimalloc + rustfs-mimalloc-sys (crates.io) - allocator_reclaim.rs: libmimalloc_sys::mi_collect → rustfs_mimalloc::MiMalloc::collect - memory_observability.rs: raw FFI mi_stats_get_json → MiMalloc::stats_json() - main.rs: heap ownership tests use Heap::contains() (V3 API) - deny.toml: remove xonatius/mimalloc_rust.git from allow-git Co-authored-by: heihutu <heihutu@gmail.com>	2026-08-23 12:07:25 +08:00
houseme	84eb5aebef	fix(ecstore): remove inline write debug noise (#6408 ) * fix(ecstore): remove inline write debug noise Co-Authored-By: heihutu <heihutu@gmail.com> * fix(ecstore): satisfy warning-as-error lints Co-Authored-By: heihutu <heihutu@gmail.com> --------- Co-authored-by: heihutu <heihutu@gmail.com>	2026-08-23 12:07:20 +08:00
唐小鸭	f8203b43f9	fix(admin): bound site replication lifecycle lock and parallelize add preflight The site replication add preflight probed peer sites serially while holding the process-wide lifecycle lock, so k unreachable sites held the lock for k peer-request timeouts, and every concurrent add/remove/refresh waited on an unbounded lock acquire for the whole time. Probe all sites concurrently (matching the file's other peer fan-outs) so k unreachable sites cost roughly one timeout, and bound the lifecycle lock acquire at 30s, returning a retryable 503 to waiters instead of hanging indefinitely. Regression tests pin the preflight fan-out concurrency, the bounded acquire's 503, and the 10s/3s peer client timeout constants. Refs rustfs/backlog#1952, rustfs/backlog#1946, rustfs/backlog#1889	2026-08-21 18:59:59 +08:00