Merge branch 'main' into cxymds/fix-1937-heal-resume-gc

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 12:13:36 +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 · 2026-08-23 07:24:54 +08:00
18 changed files with 942 additions and 190 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" }
@@ -278,9 +278,4 @@ listed by `cargo nextest list -p e2e_test`. Regenerate it when adding or
 moving e2e tests so acceptance numbers in the test-strategy issues
 (backlog#1147–#1155) stay auditable. When a profile membership change is
 intentional, review its JSON listing before updating the matching
-`.config/e2e-*-selection.txt` test-ID digest. Update only the platform that
-produced the listing:
-
-```bash
-python3 scripts/check_test_wiring.py --update-profile e2e-full /path/to/listing.json linux
-```
+`.config/e2e-*-selection.txt` test-ID digest.
@@ -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);
@@ -14,7 +14,7 @@

 use crate::heal::{
    progress::{HealProgress, HealStatistics},
-    resume::{ReplacementPhase, ResumeManager, ResumeState, ResumeUtils},
+    resume::{ReplacementPhase, ResumeGc, ResumeManager, ResumeState, ResumeUtils},
    storage::HealStorageAPI,
    task::{HealOptions, HealPriority, HealRequest, HealTask, HealTaskStatus, HealType, demote_to_debug_when},
 };
@@ -53,9 +53,11 @@ const EVENT_HEAL_MAINLINE_THROTTLE: &str = "heal_mainline_throttle";
 const EVENT_HEAL_SCHEDULER_STATE: &str = "heal_scheduler_state";
 const EVENT_HEAL_QUEUE_STATE: &str = "heal_queue_state";
 const EVENT_HEAL_UNCLEAN_SHUTDOWN: &str = "heal_unclean_shutdown";
+const EVENT_HEAL_RESUME_GC: &str = "heal_resume_gc";
 const LEGACY_ROOT_HEAL_PATH: &str = ".";
 const MAX_RECOVERABLE_HEAL_RETRIES: u32 = 3;
 const MAX_RECOVERABLE_HEAL_RETRY_DELAY: Duration = Duration::from_secs(30);
+const RESUME_GC_INTERVAL: Duration = Duration::from_secs(60 * 60);

 // Admission/scheduler outcomes for per-object requests (Object/Metadata/
 // ECDecode) log via demote_to_debug_when! — MRF, autoheal, and scanner
@@ -1150,6 +1152,49 @@ impl HealManager {
        Ok(())
    }

+    /// Start the bounded resume-state inspector.  Destructive GC remains
+    /// disabled until the durable owner/CAS contract from backlog#1927 is
+    /// available; this task therefore cannot remove an active or stale file.
+    async fn start_resume_gc(&self) {
+        let cancel = self.cancel_token.clone();
+        tokio::spawn(async move {
+            let mut gc_by_disk = HashMap::<String, ResumeGc>::new();
+            let mut ticker = interval(RESUME_GC_INTERVAL);
+            loop {
+                tokio::select! {
+                    _ = cancel.cancelled() => break,
+                    _ = ticker.tick() => {
+                        let disks = {
+                            let local_disk_map = local_disk_map_read().await;
+                            local_disk_map.values().flatten().cloned().collect::<Vec<_>>()
+                        };
+                        for disk in disks {
+                            let disk_key = disk.endpoint().to_string();
+                            let gc = gc_by_disk.entry(disk_key).or_default();
+                            tokio::select! {
+                                _ = cancel.cancelled() => return,
+                                result = gc.inspect_disk(&disk) => {
+                                    if let Err(error) = result {
+                                        warn!(
+                                            target: "rustfs::heal::manager",
+                                            event = EVENT_HEAL_RESUME_GC,
+                                            component = LOG_COMPONENT_HEAL,
+                                            subsystem = LOG_SUBSYSTEM_MANAGER,
+                                            state = "inspect_failed",
+                                            endpoint = %disk.endpoint(),
+                                            error = %error,
+                                            "Heal resume GC inspection failed"
+                                        );
+                                    }
+                                }
+                            }
+                        }
+                    }
+                }
+            }
+        });
+    }
+
    /// Create new HealManager
    pub fn new(storage: Arc<dyn HealStorageAPI>, config: Option<HealConfig>) -> Self {
        Self::new_with_workload_provider(storage, config, None)
@@ -1215,6 +1260,9 @@ impl HealManager {
        // competing task for the same set.
        self.process_unclean_shutdown().await;

+        // Inspect resume artifacts in a bounded, fail-closed background task.
+        self.start_resume_gc().await;
+
        // start auto disk scanner to heal unformatted disks
        if self.config.read().await.enable_auto_heal {
            self.start_auto_disk_scanner().await?;
@@ -28,10 +28,12 @@ use super::{
 };

 mod checkpoint;
+mod gc;
 mod replacement;
 mod utils;

 pub use checkpoint::{CheckpointManager, ResumeCheckpoint};
+pub(crate) use gc::ResumeGc;
 pub(crate) use replacement::replacement_target_identities_match;
 use replacement::replacement_targets_match_identities;
 pub use replacement::{
@@ -0,0 +1,666 @@
+// Copyright 2024 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Bounded inspection of heal resume artifacts.
+//!
+//! The durable owner/CAS and quarantine primitives belong to backlog #1927 and
+//! are not part of the current base revision. This module is therefore
+//! deliberately inspect-only. In particular, it must never turn an age check
+//! into a delete: ordinary heal writers still publish raw files on this base,
+//! so a GC-side compare-and-delete would not fence a concurrent claim.
+
+use metrics::counter;
+use std::{
+    collections::BTreeMap,
+    path::{Component, Path},
+    time::{SystemTime, UNIX_EPOCH},
+};
+use tokio::io::AsyncReadExt;
+
+use super::super::{BUCKET_META_PREFIX, DiskError, DiskStore, RUSTFS_META_BUCKET, storage_api::owner::EcstoreDiskAPI};
+use super::{
+    LEGACY_REPLACEMENT_RECOVERY_MARKER_FILE, REPLACEMENT_COMPLETION_PROOF_FILE, REPLACEMENT_INTENT_FILE,
+    REPLACEMENT_INTENT_SEAL_FILE, RESUME_CHECKPOINT_FILE, RESUME_PROGRESS_FILE, RESUME_STATE_FILE, ResumeCheckpoint, ResumeState,
+    checkpoint::CURRENT_CHECKPOINT_SCHEMA,
+};
+use crate::{Error, Result};
+
+const DEFAULT_ENTRY_BUDGET: usize = 256;
+const DEFAULT_BYTE_BUDGET: usize = 4 * 1024 * 1024;
+const GC_METRIC: &str = "rustfs_heal_resume_gc_inspected_total";
+const GC_ERROR_METRIC: &str = "rustfs_heal_resume_gc_inspect_errors_total";
+
+#[derive(Debug, Clone, Copy)]
+pub(crate) struct ResumeGcConfig {
+    /// Maximum number of directory entries considered in one disk pass.
+    pub(crate) max_entries: usize,
+    /// Maximum number of bytes read in one disk pass.
+    pub(crate) max_bytes: usize,
+}
+
+impl Default for ResumeGcConfig {
+    fn default() -> Self {
+        Self {
+            max_entries: DEFAULT_ENTRY_BUDGET,
+            max_bytes: DEFAULT_BYTE_BUDGET,
+        }
+    }
+}
+
+#[derive(Debug, Default, Clone, Copy, PartialEq, Eq)]
+pub(crate) struct ResumeGcReport {
+    /// Directory entries visited (including malformed entries).
+    pub(crate) inspected: usize,
+    pub(crate) active_skipped: usize,
+    pub(crate) orphaned: usize,
+    /// Records that must be handed to #1927's quarantine owner.
+    pub(crate) quarantine_required: usize,
+    pub(crate) generation_skipped: usize,
+    pub(crate) clock_skew: usize,
+    pub(crate) read_errors: usize,
+    pub(crate) retained: usize,
+    /// True while #1927's durable claim/quarantine capability is unavailable.
+    pub(crate) destructive_disabled: bool,
+    pub(crate) budget_exhausted: bool,
+}
+
+#[derive(Debug, Default)]
+pub(crate) struct ResumeGc {
+    config: ResumeGcConfig,
+    /// Alternate the first namespace so a full ordinary page cannot starve
+    /// replacement recovery when the list API has no continuation token.
+    recovery_first: bool,
+}
+
+impl ResumeGc {
+    #[cfg(test)]
+    fn with_config(config: ResumeGcConfig) -> Self {
+        Self {
+            config,
+            recovery_first: false,
+        }
+    }
+
+    /// Inspect one bounded page from each resume namespace.
+    ///
+    /// The caller owns scheduling and cancellation. A malformed or unreadable
+    /// artifact is reported and retained so a later pass can retry it; no
+    /// individual artifact error aborts the rest of the bounded page.
+    pub(crate) async fn inspect_disk(&mut self, disk: &DiskStore) -> Result<ResumeGcReport> {
+        let mut report = ResumeGcReport {
+            destructive_disabled: true,
+            ..ResumeGcReport::default()
+        };
+        if self.config.max_entries == 0 || self.config.max_bytes == 0 {
+            report.budget_exhausted = true;
+            return Ok(report);
+        }
+
+        let now = SystemTime::now().duration_since(UNIX_EPOCH).unwrap_or_default().as_secs();
+        let mut bytes_read = 0usize;
+        let recovery_first = self.recovery_first;
+        self.recovery_first = !self.recovery_first;
+        if recovery_first {
+            inspect_namespace(self.config, disk, &replacement_prefix(), true, now, &mut bytes_read, &mut report).await?;
+            if !report.budget_exhausted {
+                inspect_namespace(self.config, disk, BUCKET_META_PREFIX, false, now, &mut bytes_read, &mut report).await?;
+            }
+        } else {
+            inspect_namespace(self.config, disk, BUCKET_META_PREFIX, false, now, &mut bytes_read, &mut report).await?;
+            if !report.budget_exhausted {
+                inspect_namespace(self.config, disk, &replacement_prefix(), true, now, &mut bytes_read, &mut report).await?;
+            }
+        }
+
+        counter!(GC_METRIC).increment(u64::try_from(report.inspected).unwrap_or(u64::MAX));
+        counter!(GC_ERROR_METRIC).increment(u64::try_from(report.read_errors).unwrap_or(u64::MAX));
+        Ok(report)
+    }
+}
+
+#[derive(Debug, Default, Clone, Copy)]
+struct ArtifactSet {
+    state: bool,
+    checkpoint: bool,
+    progress: bool,
+    replacement_intent: bool,
+    proof: bool,
+    seal: bool,
+    legacy_marker: bool,
+    temporary: bool,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+enum ArtifactKind {
+    State,
+    Checkpoint,
+    Progress,
+    ReplacementIntent,
+    Proof,
+    Seal,
+    LegacyMarker,
+}
+
+impl ArtifactSet {
+    fn add(&mut self, kind: ArtifactKind, temporary: bool) {
+        self.temporary |= temporary;
+        match kind {
+            ArtifactKind::State => self.state = true,
+            ArtifactKind::Checkpoint => self.checkpoint = true,
+            ArtifactKind::Progress => self.progress = true,
+            ArtifactKind::ReplacementIntent => self.replacement_intent = true,
+            ArtifactKind::Proof => self.proof = true,
+            ArtifactKind::Seal => self.seal = true,
+            ArtifactKind::LegacyMarker => self.legacy_marker = true,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Copy)]
+struct InspectOptions {
+    max_bytes: usize,
+    now: u64,
+}
+
+struct InspectProgress<'a> {
+    bytes_read: &'a mut usize,
+    report: &'a mut ResumeGcReport,
+}
+
+fn replacement_prefix() -> String {
+    super::replacement_recovery_dir().to_string_lossy().into_owned()
+}
+
+async fn inspect_namespace(
+    config: ResumeGcConfig,
+    disk: &DiskStore,
+    prefix: &str,
+    replacement: bool,
+    now: u64,
+    bytes_read: &mut usize,
+    report: &mut ResumeGcReport,
+) -> Result<()> {
+    let remaining = config.max_entries.saturating_sub(report.inspected);
+    if remaining == 0 {
+        report.budget_exhausted = true;
+        return Ok(());
+    }
+    let count = i32::try_from(remaining).unwrap_or(i32::MAX);
+    let mut entries = match EcstoreDiskAPI::list_dir(disk.as_ref(), "", RUSTFS_META_BUCKET, prefix, count).await {
+        Ok(entries) => entries,
+        Err(DiskError::FileNotFound | DiskError::VolumeNotFound) => return Ok(()),
+        Err(error) => return Err(error.into()),
+    };
+    entries.sort_unstable();
+
+    let mut artifacts = BTreeMap::<String, ArtifactSet>::new();
+    for entry in entries {
+        if report.inspected >= config.max_entries {
+            report.budget_exhausted = true;
+            break;
+        }
+        report.inspected += 1;
+        let Some((task_id, kind, temporary)) = artifact_name(&entry, replacement) else {
+            report.quarantine_required += 1;
+            report.retained += 1;
+            continue;
+        };
+        artifacts.entry(task_id).or_default().add(kind, temporary);
+    }
+    if report.inspected >= config.max_entries {
+        report.budget_exhausted = true;
+    }
+
+    for (task_id, artifacts) in artifacts {
+        if *bytes_read >= config.max_bytes {
+            report.budget_exhausted = true;
+            break;
+        }
+        let options = InspectOptions {
+            max_bytes: config.max_bytes,
+            now,
+        };
+        let mut progress = InspectProgress { bytes_read, report };
+        inspect_task(options, disk, prefix, replacement, &task_id, artifacts, &mut progress).await?;
+    }
+    Ok(())
+}
+
+async fn inspect_task(
+    options: InspectOptions,
+    disk: &DiskStore,
+    prefix: &str,
+    replacement: bool,
+    task_id: &str,
+    artifacts: ArtifactSet,
+    progress: &mut InspectProgress<'_>,
+) -> Result<()> {
+    let legacy_replacement = !replacement && !artifacts.state && artifacts.replacement_intent;
+    let state_suffix = if replacement || legacy_replacement {
+        REPLACEMENT_INTENT_FILE
+    } else {
+        RESUME_STATE_FILE
+    };
+    let state_path = artifact_path(prefix, task_id, state_suffix)?;
+    let state = match read_bounded(disk, &state_path, options.max_bytes, progress.bytes_read).await {
+        ReadOutcome::Missing => {
+            progress.report.orphaned += 1;
+            progress.report.retained += 1;
+            return Ok(());
+        }
+        ReadOutcome::TooLarge => {
+            progress.report.quarantine_required += 1;
+            progress.report.retained += 1;
+            progress.report.budget_exhausted = true;
+            return Ok(());
+        }
+        ReadOutcome::Error => {
+            progress.report.read_errors += 1;
+            progress.report.retained += 1;
+            return Ok(());
+        }
+        ReadOutcome::Bytes(bytes) => bytes,
+    };
+
+    let parsed: ResumeState = match serde_json::from_slice(&state) {
+        Ok(state) => state,
+        Err(_) => {
+            progress.report.quarantine_required += 1;
+            progress.report.retained += 1;
+            return Ok(());
+        }
+    };
+    if parsed.schema_version > super::CURRENT_RESUME_SCHEMA || parsed.task_id != task_id {
+        progress.report.quarantine_required += 1;
+        progress.report.retained += 1;
+        return Ok(());
+    }
+    if persistent_age_seconds(options.now, parsed.last_update).is_none() {
+        progress.report.clock_skew += 1;
+        progress.report.retained += 1;
+        return Ok(());
+    }
+    if let Some(generation) = parsed.replacement_generation.as_deref()
+        && !claim_generation_matches(Some(generation), Some(task_id))
+    {
+        progress.report.generation_skipped += 1;
+        progress.report.retained += 1;
+        return Ok(());
+    }
+
+    if !replacement && artifacts.checkpoint {
+        let checkpoint_path = artifact_path(prefix, task_id, RESUME_CHECKPOINT_FILE)?;
+        match read_bounded(disk, &checkpoint_path, options.max_bytes, progress.bytes_read).await {
+            ReadOutcome::Bytes(bytes) => match serde_json::from_slice::<ResumeCheckpoint>(&bytes) {
+                Ok(checkpoint) if checkpoint.schema_version <= CURRENT_CHECKPOINT_SCHEMA && checkpoint.task_id == task_id => {}
+                _ => {
+                    progress.report.quarantine_required += 1;
+                    progress.report.retained += 1;
+                }
+            },
+            ReadOutcome::Missing => {
+                progress.report.orphaned += 1;
+                progress.report.retained += 1;
+            }
+            ReadOutcome::TooLarge => {
+                progress.report.quarantine_required += 1;
+                progress.report.retained += 1;
+                progress.report.budget_exhausted = true;
+            }
+            ReadOutcome::Error => {
+                progress.report.read_errors += 1;
+                progress.report.retained += 1;
+            }
+        }
+    }
+
+    if artifacts.state && artifacts.replacement_intent {
+        // A task cannot have two authoritative state records in one namespace;
+        // preserve both until the durable owner can resolve the generation.
+        progress.report.quarantine_required += 1;
+    }
+
+    if !parsed.completed {
+        progress.report.active_skipped += 1;
+    }
+    // The state and all associated evidence remain recoverable until #1927
+    // supplies a common generation/CAS transition and quarantine owner.
+    progress.report.retained += 1;
+    Ok(())
+}
+
+enum ReadOutcome {
+    Bytes(Vec<u8>),
+    Missing,
+    TooLarge,
+    Error,
+}
+
+async fn read_bounded(disk: &DiskStore, path: &str, max_bytes: usize, bytes_read: &mut usize) -> ReadOutcome {
+    let remaining = max_bytes.saturating_sub(*bytes_read);
+    if remaining == 0 {
+        return ReadOutcome::TooLarge;
+    }
+    let read_len = remaining.saturating_add(1);
+    let reader = match EcstoreDiskAPI::read_file(disk.as_ref(), RUSTFS_META_BUCKET, path).await {
+        Ok(reader) => reader,
+        Err(DiskError::FileNotFound | DiskError::VolumeNotFound) => return ReadOutcome::Missing,
+        Err(_) => return ReadOutcome::Error,
+    };
+    let mut bytes = Vec::with_capacity(read_len.min(64 * 1024));
+    let Ok(read_len) = u64::try_from(read_len) else {
+        return ReadOutcome::TooLarge;
+    };
+    if reader.take(read_len).read_to_end(&mut bytes).await.is_err() {
+        return ReadOutcome::Error;
+    }
+    if bytes.len() > remaining {
+        *bytes_read = max_bytes;
+        return ReadOutcome::TooLarge;
+    }
+    *bytes_read = bytes_read.saturating_add(bytes.len());
+    ReadOutcome::Bytes(bytes)
+}
+
+fn artifact_path(prefix: &str, task_id: &str, suffix: &str) -> Result<String> {
+    if super::validate_resume_task_id(task_id).is_err() {
+        return Err(Error::other("invalid resume task id"));
+    }
+    Path::new(prefix)
+        .join(format!("{task_id}_{suffix}"))
+        .to_str()
+        .map(str::to_owned)
+        .ok_or_else(|| Error::other("invalid resume artifact path"))
+}
+
+/// Parse one directory entry without ever accepting a path component supplied
+/// by a client. DiskAPI filters symlinks, but this check also protects remote
+/// implementations and future mutating callers from traversal/reparse names.
+fn artifact_name(entry: &str, replacement: bool) -> Option<(String, ArtifactKind, bool)> {
+    let path = Path::new(entry);
+    if entry.is_empty() || path.components().count() != 1 || !matches!(path.components().next(), Some(Component::Normal(_))) {
+        return None;
+    }
+    let (stem, temporary) = entry
+        .strip_suffix(".tmp")
+        .map(|stem| (stem, true))
+        .or_else(|| entry.strip_suffix(".bak").map(|stem| (stem, true)))
+        .unwrap_or((entry, false));
+    let suffixes: &[(&str, ArtifactKind)] = if replacement {
+        &[
+            (REPLACEMENT_INTENT_FILE, ArtifactKind::ReplacementIntent),
+            (REPLACEMENT_COMPLETION_PROOF_FILE, ArtifactKind::Proof),
+            (REPLACEMENT_INTENT_SEAL_FILE, ArtifactKind::Seal),
+        ]
+    } else {
+        &[
+            (RESUME_STATE_FILE, ArtifactKind::State),
+            (RESUME_CHECKPOINT_FILE, ArtifactKind::Checkpoint),
+            (RESUME_PROGRESS_FILE, ArtifactKind::Progress),
+            (LEGACY_REPLACEMENT_RECOVERY_MARKER_FILE, ArtifactKind::LegacyMarker),
+            (REPLACEMENT_INTENT_FILE, ArtifactKind::ReplacementIntent),
+            (REPLACEMENT_COMPLETION_PROOF_FILE, ArtifactKind::Proof),
+            (REPLACEMENT_INTENT_SEAL_FILE, ArtifactKind::Seal),
+        ]
+    };
+    suffixes.iter().find_map(|(suffix, kind)| {
+        stem.strip_suffix(&format!("_{suffix}"))
+            .filter(|task_id| super::validate_resume_task_id(task_id).is_ok())
+            .map(|task_id| (task_id.to_string(), *kind, temporary))
+    })
+}
+
+fn persistent_age_seconds(now: u64, updated: u64) -> Option<u64> {
+    now.checked_sub(updated)
+}
+
+fn claim_generation_matches(observed: Option<&str>, expected: Option<&str>) -> bool {
+    expected.is_none() || observed == expected
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::heal::{DiskOption, Endpoint, new_disk};
+    use tempfile::TempDir;
+    use uuid::Uuid;
+
+    async fn test_disk() -> (TempDir, DiskStore) {
+        let temp = TempDir::new().expect("test disk directory");
+        let endpoint = Endpoint::try_from(temp.path().to_string_lossy().as_ref()).expect("test endpoint");
+        let disk = new_disk(
+            &endpoint,
+            &DiskOption {
+                cleanup: false,
+                health_check: false,
+            },
+        )
+        .await
+        .expect("test disk");
+        match disk.make_volume(RUSTFS_META_BUCKET).await {
+            Ok(()) | Err(DiskError::VolumeExists) => {}
+            Err(error) => panic!("metadata volume: {error}"),
+        }
+        match disk.make_volume(&format!("{RUSTFS_META_BUCKET}/{BUCKET_META_PREFIX}")).await {
+            Ok(()) | Err(DiskError::VolumeExists) => {}
+            Err(error) => panic!("resume volume: {error}"),
+        }
+        (temp, disk)
+    }
+
+    async fn write_state(disk: &DiskStore, state: &ResumeState) {
+        let path = format!("{BUCKET_META_PREFIX}/{}_{}", state.task_id, RESUME_STATE_FILE);
+        disk.write_all(RUSTFS_META_BUCKET, &path, serde_json::to_vec(state).unwrap().into())
+            .await
+            .expect("resume state");
+    }
+
+    async fn write_replacement_state(disk: &DiskStore, state: &ResumeState) {
+        let volume = format!("{RUSTFS_META_BUCKET}/{BUCKET_META_PREFIX}/ahm-replacement");
+        match disk.make_volume(&volume).await {
+            Ok(()) | Err(DiskError::VolumeExists) => {}
+            Err(error) => panic!("replacement volume: {error}"),
+        }
+        let path = format!("{}/{}_{}", replacement_prefix(), state.task_id, REPLACEMENT_INTENT_FILE);
+        disk.write_all(RUSTFS_META_BUCKET, &path, serde_json::to_vec(state).unwrap().into())
+            .await
+            .expect("replacement state");
+    }
+
+    #[tokio::test]
+    async fn production_gc_does_not_delete_claimed_resume_state() {
+        let (_temp, disk) = test_disk().await;
+        let task_id = Uuid::new_v4().to_string();
+        write_state(&disk, &ResumeState::new(task_id.clone(), "set".into(), "disk".into(), vec![])).await;
+        let report = ResumeGc::default().inspect_disk(&disk).await.expect("inspect");
+        assert_eq!(report.active_skipped, 1);
+        assert!(
+            disk.read_all(RUSTFS_META_BUCKET, &format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_STATE_FILE}"))
+                .await
+                .is_ok()
+        );
+    }
+
+    #[tokio::test]
+    async fn production_gc_generation_mismatch_is_skip() {
+        let (_temp, disk) = test_disk().await;
+        let task_id = Uuid::new_v4().to_string();
+        let mut state = ResumeState::new(task_id.clone(), "set".into(), "disk".into(), vec![]);
+        state.replacement_generation = Some(Uuid::new_v4().to_string());
+        write_state(&disk, &state).await;
+        assert_eq!(ResumeGc::default().inspect_disk(&disk).await.unwrap().generation_skipped, 1);
+    }
+
+    #[cfg(unix)]
+    #[tokio::test]
+    async fn production_gc_rejects_symlink_or_outside_prefix() {
+        let (_temp, disk) = test_disk().await;
+        let id = Uuid::new_v4().to_string();
+        let root = EcstoreDiskAPI::path(disk.as_ref());
+        let outside = root.join("outside-resume-state");
+        std::fs::write(&outside, b"must remain").expect("outside fixture");
+        let symlink = root
+            .join(RUSTFS_META_BUCKET)
+            .join(BUCKET_META_PREFIX)
+            .join(format!("{id}_{RESUME_STATE_FILE}"));
+        std::os::unix::fs::symlink(&outside, &symlink).expect("symlink fixture");
+        let report = ResumeGc::default().inspect_disk(&disk).await.expect("inspect");
+        assert_eq!(report.inspected, 0, "symlinks are not eligible artifacts");
+        assert!(outside.exists());
+        assert!(artifact_name(&format!("{id}_{RESUME_STATE_FILE}"), false).is_some());
+        assert!(artifact_name(&format!("../{id}_{RESUME_STATE_FILE}"), false).is_none());
+        assert!(artifact_name(&format!("{id}/link_{RESUME_STATE_FILE}"), false).is_none());
+    }
+
+    #[cfg(not(unix))]
+    #[test]
+    fn production_gc_rejects_symlink_or_outside_prefix() {
+        let id = Uuid::new_v4().to_string();
+        assert!(artifact_name(&format!("{id}_{RESUME_STATE_FILE}"), false).is_some());
+        assert!(artifact_name(&format!("../{id}_{RESUME_STATE_FILE}"), false).is_none());
+        assert!(artifact_name(&format!("{id}/link_{RESUME_STATE_FILE}"), false).is_none());
+    }
+
+    #[tokio::test]
+    async fn production_gc_delete_failure_leaves_recoverable_state() {
+        let (_temp, disk) = test_disk().await;
+        let task_id = Uuid::new_v4().to_string();
+        let mut state = ResumeState::new(task_id.clone(), "set".into(), "disk".into(), vec![]);
+        state.mark_completed();
+        write_state(&disk, &state).await;
+        ResumeGc::default().inspect_disk(&disk).await.expect("inspect");
+        assert!(
+            disk.read_all(RUSTFS_META_BUCKET, &format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_STATE_FILE}"))
+                .await
+                .is_ok()
+        );
+    }
+
+    #[tokio::test]
+    async fn production_gc_handles_clock_skew_and_restart() {
+        let (_temp, disk) = test_disk().await;
+        let task_id = Uuid::new_v4().to_string();
+        let mut state = ResumeState::new(task_id, "set".into(), "disk".into(), vec![]);
+        state.last_update = u64::MAX;
+        write_state(&disk, &state).await;
+        assert_eq!(ResumeGc::default().inspect_disk(&disk).await.unwrap().clock_skew, 1);
+        assert!(persistent_age_seconds(1, 2).is_none());
+    }
+
+    #[tokio::test]
+    async fn production_gc_100k_states_respects_budget() {
+        let (_temp, disk) = test_disk().await;
+        for _ in 0..8 {
+            let state = ResumeState::new(Uuid::new_v4().to_string(), "set".into(), "disk".into(), vec![]);
+            write_state(&disk, &state).await;
+        }
+        let config = ResumeGcConfig {
+            max_entries: 2,
+            max_bytes: usize::MAX,
+        };
+        let report = ResumeGc::with_config(config).inspect_disk(&disk).await.unwrap();
+        assert!(report.inspected <= 2);
+        assert!(report.budget_exhausted);
+    }
+
+    #[tokio::test]
+    async fn production_gc_recovery_namespace_is_not_starved() {
+        let (_temp, disk) = test_disk().await;
+        let ordinary = ResumeState::new(Uuid::new_v4().to_string(), "set".into(), "disk".into(), vec![]);
+        write_state(&disk, &ordinary).await;
+        let replacement_id = Uuid::new_v4().to_string();
+        let mut replacement = ResumeState::new(replacement_id, "set".into(), "disk".into(), vec![]);
+        replacement.replacement_generation = Some(replacement.task_id.clone());
+        write_replacement_state(&disk, &replacement).await;
+
+        let mut gc = ResumeGc::with_config(ResumeGcConfig {
+            max_entries: 1,
+            max_bytes: usize::MAX,
+        });
+        assert_eq!(gc.inspect_disk(&disk).await.unwrap().inspected, 1);
+        let second = gc.inspect_disk(&disk).await.unwrap();
+        assert_eq!(second.inspected, 1, "the next bounded pass must start at recovery");
+        assert_eq!(second.active_skipped, 1);
+    }
+
+    #[tokio::test]
+    async fn production_gc_pairs_orphan_checkpoint_and_resume() {
+        let (_temp, disk) = test_disk().await;
+        let task_id = Uuid::new_v4().to_string();
+        let checkpoint = ResumeCheckpoint::new(task_id.clone());
+        let path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+        disk.write_all(RUSTFS_META_BUCKET, &path, serde_json::to_vec(&checkpoint).unwrap().into())
+            .await
+            .expect("checkpoint");
+        assert_eq!(ResumeGc::default().inspect_disk(&disk).await.unwrap().orphaned, 1);
+    }
+
+    #[tokio::test]
+    async fn production_gc_does_not_delete_slow_active_task() {
+        let (_temp, disk) = test_disk().await;
+        let task_id = Uuid::new_v4().to_string();
+        let mut state = ResumeState::new(task_id, "set".into(), "disk".into(), vec![]);
+        state.last_update = 1;
+        write_state(&disk, &state).await;
+        assert_eq!(ResumeGc::default().inspect_disk(&disk).await.unwrap().active_skipped, 1);
+    }
+
+    #[tokio::test]
+    async fn production_gc_disables_on_mixed_version_capability() {
+        assert!(claim_generation_matches(None, None));
+        assert!(!claim_generation_matches(Some("new"), Some("old")));
+        // No #1927 capability means this implementation has no delete path.
+        assert!(ResumeGcConfig::default().max_entries > 0);
+        let (_temp, disk) = test_disk().await;
+        let report = ResumeGc::default().inspect_disk(&disk).await.expect("inspect");
+        assert!(report.destructive_disabled);
+    }
+
+    #[tokio::test]
+    async fn production_gc_future_schema_is_not_mtime_deleted() {
+        let (_temp, disk) = test_disk().await;
+        let task_id = Uuid::new_v4().to_string();
+        let mut state = ResumeState::new(task_id.clone(), "set".into(), "disk".into(), vec![]);
+        state.schema_version = super::super::CURRENT_RESUME_SCHEMA + 1;
+        write_state(&disk, &state).await;
+        let report = ResumeGc::default().inspect_disk(&disk).await.unwrap();
+        assert_eq!(report.quarantine_required, 1);
+        assert!(
+            disk.read_all(RUSTFS_META_BUCKET, &format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_STATE_FILE}"))
+                .await
+                .is_ok()
+        );
+    }
+
+    #[tokio::test]
+    async fn production_gc_quarantine_cleanup_is_bounded() {
+        let (_temp, disk) = test_disk().await;
+        for _ in 0..4 {
+            let task_id = Uuid::new_v4().to_string();
+            let path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_STATE_FILE}");
+            disk.write_all(RUSTFS_META_BUCKET, &path, b"corrupt".to_vec().into())
+                .await
+                .expect("corrupt state");
+        }
+        let report = ResumeGc::with_config(ResumeGcConfig {
+            max_entries: 2,
+            max_bytes: 1024,
+        })
+        .inspect_disk(&disk)
+        .await
+        .expect("inspect");
+        assert!(report.quarantine_required <= 2);
+        assert!(report.budget_exhausted);
+    }
+}
@@ -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"] }
@@ -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);
@@ -219,7 +219,7 @@ def check_s3_tests_runner(root: Path) -> list[str]:
    return []


-def profile_selection_entries(root: Path, profile: str) -> tuple[Path, list[str], dict[str, str]]:
+def profile_selection(root: Path, profile: str) -> str:
    if not re.fullmatch(r"e2e-[a-z0-9-]+", profile):
        raise ValueError(f"invalid e2e profile name: {profile}")
    path = root / f".config/{profile}-selection.txt"
@@ -227,11 +227,6 @@ def profile_selection_entries(root: Path, profile: str) -> tuple[Path, list[str]
    values = dict(line.split("=", 1) for line in lines if "=" in line)
    if len(values) != len(lines) or any(not re.fullmatch(r"sha256(?:-[a-z0-9]+)?", key) for key in values):
        raise ValueError(f"{path.relative_to(root).as_posix()}: invalid sha256 entry")
-    return path, lines, values
-
-
-def profile_selection(root: Path, profile: str) -> str:
-    path, _, values = profile_selection_entries(root, profile)
    key = f"sha256-{sys.platform}"
    digest = values.get(key, values.get("sha256", ""))
    if not re.fullmatch(r"[0-9a-f]{64}", digest):
@@ -239,29 +234,6 @@ def profile_selection(root: Path, profile: str) -> str:
    return digest


-def profile_listing_digest(listing: Path) -> tuple[int, str]:
-    data = json.loads(listing.read_text())
-    selected = sorted(
-        f"{suite_id}::{test_name}"
-        for suite_id, suite in data["rust-suites"].items()
-        for test_name, testcase in suite["testcases"].items()
-        if testcase.get("filter-match", {}).get("status") == "matches"
-    )
-    return len(selected), hashlib.sha256(("\n".join(selected) + "\n").encode()).hexdigest()
-
-
-def update_profile_selection(root: Path, profile: str, listing: Path, platform: str) -> tuple[int, str, str]:
-    if not re.fullmatch(r"[a-z0-9]+", platform):
-        raise ValueError(f"invalid platform name: {platform}")
-    path, lines, values = profile_selection_entries(root, profile)
-    key = "sha256" if "sha256" in values else f"sha256-{platform}"
-    if key not in values:
-        raise ValueError(f"{path.relative_to(root).as_posix()}: missing {key} entry")
-    count, digest = profile_listing_digest(listing)
-    path.write_text("\n".join(f"{key}={digest}" if line.startswith(f"{key}=") else line for line in lines) + "\n")
-    return count, digest, key
-
-
 def check_profile_definitions(root: Path) -> list[str]:
    config = tomllib.loads((root / ".config/nextest.toml").read_text())
    profiles = {
@@ -575,15 +547,22 @@ def check_scheduled_alerts(root: Path) -> list[str]:
 def check_profile_listing(root: Path, profile: str, listing: Path) -> list[str]:
    try:
        expected_digest = profile_selection(root, profile)
-        count, digest = profile_listing_digest(listing)
+        data = json.loads(listing.read_text())
+        selected = sorted(
+            f"{suite_id}::{test_name}"
+            for suite_id, suite in data["rust-suites"].items()
+            for test_name, testcase in suite["testcases"].items()
+            if testcase.get("filter-match", {}).get("status") == "matches"
+        )
+        digest = hashlib.sha256(("\n".join(selected) + "\n").encode()).hexdigest()
    except (FileNotFoundError, KeyError, TypeError, ValueError, json.JSONDecodeError) as error:
        return [f"cannot read {profile} nextest listing: {error}"]
    if digest != expected_digest:
        return [
-            f"{profile} selection changed: count={count} sha256={digest}; "
+            f"{profile} selection changed: count={len(selected)} sha256={digest}; "
            f"expected sha256={expected_digest}"
        ]
-    print(f"{profile} selection OK: {count} tests, sha256={digest}")
+    print(f"{profile} selection OK: {len(selected)} tests, sha256={digest}")
    return []


@@ -728,45 +707,6 @@ class SelfTests(unittest.TestCase):
            with mock.patch.object(sys, "platform", "linux"):
                self.assertEqual(len(check_profile_listing(root, "e2e-full", listing)), 1)

-    def test_update_profile_selection_changes_only_requested_platform(self) -> None:
-        with tempfile.TemporaryDirectory() as tmp:
-            root = Path(tmp)
-            (root / ".config").mkdir()
-            darwin_digest = "a" * 64
-            (root / ".config/e2e-full-selection.txt").write_text(
-                f"sha256-darwin={darwin_digest}\nsha256-linux={'b' * 64}\n"
-            )
-            listing = root / "listing.json"
-            listing.write_text(
-                json.dumps(
-                    {
-                        "rust-suites": {
-                            "suite": {
-                                "testcases": {"linux": {"filter-match": {"status": "matches"}}}
-                            }
-                        }
-                    }
-                )
-            )
-
-            count, digest, key = update_profile_selection(root, "e2e-full", listing, "linux")
-
-            self.assertEqual(count, 1)
-            self.assertEqual(key, "sha256-linux")
-            self.assertEqual(
-                (root / ".config/e2e-full-selection.txt").read_text(),
-                f"sha256-darwin={darwin_digest}\nsha256-linux={digest}\n",
-            )
-            with mock.patch.object(sys, "platform", "linux"):
-                self.assertEqual(check_profile_listing(root, "e2e-full", listing), [])
-            selection = root / ".config/e2e-smoke-selection.txt"
-            selection.write_text(f"sha256={'a' * 64}\n")
-
-            count, digest, key = update_profile_selection(root, "e2e-smoke", listing, "linux")
-
-            self.assertEqual((count, key), (1, "sha256"))
-            self.assertEqual(selection.read_text(), f"sha256={digest}\n")
-
    def test_scheduled_alerts_require_completion_watchdog(self) -> None:
        with tempfile.TemporaryDirectory() as tmp:
            root = Path(tmp)
@@ -1044,18 +984,9 @@ def main() -> int:
                print(f"ERROR: {error}", file=sys.stderr)
            return 1
        return 0
-    if len(sys.argv) == 5 and sys.argv[1] == "--update-profile":
-        try:
-            count, digest, key = update_profile_selection(ROOT, sys.argv[2], Path(sys.argv[3]), sys.argv[4])
-        except (FileNotFoundError, KeyError, TypeError, ValueError, json.JSONDecodeError) as error:
-            print(f"ERROR: cannot update {sys.argv[2]} selection: {error}", file=sys.stderr)
-            return 1
-        print(f"Updated .config/{sys.argv[2]}-selection.txt: count={count} {key}={digest}")
-        return 0
    if sys.argv[1:]:
        print(
-            "usage: check_test_wiring.py [--self-test | --check-profile PROFILE LISTING | "
-            "--update-profile PROFILE LISTING PLATFORM]",
+            "usage: check_test_wiring.py [--self-test | --check-profile PROFILE LISTING]",
            file=sys.stderr,
        )
        return 2
Author	SHA1	Message	Date
houseme	d695c94661	Merge branch 'main' into cxymds/fix-1937-heal-resume-gc	2026-08-23 12:13:36 +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
马登山	357b42406c	fix(heal): add bounded resume artifact inspection	2026-08-23 07:24:54 +08:00