fix(heal): cleanup consumed MRF replay journals

Do not retain Accepted or Merged replay intents as startup anchors after they have been handed to the heal manager. Only refused or still-pending replay records keep the journal on disk until a successor snapshot can persist them.

This keeps successor snapshots limited to the pending queue, which lets successful replay remove both authoritative and legacy journal paths and restores the crash-boundary tests around successor flush.

Co-Authored-By: heihutu <heihutu@gmail.com>

Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
(cherry picked from commit d5b8f49c9d)

Cargo.lock

@@ -9906,6 +9906,7 @@ dependencies = [
  "regex",
  "rmp",
  "rmp-serde",
+ "rustfs-config",
  "rustfs-utils",
  "s3s",
  "serde",

crates/config/src/constants/api.rs

@@ -100,5 +100,10 @@ pub const DEFAULT_API_MAX_CONNECTIONS: usize = 0;
 /// Example: RUSTFS_API_OBJECT_MAX_VERSIONS=50000
 pub const ENV_API_OBJECT_MAX_VERSIONS: &str = "RUSTFS_API_OBJECT_MAX_VERSIONS";
 
-/// Default for `RUSTFS_API_OBJECT_MAX_VERSIONS`.
-pub const DEFAULT_API_OBJECT_MAX_VERSIONS: u64 = 9_223_372_036_854_775_807;
+/// Default and maximum accepted value for `RUSTFS_API_OBJECT_MAX_VERSIONS`.
+#[cfg(target_pointer_width = "64")]
+pub const DEFAULT_API_OBJECT_MAX_VERSIONS: usize = 9_223_372_036_854_775_807;
+
+/// Default and maximum accepted value for `RUSTFS_API_OBJECT_MAX_VERSIONS`.
+#[cfg(not(target_pointer_width = "64"))]
+pub const DEFAULT_API_OBJECT_MAX_VERSIONS: usize = usize::MAX;

crates/filemeta/Cargo.toml

@@ -44,6 +44,7 @@ tokio = { workspace = true, features = ["io-util", "macros", "sync", "fs", "rt-m
 xxhash-rust = { workspace = true, features = ["xxh64", "xxh3"] }
 bytes = { workspace = true, features = ["serde"] }
 rustfs-utils = { workspace = true, features = ["hash", "http"] }
+rustfs-config = { workspace = true, features = ["constants"] }
 byteorder = { workspace = true }
 tracing.workspace = true
 thiserror.workspace = true

crates/filemeta/src/filemeta.rs

@@ -70,12 +70,8 @@ const _XL_FLAG_INLINE_DATA: u8 = 1 << 2;
 const META_DATA_READ_DEFAULT: usize = 4 << 10;
 const MSGP_UINT32_SIZE: usize = 5;
 
-/// Default max object versions per object, aligned with MinIO's default.
-pub const DEFAULT_OBJECT_MAX_VERSIONS: usize = if usize::BITS >= 64 {
-    9_223_372_036_854_775_807
-} else {
-    usize::MAX
-};
+/// Default max object versions per object.
+pub const DEFAULT_OBJECT_MAX_VERSIONS: usize = rustfs_config::DEFAULT_API_OBJECT_MAX_VERSIONS;
 
 static OBJECT_MAX_VERSIONS: AtomicUsize = AtomicUsize::new(DEFAULT_OBJECT_MAX_VERSIONS);
 

crates/heal/src/heal/mrf_queue.rs

@@ -516,7 +516,6 @@ async fn submit_mrf_heal_request(manager: &HealManager, intent: &MrfIntent) -> c
 
 struct MrfRuntime {
     queue: MrfQueue,
-    retained_replay_intents: Vec<MrfIntent>,
     config: MrfConsumerConfig,
     new_since_flush: usize,
     /// True while the in-memory pending set has changed since the last
@@ -536,7 +535,7 @@ impl MrfRuntime {
     fn snapshot(&self) -> (Vec<u8>, Vec<u8>) {
         let mut authoritative = Vec::new();
         let mut legacy = Vec::new();
-        for intent in self.retained_replay_intents.iter().chain(self.queue.intents()) {
+        for intent in self.queue.intents() {
             let scoped_identity =
                 !matches!(intent.kind, rustfs_common::mrf_channel::MrfKind::MetadataCorruption) && intent.scope.is_some();
             if !encode_intent(intent, &mut authoritative) {
@@ -674,11 +673,10 @@ pub async fn replay_journal_once(manager: &Arc<HealManager>) -> usize {
 struct ReplayOutcome {
     replayed: usize,
     journal_on_disk: bool,
-    retained_replay_intents: Vec<MrfIntent>,
 }
 
-fn replay_must_retain_journal(rearm_incomplete: bool, pending_depth: usize, retained_replay_depth: usize) -> bool {
-    rearm_incomplete || pending_depth > 0 || retained_replay_depth > 0
+fn replay_must_retain_journal(rearm_incomplete: bool, pending_depth: usize) -> bool {
+    rearm_incomplete || pending_depth > 0
 }
 
 /// Shared replay core: read + decode + re-arm, then drain what fits. The
@@ -700,7 +698,6 @@ async fn replay_into(
                 return ReplayOutcome {
                     replayed: 0,
                     journal_on_disk: false,
-                    retained_replay_intents: Vec::new(),
                 };
             }
         },
@@ -740,13 +737,10 @@ async fn replay_into(
 
     // Drain the replayed intents immediately; whatever the manager refuses
     // stays armed in `queue` for the consumer's retry loop.
-    let mut retained_replay_intents = Vec::new();
     if backoff_until.is_none() {
         while let Some(mut intent) = queue.pop_front() {
             match submit_mrf_heal_request(manager, &intent).await {
-                Ok(HealAdmissionResult::Accepted) | Ok(HealAdmissionResult::Merged) => {
-                    retained_replay_intents.push(intent);
-                }
+                Ok(HealAdmissionResult::Accepted) | Ok(HealAdmissionResult::Merged) => {}
                 Ok(HealAdmissionResult::Full) | Ok(HealAdmissionResult::Dropped(HealAdmissionDropReason::QueueFull)) => {
                     intent.attempts = intent.attempts.saturating_add(1);
                     if intent.attempts < MRF_MAX_ATTEMPTS {
@@ -775,7 +769,7 @@ async fn replay_into(
             }
         }
     }
-    let journal_on_disk = if replay_must_retain_journal(rearm_incomplete, queue.depth(), retained_replay_intents.len()) {
+    let journal_on_disk = if replay_must_retain_journal(rearm_incomplete, queue.depth()) {
         true
     } else {
         !delete_journals().await
@@ -783,7 +777,6 @@ async fn replay_into(
     ReplayOutcome {
         replayed,
         journal_on_disk,
-        retained_replay_intents,
     }
 }
 
@@ -793,7 +786,6 @@ async fn run_mrf_consumer(manager: Arc<HealManager>, mut receiver: mpsc::Receive
     let config = MrfConsumerConfig::default();
     let mut runtime = MrfRuntime {
         queue: MrfQueue::new(config.queue_capacity, config.journal_max_bytes),
-        retained_replay_intents: Vec::new(),
         config: config.clone(),
         new_since_flush: 0,
         dirty: false,
@@ -805,7 +797,6 @@ async fn run_mrf_consumer(manager: Arc<HealManager>, mut receiver: mpsc::Receive
     // on disk whenever any replayed intent still needs a successor snapshot.
     let replay = replay_into(&manager, &mut runtime.queue, &mut runtime.backoff_until).await;
     runtime.journal_on_disk = replay.journal_on_disk;
-    runtime.retained_replay_intents = replay.retained_replay_intents;
     // Anything still pending (e.g. the manager was full and backoff armed)
     // must be re-persisted by the next flush before replay can delete the
     // startup anchor.
@@ -823,7 +814,7 @@ async fn run_mrf_consumer(manager: Arc<HealManager>, mut receiver: mpsc::Receive
                     // provably current AND idle (a dirty or pending state
                     // gets one last persist attempt, matching the shutdown
                     // retry the unconditional flush used to provide).
-                    if runtime.dirty || runtime.queue.depth() > 0 || !runtime.retained_replay_intents.is_empty() {
+                    if runtime.dirty || runtime.queue.depth() > 0 {
                         runtime.flush().await;
                     }
                     tracing::info!(
@@ -852,7 +843,6 @@ async fn run_mrf_consumer(manager: Arc<HealManager>, mut receiver: mpsc::Receive
                 match tick_action(
                     runtime.dirty,
                     runtime.queue.depth(),
-                    runtime.retained_replay_intents.len(),
                     runtime.journal_on_disk,
                 ) {
                     TickAction::Flush => {
@@ -867,8 +857,8 @@ async fn run_mrf_consumer(manager: Arc<HealManager>, mut receiver: mpsc::Receive
                         runtime.dispatch(manager.as_ref()).await;
                     }
                     TickAction::DeleteJournal => {
-                        // Only remove a stale journal after every replayed
-                        // intent has a durable successor proof.
+                        // All replayed intents have either been accepted,
+                        // merged, or replaced by a pending successor snapshot.
                         if delete_journals().await {
                             runtime.journal_on_disk = false;
                             gauge!("rustfs_heal_mrf_journal_bytes").set(0.0);
@@ -897,13 +887,11 @@ enum TickAction {
     Idle,
 }
 
-fn tick_action(dirty: bool, depth: usize, retained_replay_depth: usize, journal_on_disk: bool) -> TickAction {
+fn tick_action(dirty: bool, depth: usize, journal_on_disk: bool) -> TickAction {
     if dirty {
         TickAction::Flush
     } else if depth > 0 {
         TickAction::Retry
-    } else if retained_replay_depth > 0 {
-        TickAction::Idle
     } else if journal_on_disk {
         TickAction::DeleteJournal
     } else {
@@ -936,73 +924,34 @@ mod tests {
 
         // Dirty dominates: a changed pending set flushes even when idle
         // otherwise.
-        assert!(matches!(tick_action(true, 0, 0, false), Flush));
-        assert!(matches!(tick_action(true, 3, 0, true), Flush));
+        assert!(matches!(tick_action(true, 0, false), Flush));
+        assert!(matches!(tick_action(true, 3, true), Flush));
 
         // Clean backlog: no rewrite, but keep draining so an expired
         // admission backoff retries on time.
-        assert!(matches!(tick_action(false, 1, 0, false), Retry));
-        assert!(matches!(tick_action(false, 2, 0, true), Retry));
-
-        // Replayed records accepted by the manager are still restart anchors
-        // until a durable successor proof can tombstone them.
-        assert!(matches!(tick_action(false, 0, 1, true), Idle));
+        assert!(matches!(tick_action(false, 1, false), Retry));
+        assert!(matches!(tick_action(false, 2, true), Retry));
 
         // Quiescent with a stale journal file on disk: remove it.
-        assert!(matches!(tick_action(false, 0, 0, true), DeleteJournal));
+        assert!(matches!(tick_action(false, 0, true), DeleteJournal));
 
         // Fully quiescent: nothing to do.
-        assert!(matches!(tick_action(false, 0, 0, false), Idle));
+        assert!(matches!(tick_action(false, 0, false), Idle));
     }
 
     #[test]
     fn replay_cleanup_retains_journal_for_unarmed_or_refused_records() {
         assert!(
-            replay_must_retain_journal(true, 0, 0),
+            replay_must_retain_journal(true, 0),
             "a rejected replay record still needs its disk anchor"
         );
         assert!(
-            replay_must_retain_journal(false, 1, 0),
+            replay_must_retain_journal(false, 1),
             "a Full admission retry must keep the startup journal until the next snapshot"
         );
         assert!(
-            replay_must_retain_journal(false, 0, 1),
-            "an accepted replay record still needs a durable successor before cleanup"
-        );
-        assert!(
-            !replay_must_retain_journal(false, 0, 0),
-            "only a fully consumed replay snapshot with no retained anchors may be deleted"
-        );
-    }
-
-    #[test]
-    fn retained_replay_anchor_remains_in_successor_snapshot() {
-        let retained = intent("accepted-replay", "object", 0);
-        let mut runtime = MrfRuntime {
-            queue: MrfQueue::new(8, 8192),
-            retained_replay_intents: vec![retained.clone()],
-            config: MrfConsumerConfig::default(),
-            new_since_flush: 0,
-            dirty: false,
-            journal_on_disk: true,
-            backoff_until: None,
-        };
-        assert_eq!(
-            runtime.queue.try_push_typed(intent("new-pending", "object", 0)),
-            MrfQueuePushResult::Enqueued
-        );
-
-        let (authoritative, legacy) = runtime.snapshot();
-        let (decoded, truncated) = decode_journal(&authoritative);
-        let (legacy_decoded, legacy_truncated) = decode_journal(&legacy);
-
-        assert_eq!(truncated, 0);
-        assert_eq!(legacy_truncated, 0);
-        assert_eq!(decoded.len(), 2);
-        assert_eq!(legacy_decoded.len(), 2);
-        assert!(
-            decoded.iter().any(|intent| intent.bucket == retained.bucket),
-            "accepted replay anchor must remain crash-replayable"
+            !replay_must_retain_journal(false, 0),
+            "only a fully consumed replay snapshot may be deleted"
         );
     }
 

rustfs/src/error.rs

@@ -496,9 +496,9 @@ impl From<StorageError> for ApiError {
 
         let message = if matches!(&err, StorageError::QuotaExceeded { .. }) {
             err.to_string()
-        } else if matches!(&err, StorageError::MaxVersionsExceeded) {
-            ApiError::error_code_to_message(&code)
-        } else if code == S3ErrorCode::InternalError && matches!(&err, StorageError::Io(_)) {
+        } else if matches!(&err, StorageError::MaxVersionsExceeded)
+            || (code == S3ErrorCode::InternalError && matches!(&err, StorageError::Io(_)))
+        {
             ApiError::error_code_to_message(&code)
         } else if code == S3ErrorCode::InternalError {
             err.to_string()

rustfs/src/startup_runtime.rs

@@ -17,7 +17,7 @@ use crate::{
     startup_runtime_hooks::{init_profiling_runtime, install_default_crypto_provider, log_startup_runtime_diagnostics},
     startup_tls_material::init_outbound_tls_material,
 };
-use rustfs_config::ENV_API_OBJECT_MAX_VERSIONS;
+use rustfs_config::{DEFAULT_API_OBJECT_MAX_VERSIONS, ENV_API_OBJECT_MAX_VERSIONS};
 use rustfs_utils::EnvParseOutcome;
 use std::io::{Error, Result};
 
@@ -32,13 +32,8 @@ pub(crate) async fn init_startup_runtime_foundation(config: &Config) -> Result<(
 
 fn init_object_max_versions_config() -> Result<()> {
     let limit = match rustfs_utils::get_env_parse_outcome::<u64>(ENV_API_OBJECT_MAX_VERSIONS) {
-        EnvParseOutcome::Absent => rustfs_filemeta::DEFAULT_OBJECT_MAX_VERSIONS,
-        EnvParseOutcome::Invalid => {
-            return Err(Error::other(format!(
-                "{ENV_API_OBJECT_MAX_VERSIONS} must be a positive integer no greater than {}",
-                usize::MAX
-            )));
-        }
+        EnvParseOutcome::Absent => DEFAULT_API_OBJECT_MAX_VERSIONS,
+        EnvParseOutcome::Invalid => return Err(object_max_versions_config_error()),
         EnvParseOutcome::Parsed(value) => object_max_versions_limit_from_u64(value)?,
     };
 
@@ -47,18 +42,20 @@ fn init_object_max_versions_config() -> Result<()> {
 
 fn object_max_versions_limit_from_u64(value: u64) -> Result<usize> {
     if value == 0 {
-        return Err(Error::other(format!(
-            "{ENV_API_OBJECT_MAX_VERSIONS} must be a positive integer no greater than {}",
-            usize::MAX
-        )));
+        return Err(object_max_versions_config_error());
     }
 
-    usize::try_from(value).map_err(|_| {
-        Error::other(format!(
-            "{ENV_API_OBJECT_MAX_VERSIONS} must be a positive integer no greater than {}",
-            usize::MAX
-        ))
-    })
+    let limit = usize::try_from(value).map_err(|_| object_max_versions_config_error())?;
+    if limit > DEFAULT_API_OBJECT_MAX_VERSIONS {
+        return Err(object_max_versions_config_error());
+    }
+    Ok(limit)
+}
+
+fn object_max_versions_config_error() -> Error {
+    Error::other(format!(
+        "{ENV_API_OBJECT_MAX_VERSIONS} must be a positive integer no greater than {DEFAULT_API_OBJECT_MAX_VERSIONS}"
+    ))
 }
 
 #[cfg(test)]