Merge branch 'main' into hs01-mrf-wiring

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

Cargo.lock

@@ -9280,6 +9280,7 @@ dependencies = [
  "s3s",
  "serde",
  "serde_json",
+ "smallvec",
  "tokio",
  "tonic",
  "tracing",
@@ -10251,6 +10252,7 @@ dependencies = [
  "rustfs-ecstore",
  "rustfs-filemeta",
  "rustfs-lock",
+ "rustfs-s3-types",
  "rustfs-storage-api",
  "rustfs-utils",
  "s3s",

crates/common/Cargo.toml

@@ -42,6 +42,7 @@ chrono = { workspace = true, features = ["serde"] }
 jiff = { workspace = true, features = ["serde"] }
 metrics = { workspace = true }
 serde = { workspace = true, features = ["derive"] }
+smallvec = { workspace = true }
 rmp-serde = { workspace = true }
 s3s = { workspace = true, features = ["minio"] }
 tracing = { workspace = true }

crates/common/src/heal_channel.rs

@@ -287,6 +287,9 @@ pub enum HealRequestSource {
     Scanner,
     AutoHeal,
     ReadRepair,
+    /// Mission Repair Feed: intents delivered by error paths and replayed
+    /// from the durable MRF journal.
+    Mrf,
 }
 
 impl HealRequestSource {
@@ -297,6 +300,7 @@ impl HealRequestSource {
             Self::Scanner => "scanner",
             Self::AutoHeal => "auto_heal",
             Self::ReadRepair => "read_repair",
+            Self::Mrf => "mrf",
         }
     }
 }

crates/common/src/lib.rs

@@ -17,8 +17,10 @@ pub mod globals;
 pub mod heal_channel;
 pub mod last_minute;
 pub mod metrics;
+pub mod mrf_channel;
 mod readiness;
 pub mod table_catalog;
+pub mod trace_bus;
 
 pub use globals::*;
 pub use readiness::{GlobalReadiness, SystemStage};

crates/common/src/mrf_channel.rs

@@ -0,0 +1,203 @@
+// 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.
+
+//! Mission Repair Feed (MRF) intent channel.
+//!
+//! Producers on error paths (read decode failure, scanner metadata
+//! corruption, partial-write recovery) hand a lightweight [`MrfIntent`] to the
+//! heal crate through a global bounded channel. Delivery is strictly
+//! non-blocking: `try_send_mrf_intent` never awaits and drops the intent
+//! (counting it) when the channel is full or uninitialized — losing one heal
+//! hint is always preferred over stalling an IO path. Durable replay of
+//! unconsumed intents is the consumer's job (see `rustfs-heal`
+//! `heal::mrf_queue`), mirroring MinIO's `.heal/mrf/list.bin`.
+
+use std::sync::{
+    Arc, OnceLock,
+    atomic::{AtomicBool, Ordering},
+};
+use tokio::sync::mpsc;
+use uuid::Uuid;
+
+/// Bounded capacity of the global MRF channel. Backpressure is resolved by
+/// dropping (and counting) intents, never by blocking the producer.
+const MRF_CHANNEL_CAPACITY: usize = 8192;
+
+/// Why an intent was produced. Drives the heal priority mapping on the
+/// consumer side (DecodeFailure -> Urgent, MetadataCorruption -> High,
+/// PartialWrite -> Normal).
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum MrfKind {
+    /// Erasure decode failed while serving a read (read path).
+    DecodeFailure,
+    /// Scanner classified object metadata as corrupt.
+    MetadataCorruption,
+    /// A write left the object with fewer committed shards than the set size.
+    PartialWrite,
+}
+
+impl MrfKind {
+    pub const fn as_str(self) -> &'static str {
+        match self {
+            MrfKind::DecodeFailure => "decode-failure",
+            MrfKind::MetadataCorruption => "metadata-corruption",
+            MrfKind::PartialWrite => "partial-write",
+        }
+    }
+}
+
+/// One repair intent. Kept deliberately small so the in-memory queue and the
+/// journal stay bounded; `bucket`/`object` are `Arc<str>` so re-arming an
+/// intent never re-allocates the strings.
+#[derive(Clone, Debug)]
+pub struct MrfIntent {
+    pub bucket: Arc<str>,
+    pub object: Arc<str>,
+    /// Version the intent targets, as raw UUID bytes.
+    pub version_id: Option<[u8; 16]>,
+    pub kind: MrfKind,
+    pub enqueued_at_ms: u64,
+    /// Times this intent has already been offered to the heal manager.
+    /// Dropped by the consumer once it reaches `MRF_MAX_ATTEMPTS`.
+    pub attempts: u8,
+}
+
+/// Consumer-side retry ceiling before an intent is given up on.
+pub const MRF_MAX_ATTEMPTS: u8 = 3;
+
+impl MrfIntent {
+    /// Rough in-memory footprint used by the queue's byte budget.
+    pub fn estimated_bytes(&self) -> usize {
+        // Struct + strings + version bytes; buckets and objects are usually
+        // far below this bound, so rounding up keeps the budget conservative.
+        64 + self.bucket.len() + self.object.len()
+    }
+}
+
+static GLOBAL_MRF_SENDER: OnceLock<mpsc::Sender<MrfIntent>> = OnceLock::new();
+
+/// Delivery kill-switch, set from `RUSTFS_HEAL_MRF_ENABLE`. Producers check
+/// this before touching the channel so the disabled path stays allocation- and
+/// sync-free.
+static MRF_DELIVERY_ENABLED: AtomicBool = AtomicBool::new(true);
+
+/// Override delivery (used at heal-runtime startup from configuration).
+pub fn set_mrf_delivery_enabled(enabled: bool) {
+    MRF_DELIVERY_ENABLED.store(enabled, Ordering::Relaxed);
+}
+
+/// Whether producers currently deliver intents.
+pub fn mrf_delivery_enabled() -> bool {
+    MRF_DELIVERY_ENABLED.load(Ordering::Relaxed)
+}
+
+/// Create the global MRF channel and return the consumer half. Fails if the
+/// channel is already initialized (the heal runtime is a singleton).
+pub fn init_mrf_channel() -> Result<mpsc::Receiver<MrfIntent>, &'static str> {
+    let (sender, receiver) = mpsc::channel(MRF_CHANNEL_CAPACITY);
+    GLOBAL_MRF_SENDER
+        .set(sender)
+        .map_err(|_| "MRF channel sender already initialized")?;
+    Ok(receiver)
+}
+
+/// Best-effort, non-blocking intent delivery from an error path.
+///
+/// Returns `true` when the intent was accepted into the channel. `false`
+/// means the intent was dropped (feature disabled, channel not yet
+/// initialized, or channel full) — callers must not retry or await; the
+/// existing read-repair / scanner heal paths remain the safety net.
+///
+/// This runs on IO error paths, so it stays synchronous and cheap: one
+/// bounded allocation for the two `Arc<str>` handles plus the channel slot.
+pub fn try_send_mrf_intent(kind: MrfKind, bucket: &str, object: &str, version_id: Option<Uuid>) -> bool {
+    if !mrf_delivery_enabled() {
+        return false;
+    }
+    let Some(sender) = GLOBAL_MRF_SENDER.get() else {
+        return false;
+    };
+    let intent = MrfIntent {
+        bucket: Arc::from(bucket),
+        object: Arc::from(object),
+        version_id: version_id.map(|vid| *vid.as_bytes()),
+        kind,
+        enqueued_at_ms: unix_now_ms(),
+        attempts: 0,
+    };
+    sender.try_send(intent).is_ok()
+}
+
+fn unix_now_ms() -> u64 {
+    // Kept trivial: the timestamp is diagnostic metadata only; wall-clock
+    // failure would be a bug rather than something to handle here.
+    std::time::SystemTime::now()
+        .duration_since(std::time::UNIX_EPOCH)
+        .map(|d| d.as_millis() as u64)
+        .unwrap_or(0)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn intents_estimate_is_conservative() {
+        let intent = MrfIntent {
+            bucket: Arc::from("bucket"),
+            object: Arc::from("object"),
+            version_id: Some([0u8; 16]),
+            kind: MrfKind::DecodeFailure,
+            enqueued_at_ms: 0,
+            attempts: 0,
+        };
+        assert!(intent.estimated_bytes() >= intent.bucket.len() + intent.object.len());
+    }
+
+    #[tokio::test]
+    async fn try_send_delivers_and_respects_capacity() {
+        let mut receiver = init_mrf_channel().expect("first initialization should succeed");
+        assert!(init_mrf_channel().is_err(), "double initialization must fail");
+
+        assert!(try_send_mrf_intent(MrfKind::DecodeFailure, "b", "o", Some(Uuid::nil())));
+        let intent = receiver.recv().await.expect("intent should arrive");
+        assert_eq!(intent.kind, MrfKind::DecodeFailure);
+        assert_eq!(intent.bucket.as_ref(), "b");
+
+        // Disable delivery: producers become no-ops.
+        set_mrf_delivery_enabled(false);
+        assert!(!try_send_mrf_intent(MrfKind::PartialWrite, "b", "o", None));
+        set_mrf_delivery_enabled(true);
+
+        // Fill the bounded channel past capacity: excess intents are dropped,
+        // never blocking.
+        let mut accepted = 0;
+        for _ in 0..(MRF_CHANNEL_CAPACITY + 64) {
+            if try_send_mrf_intent(MrfKind::PartialWrite, "b", "o", None) {
+                accepted += 1;
+            }
+        }
+        assert_eq!(accepted, MRF_CHANNEL_CAPACITY);
+    }
+
+    #[test]
+    fn try_send_without_channel_is_false() {
+        // This test may run after the tokio test above in the same process;
+        // the singleton semantics make a clean "uninitialized" case hard, so
+        // assert the flag-off behavior only.
+        set_mrf_delivery_enabled(false);
+        assert!(!try_send_mrf_intent(MrfKind::MetadataCorruption, "b", "o", None));
+        set_mrf_delivery_enabled(true);
+    }
+}

crates/common/src/trace_bus.rs

@@ -0,0 +1,333 @@
+// 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.
+
+use smallvec::SmallVec;
+use std::{
+    sync::{
+        Arc, OnceLock,
+        atomic::{AtomicUsize, Ordering},
+    },
+    time::{Duration, SystemTime},
+};
+use tokio::sync::broadcast;
+
+const DEFAULT_TRACE_BUS_CAPACITY: usize = 1024;
+const TRACE_ATTR_INLINE_CAPACITY: usize = 8;
+
+static GLOBAL_TRACE_BUS: OnceLock<TraceBus> = OnceLock::new();
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum TraceKind {
+    Heal,
+    Scanner,
+}
+
+impl TraceKind {
+    pub const fn as_str(self) -> &'static str {
+        match self {
+            Self::Heal => "heal",
+            Self::Scanner => "scanner",
+        }
+    }
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum TraceFunc {
+    HealTask,
+    HealBucket,
+    HealObject,
+    HealCheckAbandonedParts,
+    HealErasureSetPage,
+    ScannerFolder,
+    ScannerIlmAction,
+    ScannerHealCandidate,
+    Dropped,
+}
+
+impl TraceFunc {
+    pub const fn as_str(self) -> &'static str {
+        match self {
+            Self::HealTask => "heal.Task",
+            Self::HealBucket => "heal.Bucket",
+            Self::HealObject => "heal.Object",
+            Self::HealCheckAbandonedParts => "heal.CheckAbandonedParts",
+            Self::HealErasureSetPage => "heal.ErasureSetPage",
+            Self::ScannerFolder => "scanner.Folder",
+            Self::ScannerIlmAction => "scanner.IlmAction",
+            Self::ScannerHealCandidate => "scanner.HealCandidate",
+            Self::Dropped => "trace.Dropped",
+        }
+    }
+}
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum TraceVal {
+    Bool(bool),
+    U64(u64),
+    I64(i64),
+    Str(Arc<str>),
+}
+
+impl From<bool> for TraceVal {
+    fn from(value: bool) -> Self {
+        Self::Bool(value)
+    }
+}
+
+impl From<u64> for TraceVal {
+    fn from(value: u64) -> Self {
+        Self::U64(value)
+    }
+}
+
+impl From<i64> for TraceVal {
+    fn from(value: i64) -> Self {
+        Self::I64(value)
+    }
+}
+
+impl From<&str> for TraceVal {
+    fn from(value: &str) -> Self {
+        Self::Str(Arc::from(value))
+    }
+}
+
+impl From<String> for TraceVal {
+    fn from(value: String) -> Self {
+        Self::Str(Arc::from(value))
+    }
+}
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct TraceAttr {
+    pub key: &'static str,
+    pub value: TraceVal,
+}
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct TraceEvent {
+    pub kind: TraceKind,
+    pub func: TraceFunc,
+    pub time: SystemTime,
+    pub bucket: Option<Arc<str>>,
+    pub object: Option<Arc<str>>,
+    pub duration: Duration,
+    pub bytes: u64,
+    pub attrs: SmallVec<[TraceAttr; TRACE_ATTR_INLINE_CAPACITY]>,
+}
+
+impl TraceEvent {
+    pub fn new(kind: TraceKind, func: TraceFunc) -> Self {
+        Self {
+            kind,
+            func,
+            time: SystemTime::now(),
+            bucket: None,
+            object: None,
+            duration: Duration::ZERO,
+            bytes: 0,
+            attrs: SmallVec::new(),
+        }
+    }
+
+    pub fn with_bucket(mut self, bucket: impl Into<Arc<str>>) -> Self {
+        self.bucket = Some(bucket.into());
+        self
+    }
+
+    pub fn with_object(mut self, object: impl Into<Arc<str>>) -> Self {
+        self.object = Some(object.into());
+        self
+    }
+
+    pub fn with_duration(mut self, duration: Duration) -> Self {
+        self.duration = duration;
+        self
+    }
+
+    pub fn with_bytes(mut self, bytes: u64) -> Self {
+        self.bytes = bytes;
+        self
+    }
+
+    pub fn with_attr(mut self, key: &'static str, value: impl Into<TraceVal>) -> Self {
+        self.attrs.push(TraceAttr {
+            key,
+            value: value.into(),
+        });
+        self
+    }
+}
+
+#[derive(Debug)]
+pub struct TraceBus {
+    sender: broadcast::Sender<Arc<TraceEvent>>,
+    subscriber_count: Arc<AtomicUsize>,
+}
+
+impl TraceBus {
+    pub fn new(capacity: usize) -> Self {
+        let capacity = capacity.max(1);
+        let (sender, _receiver) = broadcast::channel(capacity);
+        Self {
+            sender,
+            subscriber_count: Arc::new(AtomicUsize::new(0)),
+        }
+    }
+
+    pub fn subscriber_count(&self) -> usize {
+        self.subscriber_count.load(Ordering::Acquire)
+    }
+
+    pub fn subscribe(&self) -> TraceSubscription {
+        let receiver = self.sender.subscribe();
+        self.subscriber_count.fetch_add(1, Ordering::AcqRel);
+        TraceSubscription {
+            receiver,
+            subscriber_count: Arc::clone(&self.subscriber_count),
+        }
+    }
+
+    pub fn emit(&self, build: impl FnOnce() -> TraceEvent) -> bool {
+        if self.subscriber_count() == 0 {
+            return false;
+        }
+
+        self.sender.send(Arc::new(build())).is_ok()
+    }
+}
+
+impl Default for TraceBus {
+    fn default() -> Self {
+        Self::new(DEFAULT_TRACE_BUS_CAPACITY)
+    }
+}
+
+#[derive(Debug)]
+pub struct TraceSubscription {
+    receiver: broadcast::Receiver<Arc<TraceEvent>>,
+    subscriber_count: Arc<AtomicUsize>,
+}
+
+impl TraceSubscription {
+    pub async fn recv(&mut self) -> Result<Arc<TraceEvent>, broadcast::error::RecvError> {
+        self.receiver.recv().await
+    }
+
+    pub fn try_recv(&mut self) -> Result<Arc<TraceEvent>, broadcast::error::TryRecvError> {
+        self.receiver.try_recv()
+    }
+}
+
+impl Drop for TraceSubscription {
+    fn drop(&mut self) {
+        self.subscriber_count.fetch_sub(1, Ordering::AcqRel);
+    }
+}
+
+pub fn global_trace_bus() -> &'static TraceBus {
+    GLOBAL_TRACE_BUS.get_or_init(TraceBus::default)
+}
+
+pub fn subscribe_trace_events() -> TraceSubscription {
+    global_trace_bus().subscribe()
+}
+
+pub fn trace_emit(build: impl FnOnce() -> TraceEvent) -> bool {
+    global_trace_bus().emit(build)
+}
+
+pub fn trace_subscriber_count() -> usize {
+    global_trace_bus().subscriber_count()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::sync::atomic::AtomicUsize;
+
+    #[test]
+    fn trace_emit_skips_builder_without_subscribers() {
+        let bus = TraceBus::new(4);
+        let built = AtomicUsize::new(0);
+
+        let sent = bus.emit(|| {
+            built.fetch_add(1, Ordering::Relaxed);
+            TraceEvent::new(TraceKind::Heal, TraceFunc::HealTask)
+        });
+
+        assert!(!sent);
+        assert_eq!(built.load(Ordering::Relaxed), 0);
+    }
+
+    #[tokio::test]
+    async fn trace_subscriber_receives_event() {
+        let bus = TraceBus::new(4);
+        let mut subscription = bus.subscribe();
+
+        assert!(bus.emit(|| {
+            TraceEvent::new(TraceKind::Heal, TraceFunc::HealObject)
+                .with_bucket("bucket")
+                .with_object("object")
+                .with_duration(Duration::from_millis(7))
+                .with_bytes(11)
+                .with_attr("dry", true)
+        }));
+
+        let event = subscription
+            .recv()
+            .await
+            .expect("subscriber should receive emitted trace event");
+
+        assert_eq!(event.kind, TraceKind::Heal);
+        assert_eq!(event.func, TraceFunc::HealObject);
+        assert_eq!(event.bucket.as_deref(), Some("bucket"));
+        assert_eq!(event.object.as_deref(), Some("object"));
+        assert_eq!(event.duration, Duration::from_millis(7));
+        assert_eq!(event.bytes, 11);
+        assert_eq!(
+            event.attrs.as_slice(),
+            &[TraceAttr {
+                key: "dry",
+                value: TraceVal::Bool(true)
+            }]
+        );
+    }
+
+    #[test]
+    fn trace_subscription_drop_decrements_count() {
+        let bus = TraceBus::new(4);
+        let subscription = bus.subscribe();
+
+        assert_eq!(bus.subscriber_count(), 1);
+        drop(subscription);
+        assert_eq!(bus.subscriber_count(), 0);
+    }
+
+    #[tokio::test]
+    async fn lagged_subscriber_drops_events_without_blocking_publishers() {
+        let bus = TraceBus::new(2);
+        let mut subscription = bus.subscribe();
+
+        for index in 0_u64..4 {
+            assert!(bus.emit(|| { TraceEvent::new(TraceKind::Scanner, TraceFunc::ScannerFolder).with_attr("index", index) }));
+        }
+
+        let err = subscription
+            .recv()
+            .await
+            .expect_err("receiver should observe lag instead of blocking publishers");
+        assert!(matches!(err, broadcast::error::RecvError::Lagged(_)));
+    }
+}

crates/config/src/constants/heal.rs

@@ -177,3 +177,31 @@ pub const DEFAULT_HEAL_MAINLINE_WRITE_UTILIZATION_HIGH_PERCENT: usize = 80;
 
 /// Default foreground pressure recheck delay for heal scheduler, in milliseconds.
 pub const DEFAULT_HEAL_MAINLINE_MAX_SLEEP_MS: u64 = 250;
+
+/// Environment variable that toggles the MRF (mission repair feed) intent
+/// pipeline: error paths deliver repair intents to the heal runtime, and
+/// unconsumed intents are replayed from the durable journal after a restart.
+pub const ENV_HEAL_MRF_ENABLE: &str = "RUSTFS_HEAL_MRF_ENABLE";
+
+/// Environment variable for the MRF in-memory queue capacity (intent count).
+pub const ENV_HEAL_MRF_QUEUE_SIZE: &str = "RUSTFS_HEAL_MRF_QUEUE_SIZE";
+
+/// Environment variable for the MRF journal byte budget. The journal is
+/// compacted once its on-disk size crosses this bound.
+pub const ENV_HEAL_MRF_JOURNAL_MAX_BYTES: &str = "RUSTFS_HEAL_MRF_JOURNAL_MAX_BYTES";
+
+/// Environment variable for the MRF journal replay batch size (intents per
+/// replay push round).
+pub const ENV_HEAL_MRF_REPLAY_BATCH: &str = "RUSTFS_HEAL_MRF_REPLAY_BATCH";
+
+/// Default behavior keeps the MRF intent pipeline enabled.
+pub const DEFAULT_HEAL_MRF_ENABLE: bool = true;
+
+/// Default MRF queue capacity (matches MinIO's 100k MRF list ceiling).
+pub const DEFAULT_HEAL_MRF_QUEUE_SIZE: usize = 100_000;
+
+/// Default MRF journal byte budget (8 MiB), mirroring the channel payload cap.
+pub const DEFAULT_HEAL_MRF_JOURNAL_MAX_BYTES: usize = 8 * 1024 * 1024;
+
+/// Default MRF replay batch size.
+pub const DEFAULT_HEAL_MRF_REPLAY_BATCH: usize = 256;

crates/data-usage/src/data_usage.rs

@@ -870,6 +870,157 @@ pub struct DataUsageCacheInfo {
     pub snapshot_complete: bool,
 }
 
+/// Prefix-level usage over a raw entry map — the shared core behind
+/// [`DataUsageCache::prefix_usage`], usable by any cache-shaped reader (the
+/// scanner's writer-side cache has the same map type).
+///
+/// Cache keys are cleaned literal paths (`bucket/pre/fix`), so sub-prefix
+/// names come straight off the child keys — no reverse mapping exists or is
+/// needed. A compacted prefix carries its aggregate but no children, which
+/// the `compacted` flag reports so callers can say why the breakdown is
+/// empty. `truncated` is set when the breakdown exceeded `max_entries` and
+/// was cut (largest first).
+pub fn prefix_usage_in_cache(
+    cache: &HashMap<String, DataUsageEntry>,
+    bucket: &str,
+    prefix: &str,
+    max_entries: usize,
+) -> Option<PrefixUsageQuery> {
+    let prefix = prefix.trim_matches('/');
+    let root = if prefix.is_empty() {
+        bucket.to_string()
+    } else {
+        format!("{bucket}/{prefix}")
+    };
+    let entry = cache.get(&hash_path(&root).key())?.clone();
+
+    let usage = PrefixUsageSummary::from_entry(&flatten_entry(cache, &entry, 0)?);
+
+    let child_prefix = format!("{root}/");
+    let mut sub_prefixes: Vec<PrefixUsageEntry> = entry
+        .children
+        .iter()
+        .filter_map(|child_key| {
+            let child = cache.get(child_key)?;
+            let child_flat = flatten_entry(cache, child, 1)?;
+            // Child keys are literal `bucket/pre/name` paths; a trailing
+            // slash marks a directory object and is display-only here.
+            let name = child_key
+                .strip_prefix(child_prefix.as_str())
+                .unwrap_or(child_key.as_str())
+                .trim_end_matches('/')
+                .to_string();
+            Some(PrefixUsageEntry {
+                prefix: name,
+                usage: PrefixUsageSummary::from_entry(&child_flat),
+            })
+        })
+        .collect();
+    sub_prefixes.sort_by(|left, right| {
+        right
+            .usage
+            .size
+            .cmp(&left.usage.size)
+            .then_with(|| left.prefix.cmp(&right.prefix))
+    });
+    let truncated = sub_prefixes.len() > max_entries;
+    sub_prefixes.truncate(max_entries);
+
+    Some(PrefixUsageQuery {
+        usage,
+        compacted: entry.compacted,
+        truncated,
+        sub_prefixes,
+    })
+}
+
+/// Maximum subtree depth [`flatten_entry`] will walk before declaring the
+/// cache corrupt — the same bound the scanner's checked flatten uses.
+const PREFIX_USAGE_MAX_DEPTH: usize = 1024;
+
+/// Flatten one entry's subtree into an aggregate: the free-function twin of
+/// [`DataUsageCache::flatten`], carrying the scanner checked-flatten
+/// hardening so a corrupt cache (cycles, over-deep trees, overflowing
+/// counters) yields `None` instead of unbounded recursion or wrapped totals.
+fn flatten_entry(cache: &HashMap<String, DataUsageEntry>, root: &DataUsageEntry, depth: usize) -> Option<DataUsageEntry> {
+    if depth > PREFIX_USAGE_MAX_DEPTH {
+        return None;
+    }
+    let mut flattened = DataUsageEntry::default();
+    if !flattened.checked_merge(root) {
+        return None;
+    }
+    flattened.compacted = root.compacted;
+    // The root itself is not pre-seeded: it is merged above, and a corrupt
+    // child edge pointing back at the root's own key is still terminated by
+    // the visited set on first encounter.
+    let mut visited: HashSet<&str> = HashSet::new();
+    let mut pending: Vec<(&String, usize)> = root.children.iter().map(|child| (child, depth + 1)).collect();
+    while let Some((key, child_depth)) = pending.pop() {
+        if child_depth > PREFIX_USAGE_MAX_DEPTH || !visited.insert(key.as_str()) {
+            return None;
+        }
+        let entry = cache.get(key)?;
+        if !flattened.checked_merge(entry) {
+            return None;
+        }
+        pending.extend(entry.children.iter().map(|child| (child, child_depth + 1)));
+    }
+    flattened.children.clear();
+    Some(flattened)
+}
+
+/// Flattened counters of one prefix subtree, as returned by
+/// [`DataUsageCache::prefix_usage`].
+#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, serde::Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct PrefixUsageSummary {
+    pub size: u64,
+    pub objects: u64,
+    pub versions: u64,
+    pub delete_markers: u64,
+}
+
+impl PrefixUsageSummary {
+    fn from_entry(entry: &DataUsageEntry) -> Self {
+        Self {
+            size: entry.size as u64,
+            objects: entry.objects as u64,
+            versions: entry.versions as u64,
+            delete_markers: entry.delete_markers as u64,
+        }
+    }
+
+    /// Add another set's counters into this one (entries are partitioned by
+    /// set, so per-set results sum).
+    pub fn merge(&mut self, other: &Self) {
+        self.size = self.size.saturating_add(other.size);
+        self.objects = self.objects.saturating_add(other.objects);
+        self.versions = self.versions.saturating_add(other.versions);
+        self.delete_markers = self.delete_markers.saturating_add(other.delete_markers);
+    }
+}
+
+/// One first-level sub-prefix row of a [`PrefixUsageQuery`].
+#[derive(Clone, Debug, PartialEq, Eq, serde::Serialize)]
+pub struct PrefixUsageEntry {
+    pub prefix: String,
+    pub usage: PrefixUsageSummary,
+}
+
+/// Result of [`DataUsageCache::prefix_usage`].
+#[derive(Clone, Debug, Default, PartialEq, Eq, serde::Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct PrefixUsageQuery {
+    pub usage: PrefixUsageSummary,
+    /// The prefix entry was compacted by the scanner: its aggregate is valid
+    /// but no sub-prefix breakdown exists on disk.
+    pub compacted: bool,
+    /// The breakdown had more entries than `max_entries`; the largest remain.
+    pub truncated: bool,
+    pub sub_prefixes: Vec<PrefixUsageEntry>,
+}
+
 /// Read-only projection of a scanner-written `.usage-cache.bin` file.
 ///
 /// The scanner-side `DataUsageCache` (`crates/scanner/src/data_usage_define.rs`)
@@ -997,6 +1148,21 @@ impl DataUsageCache {
         }
     }
 
+    /// Prefix-level usage for one bucket subtree, plus the one-level
+    /// breakdown below it (rustfs/backlog#1872, MinIO
+    /// `loadPrefixUsageFromBackend` parity and beyond: arbitrary prefixes and
+    /// full counters instead of first-level sizes only).
+    ///
+    /// Cache keys are cleaned literal paths (`bucket/pre/fix`), so sub-prefix
+    /// names come straight off the child keys — no reverse mapping exists or
+    /// is needed. A compacted prefix carries its aggregate but no children,
+    /// which the `compacted` flag reports so callers can say why the
+    /// breakdown is empty. `truncated` is set when the breakdown exceeded
+    /// `max_entries` and was cut (largest first).
+    pub fn prefix_usage(&self, bucket: &str, prefix: &str, max_entries: usize) -> Option<PrefixUsageQuery> {
+        prefix_usage_in_cache(&self.cache, bucket, prefix, max_entries)
+    }
+
     pub fn force_compact(&mut self, limit: usize) {
         if self.cache.len() < limit {
             return;
@@ -1898,6 +2064,126 @@ mod tests {
         );
     }
 
+    /// Build a cache shaped like `bucket/{a,b/{c,d}},bucket/loose` with
+    /// distinct counters so aggregation is observable.
+    fn prefix_usage_fixture_cache() -> DataUsageCache {
+        let mut cache = DataUsageCache::default();
+        let mut insert = |path: &str, parent: &str, size: usize, objects: usize, versions: usize, delete_markers: usize| {
+            cache.replace(
+                path,
+                parent,
+                DataUsageEntry {
+                    size,
+                    objects,
+                    versions,
+                    delete_markers,
+                    ..Default::default()
+                },
+            );
+        };
+        insert("bucket", "", 0, 0, 0, 0);
+        insert("bucket/a", "bucket", 100, 1, 1, 0);
+        insert("bucket/b", "bucket", 0, 0, 0, 0);
+        insert("bucket/b/c", "bucket/b", 200, 2, 2, 1);
+        insert("bucket/b/d", "bucket/b", 40, 1, 3, 0);
+        insert("bucket/loose", "bucket", 10, 1, 1, 1);
+        cache
+    }
+
+    #[test]
+    fn prefix_usage_aggregates_bucket_root_and_one_level_below() {
+        let cache = prefix_usage_fixture_cache();
+
+        let root = cache
+            .prefix_usage("bucket", "", 100)
+            .expect("root query must find the bucket entry");
+        assert_eq!(root.usage.size, 350, "root aggregate flattens the whole subtree");
+        assert_eq!(root.usage.objects, 5);
+        assert_eq!(root.usage.versions, 7);
+        assert_eq!(root.usage.delete_markers, 2);
+        assert!(!root.compacted);
+        assert!(!root.truncated);
+        // Breakdown is one level: b (240) before a (100) before loose (10),
+        // each flattened to its own subtree total.
+        let names: Vec<(&str, u64)> = root
+            .sub_prefixes
+            .iter()
+            .map(|entry| (entry.prefix.as_str(), entry.usage.size))
+            .collect();
+        assert_eq!(names, vec![("b", 240), ("a", 100), ("loose", 10)]);
+    }
+
+    #[test]
+    fn prefix_usage_drills_into_arbitrary_prefixes() {
+        let cache = prefix_usage_fixture_cache();
+
+        let b = cache.prefix_usage("bucket", "b", 100).expect("nested prefix must resolve");
+        assert_eq!(b.usage.size, 240);
+        assert_eq!(b.usage.versions, 5);
+        let names: Vec<&str> = b.sub_prefixes.iter().map(|entry| entry.prefix.as_str()).collect();
+        assert_eq!(names, vec!["c", "d"]);
+
+        // Prefix slashes are normalized away.
+        let slashed = cache.prefix_usage("bucket", "/b/", 100).expect("slash-insensitive lookup");
+        assert_eq!(slashed.usage.size, 240);
+
+        assert!(cache.prefix_usage("bucket", "absent", 100).is_none(), "unknown prefix must be a miss");
+        assert!(cache.prefix_usage("other", "", 100).is_none(), "unknown bucket must be a miss");
+    }
+
+    #[test]
+    fn prefix_usage_reports_and_respects_truncation() {
+        let cache = prefix_usage_fixture_cache();
+        let capped = cache.prefix_usage("bucket", "", 2).expect("root query");
+        assert!(capped.truncated, "three children capped to two must flag truncation");
+        let names: Vec<&str> = capped.sub_prefixes.iter().map(|entry| entry.prefix.as_str()).collect();
+        assert_eq!(names, vec!["b", "a"], "largest prefixes survive the cut");
+    }
+
+    #[test]
+    fn prefix_usage_marks_compacted_entries() {
+        let mut cache = DataUsageCache::default();
+        cache.replace(
+            "bucket",
+            "",
+            DataUsageEntry {
+                size: 999,
+                objects: 9,
+                compacted: true,
+                ..Default::default()
+            },
+        );
+
+        let compacted = cache.prefix_usage("bucket", "", 100).expect("compacted root resolves");
+        assert!(compacted.compacted, "compaction must be visible to callers");
+        assert_eq!(compacted.usage.size, 999);
+        assert!(compacted.sub_prefixes.is_empty(), "a compacted entry carries no children");
+    }
+
+    #[test]
+    fn prefix_usage_rejects_cyclic_and_dangling_caches() {
+        // A self-referencing child (corrupt cache) must yield a miss for the
+        // whole query, not unbounded recursion.
+        let mut cache = prefix_usage_fixture_cache();
+        if let Some(entry) = cache.cache.get_mut("bucket/b") {
+            entry.children.insert("bucket/b".to_string());
+        }
+        assert!(cache.prefix_usage("bucket", "b", 100).is_none(), "a cyclic subtree must be rejected");
+        // The unaffected sibling still answers.
+        assert!(cache.prefix_usage("bucket", "a", 100).is_some());
+
+        // A child key with no entry (dangling link) is rejected rather than
+        // silently dropped: half a tree would under-report usage.
+        let mut dangling = prefix_usage_fixture_cache();
+        if let Some(entry) = dangling.cache.get_mut("bucket/b") {
+            entry.children.insert("bucket/b/ghost".to_string());
+        }
+        assert!(
+            dangling.prefix_usage("bucket", "b", 100).is_none(),
+            "a dangling child link must be rejected"
+        );
+    }
+
     #[test]
     fn hash_path_uses_portable_slash_semantics() {
         for (input, expected) in [

crates/e2e_test/src/common.rs

@@ -32,6 +32,7 @@ use rustfs_signer::sign_v4;
 use s3s::Body;
 use std::ffi::OsStr;
 use std::fs as stdfs;
+use std::io::ErrorKind;
 use std::path::{Path, PathBuf};
 use std::process::{Child, Command, Stdio};
 use std::sync::Once;
@@ -51,6 +52,11 @@ pub(crate) const FAST_DATA_USAGE_SCANNER_ENV: &[(&str, &str)] =
     &[("RUSTFS_SCANNER_CYCLE", "1"), ("RUSTFS_SCANNER_START_DELAY_SECS", "0")];
 pub const TEST_BUCKET: &str = "e2e-test-bucket";
 const RUSTFS_FULL_FEATURE: &str = "full";
+const TEST_PORT_MIN: u16 = 20_000;
+const TEST_PORT_RANGE: u16 = 40_000;
+const TEST_PORT_COUNTER_PATH: &str = "/tmp/rustfs_e2e_next_port";
+const TEST_PORT_LOCK_DIR: &str = "/tmp/rustfs_e2e_port_allocator.lock";
+const TEST_PORT_LOCK_STALE_AFTER: Duration = Duration::from_secs(30);
 
 fn capture_log_path(log_dir: &Path, temp_dir: &str) -> Option<PathBuf> {
     let temp_name = Path::new(temp_dir).file_name()?.to_string_lossy();
@@ -67,6 +73,64 @@ fn configured_capture_log_path(temp_dir: &str) -> Option<String> {
     capture_log_path(Path::new(&log_dir), temp_dir).map(|path| path.to_string_lossy().into_owned())
 }
 
+struct PortAllocatorGuard;
+
+impl PortAllocatorGuard {
+    async fn acquire() -> Result<Self, Box<dyn std::error::Error + Send + Sync>> {
+        loop {
+            match stdfs::create_dir(TEST_PORT_LOCK_DIR) {
+                Ok(()) => return Ok(Self),
+                Err(err) if err.kind() == ErrorKind::AlreadyExists => {
+                    remove_stale_port_allocator_lock();
+                    sleep(Duration::from_millis(10)).await;
+                }
+                Err(err) => return Err(err.into()),
+            }
+        }
+    }
+}
+
+impl Drop for PortAllocatorGuard {
+    fn drop(&mut self) {
+        let _ = stdfs::remove_dir(TEST_PORT_LOCK_DIR);
+    }
+}
+
+fn advance_test_port(port: u16) -> u16 {
+    let offset = (port - TEST_PORT_MIN + 1) % TEST_PORT_RANGE;
+    TEST_PORT_MIN + offset
+}
+
+fn seeded_test_port() -> u16 {
+    let offset = (Uuid::new_v4().as_u128() % u128::from(TEST_PORT_RANGE)) as u16;
+    TEST_PORT_MIN + offset
+}
+
+fn read_next_test_port() -> u16 {
+    stdfs::read_to_string(TEST_PORT_COUNTER_PATH)
+        .ok()
+        .and_then(|value| value.trim().parse::<u16>().ok())
+        .filter(|port| (TEST_PORT_MIN..TEST_PORT_MIN + TEST_PORT_RANGE).contains(port))
+        .unwrap_or_else(seeded_test_port)
+}
+
+fn remove_stale_port_allocator_lock() {
+    let Ok(metadata) = stdfs::metadata(TEST_PORT_LOCK_DIR) else {
+        return;
+    };
+    let Ok(modified) = metadata.modified() else {
+        return;
+    };
+    if modified.elapsed().is_ok_and(|elapsed| elapsed > TEST_PORT_LOCK_STALE_AFTER) {
+        let _ = stdfs::remove_dir(TEST_PORT_LOCK_DIR);
+    }
+}
+
+fn write_next_test_port(port: u16) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    stdfs::write(TEST_PORT_COUNTER_PATH, port.to_string())?;
+    Ok(())
+}
+
 pub(crate) fn capture_command_logs(
     command: &mut Command,
     log_path: Option<&str>,
@@ -508,10 +572,21 @@ impl RustFSTestEnvironment {
     /// Find an available port for the test
     pub async fn find_available_port() -> Result<u16, Box<dyn std::error::Error + Send + Sync>> {
         use std::net::TcpListener;
-        let listener = TcpListener::bind("127.0.0.1:0")?;
-        let port = listener.local_addr()?.port();
-        drop(listener);
-        Ok(port)
+        let _guard = PortAllocatorGuard::acquire().await?;
+        let mut next_port = read_next_test_port();
+
+        for _ in 0..TEST_PORT_RANGE {
+            let port = next_port;
+            next_port = advance_test_port(next_port);
+            write_next_test_port(next_port)?;
+
+            if let Ok(listener) = TcpListener::bind(("127.0.0.1", port)) {
+                drop(listener);
+                return Ok(port);
+            }
+        }
+
+        Err("no available E2E test port found".into())
     }
 
     /// Kill any existing RustFS processes

crates/ecstore/src/api/mod.rs

@@ -380,7 +380,7 @@ pub mod erasure {
 
 pub mod event {
     pub use crate::event::name::EventName;
-    pub use crate::services::event_notification::{EventArgs, register_event_dispatch_hook};
+    pub use crate::services::event_notification::{EventArgs, register_event_dispatch_hook, send_event};
 }
 
 pub mod global {
@@ -483,6 +483,7 @@ pub mod store_list {
 }
 
 pub mod storage {
+    pub use crate::core::pools::HealLifecycleExpiryContext;
     pub use crate::store::HealWalkVersion;
     pub use crate::store::{
         ECStore, all_local_disk, all_local_disk_path, find_local_disk_by_ref, init_local_disks,

crates/ecstore/src/bucket/bucket_target_sys.rs

@@ -1549,8 +1549,8 @@ impl Default for PutObjectOptions {
     }
 }
 
-#[allow(dead_code)]
 impl PutObjectOptions {
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn set_match_etag(&mut self, etag: &str) {
         if etag == "*" {
             self.custom_header
@@ -1561,6 +1561,7 @@ impl PutObjectOptions {
         }
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn set_match_etag_except(&mut self, etag: &str) {
         if etag == "*" {
             self.custom_header
@@ -1696,6 +1697,7 @@ impl PutObjectOptions {
         header
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn validate(&self, _c: Arc<TargetClient>) -> Result<(), std::io::Error> {
         //if self.checksum.is_set() {
         /*if !self.trailing_header_support {

crates/ecstore/src/bucket/lifecycle/bucket_lifecycle_ops.rs

@@ -456,16 +456,23 @@ impl<'a> LifecycleExpiryTrace<'a> {
     }
 }
 
-#[allow(dead_code)]
 impl ExpiryStats {
     pub fn missed_tasks(&self) -> i64 {
         self.missed_expiry_tasks.load(Ordering::SeqCst)
     }
 
+    #[allow(
+        dead_code,
+        reason = "asserted by this file's tests; the lib target cannot see test-only consumers (backlog#1823)"
+    )]
     fn missed_free_vers_tasks(&self) -> i64 {
         self.missed_freevers_tasks.load(Ordering::SeqCst)
     }
 
+    #[allow(
+        dead_code,
+        reason = "asserted by this file's tests; the lib target cannot see test-only consumers (backlog#1823)"
+    )]
     fn missed_tier_journal_tasks(&self) -> i64 {
         self.missed_tier_journal_tasks.load(Ordering::SeqCst)
     }

crates/ecstore/src/bucket/lifecycle/mod.rs

@@ -19,7 +19,7 @@ pub mod core;
 pub mod evaluator;
 pub mod manual_transition_job;
 mod metadata_boundary;
-pub(crate) use metadata_boundary::get_expiry_configs;
+pub(crate) use metadata_boundary::{LifecycleExpiryConfigs, get_expiry_configs};
 mod object_lock_boundary;
 pub use self::core as lifecycle;
 mod replication_sink;

crates/ecstore/src/bucket/lifecycle/tier_last_day_stats.rs

@@ -80,7 +80,10 @@ impl LastDayTierStats {
         }
     }
 
-    #[allow(dead_code)]
+    #[allow(
+        dead_code,
+        reason = "asserted by this file's tests; the lib target cannot see test-only consumers (backlog#1823)"
+    )]
     fn merge(&self, m: LastDayTierStats) -> LastDayTierStats {
         let mut cl = self.clone();
         let mut cm = m;

crates/ecstore/src/bucket/lifecycle/tier_sweeper.rs

@@ -177,9 +177,10 @@ fn should_record_remote_delete_failure(err: &std::io::Error) -> bool {
 }
 
 #[derive(Default)]
-#[allow(dead_code)]
 struct ObjSweeper {
+    #[allow(dead_code, reason = "written but never read back (backlog#1823)")]
     object: String,
+    #[allow(dead_code, reason = "written but never read back (backlog#1823)")]
     bucket: String,
     version_id: Option<Uuid>,
     versioned: bool,
@@ -191,9 +192,9 @@ struct ObjSweeper {
     remote_object: String,
 }
 
-#[allow(dead_code)]
 impl ObjSweeper {
     #[allow(clippy::new_ret_no_self)]
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     pub async fn new(bucket: &str, object: &str) -> Result<Self, std::io::Error> {
         Ok(Self {
             object: object.into(),
@@ -202,17 +203,20 @@ impl ObjSweeper {
         })
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     pub fn with_version(&mut self, vid: Option<Uuid>) -> &Self {
         self.version_id = vid.clone();
         self
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     pub fn with_versioning(&mut self, versioned: bool, suspended: bool) -> &Self {
         self.versioned = versioned;
         self.suspended = suspended;
         self
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     pub fn get_opts(&self) -> lifecycle::ObjectOpts {
         let mut opts = ObjectOpts {
             version_id: self.version_id.clone(),
@@ -226,6 +230,7 @@ impl ObjSweeper {
         opts
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     pub fn set_transition_state(&mut self, info: TransitionedObject) {
         self.transition_tier = info.tier;
         self.transition_status = info.status;
@@ -266,6 +271,7 @@ impl ObjSweeper {
         None
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     pub async fn sweep(&self, api: Arc<ECStore>) {
         let Some(je) = self.should_remove_remote_object() else {
             return;

crates/ecstore/src/bucket/quota/mod.rs

@@ -312,9 +312,7 @@ mod tests {
         }
         #[derive(Deserialize)]
         struct LegacyBucketQuota {
-            #[allow(dead_code)]
             quota: Option<u64>,
-            #[allow(dead_code)]
             quota_type: LegacyQuotaType,
         }
         let legacy = serde_json::from_slice::<LegacyBucketQuota>(&json)

crates/ecstore/src/client/api_get_object.rs

@@ -95,7 +95,6 @@ impl TransitionClient {
 }
 
 #[derive(Default)]
-#[allow(dead_code)]
 pub struct GetRequest {
     pub buffer: Vec<u8>,
     pub offset: i64,
@@ -107,11 +106,12 @@ pub struct GetRequest {
     pub setting_object_info: bool,
 }
 
-#[allow(dead_code)]
 pub struct GetResponse {
     pub size: i64,
     //pub error:       error,
+    #[allow(dead_code, reason = "written but never read back (backlog#1823)")]
     pub did_read: bool,
+    #[allow(dead_code, reason = "written but never read back (backlog#1823)")]
     pub object_info: ObjectInfo,
 }
 

crates/ecstore/src/client/api_get_options.rs

@@ -27,7 +27,6 @@ use tracing::warn;
 use crate::client::api_error_response::err_invalid_argument;
 
 #[derive(Default)]
-#[allow(dead_code)]
 pub struct AdvancedGetOptions {
     pub replication_delete_marker: bool,
     pub is_replication_ready_for_delete_marker: bool,

crates/ecstore/src/client/api_list.rs

@@ -360,7 +360,6 @@ impl TransitionClient {
 }
 
 #[derive(Default)]
-#[allow(dead_code)]
 pub struct ListObjectsOptions {
     reverse_versions: bool,
     with_versions: bool,

crates/ecstore/src/client/api_put_object.rs

@@ -137,8 +137,8 @@ impl Default for PutObjectOptions {
     }
 }
 
-#[allow(dead_code)]
 impl PutObjectOptions {
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn set_match_etag(&mut self, etag: &str) {
         if etag == "*" {
             self.custom_header.insert("If-Match", HeaderValue::from_static("*"));
@@ -149,6 +149,7 @@ impl PutObjectOptions {
         }
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn set_match_etag_except(&mut self, etag: &str) {
         if etag == "*" {
             self.custom_header.insert("If-None-Match", HeaderValue::from_static("*"));
@@ -259,6 +260,7 @@ impl PutObjectOptions {
         header
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn validate(&self, c: TransitionClient) -> Result<(), std::io::Error> {
         //if self.checksum.is_set() {
         /*if !self.trailing_header_support {

crates/ecstore/src/client/api_remove.rs

@@ -55,7 +55,6 @@ pub struct RemoveBucketOptions {
 const DELETE_RESPONSE_PREVIEW_LEN: usize = 1024;
 
 #[derive(Debug)]
-#[allow(dead_code)]
 pub struct AdvancedRemoveOptions {
     pub replication_delete_marker: bool,
     pub replication_status: ReplicationStatus,
@@ -465,10 +464,10 @@ impl TransitionClient {
 }
 
 #[derive(Debug, Default)]
-#[allow(dead_code)]
 pub struct RemoveObjectError {
+    #[allow(dead_code, reason = "written but never read back (backlog#1823)")]
     object_name: String,
-    #[allow(dead_code)]
+    #[allow(dead_code, reason = "written but never read back (backlog#1823)")]
     version_id: String,
     err: Option<std::io::Error>,
 }

crates/ecstore/src/client/checksum.rs

@@ -372,8 +372,8 @@ pub struct Checksum {
     computed: bool,
 }
 
-#[allow(dead_code)]
 impl Checksum {
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn new(t: ChecksumMode, b: &[u8]) -> Checksum {
         if t.is_set() && b.len() == t.raw_byte_len() {
             return Checksum {
@@ -385,7 +385,7 @@ impl Checksum {
         Checksum::default()
     }
 
-    #[allow(dead_code)]
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn new_checksum_string(t: ChecksumMode, s: &str) -> Result<Checksum, std::io::Error> {
         let b = match base64_decode(s.as_bytes()) {
             Ok(b) => b,
@@ -412,7 +412,7 @@ impl Checksum {
         base64_encode(&self.r)
     }
 
-    #[allow(dead_code)]
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn raw(&self) -> Option<Vec<u8>> {
         if !self.is_set() {
             return None;

crates/ecstore/src/client/object_api_utils.rs

@@ -37,16 +37,17 @@ pub struct PutObjReader {
     //pub sealMD5Fn: SealMD5CurrFn,
 }
 
-#[allow(dead_code)]
 impl PutObjReader {
     pub fn new(reader: HashReader) -> Self {
         Self { reader }
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn md5_current_hex_string(&self) -> String {
         self.reader.checksum().map(|v| v.encoded).unwrap_or_default()
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn with_encryption(&mut self, enc_reader: HashReader) -> Result<(), std::io::Error> {
         self.reader = enc_reader;
 

crates/ecstore/src/cluster/rpc/peer_s3_client.rs

@@ -214,6 +214,19 @@ fn pool_write_quorum(participant_count: usize) -> usize {
     (participant_count / 2) + 1
 }
 
+/// Error for a peer that reported `success = false` without an error payload.
+///
+/// The message must stay identical across the peers of one operation: `reduce_errs`
+/// buckets `Error::Io` by kind plus rendered message, so any per-peer detail (address,
+/// timing) would split one shared failure into single-count buckets and downgrade a real
+/// dominant error into `ErasureWriteQuorum`.
+fn peer_failure_without_details(op: &str, bucket: Option<&str>) -> Error {
+    match bucket {
+        Some(bucket) => Error::other(format!("{op}({bucket}): peer returned failure without error details")),
+        None => Error::other(format!("{op}: peer returned failure without error details")),
+    }
+}
+
 fn reduce_pool_write_quorum_errs(per_pool_errs: &[Option<Error>]) -> Option<Error> {
     if per_pool_errs.is_empty() {
         return Some(Error::ErasureWriteQuorum);
@@ -1078,7 +1091,7 @@ impl PeerS3Client for RemotePeerS3Client {
                     return if let Some(err) = response.error {
                         Err(err.into())
                     } else {
-                        Err(Error::other(""))
+                        Err(peer_failure_without_details("heal_bucket", Some(bucket)))
                     };
                 }
 
@@ -1105,7 +1118,7 @@ impl PeerS3Client for RemotePeerS3Client {
                     return if let Some(err) = response.error {
                         Err(err.into())
                     } else {
-                        Err(Error::other(""))
+                        Err(peer_failure_without_details("list_bucket", None))
                     };
                 }
                 let bucket_infos = response
@@ -1136,9 +1149,7 @@ impl PeerS3Client for RemotePeerS3Client {
                     return if let Some(err) = response.error {
                         Err(err.into())
                     } else {
-                        Err(Error::other(format!(
-                            "make_bucket({bucket}): peer returned failure without error details"
-                        )))
+                        Err(peer_failure_without_details("make_bucket", Some(bucket)))
                     };
                 }
 
@@ -1162,7 +1173,7 @@ impl PeerS3Client for RemotePeerS3Client {
                     return if let Some(err) = response.error {
                         Err(err.into())
                     } else {
-                        Err(Error::other(""))
+                        Err(peer_failure_without_details("get_bucket_info", Some(bucket)))
                     };
                 }
                 let bucket_info = serde_json::from_str::<BucketInfo>(&response.bucket_info)?;
@@ -1190,7 +1201,7 @@ impl PeerS3Client for RemotePeerS3Client {
                     return if let Some(err) = response.error {
                         Err(err.into())
                     } else {
-                        Err(Error::other(""))
+                        Err(peer_failure_without_details("delete_bucket", Some(bucket)))
                     };
                 }
 
@@ -2314,4 +2325,37 @@ mod tests {
             .collect::<Vec<_>>();
         assert_eq!(calls, vec![1, 1, 0, 0, 0, 0, 0, 0]);
     }
+
+    #[test]
+    fn peer_failure_without_details_names_operation_and_bucket() {
+        for op in ["heal_bucket", "make_bucket", "get_bucket_info", "delete_bucket"] {
+            let message = peer_failure_without_details(op, Some("ops-bucket")).to_string();
+            assert!(message.contains(op), "{op} message must name the operation: {message}");
+            assert!(message.contains("ops-bucket"), "{op} message must name the bucket: {message}");
+        }
+
+        let message = peer_failure_without_details("list_bucket", None).to_string();
+        assert!(message.contains("list_bucket"), "cluster-wide message must name the operation");
+        assert!(!message.trim().is_empty());
+    }
+
+    #[test]
+    fn peer_failure_without_details_keeps_one_reduce_errs_bucket_per_operation() {
+        // reduce_errs groups Io errors by kind plus rendered message: peers failing the
+        // same operation on the same bucket must still reach quorum as one dominant error.
+        let per_pool_errs = vec![
+            Some(peer_failure_without_details("delete_bucket", Some("shared"))),
+            Some(peer_failure_without_details("delete_bucket", Some("shared"))),
+            Some(peer_failure_without_details("delete_bucket", Some("shared"))),
+        ];
+        assert_eq!(
+            reduce_pool_write_quorum_errs(&per_pool_errs),
+            Some(peer_failure_without_details("delete_bucket", Some("shared")))
+        );
+
+        assert_ne!(
+            peer_failure_without_details("delete_bucket", Some("shared")),
+            peer_failure_without_details("get_bucket_info", Some("shared"))
+        );
+    }
 }

crates/ecstore/src/config/audit.rs

@@ -39,7 +39,6 @@ use rustfs_config::{
 };
 use std::sync::LazyLock;
 
-#[allow(dead_code)]
 #[allow(clippy::declare_interior_mutable_const)]
 /// Default KVS for audit webhook settings.
 pub static DEFAULT_AUDIT_WEBHOOK_KVS: LazyLock<KVS> = LazyLock::new(|| {
@@ -117,7 +116,6 @@ pub static DEFAULT_AUDIT_WEBHOOK_KVS: LazyLock<KVS> = LazyLock::new(|| {
     ])
 });
 
-#[allow(dead_code)]
 #[allow(clippy::declare_interior_mutable_const)]
 /// Default KVS for audit MQTT settings.
 pub static DEFAULT_AUDIT_MQTT_KVS: LazyLock<KVS> = LazyLock::new(|| {
@@ -375,7 +373,6 @@ pub static DEFAULT_AUDIT_NATS_KVS: LazyLock<KVS> = LazyLock::new(|| {
     ])
 });
 
-#[allow(dead_code)]
 pub static DEFAULT_AUDIT_PULSAR_KVS: LazyLock<KVS> = LazyLock::new(|| {
     KVS(vec![
         KV {

crates/ecstore/src/config/heal.rs

@@ -12,12 +12,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use crate::error::{Error, Result};
 use rustfs_config::server_config::{KV, KVS};
 use rustfs_config::{DEFAULT_HEAL_BITROT_CYCLE_SECS, HEAL_BITROT_CYCLE};
-use rustfs_utils::string::parse_bool;
 use std::sync::LazyLock;
-use std::time::Duration;
 
 pub static DEFAULT_KVS: LazyLock<KVS> = LazyLock::new(|| {
     KVS(vec![KV {
@@ -26,59 +23,3 @@ pub static DEFAULT_KVS: LazyLock<KVS> = LazyLock::new(|| {
         hidden_if_empty: false,
     }])
 });
-
-#[derive(Debug, Default)]
-pub struct Config {
-    pub bitrot: String,
-    pub sleep: Duration,
-    pub io_count: usize,
-    pub drive_workers: usize,
-    pub cache: Duration,
-}
-
-impl Config {
-    pub fn bitrot_scan_cycle(&self) -> Duration {
-        self.cache
-    }
-
-    pub fn get_workers(&self) -> usize {
-        self.drive_workers
-    }
-
-    pub fn update(&mut self, nopts: &Config) {
-        self.bitrot = nopts.bitrot.clone();
-        self.io_count = nopts.io_count;
-        self.sleep = nopts.sleep;
-        self.drive_workers = nopts.drive_workers;
-    }
-}
-
-const RUSTFS_BITROT_CYCLE_IN_MONTHS: u64 = 1;
-
-fn parse_bitrot_config(s: &str) -> Result<Duration> {
-    match parse_bool(s) {
-        Ok(enabled) => {
-            if enabled {
-                Ok(Duration::from_secs_f64(0.0))
-            } else {
-                Ok(Duration::from_secs_f64(-1.0))
-            }
-        }
-        Err(_) => {
-            if !s.ends_with("m") {
-                return Err(Error::other("unknown format"));
-            }
-
-            match s.trim_end_matches('m').parse::<u64>() {
-                Ok(months) => {
-                    if months < RUSTFS_BITROT_CYCLE_IN_MONTHS {
-                        return Err(Error::other(format!("minimum bitrot cycle is {RUSTFS_BITROT_CYCLE_IN_MONTHS} month(s)")));
-                    }
-
-                    Ok(Duration::from_secs(months * 30 * 24 * 60))
-                }
-                Err(err) => Err(Error::other(err)),
-            }
-        }
-    }
-}

crates/ecstore/src/config/mod.rs

@@ -16,7 +16,6 @@
 
 mod audit;
 pub mod com;
-#[allow(dead_code)]
 pub mod heal;
 mod notify;
 mod oidc;

crates/ecstore/src/core/pools.rs

@@ -16,6 +16,7 @@ use crate::bucket::replication::replication_state_from_filemeta;
 use crate::bucket::versioning_sys::BucketVersioningSys;
 use crate::bucket::{
     lifecycle::{
+        LifecycleExpiryConfigs,
         bucket_lifecycle_audit::LcEventSrc,
         bucket_lifecycle_ops::{
             LifecycleOps, apply_expiry_on_transitioned_object, apply_expiry_rule_in, eval_action_from_lifecycle,
@@ -1996,11 +1997,11 @@ impl PoolMeta {
         Ok(false)
     }
 
-    #[allow(dead_code)]
     pub fn validate(&self, pools: Vec<Arc<Sets>>) -> Result<bool> {
         struct PoolInfo {
             position: usize,
             completed: bool,
+            #[allow(dead_code, reason = "written but never read back (backlog#1823)")]
             decom_started: bool,
         }
 
@@ -2335,6 +2336,10 @@ fn lifecycle_action_removes_data_movement_version(action: IlmAction) -> bool {
     )
 }
 
+fn lifecycle_action_skips_heal_version(action: IlmAction) -> bool {
+    action.delete()
+}
+
 fn resolve_data_movement_lifecycle_expiry_result(action: IlmAction, apply_actions: bool, applied: bool) -> Result<bool> {
     if !apply_actions || applied {
         return Ok(true);
@@ -2385,7 +2390,80 @@ pub(crate) async fn should_skip_lifecycle_for_data_movement(
     }
 }
 
+pub struct HealLifecycleExpiryContext {
+    configs: LifecycleExpiryConfigs,
+}
+
 impl ECStore {
+    pub async fn load_heal_lifecycle_expiry_context(&self, bucket: &str) -> Result<Option<HealLifecycleExpiryContext>> {
+        if bucket == RUSTFS_META_BUCKET {
+            return Ok(None);
+        }
+
+        let configs = get_expiry_configs(self, bucket).await?;
+        if configs.lifecycle.is_none() {
+            return Ok(None);
+        }
+
+        Ok(Some(HealLifecycleExpiryContext { configs }))
+    }
+
+    pub async fn enqueue_heal_lifecycle_expiry(
+        self: &Arc<Self>,
+        context: &HealLifecycleExpiryContext,
+        bucket: &str,
+        object: &str,
+        version_id: Option<&str>,
+        object_info: Option<&crate::object_api::ObjectInfo>,
+    ) -> Result<bool> {
+        let Some(lifecycle_config) = context.configs.lifecycle.as_ref() else {
+            return Ok(false);
+        };
+
+        let object_info = if let Some(object_info) = object_info {
+            if object_info.bucket != bucket || object_info.name != object {
+                return Ok(false);
+            }
+            let snapshot_version_id = object_info
+                .version_id
+                .filter(|version_id| !version_id.is_nil())
+                .map(|version_id| version_id.to_string());
+            if snapshot_version_id.as_deref() != version_id {
+                return Ok(false);
+            }
+            object_info.clone()
+        } else {
+            match self
+                .get_object_info(
+                    bucket,
+                    object,
+                    &ObjectOptions {
+                        version_id: version_id.map(str::to_string),
+                        versioned: version_id.is_some(),
+                        expected_bucket_incarnation_id: Some(context.configs.bucket_incarnation_id),
+                        ..Default::default()
+                    },
+                )
+                .await
+            {
+                Ok(object_info) => object_info,
+                Err(err) if is_err_object_not_found(&err) || is_err_version_not_found(&err) => return Ok(false),
+                Err(err) => return Err(err),
+            }
+        };
+
+        let event = eval_action_from_lifecycle(lifecycle_config, context.configs.object_lock.as_deref(), &object_info).await;
+        if !lifecycle_action_skips_heal_version(event.action) {
+            return Ok(false);
+        }
+
+        if lifecycle_delete_all_versions_blocked_by_replication(self.clone(), bucket, &object_info.name, event.action).await? {
+            return Ok(false);
+        }
+
+        Ok(apply_expiry_rule_in(self.clone(), &event, &LcEventSrc::Scanner, &object_info).await)
+    }
+
     async fn save_current_pool_meta(&self) -> Result<()> {
         let _save_guard = self.pool_meta_save_gate.lock().await;
         let snapshot = {
@@ -4287,6 +4365,19 @@ mod tests {
         ));
     }
 
+    #[test]
+    fn lifecycle_action_skips_heal_version_for_every_delete_action() {
+        assert!(lifecycle_action_skips_heal_version(IlmAction::DeleteAction));
+        assert!(lifecycle_action_skips_heal_version(IlmAction::DeleteVersionAction));
+        assert!(lifecycle_action_skips_heal_version(IlmAction::DeleteRestoredAction));
+        assert!(lifecycle_action_skips_heal_version(IlmAction::DeleteRestoredVersionAction));
+        assert!(lifecycle_action_skips_heal_version(IlmAction::DeleteAllVersionsAction));
+        assert!(lifecycle_action_skips_heal_version(IlmAction::DelMarkerDeleteAllVersionsAction));
+        assert!(!lifecycle_action_skips_heal_version(IlmAction::TransitionAction));
+        assert!(!lifecycle_action_skips_heal_version(IlmAction::TransitionVersionAction));
+        assert!(!lifecycle_action_skips_heal_version(IlmAction::NoneAction));
+    }
+
     #[test]
     fn resolve_data_movement_lifecycle_expiry_result_allows_dry_run_skip() {
         let skip = resolve_data_movement_lifecycle_expiry_result(IlmAction::DeleteVersionAction, false, false)
@@ -4958,13 +5049,19 @@ fn is_disk_online_state(state: &str) -> bool {
 }
 
 #[deprecated(since = "0.1.0", note = "Use fallback_total_capacity_dedup instead")]
-#[allow(dead_code)]
+#[allow(
+    dead_code,
+    reason = "superseded by the replacement named in the comment at pools.rs:5071 (backlog#1823)"
+)]
 fn fallback_total_capacity(disks: &[rustfs_madmin::Disk]) -> usize {
     fallback_total_capacity_dedup(disks)
 }
 
 #[deprecated(since = "0.1.0", note = "Use fallback_free_capacity_dedup instead")]
-#[allow(dead_code)]
+#[allow(
+    dead_code,
+    reason = "superseded by the replacement named in the comment at pools.rs:5071 (backlog#1823)"
+)]
 fn fallback_free_capacity(disks: &[rustfs_madmin::Disk]) -> usize {
     fallback_free_capacity_dedup(disks)
 }

crates/ecstore/src/core/sets.rs

@@ -1140,11 +1140,11 @@ impl crate::storage_api_contracts::heal::HealOperations for Sets {
 
         Err(Error::DiskNotFound)
     }
-    #[tracing::instrument(skip(self))]
-    async fn check_abandoned_parts(&self, _bucket: &str, _object: &str, _opts: &HealOpts) -> Result<()> {
-        // Multipart orphan reconciliation is intentionally retained above the pool/set layers
-        // until there is a concrete caller and a stable lower-level contract to implement.
-        Err(StorageError::NotImplemented)
+    #[tracing::instrument(level = "debug", skip(self, opts), fields(bucket = %bucket, object = %object, dry_run = opts.dry_run))]
+    async fn check_abandoned_parts(&self, bucket: &str, object: &str, opts: &HealOpts) -> Result<()> {
+        self.get_disks_for_heal_object(object, opts)?
+            .check_abandoned_parts(bucket, object, opts)
+            .await
     }
 }
 
@@ -1996,7 +1996,7 @@ mod tests {
     }
 
     #[tokio::test]
-    async fn sets_check_abandoned_parts_returns_typed_not_implemented_error() {
+    async fn sets_check_abandoned_parts_rejects_invalid_set_scope() {
         let format = FormatV3::new(1, 1);
         let sets = Sets {
             id: format.id,
@@ -2021,10 +2021,21 @@ mod tests {
         };
 
         let err = sets
-            .check_abandoned_parts("bucket", "object", &HealOpts::default())
+            .check_abandoned_parts(
+                "bucket",
+                "object",
+                &HealOpts {
+                    set: Some(1),
+                    ..Default::default()
+                },
+            )
             .await
-            .expect_err("abandoned-parts ownership should stay above the pool/set storage layers");
-        assert!(matches!(err, StorageError::NotImplemented));
+            .expect_err("out-of-range abandoned-parts set scope must fail closed");
+        assert!(
+            matches!(err, StorageError::InvalidArgument(_, ref field, ref reason)
+                if field == "set" && reason.contains("invalid heal set index 1")),
+            "unexpected invalid set error: {err:?}"
+        );
     }
 
     // Builds a single-set `Sets` over `SET_DRIVE_COUNT` local temp-dir disks,

crates/ecstore/src/disk/local.rs

@@ -6562,7 +6562,7 @@ impl LocalDisk {
         Ok(f)
     }
 
-    #[allow(dead_code)]
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn get_metrics(&self) -> DiskMetrics {
         DiskMetrics::default()
     }

crates/ecstore/src/services/rebalance/types.rs

@@ -132,7 +132,6 @@ impl RebalanceStopPropagationRecord {
     }
 }
 
-#[allow(dead_code)]
 #[derive(Debug, Clone, Default)]
 pub struct DiskStat {
     pub total_space: u64,

crates/ecstore/src/services/tier/tier_config.rs

@@ -16,8 +16,16 @@ use serde::{Deserialize, Serialize};
 use std::{fmt::Display, io};
 use tracing::info;
 
+#[allow(
+    dead_code,
+    reason = "tier config wire version stamped by the parity constructors below (backlog#1823)"
+)]
 const C_TIER_CONFIG_VER: &str = "v1";
 
+#[allow(
+    dead_code,
+    reason = "tier-name validation message reached only from the parity constructors below (backlog#1823)"
+)]
 const ERR_TIER_NAME_EMPTY: &str = "remote tier name empty";
 const WASABI_US_EAST_ENDPOINT: &str = "https://s3.wasabisys.com";
 const WASABI_ALTERNATIVE_ENDPOINTS: &[(&str, &str)] = &[
@@ -264,7 +272,6 @@ impl Clone for TierConfig {
     }
 }
 
-#[allow(dead_code)]
 impl TierConfig {
     pub(crate) fn clone_with_credentials(&self) -> Self {
         Self {
@@ -284,6 +291,7 @@ impl TierConfig {
         }
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn endpoint(&self) -> String {
         match self.tier_type {
             TierType::S3 => self.s3.as_ref().map(|s| s.endpoint.clone()).unwrap_or_default(),
@@ -303,6 +311,7 @@ impl TierConfig {
         }
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn bucket(&self) -> String {
         match self.tier_type {
             TierType::S3 => self.s3.as_ref().map(|s| s.bucket.clone()).unwrap_or_default(),
@@ -322,6 +331,7 @@ impl TierConfig {
         }
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn prefix(&self) -> String {
         match self.tier_type {
             TierType::S3 => self.s3.as_ref().map(|s| s.prefix.clone()).unwrap_or_default(),
@@ -341,6 +351,7 @@ impl TierConfig {
         }
     }
 
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn region(&self) -> String {
         match self.tier_type {
             TierType::S3 => self.s3.as_ref().map(|s| s.region.clone()).unwrap_or_default(),
@@ -457,7 +468,7 @@ impl TierWasabi {
 }
 
 impl TierS3 {
-    #[allow(dead_code)]
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn create<F>(
         name: &str,
         access_key: &str,
@@ -528,7 +539,7 @@ pub struct TierMinIO {
 }
 
 impl TierMinIO {
-    #[allow(dead_code)]
+    #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     fn create<F>(
         name: &str,
         endpoint: &str,

crates/ecstore/src/services/tier/tier_gen.rs

@@ -14,7 +14,6 @@
 
 use crate::services::tier::tier::TierConfigMgr;
 
-#[allow(dead_code)]
 impl TierConfigMgr {
     pub fn msg_size(&self) -> usize {
         100

crates/ecstore/src/set_disk/core/io_primitives.rs

@@ -4860,6 +4860,14 @@ impl SetDisks {
     /// is best-effort maintenance: individual delete failures are logged and
     /// skipped rather than propagated.
     pub(crate) async fn reclaim_orphan_data_dirs(&self, bucket: &str, object: &str) -> disk::error::Result<usize> {
+        self.reclaim_orphan_data_dirs_inner(bucket, object, false).await
+    }
+
+    pub(crate) async fn dry_run_reclaim_orphan_data_dirs(&self, bucket: &str, object: &str) -> disk::error::Result<usize> {
+        self.reclaim_orphan_data_dirs_inner(bucket, object, true).await
+    }
+
+    async fn reclaim_orphan_data_dirs_inner(&self, bucket: &str, object: &str, dry_run: bool) -> disk::error::Result<usize> {
         let disks = self.get_disks_internal().await;
 
         // Phase 1 (read-only): build the referenced-data-dir union and record the
@@ -4967,6 +4975,20 @@ impl SetDisks {
                     continue;
                 }
                 let stray = format!("{object}/{dir}");
+                if dry_run {
+                    removed += 1;
+                    debug!(
+                        target: "rustfs_ecstore::set_disk",
+                        event = "heal_abandoned_parts",
+                        component = "ecstore",
+                        subsystem = "heal",
+                        state = "dry_run_matched",
+                        result = "matched",
+                        bucket, object, data_dir = %dir,
+                        "Heal abandoned parts dry-run matched orphaned data directory"
+                    );
+                    continue;
+                }
                 match disk
                     .delete(
                         bucket,

crates/ecstore/src/set_disk/mod.rs

@@ -6998,6 +6998,100 @@ mod tests {
         assert!(object_dir.join(STORAGE_FORMAT_FILE).exists(), "metadata must be preserved");
     }
 
+    async fn recv_abandoned_parts_trace(
+        trace: &mut rustfs_common::trace_bus::TraceSubscription,
+        bucket: &str,
+        object: &str,
+        state: &str,
+    ) -> rustfs_common::trace_bus::TraceEvent {
+        for _ in 0..32 {
+            let event = tokio::time::timeout(std::time::Duration::from_secs(1), trace.recv())
+                .await
+                .expect("abandoned-parts trace event should arrive")
+                .expect("trace bus should stay open");
+            if event.kind == rustfs_common::trace_bus::TraceKind::Heal
+                && event.func == rustfs_common::trace_bus::TraceFunc::HealCheckAbandonedParts
+                && event.bucket.as_deref() == Some(bucket)
+                && event.object.as_deref() == Some(object)
+                && trace_attr_string(&event, "state").as_deref() == Some(state)
+            {
+                return (*event).clone();
+            }
+        }
+
+        panic!("expected abandoned-parts trace state {state} for {bucket}/{object}");
+    }
+
+    fn trace_attr_string(event: &rustfs_common::trace_bus::TraceEvent, key: &str) -> Option<String> {
+        event.attrs.iter().find_map(|attr| {
+            if attr.key != key {
+                return None;
+            }
+            Some(match &attr.value {
+                rustfs_common::trace_bus::TraceVal::Bool(value) => value.to_string(),
+                rustfs_common::trace_bus::TraceVal::U64(value) => value.to_string(),
+                rustfs_common::trace_bus::TraceVal::I64(value) => value.to_string(),
+                rustfs_common::trace_bus::TraceVal::Str(value) => value.to_string(),
+            })
+        })
+    }
+
+    #[tokio::test]
+    async fn check_abandoned_parts_dry_run_counts_without_deleting() {
+        let mut trace = rustfs_common::trace_bus::subscribe_trace_events();
+        let (dir, disk) = make_single_local_disk().await;
+        let live = Uuid::new_v4();
+        let orphan = Uuid::new_v4();
+
+        let object_dir = dir.path().join("bucket").join("obj");
+        write_object_meta_with_data_dirs(&object_dir, "bucket", "obj", &[live]).await;
+        fs::create_dir_all(object_dir.join(live.to_string()))
+            .await
+            .expect("live data dir should be created");
+        fs::create_dir_all(object_dir.join(orphan.to_string()))
+            .await
+            .expect("orphan data dir should be created");
+
+        let set = make_set_disks_with(vec![Some(disk)]).await;
+        set.check_abandoned_parts(
+            "bucket",
+            "obj",
+            &HealOpts {
+                dry_run: true,
+                no_lock: true,
+                ..Default::default()
+            },
+        )
+        .await
+        .expect("dry-run abandoned-parts check should succeed");
+        let dry_run_trace = recv_abandoned_parts_trace(&mut trace, "bucket", "obj", "dry_run_matched").await;
+        assert_eq!(trace_attr_string(&dry_run_trace, "dry_run").as_deref(), Some("true"));
+        assert_eq!(trace_attr_string(&dry_run_trace, "data_dirs").as_deref(), Some("1"));
+
+        assert!(object_dir.join(live.to_string()).exists(), "referenced data dir must be preserved");
+        assert!(object_dir.join(orphan.to_string()).exists(), "dry-run must not remove orphaned data dir");
+
+        set.check_abandoned_parts(
+            "bucket",
+            "obj",
+            &HealOpts {
+                no_lock: true,
+                ..Default::default()
+            },
+        )
+        .await
+        .expect("abandoned-parts check should reclaim stale data dir");
+        let reclaim_trace = recv_abandoned_parts_trace(&mut trace, "bucket", "obj", "reclaimed").await;
+        assert_eq!(trace_attr_string(&reclaim_trace, "dry_run").as_deref(), Some("false"));
+        assert_eq!(trace_attr_string(&reclaim_trace, "data_dirs").as_deref(), Some("1"));
+
+        assert!(
+            object_dir.join(live.to_string()).exists(),
+            "referenced data dir must remain after reclaim"
+        );
+        assert!(!object_dir.join(orphan.to_string()).exists(), "orphaned data dir must be removed");
+    }
+
     #[tokio::test]
     async fn reclaim_orphan_data_dirs_recovers_deferred_cleanup_after_restart() {
         let (dir, disk) = make_single_local_disk().await;
@@ -12233,11 +12327,18 @@ mod tests {
             .expect_err("unsupported copy_object_part should return a typed error");
         assert!(matches!(copy_part_err, StorageError::NotImplemented));
 
-        let abandoned_err = set_disks
-            .check_abandoned_parts("bucket", "object", &HealOpts::default())
+        set_disks
+            .check_abandoned_parts(
+                "bucket",
+                "object",
+                &HealOpts {
+                    dry_run: true,
+                    no_lock: true,
+                    ..Default::default()
+                },
+            )
             .await
-            .expect_err("abandoned-parts check should stay in the upper reconciliation layer");
-        assert!(matches!(abandoned_err, StorageError::NotImplemented));
+            .expect("abandoned-parts check should be callable on empty disk sets");
     }
 
     #[tokio::test]

crates/ecstore/src/set_disk/ops/heal.rs

@@ -16,6 +16,7 @@ use super::super::*;
 use crate::disk::disk_store::DiskStoreRenameDataExt;
 use crate::io_support::bitrot::object_mmap_read_enabled;
 use crate::storage_api_contracts::namespace::NamespaceLocking as _;
+use rustfs_common::trace_bus::{TraceEvent, TraceFunc, TraceKind, trace_emit};
 use tracing::trace;
 
 const LOG_COMPONENT_ECSTORE: &str = "ecstore";
@@ -2057,11 +2058,61 @@ impl crate::storage_api_contracts::heal::HealOperations for SetDisks {
         Err(Error::DiskNotFound)
     }
 
-    #[tracing::instrument(skip(self))]
-    async fn check_abandoned_parts(&self, _bucket: &str, _object: &str, _opts: &HealOpts) -> Result<()> {
-        // Multipart orphan reconciliation is intentionally retained above the set layer
-        // until there is a concrete caller and a stable lower-level contract to implement.
-        Err(StorageError::NotImplemented)
+    #[tracing::instrument(level = "debug", skip(self, opts), fields(bucket = %bucket, object = %object, dry_run = opts.dry_run))]
+    async fn check_abandoned_parts(&self, bucket: &str, object: &str, opts: &HealOpts) -> Result<()> {
+        let started_at = std::time::Instant::now();
+        let _write_lock_guard = if !opts.no_lock {
+            let ns_lock = self.new_ns_lock(bucket, object).await?;
+            Some(
+                ns_lock
+                    .get_write_lock(get_lock_acquire_timeout())
+                    .await
+                    .map_err(|e| self.map_namespace_lock_error(bucket, object, "write", e))?,
+            )
+        } else {
+            None
+        };
+
+        let removed = if opts.dry_run {
+            self.dry_run_reclaim_orphan_data_dirs(bucket, object).await?
+        } else {
+            self.reclaim_orphan_data_dirs(bucket, object).await?
+        };
+        let state = if opts.dry_run && removed > 0 {
+            "dry_run_matched"
+        } else if removed > 0 {
+            "reclaimed"
+        } else {
+            "checked"
+        };
+        let data_dirs = u64::try_from(removed).unwrap_or(u64::MAX);
+
+        trace_emit(|| {
+            TraceEvent::new(TraceKind::Heal, TraceFunc::HealCheckAbandonedParts)
+                .with_bucket(bucket)
+                .with_object(object)
+                .with_duration(started_at.elapsed())
+                .with_attr("state", state)
+                .with_attr("dry_run", opts.dry_run)
+                .with_attr("data_dirs", data_dirs)
+        });
+
+        if removed > 0 {
+            trace!(
+                event = "heal_abandoned_parts",
+                component = LOG_COMPONENT_ECSTORE,
+                subsystem = LOG_SUBSYSTEM_HEAL,
+                state = if opts.dry_run { "dry_run_matched" } else { "reclaimed" },
+                result = "ok",
+                bucket,
+                object,
+                dry_run = opts.dry_run,
+                data_dirs = removed,
+                "Heal abandoned parts checked object data directories"
+            );
+        }
+
+        Ok(())
     }
 }
 
@@ -3246,4 +3297,223 @@ mod heal_result_report_tests {
         assert!(result.detail.contains("part 1"));
         assert!(result.detail.contains("bitrot_failure=true"));
     }
+
+    // HS-12 (backlog#1874): a versioned DELETE racing an object heal must never
+    // resurrect the deleted version. The heal has real reconstruction work (a
+    // shard of the doomed version is removed), so both sides touch the same
+    // (bucket, object, data_dir); whichever order the ns write lock serializes
+    // them in, the committed delete must win.
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn heal_racing_version_delete_never_resurrects_the_deleted_version() {
+        let (temp_dirs, disks, set) = hermetic_set_disks_isolated(4).await;
+        let bucket = "heal-race-delete-no-resurrect";
+        let object = "object.bin";
+        set.make_bucket(
+            bucket,
+            &MakeBucketOptions {
+                versioning_enabled: true,
+                ..Default::default()
+            },
+        )
+        .await
+        .expect("versioned bucket should be created");
+
+        let mut first_reader = PutObjReader::from_vec(vec![0x11; 1024 * 1024]);
+        let first_info = set
+            .put_object(
+                bucket,
+                object,
+                &mut first_reader,
+                &ObjectOptions {
+                    versioned: true,
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("first version should be written");
+        let first_version = first_info
+            .version_id
+            .expect("versioned put should return the first version id")
+            .to_string();
+
+        let mut second_reader = PutObjReader::from_vec(vec![0x22; 1024 * 1024]);
+        let second_info = set
+            .put_object(
+                bucket,
+                object,
+                &mut second_reader,
+                &ObjectOptions {
+                    versioned: true,
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("second version should be written");
+        let second_version = second_info
+            .version_id
+            .expect("versioned put should return the second version id")
+            .to_string();
+
+        // Damage one shard of the doomed version so the racing heal performs an
+        // actual reconstruction over its data dir instead of an early exit.
+        let doomed_source = disks[0]
+            .read_version("", bucket, object, &first_version, &ReadOptions::default())
+            .await
+            .expect("doomed version metadata should be readable");
+        let doomed_data_dir = doomed_source
+            .data_dir
+            .expect("non-inline version should have a data directory");
+        tokio::fs::remove_file(
+            temp_dirs[1]
+                .path()
+                .join(bucket)
+                .join(object)
+                .join(doomed_data_dir.to_string())
+                .join("part.1"),
+        )
+        .await
+        .expect("shard damage should be injected before the race");
+
+        let delete_set = set.clone();
+        let (delete_res, heal_res) = tokio::join!(
+            async {
+                delete_set
+                    .delete_object(
+                        bucket,
+                        object,
+                        ObjectOptions {
+                            versioned: true,
+                            version_id: Some(first_version.clone()),
+                            object_lock_config_snapshot: Some(Arc::new(crate::set_disk::ObjectLockConfigSnapshot::new(
+                                crate::bucket::metadata_sys::ObjectLockConfigState::ConfirmedAbsent,
+                            ))),
+                            ..Default::default()
+                        },
+                    )
+                    .await
+            },
+            async {
+                set.heal_object(
+                    bucket,
+                    object,
+                    "",
+                    &HealOpts {
+                        scan_mode: HealScanMode::Deep,
+                        ..Default::default()
+                    },
+                )
+                .await
+            },
+        );
+        delete_res.expect("version delete must succeed under lock serialization");
+        // The heal may legitimately report a transient failure when the version
+        // it was rebuilding disappears mid-flight; only the end state matters.
+        drop(heal_res);
+
+        let resurrected = set
+            .get_object_info(
+                bucket,
+                object,
+                &ObjectOptions {
+                    versioned: true,
+                    version_id: Some(first_version.clone()),
+                    ..Default::default()
+                },
+            )
+            .await;
+        assert!(
+            matches!(&resurrected, Err(Error::FileVersionNotFound) | Err(Error::ObjectNotFound(..))),
+            "a racing heal must not resurrect the deleted version: {resurrected:?}"
+        );
+
+        let survivor = set
+            .get_object_info(
+                bucket,
+                object,
+                &ObjectOptions {
+                    versioned: true,
+                    version_id: Some(second_version.clone()),
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("surviving version must remain readable after the race");
+        assert_eq!(survivor.size, 1024 * 1024, "survivor size must be intact");
+    }
+
+    // HS-12 (backlog#1874): unversioned overwrite commits race a Deep heal on
+    // the same object. The overwrite's post-commit tail deletes the replaced
+    // data dir without the ns lock (object.rs commit tail), which is exactly
+    // the intersection the audit flagged: the heal must tolerate the tail race
+    // (retryable outcome) and every committed overwrite must survive — the
+    // final current version is exactly the last payload written.
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn heal_racing_unversioned_overwrites_preserves_the_last_commit() {
+        let (temp_dirs, disks, set) = hermetic_set_disks_isolated(4).await;
+        let bucket = "heal-race-put-overwrite";
+        let object = "object.bin";
+        set.make_bucket(bucket, &MakeBucketOptions::default())
+            .await
+            .expect("bucket should be created");
+
+        const ROUNDS: usize = 8;
+        const PAYLOAD_SIZE: usize = 256 * 1024;
+        let mut last_etag = String::new();
+        for round in 0..ROUNDS {
+            // Give the heal something to rebuild on alternating rounds: remove a
+            // shard of the current data dir right before the race.
+            if round % 2 == 1 {
+                let current = disks[2]
+                    .read_version("", bucket, object, "", &ReadOptions::default())
+                    .await
+                    .expect("current metadata should be readable");
+                if let Some(data_dir) = current.data_dir {
+                    let shard = temp_dirs[3]
+                        .path()
+                        .join(bucket)
+                        .join(object)
+                        .join(data_dir.to_string())
+                        .join("part.1");
+                    if shard.exists() {
+                        tokio::fs::remove_file(&shard)
+                            .await
+                            .expect("shard damage should be injectable mid-race");
+                    }
+                }
+            }
+
+            let payload = vec![round as u8; PAYLOAD_SIZE];
+            let mut put_reader = PutObjReader::from_vec(payload);
+            let put_opts = ObjectOptions::default();
+            let heal_opts = HealOpts {
+                scan_mode: HealScanMode::Deep,
+                ..Default::default()
+            };
+            let (put_res, heal_res) = tokio::join!(
+                set.put_object(bucket, object, &mut put_reader, &put_opts),
+                set.heal_object(bucket, object, "", &heal_opts),
+            );
+            let put_info = put_res.expect("overwrite must succeed under lock serialization");
+            last_etag = put_info.etag.clone().unwrap_or_default();
+            // Heal outcome is unconstrained (may hit the tail race and report a
+            // retryable error); the invariant is checked on the end state.
+            drop(heal_res);
+        }
+
+        let final_info = set
+            .get_object_info(bucket, object, &ObjectOptions::default())
+            .await
+            .expect("object must remain readable after the race loop");
+        assert_eq!(
+            final_info.size, PAYLOAD_SIZE as i64,
+            "final current version must be the last committed overwrite"
+        );
+        assert_eq!(
+            final_info.etag.unwrap_or_default(),
+            last_etag,
+            "the racing heal loop must never leave a stale or resurrected current version"
+        );
+    }
 }

crates/ecstore/src/set_disk/ops/heal_walk.rs

@@ -23,6 +23,7 @@
 //! per-version `SetDisks::heal_object`.
 
 use super::super::*;
+use crate::object_api::ObjectInfo;
 use std::collections::HashSet;
 use std::sync::Mutex;
 use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
@@ -39,12 +40,16 @@ const BACKGROUND_WALKDIR_STALL_TIMEOUT: Duration = Duration::from_secs(60);
 /// it must not gate healing logic — the delete-marker vs data path is chosen
 /// inside `ops/heal.rs` from the resolved latest metadata. `version_id` is
 /// normalized (nil/absent UUID => `None`).
-#[derive(Debug, Clone, PartialEq, Eq)]
+#[derive(Debug, Clone)]
 pub struct HealWalkVersion {
     /// object key
     pub name: String,
     /// normalized version id (`None` when the version is nil/absent)
     pub version_id: Option<String>,
+    /// version modification time as Unix nanoseconds
+    pub mod_time_unix_nanos: Option<i128>,
+    /// object snapshot for lifecycle evaluation
+    pub lifecycle_object_info: Option<ObjectInfo>,
     /// whether this version is a delete marker (observability only)
     pub is_delete_marker: bool,
 }
@@ -63,6 +68,7 @@ struct HealWalkCollector {
     bucket: String,
     batch_objects: usize,
     version_budget: usize,
+    include_lifecycle_object_info: bool,
     objects: Mutex<Vec<HealWalkObject>>,
     decode_error: Mutex<Option<DiskError>>,
     version_total: AtomicUsize,
@@ -116,10 +122,25 @@ impl HealWalkCollector {
 
         let mut versions = Vec::with_capacity(fiv.versions.len() + fiv.free_versions.len());
         for fi in fiv.versions.iter().chain(fiv.free_versions.iter()) {
+            let version_uuid = fi.version_id.filter(|version_id| !version_id.is_nil());
+            let lifecycle_object_info = if self.include_lifecycle_object_info {
+                let mut lifecycle_fi = fi.clone();
+                lifecycle_fi.version_id = version_uuid;
+                Some(ObjectInfo::from_file_info(
+                    &lifecycle_fi,
+                    &self.bucket,
+                    &entry.name,
+                    version_uuid.is_some(),
+                ))
+            } else {
+                None
+            };
             versions.push(HealWalkVersion {
                 name: entry.name.clone(),
                 // Normalize: nil/absent version id => None.
-                version_id: fi.version_id.filter(|u| !u.is_nil()).map(|u| u.to_string()),
+                version_id: version_uuid.map(|u| u.to_string()),
+                mod_time_unix_nanos: fi.mod_time.map(|mod_time| mod_time.unix_timestamp_nanos()),
+                lifecycle_object_info,
                 is_delete_marker: fi.deleted,
             });
         }
@@ -173,11 +194,26 @@ impl HealWalkCollector {
                 }
             };
             for fi in fiv.versions.iter().chain(fiv.free_versions.iter()) {
-                let vid = fi.version_id.filter(|u| !u.is_nil()).map(|u| u.to_string());
+                let version_uuid = fi.version_id.filter(|version_id| !version_id.is_nil());
+                let vid = version_uuid.map(|u| u.to_string());
                 if seen.insert(vid.clone()) {
+                    let lifecycle_object_info = if self.include_lifecycle_object_info {
+                        let mut lifecycle_fi = fi.clone();
+                        lifecycle_fi.version_id = version_uuid;
+                        Some(ObjectInfo::from_file_info(
+                            &lifecycle_fi,
+                            &self.bucket,
+                            &entry.name,
+                            version_uuid.is_some(),
+                        ))
+                    } else {
+                        None
+                    };
                     versions.push(HealWalkVersion {
                         name: entry.name.clone(),
                         version_id: vid,
+                        mod_time_unix_nanos: fi.mod_time.map(|mod_time| mod_time.unix_timestamp_nanos()),
+                        lifecycle_object_info,
                         is_delete_marker: fi.deleted,
                     });
                 }
@@ -255,6 +291,7 @@ impl SetDisks {
         forward_to: Option<&str>,
         batch_objects: usize,
         version_budget: usize,
+        include_lifecycle_object_info: bool,
     ) -> disk::error::Result<(Vec<HealWalkVersion>, Option<String>, bool)> {
         assert!(batch_objects >= 2, "heal_walk_versions_page requires batch_objects >= 2");
 
@@ -264,6 +301,7 @@ impl SetDisks {
             bucket: bucket.to_string(),
             batch_objects,
             version_budget: version_budget.max(1),
+            include_lifecycle_object_info,
             objects: Mutex::new(Vec::new()),
             decode_error: Mutex::new(None),
             version_total: AtomicUsize::new(0),
@@ -347,6 +385,7 @@ mod tests {
             bucket: "bucket".to_string(),
             batch_objects: 2,
             version_budget: 2,
+            include_lifecycle_object_info: false,
             objects: Mutex::new(Vec::new()),
             decode_error: Mutex::new(None),
             version_total: AtomicUsize::new(0),
@@ -388,6 +427,8 @@ mod tests {
         HealWalkVersion {
             name: name.to_string(),
             version_id: Some(id.to_string()),
+            mod_time_unix_nanos: None,
+            lifecycle_object_info: None,
             is_delete_marker: dm,
         }
     }
@@ -491,6 +532,7 @@ mod tests {
             bucket: "bucket".to_string(),
             batch_objects: 1000,
             version_budget: 10_000,
+            include_lifecycle_object_info: false,
             objects: Mutex::new(Vec::new()),
             version_total: AtomicUsize::new(0),
             decode_error: Mutex::new(None),
@@ -567,7 +609,7 @@ mod tests {
         .expect("corrupt test metadata should be written");
 
         let error = set_disks
-            .heal_walk_versions_page(bucket, "", None, 2, 2)
+            .heal_walk_versions_page(bucket, "", None, 2, 2, false)
             .await
             .expect_err("semantic metadata corruption must fail the heal disk walk");
 

crates/ecstore/src/set_disk/ops/object.rs

@@ -5845,6 +5845,14 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
 
     #[tracing::instrument(skip(self))]
     async fn add_partial(&self, bucket: &str, object: &str, version_id: &str) -> Result<()> {
+        // MRF journal intent: partial-write recovery must survive a restart
+        // (HS-01); the heal request below remains the in-memory fast path.
+        rustfs_common::mrf_channel::try_send_mrf_intent(
+            rustfs_common::mrf_channel::MrfKind::PartialWrite,
+            bucket,
+            object,
+            uuid::Uuid::try_parse(version_id).ok(),
+        );
         let mut request = rustfs_common::heal_channel::create_heal_request_with_options(
             bucket.to_string(),
             Some(object.to_string()),

crates/ecstore/src/set_disk/read.rs

@@ -1077,6 +1077,15 @@ impl SetDisks {
                             "Recoverable decode error triggered read repair"
                         );
                         let version_id = fi.version_id.as_ref().map(ToString::to_string);
+                        // MRF journal intent: keeps a durable Urgent ECDecode
+                        // request alive across restarts even when the in-memory
+                        // read-repair request is dropped or lost (HS-01).
+                        rustfs_common::mrf_channel::try_send_mrf_intent(
+                            rustfs_common::mrf_channel::MrfKind::DecodeFailure,
+                            bucket,
+                            object,
+                            fi.version_id,
+                        );
                         submit_read_repair_heal(
                             bucket,
                             object,

crates/ecstore/src/store/heal.rs

@@ -18,6 +18,7 @@ use tracing::trace;
 
 const LOG_COMPONENT_ECSTORE: &str = "ecstore";
 const LOG_SUBSYSTEM_HEAL: &str = "heal";
+const EVENT_HEAL_ABANDONED_PARTS: &str = "heal_abandoned_parts";
 const EVENT_HEAL_FORMAT_COMPLETED: &str = "heal_format_completed";
 const EVENT_HEAL_OBJECT_STARTED: &str = "heal_object_started";
 
@@ -256,13 +257,40 @@ impl ECStore {
 
     #[instrument(skip(self))]
     pub(super) async fn handle_check_abandoned_parts(&self, bucket: &str, object: &str, opts: &HealOpts) -> Result<()> {
-        let _ = (bucket, object, opts);
-        // Stale multipart reconciliation is already owned by the lifecycle-driven
-        // background cleanup path in `bucket_lifecycle_ops.rs`. There is currently
-        // no stable object-heal contract that should fan this request out through
-        // pool/set storage layers, so keep the placeholder explicit at the ECStore
-        // boundary instead of dispatching into lower layers.
-        Err(StorageError::NotImplemented)
+        let object = encode_dir_object(object);
+        let pools = self.get_pools_for_heal_object(opts)?;
+
+        let mut futures = Vec::with_capacity(pools.len());
+        for pool in pools.iter() {
+            futures.push(pool.check_abandoned_parts(bucket, &object, opts));
+        }
+
+        let mut first_error = None;
+        for result in join_all(futures).await {
+            if let Err(err) = result
+                && first_error.is_none()
+            {
+                first_error = Some(err);
+            }
+        }
+
+        if let Some(err) = first_error {
+            return Err(err);
+        }
+
+        trace!(
+            event = EVENT_HEAL_ABANDONED_PARTS,
+            component = LOG_COMPONENT_ECSTORE,
+            subsystem = LOG_SUBSYSTEM_HEAL,
+            state = "completed",
+            result = "ok",
+            bucket,
+            object,
+            dry_run = opts.dry_run,
+            "Heal abandoned parts completed"
+        );
+
+        Ok(())
     }
 }
 

crates/ecstore/src/store/heal_walk.rs

@@ -34,6 +34,7 @@ impl ECStore {
         forward_to: Option<&str>,
         batch_objects: usize,
         version_budget: usize,
+        include_lifecycle_object_info: bool,
     ) -> Result<(Vec<HealWalkVersion>, Option<String>, bool)> {
         if pool_idx >= self.pools.len() || set_idx >= self.pools[pool_idx].disk_set.len() {
             return Err(Error::other(format!(
@@ -43,7 +44,7 @@ impl ECStore {
         }
 
         self.pools[pool_idx].disk_set[set_idx]
-            .heal_walk_versions_page(bucket, prefix, forward_to, batch_objects, version_budget)
+            .heal_walk_versions_page(bucket, prefix, forward_to, batch_objects, version_budget, include_lifecycle_object_info)
             .await
             .map_err(Error::from)
     }

crates/ecstore/src/store/mod.rs

@@ -216,6 +216,16 @@ impl std::fmt::Debug for ECStore {
 /// These delegate to the process-global statics. No local state — the globals
 /// remain the single source of truth until the migration is complete.
 impl ECStore {
+    /// Every erasure set across all pools, pool-major order.
+    ///
+    /// Read-only queries that must consult each set's own copy of a
+    /// per-bucket object (e.g. the scanner's `.usage-cache.bin`) iterate
+    /// this instead of the hash-routed store path, which would always land
+    /// on one set (rustfs/backlog#1872).
+    pub fn all_set_disks(&self) -> Vec<Arc<crate::set_disk::SetDisks>> {
+        self.pools.iter().flat_map(|pool| pool.disk_set.iter().cloned()).collect()
+    }
+
     /// Get server configuration (delegates to global)
     pub fn get_server_config(&self) -> Option<Config> {
         runtime_sources::server_config()

crates/heal/Cargo.toml

@@ -89,6 +89,8 @@ async-trait = { workspace = true }
 futures = { workspace = true }
 metrics = { workspace = true }
 base64 = { workspace = true }
+bytes = { workspace = true }
+crc-fast = { workspace = true }
 
 [dev-dependencies]
 serde_json = { workspace = true, features = ["raw_value"] }

crates/heal/src/heal/channel.rs

@@ -612,7 +612,8 @@ impl HealChannelProcessor {
             HealRequestSource::Admin
             | HealRequestSource::AutoHeal
             | HealRequestSource::Internal
-            | HealRequestSource::ReadRepair => true,
+            | HealRequestSource::ReadRepair
+            | HealRequestSource::Mrf => true,
         });
 
         // Build HealOptions with all available fields
@@ -767,6 +768,7 @@ mod tests {
             _bucket: &str,
             _prefix: &str,
             _continuation_token: Option<&str>,
+            _include_lifecycle_object_info: bool,
         ) -> crate::Result<(Vec<crate::heal::storage::HealListItem>, Option<String>, bool)> {
             Ok((vec![], None, false))
         }

crates/heal/src/heal/erasure_healer.rs

@@ -23,13 +23,14 @@ use crate::heal::{
 };
 use crate::{Error, Result};
 use futures::{StreamExt, stream::FuturesUnordered};
-use metrics::gauge;
+use metrics::{counter, gauge};
 use rustfs_common::heal_channel::{HealOpts, HealRequestSource, HealScanMode};
 use rustfs_madmin::heal_commands::HealResultItem;
 use std::sync::{
     Arc,
     atomic::{AtomicUsize, Ordering},
 };
+use std::time::{Duration, UNIX_EPOCH};
 use tokio::sync::{RwLock, Semaphore};
 use tracing::{debug, error, warn};
 
@@ -47,6 +48,21 @@ enum HealObjectOutcome {
     Failed,
 }
 
+fn result_object_size_u64(result: &HealResultItem) -> u64 {
+    u64::try_from(result.object_size).unwrap_or(u64::MAX)
+}
+
+const NEW_VERSION_SKIP_GRACE_SECS: u64 = 60;
+const NANOS_PER_SECOND: i128 = 1_000_000_000;
+
+fn should_skip_new_version(mod_time_unix_nanos: Option<i128>, started_at_secs: u64) -> bool {
+    let Some(mod_time_unix_nanos) = mod_time_unix_nanos else {
+        return false;
+    };
+    let cutoff_secs = started_at_secs.saturating_add(NEW_VERSION_SKIP_GRACE_SECS);
+    mod_time_unix_nanos > i128::from(cutoff_secs).saturating_mul(NANOS_PER_SECOND)
+}
+
 struct PageConcurrencyGuard {
     in_flight: Arc<AtomicUsize>,
     set_label: String,
@@ -492,6 +508,7 @@ impl ErasureSetHealer {
                     &mut skipped_objects,
                     resume_manager,
                     checkpoint_manager,
+                    state.start_time,
                 )
                 .await;
 
@@ -658,6 +675,7 @@ impl ErasureSetHealer {
         skipped_objects: &mut u64,
         resume_manager: &ResumeManager,
         checkpoint_manager: &CheckpointManager,
+        started_at_secs: u64,
     ) -> Result<()> {
         debug!(
             target: "rustfs::heal::erasure_healer",
@@ -710,6 +728,7 @@ impl ErasureSetHealer {
         // The end-of-pass summary reports the full failed/skipped counts.
         let mut transient_skip_samples_logged = 0_u64;
         let mut failure_samples_logged = 0_u64;
+        let mut bytes_processed = self.progress.read().await.bytes_processed;
 
         // backlog#920: select the per-erasure-set DISK-WALK union enumerator when
         // the scan is Deep OR the request came from AutoHeal — these are the paths
@@ -718,17 +737,25 @@ impl ErasureSetHealer {
         // which stays the default.
         let use_disk_walk =
             matches!(self.heal_opts.scan_mode, HealScanMode::Deep) || matches!(self.source, HealRequestSource::AutoHeal);
+        let lifecycle_expiry_context = self.storage.load_heal_lifecycle_expiry_context(bucket).await?;
+        let include_lifecycle_object_info = lifecycle_expiry_context.is_some();
 
         loop {
             self.verify_replacement_identity_fence("page scan").await?;
             // Get one page of object versions
             let (objects, next_token, is_truncated) = if use_disk_walk {
                 self.storage
-                    .list_versions_for_heal_page_disk_walk(set_disk_id, bucket, "", continuation_token.as_deref())
+                    .list_versions_for_heal_page_disk_walk(
+                        set_disk_id,
+                        bucket,
+                        "",
+                        continuation_token.as_deref(),
+                        include_lifecycle_object_info,
+                    )
                     .await?
             } else {
                 self.storage
-                    .list_objects_for_heal_page(bucket, "", continuation_token.as_deref())
+                    .list_objects_for_heal_page(bucket, "", continuation_token.as_deref(), include_lifecycle_object_info)
                     .await?
             };
             let page_is_empty = objects.is_empty();
@@ -736,6 +763,7 @@ impl ErasureSetHealer {
             let page_resume_index = *current_object_index;
             let semaphore = Arc::new(Semaphore::new(page_concurrency_limit));
             let mut page_tasks = FuturesUnordered::new();
+            let mut completed_in_page = 0usize;
 
             // Capture the last version identity of this page for the anti-loop guard.
             let page_last = objects.last().map(|item| (item.name.clone(), item.version_id.clone()));
@@ -751,6 +779,75 @@ impl ErasureSetHealer {
                     continue;
                 }
 
+                if should_skip_new_version(item.mod_time_unix_nanos, started_at_secs) {
+                    checkpoint_manager.add_processed_object(key).await?;
+                    *processed_objects = processed_objects.saturating_add(1);
+                    completed_in_page = completed_in_page.saturating_add(1);
+                    counter!("rustfs_heal_skipped_new_versions_total").increment(1);
+                    {
+                        let mut progress = self.progress.write().await;
+                        progress.record_skipped_new_version();
+                        progress.set_current_object(Some(format!("skipped_new: {bucket}/{}", item.name)));
+                        progress.update_progress(*processed_objects, *successful_objects, *failed_objects, bytes_processed);
+                    }
+                    debug!(
+                        target: "rustfs::heal::erasure_healer",
+                        event = EVENT_HEAL_ERASURE_OBJECT_STATE,
+                        component = LOG_COMPONENT_HEAL,
+                        subsystem = LOG_SUBSYSTEM_ERASURE_HEALER,
+                        set_disk_id,
+                        bucket,
+                        object = %item.name,
+                        version_id = ?item.version_id,
+                        state = "skipped_new_version",
+                        "Erasure set object version skipped because it was written after heal started"
+                    );
+                    if completed_in_page.is_multiple_of(100) {
+                        checkpoint_manager.update_position(bucket_index, page_resume_index).await?;
+                    }
+                    continue;
+                }
+
+                if let Some(context) = lifecycle_expiry_context.as_ref()
+                    && self
+                        .storage
+                        .enqueue_heal_lifecycle_expiry(
+                            context,
+                            bucket,
+                            &item.name,
+                            item.version_id.as_deref(),
+                            item.lifecycle_object_info.as_ref(),
+                        )
+                        .await?
+                {
+                    checkpoint_manager.add_processed_object(key).await?;
+                    *processed_objects = processed_objects.saturating_add(1);
+                    completed_in_page = completed_in_page.saturating_add(1);
+                    counter!("rustfs_heal_skipped_ilm_expired_total").increment(1);
+                    {
+                        let mut progress = self.progress.write().await;
+                        progress.record_skipped_ilm_expired();
+                        progress.set_current_object(Some(format!("skipped_ilm: {bucket}/{}", item.name)));
+                        progress.update_progress(*processed_objects, *successful_objects, *failed_objects, bytes_processed);
+                    }
+                    debug!(
+                        target: "rustfs::heal::erasure_healer",
+                        event = EVENT_HEAL_ERASURE_OBJECT_STATE,
+                        component = LOG_COMPONENT_HEAL,
+                        subsystem = LOG_SUBSYSTEM_ERASURE_HEALER,
+                        set_disk_id,
+                        bucket,
+                        object = %item.name,
+                        version_id = ?item.version_id,
+                        state = "skipped_ilm_expired",
+                        "Erasure set object version skipped because lifecycle expiry was queued"
+                    );
+                    if completed_in_page.is_multiple_of(100) {
+                        checkpoint_manager.update_position(bucket_index, page_resume_index).await?;
+                    }
+                    continue;
+                }
+
                 resume_manager
                     .set_current_item(Some(bucket.to_string()), Some(item.name.clone()))
                     .await?;
@@ -777,7 +874,7 @@ impl ErasureSetHealer {
 
                     let _permit = match permit {
                         Ok(permit) => permit,
-                        Err(err) => return (dedup_key, object_name, version_id, Err(err)),
+                        Err(err) => return (dedup_key, object_name, version_id, (0, Err(err))),
                     };
 
                     let _in_flight_guard = PageConcurrencyGuard::new(in_flight, set_label);
@@ -788,7 +885,7 @@ impl ErasureSetHealer {
                     // recorded as skipped-ok rather than failed. The delete-marker
                     // vs data path is chosen internally in ops/heal.rs.
                     let result = if cancel_token.is_cancelled() {
-                        Err(Error::TaskCancelled)
+                        (0, Err(Error::TaskCancelled))
                     } else {
                         match storage
                             .heal_object(&bucket_name, &object_name, version_id.as_deref(), &heal_opts)
@@ -797,8 +894,9 @@ impl ErasureSetHealer {
                             Ok((result, None))
                                 if target_outcomes_complete(&result, &target_endpoints) =>
                             {
+                                let object_size = result_object_size_u64(&result);
                                 if !replacement_commit_evidence_required {
-                                    Ok(true)
+                                    (object_size, Ok(true))
                                 } else {
                                     match storage
                                         .replacement_targets_have_version(
@@ -810,27 +908,42 @@ impl ErasureSetHealer {
                                         )
                                         .await
                                     {
-                                        Ok(true) => Ok(true),
-                                        Ok(false) => Err(Error::transient_skip(format!(
+                                        Ok(true) => (object_size, Ok(true)),
+                                        Ok(false) => (object_size, Err(Error::transient_skip(format!(
                                             "Skipped heal for {bucket_name}/{object_name} because replacement target readback did not confirm the committed version"
-                                        ))),
-                                        Err(err) => Err(Error::transient_skip(format!(
+                                        )))),
+                                        Err(err) => (object_size, Err(Error::transient_skip(format!(
                                             "Skipped heal for {bucket_name}/{object_name} because replacement target readback failed: {err}"
-                                        ))),
+                                        )))),
                                     }
                                 }
-                            }
-                            Ok((_result, None)) if !target_endpoints.is_empty() => Err(Error::transient_skip(format!(
-                                "Skipped heal for {bucket_name}/{object_name} because a replacement target was not committed"
-                            ))),
-                            Ok((_result, None)) => Ok(true),
-                            Ok((_, Some(err))) if is_missing_object_dir_heal_result(&object_name, &err) => Ok(false),
-                            Ok((_, Some(err))) | Err(err) => match Self::classify_heal_object_error(&err) {
-                                HealObjectOutcome::Absent => Ok(false),
-                                HealObjectOutcome::Transient => Err(Error::transient_skip(format!(
-                                    "Skipped heal for {bucket_name}/{object_name} due to transient error: {err}"
+                            },
+                            Ok((result, None)) if !target_endpoints.is_empty() => (
+                                result_object_size_u64(&result),
+                                Err(Error::transient_skip(format!(
+                                    "Skipped heal for {bucket_name}/{object_name} because a replacement target was not committed"
                                 ))),
-                                HealObjectOutcome::Failed => Err(err),
+                            ),
+                            Ok((result, None)) => (result_object_size_u64(&result), Ok(true)),
+                            Ok((result, Some(err))) if is_missing_object_dir_heal_result(&object_name, &err) => {
+                                (result_object_size_u64(&result), Ok(false))
+                            }
+                            Ok((result, Some(err))) => {
+                                let object_size = result_object_size_u64(&result);
+                                match Self::classify_heal_object_error(&err) {
+                                    HealObjectOutcome::Absent => (object_size, Ok(false)),
+                                    HealObjectOutcome::Transient => (object_size, Err(Error::transient_skip(format!(
+                                        "Skipped heal for {bucket_name}/{object_name} due to transient error: {err}"
+                                    )))),
+                                    HealObjectOutcome::Failed => (object_size, Err(err)),
+                                }
+                            }
+                            Err(err) => match Self::classify_heal_object_error(&err) {
+                                HealObjectOutcome::Absent => (0, Ok(false)),
+                                HealObjectOutcome::Transient => (0, Err(Error::transient_skip(format!(
+                                    "Skipped heal for {bucket_name}/{object_name} due to transient error: {err}"
+                                )))),
+                                HealObjectOutcome::Failed => (0, Err(err)),
                             },
                         }
                     };
@@ -839,11 +952,12 @@ impl ErasureSetHealer {
                 });
             }
 
-            let mut completed_in_page = 0usize;
             while let Some((key, object, version_id, result)) = page_tasks.next().await {
+                let (object_size, result) = result;
                 match result {
                     Ok(true) => {
                         *successful_objects += 1;
+                        bytes_processed = bytes_processed.saturating_add(object_size);
                         checkpoint_manager.add_processed_object(key).await?;
                         debug!(
                             target: "rustfs::heal::erasure_healer",
@@ -861,6 +975,7 @@ impl ErasureSetHealer {
                     Ok(false) => {
                         checkpoint_manager.add_processed_object(key).await?;
                         *successful_objects += 1;
+                        bytes_processed = bytes_processed.saturating_add(object_size);
                         debug!(
                             target: "rustfs::heal::erasure_healer",
                             event = EVENT_HEAL_ERASURE_OBJECT_STATE,
@@ -877,6 +992,7 @@ impl ErasureSetHealer {
                     Err(err @ Error::TaskCancelled) | Err(err @ Error::TaskTimeout) => return Err(err),
                     Err(Error::TransientSkip { message }) => {
                         *skipped_objects += 1;
+                        bytes_processed = bytes_processed.saturating_add(object_size);
                         checkpoint_manager.add_skipped_object(key).await?;
                         demote_to_debug_when!(!take_failure_log_sample(&mut transient_skip_samples_logged), warn, target: "rustfs::heal::erasure_healer", {
                             event = EVENT_HEAL_ERASURE_OBJECT_STATE,
@@ -893,6 +1009,7 @@ impl ErasureSetHealer {
                     }
                     Err(err) => {
                         *failed_objects += 1;
+                        bytes_processed = bytes_processed.saturating_add(object_size);
                         checkpoint_manager.add_failed_object(key).await?;
                         demote_to_debug_when!(!take_failure_log_sample(&mut failure_samples_logged), warn, target: "rustfs::heal::erasure_healer", {
                             event = EVENT_HEAL_ERASURE_OBJECT_STATE,
@@ -911,6 +1028,11 @@ impl ErasureSetHealer {
 
                 *processed_objects += 1;
                 completed_in_page += 1;
+                {
+                    let mut progress = self.progress.write().await;
+                    progress.set_current_object(Some(format!("{bucket}/{object}")));
+                    progress.update_progress(*processed_objects, *successful_objects, *failed_objects, bytes_processed);
+                }
 
                 if completed_in_page.is_multiple_of(100) {
                     checkpoint_manager.update_position(bucket_index, page_resume_index).await?;
@@ -964,7 +1086,9 @@ impl ErasureSetHealer {
         progress.objects_scanned = state.total_objects;
         progress.objects_healed = state.successful_objects;
         progress.objects_failed = state.failed_objects;
-        progress.bytes_processed = 0; // set to 0 for now, can be extended later
+        progress.bytes_processed = 0; // Resume state tracks object counts, not byte counters.
+        progress.start_time = UNIX_EPOCH.checked_add(Duration::from_secs(state.start_time));
+        progress.last_update_time = UNIX_EPOCH.checked_add(Duration::from_secs(state.last_update));
         progress.set_current_object(state.current_object.clone());
     }
 }
@@ -1135,13 +1259,15 @@ mod resume_loop_tests {
     //! that emits programmable multi-version pages. These exercise the real loop
     //! logic (cursor seeding, per-version dedup, anti-loop guard, absence
     //! handling) — not merely a mock's own output.
-    use super::{ErasureSetHealer, target_outcomes_complete};
+    use super::{
+        ErasureSetHealer, NANOS_PER_SECOND, NEW_VERSION_SKIP_GRACE_SECS, should_skip_new_version, target_outcomes_complete,
+    };
     use crate::heal::progress::HealProgress;
     use crate::heal::resume::{
         CheckpointManager, RESUME_CHECKPOINT_FILE, ReplacementTargetIdentity, ResumeDeleteFailure, ResumeManager, ResumeUtils,
         compose_key,
     };
-    use crate::heal::storage::{DiskStatus, HealListItem, HealObjectInfo, HealStorageAPI};
+    use crate::heal::storage::{DiskStatus, HealLifecycleExpiryContext, HealListItem, HealObjectInfo, HealStorageAPI};
     use crate::heal::storage_api::status::BucketInfo;
     use crate::heal::{
         BUCKET_META_PREFIX, DiskOption, DiskStore, EcstoreError, Endpoint, HealDiskExt as _, RUSTFS_META_BUCKET, new_disk,
@@ -1149,7 +1275,7 @@ mod resume_loop_tests {
     use crate::{Error, Result};
     use rustfs_common::heal_channel::{HealOpts, HealRequestSource};
     use rustfs_madmin::heal_commands::{HealDriveInfo, HealResultItem, Infos};
-    use std::collections::{HashMap, VecDeque};
+    use std::collections::{HashMap, HashSet, VecDeque};
     use std::sync::atomic::{AtomicBool, Ordering};
     use std::sync::{Arc, Mutex};
     use tempfile::TempDir;
@@ -1160,10 +1286,37 @@ mod resume_loop_tests {
         HealListItem {
             name: name.to_string(),
             version_id: version.map(str::to_string),
+            mod_time_unix_nanos: None,
+            lifecycle_object_info: None,
             is_delete_marker: delete_marker,
         }
     }
 
+    fn item_with_mod_time(name: &str, version: Option<&str>, mod_time_secs: u64) -> HealListItem {
+        HealListItem {
+            name: name.to_string(),
+            version_id: version.map(str::to_string),
+            mod_time_unix_nanos: Some(i128::from(mod_time_secs).saturating_mul(NANOS_PER_SECOND)),
+            lifecycle_object_info: None,
+            is_delete_marker: false,
+        }
+    }
+
+    #[test]
+    fn new_version_filter_respects_grace_boundary() {
+        let started_at = 1_700_000_000;
+
+        assert!(!should_skip_new_version(None, started_at));
+        assert!(!should_skip_new_version(
+            Some(i128::from(started_at + NEW_VERSION_SKIP_GRACE_SECS).saturating_mul(NANOS_PER_SECOND)),
+            started_at,
+        ));
+        assert!(should_skip_new_version(
+            Some(i128::from(started_at + NEW_VERSION_SKIP_GRACE_SECS + 1).saturating_mul(NANOS_PER_SECOND)),
+            started_at,
+        ));
+    }
+
     #[test]
     fn target_outcomes_require_each_requested_endpoint_once_and_ok() {
         let result = HealResultItem {
@@ -1246,8 +1399,10 @@ mod resume_loop_tests {
         /// Target-specific physical readback evidence per `compose_key`; the
         /// fake models a healthy backend unless a test explicitly revokes it.
         replacement_commit_evidence: Mutex<HashMap<String, ReplacementCommitEvidence>>,
+        lifecycle_expired: Mutex<HashSet<String>>,
         /// every heal_object call recorded as (name, version_id)
         heal_calls: Mutex<Vec<(String, Option<String>)>>,
+        list_include_lifecycle_object_info: Mutex<Vec<bool>>,
         replacement_target_identity_sequences: Mutex<VecDeque<Vec<ReplacementTargetIdentity>>>,
         fail_listing: AtomicBool,
     }
@@ -1274,9 +1429,15 @@ mod resume_loop_tests {
                 .unwrap()
                 .insert(compose_key(name, version), ReplacementCommitEvidence::Error(message.to_string()));
         }
+        fn set_lifecycle_expired(&self, name: &str, version: Option<&str>) {
+            self.lifecycle_expired.lock().unwrap().insert(compose_key(name, version));
+        }
         fn calls(&self) -> Vec<(String, Option<String>)> {
             self.heal_calls.lock().unwrap().clone()
         }
+        fn list_include_lifecycle_object_info_calls(&self) -> Vec<bool> {
+            self.list_include_lifecycle_object_info.lock().unwrap().clone()
+        }
         fn fail_listing(&self) {
             self.fail_listing.store(true, Ordering::SeqCst);
         }
@@ -1330,6 +1491,23 @@ mod resume_loop_tests {
         async fn get_object_checksum(&self, _b: &str, _o: &str) -> Result<Option<String>> {
             Ok(None)
         }
+        async fn load_heal_lifecycle_expiry_context(&self, _bucket: &str) -> Result<Option<HealLifecycleExpiryContext>> {
+            Ok((!self.lifecycle_expired.lock().unwrap().is_empty()).then(HealLifecycleExpiryContext::test))
+        }
+        async fn enqueue_heal_lifecycle_expiry(
+            &self,
+            _context: &HealLifecycleExpiryContext,
+            _bucket: &str,
+            object: &str,
+            version_id: Option<&str>,
+            _object_info: Option<&HealObjectInfo>,
+        ) -> Result<bool> {
+            Ok(self
+                .lifecycle_expired
+                .lock()
+                .unwrap()
+                .contains(&compose_key(object, version_id)))
+        }
         async fn heal_object(
             &self,
             _bucket: &str,
@@ -1386,7 +1564,12 @@ mod resume_loop_tests {
             _bucket: &str,
             _prefix: &str,
             continuation_token: Option<&str>,
+            include_lifecycle_object_info: bool,
         ) -> Result<(Vec<HealListItem>, Option<String>, bool)> {
+            self.list_include_lifecycle_object_info
+                .lock()
+                .unwrap()
+                .push(include_lifecycle_object_info);
             if self.fail_listing.load(Ordering::SeqCst) {
                 return Err(Error::other("injected listing failure"));
             }
@@ -1476,6 +1659,7 @@ mod resume_loop_tests {
 
     /// Drive one bucket heal pass; returns (processed, successful, failed, skipped, result).
     async fn run(env: &Env) -> (u64, u64, u64, u64, Result<()>) {
+        let state = env.resume.get_state().await;
         let mut current_object_index = 0usize;
         let mut processed = 0u64;
         let mut successful = 0u64;
@@ -1494,6 +1678,7 @@ mod resume_loop_tests {
                 &mut skipped,
                 &env.resume,
                 &env.checkpoint,
+                state.start_time,
             )
             .await;
         (processed, successful, failed, skipped, result)
@@ -1559,6 +1744,7 @@ mod resume_loop_tests {
         let mut successful = 0;
         let mut failed = 0;
         let mut skipped = 0;
+        let started_at = env.resume.get_state().await.start_time;
 
         let error = healer
             .heal_bucket_with_resume(
@@ -1572,6 +1758,7 @@ mod resume_loop_tests {
                 &mut skipped,
                 &env.resume,
                 &env.checkpoint,
+                started_at,
             )
             .await
             .expect_err("a remounted target must not begin a new page scan");
@@ -1641,6 +1828,109 @@ mod resume_loop_tests {
         assert_eq!(skipped, 0);
     }
 
+    #[tokio::test]
+    async fn erasure_set_progress_accumulates_healed_object_bytes() {
+        let env = make_env().await;
+        env.storage.set_page(
+            None,
+            Page {
+                items: vec![item("first", Some("v1"), false), item("second", Some("v2"), false)],
+                next: None,
+                truncated: false,
+            },
+        );
+        env.storage.set_result(
+            "first",
+            Some("v1"),
+            HealResultItem {
+                object_size: 1024,
+                ..Default::default()
+            },
+        );
+        env.storage.set_result(
+            "second",
+            Some("v2"),
+            HealResultItem {
+                object_size: 2048,
+                ..Default::default()
+            },
+        );
+
+        let (processed, successful, failed, skipped, result) = run(&env).await;
+
+        result.expect("page heal should succeed");
+        assert_eq!(processed, 2);
+        assert_eq!(successful, 2);
+        assert_eq!(failed, 0);
+        assert_eq!(skipped, 0);
+        let progress = env.healer.progress.read().await;
+        assert_eq!(progress.objects_scanned, 2);
+        assert_eq!(progress.objects_healed, 2);
+        assert_eq!(progress.objects_failed, 0);
+        assert_eq!(progress.bytes_processed, 3072);
+        assert!(matches!(progress.current_object.as_deref(), Some("b/first" | "b/second")));
+    }
+
+    #[tokio::test]
+    async fn erasure_set_skips_versions_written_after_heal_started() {
+        let env = make_env().await;
+        let started_at = env.resume.get_state().await.start_time;
+        env.storage.set_page(
+            None,
+            Page {
+                items: vec![
+                    item_with_mod_time("old", Some("v1"), started_at + NEW_VERSION_SKIP_GRACE_SECS),
+                    item_with_mod_time("new", Some("v2"), started_at + NEW_VERSION_SKIP_GRACE_SECS + 1),
+                ],
+                next: None,
+                truncated: false,
+            },
+        );
+
+        let (processed, successful, failed, skipped, result) = run(&env).await;
+
+        result.expect("page heal should succeed");
+        assert_eq!(processed, 2);
+        assert_eq!(successful, 1);
+        assert_eq!(failed, 0);
+        assert_eq!(skipped, 0);
+        assert_eq!(env.storage.calls(), vec![("old".to_string(), Some("v1".to_string()))]);
+        let progress = env.healer.progress.read().await;
+        assert_eq!(progress.skipped_new_versions, 1);
+        assert_eq!(progress.objects_scanned, 2);
+        assert_eq!(progress.objects_healed, 1);
+        assert_eq!(progress.objects_failed, 0);
+    }
+
+    #[tokio::test]
+    async fn erasure_set_skips_versions_queued_for_lifecycle_expiry() {
+        let env = make_env().await;
+        env.storage.set_page(
+            None,
+            Page {
+                items: vec![item("expired", Some("v1"), false), item("kept", Some("v2"), false)],
+                next: None,
+                truncated: false,
+            },
+        );
+        env.storage.set_lifecycle_expired("expired", Some("v1"));
+
+        let (processed, successful, failed, skipped, result) = run(&env).await;
+
+        result.expect("page heal should succeed");
+        assert_eq!(processed, 2);
+        assert_eq!(successful, 1);
+        assert_eq!(failed, 0);
+        assert_eq!(skipped, 0);
+        assert_eq!(env.storage.calls(), vec![("kept".to_string(), Some("v2".to_string()))]);
+        assert_eq!(env.storage.list_include_lifecycle_object_info_calls(), vec![true]);
+        let progress = env.healer.progress.read().await;
+        assert_eq!(progress.skipped_ilm_expired, 1);
+        assert_eq!(progress.objects_scanned, 2);
+        assert_eq!(progress.objects_healed, 1);
+        assert_eq!(progress.objects_failed, 0);
+    }
+
     #[tokio::test]
     async fn bucket_listing_failure_does_not_mark_set_completed() {
         let env = make_env().await;

crates/heal/src/heal/manager.rs

@@ -270,6 +270,7 @@ pub struct HealSourceCounts {
     pub auto_heal: u64,
     pub internal: u64,
     pub read_repair: u64,
+    pub mrf: u64,
 }
 
 impl HealSourceCounts {
@@ -280,6 +281,7 @@ impl HealSourceCounts {
             HealRequestSource::AutoHeal => self.auto_heal += 1,
             HealRequestSource::Internal => self.internal += 1,
             HealRequestSource::ReadRepair => self.read_repair += 1,
+            HealRequestSource::Mrf => self.mrf += 1,
         }
     }
 }
@@ -2385,8 +2387,27 @@ impl HealManager {
             snapshot.objects_scanned = snapshot.objects_scanned.saturating_add(progress.objects_scanned);
             snapshot.objects_healed = snapshot.objects_healed.saturating_add(progress.objects_healed);
             snapshot.objects_failed = snapshot.objects_failed.saturating_add(progress.objects_failed);
+            snapshot.skipped_new_versions = snapshot.skipped_new_versions.saturating_add(progress.skipped_new_versions);
+            snapshot.skipped_ilm_expired = snapshot.skipped_ilm_expired.saturating_add(progress.skipped_ilm_expired);
+            snapshot.objects_total_count = snapshot.objects_total_count.saturating_add(progress.objects_total_count);
+            snapshot.objects_total_size = snapshot.objects_total_size.saturating_add(progress.objects_total_size);
             snapshot.bytes_processed = snapshot.bytes_processed.saturating_add(progress.bytes_processed);
+            snapshot.start_time = match (snapshot.start_time, progress.start_time) {
+                (Some(current), Some(next)) => Some(current.min(next)),
+                (None, next) => next,
+                (current, None) => current,
+            };
+            snapshot.last_update_time = match (snapshot.last_update_time, progress.last_update_time) {
+                (Some(current), Some(next)) => Some(current.max(next)),
+                (None, next) => next,
+                (current, None) => current,
+            };
+            if progress.current_object.is_some() {
+                snapshot.current_object = progress.current_object;
+            }
         }
+        snapshot.refresh_progress_percentage();
+        snapshot.refresh_estimated_completion_time();
         Some(snapshot)
     }
 
@@ -3208,6 +3229,7 @@ impl HealManager {
                         } else {
                             completed_task.get_status().await
                         };
+                        let completed_progress = completed_task.get_progress().await;
                         let completed_status_entry = CompletedHealStatus {
                             heal_type: completed_task.heal_type.clone(),
                             status: completed_status.clone(),
@@ -3223,6 +3245,7 @@ impl HealManager {
                         match completed_status {
                             HealTaskStatus::Completed => {
                                 stats.update_task_completion(true);
+                                stats.add_healed_objects(completed_progress.objects_healed, completed_progress.bytes_processed);
                             }
                             HealTaskStatus::Retrying { .. } => {}
                             _ => {
@@ -3749,6 +3772,7 @@ mod tests {
             _bucket: &str,
             _prefix: &str,
             _continuation_token: Option<&str>,
+            _include_lifecycle_object_info: bool,
         ) -> Result<(Vec<crate::heal::storage::HealListItem>, Option<String>, bool)> {
             Ok((Vec::new(), None, false))
         }
@@ -5396,6 +5420,8 @@ mod tests {
         ));
         {
             let mut progress = first.progress.write().await;
+            progress.start_time = Some(SystemTime::now() - Duration::from_secs(20));
+            progress.set_total_baseline(12, 8192);
             progress.update_progress(7, 3, 1, 4096);
         }
 
@@ -5405,6 +5431,8 @@ mod tests {
         ));
         {
             let mut progress = second.progress.write().await;
+            progress.start_time = Some(SystemTime::now() - Duration::from_secs(10));
+            progress.set_total_baseline(8, 4096);
             progress.update_progress(11, 5, 2, 2048);
         }
 
@@ -5419,7 +5447,11 @@ mod tests {
         assert_eq!(progress.objects_scanned, 18);
         assert_eq!(progress.objects_healed, 8);
         assert_eq!(progress.objects_failed, 3);
+        assert_eq!(progress.objects_total_count, 20);
+        assert_eq!(progress.objects_total_size, 12288);
         assert_eq!(progress.bytes_processed, 6144);
+        assert!((progress.progress_percentage - 50.0).abs() < 0.001);
+        assert!(progress.estimated_completion_time.is_some());
     }
 
     #[tokio::test]

crates/heal/src/heal/mod.rs

@@ -16,6 +16,7 @@ pub mod channel;
 pub mod erasure_healer;
 pub mod event;
 pub mod manager;
+pub mod mrf_queue;
 pub mod progress;
 pub(crate) mod replacement_readiness;
 pub mod resume;

crates/heal/src/heal/mrf_queue.rs

@@ -0,0 +1,682 @@
+// 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.
+
+//! Mission Repair Feed (MRF) queue, journal, and consumer.
+//!
+//! Intents arriving on the global channel (see `rustfs_common::mrf_channel`)
+//! are buffered in a bounded in-memory queue, translated into prioritized
+//! heal requests, and — while they are not yet accepted by the heal manager —
+//! mirrored into a durable journal so a crash or restart can replay them.
+//! This is the RustFS counterpart of MinIO's `.heal/mrf/list.bin` replay,
+//! layered on top of (not replacing) read-repair and scanner heal.
+//!
+//! Durability model: the journal is a snapshot of the *unaccepted* pending
+//! set, rewritten on a group-commit cadence (every flush interval or flush
+//! threshold new intents). A rewrite is atomic at the record level only — a
+//! torn tail simply truncates during replay because every record carries its
+//! own CRC32. Losing the last flush window (≤500 ms) is acceptable: replayed
+//! duplicates are merged by the manager's dedup key, and read-repair remains
+//! the safety net.
+
+use super::{DiskStore, HealDiskExt as _, local_disk_map_read};
+use crate::heal::manager::HealManager;
+use metrics::{counter, gauge};
+use rustfs_common::heal_channel::{HealAdmissionDropReason, HealAdmissionResult};
+use rustfs_common::mrf_channel::{MRF_MAX_ATTEMPTS, MrfIntent};
+use std::collections::VecDeque;
+use std::sync::Arc;
+use std::time::Duration;
+use tokio::sync::mpsc;
+use uuid::Uuid;
+
+use crate::heal::task::{HealOptions, HealPriority, HealRequest, HealType};
+
+/// Journal location inside the metadata bucket, following the resume-state
+/// layout.
+pub(crate) const MRF_JOURNAL_PATH: &str = "buckets/.heal/mrf/journal.bin";
+
+/// Record format tag.
+const MRF_JOURNAL_FORMAT: u8 = 1;
+/// Record layout version.
+const MRF_JOURNAL_VERSION: u8 = 1;
+
+/// Fixed header size: format, version, kind, attempts, enqueued_at_ms,
+/// has_version flag.
+const MRF_RECORD_FIXED_HEAD: usize = 1 + 1 + 1 + 1 + 8 + 1;
+
+#[derive(Debug, Clone)]
+pub(crate) struct MrfConsumerConfig {
+    /// In-memory queue capacity in intents.
+    pub queue_capacity: usize,
+    /// Journal byte budget; a pending snapshot above this bound is rejected
+    /// oldest-first so the journal can never grow unbounded.
+    pub journal_max_bytes: usize,
+    /// How many journal intents to re-arm per replay round.
+    pub replay_batch: usize,
+    /// Group-commit cadence for the journal snapshot.
+    pub flush_interval: Duration,
+    /// New intents between flushes that force an early snapshot.
+    pub flush_threshold: usize,
+    /// Backoff after the heal manager reports a full admission.
+    pub admission_backoff: Duration,
+}
+
+impl Default for MrfConsumerConfig {
+    fn default() -> Self {
+        Self {
+            queue_capacity: rustfs_utils::get_env_usize(
+                rustfs_config::ENV_HEAL_MRF_QUEUE_SIZE,
+                rustfs_config::DEFAULT_HEAL_MRF_QUEUE_SIZE,
+            ),
+            journal_max_bytes: rustfs_utils::get_env_usize(
+                rustfs_config::ENV_HEAL_MRF_JOURNAL_MAX_BYTES,
+                rustfs_config::DEFAULT_HEAL_MRF_JOURNAL_MAX_BYTES,
+            ),
+            replay_batch: rustfs_utils::get_env_usize(
+                rustfs_config::ENV_HEAL_MRF_REPLAY_BATCH,
+                rustfs_config::DEFAULT_HEAL_MRF_REPLAY_BATCH,
+            ),
+            flush_interval: Duration::from_millis(500),
+            flush_threshold: 1000,
+            admission_backoff: Duration::from_secs(5),
+        }
+    }
+}
+
+/// Bounded pending set with count and byte ceilings. Overflow drops the
+/// incoming intent (never a resident one) and counts the loss.
+pub(crate) struct MrfQueue {
+    pending: VecDeque<MrfIntent>,
+    bytes: usize,
+    capacity: usize,
+    byte_budget: usize,
+}
+
+impl MrfQueue {
+    pub(crate) fn new(capacity: usize, byte_budget: usize) -> Self {
+        Self {
+            pending: VecDeque::new(),
+            bytes: 0,
+            capacity,
+            byte_budget,
+        }
+    }
+
+    /// Returns `false` (after counting) when either ceiling would be crossed.
+    pub(crate) fn try_push(&mut self, intent: MrfIntent) -> bool {
+        let cost = intent.estimated_bytes();
+        if self.pending.len() >= self.capacity || self.bytes + cost > self.byte_budget {
+            counter!("rustfs_heal_mrf_dropped_total", "reason" => "queue_overflow").increment(1);
+            return false;
+        }
+        self.bytes += cost;
+        self.pending.push_back(intent);
+        true
+    }
+
+    pub(crate) fn pop_front(&mut self) -> Option<MrfIntent> {
+        let intent = self.pending.pop_front()?;
+        self.bytes = self.bytes.saturating_sub(intent.estimated_bytes());
+        Some(intent)
+    }
+
+    pub(crate) fn push_back(&mut self, intent: MrfIntent) {
+        self.bytes += intent.estimated_bytes();
+        self.pending.push_back(intent);
+    }
+
+    pub(crate) fn depth(&self) -> usize {
+        self.pending.len()
+    }
+
+    pub(crate) fn bytes(&self) -> usize {
+        self.bytes
+    }
+
+    pub(crate) fn intents(&self) -> impl Iterator<Item = &MrfIntent> {
+        self.pending.iter()
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Journal record codec
+// ---------------------------------------------------------------------------
+
+/// Append one encoded record to `out`.
+pub(crate) fn encode_intent(intent: &MrfIntent, out: &mut Vec<u8>) {
+    let start = out.len();
+    out.push(MRF_JOURNAL_FORMAT);
+    out.push(MRF_JOURNAL_VERSION);
+    out.push(match intent.kind {
+        rustfs_common::mrf_channel::MrfKind::DecodeFailure => 1,
+        rustfs_common::mrf_channel::MrfKind::MetadataCorruption => 2,
+        rustfs_common::mrf_channel::MrfKind::PartialWrite => 3,
+    });
+    out.push(intent.attempts);
+    out.extend_from_slice(&intent.enqueued_at_ms.to_le_bytes());
+    match intent.version_id {
+        Some(bytes) => {
+            out.push(1);
+            out.extend_from_slice(&bytes);
+        }
+        None => out.push(0),
+    }
+    out.extend_from_slice(&(intent.bucket.len() as u32).to_le_bytes());
+    out.extend_from_slice(&(intent.object.len() as u32).to_le_bytes());
+    out.extend_from_slice(intent.bucket.as_bytes());
+    out.extend_from_slice(intent.object.as_bytes());
+    let mut hasher = crc_fast::Digest::new(crc_fast::CrcAlgorithm::Crc32IsoHdlc);
+    hasher.update(&out[start..]);
+    out.extend_from_slice(&(hasher.finalize() as u32).to_le_bytes());
+}
+
+fn decode_one(data: &[u8]) -> Option<(MrfIntent, usize)> {
+    if data.len() < MRF_RECORD_FIXED_HEAD + 8 {
+        return None;
+    }
+    if data[0] != MRF_JOURNAL_FORMAT || data[1] != MRF_JOURNAL_VERSION {
+        return None;
+    }
+    let kind = match data[2] {
+        1 => rustfs_common::mrf_channel::MrfKind::DecodeFailure,
+        2 => rustfs_common::mrf_channel::MrfKind::MetadataCorruption,
+        3 => rustfs_common::mrf_channel::MrfKind::PartialWrite,
+        _ => return None,
+    };
+    let attempts = data[3];
+    let enqueued_at_ms = u64::from_le_bytes(data[4..12].try_into().expect("slice length checked"));
+    let has_version = data[12] != 0;
+    let mut cursor = MRF_RECORD_FIXED_HEAD;
+    let version_id = if has_version {
+        if data.len() < cursor + 16 {
+            return None;
+        }
+        let bytes: [u8; 16] = data[cursor..cursor + 16].try_into().expect("slice length checked");
+        cursor += 16;
+        Some(bytes)
+    } else {
+        None
+    };
+    if data.len() < cursor + 8 {
+        return None;
+    }
+    let bucket_len = u32::from_le_bytes(data[cursor..cursor + 4].try_into().expect("slice length checked")) as usize;
+    let object_len = u32::from_le_bytes(data[cursor + 4..cursor + 8].try_into().expect("slice length checked")) as usize;
+    cursor += 8;
+    let body_end = cursor.checked_add(bucket_len)?.checked_add(object_len)?;
+    let record_end = body_end.checked_add(4)?;
+    if data.len() < record_end {
+        return None;
+    }
+    let mut hasher = crc_fast::Digest::new(crc_fast::CrcAlgorithm::Crc32IsoHdlc);
+    hasher.update(&data[..body_end]);
+    if (hasher.finalize() as u32) != u32::from_le_bytes(data[body_end..record_end].try_into().expect("slice length checked")) {
+        return None;
+    }
+    let bucket = std::sync::Arc::from(std::str::from_utf8(&data[cursor..cursor + bucket_len]).ok()?);
+    let object = std::sync::Arc::from(std::str::from_utf8(&data[cursor + bucket_len..body_end]).ok()?);
+    Some((
+        MrfIntent {
+            bucket,
+            object,
+            version_id,
+            kind,
+            enqueued_at_ms,
+            attempts,
+        },
+        record_end,
+    ))
+}
+
+/// Decode a whole journal, stopping at the first torn or corrupt record.
+/// Returns the decoded intents and the number of trailing bytes discarded.
+pub(crate) fn decode_journal(data: &[u8]) -> (Vec<MrfIntent>, usize) {
+    let mut intents = Vec::new();
+    let mut cursor = 0usize;
+    while cursor < data.len() {
+        match decode_one(&data[cursor..]) {
+            Some((intent, consumed)) => {
+                intents.push(intent);
+                cursor += consumed;
+            }
+            None => break,
+        }
+    }
+    let truncated = data.len() - cursor;
+    (intents, truncated)
+}
+
+// ---------------------------------------------------------------------------
+// Journal disk IO (all local disks, first successful read wins)
+// ---------------------------------------------------------------------------
+
+async fn journal_disks() -> Vec<DiskStore> {
+    let map = local_disk_map_read().await;
+    map.values().flatten().cloned().collect()
+}
+
+async fn read_journal() -> Option<Vec<u8>> {
+    for disk in journal_disks().await {
+        match disk.read_all(super::RUSTFS_META_BUCKET, MRF_JOURNAL_PATH).await {
+            Ok(bytes) => return Some(bytes.to_vec()),
+            Err(_) => continue,
+        }
+    }
+    None
+}
+
+async fn write_journal(data: &[u8]) {
+    let payload = bytes::Bytes::copy_from_slice(data);
+    for disk in journal_disks().await {
+        if let Err(err) = disk
+            .write_all(super::RUSTFS_META_BUCKET, MRF_JOURNAL_PATH, payload.clone())
+            .await
+        {
+            warn_mrf_journal_write(&err);
+        }
+    }
+    if !data.is_empty() {
+        counter!("rustfs_heal_mrf_journal_fsync_total").increment(1);
+    }
+    gauge!("rustfs_heal_mrf_journal_bytes").set(data.len() as f64);
+}
+
+async fn delete_journal() {
+    for disk in journal_disks().await {
+        let _ = disk
+            .delete(
+                super::RUSTFS_META_BUCKET,
+                MRF_JOURNAL_PATH,
+                crate::heal::storage_api::owner::EcstoreDeleteOptions::default(),
+            )
+            .await;
+    }
+}
+
+fn warn_mrf_journal_write(err: &super::DiskError) {
+    tracing::warn!(
+        target: "rustfs::heal::mrf",
+        error = %err,
+        "MRF journal write failed; unconsumed intents may be lost on restart"
+    );
+}
+
+// ---------------------------------------------------------------------------
+// Consumer
+// ---------------------------------------------------------------------------
+
+/// Translate an intent into the prioritized heal request the issue specifies:
+/// decode failures go Urgent ECDecode, metadata corruption goes High
+/// Metadata, partial writes go Normal object heal.
+pub(crate) fn build_heal_request(intent: &MrfIntent) -> HealRequest {
+    let bucket = intent.bucket.to_string();
+    let object = intent.object.to_string();
+    let version_id = intent.version_id.map(|bytes| Uuid::from_bytes(bytes).to_string());
+    let (heal_type, priority) = match intent.kind {
+        rustfs_common::mrf_channel::MrfKind::DecodeFailure => (
+            HealType::ECDecode {
+                bucket,
+                object,
+                version_id,
+            },
+            HealPriority::Urgent,
+        ),
+        rustfs_common::mrf_channel::MrfKind::MetadataCorruption => (HealType::Metadata { bucket, object }, HealPriority::High),
+        rustfs_common::mrf_channel::MrfKind::PartialWrite => (
+            HealType::Object {
+                bucket,
+                object,
+                version_id,
+            },
+            HealPriority::Normal,
+        ),
+    };
+    let mut request = HealRequest::new(heal_type, HealOptions::default(), priority);
+    request.source = rustfs_common::heal_channel::HealRequestSource::Mrf;
+    request
+}
+
+struct MrfRuntime {
+    queue: MrfQueue,
+    config: MrfConsumerConfig,
+    new_since_flush: usize,
+    /// True while a journal snapshot exists on disk that no longer reflects
+    /// an all-consumed pending set; the next idle tick removes it (MinIO
+    /// deletes its `list.bin` after replay for the same reason).
+    journal_on_disk: bool,
+    /// Earliest instant a full-admission retry may proceed.
+    backoff_until: Option<tokio::time::Instant>,
+}
+
+impl MrfRuntime {
+    fn record_accept(&mut self) {
+        // Accepted intents leave the pending set; the next flush persists the
+        // smaller snapshot, which is the journal's compaction.
+    }
+
+    fn snapshot(&self) -> Vec<u8> {
+        let mut buf = Vec::new();
+        for intent in self.queue.intents() {
+            encode_intent(intent, &mut buf);
+        }
+        buf
+    }
+
+    async fn flush(&mut self) {
+        write_journal(&self.snapshot()).await;
+        self.new_since_flush = 0;
+        self.journal_on_disk = true;
+    }
+
+    /// Drain pending intents into the heal manager until it is full, the
+    /// queue empties, or attempts are exhausted.
+    async fn dispatch(&mut self, manager: &HealManager) {
+        if let Some(until) = self.backoff_until {
+            if tokio::time::Instant::now() < until {
+                return;
+            }
+            self.backoff_until = None;
+        }
+        while let Some(mut intent) = self.queue.pop_front() {
+            let request = build_heal_request(&intent);
+            match manager.submit_heal_request(request).await {
+                Ok(HealAdmissionResult::Accepted) | Ok(HealAdmissionResult::Merged) => self.record_accept(),
+                Ok(HealAdmissionResult::Full) | Ok(HealAdmissionResult::Dropped(HealAdmissionDropReason::QueueFull)) => {
+                    intent.attempts = intent.attempts.saturating_add(1);
+                    if intent.attempts >= MRF_MAX_ATTEMPTS {
+                        counter!("rustfs_heal_mrf_dropped_total", "reason" => "attempts_exhausted").increment(1);
+                        continue;
+                    }
+                    self.queue.push_back(intent);
+                    self.backoff_until = Some(tokio::time::Instant::now() + self.config.admission_backoff);
+                    break;
+                }
+                Ok(HealAdmissionResult::Dropped(_)) => {
+                    counter!("rustfs_heal_mrf_dropped_total", "reason" => "admission_policy").increment(1);
+                }
+                Err(_) => {
+                    intent.attempts = intent.attempts.saturating_add(1);
+                    if intent.attempts >= MRF_MAX_ATTEMPTS {
+                        counter!("rustfs_heal_mrf_dropped_total", "reason" => "attempts_exhausted").increment(1);
+                        continue;
+                    }
+                    self.queue.push_back(intent);
+                    self.backoff_until = Some(tokio::time::Instant::now() + self.config.admission_backoff);
+                    break;
+                }
+            }
+        }
+        gauge!("rustfs_heal_mrf_queue_depth").set(self.queue.depth() as f64);
+        gauge!("rustfs_heal_mrf_queue_bytes").set(self.queue.bytes() as f64);
+    }
+}
+
+/// Initialize the global MRF channel (honoring `RUSTFS_HEAL_MRF_ENABLE`) and
+/// spawn the consumer task. Called once from the heal runtime bootstrap right
+/// after the manager started; a disabled feature or a double call is a no-op.
+/// Public for integration tests that drive the real consumer loop.
+pub fn spawn_mrf_consumer(manager: Arc<HealManager>) {
+    let enabled = rustfs_utils::get_env_bool(rustfs_config::ENV_HEAL_MRF_ENABLE, rustfs_config::DEFAULT_HEAL_MRF_ENABLE);
+    rustfs_common::mrf_channel::set_mrf_delivery_enabled(enabled);
+    if !enabled {
+        tracing::info!(
+            target: "rustfs::heal::mrf",
+            "MRF intent pipeline disabled by configuration; producers will not deliver"
+        );
+        return;
+    }
+    let receiver = match rustfs_common::mrf_channel::init_mrf_channel() {
+        Ok(receiver) => receiver,
+        Err(err) => {
+            tracing::warn!(
+                target: "rustfs::heal::mrf",
+                error = err,
+                "MRF channel initialization failed; intents will be dropped at producers"
+            );
+            return;
+        }
+    };
+    tokio::spawn(async move {
+        run_mrf_consumer(manager, receiver).await;
+    });
+    tracing::info!(target: "rustfs::heal::mrf", "MRF intent consumer started");
+}
+
+/// Replay the durable journal into a fresh pending queue and submit whatever
+/// it armed. Returns the number of intact intents replayed. Duplicates are
+/// merged by the manager's dedup key; the journal file is removed once read
+/// (torn tails truncate via the per-record CRC). Public for integration tests;
+/// the live consumer invokes this through [`replay_into`] at startup.
+pub async fn replay_journal_once(manager: &Arc<HealManager>) -> usize {
+    let config = MrfConsumerConfig::default();
+    let mut queue = MrfQueue::new(config.queue_capacity, config.journal_max_bytes);
+    let mut backoff_until: Option<tokio::time::Instant> = None;
+    replay_into(manager, &mut queue, &mut backoff_until).await
+}
+
+/// Shared replay core: read + decode + re-arm + delete, then drain what fits.
+async fn replay_into(
+    manager: &Arc<HealManager>,
+    queue: &mut MrfQueue,
+    backoff_until: &mut Option<tokio::time::Instant>,
+) -> usize {
+    let Some(data) = read_journal().await else {
+        return 0;
+    };
+    let (intents, truncated) = decode_journal(&data);
+    if truncated > 0 {
+        tracing::warn!(
+            target: "rustfs::heal::mrf",
+            truncated_bytes = truncated,
+            "MRF journal had a torn tail; truncated records were discarded"
+        );
+    }
+    counter!("rustfs_heal_mrf_replayed_total").increment(intents.len() as u64);
+    let replayed = intents.len();
+    for intent in intents {
+        queue.try_push(intent);
+    }
+    delete_journal().await;
+
+    // Drain the replayed intents immediately; whatever the manager refuses
+    // stays armed in `queue` for the consumer's retry loop.
+    if backoff_until.is_none() {
+        while let Some(mut intent) = queue.pop_front() {
+            let request = build_heal_request(&intent);
+            match manager.submit_heal_request(request).await {
+                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 {
+                        queue.push_back(intent);
+                        *backoff_until = Some(tokio::time::Instant::now());
+                    }
+                    break;
+                }
+                Ok(HealAdmissionResult::Dropped(_)) | Err(_) => {}
+            }
+        }
+    }
+    replayed
+}
+
+/// Replay the journal, then keep draining the channel into the heal manager
+/// while persisting the pending snapshot.
+async fn run_mrf_consumer(manager: Arc<HealManager>, mut receiver: mpsc::Receiver<MrfIntent>) {
+    let config = MrfConsumerConfig::default();
+    let mut runtime = MrfRuntime {
+        queue: MrfQueue::new(config.queue_capacity, config.journal_max_bytes),
+        config: config.clone(),
+        new_since_flush: 0,
+        journal_on_disk: false,
+        backoff_until: None,
+    };
+
+    // Replay: read the journal, re-arm intents (duplicates are merged by the
+    // manager's dedup key), then drop the file so the next flush starts clean.
+    replay_into(&manager, &mut runtime.queue, &mut runtime.backoff_until).await;
+
+    let mut flush_tick = tokio::time::interval(runtime.config.flush_interval);
+    flush_tick.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
+    let mut batch: Vec<MrfIntent> = Vec::with_capacity(runtime.config.replay_batch);
+
+    loop {
+        tokio::select! {
+            received = receiver.recv_many(&mut batch, runtime.config.replay_batch) => {
+                if received == 0 {
+                    // Channel closed: flush once more and stop.
+                    runtime.flush().await;
+                    tracing::info!(
+                        target: "rustfs::heal::mrf",
+                        "MRF channel closed; consumer stopped after final flush"
+                    );
+                    return;
+                }
+                for intent in batch.drain(..) {
+                    runtime.queue.try_push(intent);
+                    runtime.new_since_flush += 1;
+                }
+                runtime.dispatch(manager.as_ref()).await;
+                if runtime.new_since_flush >= runtime.config.flush_threshold {
+                    runtime.flush().await;
+                }
+            }
+            _ = flush_tick.tick() => {
+                if runtime.new_since_flush > 0 || runtime.queue.depth() > 0 {
+                    runtime.flush().await;
+                    runtime.dispatch(manager.as_ref()).await;
+                } else if runtime.journal_on_disk {
+                    // All intents consumed: remove the journal so a restart
+                    // replays nothing (mirrors MinIO's post-replay unlink).
+                    delete_journal().await;
+                    runtime.journal_on_disk = false;
+                    gauge!("rustfs_heal_mrf_journal_bytes").set(0.0);
+                }
+                gauge!("rustfs_heal_mrf_queue_depth").set(runtime.queue.depth() as f64);
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use rustfs_common::mrf_channel::{MrfIntent, MrfKind};
+    use std::sync::Arc as StdArc;
+
+    fn intent(bucket: &str, object: &str, attempts: u8) -> MrfIntent {
+        MrfIntent {
+            bucket: StdArc::from(bucket),
+            object: StdArc::from(object),
+            version_id: Some([7u8; 16]),
+            kind: MrfKind::DecodeFailure,
+            enqueued_at_ms: 1_700_000_000_000,
+            attempts,
+        }
+    }
+
+    #[test]
+    fn queue_enforces_count_and_byte_ceilings() {
+        let mut queue = MrfQueue::new(2, usize::MAX);
+        assert!(queue.try_push(intent("b", "o", 0)));
+        assert!(queue.try_push(intent("b", "o", 0)));
+        assert!(!queue.try_push(intent("b", "o", 0)), "count ceiling must drop");
+
+        let mut tiny = MrfQueue::new(usize::MAX, intent("bucket", "object", 0).estimated_bytes());
+        assert!(tiny.try_push(intent("bucket", "object", 0)));
+        assert!(
+            !tiny.try_push(intent("bucket", "object", 0)),
+            "byte budget must drop before the second intent fits"
+        );
+    }
+
+    #[test]
+    fn journal_roundtrip_preserves_intents() {
+        let intents = vec![
+            intent("bucket-a", "object/a", 0),
+            intent("bucket-b", "object/b", 2),
+            MrfIntent {
+                bucket: StdArc::from("bucket-c"),
+                object: StdArc::from("object/c"),
+                version_id: None,
+                kind: MrfKind::MetadataCorruption,
+                enqueued_at_ms: 5,
+                attempts: 1,
+            },
+        ];
+        let mut buf = Vec::new();
+        for intent in &intents {
+            encode_intent(intent, &mut buf);
+        }
+        let (decoded, truncated) = decode_journal(&buf);
+        assert_eq!(truncated, 0);
+        assert_eq!(decoded.len(), intents.len());
+        for (left, right) in decoded.iter().zip(intents.iter()) {
+            assert_eq!(left.bucket, right.bucket);
+            assert_eq!(left.object, right.object);
+            assert_eq!(left.version_id, right.version_id);
+            assert_eq!(left.kind, right.kind);
+            assert_eq!(left.attempts, right.attempts);
+        }
+    }
+
+    #[test]
+    fn journal_torn_tail_is_truncated() {
+        let mut buf = Vec::new();
+        encode_intent(&intent("b", "o", 0), &mut buf);
+        let mut torn = buf.clone();
+        torn.extend_from_slice(&buf[..buf.len() / 2]);
+
+        let (decoded, truncated) = decode_journal(&torn);
+        assert_eq!(decoded.len(), 1, "the intact record must survive");
+        assert!(truncated > 0, "the partial tail must be discarded");
+
+        // A corrupted body (CRC mismatch) also truncates from that record on.
+        let mut corrupt = buf.clone();
+        let mid = MRF_RECORD_FIXED_HEAD + 4;
+        corrupt[mid] ^= 0xff;
+        let (decoded, truncated) = decode_journal(&corrupt);
+        assert!(decoded.is_empty());
+        assert_eq!(truncated, corrupt.len());
+    }
+
+    #[test]
+    fn heal_request_mapping_follows_priority_matrix() {
+        let decode = build_heal_request(&intent("b", "o", 0));
+        assert!(matches!(decode.heal_type, HealType::ECDecode { .. }));
+        assert_eq!(decode.priority, HealPriority::Urgent);
+
+        let metadata = build_heal_request(&MrfIntent {
+            bucket: StdArc::from("b"),
+            object: StdArc::from("o"),
+            version_id: None,
+            kind: MrfKind::MetadataCorruption,
+            enqueued_at_ms: 0,
+            attempts: 0,
+        });
+        assert!(matches!(metadata.heal_type, HealType::Metadata { .. }));
+        assert_eq!(metadata.priority, HealPriority::High);
+
+        let partial = build_heal_request(&MrfIntent {
+            bucket: StdArc::from("b"),
+            object: StdArc::from("o"),
+            version_id: None,
+            kind: MrfKind::PartialWrite,
+            enqueued_at_ms: 0,
+            attempts: 0,
+        });
+        assert!(matches!(partial.heal_type, HealType::Object { .. }));
+        assert_eq!(partial.priority, HealPriority::Normal);
+    }
+}

crates/heal/src/heal/progress.rs

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 use serde::{Deserialize, Serialize};
-use std::time::SystemTime;
+use std::time::{Duration, SystemTime};
 
 #[derive(Debug, Default, Clone, Serialize, Deserialize)]
 #[serde(rename_all = "camelCase")]
@@ -24,6 +24,14 @@ pub struct HealProgress {
     pub objects_healed: u64,
     /// Objects failed
     pub objects_failed: u64,
+    /// Versions skipped because they were written after this heal started
+    pub skipped_new_versions: u64,
+    /// Versions skipped because lifecycle already selected them for expiry
+    pub skipped_ilm_expired: u64,
+    /// Baseline object count from the latest complete usage snapshot
+    pub objects_total_count: u64,
+    /// Baseline object bytes from the latest complete usage snapshot
+    pub objects_total_size: u64,
     /// Bytes processed
     pub bytes_processed: u64,
     /// Current object
@@ -54,10 +62,56 @@ impl HealProgress {
         self.bytes_processed = bytes;
         self.last_update_time = Some(SystemTime::now());
 
-        // calculate progress percentage
-        let total = scanned + healed + failed;
+        self.refresh_progress_percentage();
+        self.refresh_estimated_completion_time();
+    }
+
+    pub fn set_total_baseline(&mut self, objects_total_count: u64, objects_total_size: u64) {
+        self.objects_total_count = objects_total_count;
+        self.objects_total_size = objects_total_size;
+        self.last_update_time = Some(SystemTime::now());
+        self.refresh_progress_percentage();
+        self.refresh_estimated_completion_time();
+    }
+
+    pub fn record_skipped_new_version(&mut self) {
+        self.skipped_new_versions = self.skipped_new_versions.saturating_add(1);
+        self.last_update_time = Some(SystemTime::now());
+        self.refresh_progress_percentage();
+        self.refresh_estimated_completion_time();
+    }
+
+    pub fn record_skipped_ilm_expired(&mut self) {
+        self.skipped_ilm_expired = self.skipped_ilm_expired.saturating_add(1);
+        self.last_update_time = Some(SystemTime::now());
+        self.refresh_progress_percentage();
+        self.refresh_estimated_completion_time();
+    }
+
+    fn completed_for_baseline(&self) -> u64 {
+        self.objects_healed
+            .saturating_add(self.objects_failed)
+            .saturating_add(self.skipped_new_versions)
+            .saturating_add(self.skipped_ilm_expired)
+    }
+
+    pub(crate) fn refresh_progress_percentage(&mut self) {
+        if self.objects_total_size > 0 {
+            self.progress_percentage = ((self.bytes_processed as f64 / self.objects_total_size as f64) * 100.0).min(100.0);
+            return;
+        }
+        if self.objects_total_count > 0 {
+            let completed = self.completed_for_baseline();
+            self.progress_percentage = ((completed as f64 / self.objects_total_count as f64) * 100.0).min(100.0);
+            return;
+        }
+
+        let total = self
+            .objects_scanned
+            .saturating_add(self.objects_healed)
+            .saturating_add(self.objects_failed);
         if total > 0 {
-            self.progress_percentage = (healed as f64 / total as f64) * 100.0;
+            self.progress_percentage = (self.objects_healed as f64 / total as f64) * 100.0;
         }
     }
 
@@ -66,9 +120,36 @@ impl HealProgress {
         self.last_update_time = Some(SystemTime::now());
     }
 
+    pub fn refresh_estimated_completion_time(&mut self) {
+        let Some(start_time) = self.start_time else {
+            self.estimated_completion_time = None;
+            return;
+        };
+        if self.is_completed() || !(0.0..100.0).contains(&self.progress_percentage) || self.bytes_processed == 0 {
+            self.estimated_completion_time = None;
+            return;
+        }
+
+        let elapsed = match SystemTime::now().duration_since(start_time) {
+            Ok(elapsed) if !elapsed.is_zero() => elapsed,
+            _ => {
+                self.estimated_completion_time = None;
+                return;
+            }
+        };
+        let estimated_total_secs = elapsed.as_secs_f64() * 100.0 / self.progress_percentage;
+        self.estimated_completion_time = start_time.checked_add(Duration::from_secs_f64(estimated_total_secs));
+    }
+
     pub fn is_completed(&self) -> bool {
-        self.progress_percentage >= 100.0
-            || self.objects_scanned > 0 && self.objects_healed + self.objects_failed >= self.objects_scanned
+        if self.progress_percentage >= 100.0 {
+            return true;
+        }
+        if self.objects_total_count > 0 || self.objects_total_size > 0 {
+            return false;
+        }
+
+        self.objects_scanned > 0 && self.objects_healed.saturating_add(self.objects_failed) >= self.objects_scanned
     }
 
     pub fn get_success_rate(&self) -> f64 {
@@ -158,6 +239,10 @@ mod tests {
         assert_eq!(progress.objects_scanned, 0);
         assert_eq!(progress.objects_healed, 0);
         assert_eq!(progress.objects_failed, 0);
+        assert_eq!(progress.skipped_new_versions, 0);
+        assert_eq!(progress.skipped_ilm_expired, 0);
+        assert_eq!(progress.objects_total_count, 0);
+        assert_eq!(progress.objects_total_size, 0);
         assert_eq!(progress.bytes_processed, 0);
         assert_eq!(progress.progress_percentage, 0.0);
         assert!(progress.start_time.is_some());
@@ -181,6 +266,73 @@ mod tests {
         assert!(progress.last_update_time.is_some());
     }
 
+    #[test]
+    fn test_heal_progress_estimates_completion_time_from_progress() {
+        let mut progress = HealProgress::new();
+        progress.start_time = Some(SystemTime::now() - Duration::from_secs(10));
+
+        progress.update_progress(100, 25, 0, 4096);
+
+        let eta = progress
+            .estimated_completion_time
+            .expect("partial byte progress should estimate completion");
+        assert!(eta > SystemTime::now());
+    }
+
+    #[test]
+    fn test_heal_progress_uses_byte_baseline_for_percentage() {
+        let mut progress = HealProgress::new();
+        progress.set_total_baseline(10, 8192);
+
+        progress.update_progress(100, 25, 0, 4096);
+
+        assert!((progress.progress_percentage - 50.0).abs() < 0.001);
+    }
+
+    #[test]
+    fn test_heal_progress_uses_object_baseline_when_bytes_unknown() {
+        let mut progress = HealProgress::new();
+        progress.set_total_baseline(10, 0);
+
+        progress.update_progress(100, 3, 2, 0);
+
+        assert!((progress.progress_percentage - 50.0).abs() < 0.001);
+    }
+
+    #[test]
+    fn test_heal_progress_counts_skipped_versions_for_object_baseline() {
+        let mut progress = HealProgress::new();
+        progress.set_total_baseline(10, 0);
+
+        progress.update_progress(100, 3, 2, 0);
+        progress.record_skipped_new_version();
+
+        assert_eq!(progress.skipped_new_versions, 1);
+        assert!((progress.progress_percentage - 60.0).abs() < 0.001);
+    }
+
+    #[test]
+    fn test_heal_progress_does_not_estimate_completion_without_bytes() {
+        let mut progress = HealProgress::new();
+        progress.start_time = Some(SystemTime::now() - Duration::from_secs(10));
+
+        progress.update_progress(100, 25, 0, 0);
+
+        assert!(progress.estimated_completion_time.is_none());
+    }
+
+    #[test]
+    fn test_heal_progress_with_baseline_is_not_completed_by_processed_count() {
+        let mut progress = HealProgress::new();
+        progress.start_time = Some(SystemTime::now() - Duration::from_secs(10));
+        progress.set_total_baseline(10, 8192);
+
+        progress.update_progress(1, 1, 0, 1024);
+
+        assert!(!progress.is_completed());
+        assert!(progress.estimated_completion_time.is_some());
+    }
+
     #[test]
     fn test_heal_progress_update_progress_zero_total() {
         let mut progress = HealProgress::new();
@@ -251,6 +403,8 @@ mod tests {
         assert_eq!(json["objectsScanned"], 10);
         assert_eq!(json["objectsHealed"], 8);
         assert_eq!(json["objectsFailed"], 2);
+        assert_eq!(json["skippedNewVersions"], 0);
+        assert_eq!(json["skippedIlmExpired"], 0);
         assert_eq!(json["bytesProcessed"], 1024);
         assert_eq!(json["currentObject"], "test-bucket/test-object");
         assert!(json["progressPercentage"].is_number());

crates/heal/src/heal/storage.rs

@@ -22,6 +22,7 @@ use serde::{Deserialize, Serialize};
 use std::sync::Arc;
 use tracing::{debug, error, warn};
 
+use super::storage_api::owner::{EcstoreHealLifecycleExpiryContext, ecstore_load_admin_data_usage_from_backend_cached};
 use super::storage_api::storage::{
     BucketInfo, BucketOperations, DiskSetSelector, HealOperations as _, ListOperations as _, ObjectIO as _,
     ObjectOperations as _, StorageAdminApi,
@@ -29,6 +30,37 @@ use super::storage_api::storage::{
 use super::{DiskStore, ECStore, Endpoint, HealDiskExt as _, StorageError, resume::ReplacementTargetIdentity};
 pub use super::{HealObjectInfo, HealObjectOptions, HealPutObjReader};
 
+#[derive(Debug, Clone, Copy, Default, PartialEq, Eq)]
+pub struct HealBucketUsageBaseline {
+    pub objects_count: u64,
+    pub bytes: u64,
+}
+
+pub struct HealLifecycleExpiryContext {
+    inner: HealLifecycleExpiryContextInner,
+}
+
+enum HealLifecycleExpiryContextInner {
+    Ecstore(EcstoreHealLifecycleExpiryContext),
+    #[allow(dead_code)]
+    Test,
+}
+
+impl HealLifecycleExpiryContext {
+    fn ecstore(inner: EcstoreHealLifecycleExpiryContext) -> Self {
+        Self {
+            inner: HealLifecycleExpiryContextInner::Ecstore(inner),
+        }
+    }
+
+    #[cfg(test)]
+    pub(crate) fn test() -> Self {
+        Self {
+            inner: HealLifecycleExpiryContextInner::Test,
+        }
+    }
+}
+
 const LOG_COMPONENT_HEAL: &str = "heal";
 const LOG_SUBSYSTEM_STORAGE: &str = "storage";
 const EVENT_HEAL_STORAGE_OBJECT_IO: &str = "heal_storage_object_io";
@@ -272,6 +304,10 @@ pub struct HealListItem {
     pub name: String,
     /// normalized version id (`None` when the version is nil/absent)
     pub version_id: Option<String>,
+    /// version modification time as Unix nanoseconds
+    pub mod_time_unix_nanos: Option<i128>,
+    /// object snapshot for lifecycle evaluation
+    pub lifecycle_object_info: Option<HealObjectInfo>,
     /// whether this version is a delete marker (observability only)
     pub is_delete_marker: bool,
 }
@@ -329,6 +365,28 @@ pub trait HealStorageAPI: Send + Sync {
     /// Get bucket info
     async fn get_bucket_info(&self, bucket: &str) -> Result<Option<BucketInfo>>;
 
+    /// Aggregate usage-cache baselines for the requested buckets.
+    async fn erasure_set_usage_baseline(&self, _buckets: &[String]) -> Result<Option<HealBucketUsageBaseline>> {
+        Ok(None)
+    }
+
+    /// Load per-bucket lifecycle expiry context for heal skips.
+    async fn load_heal_lifecycle_expiry_context(&self, _bucket: &str) -> Result<Option<HealLifecycleExpiryContext>> {
+        Ok(None)
+    }
+
+    /// Queue lifecycle expiry for a version that heal can skip.
+    async fn enqueue_heal_lifecycle_expiry(
+        &self,
+        _context: &HealLifecycleExpiryContext,
+        _bucket: &str,
+        _object: &str,
+        _version_id: Option<&str>,
+        _object_info: Option<&HealObjectInfo>,
+    ) -> Result<bool> {
+        Ok(false)
+    }
+
     /// Fix bucket metadata
     async fn heal_bucket_metadata(&self, bucket: &str) -> Result<()>;
 
@@ -409,6 +467,7 @@ pub trait HealStorageAPI: Send + Sync {
         bucket: &str,
         prefix: &str,
         continuation_token: Option<&str>,
+        include_lifecycle_object_info: bool,
     ) -> Result<(Vec<HealListItem>, Option<String>, bool)>;
 
     /// List versions for healing via a per-erasure-set DISK-WALK union enumerator
@@ -427,8 +486,10 @@ pub trait HealStorageAPI: Send + Sync {
         bucket: &str,
         prefix: &str,
         continuation_token: Option<&str>,
+        include_lifecycle_object_info: bool,
     ) -> Result<(Vec<HealListItem>, Option<String>, bool)> {
-        self.list_objects_for_heal_page(bucket, prefix, continuation_token).await
+        self.list_objects_for_heal_page(bucket, prefix, continuation_token, include_lifecycle_object_info)
+            .await
     }
 
     /// Get disk for resume functionality.
@@ -1021,6 +1082,85 @@ impl HealStorageAPI for ECStoreHealStorage {
         }
     }
 
+    async fn erasure_set_usage_baseline(&self, buckets: &[String]) -> Result<Option<HealBucketUsageBaseline>> {
+        if buckets.is_empty() {
+            return Ok(None);
+        }
+
+        let info = match ecstore_load_admin_data_usage_from_backend_cached(self.ecstore.clone()).await {
+            Ok(info) if info.is_complete_bucket_usage_snapshot() => info,
+            Ok(_) | Err(_) => return Ok(None),
+        };
+
+        let mut baseline = HealBucketUsageBaseline::default();
+        for bucket in buckets {
+            if let Some(usage) = info.buckets_usage.get(bucket) {
+                baseline.objects_count = baseline.objects_count.saturating_add(usage.objects_count);
+                baseline.bytes = baseline.bytes.saturating_add(usage.size);
+            }
+        }
+
+        Ok(Some(baseline))
+    }
+
+    async fn load_heal_lifecycle_expiry_context(&self, bucket: &str) -> Result<Option<HealLifecycleExpiryContext>> {
+        match self.ecstore.load_heal_lifecycle_expiry_context(bucket).await {
+            Ok(Some(context)) => Ok(Some(HealLifecycleExpiryContext::ecstore(context))),
+            Ok(None) => Ok(None),
+            Err(err) => {
+                debug!(
+                    target: "rustfs::heal::storage",
+                    event = EVENT_HEAL_STORAGE_ADMIN_OP,
+                    component = LOG_COMPONENT_HEAL,
+                    subsystem = LOG_SUBSYSTEM_STORAGE,
+                    operation = "load_heal_lifecycle_expiry_context",
+                    bucket,
+                    result = "failed",
+                    error = %err,
+                    "Heal storage lifecycle expiry context load failed"
+                );
+                Ok(None)
+            }
+        }
+    }
+
+    async fn enqueue_heal_lifecycle_expiry(
+        &self,
+        context: &HealLifecycleExpiryContext,
+        bucket: &str,
+        object: &str,
+        version_id: Option<&str>,
+        object_info: Option<&HealObjectInfo>,
+    ) -> Result<bool> {
+        let context = match &context.inner {
+            HealLifecycleExpiryContextInner::Ecstore(context) => context,
+            HealLifecycleExpiryContextInner::Test => return Ok(false),
+        };
+        match self
+            .ecstore
+            .enqueue_heal_lifecycle_expiry(context, bucket, object, version_id, object_info)
+            .await
+        {
+            Ok(queued) => Ok(queued),
+            Err(err) => {
+                debug!(
+                    target: "rustfs::heal::storage",
+                    event = EVENT_HEAL_STORAGE_ADMIN_OP,
+                    component = LOG_COMPONENT_HEAL,
+                    subsystem = LOG_SUBSYSTEM_STORAGE,
+                    operation = "enqueue_heal_lifecycle_expiry",
+                    bucket,
+                    object,
+                    version_id = ?version_id,
+                    result = "failed",
+                    error = %err,
+                    "Heal storage lifecycle expiry check failed"
+                );
+                Ok(false)
+            }
+        }
+    }
+
     async fn heal_bucket_metadata(&self, bucket: &str) -> Result<()> {
         debug!(
             target: "rustfs::heal::storage",
@@ -1436,7 +1576,7 @@ impl HealStorageAPI for ECStoreHealStorage {
 
         loop {
             let (page_objects, next_token, is_truncated) = self
-                .list_objects_for_heal_page(bucket, prefix, continuation_token.as_deref())
+                .list_objects_for_heal_page(bucket, prefix, continuation_token.as_deref(), false)
                 .await?;
 
             all_objects.extend(page_objects);
@@ -1471,6 +1611,7 @@ impl HealStorageAPI for ECStoreHealStorage {
         bucket: &str,
         prefix: &str,
         continuation_token: Option<&str>,
+        include_lifecycle_object_info: bool,
     ) -> Result<(Vec<HealListItem>, Option<String>, bool)> {
         debug!(
             target: "rustfs::heal::storage",
@@ -1522,10 +1663,19 @@ impl HealStorageAPI for ECStoreHealStorage {
         let page_objects: Vec<HealListItem> = list_info
             .objects
             .into_iter()
-            .map(|obj| HealListItem {
-                name: obj.name,
-                version_id: obj.version_id.filter(|u| !u.is_nil()).map(|u| u.to_string()),
-                is_delete_marker: obj.delete_marker,
+            .map(|mut obj| {
+                obj.version_id = obj.version_id.filter(|u| !u.is_nil());
+                let version_id = obj.version_id.map(|u| u.to_string());
+                let mod_time_unix_nanos = obj.mod_time.map(|mod_time| mod_time.unix_timestamp_nanos());
+                let is_delete_marker = obj.delete_marker;
+                let lifecycle_object_info = include_lifecycle_object_info.then(|| obj.clone());
+                HealListItem {
+                    name: obj.name,
+                    version_id,
+                    mod_time_unix_nanos,
+                    lifecycle_object_info,
+                    is_delete_marker,
+                }
             })
             .collect();
         let page_count = page_objects.len();
@@ -1562,6 +1712,7 @@ impl HealStorageAPI for ECStoreHealStorage {
         bucket: &str,
         prefix: &str,
         continuation_token: Option<&str>,
+        include_lifecycle_object_info: bool,
     ) -> Result<(Vec<HealListItem>, Option<String>, bool)> {
         // Per-page bounds for the disk-walk union enumerator. Objects are atomic
         // (never split across pages), so version_budget only bounds how many
@@ -1590,7 +1741,16 @@ impl HealStorageAPI for ECStoreHealStorage {
 
         let (versions, next_forward, is_truncated) = self
             .ecstore
-            .heal_walk_versions_page(pool_idx, set_idx, bucket, prefix, forward_to.as_deref(), BATCH_OBJECTS, VERSION_BUDGET)
+            .heal_walk_versions_page(
+                pool_idx,
+                set_idx,
+                bucket,
+                prefix,
+                forward_to.as_deref(),
+                BATCH_OBJECTS,
+                VERSION_BUDGET,
+                include_lifecycle_object_info,
+            )
             .await
             .map_err(|e| {
                 error!(
@@ -1614,6 +1774,8 @@ impl HealStorageAPI for ECStoreHealStorage {
             .map(|v| HealListItem {
                 name: v.name,
                 version_id: v.version_id,
+                mod_time_unix_nanos: v.mod_time_unix_nanos,
+                lifecycle_object_info: v.lifecycle_object_info,
                 is_delete_marker: v.is_delete_marker,
             })
             .collect();

crates/heal/src/heal/storage_api.rs

@@ -12,7 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-pub(crate) use rustfs_ecstore::api::data_usage::DATA_USAGE_CACHE_NAME as ECSTORE_DATA_USAGE_CACHE_NAME;
+pub(crate) use rustfs_ecstore::api::data_usage::{
+    DATA_USAGE_CACHE_NAME as ECSTORE_DATA_USAGE_CACHE_NAME,
+    load_admin_data_usage_from_backend_cached as ecstore_load_admin_data_usage_from_backend_cached,
+};
 pub(crate) use rustfs_ecstore::api::disk::endpoint::Endpoint as EcstoreEndpoint;
 pub(crate) use rustfs_ecstore::api::disk::error::{DiskError as EcstoreDiskError, Result as EcstoreDiskResult};
 pub(crate) use rustfs_ecstore::api::disk::{
@@ -25,7 +28,9 @@ pub(crate) use rustfs_ecstore::api::disk::{
 pub(crate) use rustfs_ecstore::api::disk::{DiskOption as EcstoreDiskOption, new_disk as ecstore_new_disk};
 pub(crate) use rustfs_ecstore::api::error::{Error as EcstoreErrorType, StorageError as EcstoreStorageError};
 pub(crate) use rustfs_ecstore::api::runtime::local_disk_map_read as ecstore_local_disk_map_read;
-pub(crate) use rustfs_ecstore::api::storage::ECStore as EcstoreStore;
+pub(crate) use rustfs_ecstore::api::storage::{
+    ECStore as EcstoreStore, HealLifecycleExpiryContext as EcstoreHealLifecycleExpiryContext,
+};
 use rustfs_storage_api as storage_contracts;
 
 pub(crate) mod owner {
@@ -34,8 +39,8 @@ pub(crate) mod owner {
     pub(crate) use super::{
         ECSTORE_BUCKET_META_PREFIX, ECSTORE_DATA_USAGE_CACHE_NAME, ECSTORE_HEALING_MARKER_PATH, ECSTORE_RUSTFS_META_BUCKET,
         EcstoreConditionalFileUpdate, EcstoreDeleteOptions, EcstoreDiskAPI, EcstoreDiskBytes, EcstoreDiskError,
-        EcstoreDiskResult, EcstoreDiskStore, EcstoreEndpoint, EcstoreErrorType, EcstoreStorageError, EcstoreStore,
-        ecstore_local_disk_map_read,
+        EcstoreDiskResult, EcstoreDiskStore, EcstoreEndpoint, EcstoreErrorType, EcstoreHealLifecycleExpiryContext,
+        EcstoreStorageError, EcstoreStore, ecstore_load_admin_data_usage_from_backend_cached, ecstore_local_disk_map_read,
     };
 
     #[cfg(test)]

crates/heal/src/heal/task.rs

@@ -19,11 +19,12 @@ use crate::heal::{
     resume::{
         CheckpointManager, ReplacementPhase, ReplacementTargetIdentity, ResumeManager, replacement_target_identities_match,
     },
-    storage::{HealStorageAPI, next_heal_listing_token},
+    storage::{HealBucketUsageBaseline, HealStorageAPI, next_heal_listing_token},
 };
 use crate::{Error, Result};
 use metrics::{counter, histogram};
 use rustfs_common::heal_channel::{HealOpts, HealRequestSource, HealScanMode};
+use rustfs_common::trace_bus::{TraceEvent, TraceFunc, TraceKind, trace_emit};
 use rustfs_madmin::heal_commands::HealResultItem;
 use rustfs_utils::path::SLASH_SEPARATOR;
 use serde::{Deserialize, Serialize};
@@ -178,6 +179,17 @@ pub enum HealPriority {
     Urgent = 3,
 }
 
+impl HealPriority {
+    fn as_str(self) -> &'static str {
+        match self {
+            Self::Low => "low",
+            Self::Normal => "normal",
+            Self::High => "high",
+            Self::Urgent => "urgent",
+        }
+    }
+}
+
 /// Heal options
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct HealOptions {
@@ -498,6 +510,61 @@ impl HealTask {
         }
     }
 
+    fn emit_trace_task_state(&self, state: &'static str, duration: Duration, error: Option<&Error>) {
+        trace_emit(|| {
+            let mut event = TraceEvent::new(TraceKind::Heal, TraceFunc::HealTask)
+                .with_duration(duration)
+                .with_attr("task_id", self.id.as_str())
+                .with_attr("heal_type", self.heal_type.log_kind())
+                .with_attr("state", state)
+                .with_attr("source", self.source.as_str())
+                .with_attr("priority", self.priority.as_str())
+                .with_attr("retry_attempts", u64::from(self.retry_attempts))
+                .with_attr("dry_run", self.options.dry_run);
+
+            event = match &self.heal_type {
+                HealType::Cluster => event,
+                HealType::Object {
+                    bucket,
+                    object,
+                    version_id,
+                } => {
+                    let event = event.with_bucket(bucket.as_str()).with_object(object.as_str());
+                    match version_id {
+                        Some(version_id) => event.with_attr("version_id", version_id.as_str()),
+                        None => event,
+                    }
+                }
+                HealType::Bucket { bucket } => event.with_bucket(bucket.as_str()),
+                HealType::Prefix { bucket, prefix } => event.with_bucket(bucket.as_str()).with_object(prefix.as_str()),
+                HealType::ErasureSet { buckets, set_disk_id } => {
+                    let bucket_count = u64::try_from(buckets.len()).unwrap_or(u64::MAX);
+                    event
+                        .with_attr("set_disk_id", set_disk_id.as_str())
+                        .with_attr("bucket_count", bucket_count)
+                }
+                HealType::Metadata { bucket, object } => event.with_bucket(bucket.as_str()).with_object(object.as_str()),
+                HealType::ECDecode {
+                    bucket,
+                    object,
+                    version_id,
+                } => {
+                    let event = event.with_bucket(bucket.as_str()).with_object(object.as_str());
+                    match version_id {
+                        Some(version_id) => event.with_attr("version_id", version_id.as_str()),
+                        None => event,
+                    }
+                }
+                HealType::MRF { meta_path } => event.with_object(meta_path.as_str()),
+            };
+
+            match error {
+                Some(error) => event.with_attr("error", error.to_string()),
+                None => event,
+            }
+        });
+    }
+
     async fn remaining_timeout(&self) -> Result<Option<Duration>> {
         if let Some(total) = self.options.timeout {
             let start_instant = { *self.task_start_instant.read().await };
@@ -717,6 +784,7 @@ impl HealTask {
             queue_delay = ?queue_delay,
             "Heal task started"
         });
+        self.emit_trace_task_state("started", Duration::ZERO, None);
 
         let result = match &self.heal_type {
             HealType::Cluster => self.heal_cluster().await,
@@ -805,6 +873,14 @@ impl HealTask {
             }
         }
 
+        let terminal_state = match &result {
+            Ok(_) => "completed",
+            Err(Error::TaskCancelled) => "cancelled",
+            Err(Error::TaskTimeout) => "timed_out",
+            Err(_) => "failed",
+        };
+        self.emit_trace_task_state(terminal_state, start_instant.elapsed(), result.as_ref().err());
+
         result
     }
 
@@ -1535,7 +1611,7 @@ impl HealTask {
             let (objects, next_token, is_truncated) = self
                 .await_with_control(
                     self.storage
-                        .list_objects_for_heal_page(bucket, prefix, continuation_token.as_deref()),
+                        .list_objects_for_heal_page(bucket, prefix, continuation_token.as_deref(), false),
                 )
                 .await?;
 
@@ -1697,6 +1773,23 @@ impl HealTask {
         Ok(())
     }
 
+    async fn apply_erasure_set_usage_baseline(&self, buckets: &[String]) -> Result<()> {
+        let baseline = match self
+            .await_with_control(self.storage.erasure_set_usage_baseline(buckets))
+            .await
+        {
+            Ok(Some(baseline)) => baseline,
+            Ok(None) => return Ok(()),
+            Err(err @ Error::TaskCancelled) | Err(err @ Error::TaskTimeout) => return Err(err),
+            Err(_) => return Ok(()),
+        };
+
+        let HealBucketUsageBaseline { objects_count, bytes } = baseline;
+        let mut progress = self.progress.write().await;
+        progress.set_total_baseline(objects_count, bytes);
+        Ok(())
+    }
+
     async fn heal_metadata(&self, bucket: &str, object: &str) -> Result<()> {
         debug!(
             target: "rustfs::heal::task",
@@ -2298,6 +2391,8 @@ impl HealTask {
             None
         };
 
+        self.apply_erasure_set_usage_baseline(&buckets).await?;
+
         let healing_marker = format!("{set_disk_id}:{}", self.id);
         if let Some((disk, resume_manager, _)) = replacement_resume.as_ref() {
             let state = resume_manager.get_state().await;
@@ -2602,7 +2697,8 @@ impl HealTask {
 
         {
             let mut progress = self.progress.write().await;
-            progress.update_progress(4, 4, 0, 0);
+            let bytes_processed = progress.bytes_processed;
+            progress.update_progress(4, 4, 0, bytes_processed);
         }
 
         match result {
@@ -2658,6 +2754,7 @@ mod tests {
     use super::super::{DiskOption, DiskStore, Endpoint, HealDiskExt as _, new_disk};
     use super::*;
     use crate::heal::storage::{DiskStatus, HealListItem, HealObjectInfo};
+    use rustfs_common::trace_bus::{TraceEvent, TraceFunc, TraceKind, TraceSubscription, TraceVal, subscribe_trace_events};
     use rustfs_madmin::heal_commands::{HealDriveInfo, HealResultItem, Infos};
     use std::collections::{HashMap, VecDeque};
     use std::sync::Mutex;
@@ -3203,6 +3300,8 @@ mod tests {
         block_heal_object: Mutex<bool>,
         resume_disk: Mutex<Option<DiskStore>>,
         replacement_resume_disk: Mutex<Option<DiskStore>>,
+        usage_baseline: Mutex<Option<HealBucketUsageBaseline>>,
+        usage_baseline_error: Mutex<bool>,
     }
 
     #[test]
@@ -3265,11 +3364,69 @@ mod tests {
         assert_eq!(samples_logged, MAX_BUCKET_FAILURE_LOG_SAMPLES);
     }
 
+    #[tokio::test]
+    async fn execute_emits_heal_trace_task_state() {
+        let mut trace = subscribe_trace_events();
+        let storage = Arc::new(MockStorage::default());
+        let task = HealTask::from_request(
+            HealRequest::object("bucket-a".to_string(), "object-a".to_string(), Some("version-a".to_string())),
+            storage,
+        );
+
+        task.execute().await.expect("mock object heal should complete");
+
+        let started = recv_trace_task_state(&mut trace, &task.id, "started").await;
+        assert_eq!(started.kind, TraceKind::Heal);
+        assert_eq!(started.func, TraceFunc::HealTask);
+        assert_eq!(started.bucket.as_deref(), Some("bucket-a"));
+        assert_eq!(started.object.as_deref(), Some("object-a"));
+        assert_eq!(trace_attr_string(&started, "heal_type").as_deref(), Some("object"));
+        assert_eq!(trace_attr_string(&started, "source").as_deref(), Some("internal"));
+        assert_eq!(trace_attr_string(&started, "version_id").as_deref(), Some("version-a"));
+
+        let completed = recv_trace_task_state(&mut trace, &task.id, "completed").await;
+        assert_eq!(completed.kind, TraceKind::Heal);
+        assert_eq!(completed.func, TraceFunc::HealTask);
+        assert_eq!(trace_attr_string(&completed, "state").as_deref(), Some("completed"));
+    }
+
+    async fn recv_trace_task_state(trace: &mut TraceSubscription, task_id: &str, state: &str) -> TraceEvent {
+        for _ in 0..32 {
+            let event = tokio::time::timeout(Duration::from_secs(1), trace.recv())
+                .await
+                .expect("trace event should arrive")
+                .expect("trace bus should stay open");
+            if trace_attr_string(&event, "task_id").as_deref() == Some(task_id)
+                && trace_attr_string(&event, "state").as_deref() == Some(state)
+            {
+                return (*event).clone();
+            }
+        }
+
+        panic!("expected trace state {state} for task {task_id}");
+    }
+
+    fn trace_attr_string(event: &TraceEvent, key: &str) -> Option<String> {
+        event.attrs.iter().find_map(|attr| {
+            if attr.key != key {
+                return None;
+            }
+            Some(match &attr.value {
+                TraceVal::Bool(value) => value.to_string(),
+                TraceVal::U64(value) => value.to_string(),
+                TraceVal::I64(value) => value.to_string(),
+                TraceVal::Str(value) => value.to_string(),
+            })
+        })
+    }
+
     /// Build a latest, non-delete-marker heal list item with no version id.
     fn heal_item(name: &str) -> HealListItem {
         HealListItem {
             name: name.to_string(),
             version_id: None,
+            mod_time_unix_nanos: None,
+            lifecycle_object_info: None,
             is_delete_marker: false,
         }
     }
@@ -3357,6 +3514,13 @@ mod tests {
             }))
         }
 
+        async fn erasure_set_usage_baseline(&self, _buckets: &[String]) -> Result<Option<HealBucketUsageBaseline>> {
+            if *self.usage_baseline_error.lock().unwrap() {
+                return Err(Error::Other("usage baseline unavailable".to_string()));
+            }
+            Ok(*self.usage_baseline.lock().unwrap())
+        }
+
         async fn heal_bucket_metadata(&self, _bucket: &str) -> Result<()> {
             Ok(())
         }
@@ -3540,6 +3704,7 @@ mod tests {
             bucket: &str,
             prefix: &str,
             continuation_token: Option<&str>,
+            _include_lifecycle_object_info: bool,
         ) -> Result<(Vec<HealListItem>, Option<String>, bool)> {
             self.listed_prefixes.lock().unwrap().push(prefix.to_string());
             if *self.truncate_without_token.lock().unwrap() {
@@ -4654,6 +4819,73 @@ mod tests {
         assert!(storage.object_heal_opts.lock().unwrap().is_empty());
     }
 
+    #[tokio::test]
+    async fn erasure_set_heal_applies_usage_baseline_to_progress() {
+        let temp = TempDir::new().expect("temporary directory should be created");
+        let disk = make_resume_disk(&temp).await;
+        let storage = Arc::new(MockStorage {
+            resume_disk: Mutex::new(Some(disk)),
+            usage_baseline: Mutex::new(Some(HealBucketUsageBaseline {
+                objects_count: 10,
+                bytes: 8,
+            })),
+            ..Default::default()
+        });
+        let request = HealRequest::new(
+            HealType::ErasureSet {
+                buckets: vec!["bucket-a".to_string()],
+                set_disk_id: "pool_0_set_0".to_string(),
+            },
+            HealOptions {
+                timeout: None,
+                ..Default::default()
+            },
+            HealPriority::Normal,
+        );
+        let task = HealTask::from_request(request, storage);
+
+        task.heal_erasure_set(vec!["bucket-a".to_string()], "pool_0_set_0".to_string())
+            .await
+            .expect("erasure set heal should complete");
+
+        let progress = task.get_progress().await;
+        assert_eq!(progress.objects_total_count, 10);
+        assert_eq!(progress.objects_total_size, 8);
+        assert_eq!(progress.bytes_processed, 2);
+        assert!((progress.progress_percentage - 25.0).abs() < 0.001);
+    }
+
+    #[tokio::test]
+    async fn erasure_set_heal_ignores_usage_baseline_errors() {
+        let temp = TempDir::new().expect("temporary directory should be created");
+        let disk = make_resume_disk(&temp).await;
+        let storage = Arc::new(MockStorage {
+            resume_disk: Mutex::new(Some(disk)),
+            usage_baseline_error: Mutex::new(true),
+            ..Default::default()
+        });
+        let request = HealRequest::new(
+            HealType::ErasureSet {
+                buckets: vec!["bucket-a".to_string()],
+                set_disk_id: "pool_0_set_0".to_string(),
+            },
+            HealOptions {
+                timeout: None,
+                ..Default::default()
+            },
+            HealPriority::Normal,
+        );
+        let task = HealTask::from_request(request, storage);
+
+        task.heal_erasure_set(vec!["bucket-a".to_string()], "pool_0_set_0".to_string())
+            .await
+            .expect("usage baseline failures should not fail erasure set heal");
+
+        let progress = task.get_progress().await;
+        assert_eq!(progress.objects_total_count, 0);
+        assert_eq!(progress.objects_total_size, 0);
+    }
+
     #[tokio::test]
     async fn resumable_erasure_set_execution_is_cancelled_while_object_heal_is_pending() {
         let temp = TempDir::new().expect("temporary directory should be created");

crates/heal/src/lib.rs

@@ -158,6 +158,10 @@ pub async fn init_heal_manager_with_workload_provider(
             return Err(err);
         }
 
+        // Start the MRF intent consumer (error-path repair intents + durable
+        // journal replay) now that the manager can accept submissions.
+        heal::mrf_queue::spawn_mrf_consumer(heal_manager.clone());
+
         #[cfg(test)]
         test_hook_after_manager_start().await;
 
@@ -445,6 +449,7 @@ mod tests {
             _bucket: &str,
             _prefix: &str,
             _continuation_token: Option<&str>,
+            _include_lifecycle_object_info: bool,
         ) -> Result<(Vec<HealListItem>, Option<String>, bool), Error> {
             Ok((Vec::new(), None, false))
         }

crates/heal/tests/heal_b5_versioned_regression_test.rs

@@ -176,7 +176,7 @@ async fn enumerate_all_versions(heal_storage: &Arc<ECStoreHealStorage>, bucket:
     let mut token: Option<String> = None;
     loop {
         let (page, next, truncated) = heal_storage
-            .list_objects_for_heal_page(bucket, "", token.as_deref())
+            .list_objects_for_heal_page(bucket, "", token.as_deref(), false)
             .await
             .expect("list_objects_for_heal_page failed");
         items.extend(page);

crates/heal/tests/heal_b920_subquorum_union_test.rs

@@ -166,7 +166,7 @@ async fn enumerate_b5(heal_storage: &Arc<ECStoreHealStorage>, bucket: &str) -> V
     let mut token: Option<String> = None;
     loop {
         let (page, next, truncated) = heal_storage
-            .list_objects_for_heal_page(bucket, "", token.as_deref())
+            .list_objects_for_heal_page(bucket, "", token.as_deref(), false)
             .await
             .expect("b5 list page failed");
         items.extend(page);
@@ -187,7 +187,7 @@ async fn enumerate_disk_walk(heal_storage: &Arc<ECStoreHealStorage>, bucket: &st
     let mut token: Option<String> = None;
     loop {
         let (page, next, truncated) = heal_storage
-            .list_versions_for_heal_page_disk_walk(SET_DISK_ID, bucket, "", token.as_deref())
+            .list_versions_for_heal_page_disk_walk(SET_DISK_ID, bucket, "", token.as_deref(), false)
             .await
             .expect("disk-walk list page failed");
         items.extend(page);
@@ -418,7 +418,7 @@ mod serial_tests {
         let mut pages = 0usize;
         loop {
             let (versions, next_forward, truncated) = ecstore
-                .heal_walk_versions_page(0, 0, bucket, "", forward.as_deref(), 2, 100_000)
+                .heal_walk_versions_page(0, 0, bucket, "", forward.as_deref(), 2, 100_000, false)
                 .await
                 .expect("heal_walk_versions_page failed");
             pages += 1;

crates/heal/tests/heal_bug_fixes_test.rs

@@ -242,6 +242,7 @@ fn test_heal_task_status_atomic_update() {
             _bucket: &str,
             _prefix: &str,
             _continuation_token: Option<&str>,
+            _include_lifecycle_object_info: bool,
         ) -> rustfs_heal::Result<(Vec<HealListItem>, Option<String>, bool)> {
             Ok((vec![], None, false))
         }
@@ -385,6 +386,7 @@ async fn test_heal_task_transient_object_exists_skip_avoids_recreate() {
             _bucket: &str,
             _prefix: &str,
             _continuation_token: Option<&str>,
+            _include_lifecycle_object_info: bool,
         ) -> rustfs_heal::Result<(Vec<HealListItem>, Option<String>, bool)> {
             Ok((Vec::new(), None, false))
         }

crates/heal/tests/mrf_pipeline_test.rs

@@ -0,0 +1,189 @@
+// 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.
+
+//! HS-01 (rustfs/backlog#1865): MRF intent pipeline integration tests.
+//!
+//! Drives the real consumer loop (`spawn_mrf_consumer`) against a real
+//! 4-disk `ECStore` heal storage and a `HealManager` that has not started its
+//! scheduler, so submitted intents stay observable in the admission queue.
+//! Under `cargo nextest` each test runs in its own process, which keeps the
+//! process-global MRF channel singleton safe.
+
+use rustfs_common::mrf_channel::{self, MrfKind};
+use rustfs_heal::heal::{
+    manager::{HealConfig, HealManager},
+    mrf_queue,
+    storage::{ECStoreHealStorage, HealStorageAPI},
+};
+use serial_test::serial;
+use std::{path::Path, sync::Arc, time::Duration};
+
+mod storage_api;
+
+use storage_api::endpoint_index::{Endpoint, EndpointServerPools, Endpoints, PoolEndpoints, init_local_disks};
+
+const META_BUCKET: &str = ".rustfs.sys";
+const JOURNAL_REL: &str = "buckets/.heal/mrf/journal.bin";
+
+async fn heal_env() -> (Vec<std::path::PathBuf>, Arc<dyn HealStorageAPI>) {
+    let env = rustfs_test_utils::TestECStoreEnv::builder()
+        .prefix("rustfs_heal_mrf_test")
+        .build()
+        .await;
+    let heal_storage: Arc<dyn HealStorageAPI> = Arc::new(ECStoreHealStorage::new(env.ecstore.clone()));
+    (env.disk_paths, heal_storage)
+}
+
+fn make_manager(storage: Arc<dyn HealStorageAPI>) -> Arc<HealManager> {
+    Arc::new(HealManager::new(
+        storage,
+        Some(HealConfig {
+            // Keep the scheduler from draining the queue before assertions.
+            heal_interval: Duration::from_secs(3600),
+            enable_auto_heal: false,
+            ..Default::default()
+        }),
+    ))
+}
+
+/// Encode one journal record independently of the implementation, so a format
+/// drift between writer and this fixture fails loudly here.
+fn journal_record(kind: u8, bucket: &str, object: &str, version: Option<[u8; 16]>, attempts: u8) -> Vec<u8> {
+    let mut body = vec![1u8, 1, kind, attempts];
+    body.extend_from_slice(&1_700_000_000_000u64.to_le_bytes());
+    match version {
+        Some(bytes) => {
+            body.push(1);
+            body.extend_from_slice(&bytes);
+        }
+        None => body.push(0),
+    }
+    body.extend_from_slice(&(bucket.len() as u32).to_le_bytes());
+    body.extend_from_slice(&(object.len() as u32).to_le_bytes());
+    body.extend_from_slice(bucket.as_bytes());
+    body.extend_from_slice(object.as_bytes());
+    let mut hasher = crc_fast::Digest::new(crc_fast::CrcAlgorithm::Crc32IsoHdlc);
+    hasher.update(&body);
+    body.extend_from_slice(&(hasher.finalize() as u32).to_le_bytes());
+    body
+}
+
+fn write_journal_to_disks(disk_paths: &[std::path::PathBuf], data: &[u8]) {
+    for path in disk_paths {
+        let journal = path.join(META_BUCKET).join(JOURNAL_REL);
+        std::fs::create_dir_all(journal.parent().expect("journal parent")).expect("create journal dir");
+        std::fs::write(&journal, data).expect("write journal fixture");
+    }
+}
+
+async fn wait_until<F, Fut>(deadline: Duration, mut probe: F) -> bool
+where
+    F: FnMut() -> Fut,
+    Fut: std::future::Future<Output = bool>,
+{
+    let start = std::time::Instant::now();
+    while start.elapsed() < deadline {
+        if probe().await {
+            return true;
+        }
+        tokio::time::sleep(Duration::from_millis(50)).await;
+    }
+    false
+}
+
+/// A decode-failure intent delivered on the global channel must surface in the
+/// heal manager as an Urgent request attributed to the MRF source.
+#[tokio::test]
+#[serial]
+async fn decode_failure_intent_maps_to_urgent_mrf_heal_request() {
+    let (_disk_paths, storage) = heal_env().await;
+    let manager = make_manager(storage);
+
+    mrf_queue::spawn_mrf_consumer(manager.clone());
+
+    assert!(
+        mrf_channel::try_send_mrf_intent(MrfKind::DecodeFailure, "mrf-bucket", "mrf-object", None),
+        "intent should be accepted while the consumer holds the channel"
+    );
+
+    let appeared = wait_until(Duration::from_secs(10), || async {
+        let snapshot = manager.operations_snapshot().await;
+        snapshot.queued_by_source.mrf >= 1 && snapshot.queued_by_priority.urgent >= 1
+    })
+    .await;
+    assert!(
+        appeared,
+        "MRF intent must reach the manager queue as an Urgent request (snapshot: {:?})",
+        manager.operations_snapshot().await
+    );
+}
+
+/// A journal left behind by a previous process must be replayed into the
+/// manager queue and then removed, and a torn tail must not block replay of
+/// the intact records.
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+#[serial]
+async fn journal_replay_arms_intents_and_deletes_the_file() {
+    let (disk_paths, storage) = heal_env().await;
+
+    // The journal reader resolves disks through the process-local disk map;
+    // register the environment's disks the same way server startup does.
+    let mut endpoints: Vec<Endpoint> = disk_paths
+        .iter()
+        .map(|p| Endpoint::try_from(p.to_string_lossy().as_ref()).expect("endpoint from disk path"))
+        .collect();
+    for (i, endpoint) in endpoints.iter_mut().enumerate() {
+        endpoint.set_pool_index(0);
+        endpoint.set_set_index(0);
+        endpoint.set_disk_index(i);
+    }
+    let pool = PoolEndpoints {
+        legacy: false,
+        set_count: 1,
+        drives_per_set: endpoints.len(),
+        endpoints: Endpoints::from(endpoints),
+        cmd_line: "mrf-test".to_string(),
+        platform: String::new(),
+    };
+    init_local_disks(EndpointServerPools::from(vec![pool]))
+        .await
+        .expect("local disks should register");
+
+    let mut journal = journal_record(1, "replay-bucket", "replay-object", Some([9u8; 16]), 0);
+    journal.extend(journal_record(3, "replay-bucket", "partial-object", None, 1));
+    // Torn tail: a third record truncated mid-way must not block the two
+    // intact records above.
+    journal.extend_from_slice(&journal_record(2, "replay-bucket", "metadata-object", None, 0)[..8]);
+    write_journal_to_disks(&disk_paths, &journal);
+
+    let manager = make_manager(storage);
+    // Replay directly (not via the process-global channel consumer, which the
+    // sibling test already claimed in this process under plain `cargo test`).
+    let replayed = mrf_queue::replay_journal_once(&manager).await;
+    assert_eq!(replayed, 2, "the two intact records must be replayed");
+
+    let snapshot = manager.operations_snapshot().await;
+    assert_eq!(snapshot.queued_by_source.mrf, 2, "replayed intents must be attributed to the MRF source");
+
+    assert!(
+        disk_paths
+            .iter()
+            .all(|path| !Path::new(path).join(META_BUCKET).join(JOURNAL_REL).exists()),
+        "the journal file must be removed after a successful replay"
+    );
+
+    let snapshot = manager.operations_snapshot().await;
+    assert_eq!(snapshot.queued_by_priority.urgent, 1, "the decode-failure record must replay as Urgent");
+    assert!(snapshot.queued_by_priority.normal >= 1, "the partial-write record must replay as Normal");
+}

crates/madmin/src/service_commands.rs

@@ -43,7 +43,7 @@ pub struct ServiceTraceOpts {
 
 #[allow(dead_code)]
 impl ServiceTraceOpts {
-    fn trace_types(&self) -> TraceType {
+    pub fn trace_types(&self) -> TraceType {
         let mut tt = TraceType::default();
         tt.set_if(self.s3, &TraceType::S3);
         tt.set_if(self.internal, &TraceType::INTERNAL);
@@ -72,6 +72,14 @@ impl ServiceTraceOpts {
         tt
     }
 
+    pub fn only_errors(&self) -> bool {
+        self.only_errors
+    }
+
+    pub fn threshold(&self) -> Duration {
+        self.threshold
+    }
+
     pub fn parse_params(&mut self, uri: &Uri) -> Result<(), String> {
         let query_pairs: HashMap<_, _> = uri
             .query()

crates/scanner/Cargo.toml

@@ -108,6 +108,9 @@ temp-env = { workspace = true }
 tempfile = { workspace = true }
 uuid = { workspace = true, features = ["v4", "serde", "fast-rng", "macro-diagnostics"] }
 tokio = { workspace = true, features = ["test-util", "fs", "rt-multi-thread"] }
+# Test-only: pins the emitted scanner alert wire names against the canonical
+# EventName string forms subscribers configure (rustfs/backlog#1868).
+rustfs-s3-types.workspace = true
 # Enables the shared MockWarmBackend / xl.meta assertion helpers exposed via
 # the ecstore `api::tier::test_util` facade module (rustfs/backlog#1148 ilm-6).
 rustfs-ecstore = { workspace = true, features = ["test-util"] }

crates/scanner/src/data_usage_define.rs

@@ -28,7 +28,8 @@ use rustfs_common::heal_channel::HealScanMode;
 use rustfs_config::ENV_SCANNER_CACHE_SAVE_TIMEOUT_SECS;
 pub use rustfs_data_usage::{
     AllTierStats, BucketTargetUsageInfo, BucketUsageInfo, DATA_USAGE_OBJECT_NAME, DATA_USAGE_OBSERVED_OBJECT_NAME,
-    DataUsageEntry, DataUsageHash, DataUsageHashMap, DataUsageInfo, LEGACY_DATA_USAGE_OBJECT_NAME, TierStats, hash_path,
+    DataUsageEntry, DataUsageHash, DataUsageHashMap, DataUsageInfo, LEGACY_DATA_USAGE_OBJECT_NAME, PrefixUsageEntry,
+    PrefixUsageQuery, PrefixUsageSummary, TierStats, hash_path, prefix_usage_in_cache,
 };
 use rustfs_utils::path::{SLASH_SEPARATOR, path_join_buf};
 use tokio::time::{Duration, Instant, sleep, timeout};
@@ -430,6 +431,13 @@ pub(crate) enum DataUsageCachePrepareOutcome {
 }
 
 impl DataUsageCache {
+    /// Prefix-level usage query over this (writer-side) cache; see
+    /// [`prefix_usage_in_cache`] for the semantics
+    /// (rustfs/backlog#1872).
+    pub fn prefix_usage(&self, bucket: &str, prefix: &str, max_entries: usize) -> Option<PrefixUsageQuery> {
+        prefix_usage_in_cache(&self.cache, bucket, prefix, max_entries)
+    }
+
     pub(crate) fn prepare_for_scan(
         &mut self,
         name: &str,

crates/scanner/src/lib.rs

@@ -53,6 +53,7 @@ use tokio_util::sync::CancellationToken;
 
 pub mod data_usage_define;
 pub mod error;
+pub mod prefix_usage;
 mod remote_scanner;
 pub mod runtime_config;
 pub mod scanner;
@@ -64,6 +65,7 @@ pub(crate) mod storage_api;
 
 pub use data_usage_define::*;
 pub use error::ScannerError;
+pub use prefix_usage::{BucketPrefixUsageResponse, bucket_prefix_usage, invalidate_prefix_usage_cache};
 pub use remote_scanner::{
     NS_SCANNER_MAX_REQUEST_BODY_SIZE, RemoteScannerAdmission, RemoteScannerRequest, admit_remote_scanner_request,
     claim_remote_scanner_request, decode_remote_scanner_request, preflight_remote_scanner_request,

crates/scanner/src/prefix_usage.rs

@@ -0,0 +1,349 @@
+// 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.
+
+//! Prefix-level bucket usage for admin/console consumers (rustfs/backlog#1872,
+//! MinIO `loadPrefixUsageFromBackend` parity).
+//!
+//! The per-bucket, per-set `.usage-cache.bin` objects already hold a
+//! path-keyed prefix tree; this module reads every set's copy through that
+//! set's own object layer (the hash-routed store path would always land on
+//! one set), aggregates the overlapping trees, and serves the result from a
+//! bounded 30-second cache. Bucket writes poke the cache through the
+//! dirty-usage hook so a fresh scan is visible immediately.
+
+use crate::data_usage_define::{DATA_USAGE_CACHE_NAME, DataUsageCache};
+use crate::error::ScannerError;
+use crate::storage_api::owner::{
+    EcstoreSetDisks, EcstoreStore, ecstore_is_reserved_or_invalid_bucket, ecstore_resolve_object_store_handle,
+};
+use futures::future::join_all;
+use rustfs_data_usage::{PrefixUsageEntry, PrefixUsageSummary};
+use std::collections::HashMap;
+use std::sync::{Arc, Mutex};
+use std::time::{Duration, SystemTime};
+use tracing::{debug, warn};
+
+const LOG_COMPONENT_SCANNER: &str = "scanner";
+const LOG_SUBSYSTEM_PREFIX_USAGE: &str = "prefix_usage";
+const EVENT_PREFIX_USAGE_CACHE_STATE: &str = "prefix_usage_cache_state";
+
+/// How long a computed breakdown stays fresh. MinIO uses the same 30s for
+/// its prefix-usage cache; bucket writes additionally invalidate on the spot.
+const CACHE_TTL: Duration = Duration::from_secs(30);
+/// Hard entry cap for the result cache; exceeded, expired entries go first
+/// and the map clears rather than growing past the bound.
+const CACHE_MAX_ENTRIES: usize = 128;
+/// Per-set cache read budget. The underlying loader retries for up to a
+/// minute per attempt on backend errors — far too long for an admin GET, so
+/// a slow set degrades to "not reporting" instead of stalling the caller.
+const PER_SET_LOAD_TIMEOUT: Duration = Duration::from_secs(5);
+
+/// Aggregated prefix-usage answer across every erasure set.
+#[derive(Clone, Debug, PartialEq, serde::Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct BucketPrefixUsageResponse {
+    pub bucket: String,
+    pub prefix: String,
+    pub usage: PrefixUsageSummary,
+    /// Every reporting set's prefix entry was compacted: the aggregate is
+    /// valid, the sub-prefix breakdown is empty on disk.
+    pub compacted: bool,
+    /// The sub-prefix breakdown is incomplete: at least one reporting set
+    /// had the prefix compacted (or absent while others found it), so its
+    /// objects cannot be attributed to a sub-prefix.
+    pub sub_prefixes_partial: bool,
+    /// The breakdown exceeded the caller's entry limit; largest remain.
+    pub truncated: bool,
+    pub sub_prefixes: Vec<PrefixUsageEntry>,
+    /// Sets whose cache held this bucket and prefix.
+    pub sets_reporting: usize,
+    pub sets_total: usize,
+    /// Newest `last_update` across reporting sets, unix seconds.
+    pub last_update_unix_secs: Option<u64>,
+}
+
+#[derive(Clone)]
+struct CachedResponse {
+    computed_at: std::time::Instant,
+    response: Arc<BucketPrefixUsageResponse>,
+}
+
+/// Cache key: (lowercased bucket, normalized prefix, max entries).
+type PrefixUsageCacheKey = (String, String, usize);
+type PrefixUsageCacheMap = Option<HashMap<PrefixUsageCacheKey, CachedResponse>>;
+
+static PREFIX_USAGE_CACHE: Mutex<PrefixUsageCacheMap> = Mutex::new(None);
+
+/// Drop cached results for `bucket` (empty string clears everything). Wired
+/// into the dirty-usage recording path so a write makes the next prefix
+/// query recompute instead of serving up to `CACHE_TTL` seconds of stale
+/// numbers.
+pub fn invalidate_prefix_usage_cache(bucket: &str) {
+    let mut guard = PREFIX_USAGE_CACHE.lock().unwrap_or_else(|poison| poison.into_inner());
+    let Some(map) = guard.as_mut() else {
+        return;
+    };
+    if bucket.is_empty() {
+        map.clear();
+        return;
+    }
+    map.retain(|(cached_bucket, ..), _| !cached_bucket.eq_ignore_ascii_case(bucket));
+}
+
+/// Query prefix usage for `bucket` (arbitrary `prefix`, empty = whole
+/// bucket), merging every erasure set's own cache copy. `max_entries` bounds
+/// the sub-prefix rows (largest first).
+pub async fn bucket_prefix_usage(
+    bucket: &str,
+    prefix: &str,
+    max_entries: usize,
+) -> Result<BucketPrefixUsageResponse, ScannerError> {
+    if ecstore_is_reserved_or_invalid_bucket(bucket, true) {
+        return Err(ScannerError::Other(format!("invalid bucket name: {bucket}")));
+    }
+    let normalized_prefix = prefix.trim_matches('/').to_string();
+    let cache_key = (bucket.to_ascii_lowercase(), normalized_prefix.clone(), max_entries);
+    if let Some(response) = lookup_cached(&cache_key) {
+        return Ok((*response).clone());
+    }
+
+    let store = ecstore_resolve_object_store_handle()
+        .ok_or_else(|| ScannerError::Other("object store is not initialized".to_string()))?;
+    let response = Arc::new(compute_prefix_usage(store, bucket, &normalized_prefix, max_entries).await);
+    store_cached(cache_key, response.clone());
+    Ok((*response).clone())
+}
+
+async fn compute_prefix_usage(
+    store: Arc<EcstoreStore>,
+    bucket: &str,
+    prefix: &str,
+    max_entries: usize,
+) -> BucketPrefixUsageResponse {
+    let sets: Vec<Arc<EcstoreSetDisks>> = store.all_set_disks();
+    let sets_total = sets.len();
+    let cache_name = format!("{bucket}/{DATA_USAGE_CACHE_NAME}");
+
+    let per_set = join_all(sets.into_iter().map(|set| {
+        let cache_name = cache_name.clone();
+        async move {
+            let mut cache = DataUsageCache::default();
+            // A set that has never scanned this bucket (or cannot be read
+            // within the budget) reports nothing — the remaining sets still
+            // produce a usable, flagged answer.
+            let loaded = match tokio::time::timeout(PER_SET_LOAD_TIMEOUT, cache.load(set, &cache_name)).await {
+                Ok(Ok(())) => cache,
+                Ok(Err(err)) => {
+                    debug!(
+                        target: "rustfs::scanner::prefix_usage",
+                        event = EVENT_PREFIX_USAGE_CACHE_STATE,
+                        component = LOG_COMPONENT_SCANNER,
+                        subsystem = LOG_SUBSYSTEM_PREFIX_USAGE,
+                        bucket = %bucket,
+                        state = "set_load_failed",
+                        error = %err,
+                        "Prefix usage set cache load failed"
+                    );
+                    return None;
+                }
+                Err(_) => {
+                    warn!(
+                        target: "rustfs::scanner::prefix_usage",
+                        event = EVENT_PREFIX_USAGE_CACHE_STATE,
+                        component = LOG_COMPONENT_SCANNER,
+                        subsystem = LOG_SUBSYSTEM_PREFIX_USAGE,
+                        bucket = %bucket,
+                        state = "set_load_timeout",
+                        "Prefix usage set cache load timed out"
+                    );
+                    return None;
+                }
+            };
+            if loaded.info.name != bucket {
+                // Empty or stale-scoped cache: this set has no data for the bucket.
+                return None;
+            }
+            let last_update = loaded.info.last_update;
+            let query = loaded.prefix_usage(bucket, prefix, max_entries);
+            Some((query, last_update))
+        }
+    }))
+    .await;
+
+    let mut usage = PrefixUsageSummary::default();
+    let mut sub_prefix_map: HashMap<String, PrefixUsageSummary> = HashMap::new();
+    let mut sets_reporting = 0usize;
+    let mut reporting_but_absent = 0usize;
+    let mut any_compacted = false;
+    let mut all_compacted = true;
+    let mut truncated = false;
+    let mut last_update: Option<SystemTime> = None;
+
+    for (query, set_last_update) in per_set.into_iter().flatten() {
+        // last_update counts every set that has scanned the bucket, even
+        // when the prefix itself is absent on that set.
+        if let Some(set_last_update) = set_last_update
+            && last_update.map(|current| set_last_update > current).unwrap_or(true)
+        {
+            last_update = Some(set_last_update);
+        }
+        let Some(query) = query else {
+            // The set knows the bucket but not this prefix: legitimate when
+            // the prefix's objects all hash to other sets, but it means the
+            // breakdown below cannot attribute that set's (zero) objects.
+            reporting_but_absent += 1;
+            continue;
+        };
+        sets_reporting += 1;
+        usage.merge(&query.usage);
+        if query.compacted {
+            any_compacted = true;
+        } else {
+            all_compacted = false;
+        }
+        truncated |= query.truncated;
+        for entry in query.sub_prefixes {
+            sub_prefix_map.entry(entry.prefix).or_default().merge(&entry.usage);
+        }
+    }
+
+    let mut sub_prefixes: Vec<PrefixUsageEntry> = sub_prefix_map
+        .into_iter()
+        .map(|(prefix, usage)| PrefixUsageEntry { prefix, usage })
+        .collect();
+    sub_prefixes.sort_by(|left, right| {
+        right
+            .usage
+            .size
+            .cmp(&left.usage.size)
+            .then_with(|| left.prefix.cmp(&right.prefix))
+    });
+    // Merged rows can exceed max_entries only when per-set truncation
+    // already flagged; enforce the caller bound on the merged view too.
+    if sub_prefixes.len() > max_entries {
+        truncated = true;
+        sub_prefixes.truncate(max_entries);
+    }
+
+    let found = sets_reporting > 0;
+    BucketPrefixUsageResponse {
+        bucket: bucket.to_string(),
+        prefix: prefix.to_string(),
+        usage,
+        compacted: found && all_compacted,
+        sub_prefixes_partial: any_compacted || reporting_but_absent > 0,
+        truncated,
+        sub_prefixes,
+        sets_reporting,
+        sets_total,
+        last_update_unix_secs: last_update
+            .and_then(|time| time.duration_since(SystemTime::UNIX_EPOCH).ok())
+            .map(|dur| dur.as_secs()),
+    }
+}
+
+fn lookup_cached(key: &(String, String, usize)) -> Option<Arc<BucketPrefixUsageResponse>> {
+    let mut guard = PREFIX_USAGE_CACHE.lock().unwrap_or_else(|poison| poison.into_inner());
+    let map = guard.as_mut()?;
+    let cached = map.get(key)?;
+    if cached.computed_at.elapsed() > CACHE_TTL {
+        map.remove(key);
+        return None;
+    }
+    Some(cached.response.clone())
+}
+
+fn store_cached(key: (String, String, usize), response: Arc<BucketPrefixUsageResponse>) {
+    let mut guard = PREFIX_USAGE_CACHE.lock().unwrap_or_else(|poison| poison.into_inner());
+    let map = guard.get_or_insert_with(HashMap::new);
+    // Bound the cache: drop expired entries first, and if the cap is still
+    // exceeded clear wholesale — the next queries recompute in milliseconds.
+    if map.len() >= CACHE_MAX_ENTRIES {
+        map.retain(|_, cached| cached.computed_at.elapsed() <= CACHE_TTL);
+        if map.len() >= CACHE_MAX_ENTRIES {
+            map.clear();
+        }
+    }
+    map.insert(
+        key,
+        CachedResponse {
+            computed_at: std::time::Instant::now(),
+            response,
+        },
+    );
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{CACHE_MAX_ENTRIES, PREFIX_USAGE_CACHE, invalidate_prefix_usage_cache, store_cached};
+    use rustfs_data_usage::PrefixUsageSummary;
+
+    fn response(bucket: &str) -> super::BucketPrefixUsageResponse {
+        super::BucketPrefixUsageResponse {
+            bucket: bucket.to_string(),
+            prefix: String::new(),
+            usage: PrefixUsageSummary::default(),
+            compacted: false,
+            sub_prefixes_partial: false,
+            truncated: false,
+            sub_prefixes: Vec::new(),
+            sets_reporting: 1,
+            sets_total: 1,
+            last_update_unix_secs: None,
+        }
+    }
+
+    fn seed(bucket: &str, prefix: &str) {
+        store_cached(
+            (bucket.to_ascii_lowercase(), prefix.to_string(), 10),
+            std::sync::Arc::new(response(bucket)),
+        );
+    }
+
+    fn contains(bucket: &str, prefix: &str) -> bool {
+        PREFIX_USAGE_CACHE
+            .lock()
+            .unwrap_or_else(|poison| poison.into_inner())
+            .as_ref()
+            .is_some_and(|map| map.contains_key(&(bucket.to_ascii_lowercase(), prefix.to_string(), 10)))
+    }
+
+    /// All cache tests run inside one test to keep the process-global map
+    /// free of cross-test ordering (the flake class this module avoids).
+    #[test]
+    fn invalidation_scopes_to_bucket_and_cache_stays_bounded() {
+        invalidate_prefix_usage_cache("");
+        seed("alpha", "x");
+        seed("beta", "y");
+
+        // Case-insensitive bucket scoping.
+        invalidate_prefix_usage_cache("ALPHA");
+        assert!(!contains("alpha", "x"));
+        assert!(contains("beta", "y"));
+
+        // Wholesale clear.
+        invalidate_prefix_usage_cache("");
+        assert!(!contains("beta", "y"));
+
+        // Hard cap: overflow clears rather than grows.
+        for index in 0..=(CACHE_MAX_ENTRIES / 2) {
+            let bucket = format!("cap-bucket-{index}");
+            seed(&bucket, "a");
+            seed(&bucket, "b");
+        }
+        let guard = PREFIX_USAGE_CACHE.lock().unwrap_or_else(|poison| poison.into_inner());
+        let map = guard.as_ref().expect("seeded");
+        assert!(map.len() <= CACHE_MAX_ENTRIES, "cache must stay bounded, got {}", map.len());
+    }
+}

crates/scanner/src/scanner_folder.rs

@@ -12,10 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::collections::HashSet;
+use std::collections::{HashMap, HashSet};
 use std::fs::FileType;
 use std::io::ErrorKind;
-use std::sync::{Arc, Once};
+use std::sync::{Arc, Mutex, Once};
 use std::time::{Duration, Instant, SystemTime};
 
 use crate::ReplTargetSizeSummary;
@@ -32,6 +32,7 @@ use crate::scanner_io::{
     SCANNER_SKIP_FILE_ERROR, ScannerIODisk as _, is_scanner_metadata_corrupt_error, is_scanner_metadata_transient_error,
 };
 use crate::sleeper::DynamicSleeper;
+use crate::storage_api::owner::{EcstoreEventArgs, ecstore_send_event};
 use metrics::{counter, describe_counter};
 use rustfs_common::heal_channel::{
     HEAL_DELETE_DANGLING, HealAdmissionDropReason, HealAdmissionResult, HealChannelPriority, HealChannelRequest,
@@ -41,6 +42,7 @@ use rustfs_common::metrics::{
     CloseDiskGuard, IlmAction, Metric, Metrics, ScannerReplicationRepairKind, ScannerSourceWorkUpdate, ScannerWorkSource,
     UpdateCurrentPathFn, current_path_updater, global_metrics,
 };
+use rustfs_common::trace_bus::{TraceEvent, TraceFunc, TraceKind, trace_emit, trace_subscriber_count};
 use rustfs_filemeta::{MetaCacheEntries, MetaCacheEntry, MetadataResolutionParams};
 use rustfs_utils::path::{SLASH_SEPARATOR, path_join_buf};
 use s3s::dto::{BucketLifecycleConfiguration, ObjectLockConfiguration};
@@ -97,6 +99,101 @@ const METRIC_SCANNER_EXCESS_FOLDERS_TOTAL: &str = "rustfs_scanner_excess_folders
 const METRIC_SCANNER_PENDING_HEAL_PRUNE_TOTAL: &str = "rustfs_scanner_pending_heal_prune_total";
 const METRIC_SCANNER_PENDING_HEAL_MALFORMED_TOTAL: &str = "rustfs_scanner_pending_heal_malformed_total";
 const MAX_PENDING_SCANNER_HEAL_RETRIES_PER_BUCKET: usize = 128;
+
+// --- scanner excess alerts as S3 notification events (rustfs/backlog#1868) --
+//
+// The excess-versions / excess-version-size / excess-folders alerts were
+// metrics-and-logs only; subscribers (consoles, external auditors) had no way
+// to hear them. MinIO emits s3:ObjectManyVersions / s3:ObjectLargeVersions /
+// s3:PrefixManyFolders for the same conditions — RustFS carries those as
+// EventName::Scanner* with the wire names below. Without a cooldown a single
+// over-threshold object would re-emit on every scan cycle (~a minute), so
+// emissions are edge-held per (kind, bucket, object) for 24h.
+
+/// `s3:Scanner:ManyVersions` (MinIO `s3:ObjectManyVersions`).
+pub const EVENT_SCANNER_MANY_VERSIONS: &str = "s3:Scanner:ManyVersions";
+/// `s3:Scanner:LargeVersions` (MinIO `s3:ObjectLargeVersions`).
+pub const EVENT_SCANNER_LARGE_VERSIONS: &str = "s3:Scanner:LargeVersions";
+/// `s3:Scanner:BigPrefix` (MinIO `s3:PrefixManyFolders`).
+pub const EVENT_SCANNER_BIG_PREFIX: &str = "s3:Scanner:BigPrefix";
+const ENV_SCANNER_ALERT_COOLDOWN_SECS: &str = "RUSTFS_SCANNER_ALERT_COOLDOWN_SECS";
+const DEFAULT_SCANNER_ALERT_COOLDOWN_SECS: u64 = 86_400;
+/// Hard cap on distinct cooldown keys; a pathological number of over-threshold
+/// objects clears the map wholesale instead of growing without bound (the
+/// worst case is one re-emission per still-hot key per scan cycle).
+const MAX_SCANNER_ALERT_COOLDOWN_KEYS: usize = 4096;
+
+/// Distinct alert kinds sharing one cooldown map.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+enum ScannerAlertKind {
+    ManyVersions,
+    LargeVersions,
+    BigPrefix,
+}
+
+type ScannerAlertCooldownKey = (ScannerAlertKind, String, String);
+type ScannerAlertCooldownMap = HashMap<ScannerAlertCooldownKey, Instant>;
+
+static SCANNER_ALERT_EMISSION_COOLDOWN: Mutex<Option<ScannerAlertCooldownMap>> = Mutex::new(None);
+
+fn scanner_alert_cooldown() -> Duration {
+    let raw = std::env::var(ENV_SCANNER_ALERT_COOLDOWN_SECS)
+        .ok()
+        .and_then(|v| v.parse::<u64>().ok());
+    Duration::from_secs(raw.unwrap_or(DEFAULT_SCANNER_ALERT_COOLDOWN_SECS))
+}
+
+/// Edge-held emission gate: returns `true` (and records the cooldown) only
+/// when this (kind, bucket, object) last fired longer than the cooldown ago —
+/// or never. Metrics and logs stay level-triggered every cycle; only the
+/// notification events are held back.
+fn scanner_alert_emission_allows(kind: ScannerAlertKind, bucket: &str, object: &str, cooldown: Duration) -> bool {
+    let key = (kind, bucket.to_string(), object.to_string());
+    let mut guard = SCANNER_ALERT_EMISSION_COOLDOWN
+        .lock()
+        .unwrap_or_else(|poison| poison.into_inner());
+    let guard = guard.get_or_insert_with(ScannerAlertCooldownMap::new);
+    let now = Instant::now();
+    // Expired entries leave first; the cap is still exceeded only when live
+    // keys alone overflow it, in which case a wholesale clear trades one
+    // extra emission per hot key for a hard memory bound.
+    if guard.len() >= MAX_SCANNER_ALERT_COOLDOWN_KEYS {
+        guard.retain(|_, fired_at| now.duration_since(*fired_at) < cooldown);
+        if guard.len() >= MAX_SCANNER_ALERT_COOLDOWN_KEYS {
+            guard.clear();
+        }
+    }
+    match guard.get(&key) {
+        Some(fired_at) if now.duration_since(*fired_at) < cooldown => false,
+        _ => {
+            guard.insert(key, now);
+            true
+        }
+    }
+}
+
+/// Emit a scanner alert as an S3 notification event through the standard
+/// dispatch pipeline. Fire-and-forget: the notify layer owns delivery,
+/// retry, and target filtering; the scanner never waits on it.
+fn emit_scanner_alert_event(event_name: &str, bucket: &str, object: &str, size: i64, details: &[(&str, String)]) {
+    let mut req_params = HashMap::with_capacity(details.len());
+    for (key, value) in details {
+        req_params.insert((*key).to_string(), value.clone());
+    }
+    ecstore_send_event(EcstoreEventArgs {
+        event_name: event_name.to_string(),
+        bucket_name: bucket.to_string(),
+        object: crate::ScannerObjectInfo {
+            bucket: bucket.to_string(),
+            name: object.to_string(),
+            size,
+            ..Default::default()
+        },
+        req_params,
+        user_agent: "Scanner".to_string(),
+        ..Default::default()
+    });
+}
 const MAX_PENDING_SCANNER_HEALS_PER_BUCKET: usize = 10_000;
 
 static SCANNER_INLINE_HEAL_WARN_ONCE: Once = Once::new();
@@ -430,6 +527,113 @@ fn non_negative_i64_to_u64(value: i64) -> u64 {
     value.max(0) as u64
 }
 
+fn trace_start_instant() -> Option<Instant> {
+    (trace_subscriber_count() > 0).then(Instant::now)
+}
+
+fn emit_scanner_folder_trace(root: &str, folder: &str, objects: u64, started_at: Option<Instant>, state: &'static str) {
+    let Some(started_at) = started_at else {
+        return;
+    };
+
+    trace_emit(|| {
+        let (bucket, prefix) = path2_bucket_object_with_base_path(root, folder);
+        TraceEvent::new(TraceKind::Scanner, TraceFunc::ScannerFolder)
+            .with_bucket(bucket)
+            .with_object(prefix)
+            .with_duration(started_at.elapsed())
+            .with_attr("state", state)
+            .with_attr("objects", objects)
+    });
+}
+
+fn emit_scanner_ilm_action_trace(
+    bucket: &str,
+    object: &str,
+    action: IlmAction,
+    count: u64,
+    queued: bool,
+    started_at: Option<Instant>,
+) {
+    let Some(started_at) = started_at else {
+        return;
+    };
+
+    let state = if queued { "queued" } else { "not_queued" };
+    trace_emit(|| {
+        TraceEvent::new(TraceKind::Scanner, TraceFunc::ScannerIlmAction)
+            .with_bucket(bucket)
+            .with_object(object)
+            .with_duration(started_at.elapsed())
+            .with_attr("state", state)
+            .with_attr("action", action.as_str())
+            .with_attr("count", count)
+            .with_attr("queued", queued)
+    });
+}
+
+struct ScannerHealCandidateTraceContext {
+    bucket: String,
+    object: Option<String>,
+    version_id: Option<String>,
+    scan_mode: Option<HealScanMode>,
+    started_at: Instant,
+}
+
+fn scanner_heal_candidate_trace_context(request: &HealChannelRequest) -> Option<ScannerHealCandidateTraceContext> {
+    let started_at = trace_start_instant()?;
+    Some(ScannerHealCandidateTraceContext {
+        bucket: request.bucket.clone(),
+        object: request.object_prefix.clone(),
+        version_id: request.object_version_id.clone(),
+        scan_mode: request.scan_mode,
+        started_at,
+    })
+}
+
+struct ScannerHealCandidateTrace<'a> {
+    candidate_type: &'static str,
+    bucket: &'a str,
+    object: Option<&'a str>,
+    version_id: Option<&'a str>,
+    priority: HealChannelPriority,
+    scan_mode: Option<HealScanMode>,
+    result: Result<HealAdmissionResult, &'a str>,
+    started_at: Instant,
+}
+
+fn emit_scanner_heal_candidate_trace(trace: ScannerHealCandidateTrace<'_>) {
+    trace_emit(|| {
+        let (state, admission, error) = match trace.result {
+            Ok(result) if result.is_admitted() => ("admitted", describe_heal_admission(result), None),
+            Ok(result) => ("not_admitted", describe_heal_admission(result), None),
+            Err(error) => ("submit_failed", "channel_error".to_string(), Some(error)),
+        };
+        let mut event = TraceEvent::new(TraceKind::Scanner, TraceFunc::ScannerHealCandidate)
+            .with_bucket(trace.bucket)
+            .with_duration(trace.started_at.elapsed())
+            .with_attr("state", state)
+            .with_attr("candidate_type", trace.candidate_type)
+            .with_attr("priority", heal_priority_label(trace.priority))
+            .with_attr("admission", admission);
+
+        if let Some(object) = trace.object {
+            event = event.with_object(object);
+        }
+        if let Some(version_id) = trace.version_id {
+            event = event.with_attr("version_id", version_id);
+        }
+        if let Some(scan_mode) = trace.scan_mode {
+            event = event.with_attr("scan_mode", scan_mode.as_str());
+        }
+        if let Some(error) = error {
+            event = event.with_attr("error", error);
+        }
+
+        event
+    });
+}
+
 fn apply_scanner_size_summary(into: &mut DataUsageEntry, summary: &SizeSummary) {
     into.size = into.size.saturating_add(summary.total_size);
     into.versions = into.versions.saturating_add(summary.versions);
@@ -677,9 +881,22 @@ async fn send_scanner_heal_request(
     request: HealChannelRequest,
 ) -> Result<HealAdmissionResult, ScannerError> {
     let priority = request.priority;
+    let trace_context = scanner_heal_candidate_trace_context(&request);
     match send_heal_request_with_admission(request).await {
         Ok(result) => {
             record_heal_candidate_admission(candidate_type, priority, result);
+            if let Some(trace_context) = trace_context.as_ref() {
+                emit_scanner_heal_candidate_trace(ScannerHealCandidateTrace {
+                    candidate_type,
+                    bucket: &trace_context.bucket,
+                    object: trace_context.object.as_deref(),
+                    version_id: trace_context.version_id.as_deref(),
+                    priority,
+                    scan_mode: trace_context.scan_mode,
+                    result: Ok(result),
+                    started_at: trace_context.started_at,
+                });
+            }
             Ok(result)
         }
         Err(err) => {
@@ -690,6 +907,18 @@ async fn send_scanner_heal_request(
                 "result" => "channel_error".to_string()
             )
             .increment(1);
+            if let Some(trace_context) = trace_context.as_ref() {
+                emit_scanner_heal_candidate_trace(ScannerHealCandidateTrace {
+                    candidate_type,
+                    bucket: &trace_context.bucket,
+                    object: trace_context.object.as_deref(),
+                    version_id: trace_context.version_id.as_deref(),
+                    priority,
+                    scan_mode: trace_context.scan_mode,
+                    result: Err(err.as_str()),
+                    started_at: trace_context.started_at,
+                });
+            }
             Err(ScannerError::Other(err))
         }
     }
@@ -905,7 +1134,9 @@ impl ScannerItem {
                             "Scanner lifecycle action dispatched"
                         );
                         let done_ilm = Metrics::time_ilm(event.action);
+                        let trace_started_at = trace_start_instant();
                         let queued = apply_expiry_rule(event, &LcEventSrc::Scanner, oi).await;
+                        emit_scanner_ilm_action_trace(&self.bucket, &oi.name, event.action, 1, queued, trace_started_at);
                         if record_scanner_ilm_action_if_queued(global_metrics(), event.action, 1, queued) {
                             done_ilm(1)();
                             remaining_versions = 0;
@@ -957,7 +1188,9 @@ impl ScannerItem {
                             "Scanner lifecycle action dispatched"
                         );
                         let done_ilm = Metrics::time_ilm(event.action);
+                        let trace_started_at = trace_start_instant();
                         let queued = apply_expiry_rule(event, &LcEventSrc::Scanner, oi).await;
+                        emit_scanner_ilm_action_trace(&self.bucket, &oi.name, event.action, 1, queued, trace_started_at);
                         if record_scanner_ilm_action_if_queued(global_metrics(), event.action, 1, queued) {
                             done_ilm(1)();
                             if !versioning_config.prefix_enabled(&self.object_path()) && event.action == IlmAction::DeleteAction {
@@ -995,7 +1228,9 @@ impl ScannerItem {
                             "Scanner lifecycle action dispatched"
                         );
                         let done_ilm = Metrics::time_ilm(event.action);
+                        let trace_started_at = trace_start_instant();
                         let queued = apply_transition_rule(event, &LcEventSrc::Scanner, oi).await;
+                        emit_scanner_ilm_action_trace(&self.bucket, &oi.name, event.action, 1, queued, trace_started_at);
                         if record_scanner_ilm_action_if_queued(global_metrics(), event.action, 1, queued) {
                             done_ilm(1)();
                         }
@@ -1019,7 +1254,21 @@ impl ScannerItem {
             let action = event.action;
             let count = u64::try_from(to_delete_objs.len()).unwrap_or(u64::MAX);
             let done_ilm = Metrics::time_ilm(action);
+            let trace_started_at = trace_start_instant();
             let queued = enqueue_runtime_newer_noncurrent(&self.bucket, to_delete_objs, event, &LcEventSrc::Scanner).await;
+            if let Some(trace_started_at) = trace_started_at {
+                let state = if queued { "queued" } else { "not_queued" };
+                trace_emit(|| {
+                    TraceEvent::new(TraceKind::Scanner, TraceFunc::ScannerIlmAction)
+                        .with_bucket(self.bucket.as_str())
+                        .with_object(self.object_path())
+                        .with_duration(trace_started_at.elapsed())
+                        .with_attr("state", state)
+                        .with_attr("action", action.as_str())
+                        .with_attr("count", count)
+                        .with_attr("queued", queued)
+                });
+            }
             if record_scanner_ilm_action_if_queued(global_metrics(), action, count, queued) {
                 done_ilm(count)();
                 remaining_versions = remaining_versions.saturating_sub(noncurrent_accounting.len());
@@ -1197,6 +1446,7 @@ impl ScannerItem {
     fn alert_excessive_versions(&self, remaining_versions: usize, cumulative_size: i64) {
         ensure_scanner_alert_metrics_registered();
         let (too_many_versions, too_large_versions) = should_alert_excessive_versions(remaining_versions, cumulative_size);
+        let object_path = self.object_path();
         if too_many_versions {
             global_metrics().record_scanner_source_executed(ScannerWorkSource::Alerts, 1);
             counter!(
@@ -1204,13 +1454,26 @@ impl ScannerItem {
                 "bucket" => self.bucket.clone()
             )
             .increment(1);
+            if scanner_alert_emission_allows(ScannerAlertKind::ManyVersions, &self.bucket, &object_path, scanner_alert_cooldown())
+            {
+                emit_scanner_alert_event(
+                    EVENT_SCANNER_MANY_VERSIONS,
+                    &self.bucket,
+                    &object_path,
+                    cumulative_size,
+                    &[
+                        ("versions", remaining_versions.to_string()),
+                        ("threshold", scanner_excess_versions_threshold().to_string()),
+                    ],
+                );
+            }
             warn!(
                 target: "rustfs::scanner::folder",
                 event = EVENT_SCANNER_ALERT_STATE,
                 component = LOG_COMPONENT_SCANNER,
                 subsystem = LOG_SUBSYSTEM_FOLDER,
                 bucket = %self.bucket,
-                object = %self.object_path(),
+                object = %object_path,
                 versions = remaining_versions,
                 threshold = scanner_excess_versions_threshold(),
                 state = "excess_versions",
@@ -1224,13 +1487,31 @@ impl ScannerItem {
                 "bucket" => self.bucket.clone()
             )
             .increment(1);
+            if scanner_alert_emission_allows(
+                ScannerAlertKind::LargeVersions,
+                &self.bucket,
+                &object_path,
+                scanner_alert_cooldown(),
+            ) {
+                emit_scanner_alert_event(
+                    EVENT_SCANNER_LARGE_VERSIONS,
+                    &self.bucket,
+                    &object_path,
+                    cumulative_size,
+                    &[
+                        ("versions", remaining_versions.to_string()),
+                        ("cumulativeSize", cumulative_size.to_string()),
+                        ("threshold", scanner_excess_version_size_threshold().to_string()),
+                    ],
+                );
+            }
             warn!(
                 target: "rustfs::scanner::folder",
                 event = EVENT_SCANNER_ALERT_STATE,
                 component = LOG_COMPONENT_SCANNER,
                 subsystem = LOG_SUBSYSTEM_FOLDER,
                 bucket = %self.bucket,
-                object = %self.object_path(),
+                object = %object_path,
                 versions = remaining_versions,
                 cumulative_size,
                 threshold = scanner_excess_version_size_threshold(),
@@ -1611,6 +1892,15 @@ impl FolderScanner {
             "root" => self.root.clone()
         )
         .increment(1);
+        if scanner_alert_emission_allows(ScannerAlertKind::BigPrefix, &self.root, folder, scanner_alert_cooldown()) {
+            emit_scanner_alert_event(
+                EVENT_SCANNER_BIG_PREFIX,
+                &self.root,
+                folder,
+                0,
+                &[("folders", total_folders.to_string()), ("threshold", threshold.to_string())],
+            );
+        }
         warn!(
             target: "rustfs::scanner::folder",
             event = EVENT_SCANNER_ALERT_STATE,
@@ -1830,6 +2120,7 @@ impl FolderScanner {
         into: &mut DataUsageEntry,
     ) -> Result<(), ScannerError> {
         let done_folder = Metrics::time(Metric::ScanFolder);
+        let trace_started_at = trace_start_instant();
 
         if ctx.is_cancelled() {
             return Err(ScannerError::Other("Operation cancelled".to_string()));
@@ -2187,6 +2478,15 @@ impl FolderScanner {
                         }
 
                         if let GetSizeFailureAction::HealMetadata { object } = failure_action {
+                            // MRF journal intent: durable High-priority Metadata
+                            // heal across restarts (HS-01); the scanner heal
+                            // request below stays as the immediate path.
+                            rustfs_common::mrf_channel::try_send_mrf_intent(
+                                rustfs_common::mrf_channel::MrfKind::MetadataCorruption,
+                                &item.bucket,
+                                &object,
+                                None,
+                            );
                             self.send_required_scanner_heal_request(
                                 PendingScannerHealKind::Object,
                                 item.bucket.clone(),
@@ -2895,6 +3195,8 @@ impl FolderScanner {
         }
 
         done_folder();
+        let scanned_objects = u64::try_from(into.objects).unwrap_or(u64::MAX);
+        emit_scanner_folder_trace(&self.root, &folder.name, scanned_objects, trace_started_at, "completed");
 
         Ok(())
     }
@@ -3076,6 +3378,90 @@ mod tests {
     #[cfg(unix)]
     use std::os::unix::fs::{PermissionsExt, symlink};
     use std::sync::Mutex;
+
+    /// Reset the process-global alert cooldown map; test-only.
+    fn reset_alert_cooldowns() {
+        *SCANNER_ALERT_EMISSION_COOLDOWN
+            .lock()
+            .unwrap_or_else(|poison| poison.into_inner()) = Some(ScannerAlertCooldownMap::new());
+    }
+
+    /// The emitted event-name strings must be exactly what `EventName`
+    /// serializes, or a bucket notification subscribed to the documented name
+    /// would silently never match (rustfs/backlog#1868).
+    #[test]
+    fn scanner_alert_wire_names_match_canonical_event_names() {
+        use rustfs_s3_types::EventName;
+        assert_eq!(EVENT_SCANNER_MANY_VERSIONS, EventName::ScannerManyVersions.to_string());
+        assert_eq!(EVENT_SCANNER_LARGE_VERSIONS, EventName::ScannerLargeVersions.to_string());
+        assert_eq!(EVENT_SCANNER_BIG_PREFIX, EventName::ScannerBigPrefix.to_string());
+    }
+
+    fn cooldown_map_len() -> usize {
+        SCANNER_ALERT_EMISSION_COOLDOWN
+            .lock()
+            .unwrap_or_else(|poison| poison.into_inner())
+            .as_ref()
+            .map(|map| map.len())
+            .unwrap_or(0)
+    }
+
+    /// Backdate every recorded cooldown so the next check fires again.
+    fn expire_all_alert_cooldowns(cooldown: Duration) {
+        let now = Instant::now();
+        let mut guard = SCANNER_ALERT_EMISSION_COOLDOWN
+            .lock()
+            .unwrap_or_else(|poison| poison.into_inner());
+        if let Some(map) = guard.as_mut() {
+            for fired_at in map.values_mut() {
+                if let Some(expired) = now.checked_sub(cooldown + Duration::from_secs(1)) {
+                    *fired_at = expired;
+                }
+            }
+        }
+    }
+
+    /// The emission gate is the only thing standing between an over-threshold
+    /// object and one S3 event per scan cycle, so its edge semantics get
+    /// pinned directly. All scenarios share one #[test] because the cooldown
+    /// map is process-global and parallel tests would read each other's
+    /// firings.
+    #[test]
+    fn scanner_alert_emission_is_edge_held_per_key_and_bounded() {
+        reset_alert_cooldowns();
+        let cooldown = Duration::from_secs(3600);
+
+        // First firing allows, an immediate re-check is held.
+        assert!(scanner_alert_emission_allows(ScannerAlertKind::ManyVersions, "bkt", "obj", cooldown));
+        assert!(!scanner_alert_emission_allows(ScannerAlertKind::ManyVersions, "bkt", "obj", cooldown));
+
+        // Different kind, object, and bucket are independent keys.
+        assert!(scanner_alert_emission_allows(ScannerAlertKind::LargeVersions, "bkt", "obj", cooldown));
+        assert!(scanner_alert_emission_allows(ScannerAlertKind::ManyVersions, "bkt", "other", cooldown));
+        assert!(scanner_alert_emission_allows(ScannerAlertKind::ManyVersions, "other", "obj", cooldown));
+        assert_eq!(cooldown_map_len(), 4);
+
+        // After the cooldown elapses the same key fires again.
+        expire_all_alert_cooldowns(cooldown);
+        assert!(scanner_alert_emission_allows(ScannerAlertKind::ManyVersions, "bkt", "obj", cooldown));
+
+        // A zero cooldown degenerates to always-emit (operators may want that).
+        assert!(scanner_alert_emission_allows(ScannerAlertKind::BigPrefix, "bkt", "dir", Duration::ZERO));
+        assert!(scanner_alert_emission_allows(ScannerAlertKind::BigPrefix, "bkt", "dir", Duration::ZERO));
+
+        // Hard bound: overflow the cap with zero-cooldown keys and confirm the
+        // map clears rather than growing past it.
+        reset_alert_cooldowns();
+        for index in 0..=(MAX_SCANNER_ALERT_COOLDOWN_KEYS + 8) {
+            let _ = scanner_alert_emission_allows(ScannerAlertKind::BigPrefix, "bkt", &format!("dir-{index}"), Duration::ZERO);
+        }
+        assert!(
+            cooldown_map_len() <= MAX_SCANNER_ALERT_COOLDOWN_KEYS,
+            "cooldown map must stay bounded, got {}",
+            cooldown_map_len()
+        );
+    }
+
     use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
     use temp_env::{with_var, with_var_unset};
     use tracing_subscriber::fmt::MakeWriter;
@@ -4400,6 +4786,104 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn scanner_trace_helpers_emit_expected_events() {
+        let mut trace = rustfs_common::trace_bus::subscribe_trace_events();
+
+        emit_scanner_folder_trace(
+            "/tmp/rustfs-scanner-trace",
+            "/tmp/rustfs-scanner-trace/bucket-a/folder-a",
+            7,
+            Some(Instant::now()),
+            "completed",
+        );
+        let folder = recv_scanner_trace_event(
+            &mut trace,
+            TraceFunc::ScannerFolder,
+            Some("bucket-a"),
+            Some("folder-a"),
+            Some("completed"),
+        )
+        .await;
+        assert_eq!(trace_attr_string(&folder, "objects").as_deref(), Some("7"));
+
+        emit_scanner_ilm_action_trace("bucket-a", "object-a", IlmAction::DeleteAction, 2, true, Some(Instant::now()));
+        let ilm = recv_scanner_trace_event(
+            &mut trace,
+            TraceFunc::ScannerIlmAction,
+            Some("bucket-a"),
+            Some("object-a"),
+            Some("queued"),
+        )
+        .await;
+        assert_eq!(trace_attr_string(&ilm, "action").as_deref(), Some("delete"));
+        assert_eq!(trace_attr_string(&ilm, "count").as_deref(), Some("2"));
+        assert_eq!(trace_attr_string(&ilm, "queued").as_deref(), Some("true"));
+
+        emit_scanner_heal_candidate_trace(ScannerHealCandidateTrace {
+            candidate_type: "object",
+            bucket: "bucket-a",
+            object: Some("object-a"),
+            version_id: Some("version-a"),
+            priority: HealChannelPriority::High,
+            scan_mode: Some(HealScanMode::Deep),
+            result: Ok(HealAdmissionResult::Merged),
+            started_at: Instant::now(),
+        });
+        let heal_candidate = recv_scanner_trace_event(
+            &mut trace,
+            TraceFunc::ScannerHealCandidate,
+            Some("bucket-a"),
+            Some("object-a"),
+            Some("admitted"),
+        )
+        .await;
+        assert_eq!(trace_attr_string(&heal_candidate, "candidate_type").as_deref(), Some("object"));
+        assert_eq!(trace_attr_string(&heal_candidate, "priority").as_deref(), Some("high"));
+        assert_eq!(trace_attr_string(&heal_candidate, "scan_mode").as_deref(), Some("deep"));
+        assert_eq!(trace_attr_string(&heal_candidate, "version_id").as_deref(), Some("version-a"));
+        assert_eq!(trace_attr_string(&heal_candidate, "admission").as_deref(), Some("merged"));
+    }
+
+    async fn recv_scanner_trace_event(
+        trace: &mut rustfs_common::trace_bus::TraceSubscription,
+        func: TraceFunc,
+        bucket: Option<&str>,
+        object: Option<&str>,
+        state: Option<&str>,
+    ) -> TraceEvent {
+        for _ in 0..32 {
+            let event = tokio::time::timeout(Duration::from_secs(1), trace.recv())
+                .await
+                .expect("scanner trace event should arrive")
+                .expect("trace bus should stay open");
+            if event.kind == TraceKind::Scanner
+                && event.func == func
+                && event.bucket.as_deref() == bucket
+                && event.object.as_deref() == object
+                && state.is_none_or(|state| trace_attr_string(&event, "state").as_deref() == Some(state))
+            {
+                return (*event).clone();
+            }
+        }
+
+        panic!("expected scanner trace event {func:?} for bucket {bucket:?} object {object:?}");
+    }
+
+    fn trace_attr_string(event: &TraceEvent, key: &str) -> Option<String> {
+        event.attrs.iter().find_map(|attr| {
+            if attr.key != key {
+                return None;
+            }
+            Some(match &attr.value {
+                rustfs_common::trace_bus::TraceVal::Bool(value) => value.to_string(),
+                rustfs_common::trace_bus::TraceVal::U64(value) => value.to_string(),
+                rustfs_common::trace_bus::TraceVal::I64(value) => value.to_string(),
+                rustfs_common::trace_bus::TraceVal::Str(value) => value.to_string(),
+            })
+        })
+    }
+
     #[test]
     fn test_build_high_priority_heal_admission_error_contains_context() {
         let err = build_high_priority_heal_admission_error(

crates/scanner/src/scanner_io.rs

@@ -231,6 +231,10 @@ pub fn record_dirty_usage_bucket(bucket: &str) {
         dirty_buckets.len()
     };
     global_metrics().record_scanner_dirty_usage_pending(usize_to_u64_saturated(pending_buckets));
+    // A write invalidates this bucket's prefix-usage answers on the spot so
+    // admin/console consumers never ride the full TTL after a change
+    // (rustfs/backlog#1872).
+    crate::prefix_usage::invalidate_prefix_usage_cache(bucket);
     DIRTY_USAGE_BUCKET_NOTIFY.notify_one();
 }
 

crates/scanner/src/storage_api.rs

@@ -78,6 +78,7 @@ pub(crate) use rustfs_ecstore::api::disk::{
 pub(crate) use rustfs_ecstore::api::error::{
     Error as EcstoreErrorType, Result as EcstoreResultType, StorageError as EcstoreStorageError,
 };
+pub(crate) use rustfs_ecstore::api::event::{EventArgs as EcstoreEventArgs, send_event as ecstore_send_event};
 #[cfg(test)]
 pub(crate) use rustfs_ecstore::api::layout::{
     EndpointServerPools as EcstoreEndpointServerPools, Endpoints as EcstoreEndpoints, PoolEndpoints as EcstorePoolEndpoints,
@@ -110,8 +111,8 @@ pub(crate) mod owner {
         ECSTORE_BUCKET_META_PREFIX, ECSTORE_RUSTFS_META_BUCKET, ECSTORE_STORAGE_FORMAT_FILE, ECSTORE_STORAGECLASS_RRS,
         ECSTORE_STORAGECLASS_STANDARD, ECSTORE_TRANSITION_COMPLETE, EcstoreBucketTargetSys, EcstoreBucketVersioningSys,
         EcstoreDisk, EcstoreDiskAPI, EcstoreDiskBytes, EcstoreDiskError, EcstoreDiskInfo, EcstoreDiskInfoOptions,
-        EcstoreDiskLocation, EcstoreDiskResult, EcstoreErrorType, EcstoreEvaluator, EcstoreEvent, EcstoreLcEventSrc,
-        EcstoreLifecycle, EcstoreListPathRawOptions, EcstoreNsScannerOpenRequest, EcstoreObjectOpts,
+        EcstoreDiskLocation, EcstoreDiskResult, EcstoreErrorType, EcstoreEvaluator, EcstoreEvent, EcstoreEventArgs,
+        EcstoreLcEventSrc, EcstoreLifecycle, EcstoreListPathRawOptions, EcstoreNsScannerOpenRequest, EcstoreObjectOpts,
         EcstoreReplicationConfigurationExt, EcstoreReplicationScannerBridge, EcstoreResultType, EcstoreScanGuard,
         EcstoreSetDisks, EcstoreStorageError, EcstoreStore, EcstoreTierConfig, EcstoreVersioningApi,
         ScannerReplicationHealObject, ScannerReplicationHealResult, ScannerReplicationQueueAdmission, ecstore_apply_expiry_rule,
@@ -121,7 +122,7 @@ pub(crate) mod owner {
         ecstore_is_erasure_sd, ecstore_is_reserved_or_invalid_bucket, ecstore_list_path_raw,
         ecstore_object_opts_from_object_info, ecstore_path2_bucket_object, ecstore_path2_bucket_object_with_base_path,
         ecstore_read_config, ecstore_replace_bucket_usage_memory_from_info, ecstore_resolve_object_store_handle,
-        ecstore_save_config, scanner_replication_config_for_lifecycle_eval,
+        ecstore_save_config, ecstore_send_event, scanner_replication_config_for_lifecycle_eval,
     };
 
     #[cfg(test)]

docs/operations/heal-concurrency-safety-notes-zh.md

@@ -0,0 +1,115 @@
+# Heal 并发安全说明（对象级 healing 标记对标审计结论）
+
+对应 backlog rustfs/backlog#1874（父 #1862，HS-12）。本文回答一个问题：MinIO 在 heal
+期间对对象打 `x-minio-healing:true` 元数据标记以防"heal 提交与并发删除/版本清理互毁"
+（cmd/xl-storage.go RenameData 的 healing 分支），RustFS 是否需要同款防御。
+
+**结论：不需要。** RustFS 不存在 MinIO 用 healing 标记防御的那类竞争：所有会触达同一
+`(bucket, object)` 提交面的路径都在同一把对象级 namespace 写锁上互斥，且 heal 的锁
+guard 覆盖 rename 提交全程；MinIO 需要标记的根因（RenameData 提交内部与版本清理逻辑
+交错）在 RustFS 的提交模型中不存在。RustFS 已有一个瞬态 healing 旗标用于另一目的
+（见下文 §2），并有并发不变量回归测试锁定本结论（§5）。
+
+## 1. 两个防御模型的对照
+
+MinIO：heal 时对对象写 `x-minio-healing:true`（持久元数据标记），后续任何 RenameData
+提交看到该标记就跳过版本清理/legacy purge 逻辑——防御发生在锁外，靠元数据让路。
+
+RustFS：三层防御，全部不依赖持久对象标记：
+
+1. **锁内互斥**：heal 与一切前台/后台写路径的提交点在同一把 `(bucket, object)` ns 写锁
+   上串行（分布式部署为 quorum 锁 RPC，单机为进程内锁管理器；锁粒度是对象级，version
+   恒为 None）。
+2. **提交模型隔离**：rename_data 提交内没有会与 heal 交错的版本清理逻辑；被替换旧版本
+   的 data_dir 物理删除被移出提交临界区（commit tail），且只删已被新提交替换的 unshared
+   目录。
+3. **瞬态 healing 旗标**：`FileInfo::set_healing`（crates/filemeta/src/fileinfo.rs）在
+   heal 提交的内存 FileInfo 上打 `"healing"` 内部键，rename_data 据此允许先清空 stale
+   目标 data_dir 再 rename——解决 heal 复用 data_dir 做 in-place 修复时 rename(2) 无法
+   替换非空目录的文件系统语义冲突（EEXIST/ENOTEMPTY）。该键是瞬态的，不落盘
+   （`is_skip_meta_key`），与 MinIO 的持久标记目的不同。非 heal 提交撞上非空目标
+   data_dir 会显式失败，有测试锁定两个方向的行为。
+
+## 2. 交点矩阵
+
+中心路径：`heal_object_with_explicit_version_regen`（crates/ecstore/src/set_disk/ops/heal.rs，
+下称 heal.rs）在入口取 `(bucket, object)` ns 写锁，guard 绑定到函数作用域末尾，覆盖
+quorum 元数据读取 → EC 重建 → 逐盘 rename 提交 → tmp 清理 → HEAL_RENAME_INCOMPLETE
+部分提交返回 → 孤儿 data_dir 回收的全过程。并发侧逐交点判定：
+
+| # | 并发路径 | 并发侧锁 | 判定 | 关键证据 |
+|---|---|---|---|---|
+| 1 | PUT 对象提交 | `put_object_commit` 对象写锁，rename_data 在锁内 | 同锁串行 | ops/object.rs 提交锁段 + rename 调用点 |
+| 2 | PUT 旧 data_dir tail 清理 | drop 对象锁后的 `commit_rename_data_dir`，无锁 | 无锁并发，语义安全（见 §3.1） | object.rs drop 后 tail 段；io_primitives.rs |
+| 3 | DELETE 单对象/版本 | `delete_object` 对象写锁，delete_version 在锁内 | 同锁串行 | object.rs delete_object 锁段 |
+| 4 | DELETE 批量 | 批量逐对象写锁（dist 走批量锁 RPC） | 同锁串行 | object.rs delete_objects 锁段 |
+| 5 | CompleteMultipart | 对象写锁 + upload 路径锁双锁，rename 在锁内 | 同锁串行 | ops/multipart.rs 提交锁段 |
+| 6 | CompleteMultipart tail 清理 | drop 对象锁后的旧 data_dir 删除 | 无锁并发，语义安全（见 §3.1） | multipart.rs drop 后 tail 段 |
+| 7 | AbortMultipart | 仅 multipart bucket 的 upload 路径锁 | 锁 key 不相交，但资源不相交（abort 不触对象 data_dir/xl.meta）→ 无实际交点 | multipart.rs abort 锁段 |
+| 8 | ILM expiry（含 DeleteAllVersions） | DeleteAllVersions 走 `delete_prefix_object=true` → 仍取对象锁；FreeVersionTask 显式取锁；noncurrent 批量走批量锁 | 同锁串行 | bucket_lifecycle_ops.rs 消费端链路 |
+| 9 | 纯 prefix 删除（绕锁能力面） | `delete_prefix`-only 不取子对象锁 | 无锁并发，但生产调用方为零（见 §3.2） | object.rs delete_object 锁条件 |
+| 10 | 孤儿 data_dir 回收 reclaim_orphan_data_dirs | 函数本体无锁；唯一生产调用方在 heal 锁内 | heal 流程内=锁内串行 | heal.rs 收尾调用；io_primitives.rs |
+| 11 | 旧清理 receipt 对账 reconcile_old_data_cleanup_receipts | 函数本体无锁；调用点在 heal 锁内 + epoch fence 防误删 | 锁内串行 | object.rs 对账函数 |
+| 12 | replication | 数据面为远端 HTTP 写（不落本地盘）；本地元数据回写走对象锁 | 同锁串行 / 无交点 | replication_resyncer.rs 链路 |
+| 13 | data_movement / rebalance / decommission 源清理 | 显式取对象锁 + 版本未变复核 + guard 复用（no_lock 只是复用已持锁） | 同锁串行 | data_movement/mod.rs 源清理 |
+| 14 | copy_object | 目标对象锁 / 走 put 链锁 | 同锁串行 | object.rs copy_object 锁段 |
+| 15 | 另一 heal 任务（跨 HealType/force_start） | dedup key 跨类型不相交 + force_start 跳过去重 → 任务级可并发 | 最终在 ns 写锁上串行 | heal/manager.rs dedup key 构成 |
+| 16 | admin `no_lock=true` heal | 客户端可控绕锁 | 无锁并发，明示运维选项（见 §3.3） | admin/handlers/heal.rs 透传 |
+| 17 | stale multipart 清理 | multipart bucket 的 upload 路径锁 | 资源不相交 → 无交点 | bucket_lifecycle_ops.rs 清理链路 |
+
+## 3. 残留窗口定性
+
+### 3.1 PUT/CompleteMultipart commit tail（交点 2/6）
+
+写路径提交成功、释放对象锁之后，才 best-effort 删除被替换的旧 data_dir（注释明示有意
+不阻塞下一操作）。该删除与并发 heal 对同一旧 data_dir 的读取/重建存在竞态窗口，但语义
+安全：
+
+- 删除目标是已被新提交替换的 unshared data_dir；heal 的 canonical 元数据来自 quorum
+  仲裁（ETag/mod_time），此时 quorum 已指向新版本，heal 不会把已替换版本当作 canonical
+  复活；
+- 竞态最坏后果 = heal 当轮对旧版本的一次 transient 失败/空转，重试轮自然收敛；清理
+  residue 会上报并重新入队 heal（`report_old_data_dir_cleanup`）；
+- 换盘重建等长 heal 走 per-version 显式版本请求，quorum 元数据在锁内读取，不受 tail
+  影响。
+
+### 3.2 纯 prefix 删除（交点 9）
+
+`delete_prefix && !delete_prefix_object` 的路径不取子对象锁（对象名空间锁无法保护前缀
+递归删除），与并发 heal 存在理论复活窗口（heal 在 prefix 删除进行中依据旧 quorum 元
+数据重建某版本）。全仓库核对结论：该路径的**生产调用方为零**——所有生产 `delete_prefix:
+true` 调用点均同时设置 `delete_prefix_object: true`（从而取对象锁）或在测试模块内。这
+是 API 能力面的暴露而非行为风险。若未来有调用方需要纯 prefix 删除，须在调用点证明与
+heal/scanner 的隔离（例如 bucket 级停扫围栏）。
+
+### 3.3 admin `no_lock=true`（交点 16）
+
+admin heal 请求可透传客户端 `nolock` 参数绕过 ns 锁（与 MinIO madmin 的同名选项对齐）。
+这是运维明示选项：使用即自负与并发写的竞争责任。文档化即可，不建议收紧。
+
+## 4. heal 侧自身的不变量保障
+
+- dedup key 跨 HealType 不相交（object/metadata/mrf/ecdecode/prefix 各自键面）+ admin
+  `force_start` 可跳过去重 → 同对象可能同时存在多个 heal 任务，但它们的执行体全部在
+  `heal_object` 入口的 ns 写锁上串行（生产入口均 `no_lock=false`）；
+- read-repair 的本地 TTL 预留只去重自身来源，不拦截其他来源的 heal——同样由 ns 锁兜底；
+- healing 旗标不落盘，故不存在"标记残留导致后续提交错误让路"的反向风险。
+
+## 5. 回归测试
+
+以下两个并发不变量测试随本审计加入 `crates/ecstore/src/set_disk/ops/heal.rs` 测试模块：
+
+- `heal_racing_version_delete_never_resurrects_the_deleted_version`：注入 doomed 版本
+  shard 损坏后，版本化 DELETE 与 Deep heal 真并发（同一把锁争用），断言已删除版本不被
+  复活、存活版本完好；
+- `heal_racing_unversioned_overwrites_preserves_the_last_commit`：非版本化覆盖提交（激活
+  commit tail 旧 data_dir 删除）与 Deep heal 循环竞态，断言最终 current 恰为最后一次
+  提交（etag 级一致）。
+
+## 6. 结论
+
+MinIO 的 `x-minio-healing` 是锁外元数据防御，前提是其 RenameData 提交内部存在与 heal
+交错的版本清理逻辑；RustFS 的提交模型把这类交错从根上消除（提交面锁内互斥 + 清理外
+移到 tail + tail 只删 unshared 旧目录），因此引入持久对象级 healing 标记没有对应的竞争
+可防，反而会引入 FileInfo 落盘格式变更与标记残留清理两类新成本。维持现状，本对标疑点
+关闭。

docs/operations/scanner-excess-alerts.md

@@ -0,0 +1,37 @@
+# Scanner Excess Alerts: Metrics, S3 Events, and Thresholds
+
+> 中文版：[scanner-excess-alerts_zh.md](scanner-excess-alerts_zh.md)
+
+Date: 2026-08-18 (rustfs/backlog#1868 / HS-04; includes the HS-15 threshold-delta notes)
+
+The background scanner detects three classes of "excess" conditions while it walks buckets and surfaces them as alerts. This page documents each alert's trigger condition, the subscribable S3 event, the cooldown semantics, and the threshold differences versus MinIO — for operators debugging alerts and for event consumers wiring up subscriptions.
+
+## The three alerts
+
+| Alert | Trigger (per scan cycle) | Metric | S3 event (RustFS wire name) | MinIO event name |
+|---|---|---|---|---|
+| Excess versions | Retained versions of one object ≥ `scanner:alert_excess_versions` | `rustfs_scanner_excess_object_versions_total{bucket}` | `s3:Scanner:ManyVersions` | `s3:ObjectManyVersions` |
+| Excess version size | Cumulative bytes of all versions of one object ≥ `scanner:alert_excess_version_size` | `rustfs_scanner_excess_object_version_size_total{bucket}` | `s3:Scanner:LargeVersions` | `s3:ObjectLargeVersions` |
+| Excess folders | Direct subfolders of one directory > `scanner:alert_excess_folders` | `rustfs_scanner_excess_folders_total{root}` | `s3:Scanner:BigPrefix` | `s3:PrefixManyFolders` |
+
+Subscribe like any bucket notification: configure a notification on the target bucket with the RustFS wire name above (or the `s3:Scanner:*` wildcard). Events carry `UserAgent: Scanner` as their origin marker, and `req_params` holds the observed value and the threshold (`versions` / `cumulativeSize` / `folders` / `threshold`), so consumers can judge severity directly.
+
+## Metrics and events fire on different cadences
+
+- **Metrics and structured logs are level-triggered**: as long as the object stays over the threshold, every scan cycle counts and logs it (default cycle ≈ 60s; see `scanner:speed`).
+- **S3 events are edge-triggered with a cooldown**: the same (alert kind, bucket, object) emits at most once per cooldown window — 24 hours by default (`RUSTFS_SCANNER_ALERT_COOLDOWN_SECS`; set it to 0 to emit every cycle). When the window lapses and the object is still over the threshold, the event fires again. The cooldown table lives in process memory with a 4096-entry hard cap; on overflow it is cleared and rebuilt (worst case: one extra emission per still-hot key).
+- A process restart resets the cooldown (every still-over-threshold object emits once more after a restart) — deliberately: restarts usually accompany incident response, and the re-emission buys visibility.
+
+## Threshold defaults and the MinIO deltas (HS-15)
+
+| Config key | ENV | RustFS default | MinIO default | Notes |
+|---|---|---|---|---|
+| `scanner:alert_excess_versions` | `RUSTFS_SCANNER_ALERT_EXCESS_VERSIONS` | 100 | 100 | Identical |
+| `scanner:alert_excess_version_size` | `RUSTFS_SCANNER_ALERT_EXCESS_VERSION_SIZE` | 1 TiB | 1 TB | Same order of magnitude; different unit basis (TiB vs TB) |
+| `scanner:alert_excess_folders` | `RUSTFS_SCANNER_ALERT_EXCESS_FOLDERS` | 65538 | 50000 | **Deliberate divergence**: 65538 tolerates the Proxmox Backup Server chunk layout (65536 chunks per directory plus the directory's own entries); MinIO's 50000 would fire continuously for PBS users. Set it to 50000 explicitly to match MinIO behavior |
+
+All three keys accept both env and admin config (`PUT /rustfs/admin/v3/config`, `scanner` subsystem); hot updates take effect immediately.
+
+## Why the event names are mapped
+
+RustFS's event enum (`rustfs_s3_types::EventName::ScannerManyVersions/LargeVersions/BigPrefix`) keeps the repo's established `s3:Scanner:*` wire names (literally different from MinIO's `s3:ObjectManyVersions`; the enum comments preserve the mapping). Subscribers should use the RustFS wire names in this page. If you need MinIO-literal compatibility, map the names on the console/consumer side — do not change the published wire names.

docs/operations/scanner-excess-alerts_zh.md

@@ -0,0 +1,37 @@
+# Scanner 超限告警：指标、S3 事件与阈值
+
+> English version: [scanner-excess-alerts.md](scanner-excess-alerts.md)
+
+日期：2026-08-18（rustfs/backlog#1868 / HS-04，含 HS-15 阈值差异说明）
+
+后台 scanner 在扫描过程中检测三类"超限"状态并对外告警。本文说明每类告警的触发条件、可订阅的 S3 事件、冷却语义，以及与 MinIO 的阈值差异，供运维排障与事件消费方对接。
+
+## 三类告警
+
+| 告警 | 触发条件（任一扫描周期） | 指标 | S3 事件（RustFS wire 名） | MinIO 对应事件名 |
+|---|---|---|---|---|
+| 版本数超限 | 单对象保留版本数 ≥ `scanner:alert_excess_versions` | `rustfs_scanner_excess_object_versions_total{bucket}` | `s3:Scanner:ManyVersions` | `s3:ObjectManyVersions` |
+| 版本总大小超限 | 单对象全部版本累计字节 ≥ `scanner:alert_excess_version_size` | `rustfs_scanner_excess_object_version_size_total{bucket}` | `s3:Scanner:LargeVersions` | `s3:ObjectLargeVersions` |
+| 子目录数超限 | 单目录直接子目录数 > `scanner:alert_excess_folders` | `rustfs_scanner_excess_folders_total{root}` | `s3:Scanner:BigPrefix` | `s3:PrefixManyFolders` |
+
+订阅方式与普通桶通知一致：对目标桶配置 notification，事件名填上表 RustFS wire 名（或通配 `s3:Scanner:*`）。事件以 `UserAgent: Scanner` 标记来源，`req_params` 携带实际值与阈值（`versions` / `cumulativeSize` / `folders` / `threshold`），便于消费方直接判断严重程度。
+
+## 指标与事件的触发节奏不同
+
+- **指标与结构化日志是电平触发**：只要对象仍在阈值之上，每个扫描周期都会计数/打日志（默认周期约 60s，见 `scanner:speed`）。
+- **S3 事件是边沿触发 + 冷却**：同一 (告警类型， 桶, 对象) 在冷却窗口内只发一次，默认 24 小时（`RUSTFS_SCANNER_ALERT_COOLDOWN_SECS`，设 0 表示每周期都发）。窗口过后对象仍超限会再次发出。冷却表在进程内有 4096 条硬顶，超限清空重建（最坏情况是每个仍超限的 key 多发一次）。
+- 进程重启会重置冷却（重启后每个仍超限的对象会再发一次）——这是有意为之：重启常伴随排障，重发提供可见性。
+
+## 阈值默认值与 MinIO 差异（HS-15）
+
+| 配置键 | ENV | RustFS 默认 | MinIO 默认 | 差异说明 |
+|---|---|---|---|---|
+| `scanner:alert_excess_versions` | `RUSTFS_SCANNER_ALERT_EXCESS_VERSIONS` | 100 | 100 | 一致 |
+| `scanner:alert_excess_version_size` | `RUSTFS_SCANNER_ALERT_EXCESS_VERSION_SIZE` | 1 TiB | 1 TB | 语义同量级，单位口径不同（TiB vs TB） |
+| `scanner:alert_excess_folders` | `RUSTFS_SCANNER_ALERT_EXCESS_FOLDERS` | 65538 | 50000 | **有意差异**：65538 兼容 Proxmox Backup Server 的 chunk 布局（每目录 65536 个 chunk + 目录自身条目），按 MinIO 的 50000 会对 PBS 用户持续误报。如需与 MinIO 行为一致可显式配置为 50000 |
+
+三个键均支持 env 与 admin config（`PUT /rustfs/admin/v3/config` 的 `scanner` 子系统）双通道，热更新即时生效。
+
+## 事件名映射的由来
+
+RustFS 的事件枚举（`rustfs_s3_types::EventName::ScannerManyVersions/LargeVersions/BigPrefix`）沿用仓库既有 wire 名 `s3:Scanner:*`（与 MinIO 的 `s3:ObjectManyVersions` 字面不同，枚举注释中保留了映射关系）。订阅方应以本文的 RustFS wire 名为准；如需 MinIO 字面兼容，请在 console/消费侧做名称映射，不要修改已发布的 wire 名。

helm/README.md

@@ -273,6 +273,10 @@ uer. `ClusterIssuer` or `Issuer`. |
 | gatewayApi.listeners.http.port| int | `8000` | Gateway API http listener port. |
 | gatewayApi.listeners.https.name | string | `websecure` | Gateway API https listener name. |
 | gatewayApi.listeners.https.port| int | `8443` | Gateway API https listener port. |
+| gatewayApi.listeners.tls.enabled | bool | `false` | Enable a TLS passthrough listener and generate a TLSRoute. |
+| gatewayApi.listeners.tls.name | string | `tls` | Gateway API TLS passthrough listener name. |
+| gatewayApi.listeners.tls.port | int | `443` | Gateway API TLS passthrough listener port. |
+| gatewayApi.listeners.tls.backendPort | int | `null` | Backend service port that terminates TLS; defaults to the console port. |
 | gatewayApi.hostname | string | Hostname to access RustFS via gateway api. |
 | gatewayApi.secretName | string | Secret tls to via RustFS using HTTPS. |
 | gatewayApi.existingGateway.name | string | `""` |  The existing gateway name, instead of creating a new one. |
@@ -447,6 +451,8 @@ rustfs-route   ["example.rustfs.com"]   172m
 
 Then, via RustFS instance via `https://example.rustfs.com` or `http://example.rustfs.com`.
 
+For end-to-end encryption, set `gatewayApi.listeners.tls.enabled` to `true`. The chart then adds a `TLS` listener with `tls.mode: Passthrough` to the `Gateway` and generates a `TLSRoute` that forwards the encrypted stream to the RustFS service, where TLS is terminated on the backend side. Note that backend TLS termination must be configured on RustFS itself (for example `RUSTFS_TLS_PATH` pointing to server certificates), and the installed Gateway API CRDs must include `TLSRoute`.
+
 # Uninstall
 
 Uninstalling the rustfs installation with command,

helm/rustfs/templates/gateway-api/gateway.yml

@@ -26,5 +26,15 @@ spec:
           - name: {{ include "rustfs.fullname" $ }}-tls
             kind: Secret
   {{- end }}
+  {{- if .tls.enabled }}
+    - name: {{ .tls.name }}
+      port: {{ .tls.port }}
+      protocol: TLS
+      tls:
+        mode: Passthrough
+      allowedRoutes:
+        namespaces:
+          from: Same
+  {{- end }}
   {{- end }}
 {{- end }}

helm/rustfs/templates/gateway-api/tlsroute.yml

@@ -0,0 +1,25 @@
+{{- if and .Values.gatewayApi.enabled .Values.gatewayApi.listeners.tls.enabled }}
+apiVersion: gateway.networking.k8s.io/v1
+kind: TLSRoute
+metadata:
+  name: {{ include "rustfs.fullname" . }}-tlsroute
+  namespace: {{ .Release.Namespace }}
+spec:
+  parentRefs:
+    {{- if .Values.gatewayApi.existingGateway.name }}
+    - name: {{ .Values.gatewayApi.existingGateway.name }}
+      {{- if .Values.gatewayApi.existingGateway.namespace }}
+      namespace: {{ .Values.gatewayApi.existingGateway.namespace }}
+      {{- end }}
+      sectionName: {{ .Values.gatewayApi.listeners.tls.name }}
+    {{- else }}
+    - name: {{ include "rustfs.fullname" $ }}-gateway
+      sectionName: {{ .Values.gatewayApi.listeners.tls.name }}
+    {{- end }}
+  hostnames:
+    - {{ .Values.gatewayApi.hostname }}
+  rules:
+    - backendRefs:
+        - name: {{ include "rustfs.fullname" . }}-svc
+          port: {{ .Values.gatewayApi.listeners.tls.backendPort | default .Values.service.console.port }}
+{{- end }}

helm/rustfs/values.yaml

@@ -369,6 +369,12 @@ gatewayApi:
     https:
       name: websecure
       port: 8443
+    tls: # Optional TLS passthrough listener; renders a TLSRoute so TLS terminates at the RustFS backend.
+      enabled: false
+      name: tls
+      port: 443
+      # Service port that terminates TLS on the backend; defaults to the console port.
+      backendPort: null
   hostname: example.rustfs.com
   httpToHttpsRedirect: true
   existingGateway:

rustfs/src/admin/handlers/heal.rs

@@ -317,6 +317,7 @@ fn add_source_counts(total: &mut rustfs_heal::HealSourceCounts, next: rustfs_hea
     total.auto_heal = total.auto_heal.saturating_add(next.auto_heal);
     total.internal = total.internal.saturating_add(next.internal);
     total.read_repair = total.read_repair.saturating_add(next.read_repair);
+    total.mrf = total.mrf.saturating_add(next.mrf);
 }
 
 fn add_operations(total: &mut rustfs_heal::HealOperationsSnapshot, next: rustfs_heal::HealOperationsSnapshot) {
@@ -2353,6 +2354,7 @@ mod tests {
             auto_heal: value,
             internal: value,
             read_repair: value,
+            mrf: value,
         };
         let operations = |value| rustfs_heal::HealOperationsSnapshot {
             queue_length: value,

rustfs/src/admin/handlers/mod.rs

@@ -64,6 +64,7 @@ mod target_descriptor;
 pub mod tier;
 pub mod tls_debug;
 pub mod trace;
+pub mod usage_prefix;
 pub mod user;
 pub mod user_iam;
 pub mod user_lifecycle;

rustfs/src/admin/handlers/profile_admin.rs

@@ -23,18 +23,24 @@ use futures::{Stream, StreamExt};
 use http::{HeaderMap, HeaderValue};
 use hyper::{Method, StatusCode};
 use matchit::Params;
+use regex::Regex;
+use rustfs_common::trace_bus::{TraceEvent, TraceKind, TraceVal, subscribe_trace_events};
 use rustfs_madmin::service_commands::ServiceTraceOpts;
+use rustfs_madmin::trace::TraceType;
 use rustfs_policy::policy::action::{Action, AdminAction};
 use s3s::header::CONTENT_TYPE;
 use s3s::stream::{ByteStream, DynByteStream};
 use s3s::{Body, S3Request, S3Response, S3Result, StdError, s3_error};
 use serde::Serialize;
+use std::collections::HashMap;
 use std::pin::Pin;
 use std::task::{Context, Poll};
-use std::time::Duration;
+use std::time::{Duration, SystemTime};
+use time::{OffsetDateTime, format_description::well_known::Rfc3339};
 use tokio::sync::mpsc;
 use tokio_stream::wrappers::ReceiverStream;
 use tracing::error;
+use url::form_urlencoded;
 
 #[derive(Serialize)]
 struct ProfileStatus {
@@ -206,16 +212,164 @@ impl Stream for TraceStream {
 
 impl ByteStream for TraceStream {}
 
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+struct TraceKindFilter {
+    heal: bool,
+    scanner: bool,
+}
+
+impl TraceKindFilter {
+    const ALL_SUPPORTED: Self = Self {
+        heal: true,
+        scanner: true,
+    };
+
+    fn from_request(uri: &hyper::Uri, trace_types: TraceType) -> S3Result<Self> {
+        let mut has_kind = false;
+        let mut filter = Self {
+            heal: false,
+            scanner: false,
+        };
+
+        for (key, value) in trace_query_pairs(uri) {
+            if key != "kind" {
+                continue;
+            }
+            has_kind = true;
+            for item in value.split(',') {
+                match item.trim().to_ascii_lowercase().as_str() {
+                    "heal" | "healing" => filter.heal = true,
+                    "scanner" => filter.scanner = true,
+                    "all" => return Ok(Self::ALL_SUPPORTED),
+                    _ => return Err(s3_error!(InvalidRequest, "invalid trace kind")),
+                }
+            }
+        }
+
+        if has_kind {
+            return Ok(filter);
+        }
+
+        if trace_types.mask() == 0 || trace_query_flag(uri, "all") {
+            return Ok(Self::ALL_SUPPORTED);
+        }
+
+        Ok(Self {
+            heal: trace_types.overlaps(&TraceType::HEALING),
+            scanner: trace_types.overlaps(&TraceType::SCANNER),
+        })
+    }
+
+    const fn matches(self, kind: TraceKind) -> bool {
+        match kind {
+            TraceKind::Heal => self.heal,
+            TraceKind::Scanner => self.scanner,
+        }
+    }
+}
+
+#[derive(Debug)]
+struct TraceStreamFilter {
+    kinds: TraceKindFilter,
+    regex: Option<Regex>,
+    threshold: Duration,
+}
+
+impl TraceStreamFilter {
+    fn from_request(uri: &hyper::Uri, opts: &ServiceTraceOpts) -> S3Result<Self> {
+        if opts.only_errors() {
+            return Err(s3_error!(
+                InvalidRequest,
+                "trace error-only filter is not supported for heal/scanner trace"
+            ));
+        }
+
+        Ok(Self {
+            kinds: TraceKindFilter::from_request(uri, opts.trace_types())?,
+            regex: trace_regex_filter(uri)?,
+            threshold: opts.threshold(),
+        })
+    }
+
+    fn matches_kind(&self, kind: TraceKind) -> bool {
+        self.kinds.matches(kind)
+    }
+
+    fn matches_record(&self, record: &TraceWireRecord) -> bool {
+        record.duration >= self.threshold && self.regex.as_ref().is_none_or(|regex| record.matches_regex(regex))
+    }
+}
+
+#[derive(Serialize)]
+struct TraceWireRecord {
+    #[serde(rename = "type")]
+    trace_type: u64,
+    #[serde(rename = "nodename")]
+    node_name: String,
+    #[serde(rename = "funcname")]
+    func_name: String,
+    #[serde(rename = "time")]
+    time: String,
+    #[serde(rename = "path")]
+    path: String,
+    #[serde(rename = "dur")]
+    duration: Duration,
+    #[serde(rename = "bytes", skip_serializing_if = "Option::is_none")]
+    bytes: Option<i64>,
+    #[serde(rename = "msg", skip_serializing_if = "Option::is_none")]
+    message: Option<String>,
+    #[serde(rename = "custom", skip_serializing_if = "Option::is_none")]
+    custom: Option<HashMap<String, String>>,
+}
+
+impl TraceWireRecord {
+    fn from_event(node_name: &str, event: &TraceEvent) -> Self {
+        Self {
+            trace_type: trace_type_mask(event.kind),
+            node_name: node_name.to_owned(),
+            func_name: event.func.as_str().to_owned(),
+            time: trace_time_string(event.time),
+            path: trace_path(event),
+            duration: event.duration,
+            bytes: trace_bytes(event.bytes),
+            message: None,
+            custom: trace_custom_attrs(event),
+        }
+    }
+
+    fn dropped(node_name: &str, dropped: u64) -> Self {
+        let mut custom = HashMap::new();
+        custom.insert("dropped_events".to_string(), dropped.to_string());
+
+        Self {
+            trace_type: 0,
+            node_name: node_name.to_owned(),
+            func_name: "trace.Dropped".to_string(),
+            time: trace_time_string(SystemTime::now()),
+            path: String::new(),
+            duration: Duration::ZERO,
+            bytes: None,
+            message: Some("trace subscriber lagged".to_string()),
+            custom: Some(custom),
+        }
+    }
+
+    fn matches_regex(&self, regex: &Regex) -> bool {
+        regex.is_match(&self.func_name)
+            || regex.is_match(&self.path)
+            || self.message.as_ref().is_some_and(|message| regex.is_match(message))
+            || self
+                .custom
+                .as_ref()
+                .is_some_and(|custom| custom.iter().any(|(key, value)| regex.is_match(key) || regex.is_match(value)))
+    }
+}
+
 /// `GET /v3/trace` — stream real-time server trace events.
 ///
-/// RustFS emits diagnostics through the `tracing` pipeline but does not expose
-/// an in-process subscriber that can fan trace events out to an admin client
-/// (there is no request-trace broadcast channel). Rather than return an opaque
-/// `501` — which would make `mc admin trace` fail to connect — this honors the
-/// streaming NDJSON contract: it validates the requested trace filters, opens
-/// the stream, emits a single capability record explaining that live tracing is
-/// not wired, then holds the connection open with keep-alives. No fabricated
-/// trace records are ever sent.
+/// RustFS currently publishes heal and scanner diagnostics through the common
+/// trace bus. The admin endpoint exposes those events as MinIO-shaped NDJSON
+/// records while keeping unsupported trace classes filtered out.
 pub struct TraceHandler {}
 
 #[async_trait::async_trait]
@@ -228,24 +382,13 @@ impl Operation for TraceHandler {
         let mut opts = ServiceTraceOpts::default();
         opts.parse_params(&req.uri)
             .map_err(|_| s3_error!(InvalidRequest, "invalid trace parameters"))?;
+        let filter = TraceStreamFilter::from_request(&req.uri, &opts)?;
 
         let node_name = sysinfo::System::host_name().unwrap_or_else(|| "rustfs".to_string());
-        let (tx, rx) = mpsc::channel::<Result<Bytes, StdError>>(8);
+        let mut subscription = subscribe_trace_events();
+        let (tx, rx) = mpsc::channel::<Result<Bytes, StdError>>(64);
 
         spawn_traced(async move {
-            let notice = serde_json::json!({
-                "nodename": node_name,
-                "funcname": "admin.Trace",
-                "msg": "RustFS does not expose an in-process trace-event subscriber; live tracing is not yet available",
-                "err": "trace_streaming_unsupported",
-            });
-            if let Ok(mut encoded) = serde_json::to_vec(&notice) {
-                encoded.push(b'\n');
-                if tx.send(Ok(Bytes::from(encoded))).await.is_err() {
-                    return;
-                }
-            }
-
             let mut ticker = tokio::time::interval(Duration::from_secs(15));
             ticker.tick().await;
             loop {
@@ -256,6 +399,26 @@ impl Operation for TraceHandler {
                             break;
                         }
                     }
+                    received = subscription.recv() => {
+                        match received {
+                            Ok(event) => {
+                                if !filter.matches_kind(event.kind) {
+                                    continue;
+                                }
+                                let record = TraceWireRecord::from_event(&node_name, &event);
+                                if filter.matches_record(&record) && send_trace_record(&tx, &record).await.is_err() {
+                                    break;
+                                }
+                            }
+                            Err(tokio::sync::broadcast::error::RecvError::Lagged(dropped)) => {
+                                let record = TraceWireRecord::dropped(&node_name, dropped);
+                                if send_trace_record(&tx, &record).await.is_err() {
+                                    break;
+                                }
+                            }
+                            Err(tokio::sync::broadcast::error::RecvError::Closed) => break,
+                        }
+                    }
                 }
             }
         });
@@ -269,17 +432,115 @@ impl Operation for TraceHandler {
     }
 }
 
+async fn send_trace_record(tx: &mpsc::Sender<Result<Bytes, StdError>>, record: &TraceWireRecord) -> Result<(), ()> {
+    let Some(encoded) = encode_ndjson(record) else {
+        return Ok(());
+    };
+    tx.send(Ok(encoded)).await.map_err(|_| ())
+}
+
+fn encode_ndjson(value: &impl Serialize) -> Option<Bytes> {
+    let mut encoded = serde_json::to_vec(value).ok()?;
+    encoded.push(b'\n');
+    Some(Bytes::from(encoded))
+}
+
+fn trace_query_pairs(uri: &hyper::Uri) -> impl Iterator<Item = (String, String)> + '_ {
+    uri.query()
+        .into_iter()
+        .flat_map(|query| form_urlencoded::parse(query.as_bytes()))
+        .map(|(key, value)| (key.into_owned(), value.into_owned()))
+}
+
+fn trace_query_flag(uri: &hyper::Uri, flag: &str) -> bool {
+    trace_query_pairs(uri).any(|(key, value)| key == flag && value == "true")
+}
+
+fn trace_regex_filter(uri: &hyper::Uri) -> S3Result<Option<Regex>> {
+    trace_query_pairs(uri)
+        .find_map(|(key, value)| {
+            if key == "filter" && !value.is_empty() {
+                Some(value)
+            } else {
+                None
+            }
+        })
+        .map(|pattern| Regex::new(&pattern).map_err(|_| s3_error!(InvalidRequest, "invalid trace filter")))
+        .transpose()
+}
+
+fn trace_type_mask(kind: TraceKind) -> u64 {
+    match kind {
+        TraceKind::Heal => TraceType::HEALING.mask(),
+        TraceKind::Scanner => TraceType::SCANNER.mask(),
+    }
+}
+
+fn trace_time_string(time: SystemTime) -> String {
+    match OffsetDateTime::from(time).format(&Rfc3339) {
+        Ok(value) => value,
+        Err(_) => "1970-01-01T00:00:00Z".to_string(),
+    }
+}
+
+fn trace_path(event: &TraceEvent) -> String {
+    match (event.bucket.as_deref(), event.object.as_deref()) {
+        (Some(bucket), Some(object)) if !object.is_empty() => format!("{bucket}/{object}"),
+        (Some(bucket), _) => bucket.to_owned(),
+        (None, Some(object)) => object.to_owned(),
+        (None, None) => String::new(),
+    }
+}
+
+fn trace_bytes(bytes: u64) -> Option<i64> {
+    if bytes == 0 {
+        return None;
+    }
+
+    match i64::try_from(bytes) {
+        Ok(value) => Some(value),
+        Err(_) => Some(i64::MAX),
+    }
+}
+
+fn trace_custom_attrs(event: &TraceEvent) -> Option<HashMap<String, String>> {
+    if event.attrs.is_empty() {
+        return None;
+    }
+
+    Some(
+        event
+            .attrs
+            .iter()
+            .map(|attr| (attr.key.to_string(), trace_value_string(&attr.value)))
+            .collect(),
+    )
+}
+
+fn trace_value_string(value: &TraceVal) -> String {
+    match value {
+        TraceVal::Bool(value) => value.to_string(),
+        TraceVal::U64(value) => value.to_string(),
+        TraceVal::I64(value) => value.to_string(),
+        TraceVal::Str(value) => value.to_string(),
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::{
         ProfileControlHandler, ProfileHandler, ProfileStatusHandler, ProfilingDownloadHandler, ProfilingStartHandler,
-        TraceHandler,
+        TraceHandler, TraceKindFilter, TraceStreamFilter, TraceWireRecord,
     };
     use crate::admin::router::Operation;
     use http::{Extensions, HeaderMap, Uri};
     use hyper::Method;
     use matchit::Params;
-    use s3s::{Body, S3ErrorCode, S3Request};
+    use rustfs_common::trace_bus::{TraceEvent, TraceFunc, TraceKind};
+    use rustfs_madmin::service_commands::ServiceTraceOpts;
+    use rustfs_madmin::trace::TraceType;
+    use s3s::{Body, S3ErrorCode, S3Request, S3Result};
+    use std::time::{Duration, UNIX_EPOCH};
 
     fn build_profile_request(uri: &'static str) -> S3Request<Body> {
         S3Request {
@@ -295,6 +556,13 @@ mod tests {
         }
     }
 
+    fn build_trace_stream_filter(uri: &'static str) -> S3Result<TraceStreamFilter> {
+        let uri = Uri::from_static(uri);
+        let mut opts = ServiceTraceOpts::default();
+        opts.parse_params(&uri).expect("test trace params should parse");
+        TraceStreamFilter::from_request(&uri, &opts)
+    }
+
     #[tokio::test]
     async fn profile_handler_rejects_missing_credentials() {
         let result = ProfileHandler {}
@@ -358,4 +626,108 @@ mod tests {
             .expect_err("trace must reject anonymous requests");
         assert_eq!(err.code(), &S3ErrorCode::AccessDenied);
     }
+
+    #[test]
+    fn trace_kind_filter_supports_kind_query() {
+        let uri = Uri::from_static("/rustfs/admin/v3/trace?kind=heal");
+        let filter = TraceKindFilter::from_request(&uri, TraceType::default()).expect("kind filter should parse");
+
+        assert!(filter.matches(TraceKind::Heal));
+        assert!(!filter.matches(TraceKind::Scanner));
+    }
+
+    #[test]
+    fn trace_kind_filter_defaults_to_supported_events_without_type_flags() {
+        let uri = Uri::from_static("/rustfs/admin/v3/trace");
+        let filter = TraceKindFilter::from_request(&uri, TraceType::default()).expect("empty filter should parse");
+
+        assert!(filter.matches(TraceKind::Heal));
+        assert!(filter.matches(TraceKind::Scanner));
+    }
+
+    #[test]
+    fn trace_kind_filter_rejects_unknown_kind() {
+        let uri = Uri::from_static("/rustfs/admin/v3/trace?kind=s3");
+        let err = TraceKindFilter::from_request(&uri, TraceType::default()).expect_err("unknown kind should fail");
+
+        assert_eq!(err.code(), &S3ErrorCode::InvalidRequest);
+    }
+
+    #[test]
+    fn trace_stream_filter_matches_regex_against_path_and_attrs() {
+        let filter = build_trace_stream_filter("/rustfs/admin/v3/trace?kind=heal&filter=data/.%2Bxl.meta")
+            .expect("regex filter should parse");
+        let event = TraceEvent::new(TraceKind::Heal, TraceFunc::HealObject)
+            .with_bucket("data")
+            .with_object("dir/xl.meta")
+            .with_attr("dry_run", true);
+        let record = TraceWireRecord::from_event("node-a", &event);
+
+        assert!(filter.matches_kind(event.kind));
+        assert!(filter.matches_record(&record));
+    }
+
+    #[test]
+    fn trace_stream_filter_rejects_invalid_regex() {
+        let err = build_trace_stream_filter("/rustfs/admin/v3/trace?kind=heal&filter=[").expect_err("invalid regex should fail");
+
+        assert_eq!(err.code(), &S3ErrorCode::InvalidRequest);
+    }
+
+    #[test]
+    fn trace_stream_filter_applies_threshold() {
+        let filter =
+            build_trace_stream_filter("/rustfs/admin/v3/trace?kind=heal&threshold=10ms").expect("threshold should parse");
+        let short = TraceWireRecord::from_event(
+            "node-a",
+            &TraceEvent::new(TraceKind::Heal, TraceFunc::HealObject).with_duration(Duration::from_millis(9)),
+        );
+        let long = TraceWireRecord::from_event(
+            "node-a",
+            &TraceEvent::new(TraceKind::Heal, TraceFunc::HealObject).with_duration(Duration::from_millis(10)),
+        );
+
+        assert!(!filter.matches_record(&short));
+        assert!(filter.matches_record(&long));
+    }
+
+    #[test]
+    fn trace_stream_filter_rejects_error_only_filter() {
+        let err = build_trace_stream_filter("/rustfs/admin/v3/trace?kind=heal&err=true").expect_err("err filter should fail");
+
+        assert_eq!(err.code(), &S3ErrorCode::InvalidRequest);
+    }
+
+    #[test]
+    fn trace_wire_record_contains_madmin_trace_fields() {
+        let event = TraceEvent::new(TraceKind::Heal, TraceFunc::HealObject)
+            .with_bucket("bucket")
+            .with_object("object")
+            .with_duration(Duration::from_millis(3))
+            .with_bytes(17)
+            .with_attr("dry", true);
+        let mut record = TraceWireRecord::from_event("node-a", &event);
+        record.time = "1970-01-01T00:00:00Z".to_string();
+
+        let value = serde_json::to_value(&record).expect("trace record should serialize");
+
+        assert_eq!(value["type"], TraceType::HEALING.mask());
+        assert_eq!(value["nodename"], "node-a");
+        assert_eq!(value["funcname"], "heal.Object");
+        assert_eq!(value["time"], "1970-01-01T00:00:00Z");
+        assert_eq!(value["path"], "bucket/object");
+        assert_eq!(value["bytes"], 17);
+        assert_eq!(value["custom"]["dry"], "true");
+    }
+
+    #[test]
+    fn trace_wire_record_formats_epoch_time() {
+        let event = TraceEvent {
+            time: UNIX_EPOCH,
+            ..TraceEvent::new(TraceKind::Scanner, TraceFunc::ScannerFolder)
+        };
+        let record = TraceWireRecord::from_event("node-a", &event);
+
+        assert_eq!(record.time, "1970-01-01T00:00:00Z");
+    }
 }

rustfs/src/admin/handlers/site_replication.rs

@@ -6023,7 +6023,10 @@ fn edit_generation_wall_clock() -> u64 {
 /// node's clock behind the clock that fed the previous lifetime) mints
 /// below the stale mark and the origin stays fenced — but only until real
 /// time passes the previous lifetime's last allocation, because every later
-/// allocation takes the wall-clock floor again. Bounded by the skew,
+/// allocation takes the wall-clock floor again (and never longer than
+/// [`PEER_EDIT_FENCE_STALENESS_WINDOW_NANOS`]: a regression past the window
+/// leaves the mark implausibly distant and the origin runs unfenced
+/// immediately). Bounded by the skew,
 /// self-healing, and no rollback window beyond the plain counter's: a
 /// delivery applies only at or above the receiver's mark, so the one
 /// cross-lifetime interleaving that can apply stale content — a
@@ -6063,6 +6066,52 @@ fn peer_edit_fence(queries: &HashMap<String, String>) -> Option<(String, u64)> {
     Some((origin.clone(), generation))
 }
 
+/// How far below the recorded high-water mark a delivery may sit and still
+/// be fenced as stale. The distance a GENUINE superseded delivery can trail
+/// its origin's mark is small: retransmissions re-run the sender flow and
+/// mint a fresh generation (the retry queue keys on the bare path and never
+/// replays a fenced URL), so only an in-flight straggler of the losing
+/// fan-out race trails the mark, by delivery latency — minutes at the
+/// outside. A mark further above than this window cannot be explained by
+/// any genuine race, only by a forged fence (the shared service account
+/// lets any peer stamp any origin) or by a persisted clock excursion the
+/// origin has since left behind — and fencing on it would silently drop the
+/// origin's real edits, so the stale check ignores it instead.
+const PEER_EDIT_FENCE_STALENESS_WINDOW_NANOS: u64 = 24 * 60 * 60 * 1_000_000_000;
+
+/// Whether an incoming fence may be honoured, as far as this site can vouch
+/// for it. The sender's identity is unverifiable (shared service account),
+/// so the check runs over what the receiving state knows: the claimed origin
+/// must be a site this state currently replicates with — the same membership
+/// rule the load-time mark pruning applies, so every mark recorded behind
+/// this check is one a reload would keep — and not this site itself, which
+/// never delivers edits to itself. The caller IGNORES an inadmissible fence
+/// rather than failing the request: the delivery applies exactly as an
+/// unstamped (pre-fence) delivery would, no high-water mark is read or
+/// written, and the worst a forged fence achieves is forfeiting an ordering
+/// guarantee its sender was never owed. The generation itself is NOT
+/// bounded here: a genuine origin whose hybrid clock persisted a wall-clock
+/// excursion allocates arbitrarily far in the future, and refusing to
+/// record its marks would strip the ordering fence from exactly the
+/// deliveries that still race — the staleness window on the read side is
+/// what defuses forged marks instead.
+fn peer_edit_fence_is_admissible(state: &SiteReplicationState, local_deployment_id: &str, fence: &(String, u64)) -> bool {
+    let (origin, generation) = fence;
+    if origin != local_deployment_id && state.peers.contains_key(origin) {
+        return true;
+    }
+    warn!(
+        event = EVENT_ADMIN_SITE_REPLICATION_STATE,
+        component = LOG_COMPONENT_ADMIN,
+        subsystem = LOG_SUBSYSTEM_SITE_REPLICATION,
+        result = "fence_origin_not_a_remote_peer",
+        origin = %origin,
+        generation = *generation,
+        "ignoring inadmissible peer-edit fence"
+    );
+    false
+}
+
 /// True when a strictly newer edit from the same origin site already landed
 /// here. No lock on the sending side can order deliveries issued by two
 /// nodes of that site, so ordering is decided here, on the generation the
@@ -6070,11 +6119,42 @@ fn peer_edit_fence(queries: &HashMap<String, String>) -> Option<(String, u64)> {
 /// stale: one edit legitimately fans out several deliveries under a single
 /// generation (the ILM-expiry edit sends every peer's record), and a replay of
 /// an applied delivery re-applies the same edit idempotently.
+///
+/// A mark more than [`PEER_EDIT_FENCE_STALENESS_WINDOW_NANOS`] above the
+/// delivery is implausible and does NOT fence: the shared service account
+/// means any peer can stamp any origin, so a forged `u64::MAX`-scale mark
+/// would otherwise silently swallow the origin's genuine edits for good.
+/// Bounding the fence by distance instead of by an absolute ceiling keeps
+/// ordering intact wherever the origin's clock actually operates — two
+/// racing deliveries trail each other by seconds whether the hybrid clock
+/// tracks wall time or persists a long-gone excursion far ahead of it —
+/// while a mark no genuine race can explain merely downgrades the origin to
+/// unfenced (pre-fence) delivery instead of dropping its edits. (One genuine
+/// shape does land out here: a plain-counter straggler arriving after its
+/// origin's first hybrid-clock edit. It gets the same downgrade — applied
+/// unfenced — once, at upgrade time; fencing it instead would silence the
+/// mirror case, a hybrid-clock origin downgraded back to the plain counter.)
 fn peer_edit_delivery_is_stale(state: &SiteReplicationState, origin: &str, generation: u64) -> bool {
-    state
-        .applied_edit_generations
-        .get(origin)
-        .is_some_and(|applied| *applied > generation)
+    let Some(applied) = state.applied_edit_generations.get(origin) else {
+        return false;
+    };
+    if *applied <= generation {
+        return false;
+    }
+    if *applied - generation > PEER_EDIT_FENCE_STALENESS_WINDOW_NANOS {
+        warn!(
+            event = EVENT_ADMIN_SITE_REPLICATION_STATE,
+            component = LOG_COMPONENT_ADMIN,
+            subsystem = LOG_SUBSYSTEM_SITE_REPLICATION,
+            result = "fence_mark_beyond_staleness_window",
+            origin,
+            generation,
+            applied_mark = *applied,
+            "ignoring implausibly distant peer-edit high-water mark"
+        );
+        return false;
+    }
+    true
 }
 
 fn record_applied_peer_edit_generation(state: &mut SiteReplicationState, origin: &str, generation: u64) {
@@ -10698,6 +10778,11 @@ impl Operation for SRPeerEditHandler {
         let outcome = update_site_replication_state_when_changed(move |state| {
             let mut incoming = incoming;
             let local_peer = local_peer_at_endpoint(commit_endpoint, state);
+            // The fence is self-reported — the shared service account means
+            // the sender cannot be identified — so it is honoured only after
+            // the admissibility check, against the same state it will gate.
+            let commit_fence =
+                commit_fence.filter(|fence| peer_edit_fence_is_admissible(state, &local_peer.deployment_id, fence));
             // Ordering fence: the sending site allocates the generation under
             // its state-object lock, so a delivery that lost the race carries
             // a generation this site has already passed. Applying it would
@@ -13393,6 +13478,15 @@ mod tests {
             handler_block.contains("record_applied_peer_edit_generation(state, origin, *generation);"),
             "SRPeerEditHandler must record the applied generation so later stale deliveries are recognised"
         );
+        // Fence hardening: origin and generation are self-reported by a
+        // caller the shared service account cannot identify, so the handler
+        // must pass the fence through the admissibility check — against the
+        // same state the fence gates, i.e. inside the transaction — before
+        // reading or raising any high-water mark.
+        assert!(
+            handler_block.contains(".filter(|fence| peer_edit_fence_is_admissible(state, &local_peer.deployment_id, fence))"),
+            "SRPeerEditHandler must admit a fence only through peer_edit_fence_is_admissible inside the state transaction"
+        );
         // P1-15 PR2: both halves of the fence and the edit they fence share
         // ONE transaction. Checking the fence against a state read outside the
         // lock would let the check pass on one snapshot and the write land on
@@ -14769,6 +14863,121 @@ mod tests {
         assert!(peer_edit_delivery_is_stale(&state, origin, generation - 1));
     }
 
+    /// A fence is self-reported: every site authenticates peer traffic with
+    /// the same site-replicator credential, so a compromised peer can stamp
+    /// ANY origin with ANY generation. An origin the receiver does not
+    /// replicate with — or the receiver itself — is ignored and plants no
+    /// mark; a mark a compromised peer plants for a CURRENT origin cannot
+    /// silence that origin, because the staleness window refuses to fence on
+    /// a mark implausibly far above the genuine deliveries.
+    #[test]
+    fn forged_peer_edit_fences_cannot_poison_the_high_water_marks() {
+        let mut state = SiteReplicationState {
+            peers: BTreeMap::from([
+                (
+                    "site-local".to_string(),
+                    PeerInfo {
+                        deployment_id: "site-local".to_string(),
+                        ..peer("local", "https://local.example:9000")
+                    },
+                ),
+                (
+                    "site-victim".to_string(),
+                    PeerInfo {
+                        deployment_id: "site-victim".to_string(),
+                        ..peer("victim", "https://victim.example:9000")
+                    },
+                ),
+            ]),
+            ..Default::default()
+        };
+        // An origin outside the current membership is refused outright...
+        let unknown = ("site-unknown".to_string(), 4u64);
+        assert!(!peer_edit_fence_is_admissible(&state, "site-local", &unknown));
+
+        // No site delivers edits to itself: a fence claiming the receiver as
+        // its origin is forged by construction, current peer or not.
+        let own = ("site-local".to_string(), 4u64);
+        assert!(!peer_edit_fence_is_admissible(&state, "site-local", &own));
+
+        // A current remote peer's fence is admitted and works end to end.
+        let genuine = ("site-victim".to_string(), 1u64);
+        assert!(peer_edit_fence_is_admissible(&state, "site-local", &genuine));
+        assert!(!peer_edit_delivery_is_stale(&state, &genuine.0, genuine.1));
+        record_applied_peer_edit_generation(&mut state, &genuine.0, genuine.1);
+        assert_eq!(state.applied_edit_generations.get("site-victim"), Some(&1));
+
+        // A forged u64::MAX-scale mark CAN be recorded — the shared service
+        // account means the receiver cannot tell the stamp was forged — but
+        // it is inert: the victim's genuine hybrid-clock deliveries sit far
+        // more than the staleness window below it, so they keep applying
+        // instead of being silently acked-and-dropped.
+        record_applied_peer_edit_generation(&mut state, "site-victim", u64::MAX);
+        assert!(!peer_edit_delivery_is_stale(&state, "site-victim", edit_generation_wall_clock()));
+    }
+
+    /// The staleness window bounds the fence by DISTANCE from the mark, not
+    /// by an absolute clock ceiling, so ordering must hold wherever the
+    /// origin's hybrid clock actually operates. The regression that matters:
+    /// a temporary wall-clock excursion far in the future is persisted by
+    /// `next_peer_edit_generation` (`max(now, prev + 1)` never comes back
+    /// down), and two later edits g+1 then g can arrive in reverse order —
+    /// g must still be fenced, even though both generations dwarf the
+    /// receiver's clock. Conversely a mark further above a delivery than any
+    /// genuine race can explain must not fence it.
+    #[test]
+    fn peer_edit_fence_orders_a_persisted_future_clock_and_defuses_distant_marks() {
+        let mut state = SiteReplicationState {
+            peers: BTreeMap::from([(
+                "site-origin".to_string(),
+                PeerInfo {
+                    deployment_id: "site-origin".to_string(),
+                    ..peer("origin", "https://origin.example:9000")
+                },
+            )]),
+            ..Default::default()
+        };
+
+        // The origin's clock once jumped ten years ahead; the hybrid clock
+        // keeps allocating from there long after the clock was corrected.
+        let excursion = edit_generation_wall_clock() + 10 * 365 * 24 * 60 * 60 * 1_000_000_000;
+        let fence = ("site-origin".to_string(), excursion + 1);
+        assert!(peer_edit_fence_is_admissible(&state, "site-local", &fence));
+        record_applied_peer_edit_generation(&mut state, &fence.0, fence.1);
+
+        // The reverse delivery of the race: g arrives after g+1 landed.
+        // Without the fence it would commit last and roll g+1 back.
+        assert!(peer_edit_delivery_is_stale(&state, "site-origin", excursion));
+        // Equal generation (same edit's fan-out or a replay) still applies,
+        // as does the next edit.
+        assert!(!peer_edit_delivery_is_stale(&state, "site-origin", excursion + 1));
+        assert!(!peer_edit_delivery_is_stale(&state, "site-origin", excursion + 2));
+
+        // The window's exact boundary: a delivery trailing the mark by the
+        // full window is still fenced; one nanosecond further is not — that
+        // distance is no longer explicable by a genuine race, only by a
+        // forged mark or an excursion the origin has left behind.
+        let mark = fence.1;
+        // A straggler trailing by a concrete hour must still be fenced —
+        // pins the window's real magnitude, not just its symbolic boundary.
+        assert!(peer_edit_delivery_is_stale(&state, "site-origin", mark - 60 * 60 * 1_000_000_000));
+        assert!(peer_edit_delivery_is_stale(
+            &state,
+            "site-origin",
+            mark - PEER_EDIT_FENCE_STALENESS_WINDOW_NANOS
+        ));
+        assert!(!peer_edit_delivery_is_stale(
+            &state,
+            "site-origin",
+            mark - PEER_EDIT_FENCE_STALENESS_WINDOW_NANOS - 1
+        ));
+
+        // A pre-hybrid plain-counter origin trails such a mark by eons: it
+        // is not fenced (the rc.2-era downgrade case), it just runs
+        // unfenced until its counter regime catches up.
+        assert!(!peer_edit_delivery_is_stale(&state, "site-origin", 3));
+    }
+
     /// P1-15 review follow-up: a site that leaves the mesh drops below two
     /// peers, which clears its state object and restarts its generation
     /// counter at zero. A mark left over from its previous membership would

rustfs/src/admin/handlers/system.rs

@@ -1158,10 +1158,10 @@ impl Operation for RuntimeCapabilitiesHandler {
     }
 }
 
-/// Authorization gate for GET datausageinfo: any-of the dedicated admin action
-/// OR the bucket listing action. Pinned by a unit test so the gate cannot
-/// silently narrow or widen (rustfs/backlog#1306).
-fn data_usage_info_gate_actions() -> Vec<Action> {
+/// Authorization gate for GET datausageinfo (and prefix usage): any-of the
+/// dedicated admin action OR the bucket listing action. Pinned by a unit test
+/// so the gate cannot silently narrow or widen (rustfs/backlog#1306).
+pub(crate) fn data_usage_info_gate_actions() -> Vec<Action> {
     vec![
         Action::AdminAction(AdminAction::DataUsageInfoAdminAction),
         Action::S3Action(S3Action::ListBucketAction),

rustfs/src/admin/handlers/usage_prefix.rs

@@ -0,0 +1,142 @@
+// 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.
+
+//! Prefix-level bucket usage admin handler (rustfs/backlog#1872).
+//!
+//! `GET /rustfs/admin/v3/usage/{bucket}?prefix=&max-entries=` answers
+//! "what does this bucket / this prefix hold" from the scanner's per-set
+//! usage caches, with a one-level sub-prefix breakdown — the data console
+//! buckets view MinIO serves from `loadPrefixUsageFromBackend`.
+
+use crate::admin::auth::validate_admin_request;
+use crate::admin::handlers::system::data_usage_info_gate_actions;
+use crate::admin::router::{AdminOperation, Operation, S3Router};
+use crate::auth::{check_key_valid, get_session_token};
+use crate::server::{ADMIN_PREFIX, RemoteAddr};
+use http::{HeaderMap, HeaderValue, StatusCode};
+use hyper::Method;
+use matchit::Params;
+use s3s::header::CONTENT_TYPE;
+use s3s::{Body, S3Error, S3ErrorCode, S3Request, S3Response, S3Result, s3_error};
+
+const JSON_CONTENT_TYPE: &str = "application/json";
+const DEFAULT_MAX_ENTRIES: usize = 1000;
+const MAX_ENTRIES_LIMIT: usize = 10_000;
+
+pub struct BucketPrefixUsageHandler {}
+
+pub fn register_usage_prefix_route(r: &mut S3Router<AdminOperation>) -> std::io::Result<()> {
+    r.insert(
+        Method::GET,
+        format!("{}{}", ADMIN_PREFIX, "/v3/usage/{bucket}").as_str(),
+        AdminOperation(&BucketPrefixUsageHandler {}),
+    )?;
+    Ok(())
+}
+
+/// Parse `prefix` and `max-entries` from the query string. Unknown keys are
+/// rejected so a typo'd parameter cannot silently change the answer's shape.
+fn parse_usage_prefix_query(query: Option<&str>) -> S3Result<(String, usize)> {
+    let mut prefix: Option<String> = None;
+    let mut max_entries: Option<usize> = None;
+    for (key, value) in url::form_urlencoded::parse(query.unwrap_or_default().as_bytes()) {
+        match key.as_ref() {
+            "prefix" => prefix = Some(value.into_owned()),
+            "max-entries" => {
+                max_entries = Some(
+                    value
+                        .parse::<usize>()
+                        .map_err(|_| s3_error!(InvalidArgument, "max-entries must be a positive integer"))?,
+                );
+            }
+            other => return Err(s3_error!(InvalidArgument, "unknown query parameter: {other}")),
+        }
+    }
+    let max_entries = max_entries.unwrap_or(DEFAULT_MAX_ENTRIES).clamp(1, MAX_ENTRIES_LIMIT);
+    Ok((prefix.unwrap_or_default(), max_entries))
+}
+
+#[async_trait::async_trait]
+impl Operation for BucketPrefixUsageHandler {
+    async fn call(&self, req: S3Request<Body>, params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
+        let Some(input_cred) = req.credentials else {
+            return Err(s3_error!(InvalidRequest, "get cred failed"));
+        };
+
+        let (cred, owner) =
+            check_key_valid(get_session_token(&req.uri, &req.headers).unwrap_or_default(), &input_cred.access_key).await?;
+
+        let remote_addr = req.extensions.get::<Option<RemoteAddr>>().and_then(|opt| opt.map(|a| a.0));
+        validate_admin_request(&req.headers, &cred, owner, false, data_usage_info_gate_actions(), remote_addr).await?;
+
+        let bucket = params.get("bucket").unwrap_or_default().to_string();
+        if bucket.is_empty() {
+            return Err(s3_error!(InvalidRequest, "bucket path parameter is required"));
+        }
+        let (prefix, max_entries) = parse_usage_prefix_query(req.uri.query())?;
+
+        // Authorization is bucket-scoped by the same any-of gate as the
+        // datausageinfo route; the bucket name itself is validated by the
+        // scanner layer, which rejects reserved/invalid names.
+        let response = rustfs_scanner::bucket_prefix_usage(&bucket, &prefix, max_entries)
+            .await
+            .map_err(|err| s3_error!(InvalidArgument, "{}", err))?;
+
+        let data = serde_json::to_vec(&response)
+            .map_err(|_| S3Error::with_message(S3ErrorCode::InternalError, "parse prefix usage failed"))?;
+        let mut header = HeaderMap::new();
+        header.insert(CONTENT_TYPE, HeaderValue::from_static(JSON_CONTENT_TYPE));
+
+        Ok(S3Response::with_headers((StatusCode::OK, Body::from(data)), header))
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{DEFAULT_MAX_ENTRIES, MAX_ENTRIES_LIMIT, parse_usage_prefix_query};
+    use s3s::S3Error;
+
+    fn query(raw: &str) -> Result<(String, usize), S3Error> {
+        parse_usage_prefix_query(Some(raw))
+    }
+
+    #[test]
+    fn defaults_apply_when_no_query_is_given() {
+        assert_eq!(parse_usage_prefix_query(None).unwrap(), (String::new(), DEFAULT_MAX_ENTRIES));
+        assert_eq!(query("").unwrap(), (String::new(), DEFAULT_MAX_ENTRIES));
+    }
+
+    #[test]
+    fn prefix_round_trips_url_encoded_characters() {
+        let (prefix, _) = query("prefix=pre%2Ffix%20name").unwrap();
+        assert_eq!(prefix, "pre/fix name");
+    }
+
+    #[test]
+    fn max_entries_parses_and_clamps_to_documented_bounds() {
+        assert_eq!(query("max-entries=5").unwrap().1, 5);
+        assert_eq!(query("max-entries=0").unwrap().1, 1, "zero must clamp up, not mean unlimited");
+        assert_eq!(query("max-entries=99999999").unwrap().1, MAX_ENTRIES_LIMIT);
+        assert!(query("max-entries=-3").is_err());
+        assert!(query("max-entries=abc").is_err());
+    }
+
+    #[test]
+    fn unknown_parameters_are_rejected_not_ignored() {
+        assert!(
+            query("prefixes=x").is_err(),
+            "a typo'd parameter must fail the request, not widen the query"
+        );
+    }
+}

rustfs/src/admin/mod.rs

@@ -40,7 +40,8 @@ use handlers::{
     audit, batch_job, bucket_meta, cluster_snapshot, config_admin, diagnostics, durability as durability_handler, extensions,
     heal, health, idp_compat, ilm_transition, inspect_archive, kms, module_switch, object_data_cache, object_zip_download, oidc,
     plugins_catalog, plugins_instances, pools, profile_admin, quota as quota_handler, rebalance,
-    replication as replication_handler, scanner, site_replication, sts, system, table_catalog, tier, tls_debug, user,
+    replication as replication_handler, scanner, site_replication, sts, system, table_catalog, tier, tls_debug, usage_prefix,
+    user,
 };
 use router::{AdminOperation, S3Router};
 use s3s::route::S3Route;
@@ -80,6 +81,7 @@ fn register_admin_routes(r: &mut S3Router<AdminOperation>) -> std::io::Result<()
     bucket_meta::register_bucket_meta_route(r)?;
     config_admin::register_config_route(r)?;
     scanner::register_scanner_route(r)?;
+    usage_prefix::register_usage_prefix_route(r)?;
     ilm_transition::register_ilm_transition_route(r)?;
     object_data_cache::register_object_data_cache_route(r)?;
     audit::register_audit_target_route(r)?;

rustfs/src/admin/route_policy.rs

@@ -1558,6 +1558,11 @@ pub const DEFERRED_ADMIN_ROUTE_POLICIES: &[DeferredAdminRoutePolicy] = &[
         "/rustfs/admin/v3/datausageinfo",
         DeferredRoutePolicyReason::MultipleActions,
     ),
+    deferred(
+        HttpMethod::Get,
+        "/rustfs/admin/v3/usage/{bucket}",
+        DeferredRoutePolicyReason::MultipleActions,
+    ),
     deferred(
         HttpMethod::Post,
         "/rustfs/admin/v3/object-zip-downloads",

rustfs/src/admin/route_registration_test.rs

@@ -172,6 +172,7 @@ fn expected_admin_route_matrix() -> Vec<RouteMatrixEntry> {
         admin_route(Method::POST, "/v4/inspect/archive"),
         admin_route(Method::GET, "/v3/storageinfo"),
         admin_route(Method::GET, "/v3/datausageinfo"),
+        admin_route_sample(Method::GET, "/v3/usage/{bucket}", "/v3/usage/test-bucket"),
         admin_route(Method::GET, "/v3/metrics"),
         admin_route(Method::GET, "/v3/object-data-cache/stats"),
         admin_route(Method::POST, "/v3/object-data-cache/flush"),

rustfs/src/storage/rpc/node_service.rs

@@ -2420,6 +2420,7 @@ mod tests {
             _bucket: &str,
             _prefix: &str,
             _continuation_token: Option<&str>,
+            _include_lifecycle_object_info: bool,
         ) -> rustfs_heal::Result<(Vec<rustfs_heal::heal::storage::HealListItem>, Option<String>, bool)> {
             Ok((Vec::new(), None, false))
         }