Merge branch 'main' into cxymds/fix-1927-durable-checkpoint

Signed-off-by: houseme <housemecn@gmail.com>

.github/workflows/nightly-gnu.yml

@@ -39,10 +39,11 @@ jobs:
     env:
       FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
     steps:
-      - name: Checkout repository
+      - name: Checkout main branch
         uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
         with:
           persist-credentials: false
+          ref: main
 
       - name: Setup Rust environment
         uses: ./.github/actions/setup
@@ -88,10 +89,11 @@ jobs:
       # either casing.
       NO_PROXY: 127.0.0.1,localhost
     steps:
-      - name: Checkout repository
+      - name: Checkout main branch
         uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
         with:
           persist-credentials: false
+          ref: main
 
       - name: Setup Rust environment
         uses: ./.github/actions/setup
@@ -176,10 +178,11 @@ jobs:
       FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
       NO_PROXY: 127.0.0.1,localhost
     steps:
-      - name: Checkout repository
+      - name: Checkout main branch
         uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
         with:
           persist-credentials: false
+          ref: main
 
       - name: Setup Rust environment
         uses: ./.github/actions/setup

Cargo.lock

@@ -9564,6 +9564,7 @@ dependencies = [
  "serde",
  "serde_json",
  "serial_test",
+ "sha2 0.11.0",
  "temp-env",
  "tempfile",
  "thiserror 2.0.20",

crates/data-usage/src/data_usage.rs

@@ -12,13 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use serde::{
-    Deserialize, Serialize,
-    de::{IgnoredAny, SeqAccess, Visitor},
-    ser::SerializeMap as _,
-};
+use serde::{Deserialize, Serialize, ser::SerializeMap as _};
 use std::{
-    borrow::Cow,
     collections::{HashMap, HashSet},
     hash::{DefaultHasher, Hash, Hasher},
     time::{Duration, SystemTime},
@@ -53,11 +48,6 @@ pub const DATA_USAGE_OBSERVED_OBJECT_NAME: &str = ".usage.observed.json";
 // RUSTFS_COMPAT_TODO(scanner-usage-v2): keep .usage.json readable and removable during rolling upgrades from pre-v2 scanners. Remove after supported direct-upgrade sources all write .usage.v2.json.
 pub const LEGACY_DATA_USAGE_OBJECT_NAME: &str = ".usage.json";
 
-/// Fixed bucket for objects whose storage class is not in the scanner's
-/// cycle-local tier registry.  Keeping this key fixed prevents untrusted or
-/// stale tier names from growing persisted per-tier maps without bound.
-pub const UNKNOWN_TIER: &str = "UNKNOWN_TIER";
-
 /// Returns true when `existing_last_update` is ahead of `now` by more than
 /// [`USAGE_LAST_UPDATE_FUTURE_TOLERANCE`], i.e. the persisted timestamp cannot be
 /// trusted for staleness comparisons and a fresh snapshot save must be allowed.
@@ -88,301 +78,12 @@ impl TierStats {
             && self.num_objects.checked_add(u.num_objects).is_some()
     }
 
-    /// Add tier counters without allowing a counter to wrap.
-    pub fn checked_add(&self, u: &TierStats) -> Option<TierStats> {
-        Some(TierStats {
-            total_size: self.total_size.checked_add(u.total_size)?,
-            num_versions: self.num_versions.checked_add(u.num_versions)?,
-            num_objects: self.num_objects.checked_add(u.num_objects)?,
-        })
-    }
-
     /// True when this tier contributed nothing, i.e. merging it is a no-op.
     pub fn is_empty(&self) -> bool {
         self.total_size == 0 && self.num_versions == 0 && self.num_objects == 0
     }
 }
 
-/// Bounded diagnostics for objects whose tier is absent from the cycle
-/// registry. Counters are authoritative; diagnostics are only a small,
-/// redacted reconciliation aid and may be dropped at the configured caps.
-pub const UNKNOWN_TIER_DIAGNOSTIC_ENTRY_CAP: usize = 64;
-pub const UNKNOWN_TIER_DIAGNOSTIC_BYTE_CAP: usize = 4096;
-pub const UNKNOWN_TIER_DIAGNOSTIC_TTL: Duration = Duration::from_secs(60 * 60);
-const UNKNOWN_TIER_DIAGNOSTIC_KEY_BYTES: usize = 256;
-const UNKNOWN_TIER_DIAGNOSTIC_INPUT_CAP: usize = UNKNOWN_TIER_DIAGNOSTIC_ENTRY_CAP * 4;
-
-#[derive(Clone, Debug, Default, Serialize, PartialEq, Eq)]
-pub struct UnknownTierStats {
-    /// Logical bytes retained in the scanner's normal usage total.
-    pub unknown_bytes: u64,
-    /// Physical bytes recorded in the per-tier accounting dimension.
-    ///
-    /// Older writers only had `unknown_bytes`; decoding those snapshots keeps
-    /// this field at zero and the scanner fills it for new observations.
-    #[serde(default)]
-    pub unknown_physical_bytes: u64,
-    pub unknown_objects: u64,
-    pub unknown_versions: u64,
-    pub diagnostics_dropped: u64,
-    #[serde(default)]
-    pub diagnostics: Vec<String>,
-    #[serde(default)]
-    pub diagnostics_at: Option<SystemTime>,
-    /// A saturating update occurred; this snapshot cannot prove conservation.
-    /// The field is persisted so a restarted scanner cannot mistake a
-    /// saturated legacy aggregate for exact accounting evidence.
-    #[serde(default)]
-    pub counter_overflowed: bool,
-}
-
-#[derive(Default)]
-struct BoundedDiagnostics {
-    entries: Vec<String>,
-    dropped: u64,
-}
-
-impl<'de> Deserialize<'de> for BoundedDiagnostics {
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'de>,
-    {
-        struct BoundedDiagnosticsVisitor;
-
-        impl<'de> Visitor<'de> for BoundedDiagnosticsVisitor {
-            type Value = BoundedDiagnostics;
-
-            fn expecting(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-                formatter.write_str("a sequence of bounded tier diagnostics")
-            }
-
-            fn visit_seq<A>(self, mut sequence: A) -> Result<Self::Value, A::Error>
-            where
-                A: SeqAccess<'de>,
-            {
-                let mut bounded = BoundedDiagnostics::default();
-                let mut bytes = 0_usize;
-                let mut inspected = 0_usize;
-                loop {
-                    if bounded.entries.len() >= UNKNOWN_TIER_DIAGNOSTIC_ENTRY_CAP
-                        || bytes >= UNKNOWN_TIER_DIAGNOSTIC_BYTE_CAP
-                        || inspected >= UNKNOWN_TIER_DIAGNOSTIC_INPUT_CAP
-                    {
-                        if sequence.next_element::<IgnoredAny>()?.is_none() {
-                            break;
-                        }
-                        bounded.dropped = bounded.dropped.saturating_add(1);
-                        continue;
-                    }
-                    let Some(diagnostic) = sequence.next_element::<Cow<'de, str>>()? else {
-                        break;
-                    };
-                    inspected = inspected.saturating_add(1);
-                    if !is_redacted_tier_diagnostic(&diagnostic)
-                        || bytes.saturating_add(diagnostic.len()) > UNKNOWN_TIER_DIAGNOSTIC_BYTE_CAP
-                        || bounded.entries.iter().any(|entry| entry == &diagnostic)
-                    {
-                        bounded.dropped = bounded.dropped.saturating_add(1);
-                        continue;
-                    }
-                    bytes = bytes.saturating_add(diagnostic.len());
-                    bounded.entries.push(diagnostic.into_owned());
-                }
-                Ok(bounded)
-            }
-        }
-
-        deserializer.deserialize_seq(BoundedDiagnosticsVisitor)
-    }
-}
-
-#[derive(Deserialize)]
-struct UnknownTierStatsWire {
-    #[serde(default)]
-    unknown_bytes: u64,
-    #[serde(default)]
-    unknown_physical_bytes: u64,
-    #[serde(default)]
-    unknown_objects: u64,
-    #[serde(default)]
-    unknown_versions: u64,
-    #[serde(default)]
-    diagnostics_dropped: u64,
-    #[serde(default)]
-    diagnostics: BoundedDiagnostics,
-    #[serde(default)]
-    diagnostics_at: Option<SystemTime>,
-    #[serde(default)]
-    counter_overflowed: bool,
-}
-
-fn is_redacted_tier_diagnostic(diagnostic: &str) -> bool {
-    diagnostic
-        .strip_prefix("tier-hash:")
-        .is_some_and(|digest| digest.len() == 16 && digest.bytes().all(|byte| byte.is_ascii_hexdigit()))
-}
-
-fn diagnostics_expired(at: SystemTime, now: SystemTime) -> bool {
-    now.duration_since(at).map_or(true, |age| age > UNKNOWN_TIER_DIAGNOSTIC_TTL)
-}
-
-fn checked_saturating_add(left: u64, right: u64, overflowed: &mut bool) -> u64 {
-    match left.checked_add(right) {
-        Some(value) => value,
-        None => {
-            *overflowed = true;
-            u64::MAX
-        }
-    }
-}
-
-impl<'de> Deserialize<'de> for UnknownTierStats {
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'de>,
-    {
-        let wire = UnknownTierStatsWire::deserialize(deserializer)?;
-        let mut stats = Self {
-            unknown_bytes: wire.unknown_bytes,
-            unknown_physical_bytes: wire.unknown_physical_bytes,
-            unknown_objects: wire.unknown_objects,
-            unknown_versions: wire.unknown_versions,
-            counter_overflowed: wire.counter_overflowed,
-            diagnostics_dropped: wire.diagnostics_dropped,
-            diagnostics: wire.diagnostics.entries,
-            diagnostics_at: wire.diagnostics_at,
-        };
-        stats.diagnostics_dropped =
-            checked_saturating_add(stats.diagnostics_dropped, wire.diagnostics.dropped, &mut stats.counter_overflowed);
-        if stats
-            .diagnostics_at
-            .is_some_and(|at| diagnostics_expired(at, SystemTime::now()))
-        {
-            stats.diagnostics.clear();
-            stats.diagnostics_at = None;
-        }
-        Ok(stats)
-    }
-}
-
-impl UnknownTierStats {
-    /// Record one observation where the logical and physical dimensions are
-    /// the same. Kept as a small compatibility helper for callers that only
-    /// have one size value.
-    pub fn record(&mut self, tier: &str, bytes: u64, versions: u64, objects: u64) {
-        self.record_dimensions(tier, bytes, bytes, versions, objects);
-    }
-
-    /// Record one observation without conflating logical usage with physical
-    /// tier bytes. Both counters are saturating so malformed metadata cannot
-    /// wrap an aggregate.
-    pub fn record_dimensions(&mut self, tier: &str, logical_bytes: u64, physical_bytes: u64, versions: u64, objects: u64) {
-        self.unknown_bytes = checked_saturating_add(self.unknown_bytes, logical_bytes, &mut self.counter_overflowed);
-        self.unknown_physical_bytes =
-            checked_saturating_add(self.unknown_physical_bytes, physical_bytes, &mut self.counter_overflowed);
-        self.unknown_objects = checked_saturating_add(self.unknown_objects, objects, &mut self.counter_overflowed);
-        self.unknown_versions = checked_saturating_add(self.unknown_versions, versions, &mut self.counter_overflowed);
-
-        let digest = {
-            let mut hasher = DefaultHasher::new();
-            // Bound hashing work for hostile metadata while retaining enough
-            // length/prefix entropy to reconcile repeated observations.
-            hasher.write_u64(u64::try_from(tier.len()).unwrap_or(u64::MAX));
-            let bounded = &tier.as_bytes()[..tier.len().min(UNKNOWN_TIER_DIAGNOSTIC_KEY_BYTES)];
-            hasher.write(bounded);
-            hasher.write_u8(u8::from(bounded.iter().any(|byte| byte.is_ascii_control())));
-            format!("tier-hash:{:016x}", hasher.finish())
-        };
-        let now = SystemTime::now();
-        if self.diagnostics_at.is_some_and(|at| diagnostics_expired(at, now)) {
-            self.diagnostics.clear();
-        }
-        self.diagnostics_at = Some(now);
-        if self.diagnostics.iter().any(|entry| entry == &digest) {
-            return;
-        }
-        let current_bytes: usize = self.diagnostics.iter().map(String::len).sum();
-        if self.diagnostics.len() >= UNKNOWN_TIER_DIAGNOSTIC_ENTRY_CAP
-            || current_bytes.saturating_add(digest.len()) > UNKNOWN_TIER_DIAGNOSTIC_BYTE_CAP
-        {
-            self.diagnostics_dropped = checked_saturating_add(1, self.diagnostics_dropped, &mut self.counter_overflowed);
-            return;
-        }
-        self.diagnostics.push(digest);
-    }
-
-    pub fn merge(&mut self, other: &Self) {
-        self.unknown_bytes = checked_saturating_add(self.unknown_bytes, other.unknown_bytes, &mut self.counter_overflowed);
-        self.unknown_physical_bytes =
-            checked_saturating_add(self.unknown_physical_bytes, other.unknown_physical_bytes, &mut self.counter_overflowed);
-        self.unknown_objects = checked_saturating_add(self.unknown_objects, other.unknown_objects, &mut self.counter_overflowed);
-        self.unknown_versions =
-            checked_saturating_add(self.unknown_versions, other.unknown_versions, &mut self.counter_overflowed);
-        self.diagnostics_dropped =
-            checked_saturating_add(self.diagnostics_dropped, other.diagnostics_dropped, &mut self.counter_overflowed);
-        self.counter_overflowed |= other.counter_overflowed;
-        let now = SystemTime::now();
-        if self.diagnostics_at.is_some_and(|at| diagnostics_expired(at, now)) {
-            self.diagnostics.clear();
-            self.diagnostics_at = None;
-        }
-        if !other.diagnostics_at.is_some_and(|at| diagnostics_expired(at, now)) {
-            for diagnostic in &other.diagnostics {
-                if !is_redacted_tier_diagnostic(diagnostic) {
-                    self.diagnostics_dropped = checked_saturating_add(1, self.diagnostics_dropped, &mut self.counter_overflowed);
-                    continue;
-                }
-                if self.diagnostics.iter().any(|entry| entry == diagnostic) {
-                    continue;
-                }
-                let current_bytes: usize = self.diagnostics.iter().map(String::len).sum();
-                if self.diagnostics.len() >= UNKNOWN_TIER_DIAGNOSTIC_ENTRY_CAP
-                    || current_bytes.saturating_add(diagnostic.len()) > UNKNOWN_TIER_DIAGNOSTIC_BYTE_CAP
-                {
-                    self.diagnostics_dropped = checked_saturating_add(1, self.diagnostics_dropped, &mut self.counter_overflowed);
-                    continue;
-                }
-                self.diagnostics.push(diagnostic.clone());
-            }
-        }
-        if !self.diagnostics.is_empty() {
-            self.diagnostics_at = Some(now);
-        }
-    }
-
-    pub fn fits_add(&self, other: &Self) -> bool {
-        !self.counter_overflowed
-            && !other.counter_overflowed
-            && self.unknown_bytes.checked_add(other.unknown_bytes).is_some()
-            && self
-                .unknown_physical_bytes
-                .checked_add(other.unknown_physical_bytes)
-                .is_some()
-            && self.unknown_objects.checked_add(other.unknown_objects).is_some()
-            && self.unknown_versions.checked_add(other.unknown_versions).is_some()
-            && self.diagnostics_dropped.checked_add(other.diagnostics_dropped).is_some()
-    }
-
-    pub fn checked_add(&self, other: &Self) -> Option<Self> {
-        if !self.fits_add(other) {
-            return None;
-        }
-        let mut merged = self.clone();
-        merged.merge(other);
-        Some(merged)
-    }
-
-    pub fn is_empty(&self) -> bool {
-        !self.counter_overflowed
-            && self.unknown_bytes == 0
-            && self.unknown_physical_bytes == 0
-            && self.unknown_objects == 0
-            && self.unknown_versions == 0
-            && self.diagnostics_dropped == 0
-            && self.diagnostics.is_empty()
-    }
-}
-
 #[derive(Clone, Debug, Default, Serialize, Deserialize, PartialEq, Eq)]
 pub struct AllTierStats {
     pub tiers: HashMap<String, TierStats>,
@@ -423,31 +124,6 @@ impl AllTierStats {
             .iter()
             .all(|(tier, right)| self.tiers.get(tier).is_none_or(|left| left.fits_add(right)))
     }
-
-    /// Fold keys from an older cache that are no longer present in the
-    /// current registry into the fixed unknown bucket. Built-in storage
-    /// classes remain known even when no remote tier is configured.
-    pub fn fold_unknown_tiers<'a, I>(&mut self, known_tiers: I)
-    where
-        I: IntoIterator<Item = &'a str>,
-    {
-        let known: HashSet<&str> = known_tiers.into_iter().collect();
-        let mut unknown = self.tiers.remove(UNKNOWN_TIER).unwrap_or_default();
-        let retired_tiers: Vec<String> = self
-            .tiers
-            .keys()
-            .filter(|tier| tier.as_str() != "STANDARD" && tier.as_str() != "REDUCED_REDUNDANCY" && !known.contains(tier.as_str()))
-            .cloned()
-            .collect();
-        for tier in retired_tiers {
-            if let Some(stats) = self.tiers.remove(&tier) {
-                unknown = unknown.add(&stats);
-            }
-        }
-        if !unknown.is_empty() {
-            self.tiers.insert(UNKNOWN_TIER.to_string(), unknown);
-        }
-    }
 }
 
 /// Bucket target usage info provides replication statistics
@@ -535,10 +211,6 @@ pub struct DataUsageInfo {
     /// tier exists, so an absent value means "not accounted", never "zero".
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub tier_stats: Option<AllTierStats>,
-    /// Bounded diagnostics and separate logical/physical counters for objects
-    /// classified into [`UNKNOWN_TIER`].
-    #[serde(default, skip_serializing_if = "Option::is_none")]
-    pub unknown_tier_stats: Option<UnknownTierStats>,
 
     /// Total number of buckets in this cluster
     pub buckets_count: u64,
@@ -635,48 +307,6 @@ pub struct DiskUsageStatus {
     pub snapshot_exists: bool,
 }
 
-/// Independent conservation evidence for a scanner summary.
-///
-/// The totals are maintained while objects are accounted and are deliberately
-/// not reconstructed from the tier map at publish time. `*_known` records the
-/// portion represented by the corresponding accounting dimension.
-#[derive(Debug, Default, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
-pub struct TierAccountingProof {
-    pub logical_total: u64,
-    pub logical_known: u64,
-    pub physical_total: u64,
-    pub physical_known: u64,
-    #[serde(default)]
-    pub overflowed: bool,
-}
-
-impl TierAccountingProof {
-    pub fn checked_add(self, other: Self) -> Option<Self> {
-        if self.overflowed || other.overflowed {
-            return None;
-        }
-        Some(Self {
-            logical_total: self.logical_total.checked_add(other.logical_total)?,
-            logical_known: self.logical_known.checked_add(other.logical_known)?,
-            physical_total: self.physical_total.checked_add(other.physical_total)?,
-            physical_known: self.physical_known.checked_add(other.physical_known)?,
-            overflowed: false,
-        })
-    }
-
-    pub fn saturating_add(&mut self, other: Self) {
-        let Some(merged) = (*self).checked_add(other) else {
-            self.logical_total = self.logical_total.saturating_add(other.logical_total);
-            self.logical_known = self.logical_known.saturating_add(other.logical_known);
-            self.physical_total = self.physical_total.saturating_add(other.physical_total);
-            self.physical_known = self.physical_known.saturating_add(other.physical_known);
-            self.overflowed = true;
-            return;
-        };
-        *self = merged;
-    }
-}
-
 /// Size summary for a single object or group of objects
 #[derive(Debug, Default, Clone)]
 pub struct SizeSummary {
@@ -706,10 +336,6 @@ pub struct SizeSummary {
     pub repl_target_stats: HashMap<String, ReplTargetSizeSummary>,
     /// Per-tier accounting, keyed by storage class or remote tier name
     pub tier_stats: HashMap<String, TierStats>,
-    /// Counters and bounded diagnostics for unknown tiers in this summary.
-    pub unknown_tier_stats: UnknownTierStats,
-    /// Independent logical/physical conservation evidence.
-    pub tier_accounting_proof: TierAccountingProof,
 }
 
 /// Replication target size summary
@@ -801,10 +427,6 @@ impl<'de> Deserialize<'de> for SizeHistogram {
 }
 
 impl SizeHistogram {
-    pub fn is_empty(&self) -> bool {
-        self.0.iter().all(|value| *value == 0)
-    }
-
     pub fn add(&mut self, size: u64) {
         let intervals = [
             (0, 1024 - 1),                              // LESS_THAN_1024_B
@@ -919,10 +541,6 @@ impl<'de> Deserialize<'de> for VersionsHistogram {
 }
 
 impl VersionsHistogram {
-    pub fn is_empty(&self) -> bool {
-        self.0.iter().all(|value| *value == 0)
-    }
-
     pub fn add(&mut self, count: u64) {
         let intervals = [
             (0, 0),            // UNVERSIONED
@@ -1063,13 +681,6 @@ pub struct DataUsageEntry {
     /// observed tier-classified objects.
     #[serde(default)]
     pub all_tier_stats: Option<AllTierStats>,
-    /// Bounded unknown-tier reconciliation state for this cache entry.
-    #[serde(default)]
-    pub unknown_tier_stats: Option<UnknownTierStats>,
-    /// Optional conservation proof. Missing values are legacy/unproven, not
-    /// zero-valued evidence.
-    #[serde(default)]
-    pub tier_accounting_proof: Option<TierAccountingProof>,
 }
 
 impl Serialize for DataUsageEntry {
@@ -1080,9 +691,7 @@ impl Serialize for DataUsageEntry {
         // Keep entries map-encoded so older readers can ignore fields appended
         // by newer scanner versions during rolling upgrades. The derived
         // (array) encoding made any appended field a decode error for them.
-        let mut state = serializer.serialize_map(Some(
-            11 + usize::from(self.unknown_tier_stats.is_some()) + usize::from(self.tier_accounting_proof.is_some()),
-        ))?;
+        let mut state = serializer.serialize_map(Some(11))?;
         state.serialize_entry("children", &self.children)?;
         state.serialize_entry("size", &self.size)?;
         state.serialize_entry("objects", &self.objects)?;
@@ -1094,35 +703,11 @@ impl Serialize for DataUsageEntry {
         state.serialize_entry("compacted", &self.compacted)?;
         state.serialize_entry("failed_objects", &self.failed_objects)?;
         state.serialize_entry("all_tier_stats", &self.all_tier_stats)?;
-        // Keep the legacy no-unknown shape byte-for-byte stable. Once unknown
-        // accounting exists, append its map field before the optional proof.
-        if let Some(unknown_tier_stats) = self.unknown_tier_stats.as_ref() {
-            state.serialize_entry("unknown_tier_stats", unknown_tier_stats)?;
-        }
-        if let Some(proof) = self.tier_accounting_proof.as_ref() {
-            state.serialize_entry("tier_accounting_proof", proof)?;
-        }
         state.end()
     }
 }
 
 impl DataUsageEntry {
-    fn has_local_usage(&self) -> bool {
-        self.size != 0
-            || self.objects != 0
-            || self.versions != 0
-            || self.delete_markers != 0
-            || self.failed_objects != 0
-            || self.obj_sizes.0.iter().any(|value| *value != 0)
-            || self.obj_versions.0.iter().any(|value| *value != 0)
-            || self.replication_stats.as_ref().is_some_and(|stats| !stats.is_empty())
-            || self
-                .all_tier_stats
-                .as_ref()
-                .is_some_and(|stats| stats.tiers.values().any(|tier| !tier.is_empty()))
-            || self.unknown_tier_stats.as_ref().is_some_and(|stats| !stats.is_empty())
-    }
-
     pub fn add_child(&mut self, hash: &DataUsageHash) {
         if self.children.contains(&hash.key()) {
             return;
@@ -1131,8 +716,6 @@ impl DataUsageEntry {
     }
 
     pub fn merge(&mut self, other: &DataUsageEntry) {
-        let self_had_local_usage = self.has_local_usage();
-        let other_had_local_usage = other.has_local_usage();
         self.objects += other.objects;
         self.versions += other.versions;
         self.delete_markers += other.delete_markers;
@@ -1161,29 +744,6 @@ impl DataUsageEntry {
         if let Some(o_tiers) = other.all_tier_stats.as_ref().filter(|tiers| !tiers.is_empty()) {
             self.all_tier_stats.get_or_insert_with(AllTierStats::new).merge(o_tiers);
         }
-        if let Some(other_unknown) = other.unknown_tier_stats.as_ref() {
-            self.unknown_tier_stats
-                .get_or_insert_with(UnknownTierStats::default)
-                .merge(other_unknown);
-        }
-
-        self.tier_accounting_proof = match (self.tier_accounting_proof, other.tier_accounting_proof) {
-            (Some(mut left), Some(right)) => {
-                left.saturating_add(right);
-                Some(left)
-            }
-            (None, Some(right)) if !self_had_local_usage => Some(right),
-            (Some(left), None) if !other_had_local_usage => Some(left),
-            (None, None) => None,
-            _ => None,
-        };
-        // A saturated unknown-tier counter invalidates conservation evidence;
-        // never let an otherwise valid proof hide that loss of precision.
-        if self.unknown_tier_stats.as_ref().is_some_and(|stats| stats.counter_overflowed)
-            && let Some(proof) = self.tier_accounting_proof.as_mut()
-        {
-            proof.overflowed = true;
-        }
 
         self.obj_sizes.merge_from(&other.obj_sizes);
         self.obj_versions.merge_from(&other.obj_versions);
@@ -1197,15 +757,6 @@ impl DataUsageEntry {
         self.all_tier_stats.get_or_insert_with(AllTierStats::new).add_sizes(tiers);
     }
 
-    pub fn add_unknown_tier_stats(&mut self, stats: &UnknownTierStats) {
-        if stats.is_empty() {
-            return;
-        }
-        self.unknown_tier_stats
-            .get_or_insert_with(UnknownTierStats::default)
-            .merge(stats);
-    }
-
     pub fn checked_merge(&mut self, other: &DataUsageEntry) -> bool {
         let scalar_counts_fit = self.objects.checked_add(other.objects).is_some()
             && self.versions.checked_add(other.versions).is_some()
@@ -1269,21 +820,8 @@ impl DataUsageEntry {
             (_, None) | (None, Some(_)) => true,
             (Some(left), Some(right)) => left.fits_merge(right),
         };
-        let unknown_tier_stats_fit = match (&self.unknown_tier_stats, &other.unknown_tier_stats) {
-            (None, None) => true,
-            (Some(left), None) => !left.counter_overflowed,
-            (None, Some(right)) => !right.counter_overflowed,
-            (Some(left), Some(right)) => left.fits_add(right),
-        };
 
-        let proof_fit = match (self.tier_accounting_proof, other.tier_accounting_proof) {
-            (Some(left), Some(right)) => left.checked_add(right).is_some(),
-            (Some(proof), None) | (None, Some(proof)) => !proof.overflowed,
-            (None, None) => true,
-        };
-
-        if !scalar_counts_fit || !histograms_fit || !replication_fits || !tier_stats_fit || !unknown_tier_stats_fit || !proof_fit
-        {
+        if !scalar_counts_fit || !histograms_fit || !replication_fits || !tier_stats_fit {
             return false;
         }
         self.merge(other);
@@ -1801,7 +1339,6 @@ impl DataUsageCache {
             delete_markers_total_count: flat.delete_markers as u64,
             objects_total_size: flat.size as u64,
             tier_stats: flat.all_tier_stats.filter(|tiers| !tiers.is_empty()),
-            unknown_tier_stats: flat.unknown_tier_stats.filter(|stats| !stats.is_empty()),
             buckets_count: u64::try_from(buckets.len()).unwrap_or(u64::MAX),
             buckets_usage,
             usage_snapshot_complete: self.info.snapshot_complete,
@@ -2229,20 +1766,6 @@ impl SizeSummary {
         self.replica_count = self.replica_count.saturating_add(other.replica_count);
         self.pending_count = self.pending_count.saturating_add(other.pending_count);
         self.failed_count = self.failed_count.saturating_add(other.failed_count);
-        self.unknown_tier_stats.merge(&other.unknown_tier_stats);
-        self.tier_accounting_proof.saturating_add(other.tier_accounting_proof);
-        if self.unknown_tier_stats.counter_overflowed {
-            self.tier_accounting_proof.overflowed = true;
-        }
-
-        // A disk/bucket aggregate is assembled from many object summaries.
-        // Keep the per-tier dimension in lockstep with the scalar counters;
-        // dropping this map here would recreate the original silent-loss bug
-        // at the cross-disk merge boundary.
-        for (tier, stats) in &other.tier_stats {
-            let entry = self.tier_stats.entry(tier.clone()).or_default();
-            *entry = entry.add(stats);
-        }
 
         // Merge replication target stats
         for (target, stats) in &other.repl_target_stats {
@@ -2355,59 +1878,6 @@ mod tests {
         );
     }
 
-    #[test]
-    fn retired_tier_stats_fold_into_fixed_unknown_bucket() {
-        let mut stats = AllTierStats::default();
-        stats.tiers.insert(
-            "RETIRED".to_string(),
-            TierStats {
-                total_size: 9,
-                num_versions: 2,
-                num_objects: 1,
-            },
-        );
-        stats.tiers.insert(
-            "WARM".to_string(),
-            TierStats {
-                total_size: 4,
-                num_versions: 1,
-                num_objects: 1,
-            },
-        );
-
-        stats.fold_unknown_tiers(["WARM"]);
-
-        assert!(!stats.tiers.contains_key("RETIRED"));
-        assert_eq!(stats.tiers.get("WARM").map(|v| v.total_size), Some(4));
-        assert_eq!(stats.tiers.get(UNKNOWN_TIER).map(|v| v.total_size), Some(9));
-    }
-
-    #[test]
-    fn unknown_tier_stats_deserialization_keeps_diagnostics_bounded() {
-        #[derive(Serialize)]
-        struct RawUnknownTierStats {
-            unknown_bytes: u64,
-            diagnostics: Vec<String>,
-            diagnostics_dropped: u64,
-        }
-
-        let mut diagnostics = vec!["tier-hash:0123456789abcdef".to_string(); UNKNOWN_TIER_DIAGNOSTIC_ENTRY_CAP + 8];
-        diagnostics.push("raw-tier-name-that-must-not-be-exposed".to_string());
-        diagnostics.push("x".repeat(UNKNOWN_TIER_DIAGNOSTIC_BYTE_CAP + 1));
-        let encoded = rmp_serde::to_vec_named(&RawUnknownTierStats {
-            unknown_bytes: 7,
-            diagnostics,
-            diagnostics_dropped: 3,
-        })
-        .expect("unknown tier stats should encode");
-        let decoded: UnknownTierStats = rmp_serde::from_slice(&encoded).expect("unknown tier stats should decode");
-
-        assert_eq!(decoded.unknown_bytes, 7);
-        assert_eq!(decoded.diagnostics.len(), 1);
-        assert!(decoded.diagnostics_dropped >= 3);
-        assert!(decoded.diagnostics.iter().all(|entry| is_redacted_tier_diagnostic(entry)));
-    }
-
     #[test]
     fn checked_merge_rejects_overflowing_tier_totals() {
         let mut left = tier_entry(
@@ -2431,57 +1901,6 @@ mod tests {
         assert_eq!(left.all_tier_stats.expect("left is untouched").tiers["WARM"].total_size, u64::MAX);
     }
 
-    #[test]
-    fn checked_merge_rejects_unknown_counter_overflow_on_either_side() {
-        let overflowed = DataUsageEntry {
-            unknown_tier_stats: Some(UnknownTierStats {
-                counter_overflowed: true,
-                ..Default::default()
-            }),
-            tier_accounting_proof: Some(TierAccountingProof {
-                logical_total: 1,
-                logical_known: 1,
-                ..Default::default()
-            }),
-            ..Default::default()
-        };
-        let mut left = DataUsageEntry::default();
-        assert!(!left.checked_merge(&overflowed));
-
-        let mut left = overflowed.clone();
-        assert!(!left.checked_merge(&DataUsageEntry::default()));
-    }
-
-    #[test]
-    fn checked_merge_rejects_one_sided_overflowed_proof_without_mutation() {
-        let mut left = DataUsageEntry {
-            size: 1,
-            tier_accounting_proof: Some(TierAccountingProof {
-                logical_total: 1,
-                logical_known: 1,
-                overflowed: true,
-                ..Default::default()
-            }),
-            ..Default::default()
-        };
-        let legacy = DataUsageEntry {
-            size: 2,
-            ..Default::default()
-        };
-
-        assert!(!left.checked_merge(&legacy));
-        assert_eq!(left.size, 1);
-        assert_eq!(
-            left.tier_accounting_proof,
-            Some(TierAccountingProof {
-                logical_total: 1,
-                logical_known: 1,
-                overflowed: true,
-                ..Default::default()
-            })
-        );
-    }
-
     /// Entry shape released before per-tier accounting, using the derived
     /// (array) encoding those writers produced.
     #[derive(Serialize, Deserialize)]
@@ -2521,72 +1940,6 @@ mod tests {
         assert_eq!(legacy.objects, 0);
     }
 
-    #[test]
-    fn tier_accounting_proof_is_optional_and_map_encoded() {
-        let entry = DataUsageEntry {
-            tier_accounting_proof: Some(TierAccountingProof {
-                logical_total: 11,
-                logical_known: 11,
-                physical_total: 7,
-                physical_known: 7,
-                ..Default::default()
-            }),
-            ..Default::default()
-        };
-        let encoded = rmp_serde::to_vec(&entry).expect("proof-bearing entry should encode");
-        let decoded: DataUsageEntry = rmp_serde::from_slice(&encoded).expect("proof-bearing entry should decode");
-        assert_eq!(decoded.tier_accounting_proof, entry.tier_accounting_proof);
-
-        let legacy = LegacyEntry {
-            children: DataUsageHashMap::default(),
-            size: 12,
-            objects: 3,
-            versions: 4,
-            delete_markers: 1,
-            obj_sizes: SizeHistogram::default(),
-            obj_versions: VersionsHistogram::default(),
-            replication_stats: None,
-            compacted: false,
-            failed_objects: 2,
-        };
-        let legacy_bytes = rmp_serde::to_vec(&legacy).expect("legacy entry should encode");
-        let decoded: DataUsageEntry = rmp_serde::from_slice(&legacy_bytes).expect("legacy entry should decode");
-        assert!(decoded.tier_accounting_proof.is_none());
-    }
-
-    #[test]
-    fn empty_entry_merge_preserves_accounting_proof() {
-        let mut entry = DataUsageEntry {
-            tier_accounting_proof: Some(TierAccountingProof {
-                logical_total: 7,
-                logical_known: 7,
-                physical_total: 7,
-                physical_known: 7,
-                ..Default::default()
-            }),
-            ..Default::default()
-        };
-        entry.merge(&DataUsageEntry::default());
-        assert!(entry.tier_accounting_proof.is_some());
-
-        let mut empty = DataUsageEntry::default();
-        empty.merge(&entry);
-        assert_eq!(empty.tier_accounting_proof, entry.tier_accounting_proof);
-    }
-
-    #[test]
-    fn unknown_counter_overflow_is_not_empty_and_survives_roundtrip() {
-        let stats = UnknownTierStats {
-            counter_overflowed: true,
-            ..Default::default()
-        };
-        assert!(!stats.is_empty());
-        let encoded = rmp_serde::to_vec_named(&stats).expect("overflow marker should encode");
-        let decoded: UnknownTierStats = rmp_serde::from_slice(&encoded).expect("overflow marker should decode");
-        assert!(decoded.counter_overflowed);
-        assert!(!decoded.is_empty());
-    }
-
     #[test]
     fn legacy_array_encoded_entries_still_load() {
         let legacy = LegacyEntry {
@@ -3355,32 +2708,6 @@ mod tests {
         );
     }
 
-    #[test]
-    fn test_dui_filters_empty_unknown_tier_usage_from_the_flattened_tree() {
-        let root_hash = hash_path("root");
-        let bucket_hash = hash_path("bucket-a");
-        let mut cache = DataUsageCache {
-            info: DataUsageCacheInfo {
-                name: "root".to_string(),
-                ..Default::default()
-            },
-            ..Default::default()
-        };
-        cache.replace_hashed(&root_hash, &None, &DataUsageEntry::default());
-
-        let child = DataUsageEntry {
-            objects: 1,
-            unknown_tier_stats: Some(UnknownTierStats::default()),
-            ..Default::default()
-        };
-        cache.replace_hashed(&bucket_hash, &Some(root_hash), &child);
-
-        let info = cache.dui("root", &["bucket-a".to_string()]);
-
-        assert_eq!(info.objects_total_count, 1);
-        assert!(info.unknown_tier_stats.is_none());
-    }
-
     #[test]
     fn test_data_usage_entry_merge_preserves_replication_targets() {
         let mut base = DataUsageEntry {

crates/ecstore/src/api/mod.rs

@@ -456,13 +456,11 @@ pub mod rpc {
         ScannerBucketListing, ScannerPeerActivity, TONIC_RPC_PREFIX, TonicInterceptor, build_put_file_auth_trailer,
         check_and_record_signed_rpc_nonce, gen_signature_headers, gen_tonic_replay_scope_headers, gen_tonic_signature_headers,
         gen_tonic_signature_interceptor, node_service_time_out_client, node_service_time_out_client_no_auth,
-        normalize_tonic_rpc_audience, set_tonic_canonical_body_digest, sign_ns_scanner_capability,
-        sign_ns_scanner_capability_with_tier_registry_generation, sign_put_file_capability, sign_tonic_rpc_response_proof,
-        tonic_boot_epoch_challenge, tonic_boot_epoch_response_headers, tonic_rpc_auth_failure_reason,
-        verify_ns_scanner_capability, verify_ns_scanner_capability_with_tier_registry_generation, verify_put_file_auth_trailer,
-        verify_put_file_capability, verify_rpc_signature, verify_tonic_boot_epoch_response, verify_tonic_canonical_body_digest,
-        verify_tonic_mutation_body_digest, verify_tonic_rpc_response_proof, verify_tonic_rpc_signature,
-        verify_tonic_rpc_signature_with_bootstrap,
+        normalize_tonic_rpc_audience, set_tonic_canonical_body_digest, sign_ns_scanner_capability, sign_put_file_capability,
+        sign_tonic_rpc_response_proof, tonic_boot_epoch_challenge, tonic_boot_epoch_response_headers,
+        tonic_rpc_auth_failure_reason, verify_put_file_auth_trailer, verify_put_file_capability, verify_rpc_signature,
+        verify_tonic_boot_epoch_response, verify_tonic_canonical_body_digest, verify_tonic_mutation_body_digest,
+        verify_tonic_rpc_response_proof, verify_tonic_rpc_signature, verify_tonic_rpc_signature_with_bootstrap,
     };
 }
 

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

@@ -94,7 +94,6 @@ const REPLAY_CACHE_AUTO_MEMORY_PERCENT: u64 = 13;
 const REPLAY_CACHE_AUTO_RPC_RPS_PER_CPU: usize = 4096;
 const REPLAY_CACHE_AUTO_MAX_CAPACITY: usize = 33_554_432;
 const NS_SCANNER_CAPABILITY_AUTH_DOMAIN: &[u8] = b"rustfs-ns-scanner-capability-v3";
-const NS_SCANNER_TIER_REGISTRY_GENERATION_AUTH_DOMAIN: &[u8] = b"rustfs-ns-scanner-tier-registry-generation-v1";
 pub const TONIC_RPC_PREFIX: &str = "/node_service.NodeService";
 static INTERNODE_RPC_SIGNATURE_STRICT: LazyLock<bool> = LazyLock::new(|| {
     get_env_bool(
@@ -637,79 +636,40 @@ pub fn verify_put_file_capability(challenge: Uuid, server_epoch: Uuid, version:
         .map_err(|_| std::io::Error::new(std::io::ErrorKind::PermissionDenied, "Invalid put_file capability proof"))
 }
 
-fn update_ns_scanner_capability_mac(
-    mac: &mut HmacSha256,
-    challenge: Uuid,
-    server_epoch: Uuid,
-    supports_tier_registry_generation: bool,
-) {
+fn update_ns_scanner_capability_mac(mac: &mut HmacSha256, challenge: Uuid, server_epoch: Uuid) {
     mac.update(NS_SCANNER_CAPABILITY_AUTH_DOMAIN);
     mac.update(&NS_SCANNER_PROTOCOL_VERSION.to_be_bytes());
     mac.update(challenge.as_bytes());
     mac.update(server_epoch.as_bytes());
-    if supports_tier_registry_generation {
-        // The optional response capability is part of the authenticated
-        // scope. A proxy cannot turn an old/unsupported peer into a worker
-        // that receives generation-fenced scanner work.
-        mac.update(NS_SCANNER_TIER_REGISTRY_GENERATION_AUTH_DOMAIN);
-    }
 }
 
-fn generate_ns_scanner_capability_proof(
-    secret: &str,
-    challenge: Uuid,
-    server_epoch: Uuid,
-    supports_tier_registry_generation: bool,
-) -> std::io::Result<Vec<u8>> {
+fn generate_ns_scanner_capability_proof(secret: &str, challenge: Uuid, server_epoch: Uuid) -> std::io::Result<Vec<u8>> {
     if challenge.is_nil() || server_epoch.is_nil() {
         return Err(std::io::Error::other("Invalid namespace scanner capability scope"));
     }
     let mut mac =
         <HmacSha256 as KeyInit>::new_from_slice(secret.as_bytes()).map_err(|_| std::io::Error::other("Invalid RPC HMAC key"))?;
-    update_ns_scanner_capability_mac(&mut mac, challenge, server_epoch, supports_tier_registry_generation);
+    update_ns_scanner_capability_mac(&mut mac, challenge, server_epoch);
     Ok(mac.finalize().into_bytes().to_vec())
 }
 
-fn verify_ns_scanner_capability_proof(
-    secret: &str,
-    challenge: Uuid,
-    server_epoch: Uuid,
-    proof: &[u8],
-    supports_tier_registry_generation: bool,
-) -> std::io::Result<()> {
+fn verify_ns_scanner_capability_proof(secret: &str, challenge: Uuid, server_epoch: Uuid, proof: &[u8]) -> std::io::Result<()> {
     if challenge.is_nil() || server_epoch.is_nil() {
         return Err(std::io::Error::other("Invalid namespace scanner capability scope"));
     }
     let mut mac =
         <HmacSha256 as KeyInit>::new_from_slice(secret.as_bytes()).map_err(|_| std::io::Error::other("Invalid RPC HMAC key"))?;
-    update_ns_scanner_capability_mac(&mut mac, challenge, server_epoch, supports_tier_registry_generation);
+    update_ns_scanner_capability_mac(&mut mac, challenge, server_epoch);
     mac.verify_slice(proof)
         .map_err(|_| std::io::Error::new(std::io::ErrorKind::PermissionDenied, "Invalid namespace scanner capability proof"))
 }
 
 pub fn sign_ns_scanner_capability(challenge: Uuid, server_epoch: Uuid) -> std::io::Result<Vec<u8>> {
-    sign_ns_scanner_capability_with_tier_registry_generation(challenge, server_epoch, false)
+    generate_ns_scanner_capability_proof(&get_shared_secret()?, challenge, server_epoch)
 }
 
 pub fn verify_ns_scanner_capability(challenge: Uuid, server_epoch: Uuid, proof: &[u8]) -> std::io::Result<()> {
-    verify_ns_scanner_capability_with_tier_registry_generation(challenge, server_epoch, proof, false)
-}
-
-pub fn sign_ns_scanner_capability_with_tier_registry_generation(
-    challenge: Uuid,
-    server_epoch: Uuid,
-    supports_tier_registry_generation: bool,
-) -> std::io::Result<Vec<u8>> {
-    generate_ns_scanner_capability_proof(&get_shared_secret()?, challenge, server_epoch, supports_tier_registry_generation)
-}
-
-pub fn verify_ns_scanner_capability_with_tier_registry_generation(
-    challenge: Uuid,
-    server_epoch: Uuid,
-    proof: &[u8],
-    supports_tier_registry_generation: bool,
-) -> std::io::Result<()> {
-    verify_ns_scanner_capability_proof(&get_shared_secret()?, challenge, server_epoch, proof, supports_tier_registry_generation)
+    verify_ns_scanner_capability_proof(&get_shared_secret()?, challenge, server_epoch, proof)
 }
 
 #[derive(Clone, Copy)]
@@ -1749,28 +1709,13 @@ mod tests {
         let secret = "test-scanner-capability-secret";
         let challenge = Uuid::new_v4();
         let server_epoch = Uuid::new_v4();
-        let proof = generate_ns_scanner_capability_proof(secret, challenge, server_epoch, false)
-            .expect("capability proof should be generated");
+        let proof =
+            generate_ns_scanner_capability_proof(secret, challenge, server_epoch).expect("capability proof should be generated");
 
-        assert!(verify_ns_scanner_capability_proof(secret, challenge, server_epoch, &proof, false).is_ok());
-        assert!(verify_ns_scanner_capability_proof(secret, Uuid::new_v4(), server_epoch, &proof, false).is_err());
-        assert!(verify_ns_scanner_capability_proof(secret, challenge, Uuid::new_v4(), &proof, false).is_err());
-        assert!(verify_ns_scanner_capability_proof("different-secret", challenge, server_epoch, &proof, false).is_err());
-    }
-
-    #[test]
-    fn namespace_scanner_capability_proof_binds_tier_registry_generation_support() {
-        let secret = "test-scanner-capability-secret";
-        let challenge = Uuid::new_v4();
-        let server_epoch = Uuid::new_v4();
-        let proof = generate_ns_scanner_capability_proof(secret, challenge, server_epoch, true)
-            .expect("generation capability proof should be generated");
-
-        assert!(verify_ns_scanner_capability_proof(secret, challenge, server_epoch, &proof, true).is_ok());
-        assert!(verify_ns_scanner_capability_proof(secret, challenge, server_epoch, &proof, false).is_err());
-        let legacy = generate_ns_scanner_capability_proof(secret, challenge, server_epoch, false)
-            .expect("legacy capability proof should be generated");
-        assert!(verify_ns_scanner_capability_proof(secret, challenge, server_epoch, &legacy, true).is_err());
+        assert!(verify_ns_scanner_capability_proof(secret, challenge, server_epoch, &proof).is_ok());
+        assert!(verify_ns_scanner_capability_proof(secret, Uuid::new_v4(), server_epoch, &proof).is_err());
+        assert!(verify_ns_scanner_capability_proof(secret, challenge, Uuid::new_v4(), &proof).is_err());
+        assert!(verify_ns_scanner_capability_proof("different-secret", challenge, server_epoch, &proof).is_err());
     }
 
     /// Security regression for GHSA-r5qv-rc46-hv8q (internode RPC fail-closed,

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

@@ -13,18 +13,17 @@
 // limitations under the License.
 
 use crate::cluster::rpc::{
-    build_auth_headers, build_put_file_auth_trailer, verify_ns_scanner_capability_with_tier_registry_generation,
-    verify_put_file_capability,
+    build_auth_headers, build_put_file_auth_trailer, verify_ns_scanner_capability, verify_put_file_capability,
 };
 use crate::disk::error::{Error, Result};
 use crate::disk::{FileReader, FileWriter};
 use crate::storage_api_contracts::internode::{
     NS_SCANNER_BODY_SHA256_QUERY, NS_SCANNER_CAPABILITY_CHALLENGE_QUERY, NS_SCANNER_CYCLE_QUERY, NS_SCANNER_LEADER_EPOCH_QUERY,
     NS_SCANNER_PROTOCOL_VERSION, NS_SCANNER_PROTOCOL_VERSION_QUERY, NS_SCANNER_REQUEST_ID_QUERY, NS_SCANNER_SERVER_EPOCH_QUERY,
-    NS_SCANNER_SESSION_ID_QUERY, NS_SCANNER_SESSION_SEQUENCE_QUERY, NS_SCANNER_TIER_REGISTRY_GENERATION_QUERY,
-    NsScannerCapabilityResponse, PUT_FILE_AUTH_QUERY, PUT_FILE_AUTH_V1, PUT_FILE_CAPABILITY_CHALLENGE_QUERY,
-    PUT_FILE_CAPABILITY_QUERY, PUT_FILE_CAPABILITY_VERSION, PUT_FILE_NONCE_QUERY, PUT_FILE_SERVER_EPOCH_QUERY,
-    PutFileCapabilityResponse, WALK_DIR_BODY_SHA256_QUERY, WALK_DIR_STREAM_COMPLETION_QUERY, WALK_DIR_STREAM_COMPLETION_V1,
+    NS_SCANNER_SESSION_ID_QUERY, NS_SCANNER_SESSION_SEQUENCE_QUERY, NsScannerCapabilityResponse, PUT_FILE_AUTH_QUERY,
+    PUT_FILE_AUTH_V1, PUT_FILE_CAPABILITY_CHALLENGE_QUERY, PUT_FILE_CAPABILITY_QUERY, PUT_FILE_CAPABILITY_VERSION,
+    PUT_FILE_NONCE_QUERY, PUT_FILE_SERVER_EPOCH_QUERY, PutFileCapabilityResponse, WALK_DIR_BODY_SHA256_QUERY,
+    WALK_DIR_STREAM_COMPLETION_QUERY, WALK_DIR_STREAM_COMPLETION_V1,
 };
 use async_trait::async_trait;
 use http::{HeaderMap, HeaderValue, Method, header::CONTENT_TYPE};
@@ -138,12 +137,6 @@ fn put_file_capability_status_is_legacy(status: u16) -> bool {
     status == 404
 }
 
-fn ns_scanner_capability_error_allows_legacy(error: &Error) -> bool {
-    [400, 404, 405, 426]
-        .into_iter()
-        .any(|status| error.is_internode_http_status(status))
-}
-
 #[derive(Debug, Clone, Copy, Eq, PartialEq)]
 #[allow(
     dead_code,
@@ -227,7 +220,6 @@ pub struct NsScannerStreamRequest {
 #[derive(Debug, Clone)]
 pub struct NsScannerCapabilityRequest {
     pub endpoint: String,
-    pub supports_tier_registry_generation: bool,
 }
 
 /// Data-plane stream opener used by `RemoteDisk`.
@@ -260,15 +252,6 @@ pub trait InternodeDataTransport: Send + Sync + std::fmt::Debug {
     async fn probe_ns_scanner(&self, _request: NsScannerCapabilityRequest) -> Result<Uuid> {
         Err(Error::MethodNotAllowed)
     }
-    async fn probe_ns_scanner_capability(&self, request: NsScannerCapabilityRequest) -> Result<NsScannerCapabilityResponse> {
-        let server_epoch = self.probe_ns_scanner(request).await?;
-        Ok(NsScannerCapabilityResponse {
-            version: NS_SCANNER_PROTOCOL_VERSION,
-            server_epoch,
-            proof: Vec::new(),
-            supports_tier_registry_generation: None,
-        })
-    }
     // Interface facet nobody calls yet: every transport implements both, but no
     // caller negotiates on them. Kept for the internode transport split
     // (backlog#1350); deleting them would delete the seam and six impls.
@@ -352,44 +335,27 @@ impl InternodeDataTransport for TcpHttpInternodeDataTransport {
     }
 
     async fn probe_ns_scanner(&self, request: NsScannerCapabilityRequest) -> Result<Uuid> {
-        Ok(self.probe_ns_scanner_capability(request).await?.server_epoch)
-    }
-
-    async fn probe_ns_scanner_capability(&self, request: NsScannerCapabilityRequest) -> Result<NsScannerCapabilityResponse> {
-        if request.supports_tier_registry_generation {
-            return match self.probe_ns_scanner_capability_once(&request).await {
-                Ok(response) => Ok(response),
-                Err(marked_error) if ns_scanner_capability_error_allows_legacy(&marked_error) => {
-                    // A v3 peer may reject the additive query marker, ignore
-                    // it, or return its legacy proof. Retry once without the
-                    // marker and only downgrade after that legacy response is
-                    // authenticated; an unverified epoch is never trusted.
-                    let legacy_request = NsScannerCapabilityRequest {
-                        endpoint: request.endpoint.clone(),
-                        supports_tier_registry_generation: false,
-                    };
-                    match self.probe_ns_scanner_capability_once(&legacy_request).await {
-                        Ok(mut response) => {
-                            response.supports_tier_registry_generation = None;
-                            Ok(response)
-                        }
-                        Err(legacy_error) if ns_scanner_capability_error_allows_legacy(&legacy_error) => {
-                            // Some old deployments expose only the legacy
-                            // protocol response (or advertise 426). Treat
-                            // the pair as an explicit unsupported result so
-                            // the scanner can use its coordinator fallback.
-                            Err(Error::MethodNotAllowed)
-                        }
-                        Err(_) => Err(marked_error),
-                    }
-                }
-                // A server failure, network failure, or authentication error
-                // is not evidence of an old parser. Do not issue an
-                // unauthenticated legacy probe or silently downgrade.
-                Err(marked_error) => Err(marked_error),
-            };
+        let challenge = Uuid::new_v4();
+        let url = build_ns_scanner_capability_url(&request, challenge);
+        let mut headers = msgpack_headers();
+        build_auth_headers(&url, &Method::GET, &mut headers)?;
+        let reader = HttpReader::new(url, Method::GET, headers, None).await?;
+        let mut body = Vec::new();
+        reader
+            .take(u64::try_from(NS_SCANNER_MAX_CAPABILITY_RESPONSE_SIZE + 1).unwrap_or(u64::MAX))
+            .read_to_end(&mut body)
+            .await?;
+        if body.is_empty() || body.len() > NS_SCANNER_MAX_CAPABILITY_RESPONSE_SIZE {
+            return Err(Error::other("invalid remote namespace scanner capability response size"));
         }
-        self.probe_ns_scanner_capability_once(&request).await
+        let response: NsScannerCapabilityResponse =
+            rmp_serde::from_slice(&body).map_err(|_| Error::other("invalid remote namespace scanner capability response"))?;
+        if response.version != NS_SCANNER_PROTOCOL_VERSION || response.server_epoch.is_nil() {
+            return Err(Error::other("incompatible remote namespace scanner capability response"));
+        }
+        verify_ns_scanner_capability(challenge, response.server_epoch, &response.proof)
+            .map_err(|err| Error::other(format!("remote namespace scanner capability authentication failed: {err}")))?;
+        Ok(response.server_epoch)
     }
 
     fn name(&self) -> &'static str {
@@ -402,53 +368,6 @@ impl InternodeDataTransport for TcpHttpInternodeDataTransport {
 }
 
 impl TcpHttpInternodeDataTransport {
-    async fn probe_ns_scanner_capability_once(
-        &self,
-        request: &NsScannerCapabilityRequest,
-    ) -> Result<NsScannerCapabilityResponse> {
-        let challenge = Uuid::new_v4();
-        let url = build_ns_scanner_capability_url(request, challenge);
-        let mut headers = msgpack_headers();
-        build_auth_headers(&url, &Method::GET, &mut headers)?;
-        let reader = HttpReader::new(url, Method::GET, headers, None).await?;
-        let mut body = Vec::new();
-        reader
-            .take(u64::try_from(NS_SCANNER_MAX_CAPABILITY_RESPONSE_SIZE + 1).unwrap_or(u64::MAX))
-            .read_to_end(&mut body)
-            .await?;
-        if body.is_empty() || body.len() > NS_SCANNER_MAX_CAPABILITY_RESPONSE_SIZE {
-            return Err(Error::other("invalid remote namespace scanner capability response size"));
-        }
-        let mut response: NsScannerCapabilityResponse =
-            rmp_serde::from_slice(&body).map_err(|_| Error::other("invalid remote namespace scanner capability response"))?;
-        if response.version != NS_SCANNER_PROTOCOL_VERSION || response.server_epoch.is_nil() {
-            return Err(Error::other("incompatible remote namespace scanner capability response"));
-        }
-        if let Err(err) = verify_ns_scanner_capability_with_tier_registry_generation(
-            challenge,
-            response.server_epoch,
-            &response.proof,
-            request.supports_tier_registry_generation,
-        ) {
-            // A permissive older peer can ignore the additive marker and
-            // return a valid legacy-scope proof with HTTP 200. Accept that
-            // response only after independently authenticating the legacy
-            // scope; all other verification failures remain fail-closed.
-            if request.supports_tier_registry_generation && ns_scanner_capability_legacy_proof_is_valid(challenge, &response) {
-                response.supports_tier_registry_generation = None;
-                return Ok(response);
-            }
-            return Err(Error::other(format!("remote namespace scanner capability authentication failed: {err}")));
-        }
-        // The proof authenticates the requested capability scope, not the
-        // optional response field. Derive the client-facing bit from that
-        // verified scope so an intermediary cannot strip or rewrite the field
-        // and force a silent downgrade after a successful generation-bound
-        // handshake.
-        normalize_ns_scanner_capability_response(&mut response, request.supports_tier_registry_generation);
-        Ok(response)
-    }
-
     async fn put_file_auth_capability(&self, endpoint: &str) -> Result<Option<Uuid>> {
         resolve_put_file_auth_capability(endpoint, || async {
             tokio::time::timeout(PUT_FILE_CAPABILITY_PROBE_TIMEOUT, self.probe_put_file_auth(endpoint))
@@ -730,14 +649,6 @@ fn build_walk_dir_url(request: &WalkDirStreamRequest) -> String {
     )
 }
 
-fn normalize_ns_scanner_capability_response(response: &mut NsScannerCapabilityResponse, requested_generation_support: bool) {
-    response.supports_tier_registry_generation = requested_generation_support.then_some(true);
-}
-
-fn ns_scanner_capability_legacy_proof_is_valid(challenge: Uuid, response: &NsScannerCapabilityResponse) -> bool {
-    verify_ns_scanner_capability_with_tier_registry_generation(challenge, response.server_epoch, &response.proof, false).is_ok()
-}
-
 fn build_ns_scanner_url(request: &NsScannerStreamRequest) -> String {
     let body_sha256 = hex_simd::encode_to_string(Sha256::digest(&request.body), hex_simd::AsciiCase::Lower);
     format!(
@@ -764,18 +675,13 @@ fn build_ns_scanner_url(request: &NsScannerStreamRequest) -> String {
 
 fn build_ns_scanner_capability_url(request: &NsScannerCapabilityRequest, challenge: Uuid) -> String {
     format!(
-        "{}{}?{}={}&{}={}{}",
+        "{}{}?{}={}&{}={}",
         request.endpoint,
         NS_SCANNER_PATH,
         NS_SCANNER_PROTOCOL_VERSION_QUERY,
         NS_SCANNER_PROTOCOL_VERSION,
         NS_SCANNER_CAPABILITY_CHALLENGE_QUERY,
-        challenge,
-        if request.supports_tier_registry_generation {
-            format!("&{}=true", NS_SCANNER_TIER_REGISTRY_GENERATION_QUERY)
-        } else {
-            String::new()
-        }
+        challenge
     )
 }
 
@@ -888,7 +794,6 @@ mod tests {
         let probe_err = transport
             .probe_ns_scanner(NsScannerCapabilityRequest {
                 endpoint: "http://node1:9000".to_string(),
-                supports_tier_registry_generation: false,
             })
             .await
             .expect_err("legacy transport should report namespace scanner as unsupported");
@@ -1482,7 +1387,6 @@ mod tests {
         let url = build_ns_scanner_capability_url(
             &NsScannerCapabilityRequest {
                 endpoint: "http://node1:9000".to_string(),
-                supports_tier_registry_generation: false,
             },
             challenge,
         );
@@ -1495,85 +1399,6 @@ mod tests {
         );
     }
 
-    #[test]
-    fn ns_scanner_capability_url_marks_generation_support_only_when_requested() {
-        let challenge = Uuid::new_v4();
-        let url = build_ns_scanner_capability_url(
-            &NsScannerCapabilityRequest {
-                endpoint: "http://node1:9000".to_string(),
-                supports_tier_registry_generation: true,
-            },
-            challenge,
-        );
-
-        assert!(url.contains(&format!("&{}=true", NS_SCANNER_TIER_REGISTRY_GENERATION_QUERY)));
-    }
-
-    #[test]
-    fn ns_scanner_capability_legacy_fallback_requires_explicit_compatibility_status() {
-        for status in [400, 404, 405, 426] {
-            let error = Error::from(rustfs_rio::new_test_internode_http_io_error(
-                rustfs_rio::InternodeHttpErrorKind::HttpStatus(http::StatusCode::from_u16(status).expect("test status")),
-            ));
-            assert!(
-                ns_scanner_capability_error_allows_legacy(&error),
-                "status {status} should permit legacy retry"
-            );
-        }
-
-        let marked_server_error = Error::from(rustfs_rio::new_test_internode_http_io_error(
-            rustfs_rio::InternodeHttpErrorKind::HttpStatus(http::StatusCode::INTERNAL_SERVER_ERROR),
-        ));
-        let network_error = Error::from(rustfs_rio::new_test_internode_http_io_error(
-            rustfs_rio::InternodeHttpErrorKind::ConnectionRefused,
-        ));
-        let authentication_error = Error::other("remote namespace scanner capability authentication failed");
-        assert!(!ns_scanner_capability_error_allows_legacy(&marked_server_error));
-        assert!(!ns_scanner_capability_error_allows_legacy(&network_error));
-        assert!(!ns_scanner_capability_error_allows_legacy(&authentication_error));
-    }
-
-    #[test]
-    fn authenticated_ns_scanner_capability_ignores_unprotected_response_bit() {
-        let mut response = NsScannerCapabilityResponse {
-            version: NS_SCANNER_PROTOCOL_VERSION,
-            server_epoch: Uuid::new_v4(),
-            proof: Vec::new(),
-            supports_tier_registry_generation: None,
-        };
-
-        normalize_ns_scanner_capability_response(&mut response, true);
-        assert_eq!(response.supports_tier_registry_generation, Some(true));
-
-        response.supports_tier_registry_generation = Some(false);
-        normalize_ns_scanner_capability_response(&mut response, false);
-        assert_eq!(response.supports_tier_registry_generation, None);
-    }
-
-    #[test]
-    fn ns_scanner_capability_accepts_only_authenticated_legacy_scope_after_marker_mismatch() {
-        crate::runtime::sources::ensure_test_rpc_secret();
-        let challenge = Uuid::new_v4();
-        let response = NsScannerCapabilityResponse {
-            version: NS_SCANNER_PROTOCOL_VERSION,
-            server_epoch: Uuid::new_v4(),
-            proof: crate::cluster::rpc::sign_ns_scanner_capability(challenge, Uuid::new_v4())
-                .expect("placeholder proof should be generated"),
-            supports_tier_registry_generation: None,
-        };
-        // A proof bound to a different challenge cannot authorize the legacy
-        // fallback, even though the response has the expected shape.
-        assert!(!ns_scanner_capability_legacy_proof_is_valid(challenge, &response));
-
-        let server_epoch = response.server_epoch;
-        let valid_response = NsScannerCapabilityResponse {
-            proof: crate::cluster::rpc::sign_ns_scanner_capability(challenge, server_epoch)
-                .expect("legacy proof should be generated"),
-            ..response
-        };
-        assert!(ns_scanner_capability_legacy_proof_is_valid(challenge, &valid_response));
-    }
-
     #[test]
     fn transport_config_defaults_to_tcp_http() {
         let transport = build_internode_data_transport(None).unwrap();

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

@@ -35,9 +35,8 @@ pub use http_auth::{
     TONIC_RPC_PREFIX, build_auth_headers, build_put_file_auth_trailer, check_and_record_signed_rpc_nonce, gen_signature_headers,
     gen_tonic_replay_scope_headers, gen_tonic_signature_headers, normalize_tonic_rpc_audience, set_tonic_canonical_body_digest,
     set_tonic_mutation_body_digest, set_tonic_rolling_canonical_body_digest, set_tonic_rolling_mutation_body_digest,
-    sign_ns_scanner_capability, sign_ns_scanner_capability_with_tier_registry_generation, sign_put_file_capability,
-    sign_tonic_rpc_response_proof, tonic_boot_epoch_challenge, tonic_boot_epoch_response_headers, tonic_rpc_auth_failure_reason,
-    verify_ns_scanner_capability, verify_ns_scanner_capability_with_tier_registry_generation, verify_put_file_auth_trailer,
+    sign_ns_scanner_capability, sign_put_file_capability, sign_tonic_rpc_response_proof, tonic_boot_epoch_challenge,
+    tonic_boot_epoch_response_headers, tonic_rpc_auth_failure_reason, verify_ns_scanner_capability, verify_put_file_auth_trailer,
     verify_put_file_capability, verify_rpc_signature, verify_tonic_boot_epoch_response, verify_tonic_canonical_body_digest,
     verify_tonic_mutation_body_digest, verify_tonic_rpc_response_proof, verify_tonic_rpc_signature,
     verify_tonic_rpc_signature_with_bootstrap,

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

@@ -781,16 +781,14 @@ impl RemoteDisk {
         if self.health.is_faulty() {
             return Err(DiskError::FaultyDisk);
         }
-        let probe = self.data_transport.probe_ns_scanner_capability(NsScannerCapabilityRequest {
+        let probe = self.data_transport.probe_ns_scanner(NsScannerCapabilityRequest {
             endpoint: self.endpoint.grid_host(),
-            supports_tier_registry_generation: true,
         });
         let result = timeout(NS_SCANNER_CAPABILITY_PROBE_TIMEOUT, probe)
             .await
             .map_err(|_| DiskError::other("remote namespace scanner capability probe timed out"))?;
         match result {
-            Ok(response) if response.supports_tier_registry_generation == Some(true) => Ok(Some(response.server_epoch)),
-            Ok(_) => Ok(None),
+            Ok(server_epoch) => Ok(Some(server_epoch)),
             // RUSTFS_COMPAT_TODO(ns-scanner-rpc-v3): old peers and legacy transports lack the authenticated startup-epoch handshake. Remove after every supported peer implements namespace scanner protocol v3.
             Err(DiskError::MethodNotAllowed) => Ok(None),
             Err(err)
@@ -4042,21 +4040,10 @@ mod tests {
         NsScannerProbe(NsScannerCapabilityRequest),
     }
 
-    #[derive(Debug, Clone)]
+    #[derive(Debug, Clone, Default)]
     struct RecordingInternodeDataTransport {
         calls: Arc<StdMutex<Vec<RecordedTransportCall>>>,
         ns_scanner_probe_status: Arc<StdMutex<Option<u16>>>,
-        ns_scanner_generation_support: Arc<StdMutex<Option<bool>>>,
-    }
-
-    impl Default for RecordingInternodeDataTransport {
-        fn default() -> Self {
-            Self {
-                calls: Arc::default(),
-                ns_scanner_probe_status: Arc::default(),
-                ns_scanner_generation_support: Arc::new(StdMutex::new(Some(true))),
-            }
-        }
     }
 
     #[derive(Clone, Debug)]
@@ -4276,15 +4263,6 @@ mod tests {
             Self {
                 calls: Arc::default(),
                 ns_scanner_probe_status: Arc::new(StdMutex::new(Some(status))),
-                ns_scanner_generation_support: Arc::new(StdMutex::new(Some(true))),
-            }
-        }
-
-        fn with_ns_scanner_generation_support(support: Option<bool>) -> Self {
-            Self {
-                calls: Arc::default(),
-                ns_scanner_probe_status: Arc::default(),
-                ns_scanner_generation_support: Arc::new(StdMutex::new(support)),
             }
         }
 
@@ -4967,23 +4945,6 @@ mod tests {
             Ok(Uuid::from_u128(1))
         }
 
-        async fn probe_ns_scanner_capability(
-            &self,
-            request: NsScannerCapabilityRequest,
-        ) -> Result<crate::storage_api_contracts::internode::NsScannerCapabilityResponse> {
-            let server_epoch = self.probe_ns_scanner(request).await?;
-            let supports_tier_registry_generation = *self
-                .ns_scanner_generation_support
-                .lock()
-                .expect("namespace scanner generation support lock poisoned");
-            Ok(crate::storage_api_contracts::internode::NsScannerCapabilityResponse {
-                version: crate::storage_api_contracts::internode::NS_SCANNER_PROTOCOL_VERSION,
-                server_epoch,
-                proof: Vec::new(),
-                supports_tier_registry_generation,
-            })
-        }
-
         fn name(&self) -> &'static str {
             "recording"
         }
@@ -6796,22 +6757,6 @@ mod tests {
         }
     }
 
-    #[tokio::test]
-    async fn test_remote_disk_namespace_scanner_capability_falls_back_without_generation_support() {
-        for support in [None, Some(false)] {
-            let transport = RecordingInternodeDataTransport::with_ns_scanner_generation_support(support);
-            let remote_disk = new_remote_disk_with_transport(Arc::new(transport)).await;
-
-            assert_eq!(
-                remote_disk
-                    .ns_scanner_server_epoch()
-                    .await
-                    .expect("missing generation support should be classified as unsupported"),
-                None
-            );
-        }
-    }
-
     #[tokio::test]
     async fn test_remote_disk_namespace_scanner_capability_rejects_legacy_transport() {
         let remote_disk = new_remote_disk_with_transport(Arc::new(RetryingOpenReadInternodeDataTransport::default())).await;

crates/ecstore/src/disk/local.rs

@@ -8000,10 +8000,15 @@ impl DiskAPI for LocalDisk {
             use std::io::Write as _;
 
             let file_path = self.io_get_object_path(volume, path)?;
-            let lock_path = file_path.with_extension("rustfs-cas.lock");
             let path = path.to_string();
             let sync_metadata = effective_durability(volume).syncs_commit_metadata();
             return Ok(tokio::task::spawn_blocking(move || {
+                // A persistent directory lock bounds metadata growth. Removing
+                // per-target lock files can split flock ownership across inodes.
+                let lock_path = file_path
+                    .parent()
+                    .ok_or_else(|| std::io::Error::new(ErrorKind::InvalidInput, "conditional file has no parent"))?
+                    .join(".rustfs-cas.lock");
                 let lock = std::fs::OpenOptions::new()
                     .create(true)
                     .truncate(false)
@@ -8068,7 +8073,25 @@ impl DiskAPI for LocalDisk {
             .map_err(DiskError::from)??);
         }
 
-        #[cfg(not(unix))]
+        #[cfg(windows)]
+        {
+            let file_path = self.io_get_object_path(volume, path)?;
+            let sync_metadata = effective_durability(volume).syncs_commit_metadata();
+            let publication_root = self.publication_root.clone();
+            return Ok(tokio::task::spawn_blocking(move || {
+                os::compare_and_update_control_file(
+                    &file_path,
+                    expected.as_deref(),
+                    replacement.as_deref(),
+                    sync_metadata,
+                    &publication_root,
+                )
+            })
+            .await
+            .map_err(DiskError::from)??);
+        }
+
+        #[cfg(not(any(unix, windows)))]
         {
             let _ = (volume, path, expected, replacement);
             Err(DiskError::MethodNotAllowed)
@@ -21823,9 +21846,9 @@ mod test {
         assert!(matches!(results[1].as_ref().unwrap_err(), DiskError::Io(_)));
     }
 
-    #[cfg(unix)]
+    #[cfg(any(unix, windows))]
     #[tokio::test]
-    async fn conditional_file_update_never_deletes_a_new_owner() {
+    async fn windows_and_unix_conditional_file_update_never_deletes_a_new_owner() {
         use tempfile::tempdir;
 
         let dir = tempdir().expect("temp dir should be created");
@@ -21856,8 +21879,18 @@ mod test {
             disk.read_all(RUSTFS_META_BUCKET, HEALING_MARKER_PATH)
                 .await
                 .expect("new owner marker should remain"),
-            owner_b
+            owner_b.clone()
         );
+        assert_eq!(
+            disk.compare_and_update_file(RUSTFS_META_BUCKET, HEALING_MARKER_PATH, Some(owner_b), None)
+                .await
+                .expect("current owner should remove marker"),
+            ConditionalFileUpdate::Updated
+        );
+        assert!(matches!(
+            disk.read_all(RUSTFS_META_BUCKET, HEALING_MARKER_PATH).await,
+            Err(DiskError::FileNotFound)
+        ));
     }
 
     #[cfg(unix)]
@@ -21872,7 +21905,10 @@ mod test {
         let marker_path = disk
             .get_object_path(RUSTFS_META_BUCKET, HEALING_MARKER_PATH)
             .expect("marker path should resolve");
-        let lock_path = marker_path.with_extension("rustfs-cas.lock");
+        let lock_path = marker_path
+            .parent()
+            .expect("marker path should have a parent")
+            .join(".rustfs-cas.lock");
         let lock = std::fs::OpenOptions::new()
             .create(true)
             .truncate(false)
@@ -21893,6 +21929,40 @@ mod test {
         assert!(matches!(err, DiskError::Io(ref err) if err.kind() == ErrorKind::WouldBlock));
     }
 
+    #[cfg(windows)]
+    #[tokio::test]
+    async fn windows_conditional_file_update_returns_would_block_when_marker_lock_is_contended() {
+        let dir = tempfile::tempdir().expect("temp dir should be created");
+        let endpoint = Endpoint::try_from(dir.path().to_str().expect("temp dir should be utf8")).expect("endpoint should parse");
+        let disk = LocalDisk::new(&endpoint, false).await.expect("local disk should be created");
+        ensure_test_volume(&disk, RUSTFS_META_BUCKET).await;
+        let marker_path = disk
+            .get_object_path(RUSTFS_META_BUCKET, HEALING_MARKER_PATH)
+            .expect("marker path should resolve");
+        let lock_path = marker_path
+            .parent()
+            .expect("marker path should have a parent")
+            .join(".rustfs-cas.lock");
+        let lock = std::fs::OpenOptions::new()
+            .create(true)
+            .truncate(false)
+            .read(true)
+            .write(true)
+            .open(lock_path)
+            .expect("marker lock should open");
+        lock.try_lock().expect("marker lock should be held");
+
+        let err = tokio::time::timeout(
+            Duration::from_secs(1),
+            disk.compare_and_update_file(RUSTFS_META_BUCKET, HEALING_MARKER_PATH, None, Some(Bytes::from_static(b"owner"))),
+        )
+        .await
+        .expect("contended conditional update must not block")
+        .expect_err("contended conditional update must retry");
+
+        assert!(matches!(err, DiskError::Io(ref err) if err.kind() == ErrorKind::WouldBlock));
+    }
+
     #[cfg(target_os = "linux")]
     #[tokio::test]
     async fn replacement_io_paths_stay_under_the_mount_lease() {

crates/ecstore/src/disk/os.rs

@@ -12,6 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#[cfg(windows)]
+use crate::disk::ConditionalFileUpdate;
 use crate::disk::error::DiskError;
 use crate::disk::error::Result;
 use crate::disk::error_conv::to_file_error;
@@ -3458,6 +3460,89 @@ fn read_windows_relative_file(file_path: &Path, parent_guard: &ExistingBaseDirec
     Ok(Some(data))
 }
 
+#[cfg(windows)]
+pub(crate) fn compare_and_update_control_file(
+    file_path: &Path,
+    expected: Option<&[u8]>,
+    replacement: Option<&[u8]>,
+    sync_metadata: bool,
+    publication_root: &PublicationRoot,
+) -> io::Result<ConditionalFileUpdate> {
+    use windows_sys::{
+        Wdk::Storage::FileSystem::{
+            FILE_NON_DIRECTORY_FILE, FILE_OPEN, FILE_OPEN_IF, FILE_OPEN_REPARSE_POINT, FILE_SYNCHRONOUS_IO_NONALERT,
+        },
+        Win32::Storage::FileSystem::{
+            DELETE, FILE_ATTRIBUTE_NORMAL, FILE_READ_ATTRIBUTES, FILE_SHARE_READ, FILE_SHARE_WRITE, FILE_WRITE_DATA, SYNCHRONIZE,
+        },
+    };
+
+    let parent = file_path
+        .parent()
+        .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidInput, "conditional file has no parent"))?;
+    let parent_guard = lock_windows_directory_tree(parent, Some(parent), publication_root)?;
+    let lock = open_windows_relative(
+        parent_guard.last_handle()?,
+        std::ffi::OsStr::new(".rustfs-cas.lock"),
+        SYNCHRONIZE | FILE_READ_ATTRIBUTES | FILE_WRITE_DATA,
+        FILE_SHARE_READ | FILE_SHARE_WRITE,
+        FILE_OPEN_IF,
+        FILE_NON_DIRECTORY_FILE | FILE_OPEN_REPARSE_POINT | FILE_SYNCHRONOUS_IO_NONALERT,
+        FILE_ATTRIBUTE_NORMAL,
+        true,
+    )?;
+    validate_windows_owned_file(&lock)?;
+    match lock.as_file().try_lock() {
+        Ok(()) => {}
+        Err(std::fs::TryLockError::WouldBlock) => return Err(io::Error::from(io::ErrorKind::WouldBlock)),
+        Err(std::fs::TryLockError::Error(err)) => return Err(err),
+    }
+
+    let current = read_windows_relative_file(file_path, &parent_guard)?;
+    let matches = match (&current, expected) {
+        (None, None) => true,
+        (Some(current), Some(expected)) => current.as_slice() == expected,
+        _ => false,
+    };
+    if !matches {
+        return Ok(match current {
+            None => ConditionalFileUpdate::Missing,
+            Some(_) => ConditionalFileUpdate::Mismatch,
+        });
+    }
+
+    match replacement {
+        Some(replacement) => RenameDestinationPathGuard {
+            directory: parent.to_path_buf(),
+            _directory_guard: parent_guard,
+        }
+        .write_file_for_path_access(file_path, replacement, sync_metadata, sync_metadata)?,
+        None => {
+            let file_name = file_path
+                .file_name()
+                .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidInput, "conditional file must have a name"))?;
+            let file = open_windows_relative(
+                parent_guard.last_handle()?,
+                file_name,
+                DELETE | SYNCHRONIZE | FILE_READ_ATTRIBUTES,
+                FILE_SHARE_READ,
+                FILE_OPEN,
+                FILE_NON_DIRECTORY_FILE | FILE_OPEN_REPARSE_POINT | FILE_SYNCHRONOUS_IO_NONALERT,
+                0,
+                true,
+            )?;
+            validate_windows_owned_file(&file)?;
+            set_windows_file_delete_on_close(&file, true)?;
+            drop(file);
+            if sync_metadata {
+                fsync_dir_std(parent)?;
+            }
+        }
+    }
+
+    Ok(ConditionalFileUpdate::Updated)
+}
+
 #[cfg(windows)]
 fn open_windows_directory_component(
     parent: &WindowsDirectoryHandle,

crates/ecstore/src/io_support/bitrot.rs

@@ -784,24 +784,6 @@ pub(crate) fn create_deferred_bitrot_reader_with_stripe_handle(
 ///
 /// # Returns
 /// A Result containing the BitrotWriterWrapper or an error
-/// Size hint handed to `DiskAPI::create_file` for a bitrot-wrapped shard.
-///
-/// A known length is grown by one checksum per shard so the on-disk file size
-/// matches what the bitrot writer emits. A negative length is the
-/// unknown-size sentinel (`HashReader::SIZE_PRESERVE_LAYER`, used by SSE and
-/// compression) and must be preserved: `RemoteDisk::create_file` forwards it
-/// in the `put_file_stream` query, and the receiver only treats `size > 0` as
-/// a fixed body length when locating the authenticated trailer. Clamping it
-/// to `0` would claim an empty body and misframe the stream. `0` stays `0`
-/// because a genuinely empty object still means an empty body.
-fn bitrot_create_file_size(length: i64, shard_size: usize, checksum_algo: &HashAlgorithm) -> i64 {
-    if length <= 0 {
-        return length;
-    }
-    let length = length as usize;
-    (length.div_ceil(shard_size) * checksum_algo.size() + length) as i64
-}
-
 pub async fn create_bitrot_writer(
     is_inline_buffer: bool,
     disk: Option<&DiskStore>,
@@ -814,7 +796,12 @@ pub async fn create_bitrot_writer(
     let writer = if is_inline_buffer {
         CustomWriter::new_inline_buffer()
     } else if let Some(disk) = disk {
-        let length = bitrot_create_file_size(length, shard_size, &checksum_algo);
+        let length = if length > 0 {
+            let length = length as usize;
+            (length.div_ceil(shard_size) * checksum_algo.size() + length) as i64
+        } else {
+            0
+        };
 
         let file = disk.create_file("", volume, path, length).await?;
         #[cfg(feature = "hotpath")]
@@ -833,25 +820,6 @@ mod tests {
     use rustfs_rio::ChunkReader;
     use std::collections::VecDeque;
 
-    #[test]
-    fn bitrot_create_file_size_grows_known_length_by_checksums() {
-        // 10 bytes over 4-byte shards = 3 shards, each followed by a 32-byte hash.
-        assert_eq!(bitrot_create_file_size(10, 4, &HashAlgorithm::HighwayHash256), 10 + 3 * 32);
-        assert_eq!(bitrot_create_file_size(10, 4, &HashAlgorithm::None), 10);
-    }
-
-    #[test]
-    fn bitrot_create_file_size_keeps_empty_and_unknown_distinct() {
-        assert_eq!(bitrot_create_file_size(0, 4, &HashAlgorithm::HighwayHash256), 0);
-        // SSE/compression streams advertise SIZE_PRESERVE_LAYER (-1); the remote
-        // put_file_stream receiver relies on a non-positive size to parse the auth
-        // trailer from the stream tail, so the sentinel must survive untouched.
-        assert_eq!(
-            bitrot_create_file_size(rustfs_rio::HashReader::SIZE_PRESERVE_LAYER, 4, &HashAlgorithm::HighwayHash256),
-            rustfs_rio::HashReader::SIZE_PRESERVE_LAYER
-        );
-    }
-
     struct TestChunkReader {
         chunks: VecDeque<Bytes>,
     }

crates/ecstore/src/storage_api_contracts/mod.rs

@@ -27,12 +27,12 @@ pub(crate) mod internode {
         NS_SCANNER_BODY_SHA256_QUERY, NS_SCANNER_CAPABILITY_CHALLENGE_QUERY, NS_SCANNER_CYCLE_QUERY,
         NS_SCANNER_LEADER_EPOCH_QUERY, NS_SCANNER_PROTOCOL_VERSION, NS_SCANNER_PROTOCOL_VERSION_QUERY,
         NS_SCANNER_REQUEST_ID_QUERY, NS_SCANNER_SERVER_EPOCH_QUERY, NS_SCANNER_SESSION_ID_QUERY,
-        NS_SCANNER_SESSION_SEQUENCE_QUERY, NS_SCANNER_TIER_REGISTRY_GENERATION_QUERY, NsScannerCapabilityResponse,
-        PUT_FILE_AUTH_QUERY, PUT_FILE_AUTH_TRAILER_DIGEST_LEN, PUT_FILE_AUTH_TRAILER_LEN, PUT_FILE_AUTH_TRAILER_MAC_LEN,
-        PUT_FILE_AUTH_TRAILER_MAGIC, PUT_FILE_AUTH_V1, PUT_FILE_CAPABILITY_CHALLENGE_QUERY, PUT_FILE_CAPABILITY_QUERY,
-        PUT_FILE_CAPABILITY_VERSION, PUT_FILE_NONCE_QUERY, PUT_FILE_SERVER_EPOCH_QUERY, PutFileCapabilityResponse,
-        SCANNER_ACTIVITY_LEGACY_PROTOCOL_VERSION, SCANNER_ACTIVITY_PREVIOUS_PROTOCOL_VERSION, SCANNER_ACTIVITY_PROTOCOL_VERSION,
-        WALK_DIR_BODY_SHA256_QUERY, WALK_DIR_STREAM_COMPLETION_QUERY, WALK_DIR_STREAM_COMPLETION_V1,
+        NS_SCANNER_SESSION_SEQUENCE_QUERY, NsScannerCapabilityResponse, PUT_FILE_AUTH_QUERY, PUT_FILE_AUTH_TRAILER_DIGEST_LEN,
+        PUT_FILE_AUTH_TRAILER_LEN, PUT_FILE_AUTH_TRAILER_MAC_LEN, PUT_FILE_AUTH_TRAILER_MAGIC, PUT_FILE_AUTH_V1,
+        PUT_FILE_CAPABILITY_CHALLENGE_QUERY, PUT_FILE_CAPABILITY_QUERY, PUT_FILE_CAPABILITY_VERSION, PUT_FILE_NONCE_QUERY,
+        PUT_FILE_SERVER_EPOCH_QUERY, PutFileCapabilityResponse, SCANNER_ACTIVITY_LEGACY_PROTOCOL_VERSION,
+        SCANNER_ACTIVITY_PREVIOUS_PROTOCOL_VERSION, SCANNER_ACTIVITY_PROTOCOL_VERSION, WALK_DIR_BODY_SHA256_QUERY,
+        WALK_DIR_STREAM_COMPLETION_QUERY, WALK_DIR_STREAM_COMPLETION_V1,
     };
 }
 

crates/heal/Cargo.toml

@@ -91,6 +91,7 @@ metrics = { workspace = true }
 base64 = { workspace = true }
 bytes = { workspace = true }
 crc-fast = { workspace = true }
+sha2 = { workspace = true }
 
 [dev-dependencies]
 serde_json = { workspace = true, features = ["raw_value"] }

crates/heal/src/heal/erasure_healer.rs

@@ -373,6 +373,11 @@ impl ErasureSetHealer {
         set_disk_id: &str,
         buckets: &[String],
     ) -> Result<(ResumeManager, CheckpointManager)> {
+        if self.replacement_task_id.is_none() && CheckpointManager::is_blocked(&self.disk, task_id).await {
+            return Err(Error::TaskExecutionFailed {
+                message: format!("Resume task {task_id} has a blocked checkpoint"),
+            });
+        }
         // check if resume state exists
         let has_resume_state = if self.replacement_task_id.is_some() {
             ResumeManager::has_replacement_intent(&self.disk, task_id).await

crates/heal/src/heal/resume.rs

@@ -51,6 +51,7 @@ const RESUME_STATE_FILE: &str = "ahm_resume_state.json";
 const REPLACEMENT_INTENT_FILE: &str = "ahm_replacement_intent.json";
 const RESUME_PROGRESS_FILE: &str = "ahm_progress.json";
 pub(super) const RESUME_CHECKPOINT_FILE: &str = "ahm_checkpoint.json";
+pub(super) const RESUME_CHECKPOINT_BLOCKED_FILE: &str = "ahm_checkpoint.blocked";
 const REPLACEMENT_COMPLETION_PROOF_FILE: &str = "ahm_replacement_completion_proof.json";
 const REPLACEMENT_RECOVERY_DIR: &str = "ahm-replacement";
 const REPLACEMENT_INTENT_SEAL_FILE: &str = "ahm_replacement_intent_seal";

crates/heal/src/heal/resume/checkpoint.rs

@@ -13,26 +13,31 @@
 // limitations under the License.
 
 use crate::{Error, Result};
+use base64::Engine as _;
 use serde::{Deserialize, Serialize};
+use sha2::{Digest, Sha256};
 use std::collections::HashSet;
 use std::path::Path;
 use std::sync::{Arc, Mutex};
 use std::time::{SystemTime, UNIX_EPOCH};
-use tokio::sync::RwLock;
+use tokio::sync::{Mutex as AsyncMutex, RwLock};
 use tracing::{debug, warn};
 
-use super::super::{BUCKET_META_PREFIX, DiskStore, HealDiskExt as _, RUSTFS_META_BUCKET};
+use super::super::storage_api::owner::{EcstoreConditionalFileUpdate, EcstoreDiskAPI, EcstoreDiskBytes};
+use super::super::{BUCKET_META_PREFIX, DiskStore, HealDiskExt, RUSTFS_META_BUCKET};
 use super::{
-    LOG_COMPONENT_HEAL, LOG_SUBSYSTEM_RESUME, PersistThrottle, RESUME_CHECKPOINT_FILE, delete_resume_file, path_to_str,
-    validate_resume_task_id,
+    LOG_COMPONENT_HEAL, LOG_SUBSYSTEM_RESUME, PersistThrottle, RESUME_CHECKPOINT_BLOCKED_FILE, RESUME_CHECKPOINT_FILE,
+    delete_resume_file, path_to_str, validate_resume_task_id,
 };
 
 const EVENT_HEAL_CHECKPOINT_STATE: &str = "heal_checkpoint_state";
+const RESUME_CHECKPOINT_DIGEST_FILE: &str = "ahm_checkpoint.sha256";
+const CHECKPOINT_PER_VERSION_SCHEMA: u32 = 5;
 
 /// Current on-disk schema version for `ResumeCheckpoint`. Same rationale as
 /// `CURRENT_RESUME_SCHEMA`: pre-per-version dedup identities are not comparable
 /// to the new `compose_key` identities, so a stale checkpoint is discarded.
-pub(super) const CURRENT_CHECKPOINT_SCHEMA: u32 = 5;
+pub(super) const CURRENT_CHECKPOINT_SCHEMA: u32 = 6;
 
 /// resume checkpoint
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -57,6 +62,11 @@ pub struct ResumeCheckpoint {
     pub failed_objects: HashSet<String>,
     /// skipped objects
     pub skipped_objects: HashSet<String>,
+    /// Integrity digest over the checkpoint with this field set to `None`.
+    /// Keeping it in the checkpoint makes the payload and its authentication
+    /// record one CAS generation instead of two independently-written files.
+    #[serde(default)]
+    pub integrity_digest: Option<String>,
 }
 
 impl ResumeCheckpoint {
@@ -70,6 +80,7 @@ impl ResumeCheckpoint {
             processed_objects: HashSet::new(),
             failed_objects: HashSet::new(),
             skipped_objects: HashSet::new(),
+            integrity_digest: None,
         }
     }
 
@@ -116,17 +127,111 @@ pub struct CheckpointManager {
     disk: DiskStore,
     checkpoint: Arc<RwLock<ResumeCheckpoint>>,
     throttle: Mutex<PersistThrottle>,
+    save_lock: AsyncMutex<()>,
+    last_saved: Mutex<Option<EcstoreDiskBytes>>,
 }
 
 impl CheckpointManager {
+    fn blocked_path(task_id: &str) -> std::path::PathBuf {
+        Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_BLOCKED_FILE}"))
+    }
+
+    /// Return whether a checkpoint was permanently isolated after a malformed
+    /// or unsupported snapshot was observed.
+    pub(crate) async fn is_blocked(disk: &DiskStore, task_id: &str) -> bool {
+        if validate_resume_task_id(task_id).is_err() {
+            return false;
+        }
+        let blocked_path = Self::blocked_path(task_id);
+        let Ok(path) = path_to_str(&blocked_path) else {
+            return false;
+        };
+        match HealDiskExt::read_all(disk.as_ref(), RUSTFS_META_BUCKET, path).await {
+            Ok(_) => true,
+            Err(crate::heal::DiskError::FileNotFound) => false,
+            Err(_) => true,
+        }
+    }
+
+    /// Validate the checkpoint while enumerating resumable state. This reads
+    /// the checkpoint once and also isolates malformed or unsupported data.
+    pub(crate) async fn is_resumable(disk: &DiskStore, task_id: &str) -> Result<bool> {
+        validate_resume_task_id(task_id)?;
+        if Self::is_blocked(disk, task_id).await {
+            return Err(Error::InvalidCheckpoint(format!("Resume task {task_id} has a blocked checkpoint")));
+        }
+        let file_path = Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_FILE}"));
+        let Ok(path) = path_to_str(&file_path) else {
+            return Err(Error::InvalidCheckpoint("Resume checkpoint path is not valid UTF-8".to_string()));
+        };
+        match HealDiskExt::read_all(disk.as_ref(), RUSTFS_META_BUCKET, path).await {
+            Ok(bytes) if bytes.is_empty() => Ok(true),
+            Ok(bytes) => Self::load_from_data(disk.clone(), task_id, bytes.to_vec())
+                .await
+                .map(|_| true),
+            Err(crate::heal::DiskError::FileNotFound) => Ok(true),
+            Err(error) => Err(error.into()),
+        }
+    }
+
+    async fn block_invalid_snapshot(disk: &DiskStore, task_id: &str) {
+        // This marker is intentionally version-agnostic: an unsupported reader
+        // must stop selector retries until an operator cleans up the snapshot.
+        let blocked_path = Self::blocked_path(task_id);
+        let Ok(path) = path_to_str(&blocked_path) else {
+            return;
+        };
+        let result = EcstoreDiskAPI::compare_and_update_file(
+            disk.as_ref(),
+            RUSTFS_META_BUCKET,
+            path,
+            None,
+            Some(EcstoreDiskBytes::from_static(b"blocked")),
+        )
+        .await;
+        match result {
+            Ok(EcstoreConditionalFileUpdate::Updated | EcstoreConditionalFileUpdate::Mismatch) => {}
+            Ok(EcstoreConditionalFileUpdate::Missing) => warn!(
+                target: "rustfs::heal::resume",
+                event = EVENT_HEAL_CHECKPOINT_STATE,
+                component = LOG_COMPONENT_HEAL,
+                subsystem = LOG_SUBSYSTEM_RESUME,
+                task_id,
+                state = "blocked_marker_write_failed",
+                error = "marker target disappeared",
+                "Heal checkpoint could not persist its blocked marker"
+            ),
+            Err(error) => warn!(
+                target: "rustfs::heal::resume",
+                event = EVENT_HEAL_CHECKPOINT_STATE,
+                component = LOG_COMPONENT_HEAL,
+                subsystem = LOG_SUBSYSTEM_RESUME,
+                task_id,
+                state = "blocked_marker_write_failed",
+                error = %error,
+                "Heal checkpoint could not persist its blocked marker"
+            ),
+        }
+    }
+
     /// create new checkpoint manager
     pub async fn new(disk: DiskStore, task_id: String) -> Result<Self> {
         validate_resume_task_id(&task_id)?;
+        let checkpoint_volume = format!("{RUSTFS_META_BUCKET}/{BUCKET_META_PREFIX}");
+        if let Err(error) = EcstoreDiskAPI::make_volume(disk.as_ref(), &checkpoint_volume).await
+            && error != crate::heal::DiskError::VolumeExists
+        {
+            return Err(Error::TaskExecutionFailed {
+                message: format!("Failed to create checkpoint volume: {error}"),
+            });
+        }
         let checkpoint = ResumeCheckpoint::new(task_id);
         let manager = Self {
             disk,
             checkpoint: Arc::new(RwLock::new(checkpoint)),
             throttle: Mutex::new(PersistThrottle::new()),
+            save_lock: AsyncMutex::new(()),
+            last_saved: Mutex::new(None),
         };
 
         // save initial checkpoint
@@ -140,6 +245,7 @@ impl CheckpointManager {
                 error = %e,
                 "Heal checkpoint persistence failed"
             );
+            return Err(e);
         }
         Ok(manager)
     }
@@ -148,11 +254,22 @@ impl CheckpointManager {
     pub async fn load_from_disk(disk: DiskStore, task_id: &str) -> Result<Self> {
         validate_resume_task_id(task_id)?;
         let checkpoint_data = Self::read_checkpoint_file(&disk, task_id).await?;
-        let mut checkpoint: ResumeCheckpoint =
-            serde_json::from_slice(&checkpoint_data).map_err(|e| Error::TaskExecutionFailed {
-                message: format!("Failed to deserialize checkpoint: {e}"),
-            })?;
+        Self::load_from_data(disk, task_id, checkpoint_data).await
+    }
+
+    async fn load_from_data(disk: DiskStore, task_id: &str, checkpoint_data: Vec<u8>) -> Result<Self> {
+        validate_resume_task_id(task_id)?;
+        let mut checkpoint: ResumeCheckpoint = match serde_json::from_slice(&checkpoint_data) {
+            Ok(checkpoint) => checkpoint,
+            Err(error) => {
+                Self::block_invalid_snapshot(&disk, task_id).await;
+                return Err(Error::TaskExecutionFailed {
+                    message: format!("Failed to deserialize checkpoint: {error}"),
+                });
+            }
+        };
         if checkpoint.task_id != task_id {
+            Self::block_invalid_snapshot(&disk, task_id).await;
             return Err(Error::TaskExecutionFailed {
                 message: "Resume checkpoint task id does not match filename".to_string(),
             });
@@ -163,6 +280,7 @@ impl CheckpointManager {
         // identities. Discard the stale sets and position, then stamp the
         // current schema so the scan restarts cleanly.
         if checkpoint.schema_version > CURRENT_CHECKPOINT_SCHEMA {
+            Self::block_invalid_snapshot(&disk, task_id).await;
             return Err(Error::TaskExecutionFailed {
                 message: format!(
                     "Checkpoint schema {} is newer than supported schema {CURRENT_CHECKPOINT_SCHEMA}",
@@ -170,7 +288,45 @@ impl CheckpointManager {
                 ),
             });
         }
-        if checkpoint.schema_version < CURRENT_CHECKPOINT_SCHEMA {
+
+        let integrity_verified = if let Some(expected) = checkpoint.integrity_digest.as_deref() {
+            let actual = Self::checkpoint_digest(&Self::serialize_without_digest(&checkpoint)?);
+            if expected != actual {
+                Self::block_invalid_snapshot(&disk, task_id).await;
+                return Err(Error::InvalidCheckpoint(format!(
+                    "Resume checkpoint digest does not match task {task_id}"
+                )));
+            }
+            true
+        } else if checkpoint.schema_version >= CURRENT_CHECKPOINT_SCHEMA {
+            Self::block_invalid_snapshot(&disk, task_id).await;
+            return Err(Error::InvalidCheckpoint(format!(
+                "Resume checkpoint digest is missing for task {task_id}"
+            )));
+        } else {
+            let digest_path = Self::digest_path(task_id);
+            let digest_path = path_to_str(&digest_path)?;
+            match HealDiskExt::read_all(disk.as_ref(), RUSTFS_META_BUCKET, digest_path).await {
+                Ok(expected) => {
+                    let actual = Self::checkpoint_digest(&checkpoint_data);
+                    if expected.as_ref() != actual.as_bytes() {
+                        Self::block_invalid_snapshot(&disk, task_id).await;
+                        return Err(Error::InvalidCheckpoint(format!(
+                            "Resume checkpoint digest does not match task {task_id}"
+                        )));
+                    }
+                    true
+                }
+                Err(crate::heal::DiskError::FileNotFound) => false,
+                Err(error) => {
+                    return Err(Error::TaskExecutionFailed {
+                        message: format!("Failed to read checkpoint digest: {error}"),
+                    });
+                }
+            }
+        };
+
+        if checkpoint.schema_version < CHECKPOINT_PER_VERSION_SCHEMA || !integrity_verified {
             warn!(
                 target: "rustfs::heal::resume",
                 event = EVENT_HEAL_CHECKPOINT_STATE,
@@ -187,13 +343,15 @@ impl CheckpointManager {
             checkpoint.skipped_objects.clear();
             checkpoint.current_bucket_index = 0;
             checkpoint.current_object_index = 0;
-            checkpoint.schema_version = CURRENT_CHECKPOINT_SCHEMA;
         }
+        checkpoint.schema_version = CURRENT_CHECKPOINT_SCHEMA;
 
         Ok(Self {
             disk,
             checkpoint: Arc::new(RwLock::new(checkpoint)),
             throttle: Mutex::new(PersistThrottle::new()),
+            save_lock: AsyncMutex::new(()),
+            last_saved: Mutex::new(Some(EcstoreDiskBytes::from(checkpoint_data))),
         })
     }
 
@@ -204,7 +362,7 @@ impl CheckpointManager {
         }
         let file_path = Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_FILE}"));
         match path_to_str(&file_path) {
-            Ok(path_str) => match disk.read_all(RUSTFS_META_BUCKET, path_str).await {
+            Ok(path_str) => match HealDiskExt::read_all(disk.as_ref(), RUSTFS_META_BUCKET, path_str).await {
                 Ok(data) => !data.is_empty(),
                 Err(_) => false,
             },
@@ -292,6 +450,8 @@ impl CheckpointManager {
 
         let checkpoint_file = Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_FILE}"));
         delete_resume_file(&self.disk, &checkpoint_file).await?;
+        delete_resume_file(&self.disk, &Self::digest_path(&task_id)).await?;
+        delete_resume_file(&self.disk, &Self::blocked_path(&task_id)).await?;
 
         debug!(
             target: "rustfs::heal::resume",
@@ -307,21 +467,130 @@ impl CheckpointManager {
 
     /// save checkpoint to disk
     async fn save_checkpoint(&self) -> Result<()> {
-        let checkpoint = self.checkpoint.read().await;
+        // Serialize saves and take the snapshot only after acquiring the lock:
+        // a slower writer must not publish a snapshot taken before a newer one.
+        let _save_guard = self.save_lock.lock().await;
+        let checkpoint = self.checkpoint.read().await.clone();
         validate_resume_task_id(&checkpoint.task_id)?;
-        let checkpoint_data = serde_json::to_vec(&*checkpoint).map_err(|e| Error::TaskExecutionFailed {
-            message: format!("Failed to serialize checkpoint: {e}"),
-        })?;
+        let unsigned_checkpoint_data = Self::serialize_without_digest(&checkpoint)?;
+        let digest = Self::checkpoint_digest(&unsigned_checkpoint_data);
+        let mut persisted_checkpoint = checkpoint.clone();
+        persisted_checkpoint.integrity_digest = Some(digest);
+        let checkpoint_data =
+            EcstoreDiskBytes::from(serde_json::to_vec(&persisted_checkpoint).map_err(|e| Error::TaskExecutionFailed {
+                message: format!("Failed to serialize checkpoint: {e}"),
+            })?);
 
         let file_path = Path::new(BUCKET_META_PREFIX).join(format!("{}_{}", checkpoint.task_id, RESUME_CHECKPOINT_FILE));
 
         let path_str = path_to_str(&file_path)?;
-        self.disk
-            .write_all(RUSTFS_META_BUCKET, path_str, checkpoint_data.into())
+        let last_saved = self
+            .last_saved
+            .lock()
+            .map_err(|_| Error::TaskExecutionFailed {
+                message: "Checkpoint save state lock is poisoned; refusing to save".to_string(),
+            })?
+            .clone();
+        let update = EcstoreDiskAPI::compare_and_update_file(
+            self.disk.as_ref(),
+            RUSTFS_META_BUCKET,
+            path_str,
+            last_saved.clone(),
+            Some(checkpoint_data.clone()),
+        )
+        .await
+        .map_err(|e| Error::TaskExecutionFailed {
+            message: format!("Failed to save checkpoint: {e}"),
+        })?;
+
+        let expected = match update {
+            EcstoreConditionalFileUpdate::Updated => None,
+            EcstoreConditionalFileUpdate::Missing => {
+                return Err(Error::TaskExecutionFailed {
+                    message: "Checkpoint was removed after this manager saved it; refusing to recreate it".to_string(),
+                });
+            }
+            EcstoreConditionalFileUpdate::Mismatch => {
+                // A healthy manager normally completes the CAS above without
+                // another read or JSON parse. Inspect only after a mismatch so
+                // corruption and future schemas cannot be overwritten blindly.
+                let existing = match HealDiskExt::read_all(self.disk.as_ref(), RUSTFS_META_BUCKET, path_str).await {
+                    Ok(existing) => existing,
+                    Err(crate::heal::DiskError::FileNotFound) => {
+                        return Err(Error::TaskExecutionFailed {
+                            message: "Checkpoint was removed after this manager saved it; refusing to recreate it".to_string(),
+                        });
+                    }
+                    Err(error) => {
+                        return Err(Error::TaskExecutionFailed {
+                            message: format!("Failed to inspect checkpoint after CAS mismatch: {error}"),
+                        });
+                    }
+                };
+
+                if existing.is_empty() && last_saved.is_none() {
+                    Some(existing)
+                } else {
+                    let current: ResumeCheckpoint = match serde_json::from_slice(&existing) {
+                        Ok(current) => current,
+                        Err(error) => {
+                            Self::block_invalid_snapshot(&self.disk, &checkpoint.task_id).await;
+                            return Err(Error::TaskExecutionFailed {
+                                message: format!("Existing checkpoint is corrupt: {error}"),
+                            });
+                        }
+                    };
+                    if current.task_id != checkpoint.task_id {
+                        Self::block_invalid_snapshot(&self.disk, &checkpoint.task_id).await;
+                        return Err(Error::TaskExecutionFailed {
+                            message: "Existing checkpoint task id does not match filename".to_string(),
+                        });
+                    }
+                    if current.schema_version > CURRENT_CHECKPOINT_SCHEMA {
+                        Self::block_invalid_snapshot(&self.disk, &checkpoint.task_id).await;
+                        return Err(Error::TaskExecutionFailed {
+                            message: format!(
+                                "Existing checkpoint schema {} is newer than supported schema {CURRENT_CHECKPOINT_SCHEMA}",
+                                current.schema_version
+                            ),
+                        });
+                    }
+                    if last_saved.as_ref().is_none_or(|saved| saved.as_ref() != existing.as_ref()) {
+                        return Err(Error::TaskExecutionFailed {
+                            message: "Checkpoint changed since this manager loaded it; refusing to overwrite newer progress"
+                                .to_string(),
+                        });
+                    }
+                    Some(existing)
+                }
+            }
+        };
+
+        if let Some(expected) = expected {
+            match EcstoreDiskAPI::compare_and_update_file(
+                self.disk.as_ref(),
+                RUSTFS_META_BUCKET,
+                path_str,
+                Some(expected),
+                Some(checkpoint_data.clone()),
+            )
             .await
             .map_err(|e| Error::TaskExecutionFailed {
-                message: format!("Failed to save checkpoint: {e}"),
-            })?;
+                message: format!("Failed to save checkpoint after CAS mismatch: {e}"),
+            })? {
+                EcstoreConditionalFileUpdate::Updated => {}
+                EcstoreConditionalFileUpdate::Missing | EcstoreConditionalFileUpdate::Mismatch => {
+                    return Err(Error::TaskExecutionFailed {
+                        message: "Checkpoint changed while saving; refusing to overwrite newer progress".to_string(),
+                    });
+                }
+            }
+        }
+
+        let mut last_saved = self.last_saved.lock().map_err(|_| Error::TaskExecutionFailed {
+            message: "Checkpoint save state lock is poisoned after save".to_string(),
+        })?;
+        *last_saved = Some(checkpoint_data);
 
         debug!(
             target: "rustfs::heal::resume",
@@ -341,11 +610,38 @@ impl CheckpointManager {
         let file_path = Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_FILE}"));
 
         let path_str = path_to_str(&file_path)?;
-        disk.read_all(RUSTFS_META_BUCKET, path_str)
+        HealDiskExt::read_all(disk.as_ref(), RUSTFS_META_BUCKET, path_str)
             .await
             .map(|bytes| bytes.to_vec())
             .map_err(|e| Error::TaskExecutionFailed {
                 message: format!("Failed to read checkpoint file: {e}"),
             })
     }
+
+    fn serialize_without_digest(checkpoint: &ResumeCheckpoint) -> Result<Vec<u8>> {
+        let mut unsigned = checkpoint.clone();
+        unsigned.integrity_digest = None;
+        let mut value = serde_json::to_value(&unsigned).map_err(|e| Error::TaskExecutionFailed {
+            message: format!("Failed to serialize checkpoint: {e}"),
+        })?;
+        for field in ["processed_objects", "failed_objects", "skipped_objects"] {
+            let Some(values) = value.get_mut(field).and_then(serde_json::Value::as_array_mut) else {
+                return Err(Error::TaskExecutionFailed {
+                    message: format!("Failed to canonicalize checkpoint field: {field}"),
+                });
+            };
+            values.sort_by(|left, right| left.as_str().cmp(&right.as_str()));
+        }
+        serde_json::to_vec(&value).map_err(|e| Error::TaskExecutionFailed {
+            message: format!("Failed to serialize checkpoint: {e}"),
+        })
+    }
+
+    fn checkpoint_digest(checkpoint_data: &[u8]) -> String {
+        base64::engine::general_purpose::STANDARD.encode(Sha256::digest(checkpoint_data))
+    }
+
+    fn digest_path(task_id: &str) -> std::path::PathBuf {
+        Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_DIGEST_FILE}"))
+    }
 }

crates/heal/src/heal/resume/tests.rs

@@ -1600,6 +1600,32 @@ async fn test_checkpoint_schema_v4_discarded_on_load() {
     temp_dir.close().expect("remove schema test directory");
 }
 
+#[tokio::test]
+async fn downgraded_unsigned_checkpoint_resets_untrusted_progress() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone()).await.unwrap();
+    manager.add_processed_object("victim-a".to_string()).await.unwrap();
+    manager.update_position(2, 500).await.unwrap();
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    let bytes = disk.read_all(RUSTFS_META_BUCKET, &checkpoint_path).await.unwrap();
+    let mut downgraded: serde_json::Value = serde_json::from_slice(&bytes).unwrap();
+    downgraded["schema_version"] = serde_json::json!(CURRENT_CHECKPOINT_SCHEMA - 1);
+    downgraded.as_object_mut().unwrap().remove("integrity_digest");
+    downgraded["processed_objects"] = serde_json::json!(["victim-b"]);
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, serde_json::to_vec(&downgraded).unwrap().into())
+        .await
+        .expect("write downgraded checkpoint");
+
+    let manager = CheckpointManager::load_from_disk(disk, &task_id).await.unwrap();
+    let checkpoint = manager.get_checkpoint().await;
+    assert_eq!(checkpoint.schema_version, CURRENT_CHECKPOINT_SCHEMA);
+    assert_eq!(checkpoint.current_bucket_index, 0);
+    assert_eq!(checkpoint.current_object_index, 0);
+    assert!(checkpoint.processed_objects.is_empty());
+    temp_dir.close().unwrap();
+}
+
 #[tokio::test]
 async fn current_normal_resume_schema_preserves_progress() {
     let (temp_dir, disk) = schema_test_disk().await;
@@ -1675,6 +1701,369 @@ async fn future_resume_and_checkpoint_schemas_are_rejected() {
     temp_dir.close().expect("remove schema test directory");
 }
 
+#[tokio::test]
+async fn checkpoint_save_does_not_replace_a_non_empty_truncated_snapshot() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    let truncated = b"{\"schema_version\":5,\"task_id\":";
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, truncated.as_slice().into())
+        .await
+        .expect("write truncated checkpoint fixture");
+
+    let error = manager
+        .update_position(2, 7)
+        .await
+        .expect_err("a truncated checkpoint must fail closed during save");
+    assert!(error.to_string().contains("Existing checkpoint is corrupt"));
+    assert_eq!(
+        disk.read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+            .await
+            .expect("read truncated checkpoint fixture"),
+        truncated.as_slice()
+    );
+    assert!(CheckpointManager::is_blocked(&disk, &task_id).await);
+    temp_dir.close().expect("remove checkpoint save test directory");
+}
+
+#[tokio::test]
+async fn checkpoint_save_does_not_replace_a_future_schema_snapshot() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    let mut future = ResumeCheckpoint::new(task_id.clone());
+    future.schema_version = CURRENT_CHECKPOINT_SCHEMA + 1;
+    let future_bytes = serde_json::to_vec(&future).expect("serialize future checkpoint fixture");
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, future_bytes.clone().into())
+        .await
+        .expect("write future checkpoint fixture");
+
+    let error = manager
+        .update_position(2, 7)
+        .await
+        .expect_err("a future schema must fail closed during save");
+    assert!(error.to_string().contains("Existing checkpoint schema"));
+    assert_eq!(
+        disk.read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+            .await
+            .expect("read future checkpoint fixture"),
+        future_bytes
+    );
+    assert!(CheckpointManager::is_blocked(&disk, &task_id).await);
+    temp_dir.close().expect("remove future schema test directory");
+}
+
+#[tokio::test]
+async fn checkpoint_digest_rejects_same_length_progress_tampering() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    manager
+        .add_processed_object("victim-a".to_string())
+        .await
+        .expect("persist checkpoint progress");
+    manager.update_position(1, 1).await.expect("flush checkpoint progress");
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    let original = disk
+        .read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+        .await
+        .expect("read checkpoint fixture");
+    let tampered = original
+        .windows(b"victim-a".len())
+        .position(|window| window == b"victim-a")
+        .map(|index| {
+            let mut bytes = original.to_vec();
+            bytes[index..index + b"victim-a".len()].copy_from_slice(b"victim-b");
+            bytes
+        })
+        .expect("checkpoint should contain the processed object");
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, tampered.into())
+        .await
+        .expect("write tampered checkpoint fixture");
+
+    assert!(CheckpointManager::load_from_disk(disk.clone(), &task_id).await.is_err());
+    assert!(CheckpointManager::is_blocked(&disk, &task_id).await);
+    temp_dir.close().expect("remove digest test directory");
+}
+
+#[tokio::test]
+async fn checkpoint_integrity_survives_missing_legacy_sidecar() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone()).await.unwrap();
+    manager.update_position(2, 9).await.unwrap();
+
+    let digest_path = format!("{BUCKET_META_PREFIX}/{task_id}_ahm_checkpoint.sha256");
+    delete_resume_file(&disk, Path::new(&digest_path)).await.unwrap();
+
+    let restored = CheckpointManager::load_from_disk(disk, &task_id).await.unwrap();
+    let checkpoint = restored.get_checkpoint().await;
+    assert_eq!(checkpoint.current_bucket_index, 2);
+    assert_eq!(checkpoint.current_object_index, 9);
+    assert!(checkpoint.integrity_digest.is_some());
+    temp_dir.close().unwrap();
+}
+
+#[tokio::test]
+async fn checkpoint_integrity_survives_multi_object_reload() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone()).await.unwrap();
+    for index in 0..32 {
+        manager.add_processed_object(format!("processed-{index}")).await.unwrap();
+        manager.add_failed_object(format!("failed-{index}")).await.unwrap();
+        manager.add_skipped_object(format!("skipped-{index}")).await.unwrap();
+    }
+    manager.update_position(2, 9).await.unwrap();
+
+    CheckpointManager::load_from_disk(disk, &task_id)
+        .await
+        .expect("a healthy multi-object checkpoint must survive reload");
+    temp_dir.close().unwrap();
+}
+
+#[tokio::test]
+async fn checkpoint_integrity_rejects_a_removed_embedded_digest() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone()).await.unwrap();
+    manager.update_position(2, 9).await.unwrap();
+
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    let bytes = disk
+        .read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+        .await
+        .expect("read checkpoint fixture");
+    let mut value: serde_json::Value = serde_json::from_slice(&bytes).unwrap();
+    value["current_object_index"] = serde_json::json!(10);
+    value.as_object_mut().unwrap().remove("integrity_digest");
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, serde_json::to_vec(&value).unwrap().into())
+        .await
+        .expect("write tampered checkpoint fixture");
+
+    assert!(
+        CheckpointManager::load_from_disk(disk.clone(), &task_id).await.is_err(),
+        "a current checkpoint without its embedded digest must fail closed"
+    );
+    assert!(CheckpointManager::is_blocked(&disk, &task_id).await);
+    temp_dir.close().unwrap();
+}
+
+#[tokio::test]
+async fn new_checkpoint_manager_rebuilds_an_empty_snapshot() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, EcstoreDiskBytes::new())
+        .await
+        .expect("write empty checkpoint fixture");
+
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("a new manager must rebuild an empty checkpoint");
+    manager
+        .update_position(3, 11)
+        .await
+        .expect("rebuilt checkpoint must remain writable");
+    assert!(CheckpointManager::has_checkpoint(&disk, &task_id).await);
+    temp_dir.close().expect("remove empty checkpoint test directory");
+}
+
+#[tokio::test]
+async fn deleted_checkpoint_is_not_recreated_by_an_old_manager() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    manager.cleanup().await.expect("delete checkpoint fixture");
+
+    let error = manager
+        .update_position(1, 2)
+        .await
+        .expect_err("an old manager must not resurrect a deleted checkpoint");
+    assert!(error.to_string().contains("removed after this manager saved it"));
+    assert!(!CheckpointManager::has_checkpoint(&disk, &task_id).await);
+    temp_dir.close().expect("remove deleted checkpoint test directory");
+}
+
+#[cfg(unix)]
+#[tokio::test]
+async fn checkpoint_cleanup_leaves_no_task_specific_lock_artifact() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    let lock_path = Path::new(BUCKET_META_PREFIX)
+        .join(format!("{task_id}_{RESUME_CHECKPOINT_FILE}"))
+        .with_extension("rustfs-cas.lock");
+    let lock_path = temp_dir.path().join(RUSTFS_META_BUCKET).join(lock_path);
+
+    manager.cleanup().await.expect("delete checkpoint fixture");
+
+    assert!(
+        !lock_path.exists(),
+        "successful checkpoint cleanup must not leave a task-specific lock artifact"
+    );
+}
+
+#[tokio::test]
+async fn an_empty_blocked_marker_still_blocks_resume_selection() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    let blocked_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_BLOCKED_FILE}");
+    disk.write_all(RUSTFS_META_BUCKET, &blocked_path, EcstoreDiskBytes::new())
+        .await
+        .expect("write empty blocked marker fixture");
+
+    assert!(CheckpointManager::is_blocked(&disk, &task_id).await);
+    assert!(CheckpointManager::is_resumable(&disk, &task_id).await.is_err());
+    // Recovery requires replacing/cleaning the snapshot, then removing the
+    // marker; ordinary selector retries are intentionally not an unlock path.
+    manager.cleanup().await.expect("clean blocked checkpoint");
+    assert!(!CheckpointManager::is_blocked(&disk, &task_id).await);
+    temp_dir.close().expect("remove empty blocked marker test directory");
+}
+
+#[tokio::test]
+async fn resumable_selector_skips_healthy_tasks_with_blocked_markers() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let tasks = [
+        (ResumeUtils::generate_task_id(), EcstoreDiskBytes::new()),
+        (ResumeUtils::generate_task_id(), EcstoreDiskBytes::from_static(b"blocked")),
+    ];
+    for (task_id, marker) in &tasks {
+        ResumeManager::new(
+            disk.clone(),
+            task_id.clone(),
+            "erasure_set".to_string(),
+            "pool_0_set_0".to_string(),
+            vec!["bucket".to_string()],
+        )
+        .await
+        .expect("create healthy resume state");
+        CheckpointManager::new(disk.clone(), task_id.clone())
+            .await
+            .expect("create healthy checkpoint");
+        let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+        let checkpoint_bytes = disk
+            .read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+            .await
+            .expect("read healthy checkpoint before blocking");
+        let marker_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_BLOCKED_FILE}");
+        disk.write_all(RUSTFS_META_BUCKET, &marker_path, marker.clone())
+            .await
+            .expect("write blocked marker");
+
+        assert!(ResumeUtils::get_resumable_tasks(&disk).await.is_err());
+        assert_eq!(
+            disk.read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+                .await
+                .expect("read healthy checkpoint after blocking"),
+            checkpoint_bytes
+        );
+    }
+    temp_dir.close().expect("remove blocked selector test directory");
+}
+
+#[tokio::test]
+async fn stale_checkpoint_manager_cannot_overwrite_newer_progress() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let first = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create first checkpoint manager");
+    let second = CheckpointManager::load_from_disk(disk.clone(), &task_id)
+        .await
+        .expect("load second checkpoint manager");
+
+    second
+        .update_position(4, 20)
+        .await
+        .expect("persist newer checkpoint progress");
+    let error = first
+        .update_position(1, 3)
+        .await
+        .expect_err("stale checkpoint manager must not overwrite newer progress");
+    assert!(error.to_string().contains("newer progress"));
+
+    let persisted = CheckpointManager::load_from_disk(disk.clone(), &task_id)
+        .await
+        .expect("load newer checkpoint progress")
+        .get_checkpoint()
+        .await;
+    assert_eq!(persisted.current_bucket_index, 4);
+    assert_eq!(persisted.current_object_index, 20);
+    temp_dir.close().expect("remove stale manager test directory");
+}
+
+#[tokio::test]
+async fn resumable_selector_isolates_future_and_corrupt_checkpoints() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let future_task = ResumeUtils::generate_task_id();
+    let corrupt_task = ResumeUtils::generate_task_id();
+    for task_id in [&future_task, &corrupt_task] {
+        ResumeManager::new(
+            disk.clone(),
+            task_id.to_string(),
+            "erasure_set".to_string(),
+            "pool_0_set_0".to_string(),
+            vec!["bucket".to_string()],
+        )
+        .await
+        .expect("create resumable state fixture");
+    }
+
+    let future_path = format!("{BUCKET_META_PREFIX}/{future_task}_{RESUME_CHECKPOINT_FILE}");
+    let mut future = ResumeCheckpoint::new(future_task.clone());
+    future.schema_version = CURRENT_CHECKPOINT_SCHEMA + 1;
+    let future_bytes = serde_json::to_vec(&future).expect("serialize future checkpoint fixture");
+    disk.write_all(RUSTFS_META_BUCKET, &future_path, future_bytes.clone().into())
+        .await
+        .expect("write future checkpoint fixture");
+    let corrupt_path = format!("{BUCKET_META_PREFIX}/{corrupt_task}_{RESUME_CHECKPOINT_FILE}");
+    let corrupt_bytes = b"{truncated";
+    disk.write_all(RUSTFS_META_BUCKET, &corrupt_path, corrupt_bytes.as_slice().into())
+        .await
+        .expect("write corrupt checkpoint fixture");
+
+    assert!(CheckpointManager::is_resumable(&disk, &future_task).await.is_err());
+    assert!(CheckpointManager::is_resumable(&disk, &corrupt_task).await.is_err());
+    assert!(ResumeUtils::get_resumable_tasks(&disk).await.is_err());
+    for (task_id, path, bytes) in [
+        (&future_task, future_path, future_bytes),
+        (&corrupt_task, corrupt_path, corrupt_bytes.to_vec()),
+    ] {
+        assert_eq!(
+            disk.read_all(RUSTFS_META_BUCKET, &path)
+                .await
+                .expect("read isolated checkpoint bytes"),
+            bytes
+        );
+        let blocked_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_BLOCKED_FILE}");
+        assert!(
+            !disk
+                .read_all(RUSTFS_META_BUCKET, &blocked_path)
+                .await
+                .expect("read checkpoint blocked marker")
+                .is_empty()
+        );
+    }
+    temp_dir.close().expect("remove selector isolation test directory");
+}
+
 #[test]
 fn test_persist_throttle_batches_until_threshold() {
     let mut throttle = PersistThrottle::new();

crates/heal/src/heal/resume/utils.rs

@@ -21,7 +21,7 @@ use uuid::Uuid;
 use super::super::{BUCKET_META_PREFIX, DiskError, DiskStore, HealDiskExt as _, RUSTFS_META_BUCKET};
 use super::replacement::{ReplacementPhase, ReplacementRecoveryRecord};
 use super::{
-    EVENT_HEAL_RESUME_STATE, LOG_COMPONENT_HEAL, LOG_SUBSYSTEM_RESUME, REPLACEMENT_COMPLETION_PROOF_FILE,
+    CheckpointManager, EVENT_HEAL_RESUME_STATE, LOG_COMPONENT_HEAL, LOG_SUBSYSTEM_RESUME, REPLACEMENT_COMPLETION_PROOF_FILE,
     REPLACEMENT_INTENT_FILE, RESUME_STATE_FILE, ResumeManager, ResumeStateFile, is_replacement_intent, path_to_str,
     replacement_recovery_corruption_for_state_load, replacement_recovery_dir, validate_resume_task_id,
 };
@@ -67,6 +67,7 @@ impl ResumeUtils {
                 // Extract task ID from filename: {task_id}_ahm_resume_state.json
                 if let Some(task_id) = entry.strip_suffix(&format!("_{RESUME_STATE_FILE}"))
                     && validate_resume_task_id(task_id).is_ok()
+                    && CheckpointManager::is_resumable(disk, task_id).await?
                 {
                     task_ids.push(task_id.to_string());
                 }

crates/kms/tests/vault_ha_failover_live.rs

@@ -16,14 +16,14 @@
 //!
 //! `scripts/test/vault_ha_kms_live.sh` owns the official Vault containers and
 //! kills the active node while this test continuously decrypts through a
-//! surviving standby. KV2 and Transit must recover after the bounded circuit
-//! interval, use a bounded number of attempts, and leave the circuit and
-//! in-flight gauges at zero after a new leader is elected.
+//! surviving standby. KV2 and Transit requests must remain successful, use a
+//! bounded number of attempts, and leave the circuit and in-flight gauges at
+//! zero after a new leader is elected.
 
 use std::collections::HashMap;
 use std::path::{Path, PathBuf};
+use std::sync::Arc;
 use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
-use std::sync::{Arc, Mutex};
 use std::time::Duration;
 
 use metrics_util::MetricKind;
@@ -43,11 +43,6 @@ const OPERATION_ATTEMPTS: &str = "rustfs_kms_backend_operation_attempts";
 const IN_FLIGHT: &str = "rustfs_kms_backend_in_flight";
 const CIRCUIT_OPEN: &str = "rustfs_kms_backend_circuit_open";
 const MAX_ATTEMPTS: u32 = 10;
-const ATTEMPT_TIMEOUT: Duration = Duration::from_secs(2);
-const HEALTHY_PROGRESS_TIMEOUT: Duration = Duration::from_secs(20);
-// The circuit remains open for 30s after five failed attempts.
-const POST_FAILOVER_PROGRESS_TIMEOUT: Duration = Duration::from_secs(35);
-const FAILOVER_ERROR_POLL_INTERVAL: Duration = Duration::from_millis(100);
 
 type MetricEntry = (
     metrics_util::CompositeKey,
@@ -69,7 +64,7 @@ fn config(backend: KmsBackend, backend_config: BackendConfig) -> KmsConfig {
         backend,
         backend_config,
         allow_insecure_dev_defaults: true,
-        timeout: ATTEMPT_TIMEOUT,
+        timeout: Duration::from_secs(2),
         retry_attempts: MAX_ATTEMPTS,
         enable_cache: false,
         ..KmsConfig::default()
@@ -169,31 +164,14 @@ fn retryable_failures(snapshot: &[MetricEntry], operation: &str) -> u64 {
         .sum()
 }
 
-async fn wait_for_count(
-    counter: &AtomicU64,
-    failure: &Mutex<Option<String>>,
-    minimum: u64,
-    description: &str,
-    timeout: Duration,
-) {
-    tokio::time::timeout(timeout, async {
+async fn wait_for_count(counter: &AtomicU64, minimum: u64, description: &str) {
+    tokio::time::timeout(Duration::from_secs(20), async {
         while counter.load(Ordering::SeqCst) < minimum {
-            if let Some(error) = failure.lock().expect("decrypt failure lock poisoned").as_ref() {
-                panic!(
-                    "{description} worker failed after {} successful decrypts: {error}",
-                    counter.load(Ordering::SeqCst)
-                );
-            }
             tokio::time::sleep(Duration::from_millis(25)).await;
         }
     })
     .await
-    .unwrap_or_else(|_| {
-        panic!(
-            "timed out after {timeout:?} waiting for {description}: completed {}, expected {minimum}",
-            counter.load(Ordering::SeqCst)
-        )
-    });
+    .unwrap_or_else(|_| panic!("timed out waiting for {description}"));
 }
 
 async fn wait_for_file(path: &Path, description: &str) {
@@ -211,8 +189,7 @@ async fn decrypt_loop<B: KmsBackendTrait + Send + Sync + 'static>(
     request: DecryptRequest,
     expected: Vec<u8>,
     completed: Arc<AtomicU64>,
-    allow_failover_errors: Arc<AtomicBool>,
-    failure: Arc<Mutex<Option<String>>>,
+    failed: Arc<AtomicBool>,
     stop: CancellationToken,
 ) {
     while !stop.is_cancelled() {
@@ -220,18 +197,8 @@ async fn decrypt_loop<B: KmsBackendTrait + Send + Sync + 'static>(
             Ok(response) if response.plaintext == expected => {
                 completed.fetch_add(1, Ordering::SeqCst);
             }
-            Ok(_) => {
-                *failure.lock().expect("decrypt failure lock poisoned") =
-                    Some("decrypt returned unexpected plaintext".to_string());
-                return;
-            }
-            Err(rustfs_kms::KmsError::BackendError { .. } | rustfs_kms::KmsError::OperationTimedOut { .. })
-                if allow_failover_errors.load(Ordering::SeqCst) =>
-            {
-                tokio::time::sleep(FAILOVER_ERROR_POLL_INTERVAL).await;
-            }
-            Err(error) => {
-                *failure.lock().expect("decrypt failure lock poisoned") = Some(error.to_string());
+            Ok(_) | Err(_) => {
+                failed.store(true, Ordering::SeqCst);
                 return;
             }
         }
@@ -329,9 +296,7 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
     );
 
     let stop = CancellationToken::new();
-    let allow_failover_errors = Arc::new(AtomicBool::new(false));
-    let kv2_failure = Arc::new(Mutex::new(None));
-    let transit_failure = Arc::new(Mutex::new(None));
+    let failed = Arc::new(AtomicBool::new(false));
     let kv2_completed = Arc::new(AtomicU64::new(0));
     let transit_completed = Arc::new(AtomicU64::new(0));
     let kv2_worker = tokio::spawn(decrypt_loop(
@@ -339,8 +304,7 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
         kv2_request,
         kv2_data_key.plaintext_key,
         Arc::clone(&kv2_completed),
-        Arc::clone(&allow_failover_errors),
-        Arc::clone(&kv2_failure),
+        Arc::clone(&failed),
         stop.clone(),
     ));
     let transit_worker = tokio::spawn(decrypt_loop(
@@ -348,21 +312,12 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
         transit_request,
         transit_data_key.plaintext_key,
         Arc::clone(&transit_completed),
-        Arc::clone(&allow_failover_errors),
-        Arc::clone(&transit_failure),
+        Arc::clone(&failed),
         stop.clone(),
     ));
 
-    wait_for_count(&kv2_completed, &kv2_failure, 2, "two healthy KV2 decrypts", HEALTHY_PROGRESS_TIMEOUT).await;
-    wait_for_count(
-        &transit_completed,
-        &transit_failure,
-        2,
-        "two healthy Transit decrypts",
-        HEALTHY_PROGRESS_TIMEOUT,
-    )
-    .await;
-    allow_failover_errors.store(true, Ordering::SeqCst);
+    wait_for_count(&kv2_completed, 2, "two healthy KV2 decrypts").await;
+    wait_for_count(&transit_completed, 2, "two healthy Transit decrypts").await;
     std::fs::write(&marker, b"ready").expect("publish failover readiness marker");
 
     wait_for_file(&elected, "the replacement Vault leader").await;
@@ -371,39 +326,18 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
 
     let kv2_after_election = kv2_completed.load(Ordering::SeqCst) + 2;
     let transit_after_election = transit_completed.load(Ordering::SeqCst) + 2;
-    wait_for_count(
-        &kv2_completed,
-        &kv2_failure,
-        kv2_after_election,
-        "post-failover KV2 decrypts",
-        POST_FAILOVER_PROGRESS_TIMEOUT,
-    )
-    .await;
-    wait_for_count(
-        &transit_completed,
-        &transit_failure,
-        transit_after_election,
-        "post-failover Transit decrypts",
-        POST_FAILOVER_PROGRESS_TIMEOUT,
-    )
-    .await;
+    wait_for_count(&kv2_completed, kv2_after_election, "post-failover KV2 decrypts").await;
+    wait_for_count(&transit_completed, transit_after_election, "post-failover Transit decrypts").await;
 
     stop.cancel();
     kv2_worker.await.expect("KV2 decrypt worker must join");
     transit_worker.await.expect("Transit decrypt worker must join");
-    assert!(
-        kv2_failure.lock().expect("KV2 failure lock poisoned").is_none(),
-        "no KV2 decrypt may fail or return different plaintext"
-    );
-    assert!(
-        transit_failure.lock().expect("Transit failure lock poisoned").is_none(),
-        "no Transit decrypt may fail or return different plaintext"
-    );
+    assert!(!failed.load(Ordering::SeqCst), "no decrypt may fail or return different plaintext");
 }
 
 #[test]
 #[ignore = "requires a real three-node Vault Raft cluster; run scripts/test/vault_ha_kms_live.sh"]
-fn vault_raft_leader_failure_recovers_kv2_and_transit_decrypts() {
+fn vault_raft_leader_failure_preserves_kv2_and_transit_decrypts() {
     let recorder = DebuggingRecorder::new();
     let snapshotter = recorder.snapshotter();
     metrics::with_local_recorder(&recorder, || {
@@ -415,6 +349,11 @@ fn vault_raft_leader_failure_recovers_kv2_and_transit_decrypts() {
     });
     let snapshot = snapshotter.snapshot().into_vec();
 
+    assert_eq!(
+        counter_value(&snapshot, OPERATIONS_TOTAL, &[("outcome", "circuit_open")]),
+        0,
+        "a bounded leader election must not open the circuit"
+    );
     assert_eq!(
         counter_value(&snapshot, OPERATIONS_TOTAL, &[("outcome", "budget_exhausted")]),
         0,

crates/scanner/src/data_usage_define.rs

@@ -29,8 +29,7 @@ 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, PrefixUsageEntry,
-    PrefixUsageQuery, PrefixUsageSummary, ReplTargetSizeSummary, SizeSummary, TierAccountingProof, TierStats, UNKNOWN_TIER,
-    UNKNOWN_TIER_DIAGNOSTIC_BYTE_CAP, UNKNOWN_TIER_DIAGNOSTIC_ENTRY_CAP, UnknownTierStats, hash_path, prefix_usage_in_cache,
+    PrefixUsageQuery, PrefixUsageSummary, ReplTargetSizeSummary, SizeSummary, TierStats, hash_path, prefix_usage_in_cache,
 };
 use rustfs_utils::path::{SLASH_SEPARATOR, path_join_buf};
 use tokio::time::{Duration, Instant, sleep, timeout};
@@ -206,79 +205,25 @@ impl ScannerSizeSummaryExt for SizeSummary {
             self.versions = self.versions.saturating_add(1);
         }
 
-        let logical_size = size.max(0);
-        let size = usize::try_from(logical_size).unwrap_or(usize::MAX);
+        let size = usize::try_from(size.max(0)).unwrap_or(usize::MAX);
         self.total_size = self.total_size.saturating_add(size);
-        let logical_bytes = u64::try_from(logical_size).unwrap_or(u64::MAX);
-        let physical_bytes = u64::try_from(oi.size.max(0)).unwrap_or(0);
-        let mut proof = TierAccountingProof {
-            logical_total: logical_bytes,
-            logical_known: 0,
-            physical_total: physical_bytes,
-            physical_known: 0,
-            overflowed: false,
-        };
 
         if oi.transitioned_object.free_version {
-            proof.logical_known = logical_bytes;
-            proof.physical_known = physical_bytes;
-            self.tier_accounting_proof.saturating_add(proof);
             return;
         }
 
-        let tier = if oi.transitioned_object.status == TRANSITION_COMPLETE {
-            oi.transitioned_object.tier.as_str()
-        } else {
-            oi.storage_class.as_deref().unwrap_or(storageclass::STANDARD)
-        };
-
-        let builtin_tier = tier == storageclass::STANDARD || tier == storageclass::RRS;
-        let tier_registry_is_empty =
-            self.tier_stats.is_empty() || (self.tier_stats.len() == 1 && self.tier_stats.contains_key(UNKNOWN_TIER));
-        let known_tier = tier != UNKNOWN_TIER && (builtin_tier || self.tier_stats.contains_key(tier));
-
-        // With no configured tier, retain the historical empty-map shape for
-        // ordinary STANDARD/RRS objects. A non-built-in key is still an
-        // observable unknown and must create only the fixed bucket.
-        if tier_registry_is_empty && known_tier {
-            proof.logical_known = logical_bytes;
-            proof.physical_known = physical_bytes;
-            self.tier_accounting_proof.saturating_add(proof);
-            return;
+        let mut tier = oi.storage_class.clone().unwrap_or_else(|| storageclass::STANDARD.to_string());
+        if oi.transitioned_object.status == TRANSITION_COMPLETE {
+            tier = oi.transitioned_object.tier.clone();
         }
 
-        // Configured tiers and the fixed bucket are normally seeded, so the
-        // hot path can mutate them without allocating a key for every object.
-        // The fallback inserts only when a legacy/no-config summary sees its
-        // first unknown key.
-        let tier_stats = if known_tier {
-            if let Some(stats) = self.tier_stats.get_mut(tier) {
-                stats
-            } else {
-                self.tier_stats.entry(tier.to_owned()).or_default()
-            }
-        } else if let Some(stats) = self.tier_stats.get_mut(UNKNOWN_TIER) {
-            stats
-        } else {
-            self.tier_stats.entry(UNKNOWN_TIER.to_string()).or_default()
-        };
-        *tier_stats = tier_stats.add(&TierStats {
-            total_size: physical_bytes,
-            num_versions: 1,
-            num_objects: u64::from(oi.is_latest),
-        });
-        if known_tier {
-            proof.logical_known = logical_bytes;
-            proof.physical_known = physical_bytes;
+        if let Some(tier_stats) = self.tier_stats.get_mut(&tier) {
+            *tier_stats = tier_stats.add(&TierStats {
+                total_size: u64::try_from(oi.size).unwrap_or(0),
+                num_versions: 1,
+                num_objects: u64::from(oi.is_latest),
+            });
         }
-        if !known_tier {
-            self.unknown_tier_stats
-                .record_dimensions(tier, logical_bytes, physical_bytes, 1, u64::from(oi.is_latest));
-            if self.unknown_tier_stats.counter_overflowed {
-                proof.overflowed = true;
-            }
-        }
-        self.tier_accounting_proof.saturating_add(proof);
     }
 }
 
@@ -326,11 +271,6 @@ pub struct DataUsageEntryInfo {
     pub name: String,
     pub parent: String,
     pub entry: DataUsageEntry,
-    /// Registry generation used to classify this root entry. Older remote
-    /// workers omit it; callers must reject that result when a frozen cycle
-    /// requires generation fencing.
-    #[serde(default, skip_serializing_if = "Option::is_none")]
-    pub tier_registry_generation: Option<u64>,
 }
 
 #[derive(Clone, Copy, Debug, Serialize, Deserialize, PartialEq, Eq, Hash)]
@@ -404,10 +344,6 @@ pub struct DataUsageCacheInfo {
     pub scan_plan_digest: Option<DataUsageScanPlanDigest>,
     #[serde(default)]
     pub cache_key_format: u16,
-    /// Registry generation used for the completed/partial scan. This is
-    /// process-local audit data; older cache writers omit it.
-    #[serde(default, skip_serializing_if = "Option::is_none")]
-    pub tier_registry_generation: Option<u64>,
 }
 
 impl Serialize for DataUsageCacheInfo {
@@ -417,8 +353,7 @@ impl Serialize for DataUsageCacheInfo {
     {
         // Keep this metadata map-encoded so older readers can ignore fields
         // appended by newer scanner versions during rolling upgrades.
-        let field_count = 16 + usize::from(self.tier_registry_generation.is_some());
-        let mut state = serializer.serialize_map(Some(field_count))?;
+        let mut state = serializer.serialize_map(Some(16))?;
         state.serialize_entry("name", &self.name)?;
         state.serialize_entry("next_cycle", &self.next_cycle)?;
         state.serialize_entry("leader_epoch", &self.leader_epoch)?;
@@ -435,9 +370,6 @@ impl Serialize for DataUsageCacheInfo {
         state.serialize_entry("snapshot_complete", &self.snapshot_complete)?;
         state.serialize_entry("scan_plan_digest", &self.scan_plan_digest)?;
         state.serialize_entry("cache_key_format", &self.cache_key_format)?;
-        if let Some(generation) = self.tier_registry_generation {
-            state.serialize_entry("tier_registry_generation", &generation)?;
-        }
         state.end()
     }
 }
@@ -458,58 +390,6 @@ pub(crate) enum DataUsageCachePrepareOutcome {
 }
 
 impl DataUsageCache {
-    /// Reconcile tier keys loaded from an older cache against the registry
-    /// frozen for this scan. New metadata is already routed through
-    /// `UNKNOWN_TIER`; this pass handles retired keys that predate that rule.
-    /// Legacy `TierStats` carries physical bytes only, so this migration does
-    /// not manufacture a logical unknown-byte value from that physical total.
-    pub(crate) fn fold_retired_tiers(&mut self, tier_names: &[String]) {
-        let known_tiers = tier_names.iter().map(String::as_str).collect::<HashSet<_>>();
-        for entry in self.cache.values_mut() {
-            let Some(tiers) = entry.all_tier_stats.as_mut() else { continue };
-            let existing_unknown = tiers.tiers.get(UNKNOWN_TIER).cloned().unwrap_or_default();
-            let companion_present = entry.unknown_tier_stats.as_ref().is_some_and(|stats| !stats.is_empty());
-            let migrate_existing_unknown = !companion_present;
-            let mut retired = TierStats::default();
-            let mut retired_key_found = false;
-            if migrate_existing_unknown {
-                retired = retired.add(&existing_unknown);
-            }
-            for (tier, stats) in &tiers.tiers {
-                if tier != UNKNOWN_TIER
-                    && tier != storageclass::STANDARD
-                    && tier != storageclass::RRS
-                    && !known_tiers.contains(tier.as_str())
-                {
-                    retired_key_found = true;
-                    retired = retired.add(stats);
-                }
-            }
-            tiers.fold_unknown_tiers(tier_names.iter().map(String::as_str));
-            if !retired.is_empty() && !companion_present {
-                entry.add_unknown_tier_stats(&UnknownTierStats {
-                    // The legacy map stores physical bytes only. Logical
-                    // bytes remain zero until a fresh object scan observes
-                    // them under the current metadata format.
-                    unknown_physical_bytes: retired.total_size,
-                    unknown_objects: retired.num_objects,
-                    unknown_versions: retired.num_versions,
-                    ..Default::default()
-                });
-                // The legacy tier map has no logical-byte dimension, so a
-                // proof that classified this retired key as known cannot be
-                // repaired safely. Mark it unvalidated and require a fresh
-                // scan rather than guessing a logical subtraction.
-                entry.tier_accounting_proof = None;
-            } else if retired_key_found {
-                // A nonempty companion has no provenance tying it to the
-                // retired map keys. Reject the mixed cache until a fresh scan
-                // reconciles the dimensions instead of double-counting them.
-                entry.tier_accounting_proof = None;
-            }
-        }
-    }
-
     /// Prefix-level usage query over this (writer-side) cache; see
     /// [`prefix_usage_in_cache`] for the semantics
     /// (rustfs/backlog#1872).
@@ -995,7 +875,6 @@ impl DataUsageCache {
             delete_markers_total_count: flat.delete_markers as u64,
             objects_total_size: flat.size as u64,
             tier_stats: flat.all_tier_stats.filter(|tiers| !tiers.is_empty()),
-            unknown_tier_stats: flat.unknown_tier_stats.filter(|stats| !stats.is_empty()),
             buckets_count: u64::try_from(buckets.len()).unwrap_or(u64::MAX),
             buckets_usage,
             ..Default::default()

crates/scanner/src/data_usage_define/tests.rs

@@ -573,6 +573,7 @@ fn size_summary_add_saturates_all_usage_counters() {
             failed_count: usize::MAX,
         },
     );
+
     let mut increment = SizeSummary {
         total_size: 1,
         versions: 1,
@@ -587,24 +588,6 @@ fn size_summary_add_saturates_all_usage_counters() {
         failed_count: 1,
         ..Default::default()
     };
-    summary.tier_stats.insert(
-        UNKNOWN_TIER.to_string(),
-        TierStats {
-            total_size: u64::MAX,
-            num_versions: u64::MAX,
-            num_objects: u64::MAX,
-        },
-    );
-    increment.tier_stats.insert(
-        UNKNOWN_TIER.to_string(),
-        TierStats {
-            total_size: 1,
-            num_versions: 1,
-            num_objects: 1,
-        },
-    );
-    increment.unknown_tier_stats.unknown_bytes = 1;
-    increment.unknown_tier_stats.unknown_physical_bytes = 1;
     increment.repl_target_stats.insert(
         target.clone(),
         ReplTargetSizeSummary {
@@ -641,8 +624,6 @@ fn size_summary_add_saturates_all_usage_counters() {
     assert_eq!(target_summary.failed_size, i64::MAX);
     assert_eq!(target_summary.pending_count, usize::MAX);
     assert_eq!(target_summary.failed_count, usize::MAX);
-    assert_eq!(summary.tier_stats[UNKNOWN_TIER].total_size, u64::MAX);
-    assert_eq!(summary.unknown_tier_stats.unknown_bytes, 1);
 }
 
 #[test]
@@ -692,249 +673,6 @@ fn size_summary_actions_accounting_accumulates_tier_stats() {
     );
 }
 
-#[test]
-fn unknown_tier_is_bounded_and_accounted() {
-    let mut summary = SizeSummary::new();
-    summary.tier_stats.insert("WARM".to_string(), TierStats::default());
-    let object = ObjectInfo {
-        storage_class: Some("retired-tier".to_string()),
-        size: 11,
-        is_latest: true,
-        ..Default::default()
-    };
-
-    summary.actions_accounting(&object, 11, 11);
-
-    assert_eq!(summary.tier_stats.len(), 2);
-    assert_eq!(summary.tier_stats.get(UNKNOWN_TIER).map(|stats| stats.total_size), Some(11));
-    assert_eq!(summary.unknown_tier_stats.unknown_bytes, 11);
-    assert_eq!(summary.unknown_tier_stats.unknown_physical_bytes, 11);
-    assert_eq!(summary.unknown_tier_stats.unknown_objects, 1);
-    assert_eq!(summary.tier_accounting_proof.logical_total, 11);
-    assert_eq!(summary.tier_accounting_proof.logical_known, 0);
-    assert_eq!(summary.tier_accounting_proof.physical_total, 11);
-    assert_eq!(summary.tier_accounting_proof.physical_known, 0);
-    assert!(summary.unknown_tier_stats.diagnostics.len() <= UNKNOWN_TIER_DIAGNOSTIC_ENTRY_CAP);
-    assert!(summary.unknown_tier_stats.diagnostics.iter().map(String::len).sum::<usize>() <= UNKNOWN_TIER_DIAGNOSTIC_BYTE_CAP);
-    assert!(
-        summary
-            .unknown_tier_stats
-            .diagnostics
-            .iter()
-            .all(|entry| !entry.contains("retired"))
-    );
-}
-
-#[test]
-fn unknown_tier_is_accounted_when_no_remote_tier_is_configured() {
-    let mut summary = SizeSummary::new();
-    let object = ObjectInfo {
-        storage_class: Some("retired-tier".to_string()),
-        size: 3,
-        is_latest: true,
-        ..Default::default()
-    };
-
-    summary.actions_accounting(&object, 9, 9);
-
-    assert_eq!(summary.tier_stats.len(), 1);
-    assert_eq!(summary.tier_stats[UNKNOWN_TIER].total_size, 3);
-    assert_eq!(summary.unknown_tier_stats.unknown_bytes, 9);
-    assert_eq!(summary.unknown_tier_stats.unknown_physical_bytes, 3);
-
-    let standard = ObjectInfo {
-        storage_class: Some(storageclass::STANDARD.to_string()),
-        size: 4,
-        is_latest: true,
-        ..Default::default()
-    };
-    summary.actions_accounting(&standard, 4, 4);
-    assert_eq!(summary.tier_accounting_proof.logical_total, 13);
-    assert_eq!(summary.tier_accounting_proof.logical_known, 4);
-    assert_eq!(summary.tier_accounting_proof.physical_total, 3 + 4);
-    assert_eq!(summary.tier_accounting_proof.physical_known, 4);
-    assert_eq!(summary.tier_stats.len(), 1, "built-ins preserve the no-tier map shape");
-}
-
-#[test]
-fn million_unique_tier_keys_do_not_grow_stats_map() {
-    let mut summary = SizeSummary::new();
-    summary.tier_stats.insert("WARM".to_string(), TierStats::default());
-    for index in 0..1_000_000_u64 {
-        let object = ObjectInfo {
-            storage_class: Some(format!("untrusted-tier-{index}")),
-            size: 1,
-            ..Default::default()
-        };
-        summary.actions_accounting(&object, 1, 1);
-    }
-
-    assert_eq!(summary.tier_stats.len(), 2);
-    assert_eq!(summary.tier_stats[UNKNOWN_TIER].total_size, 1_000_000);
-    assert_eq!(summary.unknown_tier_stats.unknown_bytes, 1_000_000);
-    assert!(summary.unknown_tier_stats.diagnostics.len() <= UNKNOWN_TIER_DIAGNOSTIC_ENTRY_CAP);
-    assert!(summary.unknown_tier_stats.diagnostics.iter().map(String::len).sum::<usize>() <= UNKNOWN_TIER_DIAGNOSTIC_BYTE_CAP);
-}
-
-#[test]
-fn unknown_tier_never_triggers_transition() {
-    let mut summary = SizeSummary::new();
-    summary.tier_stats.insert("WARM".to_string(), TierStats::default());
-    let mut object = ObjectInfo {
-        storage_class: Some("removed-tier".to_string()),
-        size: 7,
-        ..Default::default()
-    };
-    object.transitioned_object.status = TRANSITION_COMPLETE.to_string();
-    object.transitioned_object.tier = "removed-tier".to_string();
-
-    summary.actions_accounting(&object, 7, 7);
-
-    assert_eq!(summary.tier_stats.get("removed-tier"), None);
-    assert_eq!(summary.tier_stats[UNKNOWN_TIER].total_size, 7);
-}
-
-#[test]
-fn removed_tier_survives_restart_as_unknown() {
-    let mut summary = SizeSummary::new();
-    summary.tier_stats.insert("COLD".to_string(), TierStats::default());
-    summary.tier_stats.insert(
-        "RETIRED".to_string(),
-        TierStats {
-            total_size: 5,
-            num_versions: 1,
-            num_objects: 1,
-        },
-    );
-    let object = ObjectInfo {
-        storage_class: Some("COLD".to_string()),
-        size: 5,
-        ..Default::default()
-    };
-    summary.actions_accounting(&object, 5, 5);
-    let mut entry = DataUsageEntry::default();
-    entry.add_tier_sizes(&summary.tier_stats);
-    entry.add_unknown_tier_stats(&UnknownTierStats {
-        unknown_bytes: 2,
-        unknown_physical_bytes: 2,
-        unknown_objects: 1,
-        unknown_versions: 1,
-        ..Default::default()
-    });
-    let encoded = rmp_serde::to_vec(&entry).expect("entry should encode");
-    let restored: DataUsageEntry = rmp_serde::from_slice(&encoded).expect("entry should decode");
-    assert_eq!(restored.unknown_tier_stats.as_ref().map(|stats| stats.unknown_bytes), Some(2));
-    assert_eq!(
-        restored.all_tier_stats.as_ref().expect("tier stats persisted").tiers["RETIRED"].total_size,
-        5
-    );
-
-    let mut cache = DataUsageCache::default();
-    cache.replace("bucket", "", restored);
-    cache.fold_retired_tiers(&["COLD".to_string()]);
-    let folded = cache.cache.get(&hash_path("bucket").key()).expect("folded cache entry");
-    assert_eq!(
-        folded.all_tier_stats.as_ref().expect("tier stats persisted").tiers[UNKNOWN_TIER].total_size,
-        5
-    );
-    assert_eq!(folded.unknown_tier_stats.as_ref().map(|stats| stats.unknown_bytes), Some(2));
-}
-
-#[test]
-fn retired_tier_fold_is_idempotent_and_rejects_mixed_companion_provenance() {
-    let mut cache = DataUsageCache::default();
-    let mut entry = DataUsageEntry {
-        all_tier_stats: Some(AllTierStats {
-            tiers: HashMap::from([(
-                "RETIRED".to_string(),
-                TierStats {
-                    total_size: 5,
-                    num_versions: 1,
-                    num_objects: 1,
-                },
-            )]),
-        }),
-        ..Default::default()
-    };
-    entry.unknown_tier_stats = Some(UnknownTierStats {
-        unknown_physical_bytes: 5,
-        ..Default::default()
-    });
-    entry.tier_accounting_proof = Some(TierAccountingProof {
-        physical_total: 5,
-        physical_known: 5,
-        ..Default::default()
-    });
-    cache.replace("bucket", "", entry);
-
-    cache.fold_retired_tiers(&["COLD".to_string()]);
-    let first = cache.cache.get(&hash_path("bucket").key()).expect("entry").clone();
-    assert_eq!(first.all_tier_stats.as_ref().expect("tiers").tiers[UNKNOWN_TIER].total_size, 5);
-    assert_eq!(first.unknown_tier_stats.as_ref().expect("companion").unknown_physical_bytes, 5);
-    assert!(first.tier_accounting_proof.is_none(), "mixed provenance must not publish");
-
-    cache.fold_retired_tiers(&["COLD".to_string()]);
-    let second = cache.cache.get(&hash_path("bucket").key()).expect("entry");
-    assert_eq!(second.all_tier_stats.as_ref().expect("tiers").tiers[UNKNOWN_TIER].total_size, 5);
-    assert_eq!(second.unknown_tier_stats.as_ref().expect("companion").unknown_physical_bytes, 5);
-}
-
-#[test]
-fn tier_registry_refresh_does_not_mix_cycle_generations() {
-    let first = crate::TierRegistrySnapshot {
-        generation: 1,
-        names: Arc::from(["WARM".to_string()]),
-        refresh_failed: false,
-    };
-    let second = crate::TierRegistrySnapshot {
-        generation: 2,
-        names: Arc::from(["COLD".to_string()]),
-        refresh_failed: false,
-    };
-    assert_ne!(first.generation, second.generation);
-    assert_eq!(first.names.as_ref(), ["WARM".to_string()]);
-    assert_eq!(second.names.as_ref(), ["COLD".to_string()]);
-    assert!(first.refreshed(Err(())).refresh_failed);
-    assert!(!second.refreshed(Ok(Arc::from(["HOT".to_string()]))).refresh_failed);
-    assert_eq!(first.refreshed(Err(())).generation, first.generation);
-    assert_eq!(first.refreshed(Err(())).names, first.names);
-}
-
-#[test]
-fn unknown_tier_counter_uses_checked_arithmetic() {
-    let max = TierStats {
-        total_size: u64::MAX,
-        num_versions: u64::MAX,
-        num_objects: u64::MAX,
-    };
-    assert!(max.checked_add(&TierStats::default()).is_some());
-    assert!(
-        max.checked_add(&TierStats {
-            total_size: 1,
-            ..Default::default()
-        })
-        .is_none()
-    );
-
-    let mut unknown = UnknownTierStats {
-        unknown_bytes: u64::MAX,
-        ..Default::default()
-    };
-    unknown.record("overflow", 1, 1, 1);
-    assert_eq!(unknown.unknown_bytes, u64::MAX);
-    assert_eq!(unknown.unknown_objects, 1);
-    assert!(unknown.counter_overflowed);
-    assert!(unknown.checked_add(&UnknownTierStats::default()).is_none());
-    assert!(
-        unknown
-            .checked_add(&UnknownTierStats {
-                unknown_bytes: 1,
-                ..Default::default()
-            })
-            .is_none()
-    );
-}
-
 #[test]
 fn test_data_usage_entry_merge_sums_failed_objects() {
     let mut left = DataUsageEntry {

crates/scanner/src/lib.rs

@@ -24,11 +24,8 @@
 use bytes::Bytes;
 use http::HeaderMap;
 use rustfs_config::server_config::{Config as ServerConfig, get_global_server_config as config_get_global_server_config};
-use sha2::{Digest as _, Sha256};
-use std::collections::HashMap;
 use std::path::PathBuf;
 use std::sync::Arc;
-use std::sync::LazyLock;
 use std::sync::RwLock;
 use std::time::{Duration, Instant};
 use storage_api::owner::{
@@ -97,45 +94,6 @@ static SCANNER_RUNTIME_INSTANCES: AtomicU64 = AtomicU64::new(0);
 static SCANNER_FOREGROUND_READ_ACTIVITY: AtomicU64 = AtomicU64::new(0);
 static SCANNER_FOREGROUND_STREAM_READS: AtomicU64 = AtomicU64::new(0);
 
-/// Immutable tier registry captured at the beginning of a folder scan.
-/// Generation makes it possible to prove that a result was classified against
-/// one registry even when the process-wide TTL cache refreshes concurrently.
-#[derive(Clone, Debug, PartialEq, Eq)]
-pub(crate) struct TierRegistrySnapshot {
-    pub(crate) generation: u64,
-    pub(crate) names: Arc<[String]>,
-    /// True when the last refresh attempt failed and `names` is therefore a
-    /// retained last-good snapshot rather than a newly read registry.
-    pub(crate) refresh_failed: bool,
-}
-
-impl TierRegistrySnapshot {
-    /// Apply a refresh only when the registry read succeeds. A failed refresh
-    /// retains the prior generation, preventing a transient config failure
-    /// from classifying the remainder of a scan against an empty registry.
-    pub(crate) fn refreshed(&self, names: Result<Arc<[String]>, ()>) -> Self {
-        match names {
-            Ok(names) => Self {
-                generation: self.generation.saturating_add(1),
-                names,
-                refresh_failed: false,
-            },
-            Err(()) => Self {
-                refresh_failed: true,
-                ..self.clone()
-            },
-        }
-    }
-
-    pub(crate) fn initial(names: Arc<[String]>) -> Self {
-        Self {
-            generation: 1,
-            names,
-            refresh_failed: false,
-        }
-    }
-}
-
 pub fn current_scanner_activity() -> u64 {
     SCANNER_ACTIVE_WORK_UNITS.load(Ordering::Relaxed)
 }
@@ -413,7 +371,6 @@ pub(crate) fn resolve_scanner_server_config() -> Option<ServerConfig> {
 /// How long the scanner caches the runtime tier-name list before re-reading
 /// the tier configuration manager.
 const TIER_NAME_CACHE_TTL: Duration = Duration::from_secs(30);
-const MAX_TIER_REGISTRY_NAME_BYTES: usize = 256;
 
 /// Process-wide TTL cache of runtime tier names.
 ///
@@ -426,192 +383,24 @@ const MAX_TIER_REGISTRY_NAME_BYTES: usize = 256;
 /// `TIER_NAME_CACHE_TTL` later; a removed tier can leave an all-zero
 /// `TierStats` seed behind for one cache generation, which merges harmlessly
 /// by key in per-object accounting and disappears on the next refresh.
-static TIER_NAME_CACHE: RwLock<Option<(Instant, TierRegistrySnapshot)>> = RwLock::new(None);
-static TIER_REGISTRY_GENERATION: AtomicU64 = AtomicU64::new(0);
-static TIER_CYCLE_SNAPSHOTS: LazyLock<RwLock<HashMap<(u64, u64), TierRegistrySnapshot>>> =
-    LazyLock::new(|| RwLock::new(HashMap::new()));
-static TIER_ACTIVE_CYCLES: LazyLock<RwLock<HashMap<(u64, u64), usize>>> = LazyLock::new(|| RwLock::new(HashMap::new()));
-static TIER_NAME_REFRESH_LOCK: LazyLock<tokio::sync::Mutex<()>> = LazyLock::new(|| tokio::sync::Mutex::new(()));
-
-/// Return one immutable registry snapshot for a scanner unit of work.
-pub(crate) async fn runtime_tier_registry() -> TierRegistrySnapshot {
-    {
-        let cached = TIER_NAME_CACHE.read().unwrap_or_else(|err| err.into_inner()).clone();
-        if let Some((refreshed_at, snapshot)) = cached
-            && refreshed_at.elapsed() < TIER_NAME_CACHE_TTL
-        {
-            return snapshot;
-        }
-    }
-
-    // Serialize refreshes so a slower read of the old config cannot overwrite
-    // a newer snapshot published by a concurrent caller.
-    let _refresh_guard = TIER_NAME_REFRESH_LOCK.lock().await;
-    {
-        let cached = TIER_NAME_CACHE.read().unwrap_or_else(|err| err.into_inner()).clone();
-        if let Some((refreshed_at, snapshot)) = cached
-            && refreshed_at.elapsed() < TIER_NAME_CACHE_TTL
-        {
-            return snapshot;
-        }
-    }
-
-    let previous = TIER_NAME_CACHE
-        .read()
-        .unwrap_or_else(|err| err.into_inner())
-        .as_ref()
-        .map(|(_, snapshot)| snapshot.clone());
-    let names = ecstore_get_global_tier_config_mgr()
-        .read()
-        .await
-        .list_tiers()
-        .into_iter()
-        .map(|tier| tier.name)
-        .collect::<Vec<_>>();
-    let snapshot = match validate_tier_registry_names(names) {
-        Ok(names) => {
-            let generation = next_tier_registry_generation();
-            match previous {
-                Some(previous) => TierRegistrySnapshot {
-                    generation,
-                    ..previous.refreshed(Ok(names))
-                },
-                None => TierRegistrySnapshot {
-                    generation,
-                    ..TierRegistrySnapshot::initial(names)
-                },
-            }
-        }
-        Err(()) => match previous {
-            Some(previous) => previous.refreshed(Err(())),
-            None => TierRegistrySnapshot {
-                generation: next_tier_registry_generation(),
-                names: Arc::new([]),
-                refresh_failed: true,
-            },
-        },
-    };
-    *TIER_NAME_CACHE.write().unwrap_or_else(|err| err.into_inner()) = Some((Instant::now(), snapshot.clone()));
-    snapshot
-}
-
-fn next_tier_registry_generation() -> u64 {
-    TIER_REGISTRY_GENERATION
-        .fetch_update(Ordering::AcqRel, Ordering::Relaxed, |current| Some(current.saturating_add(1)))
-        .unwrap_or(u64::MAX)
-}
-
-fn validate_tier_registry_names(mut names: Vec<String>) -> Result<Arc<[String]>, ()> {
-    if names.iter().any(|name| {
-        name.is_empty()
-            || name.len() > MAX_TIER_REGISTRY_NAME_BYTES
-            || name.bytes().any(|byte| byte.is_ascii_control())
-            || name == UNKNOWN_TIER
-            || name == storageclass::STANDARD
-            || name == storageclass::RRS
-    }) {
-        return Err(());
-    }
-    names.sort_unstable();
-    if names.windows(2).any(|pair| pair[0] == pair[1]) {
-        return Err(());
-    }
-    Ok(names.into())
-}
+static TIER_NAME_CACHE: RwLock<Option<(Instant, Arc<[String]>)>> = RwLock::new(None);
 
 /// Tier names currently registered in the tier configuration, cached for
 /// `TIER_NAME_CACHE_TTL`.
 pub(crate) async fn runtime_tier_names() -> Arc<[String]> {
-    runtime_tier_registry().await.names
-}
-
-/// Return the immutable tier registry for one scanner cycle/leader pair.
-/// Different buckets and disks belonging to the same cycle share this entry,
-/// so a TTL refresh cannot split one published cycle across generations.
-pub(crate) async fn runtime_tier_registry_for_cycle(cycle: u64, leader_epoch: u64) -> TierRegistrySnapshot {
-    let key = (cycle, leader_epoch);
-    prune_inactive_tier_cycle_snapshots(cycle, leader_epoch);
     {
-        let cached = TIER_CYCLE_SNAPSHOTS.read().unwrap_or_else(|err| err.into_inner());
-        if let Some(snapshot) = cached.get(&key) {
-            return snapshot.clone();
+        let cached = TIER_NAME_CACHE.read().unwrap_or_else(|err| err.into_inner()).clone();
+        if let Some((refreshed_at, names)) = cached
+            && refreshed_at.elapsed() < TIER_NAME_CACHE_TTL
+        {
+            return names;
         }
     }
 
-    let mut snapshot = runtime_tier_registry().await;
-    // The registry generation describes the configuration snapshot, not the
-    // scan that consumed it. Keep it stable across cycles so a healthy cache
-    // can be reused; cycle and leader fencing are carried separately by the
-    // cache metadata and scan plan.
-    snapshot.generation = tier_registry_generation(&snapshot.names);
-    let mut cached = TIER_CYCLE_SNAPSHOTS.write().unwrap_or_else(|err| err.into_inner());
-    if let Some(existing) = cached.get(&key) {
-        return existing.clone();
-    }
-    cached.insert(key, snapshot.clone());
-    snapshot
-}
-
-fn prune_inactive_tier_cycle_snapshots(cycle: u64, leader_epoch: u64) {
-    let active = TIER_ACTIVE_CYCLES.read().unwrap_or_else(|err| err.into_inner());
-    let mut snapshots = TIER_CYCLE_SNAPSHOTS.write().unwrap_or_else(|err| err.into_inner());
-    snapshots.retain(|(entry_cycle, entry_epoch), _| {
-        active.contains_key(&(*entry_cycle, *entry_epoch))
-            || *entry_epoch > leader_epoch
-            || (*entry_epoch == leader_epoch && *entry_cycle >= cycle)
-    });
-}
-
-pub(crate) struct TierRegistryCycleGuard {
-    key: (u64, u64),
-}
-
-impl Drop for TierRegistryCycleGuard {
-    fn drop(&mut self) {
-        let mut active = TIER_ACTIVE_CYCLES.write().unwrap_or_else(|err| err.into_inner());
-        if let Some(count) = active.get_mut(&self.key) {
-            *count = count.saturating_sub(1);
-            if *count == 0 {
-                active.remove(&self.key);
-            }
-        }
-    }
-}
-
-pub(crate) fn begin_tier_registry_cycle(cycle: u64, leader_epoch: u64) -> TierRegistryCycleGuard {
-    let mut active = TIER_ACTIVE_CYCLES.write().unwrap_or_else(|err| err.into_inner());
-    let count = active.entry((cycle, leader_epoch)).or_default();
-    *count = count.saturating_add(1);
-    TierRegistryCycleGuard {
-        key: (cycle, leader_epoch),
-    }
-}
-
-fn tier_registry_generation(names: &[String]) -> u64 {
-    let mut hasher = Sha256::new();
-    hasher.update(b"rustfs-tier-registry-v1");
-    for name in names {
-        hasher.update(u64::try_from(name.len()).unwrap_or(u64::MAX).to_le_bytes());
-        hasher.update(name.as_bytes());
-    }
-    let digest = hasher.finalize();
-    let mut prefix = [0_u8; 8];
-    prefix.copy_from_slice(&digest[..8]);
-    u64::from_le_bytes(prefix)
-}
-
-/// Drop cycle snapshots only after the scanner has finished publishing a
-/// cycle. In-flight or retryable cycles must retain their original registry;
-/// TTL/capacity eviction could make a later bucket in the same cycle refresh
-/// to a different generation.
-pub(crate) fn complete_tier_registry_cycle(cycle: u64, leader_epoch: u64) {
-    let active = TIER_ACTIVE_CYCLES.read().unwrap_or_else(|err| err.into_inner());
-    let mut cached = TIER_CYCLE_SNAPSHOTS.write().unwrap_or_else(|err| err.into_inner());
-    cached.retain(|(entry_cycle, entry_epoch), _| {
-        active.contains_key(&(*entry_cycle, *entry_epoch))
-            || *entry_epoch > leader_epoch
-            || (*entry_epoch == leader_epoch && *entry_cycle > cycle)
-    });
+    let tiers = ecstore_get_global_tier_config_mgr().read().await.list_tiers();
+    let names: Arc<[String]> = tiers.iter().map(|tier| tier.name.clone()).collect::<Vec<_>>().into();
+    *TIER_NAME_CACHE.write().unwrap_or_else(|err| err.into_inner()) = Some((Instant::now(), Arc::clone(&names)));
+    names
 }
 
 /// Test-only cache reset; the production cache has no invalidation hook
@@ -619,9 +408,6 @@ pub(crate) fn complete_tier_registry_cycle(cycle: u64, leader_epoch: u64) {
 #[cfg(test)]
 fn reset_tier_name_cache_for_test() {
     *TIER_NAME_CACHE.write().unwrap_or_else(|err| err.into_inner()) = None;
-    TIER_ACTIVE_CYCLES.write().unwrap_or_else(|err| err.into_inner()).clear();
-    TIER_CYCLE_SNAPSHOTS.write().unwrap_or_else(|err| err.into_inner()).clear();
-    TIER_REGISTRY_GENERATION.store(0, Ordering::Relaxed);
 }
 
 pub(crate) async fn enqueue_runtime_free_version(oi: ScannerObjectInfo) {
@@ -830,68 +616,6 @@ mod tests {
         assert!(Arc::ptr_eq(&first, &second));
     }
 
-    #[tokio::test]
-    async fn tier_registry_cycle_snapshot_stays_fixed_while_active() {
-        reset_tier_name_cache_for_test();
-        let cycle = 9_000_001;
-        let leader_epoch = 9_000_002;
-        let guard = begin_tier_registry_cycle(cycle, leader_epoch);
-        let first = runtime_tier_registry_for_cycle(cycle, leader_epoch).await;
-
-        // Simulate a TTL refresh observing a different configuration while the
-        // original cycle is still scanning. The active cycle entry must win.
-        *TIER_NAME_CACHE.write().unwrap_or_else(|err| err.into_inner()) = Some((
-            Instant::now() - TIER_NAME_CACHE_TTL - Duration::from_secs(1),
-            TierRegistrySnapshot {
-                generation: u64::MAX,
-                names: Arc::from(["COLD".to_string()]),
-                refresh_failed: false,
-            },
-        ));
-        let second = runtime_tier_registry_for_cycle(cycle, leader_epoch).await;
-        assert_eq!(second.generation, first.generation);
-        assert_eq!(second.names, first.names);
-
-        drop(guard);
-        complete_tier_registry_cycle(cycle, leader_epoch);
-        reset_tier_name_cache_for_test();
-    }
-
-    #[tokio::test]
-    async fn tier_registry_generation_survives_new_cycle_with_same_names() {
-        reset_tier_name_cache_for_test();
-        let first_cycle = 9_000_011;
-        let second_cycle = first_cycle + 1;
-        let leader_epoch = 9_000_012;
-
-        let first_guard = begin_tier_registry_cycle(first_cycle, leader_epoch);
-        let first = runtime_tier_registry_for_cycle(first_cycle, leader_epoch).await;
-        drop(first_guard);
-        complete_tier_registry_cycle(first_cycle, leader_epoch);
-
-        let second_guard = begin_tier_registry_cycle(second_cycle, leader_epoch);
-        let second = runtime_tier_registry_for_cycle(second_cycle, leader_epoch).await;
-        assert_eq!(first.names, second.names);
-        assert_eq!(first.generation, second.generation);
-
-        drop(second_guard);
-        complete_tier_registry_cycle(second_cycle, leader_epoch);
-        reset_tier_name_cache_for_test();
-    }
-
-    #[test]
-    fn invalid_tier_registry_names_fail_closed_for_refresh() {
-        assert!(validate_tier_registry_names(vec!["COLD\n".to_string()]).is_err());
-        assert!(validate_tier_registry_names(vec![UNKNOWN_TIER.to_string()]).is_err());
-        assert!(validate_tier_registry_names(vec!["COLD".to_string(), "COLD".to_string()]).is_err());
-        assert_eq!(
-            validate_tier_registry_names(vec!["WARM".to_string(), "COLD".to_string()])
-                .expect("valid registry names")
-                .as_ref(),
-            ["COLD".to_string(), "WARM".to_string()]
-        );
-    }
-
     #[test]
     fn foreground_read_guard_tracks_stream_lifetime() {
         reset_foreground_read_activity_for_test();

crates/scanner/src/remote_scanner/stream.rs

@@ -16,8 +16,8 @@
 use crate::RUSTFS_META_BUCKET;
 use crate::scanner_budget::{ScannerCycleBudget, ScannerCycleBudgetConfig};
 use crate::scanner_io::{
-    DataUsageCacheReuseOptions, DataUsageCacheScanState, ScannerDiskScanOutcome, ScannerIODisk, acquire_scanner_cache_locks,
-    cache_root_entry_info, current_cache_root_or_prepare_with_generation, scanner_set_disk_inventory,
+    DataUsageCacheScanState, ScannerDiskScanOutcome, ScannerIODisk, acquire_scanner_cache_locks, cache_root_entry_info,
+    current_cache_root_or_prepare, scanner_set_disk_inventory,
 };
 use crate::storage_api::owner::NS_SCANNER_PROTOCOL_VERSION;
 use crate::{
@@ -212,7 +212,6 @@ pub(crate) struct RemoteScannerScanSpec<'a> {
     pub(crate) session_id: Uuid,
     pub(crate) session_sequence: u64,
     pub(crate) scan_plan_digest: DataUsageScanPlanDigest,
-    pub(crate) tier_registry_generation: u64,
     pub(crate) skip_healing: bool,
     pub(crate) scan_mode: HealScanMode,
 }
@@ -223,7 +222,6 @@ struct RemoteScannerResponseExpectation<'a> {
     source: DataUsageCacheSource,
     next_cycle: u64,
     scan_plan_digest: DataUsageScanPlanDigest,
-    tier_registry_generation: u64,
 }
 
 #[derive(Debug)]
@@ -670,11 +668,6 @@ async fn scan_and_persist_local_bucket(
         scan_mode,
         ..
     } = request;
-    // Keep the worker's cycle snapshot alive through cache reuse, scanning,
-    // and persistence. Without the guard, a later cycle can prune this key
-    // while this request is still running and allow a second registry to be
-    // selected for the same cycle.
-    let _tier_cycle_guard = crate::begin_tier_registry_cycle(next_cycle, leader_epoch);
     let store = resolve_scanner_object_store_handle()
         .ok_or_else(|| RemoteScannerServerError::worker("remote namespace scanner object layer is unavailable"))?;
     validate_remote_scanner_request_fence_with_store(next_cycle, leader_epoch, store.clone())
@@ -707,25 +700,7 @@ async fn scan_and_persist_local_bucket(
     let revisions = cache.load_with_revisions(set.clone(), &cache_name).await.map_err(|err| {
         RemoteScannerServerError::worker(format!("remote namespace scanner cache load or revision lookup failed: {err}"))
     })?;
-    // Remote workers use the same cycle-frozen registry as `scan_data_folder`.
-    // Requiring its generation here prevents a cache snapshot classified by an
-    // older registry from being reused before the folder scan gets a chance to
-    // refresh it.
-    let tier_registry_generation = crate::runtime_tier_registry_for_cycle(next_cycle, leader_epoch)
-        .await
-        .generation;
-    let scan_state = current_cache_root_or_prepare_with_generation(
-        &mut cache,
-        &bucket,
-        source,
-        next_cycle,
-        leader_epoch,
-        scan_plan_digest,
-        DataUsageCacheReuseOptions {
-            require_source: true,
-            tier_registry_generation: Some(tier_registry_generation),
-        },
-    );
+    let scan_state = current_cache_root_or_prepare(&mut cache, &bucket, source, next_cycle, leader_epoch, scan_plan_digest, true);
     match scan_state {
         DataUsageCacheScanState::Current(usage) => {
             if guard.is_lock_lost() {
@@ -869,7 +844,6 @@ pub(crate) async fn scan_remote_bucket(
         session_id,
         session_sequence,
         scan_plan_digest,
-        tier_registry_generation,
         skip_healing,
         scan_mode,
     } = spec;
@@ -958,7 +932,6 @@ pub(crate) async fn scan_remote_bucket(
             source: expected_source,
             next_cycle,
             scan_plan_digest,
-            tier_registry_generation,
         },
         authenticator,
         rpc_deadline,
@@ -1014,7 +987,6 @@ where
             source: expected_source,
             next_cycle: TEST_NEXT_CYCLE,
             scan_plan_digest: expected_scan_plan_digest,
-            tier_registry_generation: 0,
         },
         authenticator,
         Instant::now() + NS_SCANNER_MAX_RPC_LIFETIME,
@@ -1114,11 +1086,6 @@ where
                         "remote namespace scanner returned usage for a different bucket plan",
                     )));
                 }
-                if complete.usage.tier_registry_generation != Some(expected.tier_registry_generation) {
-                    return Err(RemoteScannerStreamError::reconciled(StorageError::other(
-                        "remote namespace scanner returned usage for a different tier registry generation",
-                    )));
-                }
                 if !complete.usage.entry.children.is_empty() {
                     return Err(RemoteScannerStreamError::reconciled(StorageError::other(
                         "remote namespace scanner returned non-flattened bucket usage",

crates/scanner/src/remote_scanner/stream/tests.rs

@@ -195,7 +195,6 @@ fn test_usage(bucket: &str, objects: usize) -> DataUsageEntryInfo {
         name: bucket.to_string(),
         parent: crate::DATA_USAGE_ROOT.to_string(),
         entry,
-        tier_registry_generation: Some(0),
     }
 }
 
@@ -752,43 +751,6 @@ async fn complete_terminal_frame_reconciles_progress_and_usage() {
     assert_eq!(budget.progress(), (3, 2));
 }
 
-#[tokio::test]
-async fn terminal_usage_from_a_different_tier_generation_is_rejected() {
-    let request_id = Uuid::new_v4();
-    let writer_auth = FrameAuthenticator::for_test(request_id);
-    let reader_auth = FrameAuthenticator::for_test(request_id);
-    let (mut writer, reader) = tokio::io::duplex(4096);
-    tokio::spawn(async move {
-        let mut usage = test_usage("bucket", 1);
-        usage.tier_registry_generation = Some(1);
-        let mut sequence = 0;
-        write_frame(
-            &mut writer,
-            &writer_auth,
-            &mut sequence,
-            &RemoteScannerFrame::terminal(
-                RemoteScannerProgress::default(),
-                RemoteScannerFrameResult::Complete(Box::new(RemoteScannerComplete {
-                    source: TEST_SOURCE,
-                    scan_plan_digest: TEST_PLAN_DIGEST,
-                    usage,
-                    pending_maintenance_work: false,
-                })),
-            ),
-        )
-        .await
-        .expect("terminal frame should write");
-    });
-
-    let parent = CancellationToken::new();
-    let budget = ScannerCycleBudget::new(&parent, ScannerCycleBudgetConfig::default());
-    let error = consume_remote_scanner_stream(reader, parent, budget, "bucket", TEST_SOURCE, TEST_PLAN_DIGEST, reader_auth)
-        .await
-        .expect_err("generation mismatch must fail closed");
-
-    assert!(error.to_string().contains("tier registry generation"));
-}
-
 #[tokio::test]
 async fn complete_terminal_frame_after_budget_expiry_is_partial() {
     let request_id = Uuid::new_v4();

crates/scanner/src/scanner_folder.rs

@@ -55,9 +55,9 @@ use tracing::{debug, error, warn};
 use crate::{
     Disk, DiskError, DiskInfoOptions, Evaluator, Event, LcEventSrc, ListPathRawOptions, ObjectOpts, ReplicationConfig,
     ReplicationHealObject, ReplicationQueueAdmission, ReplicationStatusType, STORAGE_FORMAT_FILE, ScannerDiskExt as _,
-    ScannerLifecycleConfigExt as _, ScannerVersioningConfigExt as _, StorageError, TierRegistrySnapshot, apply_expiry_rule,
-    apply_transition_rule, enqueue_runtime_newer_noncurrent, is_reserved_or_invalid_bucket, list_path_raw, path2_bucket_object,
-    path2_bucket_object_with_base_path, queue_replication_heal, runtime_tier_registry_for_cycle, scanner_is_erasure,
+    ScannerLifecycleConfigExt as _, ScannerVersioningConfigExt as _, StorageError, apply_expiry_rule, apply_transition_rule,
+    enqueue_runtime_newer_noncurrent, is_reserved_or_invalid_bucket, list_path_raw, path2_bucket_object,
+    path2_bucket_object_with_base_path, queue_replication_heal, scanner_is_erasure,
     scanner_replication_config_for_lifecycle_eval,
 };
 use crate::{ScannerObjectInfo as ObjectInfo, ScannerObjectToDelete as ObjectToDelete};
@@ -626,20 +626,6 @@ fn apply_scanner_size_summary(into: &mut DataUsageEntry, summary: &SizeSummary)
     }
 
     into.add_tier_sizes(&summary.tier_stats);
-    into.add_unknown_tier_stats(&summary.unknown_tier_stats);
-    into.tier_accounting_proof = match (into.tier_accounting_proof, Some(summary.tier_accounting_proof)) {
-        (Some(mut current), Some(next)) => {
-            current.saturating_add(next);
-            Some(current)
-        }
-        (Some(_), None) => None,
-        (None, next) => next,
-    };
-    if into.unknown_tier_stats.as_ref().is_some_and(|stats| stats.counter_overflowed)
-        && let Some(proof) = into.tier_accounting_proof.as_mut()
-    {
-        proof.overflowed = true;
-    }
 }
 
 fn data_usage_root_has_progress(root: &DataUsageEntry) -> bool {
@@ -650,8 +636,6 @@ fn data_usage_root_has_progress(root: &DataUsageEntry) -> bool {
         || root.delete_markers > 0
         || root.failed_objects > 0
         || root.replication_stats.is_some()
-        || root.all_tier_stats.as_ref().is_some_and(|stats| !stats.is_empty())
-        || root.unknown_tier_stats.as_ref().is_some_and(|stats| !stats.is_empty())
 }
 
 fn partial_cache_is_useful(root: &DataUsageEntry, pending_heals_changed: bool) -> bool {
@@ -686,9 +670,6 @@ pub struct FolderScanner {
     budget: Arc<ScannerCycleBudget>,
     skip_heal: Arc<std::sync::atomic::AtomicBool>,
     local_disk: Arc<Disk>,
-    /// Tier registry frozen for this folder scan. A refresh applies to the
-    /// next scan and cannot mix generations in one aggregate.
-    tier_registry: TierRegistrySnapshot,
     pending_heals_changed: bool,
     #[cfg(test)]
     list_path_raw_options_observer: Option<mpsc::UnboundedSender<ListPathRawTimeoutSnapshot>>,
@@ -1348,11 +1329,7 @@ impl FolderScanner {
                     continue;
                 }
 
-                let sz = match self
-                    .local_disk
-                    .get_size_with_tier_names(item.clone(), &self.tier_registry.names)
-                    .await
-                {
+                let sz = match self.local_disk.get_size(item.clone()).await {
                     Ok(sz) => sz,
                     Err(e) => {
                         let failure_action = classify_get_size_failure(&item, &e);
@@ -2184,10 +2161,6 @@ pub async fn scan_data_folder(
 
     let failed_object_ttl = rustfs_utils::get_env_u32(ENV_FAILED_OBJECT_TTL_SECS, DEFAULT_FAILED_OBJECT_TTL_SECS) as u64;
     let failed_objects_max = rustfs_utils::get_env_u32(ENV_FAILED_OBJECTS_MAX, DEFAULT_FAILED_OBJECTS_MAX) as usize;
-    let tier_registry = runtime_tier_registry_for_cycle(cache.info.next_cycle, cache.info.leader_epoch).await;
-    let mut cache = cache;
-    cache.fold_retired_tiers(&tier_registry.names);
-    cache.info.tier_registry_generation = Some(tier_registry.generation);
 
     // Create folder scanner
     let mut scanner = FolderScanner {
@@ -2216,7 +2189,6 @@ pub async fn scan_data_folder(
         budget: budget.clone(),
         skip_heal,
         local_disk,
-        tier_registry,
         pending_heals_changed: false,
         #[cfg(test)]
         list_path_raw_options_observer: None,

crates/scanner/src/scanner_folder/item_actions.rs

@@ -289,62 +289,11 @@ impl ScannerItem {
         item.object_path()
     }
 
-    fn effective_tier(oi: &ObjectInfo) -> &str {
-        if oi.transitioned_object.status == crate::TRANSITION_COMPLETE {
-            oi.transitioned_object.tier.as_str()
-        } else {
-            oi.storage_class.as_deref().unwrap_or(crate::storageclass::STANDARD)
-        }
-    }
-
-    fn tier_name_is_known(tier: &str, tier_names: &[String]) -> bool {
-        !tier.is_empty()
-            && tier != crate::data_usage_define::UNKNOWN_TIER
-            && (tier == crate::storageclass::STANDARD
-                || tier == crate::storageclass::RRS
-                || tier_names.iter().any(|name| name == tier))
-    }
-
-    pub(crate) fn tier_is_known(oi: &ObjectInfo, tier_names: &[String]) -> bool {
-        Self::tier_name_is_known(Self::effective_tier(oi), tier_names)
-    }
-
-    fn action_requires_known_tier(action: IlmAction) -> bool {
-        matches!(
-            action,
-            IlmAction::TransitionAction
-                | IlmAction::TransitionVersionAction
-                | IlmAction::DeleteAction
-                | IlmAction::DeleteVersionAction
-                | IlmAction::DeleteRestoredAction
-                | IlmAction::DeleteRestoredVersionAction
-                | IlmAction::DeleteAllVersionsAction
-                | IlmAction::DelMarkerDeleteAllVersionsAction
-        )
-    }
-
-    fn action_blocked_by_unknown_tier(
-        action: IlmAction,
-        oi: &ObjectInfo,
-        all_versions_known: bool,
-        tier_names: &[String],
-        target: &str,
-    ) -> bool {
-        if !Self::action_requires_known_tier(action) {
-            return false;
-        }
-        !Self::tier_is_known(oi, tier_names)
-            || (action.delete_all() && !all_versions_known)
-            || (matches!(action, IlmAction::TransitionAction | IlmAction::TransitionVersionAction)
-                && !Self::tier_name_is_known(target, tier_names))
-    }
-
     pub async fn apply_actions(
         &mut self,
         object_infos: Vec<ObjectInfo>,
         lock_retention: Option<Arc<ObjectLockConfiguration>>,
         versioning_config: VersioningConfiguration,
-        tier_names: &[String],
         size_summary: &mut SizeSummary,
     ) {
         let object_path = self.object_path();
@@ -453,9 +402,6 @@ impl ScannerItem {
         let mut noncurrent_accounting: Vec<PendingScannerAccounting<'_>> = Vec::new();
         let mut cumulative_size = 0;
         let mut remaining_versions = object_infos.len();
-        let all_versions_known = object_infos
-            .iter()
-            .all(|candidate| Self::tier_is_known(candidate, tier_names));
         'eventLoop: {
             for (i, event) in events.iter().enumerate() {
                 let oi = &object_infos[i];
@@ -479,18 +425,6 @@ impl ScannerItem {
                 let mut size = actual_size;
                 let mut account_now = true;
 
-                // A retired/unknown source tier may point at a remote object
-                // that cannot be safely deleted or transitioned. Lifecycle
-                // evaluation is still useful for accounting, but all
-                // side-effecting tier actions fail closed until the registry
-                // recognizes the source again.
-                if Self::action_blocked_by_unknown_tier(event.action, oi, all_versions_known, tier_names, &event.storage_class) {
-                    size = self.heal_actions(oi, actual_size, size_summary).await;
-                    size_summary.actions_accounting(oi, size, actual_size);
-                    cumulative_size += size;
-                    continue;
-                }
-
                 match event.action {
                     IlmAction::DeleteAllVersionsAction | IlmAction::DelMarkerDeleteAllVersionsAction => {
                         debug!(
@@ -995,49 +929,4 @@ mod tests {
         assert_eq!(item.object_name, "object");
         assert_eq!(item.object_path(), "object");
     }
-
-    #[test]
-    fn unknown_tier_never_triggers_transition() {
-        let object = ObjectInfo {
-            storage_class: Some("retired-tier".to_string()),
-            ..Default::default()
-        };
-        let tier_names = ["WARM".to_string()];
-        assert!(!ScannerItem::tier_is_known(&object, &tier_names));
-        assert!(ScannerItem::action_requires_known_tier(IlmAction::TransitionAction));
-        assert!(ScannerItem::action_requires_known_tier(IlmAction::DeleteVersionAction));
-        assert!(ScannerItem::action_blocked_by_unknown_tier(
-            IlmAction::TransitionAction,
-            &object,
-            false,
-            &tier_names,
-            "WARM"
-        ));
-        assert!(!ScannerItem::action_blocked_by_unknown_tier(
-            IlmAction::NoneAction,
-            &object,
-            false,
-            &tier_names,
-            "WARM"
-        ));
-
-        let known = ObjectInfo {
-            storage_class: Some(crate::storageclass::STANDARD.to_string()),
-            ..Default::default()
-        };
-        assert!(ScannerItem::action_blocked_by_unknown_tier(
-            IlmAction::DeleteAllVersionsAction,
-            &known,
-            false,
-            &tier_names,
-            "WARM"
-        ));
-        assert!(!ScannerItem::action_blocked_by_unknown_tier(
-            IlmAction::TransitionAction,
-            &known,
-            true,
-            &tier_names,
-            crate::storageclass::STANDARD
-        ));
-    }
 }

crates/scanner/src/scanner_folder/tests.rs

@@ -325,11 +325,6 @@ async fn build_test_scanner() -> (FolderScanner, std::path::PathBuf) {
         budget: ScannerCycleBudget::new(&CancellationToken::new(), Default::default()),
         skip_heal: Arc::new(AtomicBool::new(false)),
         local_disk: disk,
-        tier_registry: crate::TierRegistrySnapshot {
-            generation: 0,
-            names: Arc::new([]),
-            refresh_failed: false,
-        },
         pending_heals_changed: false,
         list_path_raw_options_observer: None,
     };

crates/scanner/src/scanner_io.rs

@@ -57,9 +57,8 @@ use crate::storage_api::scanner_io::{BucketInfo, BucketOptions};
 use crate::{
     BucketTargetSys, BucketVersioningSys, Disk, DiskError, ECStore, EcstoreError as Error, EcstoreResult as Result,
     RUSTFS_META_BUCKET, ReplicationConfig, STORAGE_FORMAT_FILE, ScannerDiskExt as _, ScannerLifecycleConfigExt as _,
-    ScannerReplicationConfigExt as _, ScannerVersioningConfigExt as _, SetDisks, StorageError, begin_tier_registry_cycle,
-    complete_tier_registry_cycle, enqueue_runtime_free_version, get_lifecycle_config, get_object_lock_config,
-    get_replication_config, runtime_tier_names, runtime_tier_registry_for_cycle, storageclass,
+    ScannerReplicationConfigExt as _, ScannerVersioningConfigExt as _, SetDisks, StorageError, enqueue_runtime_free_version,
+    get_lifecycle_config, get_object_lock_config, get_replication_config, runtime_tier_names, storageclass,
 };
 
 pub(crate) const SCANNER_SKIP_FILE_ERROR: &str = "skip file";
@@ -144,7 +143,6 @@ pub struct ScannerBucketScanPlan {
     all_buckets: Arc<Vec<BucketInfo>>,
     digest: DataUsageScanPlanDigest,
     leader_epoch: u64,
-    tier_registry_generation: u64,
     dirty_usage_buckets: Arc<DirtyUsageBuckets>,
     bucket_failures: ScannerBucketFailureState,
     pending_maintenance_work: Arc<AtomicBool>,
@@ -343,20 +341,12 @@ pub(crate) fn cache_root_entry_info(cache: &DataUsageCache) -> std::result::Resu
         name: cache.info.name.clone(),
         parent: DATA_USAGE_ROOT.to_string(),
         entry,
-        tier_registry_generation: cache.info.tier_registry_generation,
     })
 }
 
-fn apply_bucket_result_to_cache(cache: &mut DataUsageCache, result: DataUsageEntryInfo, update_time: SystemTime) -> bool {
-    if cache.info.tier_registry_generation != result.tier_registry_generation {
-        // A result from another registry generation must never be folded into
-        // this cycle. Leaving it unapplied makes the cycle incomplete and
-        // forces the caller to re-account it under one frozen registry.
-        return false;
-    }
+fn apply_bucket_result_to_cache(cache: &mut DataUsageCache, result: DataUsageEntryInfo, update_time: SystemTime) {
     cache.replace(&result.name, &result.parent, result.entry);
     cache.info.last_update = Some(update_time);
-    true
 }
 
 fn should_publish_completed_snapshot(completed_count: usize, total_count: usize, budget_elapsed: bool, cancelled: bool) -> bool {
@@ -508,9 +498,6 @@ pub trait ScannerIODisk: Send + Sync + Debug + 'static {
     ) -> Result<ScannerDiskScanOutcome>;
 
     async fn get_size(&self, item: ScannerItem) -> Result<SizeSummary>;
-
-    /// Read one object using a registry snapshot captured at scan start.
-    async fn get_size_with_tier_names(&self, item: ScannerItem, tier_names: &[String]) -> Result<SizeSummary>;
 }
 
 #[derive(Debug)]
@@ -662,10 +649,7 @@ use cache::*;
 use dirty_usage::*;
 use guards::*;
 
-pub(crate) use cache::{
-    DataUsageCacheReuseOptions, DataUsageCacheScanState, acquire_scanner_cache_locks,
-    current_cache_root_or_prepare_with_generation,
-};
+pub(crate) use cache::{DataUsageCacheScanState, acquire_scanner_cache_locks, current_cache_root_or_prepare};
 pub use dirty_usage::{
     ScannerDirtyUsageAckError, ScannerDirtyUsageState, acknowledge_dirty_usage_generation, clear_dirty_usage_bucket,
     record_dirty_usage_bucket, record_scanner_maintenance_change, scanner_activity_epoch, scanner_dirty_usage_state,

crates/scanner/src/scanner_io/cache.rs

@@ -100,14 +100,13 @@ where
     let _ = tokio::time::timeout(SCANNER_CACHE_LOCK_LOSS_SHUTDOWN_TIMEOUT, scan).await;
 }
 
-pub(crate) fn current_cache_root_entry_with_generation(
+pub(crate) fn current_cache_root_entry(
     cache: &DataUsageCache,
     name: &str,
     source: DataUsageCacheSource,
     next_cycle: u64,
     leader_epoch: u64,
     scan_plan_digest: DataUsageScanPlanDigest,
-    tier_registry_generation: Option<u64>,
 ) -> std::result::Result<Option<DataUsageEntryInfo>, ScannerError> {
     let metadata_is_current = cache.info.name == name
         && cache.info.source == Some(source)
@@ -116,8 +115,7 @@ pub(crate) fn current_cache_root_entry_with_generation(
         && cache.info.last_update.is_some()
         && cache.info.next_cycle == next_cycle
         && cache.info.leader_epoch == leader_epoch
-        && cache.info.cache_key_format == DATA_USAGE_CACHE_KEY_FORMAT
-        && tier_registry_generation.is_none_or(|generation| cache.info.tier_registry_generation == Some(generation));
+        && cache.info.cache_key_format == DATA_USAGE_CACHE_KEY_FORMAT;
     if !metadata_is_current {
         return Ok(None);
     }
@@ -133,13 +131,6 @@ pub(crate) enum DataUsageCacheScanState {
     },
 }
 
-#[derive(Clone, Copy, Debug, Default)]
-pub(crate) struct DataUsageCacheReuseOptions {
-    pub(crate) require_source: bool,
-    pub(crate) tier_registry_generation: Option<u64>,
-}
-
-#[cfg(test)]
 pub(crate) fn current_cache_root_or_prepare(
     cache: &mut DataUsageCache,
     name: &str,
@@ -149,51 +140,11 @@ pub(crate) fn current_cache_root_or_prepare(
     scan_plan_digest: DataUsageScanPlanDigest,
     require_source: bool,
 ) -> DataUsageCacheScanState {
-    current_cache_root_or_prepare_with_generation(
-        cache,
-        name,
-        source,
-        next_cycle,
-        leader_epoch,
-        scan_plan_digest,
-        DataUsageCacheReuseOptions {
-            require_source,
-            tier_registry_generation: None,
-        },
-    )
-}
-
-pub(crate) fn current_cache_root_or_prepare_with_generation(
-    cache: &mut DataUsageCache,
-    name: &str,
-    source: DataUsageCacheSource,
-    next_cycle: u64,
-    leader_epoch: u64,
-    scan_plan_digest: DataUsageScanPlanDigest,
-    options: DataUsageCacheReuseOptions,
-) -> DataUsageCacheScanState {
-    if options.tier_registry_generation.is_some_and(|generation| {
-        cache.info.next_cycle <= next_cycle
-            && cache.info.leader_epoch <= leader_epoch
-            && cache.info.tier_registry_generation != Some(generation)
-    }) {
-        // Make prepare_for_scan take its reset path so an entry classified by
-        // an older registry cannot be reused under the new cycle generation.
-        cache.info.scan_plan_digest = None;
-    }
-    match current_cache_root_entry_with_generation(
-        cache,
-        name,
-        source,
-        next_cycle,
-        leader_epoch,
-        scan_plan_digest,
-        options.tier_registry_generation,
-    ) {
+    match current_cache_root_entry(cache, name, source, next_cycle, leader_epoch, scan_plan_digest) {
         Ok(Some(root)) => DataUsageCacheScanState::Current(Box::new(root)),
         current => DataUsageCacheScanState::Prepared {
             invalid_current: current.err(),
-            outcome: cache.prepare_for_scan(name, next_cycle, leader_epoch, source, scan_plan_digest, options.require_source),
+            outcome: cache.prepare_for_scan(name, next_cycle, leader_epoch, source, scan_plan_digest, require_source),
         },
     }
 }
@@ -208,7 +159,7 @@ pub(super) fn cache_snapshot_is_current(
     scan_plan_digest: DataUsageScanPlanDigest,
 ) -> bool {
     matches!(
-        current_cache_root_entry_with_generation(cache, name, source, next_cycle, leader_epoch, scan_plan_digest, None),
+        current_cache_root_entry(cache, name, source, next_cycle, leader_epoch, scan_plan_digest),
         Ok(Some(_))
     )
 }
@@ -217,7 +168,6 @@ pub(super) fn completed_data_usage_info(
     results: &[DataUsageCache],
     expected_sources: &HashSet<DataUsageCacheSource>,
     all_buckets: &[String],
-    tier_registry_names: &[String],
     bucket_plan_complete: bool,
     budget_elapsed: bool,
     cancelled: bool,
@@ -233,19 +183,6 @@ pub(super) fn completed_data_usage_info(
         return None;
     }
 
-    // A generation is comparable across nodes because it is derived from the
-    // frozen registry names. Cycle and leader fencing remain separate cache
-    // metadata. Legacy peers omit the generation; an all-legacy result remains
-    // readable, but mixing legacy and new (or two new generations) would make
-    // the per-tier accounting ambiguous.
-    let registry_generation = results.first()?.info.tier_registry_generation;
-    if results.iter().any(|result| match registry_generation {
-        Some(generation) => result.info.tier_registry_generation != Some(generation),
-        None => result.info.tier_registry_generation.is_some(),
-    }) {
-        return None;
-    }
-
     if results.iter().any(|result| result.root().is_none()) {
         return None;
     }
@@ -263,16 +200,9 @@ pub(super) fn completed_data_usage_info(
         if !total.checked_merge(&merged) {
             return None;
         }
-        if !tier_accounting_proof_is_publishable(&merged, registry_generation, tier_registry_names) {
-            return None;
-        }
         buckets_usage.insert(bucket.clone(), checked_bucket_usage_info(&merged)?);
     }
 
-    if !tier_accounting_proof_is_publishable(&total, registry_generation, tier_registry_names) {
-        return None;
-    }
-
     let merged_last_update = results.iter().filter_map(|result| result.info.last_update).max()?;
     let bucket_sizes = buckets_usage
         .iter()
@@ -286,7 +216,6 @@ pub(super) fn completed_data_usage_info(
         delete_markers_total_count: u64::try_from(total.delete_markers).ok()?,
         objects_total_size: u64::try_from(total.size).ok()?,
         tier_stats: total.all_tier_stats.filter(|tiers| !tiers.is_empty()),
-        unknown_tier_stats: total.unknown_tier_stats.filter(|stats| !stats.is_empty()),
         buckets_count: u64::try_from(all_buckets.len()).ok()?,
         bucket_sizes,
         buckets_usage,
@@ -296,109 +225,6 @@ pub(super) fn completed_data_usage_info(
     Some((data_usage_info, merged_last_update))
 }
 
-fn tier_accounting_proof_is_publishable(
-    entry: &DataUsageEntry,
-    registry_generation: Option<u64>,
-    tier_registry_names: &[String],
-) -> bool {
-    let has_scalar_usage = entry.size > 0
-        || entry.objects > 0
-        || entry.versions > 0
-        || entry.delete_markers > 0
-        || entry.failed_objects > 0
-        || !entry.obj_sizes.is_empty()
-        || !entry.obj_versions.is_empty()
-        || entry.replication_stats.as_ref().is_some_and(|stats| !stats.is_empty());
-    let has_tier_accounted_data = entry
-        .all_tier_stats
-        .as_ref()
-        .is_some_and(|stats| stats.tiers.values().any(|tier| !tier.is_empty()))
-        || entry.unknown_tier_stats.as_ref().is_some_and(|stats| {
-            stats.counter_overflowed
-                || stats.unknown_bytes > 0
-                || stats.unknown_physical_bytes > 0
-                || stats.unknown_objects > 0
-                || stats.unknown_versions > 0
-        });
-
-    let Some(proof) = entry.tier_accounting_proof else {
-        return !has_scalar_usage && !has_tier_accounted_data;
-    };
-    if proof.overflowed
-        || entry
-            .unknown_tier_stats
-            .as_ref()
-            .is_some_and(|stats| stats.counter_overflowed)
-        || u64::try_from(entry.size).ok() != Some(proof.logical_total)
-    {
-        return false;
-    }
-    let unknown_logical = entry.unknown_tier_stats.as_ref().map_or(0, |stats| stats.unknown_bytes);
-    let unknown_physical = entry
-        .unknown_tier_stats
-        .as_ref()
-        .map_or(0, |stats| stats.unknown_physical_bytes);
-    if proof
-        .logical_known
-        .checked_add(unknown_logical)
-        .is_none_or(|total| total != proof.logical_total)
-        || proof
-            .physical_known
-            .checked_add(unknown_physical)
-            .is_none_or(|total| total != proof.physical_total)
-    {
-        return false;
-    }
-    if registry_generation.is_some()
-        && entry.all_tier_stats.as_ref().is_some_and(|stats| {
-            stats.tiers.keys().any(|tier| {
-                tier != crate::UNKNOWN_TIER
-                    && tier != crate::storageclass::STANDARD
-                    && tier != crate::storageclass::RRS
-                    && !tier_registry_names.iter().any(|allowed| allowed == tier)
-            })
-        })
-    {
-        return false;
-    }
-
-    if !has_tier_accounted_data {
-        return true;
-    }
-
-    let Some(tiers) = entry.all_tier_stats.as_ref() else {
-        return false;
-    };
-    let map_unknown_physical = tiers.tiers.get(crate::UNKNOWN_TIER).map_or(0, |stats| stats.total_size);
-    let companion_unknown_physical = entry
-        .unknown_tier_stats
-        .as_ref()
-        .map_or(0, |stats| stats.unknown_physical_bytes);
-    if map_unknown_physical != companion_unknown_physical {
-        return false;
-    }
-
-    // A no-configuration scan intentionally stores only UNKNOWN_TIER after
-    // the first unknown object; STANDARD/RRS remain absent to preserve the
-    // historical empty-map shape. In that shape the scalar proof is the sole
-    // source of known physical bytes. Configured registries seed at least one
-    // non-UNKNOWN key, whose map total must match the proof.
-    let has_known_tier_map = tiers.tiers.keys().any(|tier| tier.as_str() != crate::UNKNOWN_TIER);
-    if !has_known_tier_map {
-        return true;
-    }
-    let Some(known_tier_physical_total) = tiers
-        .tiers
-        .iter()
-        .filter(|(tier, _)| tier.as_str() != crate::UNKNOWN_TIER)
-        .map(|(_, stats)| stats)
-        .try_fold(0_u64, |total, stats| total.checked_add(stats.total_size))
-    else {
-        return false;
-    };
-    proof.physical_known == known_tier_physical_total
-}
-
 pub(super) async fn send_cache_root_entry_info(
     bucket_result_tx: &mpsc::Sender<DataUsageEntryInfo>,
     cache: &DataUsageCache,
@@ -493,14 +319,13 @@ pub(super) async fn persist_and_publish_cache_snapshot(
         return None;
     }
     if matches!(
-        current_cache_root_entry_with_generation(
+        current_cache_root_entry(
             &persisted,
             DATA_USAGE_ROOT,
             source,
             cache_snapshot.info.next_cycle,
             cache_snapshot.info.leader_epoch,
             scan_plan_digest,
-            cache_snapshot.info.tier_registry_generation,
         ),
         Ok(Some(_))
     ) {

crates/scanner/src/scanner_io/io_cache.rs

@@ -31,7 +31,6 @@ impl ScannerIOCache for SetDisks {
             all_buckets,
             digest: scan_plan_digest,
             leader_epoch,
-            tier_registry_generation,
             dirty_usage_buckets,
             bucket_failures,
             pending_maintenance_work,
@@ -49,7 +48,6 @@ impl ScannerIOCache for SetDisks {
                     next_cycle: want_cycle,
                     last_update: Some(now),
                     leader_epoch,
-                    tier_registry_generation: Some(tier_registry_generation),
                     source: Some(source),
                     snapshot_complete: true,
                     scan_plan_digest: Some(scan_plan_digest),
@@ -220,13 +218,6 @@ impl ScannerIOCache for SetDisks {
                 "Scanner old data usage cache load failed; rebuilding from bucket caches"
             );
         }
-        // Fence a stale set aggregate before copying entries into per-bucket work caches.
-        if old_cache.info.next_cycle <= want_cycle
-            && old_cache.info.leader_epoch <= leader_epoch
-            && old_cache.info.tier_registry_generation != Some(tier_registry_generation)
-        {
-            old_cache.info.scan_plan_digest = None;
-        }
         match old_cache.prepare_for_scan(
             DATA_USAGE_ROOT,
             want_cycle,
@@ -276,7 +267,6 @@ impl ScannerIOCache for SetDisks {
                 name: DATA_USAGE_ROOT.to_string(),
                 next_cycle: want_cycle,
                 leader_epoch,
-                tier_registry_generation: Some(tier_registry_generation),
                 source: Some(source),
                 snapshot_complete: false,
                 scan_plan_digest: Some(scan_plan_digest),
@@ -338,9 +328,8 @@ impl ScannerIOCache for SetDisks {
                         };
 
                         let mut cache = cache_mutex_clone.lock().await;
-                        if apply_bucket_result_to_cache(&mut cache, result, SystemTime::now()) {
-                            completed_bucket_count_clone.fetch_add(1, Ordering::Relaxed);
-                        }
+                        apply_bucket_result_to_cache(&mut cache, result, SystemTime::now());
+                        completed_bucket_count_clone.fetch_add(1, Ordering::Relaxed);
                     }
                 }
             }
@@ -471,7 +460,6 @@ impl ScannerIOCache for SetDisks {
                                 session_id: remote_session_id,
                                 session_sequence: request_sequence,
                                 scan_plan_digest: bucket_scan_plan_digest,
-                                tier_registry_generation,
                                 skip_healing: healing,
                                 scan_mode,
                             },
@@ -687,17 +675,14 @@ impl ScannerIOCache for SetDisks {
                             continue;
                         }
                     };
-                    let scan_state = current_cache_root_or_prepare_with_generation(
+                    let scan_state = current_cache_root_or_prepare(
                         &mut cache,
                         &bucket.name,
                         source,
                         want_cycle,
                         leader_epoch,
                         bucket_scan_plan_digest,
-                        DataUsageCacheReuseOptions {
-                            require_source: require_cache_source,
-                            tier_registry_generation: Some(tier_registry_generation),
-                        },
+                        require_cache_source,
                     );
                     let outcome = match scan_state {
                         DataUsageCacheScanState::Current(root) => {
@@ -1123,7 +1108,6 @@ impl ScannerIOCache for SetDisks {
                     name: DATA_USAGE_ROOT.to_string(),
                     next_cycle: want_cycle,
                     leader_epoch,
-                    tier_registry_generation: Some(tier_registry_generation),
                     source: Some(source),
                     snapshot_complete: false,
                     scan_plan_digest: Some(scan_plan_digest),

crates/scanner/src/scanner_io/io_cycle.rs

@@ -48,7 +48,6 @@ impl ScannerIOCycle for ECStore {
         scan_mode: HealScanMode,
     ) -> Result<ScannerCycleResult> {
         let child_token = ctx.child_token();
-        let _tier_cycle_guard = begin_tier_registry_cycle(want_cycle, leader_epoch);
 
         // Check the local pool metadata before listing buckets. A failed or
         // canceled decommission remains suspended after its worker exits, so
@@ -128,8 +127,6 @@ impl ScannerIOCycle for ECStore {
             scanner_bucket_plan_digest(&all_buckets, crate::scanner::scanner_activity_snapshot_digest(&activity_before));
         let dirty_usage_snapshot = Arc::new(snapshot_dirty_usage_buckets(&all_buckets, dirty_generation_before_bucket_list));
         let cache_cycle_floor = Arc::new(AtomicU64::new(want_cycle));
-        let tier_registry = runtime_tier_registry_for_cycle(want_cycle, leader_epoch).await;
-        let tier_registry_generation = tier_registry.generation;
 
         if all_buckets.is_empty() {
             reset_set_scan_gauges();
@@ -160,9 +157,6 @@ impl ScannerIOCycle for ECStore {
             {
                 return Ok(ScannerCycleResult::new(status, None));
             }
-            if status == ScannerCycleStatus::Complete {
-                complete_tier_registry_cycle(want_cycle, leader_epoch);
-            }
             let dirty_usage_clear =
                 (status == ScannerCycleStatus::Complete).then(|| dirty_usage_snapshot.buckets.as_ref().clone());
             let remote_dirty_usage_acknowledgements = if status == ScannerCycleStatus::Complete {
@@ -255,7 +249,6 @@ impl ScannerIOCycle for ECStore {
                     all_buckets: Arc::clone(&all_buckets),
                     digest: scan_plan_digest,
                     leader_epoch,
-                    tier_registry_generation,
                     dirty_usage_buckets: dirty_usage_snapshot.buckets.clone(),
                     bucket_failures: bucket_failures.clone(),
                     pending_maintenance_work: pending_maintenance_work.clone(),
@@ -373,7 +366,6 @@ impl ScannerIOCycle for ECStore {
             &results,
             &expected_sources,
             &all_bucket_names,
-            &tier_registry.names,
             bucket_plan_complete,
             budget_elapsed,
             ctx.is_cancelled(),
@@ -399,9 +391,6 @@ impl ScannerIOCycle for ECStore {
             &failed_buckets,
         );
         result?;
-        if cycle_status == ScannerCycleStatus::Complete {
-            complete_tier_registry_cycle(want_cycle, leader_epoch);
-        }
         let remote_dirty_usage_acknowledgements = if cycle_status == ScannerCycleStatus::Complete {
             crate::scanner::scanner_dirty_usage_acknowledgements(&activity_before)
         } else {

crates/scanner/src/scanner_io/io_disk.rs

@@ -13,38 +13,29 @@
 // limitations under the License.
 /// ScannerIODisk implementation for Disk: get_size and the per-disk bucket scan.
 use super::*;
-use crate::UNKNOWN_TIER;
 
 ///
 /// Seed [`SizeSummary::tier_stats`] from the cached tier-name list.
 ///
-/// Preserves the original no-tier shape: with no tiers configured the map
-/// stays completely empty (STANDARD/RRS/UNKNOWN are not seeded either).
-/// Otherwise the standard storage classes and one fixed unknown bucket are
-/// seeded alongside every configured tier so per-object accounting never
-/// inserts an untrusted metadata key.
+/// Preserves the original seeding semantics: with no tiers configured the map
+/// stays completely empty (STANDARD/RRS are not seeded either); otherwise the
+/// standard storage classes are seeded alongside every configured tier so
+/// per-object accounting always finds its tier key.
 pub(super) fn tier_stats_template(tier_names: &[String]) -> HashMap<String, TierStats> {
-    let mut tier_stats = HashMap::with_capacity(tier_names.len() + 3);
+    let mut tier_stats = HashMap::with_capacity(tier_names.len() + 2);
     for tier_name in tier_names {
-        if tier_name != UNKNOWN_TIER {
-            tier_stats.insert(tier_name.clone(), TierStats::default());
-        }
+        tier_stats.insert(tier_name.clone(), TierStats::default());
     }
     if !tier_stats.is_empty() {
         tier_stats.insert(storageclass::STANDARD.to_string(), TierStats::default());
         tier_stats.insert(storageclass::RRS.to_string(), TierStats::default());
-        tier_stats.insert(UNKNOWN_TIER.to_string(), TierStats::default());
     }
     tier_stats
 }
 
 #[async_trait::async_trait]
 impl ScannerIODisk for Disk {
-    async fn get_size(&self, item: ScannerItem) -> Result<SizeSummary> {
-        self.get_size_with_tier_names(item, &runtime_tier_names().await).await
-    }
-
-    async fn get_size_with_tier_names(&self, mut item: ScannerItem, tier_names: &[String]) -> Result<SizeSummary> {
+    async fn get_size(&self, mut item: ScannerItem) -> Result<SizeSummary> {
         let done_object = Metrics::time(Metric::ScanObject);
 
         if !is_xl_meta_path(&item.path) {
@@ -114,13 +105,12 @@ impl ScannerIODisk for Disk {
             .map(|v| ObjectInfo::from_file_info(v, item.bucket.as_str(), object_path.as_str(), versioned))
             .collect::<Vec<ObjectInfo>>();
 
-        // The caller supplies one registry snapshot for the whole folder scan;
-        // seeding from it prevents a TTL refresh from mixing generations in a
-        // single result.
-        let mut size_summary = SizeSummary {
-            tier_stats: tier_stats_template(tier_names),
-            ..Default::default()
-        };
+        let mut size_summary = SizeSummary::default();
+
+        // Tier names come from the process-wide TTL cache; seeding from them
+        // replaces the per-object clone of every full TierConfig.
+        let tier_names = runtime_tier_names().await;
+        size_summary.tier_stats = tier_stats_template(&tier_names);
 
         let lock_config = object_lock_config_for_scanner_item(&item).await;
 
@@ -130,14 +120,12 @@ impl ScannerIODisk for Disk {
         // `object_infos`.
         global_metrics().record_scanner_versions_scanned(object_infos.len() as u64);
 
-        item.apply_actions(object_infos, lock_config, versioning_config, tier_names, &mut size_summary)
+        item.apply_actions(object_infos, lock_config, versioning_config, &mut size_summary)
             .await;
 
         if !free_version_infos.is_empty() {
             for oi in free_version_infos {
-                if ScannerItem::tier_is_known(&oi, tier_names) {
-                    enqueue_runtime_free_version(oi).await;
-                }
+                enqueue_runtime_free_version(oi).await;
             }
         }
 

crates/scanner/src/scanner_io/publish_gate_tests.rs

@@ -13,8 +13,7 @@
 // limitations under the License.
 
 use super::*;
-use crate::data_usage_define::{UNKNOWN_TIER, UnknownTierStats, hash_path};
-use rustfs_data_usage::{ReplicationAllStats, ReplicationTargetUsage, TierAccountingProof};
+use rustfs_data_usage::{ReplicationAllStats, ReplicationTargetUsage};
 
 const TEST_PLAN_DIGEST: DataUsageScanPlanDigest = DataUsageScanPlanDigest([7; 32]);
 
@@ -90,11 +89,6 @@ fn completed_root_cache(bucket: &str, objects: usize, update_secs: u64, source:
         DataUsageEntry {
             objects,
             size: objects.saturating_mul(10),
-            tier_accounting_proof: Some(TierAccountingProof {
-                logical_total: u64::try_from(objects.saturating_mul(10)).unwrap_or(u64::MAX),
-                logical_known: u64::try_from(objects.saturating_mul(10)).unwrap_or(u64::MAX),
-                ..Default::default()
-            }),
             ..Default::default()
         },
     );
@@ -108,7 +102,7 @@ fn completed_data_usage_info_for_test(
     cancelled: bool,
 ) -> Option<(DataUsageInfo, SystemTime)> {
     let expected_sources = results.iter().filter_map(|result| result.info.source).collect::<HashSet<_>>();
-    completed_data_usage_info(results, &expected_sources, all_buckets, &[], true, budget_elapsed, cancelled)
+    completed_data_usage_info(results, &expected_sources, all_buckets, true, budget_elapsed, cancelled)
 }
 
 #[test]
@@ -128,21 +122,11 @@ fn completed_data_usage_info_publishes_tier_stats_across_sets() {
     let mut first_set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
     let mut tiered = DataUsageEntry::default();
     tiered.add_tier_sizes(&warm(100, 2, 1));
-    tiered.tier_accounting_proof = Some(TierAccountingProof {
-        physical_total: 100,
-        physical_known: 100,
-        ..Default::default()
-    });
     first_set.replace("bucket-b", DATA_USAGE_ROOT, tiered);
 
     let mut second_set = completed_root_cache("bucket-b", 2, 20, DataUsageCacheSource::new(1, 0));
     let mut tiered = DataUsageEntry::default();
     tiered.add_tier_sizes(&warm(50, 1, 1));
-    tiered.tier_accounting_proof = Some(TierAccountingProof {
-        physical_total: 50,
-        physical_known: 50,
-        ..Default::default()
-    });
     second_set.replace("bucket-a", DATA_USAGE_ROOT, tiered);
 
     let (data_usage_info, _) = completed_data_usage_info_for_test(&[first_set, second_set], &all_buckets, false, false)
@@ -161,269 +145,6 @@ fn completed_data_usage_info_publishes_tier_stats_across_sets() {
     );
 }
 
-#[test]
-fn completed_data_usage_info_rejects_logical_proof_mismatch() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let mut set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    let entry = set.cache.get_mut(&hash_path("bucket-a").key()).expect("bucket entry");
-    entry.add_tier_sizes(&HashMap::from([(
-        "WARM".to_string(),
-        TierStats {
-            total_size: 10,
-            num_versions: 1,
-            num_objects: 1,
-        },
-    )]));
-    entry.tier_accounting_proof = Some(TierAccountingProof {
-        logical_total: 10,
-        logical_known: 9,
-        physical_total: 10,
-        physical_known: 10,
-        ..Default::default()
-    });
-
-    assert!(completed_data_usage_info_for_test(&[set], &all_buckets, false, false).is_none());
-}
-
-#[test]
-fn completed_data_usage_info_rejects_logical_total_size_mismatch() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let mut set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    let entry = set.cache.get_mut(&hash_path("bucket-a").key()).expect("bucket entry");
-    entry.size = 11;
-    entry.add_tier_sizes(&HashMap::from([(
-        "WARM".to_string(),
-        TierStats {
-            total_size: 10,
-            num_versions: 1,
-            num_objects: 1,
-        },
-    )]));
-    entry.tier_accounting_proof = Some(TierAccountingProof {
-        logical_total: 10,
-        logical_known: 10,
-        physical_total: 10,
-        physical_known: 10,
-        ..Default::default()
-    });
-
-    assert!(completed_data_usage_info_for_test(&[set], &all_buckets, false, false).is_none());
-}
-
-#[test]
-fn completed_data_usage_info_rejects_physical_proof_mismatch() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let mut set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    let entry = set.cache.get_mut(&hash_path("bucket-a").key()).expect("bucket entry");
-    entry.add_tier_sizes(&HashMap::from([(
-        "WARM".to_string(),
-        TierStats {
-            total_size: 10,
-            num_versions: 1,
-            num_objects: 1,
-        },
-    )]));
-    entry.tier_accounting_proof = Some(TierAccountingProof {
-        logical_total: 10,
-        logical_known: 10,
-        physical_total: 9,
-        physical_known: 9,
-        ..Default::default()
-    });
-
-    assert!(completed_data_usage_info_for_test(&[set], &all_buckets, false, false).is_none());
-}
-
-#[test]
-fn completed_data_usage_info_rejects_unknown_physical_double_accounting() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let mut set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    let entry = set.cache.get_mut(&hash_path("bucket-a").key()).expect("bucket entry");
-    entry.add_tier_sizes(&HashMap::from([(
-        UNKNOWN_TIER.to_string(),
-        TierStats {
-            total_size: 10,
-            num_versions: 1,
-            num_objects: 1,
-        },
-    )]));
-    entry.add_unknown_tier_stats(&UnknownTierStats {
-        unknown_physical_bytes: 9,
-        ..Default::default()
-    });
-    entry.tier_accounting_proof = Some(TierAccountingProof {
-        logical_total: 10,
-        logical_known: 10,
-        physical_total: 10,
-        physical_known: 10,
-        ..Default::default()
-    });
-
-    assert!(completed_data_usage_info_for_test(&[set], &all_buckets, false, false).is_none());
-}
-
-#[test]
-fn completed_data_usage_info_rejects_unknown_counter_overflow() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let mut set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    let entry = set.cache.get_mut(&hash_path("bucket-a").key()).expect("bucket entry");
-    entry.add_unknown_tier_stats(&UnknownTierStats {
-        counter_overflowed: true,
-        ..Default::default()
-    });
-
-    assert!(completed_data_usage_info_for_test(&[set], &all_buckets, false, false).is_none());
-}
-
-#[test]
-fn completed_data_usage_info_accepts_no_tier_standard_empty_map_with_proof() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-
-    let (info, _) = completed_data_usage_info_for_test(&[set], &all_buckets, false, false)
-        .expect("no-tier STANDARD/RRS usage does not require a tier map");
-    assert!(info.tier_stats.is_none());
-}
-
-#[test]
-fn completed_data_usage_info_accepts_no_tier_unknown_and_standard_shape() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let mut set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    let entry = set.cache.get_mut(&hash_path("bucket-a").key()).expect("bucket entry");
-    entry.size = 13;
-    entry.add_tier_sizes(&HashMap::from([(
-        UNKNOWN_TIER.to_string(),
-        TierStats {
-            total_size: 3,
-            num_versions: 1,
-            num_objects: 1,
-        },
-    )]));
-    entry.add_unknown_tier_stats(&UnknownTierStats {
-        unknown_bytes: 9,
-        unknown_physical_bytes: 3,
-        unknown_objects: 1,
-        unknown_versions: 1,
-        ..Default::default()
-    });
-    entry.tier_accounting_proof = Some(TierAccountingProof {
-        logical_total: 13,
-        logical_known: 4,
-        physical_total: 7,
-        physical_known: 4,
-        ..Default::default()
-    });
-
-    let (info, _) = completed_data_usage_info_for_test(&[set], &all_buckets, false, false)
-        .expect("no-tier STANDARD plus UNKNOWN should remain publishable");
-    assert_eq!(
-        info.tier_stats.expect("unknown bucket should be retained").tiers[UNKNOWN_TIER].total_size,
-        3
-    );
-}
-
-#[test]
-fn completed_data_usage_info_accepts_unknown_only_with_current_registry_generation() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let mut set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    set.info.tier_registry_generation = Some(7);
-    let entry = set.cache.get_mut(&hash_path("bucket-a").key()).expect("bucket entry");
-    entry.size = 13;
-    entry.add_tier_sizes(&HashMap::from([(
-        UNKNOWN_TIER.to_string(),
-        TierStats {
-            total_size: 3,
-            num_versions: 1,
-            num_objects: 1,
-        },
-    )]));
-    entry.add_unknown_tier_stats(&UnknownTierStats {
-        unknown_bytes: 9,
-        unknown_physical_bytes: 3,
-        unknown_objects: 1,
-        unknown_versions: 1,
-        ..Default::default()
-    });
-    entry.tier_accounting_proof = Some(TierAccountingProof {
-        logical_total: 13,
-        logical_known: 4,
-        physical_total: 7,
-        physical_known: 4,
-        ..Default::default()
-    });
-
-    let expected_sources = HashSet::from([DataUsageCacheSource::new(0, 0)]);
-    assert!(
-        completed_data_usage_info(&[set], &expected_sources, &all_buckets, &["WARM".to_string()], true, false, false,).is_some()
-    );
-}
-
-#[test]
-fn completed_data_usage_info_rejects_non_registry_tier_in_current_generation() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let mut set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    set.info.tier_registry_generation = Some(7);
-    let entry = set.cache.get_mut(&hash_path("bucket-a").key()).expect("bucket entry");
-    entry.add_tier_sizes(&HashMap::from([
-        (
-            "WARM".to_string(),
-            TierStats {
-                total_size: 4,
-                num_versions: 1,
-                num_objects: 1,
-            },
-        ),
-        (
-            "RETIRED".to_string(),
-            TierStats {
-                total_size: 6,
-                num_versions: 1,
-                num_objects: 1,
-            },
-        ),
-    ]));
-    entry.tier_accounting_proof = Some(TierAccountingProof {
-        logical_total: 10,
-        logical_known: 10,
-        physical_total: 10,
-        physical_known: 10,
-        ..Default::default()
-    });
-
-    let expected_sources = HashSet::from([DataUsageCacheSource::new(0, 0)]);
-    assert!(
-        completed_data_usage_info(&[set], &expected_sources, &all_buckets, &["WARM".to_string()], true, false, false,).is_none()
-    );
-}
-
-#[test]
-fn completed_data_usage_info_rejects_legacy_proof_missing_when_tier_accounted() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let mut set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    let entry = set.cache.get_mut(&hash_path("bucket-a").key()).expect("bucket entry");
-    entry.add_tier_sizes(&HashMap::from([(
-        "WARM".to_string(),
-        TierStats {
-            total_size: 10,
-            num_versions: 1,
-            num_objects: 1,
-        },
-    )]));
-    entry.tier_accounting_proof = None;
-
-    assert!(completed_data_usage_info_for_test(&[set], &all_buckets, false, false).is_none());
-}
-
-#[test]
-fn completed_data_usage_info_rejects_legacy_proof_missing_for_scalar_usage() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let mut set = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    let entry = set.cache.get_mut(&hash_path("bucket-a").key()).expect("bucket entry");
-    entry.tier_accounting_proof = None;
-
-    assert!(completed_data_usage_info_for_test(&[set], &all_buckets, false, false).is_none());
-}
-
 #[test]
 fn completed_data_usage_info_omits_tier_stats_without_tiered_objects() {
     let all_buckets = vec!["bucket-a".to_string()];
@@ -435,51 +156,6 @@ fn completed_data_usage_info_omits_tier_stats_without_tiered_objects() {
     assert!(data_usage_info.tier_stats.is_none());
 }
 
-#[test]
-fn completed_data_usage_info_rejects_legacy_and_new_tier_generations_mixed() {
-    let all_buckets = vec!["bucket-a".to_string()];
-    let legacy = completed_root_cache("bucket-a", 1, 10, DataUsageCacheSource::new(0, 0));
-    let mut current = completed_root_cache("bucket-a", 1, 20, DataUsageCacheSource::new(1, 0));
-    current.info.tier_registry_generation = Some(42);
-
-    assert!(
-        completed_data_usage_info_for_test(&[legacy, current], &all_buckets, false, false).is_none(),
-        "legacy and generation-tagged sets must not publish a mixed snapshot"
-    );
-}
-
-#[test]
-fn current_cache_root_with_new_tier_generation_resets_old_cache() {
-    let source = DataUsageCacheSource::new(0, 0);
-    let mut cache = completed_root_cache("bucket-a", 1, 10, source);
-    cache.info.tier_registry_generation = Some(1);
-
-    let state = current_cache_root_or_prepare_with_generation(
-        &mut cache,
-        DATA_USAGE_ROOT,
-        source,
-        0,
-        0,
-        TEST_PLAN_DIGEST,
-        DataUsageCacheReuseOptions {
-            require_source: false,
-            tier_registry_generation: Some(2),
-        },
-    );
-
-    assert!(matches!(
-        state,
-        DataUsageCacheScanState::Prepared {
-            outcome: DataUsageCachePrepareOutcome::Reset,
-            ..
-        }
-    ));
-    assert!(cache.cache.is_empty(), "old-generation entries must not be reused");
-    assert_eq!(cache.info.tier_registry_generation, None);
-    assert_eq!(cache.info.scan_plan_digest, Some(TEST_PLAN_DIGEST));
-    assert!(!cache.info.snapshot_complete);
-}
-
 #[test]
 fn completed_data_usage_info_requires_every_set_before_publish() {
     let all_buckets = vec!["bucket-a".to_string(), "bucket-b".to_string(), "bucket-empty".to_string()];
@@ -604,13 +280,6 @@ fn completed_data_usage_info_flattens_nested_bucket_entries() {
             replica_size: 2048,
             replica_count: 2,
         }),
-        tier_accounting_proof: Some(TierAccountingProof {
-            logical_total: 2048,
-            logical_known: 2048,
-            physical_total: 2048,
-            physical_known: 2048,
-            ..Default::default()
-        }),
         ..Default::default()
     };
     nested.obj_sizes.add(2048);
@@ -678,8 +347,7 @@ fn completed_data_usage_info_requires_exact_topology_sources() {
     let expected_sources = HashSet::from([DataUsageCacheSource::new(0, 0), DataUsageCacheSource::new(1, 0)]);
 
     assert!(
-        completed_data_usage_info(&[first_set, unexpected_set], &expected_sources, &all_buckets, &[], true, false, false)
-            .is_none()
+        completed_data_usage_info(&[first_set, unexpected_set], &expected_sources, &all_buckets, true, false, false).is_none()
     );
 }
 
@@ -689,7 +357,7 @@ fn completed_data_usage_info_rejects_incomplete_bucket_plan() {
     let set = completed_root_cache("bucket", 2, 10, DataUsageCacheSource::new(0, 0));
     let expected_sources = HashSet::from([DataUsageCacheSource::new(0, 0)]);
 
-    assert!(completed_data_usage_info(&[set], &expected_sources, &all_buckets, &[], false, false, false).is_none());
+    assert!(completed_data_usage_info(&[set], &expected_sources, &all_buckets, false, false, false).is_none());
 }
 
 #[test]

crates/scanner/src/scanner_io/tests.rs

@@ -23,7 +23,7 @@ use crate::storage_api::owner::{
 use crate::storage_api::scan::{BucketOperations as _, DeleteBucketOptions, MakeBucketOptions, ObjectIO as _};
 use crate::{
     DiskOption, ECStore, Endpoint, EndpointServerPools, Endpoints, InstanceContext, PoolEndpoints, ScannerObjectOptions,
-    ScannerPutObjReader, UNKNOWN_TIER, init_bucket_metadata_sys_for_scanner_tests, init_ecstore_config_for_scanner_tests,
+    ScannerPutObjReader, init_bucket_metadata_sys_for_scanner_tests, init_ecstore_config_for_scanner_tests,
     init_local_disks_with_instance_ctx, new_disk, path2_bucket_object_with_base_path,
 };
 use rustfs_filemeta::FileInfo;
@@ -986,8 +986,8 @@ fn is_xl_meta_path_accepts_forward_separator() {
 fn tier_stats_template_seeds_tiers_and_standard_classes() {
     let template = tier_stats_template(&["WARM".to_string(), "COLD".to_string()]);
 
-    assert_eq!(template.len(), 5);
-    for tier in ["WARM", "COLD", storageclass::STANDARD, storageclass::RRS, UNKNOWN_TIER] {
+    assert_eq!(template.len(), 4);
+    for tier in ["WARM", "COLD", storageclass::STANDARD, storageclass::RRS] {
         assert_eq!(template.get(tier), Some(&TierStats::default()), "missing seed for tier {tier}");
     }
 }
@@ -1328,7 +1328,7 @@ fn apply_bucket_result_to_cache_updates_bucket_entry() {
     );
 
     let update_time = SystemTime::now();
-    assert!(apply_bucket_result_to_cache(
+    apply_bucket_result_to_cache(
         &mut cache,
         DataUsageEntryInfo {
             name: "bucket".to_string(),
@@ -1338,51 +1338,12 @@ fn apply_bucket_result_to_cache_updates_bucket_entry() {
                 objects: 2,
                 ..Default::default()
             },
-            tier_registry_generation: None,
         },
         update_time,
-    ));
+    );
 
     assert_eq!(cache.info.last_update, Some(update_time));
     let entry = cache.find("bucket").expect("bucket entry should remain present");
     assert_eq!(entry.size, 10);
     assert_eq!(entry.objects, 2);
 }
-
-#[test]
-fn apply_bucket_result_to_cache_rejects_a_different_tier_generation() {
-    let mut cache = DataUsageCache {
-        info: DataUsageCacheInfo {
-            name: DATA_USAGE_ROOT.to_string(),
-            tier_registry_generation: Some(7),
-            ..Default::default()
-        },
-        ..Default::default()
-    };
-    cache.replace(
-        "bucket",
-        DATA_USAGE_ROOT,
-        DataUsageEntry {
-            size: 3,
-            ..Default::default()
-        },
-    );
-
-    let applied = apply_bucket_result_to_cache(
-        &mut cache,
-        DataUsageEntryInfo {
-            name: "bucket".to_string(),
-            parent: DATA_USAGE_ROOT.to_string(),
-            entry: DataUsageEntry {
-                size: 11,
-                ..Default::default()
-            },
-            tier_registry_generation: Some(8),
-        },
-        SystemTime::now(),
-    );
-
-    assert!(!applied);
-    assert_eq!(cache.find("bucket").map(|entry| entry.size), Some(3));
-    assert!(cache.info.last_update.is_none());
-}

crates/storage-api/src/lib.rs

@@ -46,7 +46,6 @@ pub const NS_SCANNER_SERVER_EPOCH_QUERY: &str = "ns_scanner_server_epoch";
 pub const NS_SCANNER_SESSION_ID_QUERY: &str = "ns_scanner_session_id";
 pub const NS_SCANNER_SESSION_SEQUENCE_QUERY: &str = "ns_scanner_session_sequence";
 pub const NS_SCANNER_PROTOCOL_VERSION_QUERY: &str = "ns_scanner_protocol";
-pub const NS_SCANNER_TIER_REGISTRY_GENERATION_QUERY: &str = "ns_scanner_tier_registry_generation";
 pub const NS_SCANNER_PROTOCOL_VERSION: u16 = 3;
 pub const SCANNER_ACTIVITY_LEGACY_PROTOCOL_VERSION: u32 = 0;
 pub const SCANNER_ACTIVITY_PREVIOUS_PROTOCOL_VERSION: u32 = 5;
@@ -58,8 +57,6 @@ pub struct NsScannerCapabilityResponse {
     pub version: u16,
     pub server_epoch: uuid::Uuid,
     pub proof: Vec<u8>,
-    #[serde(default, skip_serializing_if = "Option::is_none")]
-    pub supports_tier_registry_generation: Option<bool>,
 }
 
 pub mod admin;

rustfs/src/storage/rpc/http_service.rs

@@ -19,15 +19,13 @@ use crate::storage::storage_api::rpc_consumer::http_service::{
     DEFAULT_READ_BUFFER_SIZE, DeleteOptions, DiskStore, NS_SCANNER_PROTOCOL_VERSION, NsScannerCapabilityResponse,
     PUT_FILE_AUTH_TRAILER_LEN, PUT_FILE_AUTH_V1, PUT_FILE_CAPABILITY_VERSION, PutFileCapabilityResponse, StorageDiskRpcExt as _,
     WALK_DIR_STREAM_COMPLETION_V1, WalkDirOptions, check_and_record_signed_rpc_nonce, find_local_disk_by_ref,
-    sign_ns_scanner_capability_with_tier_registry_generation, sign_put_file_capability, verify_put_file_auth_trailer,
-    verify_rpc_signature,
+    sign_ns_scanner_capability, sign_put_file_capability, verify_put_file_auth_trailer, verify_rpc_signature,
 };
 #[cfg(test)]
 use crate::storage::storage_api::rpc_consumer::http_service::{
     NS_SCANNER_BODY_SHA256_QUERY, NS_SCANNER_CAPABILITY_CHALLENGE_QUERY, NS_SCANNER_CYCLE_QUERY, NS_SCANNER_LEADER_EPOCH_QUERY,
     NS_SCANNER_REQUEST_ID_QUERY, NS_SCANNER_SERVER_EPOCH_QUERY, NS_SCANNER_SESSION_ID_QUERY, NS_SCANNER_SESSION_SEQUENCE_QUERY,
-    NS_SCANNER_TIER_REGISTRY_GENERATION_QUERY, PUT_FILE_CAPABILITY_CHALLENGE_QUERY, PUT_FILE_CAPABILITY_QUERY,
-    WALK_DIR_BODY_SHA256_QUERY,
+    PUT_FILE_CAPABILITY_CHALLENGE_QUERY, PUT_FILE_CAPABILITY_QUERY, WALK_DIR_BODY_SHA256_QUERY,
 };
 use crate::storage::storage_api::runtime_sources_consumer::runtime_sources;
 use crate::storage::storage_api::tonic_rpc_auth_failure_reason;
@@ -292,8 +290,6 @@ struct NsScannerQuery {
 struct NsScannerCapabilityQuery {
     ns_scanner_protocol: Option<u16>,
     ns_scanner_challenge: Option<uuid::Uuid>,
-    #[serde(rename = "ns_scanner_tier_registry_generation")]
-    ns_scanner_tier_registry_generation: Option<bool>,
 }
 
 fn verify_ns_scanner_body_digest(query: &NsScannerQuery, body: &[u8]) -> bool {
@@ -414,9 +410,7 @@ async fn handle_internode_rpc(req: Request<Incoming>) -> Response<Body> {
         (Method::GET, WALK_DIR_PATH) | (Method::HEAD, WALK_DIR_PATH) => handle_walk_dir(req).await,
         (Method::GET, NS_SCANNER_PATH) => match parse_query::<NsScannerCapabilityQuery>(&req) {
             Ok(query) if query.ns_scanner_protocol == Some(NS_SCANNER_PROTOCOL_VERSION) => match query.ns_scanner_challenge {
-                Some(challenge) if !challenge.is_nil() => {
-                    ns_scanner_capability_response(challenge, query.ns_scanner_tier_registry_generation == Some(true))
-                }
+                Some(challenge) if !challenge.is_nil() => ns_scanner_capability_response(challenge),
                 Some(_) | None => response_with_status(StatusCode::BAD_REQUEST, "namespace scanner challenge is invalid"),
             },
             Ok(_) => response_with_status(StatusCode::UPGRADE_REQUIRED, "namespace scanner protocol is unsupported"),
@@ -472,34 +466,31 @@ fn record_internode_rpc_error(operation: Option<&'static str>) {
     }
 }
 
-fn ns_scanner_capability_response(challenge: uuid::Uuid, include_tier_registry_generation: bool) -> Response<Body> {
+fn ns_scanner_capability_response(challenge: uuid::Uuid) -> Response<Body> {
     let server_epoch = *NS_SCANNER_SERVER_EPOCH;
-    let proof =
-        match sign_ns_scanner_capability_with_tier_registry_generation(challenge, server_epoch, include_tier_registry_generation)
-        {
-            Ok(proof) => proof,
-            Err(err) => {
-                error!(
-                    event = EVENT_RPC_REQUEST_FAILED,
-                    component = LOG_COMPONENT_INTERNODE_RPC,
-                    subsystem = LOG_SUBSYSTEM_NAMESPACE_SCANNER,
-                    operation = INTERNODE_OPERATION_NS_SCANNER,
-                    result = "failed",
-                    status_code = StatusCode::UPGRADE_REQUIRED.as_u16(),
-                    rpc_path = NS_SCANNER_PATH,
-                    method = %Method::GET,
-                    reason = "capability_authentication_unavailable",
-                    error = %err,
-                    "internode rpc request failed"
-                );
-                return response_with_status(StatusCode::UPGRADE_REQUIRED, "namespace scanner RPC authentication is unavailable");
-            }
-        };
+    let proof = match sign_ns_scanner_capability(challenge, server_epoch) {
+        Ok(proof) => proof,
+        Err(err) => {
+            error!(
+                event = EVENT_RPC_REQUEST_FAILED,
+                component = LOG_COMPONENT_INTERNODE_RPC,
+                subsystem = LOG_SUBSYSTEM_NAMESPACE_SCANNER,
+                operation = INTERNODE_OPERATION_NS_SCANNER,
+                result = "failed",
+                status_code = StatusCode::UPGRADE_REQUIRED.as_u16(),
+                rpc_path = NS_SCANNER_PATH,
+                method = %Method::GET,
+                reason = "capability_authentication_unavailable",
+                error = %err,
+                "internode rpc request failed"
+            );
+            return response_with_status(StatusCode::UPGRADE_REQUIRED, "namespace scanner RPC authentication is unavailable");
+        }
+    };
     let body = match rmp_serde::to_vec_named(&NsScannerCapabilityResponse {
         version: NS_SCANNER_PROTOCOL_VERSION,
         server_epoch,
         proof,
-        supports_tier_registry_generation: include_tier_registry_generation.then_some(true),
     }) {
         Ok(body) => body,
         Err(err) => {
@@ -1685,13 +1676,12 @@ mod tests {
         LOG_SUBSYSTEM_NAMESPACE_SCANNER, LOG_SUBSYSTEM_ROUTING, NS_SCANNER_BODY_SHA256_QUERY,
         NS_SCANNER_CAPABILITY_CHALLENGE_QUERY, NS_SCANNER_CYCLE_QUERY, NS_SCANNER_LEADER_EPOCH_QUERY, NS_SCANNER_PATH,
         NS_SCANNER_REQUEST_ID_QUERY, NS_SCANNER_SERVER_EPOCH_QUERY, NS_SCANNER_SESSION_ID_QUERY,
-        NS_SCANNER_SESSION_SEQUENCE_QUERY, NS_SCANNER_TIER_REGISTRY_GENERATION_QUERY, NsScannerCapabilityResponse,
-        NsScannerQuery, PUT_FILE_AUTH_STREAM_PATH, PUT_FILE_CAPABILITY_PATH, PUT_FILE_STREAM_PATH, PutFileQuery,
-        READ_FILE_STREAM_PATH, WALK_DIR_BODY_SHA256_QUERY, WALK_DIR_PATH, WalkDirQuery, append_walk_dir_completion,
-        internode_http_operation, internode_rpc_subsystem, is_internode_rpc_path, ns_scanner_response_body,
-        ns_scanner_server_epoch_matches, put_body_size_mismatch, put_file_auth_nonce, put_file_capability_response,
-        put_file_server_epoch_matches, put_file_stage_error_message, put_file_target_lock, read_file_body_stream,
-        read_file_stream_buffer_size, remote_scanner_claim_rejection, response_with_disk_error,
+        NS_SCANNER_SESSION_SEQUENCE_QUERY, NsScannerQuery, PUT_FILE_AUTH_STREAM_PATH, PUT_FILE_CAPABILITY_PATH,
+        PUT_FILE_STREAM_PATH, PutFileQuery, READ_FILE_STREAM_PATH, WALK_DIR_BODY_SHA256_QUERY, WALK_DIR_PATH, WalkDirQuery,
+        append_walk_dir_completion, internode_http_operation, internode_rpc_subsystem, is_internode_rpc_path,
+        ns_scanner_response_body, ns_scanner_server_epoch_matches, put_body_size_mismatch, put_file_auth_nonce,
+        put_file_capability_response, put_file_server_epoch_matches, put_file_stage_error_message, put_file_target_lock,
+        read_file_body_stream, read_file_stream_buffer_size, remote_scanner_claim_rejection, response_with_disk_error,
         supports_walk_dir_stream_completion, validate_walk_dir_completion_request, verify_internode_rpc_signature,
         verify_ns_scanner_body_digest, verify_walk_dir_body_digest, walk_dir_response_body, write_authenticated_put_file,
         write_body_chunks_to_writer, write_put_file_body_chunks_to_writer,
@@ -2124,51 +2114,6 @@ mod tests {
         );
         assert!(serde_urlencoded::from_str::<NsScannerQuery>(&query).is_err());
         assert!(serde_urlencoded::from_str::<super::NsScannerCapabilityQuery>("ns_scanner_protocol=1&unexpected=true").is_err());
-        let marked =
-            format!("ns_scanner_protocol=3&ns_scanner_challenge={request_id}&{NS_SCANNER_TIER_REGISTRY_GENERATION_QUERY}=true");
-        assert_eq!(
-            serde_urlencoded::from_str::<super::NsScannerCapabilityQuery>(&marked)
-                .expect("generation marker should be accepted")
-                .ns_scanner_tier_registry_generation,
-            Some(true)
-        );
-    }
-
-    #[test]
-    fn namespace_scanner_capability_response_support_is_optional_for_old_peers() {
-        #[derive(serde::Deserialize)]
-        #[serde(deny_unknown_fields)]
-        struct LegacyCapabilityResponse {
-            version: u16,
-            server_epoch: uuid::Uuid,
-            proof: Vec<u8>,
-        }
-
-        let old = rmp_serde::to_vec_named(&NsScannerCapabilityResponse {
-            version: super::NS_SCANNER_PROTOCOL_VERSION,
-            server_epoch: uuid::Uuid::new_v4(),
-            proof: vec![1, 2, 3],
-            supports_tier_registry_generation: None,
-        })
-        .expect("old response shape should encode");
-        let decoded: NsScannerCapabilityResponse = rmp_serde::from_slice(&old).expect("old response should decode");
-        assert_eq!(decoded.supports_tier_registry_generation, None);
-        let legacy_decoded: LegacyCapabilityResponse =
-            rmp_serde::from_slice(&old).expect("legacy reader should decode old shape");
-        assert_eq!(legacy_decoded.version, super::NS_SCANNER_PROTOCOL_VERSION);
-        assert!(!legacy_decoded.server_epoch.is_nil());
-        assert_eq!(legacy_decoded.proof, vec![1, 2, 3]);
-
-        let current = rmp_serde::to_vec_named(&NsScannerCapabilityResponse {
-            version: super::NS_SCANNER_PROTOCOL_VERSION,
-            server_epoch: uuid::Uuid::new_v4(),
-            proof: vec![1, 2, 3],
-            supports_tier_registry_generation: Some(true),
-        })
-        .expect("new response shape should encode");
-        let decoded: NsScannerCapabilityResponse = rmp_serde::from_slice(&current).expect("new response should decode");
-        assert_eq!(decoded.supports_tier_registry_generation, Some(true));
-        assert!(rmp_serde::from_slice::<LegacyCapabilityResponse>(&current).is_err());
     }
 
     #[test]

rustfs/src/storage/storage_api.rs

@@ -224,8 +224,8 @@ pub(crate) mod rpc_consumer {
         pub(crate) use super::super::storage_contracts::{
             NS_SCANNER_BODY_SHA256_QUERY, NS_SCANNER_CAPABILITY_CHALLENGE_QUERY, NS_SCANNER_CYCLE_QUERY,
             NS_SCANNER_LEADER_EPOCH_QUERY, NS_SCANNER_REQUEST_ID_QUERY, NS_SCANNER_SERVER_EPOCH_QUERY,
-            NS_SCANNER_SESSION_ID_QUERY, NS_SCANNER_SESSION_SEQUENCE_QUERY, NS_SCANNER_TIER_REGISTRY_GENERATION_QUERY,
-            PUT_FILE_CAPABILITY_CHALLENGE_QUERY, PUT_FILE_CAPABILITY_QUERY, WALK_DIR_BODY_SHA256_QUERY,
+            NS_SCANNER_SESSION_ID_QUERY, NS_SCANNER_SESSION_SEQUENCE_QUERY, PUT_FILE_CAPABILITY_CHALLENGE_QUERY,
+            PUT_FILE_CAPABILITY_QUERY, WALK_DIR_BODY_SHA256_QUERY,
         };
         pub(crate) use super::super::storage_contracts::{
             NS_SCANNER_PROTOCOL_VERSION, NsScannerCapabilityResponse, PUT_FILE_AUTH_TRAILER_LEN, PUT_FILE_AUTH_V1,
@@ -233,8 +233,8 @@ pub(crate) mod rpc_consumer {
         };
         pub(crate) use super::super::{
             DeleteOptions, DiskStore, StorageDiskRpcExt, WalkDirOptions, check_and_record_signed_rpc_nonce,
-            find_local_disk_by_ref, sign_ns_scanner_capability_with_tier_registry_generation, sign_put_file_capability,
-            verify_put_file_auth_trailer, verify_rpc_signature,
+            find_local_disk_by_ref, sign_ns_scanner_capability, sign_put_file_capability, verify_put_file_auth_trailer,
+            verify_rpc_signature,
         };
     }
 
@@ -520,10 +520,9 @@ pub(crate) mod ecstore_rpc {
     pub(crate) use rustfs_ecstore::api::rpc::{
         KMS_SIGNAL_SUBSYSTEM, LocalPeerS3Client, PEER_RESTDRY_RUN, PEER_RESTSIGNAL, PEER_RESTSUB_SYS, PeerRestClient,
         PeerS3Client, SERVICE_SIGNAL_REFRESH_CONFIG, SERVICE_SIGNAL_RELOAD_DYNAMIC, TONIC_RPC_PREFIX,
-        check_and_record_signed_rpc_nonce, normalize_tonic_rpc_audience,
-        sign_ns_scanner_capability_with_tier_registry_generation, sign_put_file_capability, sign_tonic_rpc_response_proof,
-        tonic_boot_epoch_challenge, tonic_boot_epoch_response_headers, tonic_rpc_auth_failure_reason,
-        verify_put_file_auth_trailer, verify_rpc_signature, verify_tonic_canonical_body_digest,
+        check_and_record_signed_rpc_nonce, normalize_tonic_rpc_audience, sign_ns_scanner_capability, sign_put_file_capability,
+        sign_tonic_rpc_response_proof, tonic_boot_epoch_challenge, tonic_boot_epoch_response_headers,
+        tonic_rpc_auth_failure_reason, verify_put_file_auth_trailer, verify_rpc_signature, verify_tonic_canonical_body_digest,
         verify_tonic_mutation_body_digest, verify_tonic_rpc_signature_with_bootstrap,
     };
     #[cfg(test)]
@@ -1732,16 +1731,8 @@ pub(crate) fn verify_put_file_auth_trailer(
     ecstore_rpc::verify_put_file_auth_trailer(url, method, nonce, trailer)
 }
 
-pub(crate) fn sign_ns_scanner_capability_with_tier_registry_generation(
-    challenge: uuid::Uuid,
-    server_epoch: uuid::Uuid,
-    supports_tier_registry_generation: bool,
-) -> std::io::Result<Vec<u8>> {
-    ecstore_rpc::sign_ns_scanner_capability_with_tier_registry_generation(
-        challenge,
-        server_epoch,
-        supports_tier_registry_generation,
-    )
+pub(crate) fn sign_ns_scanner_capability(challenge: uuid::Uuid, server_epoch: uuid::Uuid) -> std::io::Result<Vec<u8>> {
+    ecstore_rpc::sign_ns_scanner_capability(challenge, server_epoch)
 }
 
 pub(crate) fn sign_put_file_capability(

scripts/test/vault_ha_kms_live.sh

@@ -241,7 +241,7 @@ env \
   RUSTFS_TEST_VAULT_FAILOVER_MARKER="$MARKER" \
   RUSTFS_TEST_VAULT_OLD_LEADER="$OLD_LEADER" \
   cargo test -p rustfs-kms --test vault_ha_failover_live \
-    vault_raft_leader_failure_recovers_kv2_and_transit_decrypts -- \
+    vault_raft_leader_failure_preserves_kv2_and_transit_decrypts -- \
     --ignored --nocapture --test-threads=1 &
 TEST_PID=$!