fix(scanner): own fenced publication mutations

Remove the static lease budget rejection and carry an effective absolute deadline into storage-owned publication scopes. Keep remote grants until a scope reaches a safe terminal state.

Co-Authored-By: heihutu <heihutu@gmail.com>
This commit is contained in:
houseme
2026-08-30 02:02:05 +08:00
parent 2947d22cec
commit 76d2eddbf9
10 changed files with 344 additions and 75 deletions
+49 -16
View File
@@ -420,6 +420,7 @@ struct ScannerPublicationCommitScopeInner {
/// future. A detached mutation task keeps the scope alive and therefore
/// keeps this guard alive until it reports a terminal state.
movement_permit: Mutex<Option<OwnedRwLockReadGuard<()>>>,
lease_release_safe: Arc<std::sync::atomic::AtomicBool>,
}
/// Storage-owned ownership scope for one fenced scanner metadata mutation.
@@ -453,6 +454,25 @@ impl ScannerPublicationCommitScope {
remote_lease_tokens: Vec<Uuid>,
movement_permit: OwnedRwLockReadGuard<()>,
) -> Self {
Self::new_storage_owned_with_release_flag(
expected_movement_epoch,
safe_deadline,
remote_lease_tokens,
movement_permit,
Arc::new(std::sync::atomic::AtomicBool::new(true)),
)
}
pub(crate) fn new_storage_owned_with_release_flag(
expected_movement_epoch: u64,
safe_deadline: tokio::time::Instant,
remote_lease_tokens: Vec<Uuid>,
movement_permit: OwnedRwLockReadGuard<()>,
lease_release_safe: Arc<std::sync::atomic::AtomicBool>,
) -> Self {
// Admission itself is not a safe release point. The flag becomes true
// only after the storage mutation owner reports a terminal state.
lease_release_safe.store(false, Ordering::Release);
Self {
inner: Arc::new(ScannerPublicationCommitScopeInner {
expected_movement_epoch,
@@ -462,6 +482,7 @@ impl ScannerPublicationCommitScope {
state: AtomicU8::new(SCANNER_PUBLICATION_SCOPE_ADMITTED),
completed: Notify::new(),
movement_permit: Mutex::new(Some(movement_permit)),
lease_release_safe,
}),
}
}
@@ -537,21 +558,20 @@ impl ScannerPublicationCommitScope {
}
pub fn mark_aborted_before_commit(&self) -> bool {
for expected in [SCANNER_PUBLICATION_SCOPE_ADMITTED, SCANNER_PUBLICATION_SCOPE_IN_FLIGHT] {
if self
.inner
.state
.compare_exchange(
expected,
SCANNER_PUBLICATION_SCOPE_ABORTED_BEFORE_COMMIT,
Ordering::AcqRel,
Ordering::Acquire,
)
.is_ok()
{
self.inner.completed.notify_waiters();
return true;
}
if self
.inner
.state
.compare_exchange(
SCANNER_PUBLICATION_SCOPE_ADMITTED,
SCANNER_PUBLICATION_SCOPE_ABORTED_BEFORE_COMMIT,
Ordering::AcqRel,
Ordering::Acquire,
)
.is_ok()
{
self.inner.lease_release_safe.store(true, Ordering::Release);
self.inner.completed.notify_waiters();
return true;
}
false
}
@@ -565,7 +585,12 @@ impl ScannerPublicationCommitScope {
.state
.compare_exchange(SCANNER_PUBLICATION_SCOPE_IN_FLIGHT, terminal.as_u8(), Ordering::AcqRel, Ordering::Acquire)
.is_ok()
.then(|| self.inner.completed.notify_waiters())
.then(|| {
if terminal.permits_lease_release() {
self.inner.lease_release_safe.store(true, Ordering::Release);
}
self.inner.completed.notify_waiters()
})
.is_some()
}
@@ -595,6 +620,14 @@ impl ScannerPublicationCommitScope {
}
}
impl Drop for ScannerPublicationCommitScopeInner {
fn drop(&mut self) {
if !ScannerPublicationCommitState::from_u8(self.state.load(Ordering::Acquire)).permits_lease_release() {
self.lease_release_safe.store(false, Ordering::Release);
}
}
}
#[derive(Default, Clone)]
pub struct ObjectOptions {
// Use the maximum parity (N/2), used when saving server configuration files
+44 -1
View File
@@ -105,7 +105,9 @@ use crate::{
SnapshotLeaseToken, UpdateMetadataOpts, endpoint::Endpoint, error::DiskError, format::FormatV3,
},
error::{StorageError, to_object_err},
object_api::{GetObjectReader, NamespaceLockFence, ObjectInfo, ObjectLockConfigSnapshot, PutObjReader},
object_api::{
GetObjectReader, NamespaceLockFence, ObjectInfo, ObjectLockConfigSnapshot, PutObjReader, ScannerPublicationCommitScope,
},
// event::name::EventName,
services::event_notification::{EventArgs, send_event},
store::init_format::{
@@ -3937,6 +3939,47 @@ impl SetDisks {
owner.scanner_data_usage_publication_admission_guard().await
}
/// Acquire a storage-owned scanner publication scope for this set's
/// instance movement fence. The scope keeps the read permit alive across
/// scanner future cancellation until the mutation owner drains.
pub async fn scanner_data_usage_publication_commit_scope(
&self,
expected_movement_epoch: u64,
safe_deadline: tokio::time::Instant,
remote_lease_tokens: Vec<Uuid>,
) -> Option<ScannerPublicationCommitScope> {
let (movement_permit, epoch) = self.scanner_data_usage_publication_admission_guard().await?;
if epoch != expected_movement_epoch {
return None;
}
Some(ScannerPublicationCommitScope::new_storage_owned(
epoch,
safe_deadline,
remote_lease_tokens,
movement_permit,
))
}
pub async fn scanner_data_usage_publication_commit_scope_with_release_flag(
&self,
expected_movement_epoch: u64,
safe_deadline: tokio::time::Instant,
remote_lease_tokens: Vec<Uuid>,
lease_release_safe: Arc<std::sync::atomic::AtomicBool>,
) -> Option<ScannerPublicationCommitScope> {
let (movement_permit, epoch) = self.scanner_data_usage_publication_admission_guard().await?;
if epoch != expected_movement_epoch {
return None;
}
Some(ScannerPublicationCommitScope::new_storage_owned_with_release_flag(
epoch,
safe_deadline,
remote_lease_tokens,
movement_permit,
lease_release_safe,
))
}
/// Whether both sets' namespace-lock implementations cover the same object key.
pub(crate) async fn shares_namespace_lock_domain(&self, other: &Self) -> bool {
match (self.ctx.is_dist_erasure().await, other.ctx.is_dist_erasure().await) {
+28 -2
View File
@@ -3394,7 +3394,12 @@ impl SetDisks {
let commit_tmp_dir = tmp_dir.clone();
let commit_object_lock_guard = object_lock_guard.take();
let commit_bucket_lifecycle_guard = bucket_lifecycle_guard.take();
let commit_allows_early_ack = commit_object_lock_guard.is_some();
let commit_scanner_publication_scope = opts.scanner_publication_commit_scope.clone();
// A scanner publication scope owns the movement permit until the
// complete rename fan-out drains. Keep this path synchronous so
// its terminal state is known before the coordinator releases
// remote leases.
let commit_allows_early_ack = commit_object_lock_guard.is_some() && commit_scanner_publication_scope.is_none();
let detach_commit_owner = commit_allows_early_ack || commit_bucket_lifecycle_guard.is_some() || quota_mutation_fence;
let commit_write_path_label = write_path.metric_label();
let commit_is_versioned = opts.versioned || opts.version_suspended;
@@ -3491,7 +3496,7 @@ impl SetDisks {
}
Ok(())
};
let pre_rename_result = if cancellation.is_some() || request_cancellation.is_some() {
let mut pre_rename_result = if cancellation.is_some() || request_cancellation.is_some() {
tokio::select! {
biased;
_ = wait_for_put_object_commit_cancellation(cancellation.as_ref(), request_cancellation.as_ref()) => {
@@ -3502,7 +3507,21 @@ impl SetDisks {
} else {
pre_rename.await
};
if pre_rename_result.is_ok()
&& let Some(scope) = commit_scanner_publication_scope.as_ref()
&& let Err(err) = scope.try_begin()
{
let _ = scope.mark_aborted_before_commit();
pre_rename_result = Err(Error::other(format!("scanner publication commit scope cannot start: {err:?}")));
}
if let Err(err) = pre_rename_result {
if let Some(scope) = commit_scanner_publication_scope.as_ref() {
if scope.state() == crate::object_api::ScannerPublicationCommitState::Admitted {
let _ = scope.mark_aborted_before_commit();
} else {
let _ = scope.mark_indeterminate();
}
}
SetDisks::abort_quota_reservation_after_fence(
quota_reservation,
&commit_disks,
@@ -3540,6 +3559,13 @@ impl SetDisks {
),
)
.await;
if let Some(scope) = commit_scanner_publication_scope.as_ref() {
if rename_result.is_ok() {
let _ = scope.mark_committed();
} else {
let _ = scope.mark_indeterminate();
}
}
#[cfg(any(test, feature = "test-util"))]
if rename_result.is_ok() {
pause_put_object_commit(&commit_bucket, &commit_object, PutObjectCommitPause::AfterRenameQuorum).await;
+28
View File
@@ -544,6 +544,30 @@ impl ECStore {
))
}
/// Variant used by the scanner supervisor to observe whether a scope was
/// dropped before reaching a safe terminal state. The flag is in-memory
/// only and lets the supervisor avoid releasing remote leases on an
/// indeterminate cancellation path.
pub async fn scanner_data_usage_publication_commit_scope_with_release_flag(
&self,
expected_movement_epoch: u64,
safe_deadline: tokio::time::Instant,
remote_lease_tokens: Vec<Uuid>,
lease_release_safe: Arc<std::sync::atomic::AtomicBool>,
) -> Option<ScannerPublicationCommitScope> {
let (movement_permit, epoch) = self.scanner_data_usage_publication_admission_guard().await?;
if epoch != expected_movement_epoch {
return None;
}
Some(ScannerPublicationCommitScope::new_storage_owned_with_release_flag(
epoch,
safe_deadline,
remote_lease_tokens,
movement_permit,
lease_release_safe,
))
}
/// Capture the current publication epoch without holding the movement
/// gate across backend I/O. Callers must re-admit the same epoch before a
/// mutation commits.
@@ -1521,6 +1545,10 @@ mod tests {
.expect("a second idle publication scope should be granted");
scope.try_begin().expect("scope should enter the mutation state");
scope.cancel();
assert!(
!scope.mark_aborted_before_commit(),
"an in-flight mutation cannot claim pre-commit abort without storage proof"
);
assert!(scope.mark_indeterminate());
assert_eq!(
scope.wait_for_completion().await,
+104 -5
View File
@@ -39,11 +39,11 @@ use storage_api::owner::{
EcstoreListPathRawOptions, EcstoreNsScannerOpenRequest, EcstoreObjectOpts, EcstoreReplicationConfigurationExt,
EcstoreReplicationScannerBridge, EcstoreResultType, EcstoreScanGuard, EcstoreSetDisks, EcstoreStorageError, EcstoreStore,
EcstoreVersioningApi, HTTPPreconditions, HTTPRangeSpec, ObjectIO, ObjectOperations, ObjectToDelete,
ScannerReplicationHealObject, ScannerReplicationHealResult, ScannerReplicationQueueAdmission, ecstore_apply_expiry_rule,
ecstore_apply_transition_rule, ecstore_expiry_state_handle, ecstore_get_global_tier_config_mgr, ecstore_get_lifecycle_config,
ecstore_get_object_lock_config, ecstore_get_replication_config, ecstore_invalidate_admin_data_usage_snapshot_cache,
ecstore_invalidate_data_usage_snapshot_cache, ecstore_is_erasure, ecstore_is_erasure_sd,
ecstore_is_reserved_or_invalid_bucket, ecstore_list_path_raw, ecstore_object_opts_from_object_info,
ScannerPublicationCommitScope, ScannerReplicationHealObject, ScannerReplicationHealResult, ScannerReplicationQueueAdmission,
ecstore_apply_expiry_rule, ecstore_apply_transition_rule, ecstore_expiry_state_handle, ecstore_get_global_tier_config_mgr,
ecstore_get_lifecycle_config, ecstore_get_object_lock_config, ecstore_get_replication_config,
ecstore_invalidate_admin_data_usage_snapshot_cache, ecstore_invalidate_data_usage_snapshot_cache, ecstore_is_erasure,
ecstore_is_erasure_sd, ecstore_is_reserved_or_invalid_bucket, ecstore_list_path_raw, ecstore_object_opts_from_object_info,
ecstore_path2_bucket_object, ecstore_path2_bucket_object_with_base_path, ecstore_read_config,
ecstore_replace_bucket_usage_memory_from_info, ecstore_resolve_object_store_handle, ecstore_save_config,
scanner_replication_config_for_lifecycle_eval,
@@ -55,6 +55,7 @@ use storage_api::owner::{
ecstore_new_disk,
};
use tokio_util::sync::CancellationToken;
use uuid::Uuid;
pub mod data_usage_define;
pub mod error;
@@ -804,6 +805,30 @@ pub(crate) async fn save_config_shared_with_preconditions_and_lease_fence<S>(
preconditions: HTTPPreconditions,
scanner_publication_lease_fence: Option<&str>,
) -> EcstoreResult<ScannerObjectInfo>
where
S: ScannerObjectIO,
{
save_config_shared_with_preconditions_and_lease_fence_and_scope(
api,
file,
data,
sha256hex,
preconditions,
scanner_publication_lease_fence,
None,
)
.await
}
pub(crate) async fn save_config_shared_with_preconditions_and_lease_fence_and_scope<S>(
api: Arc<S>,
file: &str,
data: Bytes,
sha256hex: Option<String>,
preconditions: HTTPPreconditions,
scanner_publication_lease_fence: Option<&str>,
scanner_publication_commit_scope: Option<ScannerPublicationCommitScope>,
) -> EcstoreResult<ScannerObjectInfo>
where
S: ScannerObjectIO,
{
@@ -822,6 +847,7 @@ where
&ScannerObjectOptions {
max_parity: true,
http_preconditions: Some(preconditions),
scanner_publication_commit_scope,
user_defined,
..Default::default()
},
@@ -886,6 +912,27 @@ pub trait ScannerConfigObjectDelete: Send + Sync + std::fmt::Debug + 'static {
async fn scanner_data_usage_publication_admission(&self) -> Option<ScannerDataUsagePublicationAdmission> {
None
}
/// Acquire a storage-owned scope for a fenced scanner metadata mutation.
/// Implementations without a storage movement owner fail closed.
async fn scanner_data_usage_publication_commit_scope(
&self,
_expected_movement_epoch: u64,
_safe_deadline: tokio::time::Instant,
_remote_lease_tokens: Vec<Uuid>,
) -> Option<ScannerPublicationCommitScope> {
None
}
async fn scanner_data_usage_publication_commit_scope_with_release_flag(
&self,
_expected_movement_epoch: u64,
_safe_deadline: tokio::time::Instant,
_remote_lease_tokens: Vec<Uuid>,
_lease_release_safe: Arc<std::sync::atomic::AtomicBool>,
) -> Option<ScannerPublicationCommitScope> {
None
}
}
pub struct ScannerDataUsagePublicationAdmission {
@@ -929,6 +976,32 @@ impl ScannerConfigObjectDelete for ECStore {
let (read_guard, epoch) = self.scanner_data_usage_publication_admission_guard().await?;
Some(ScannerDataUsagePublicationAdmission::fenced(read_guard, epoch))
}
async fn scanner_data_usage_publication_commit_scope(
&self,
expected_movement_epoch: u64,
safe_deadline: tokio::time::Instant,
remote_lease_tokens: Vec<Uuid>,
) -> Option<ScannerPublicationCommitScope> {
self.scanner_data_usage_publication_commit_scope(expected_movement_epoch, safe_deadline, remote_lease_tokens)
.await
}
async fn scanner_data_usage_publication_commit_scope_with_release_flag(
&self,
expected_movement_epoch: u64,
safe_deadline: tokio::time::Instant,
remote_lease_tokens: Vec<Uuid>,
lease_release_safe: Arc<std::sync::atomic::AtomicBool>,
) -> Option<ScannerPublicationCommitScope> {
self.scanner_data_usage_publication_commit_scope_with_release_flag(
expected_movement_epoch,
safe_deadline,
remote_lease_tokens,
lease_release_safe,
)
.await
}
}
#[async_trait::async_trait]
@@ -946,6 +1019,32 @@ impl ScannerConfigObjectDelete for SetDisks {
let (read_guard, epoch) = self.scanner_data_usage_publication_admission_guard().await?;
Some(ScannerDataUsagePublicationAdmission::fenced(read_guard, epoch))
}
async fn scanner_data_usage_publication_commit_scope(
&self,
expected_movement_epoch: u64,
safe_deadline: tokio::time::Instant,
remote_lease_tokens: Vec<Uuid>,
) -> Option<ScannerPublicationCommitScope> {
self.scanner_data_usage_publication_commit_scope(expected_movement_epoch, safe_deadline, remote_lease_tokens)
.await
}
async fn scanner_data_usage_publication_commit_scope_with_release_flag(
&self,
expected_movement_epoch: u64,
safe_deadline: tokio::time::Instant,
remote_lease_tokens: Vec<Uuid>,
lease_release_safe: Arc<std::sync::atomic::AtomicBool>,
) -> Option<ScannerPublicationCommitScope> {
self.scanner_data_usage_publication_commit_scope_with_release_flag(
expected_movement_epoch,
safe_deadline,
remote_lease_tokens,
lease_release_safe,
)
.await
}
}
#[cfg(test)]
+20 -23
View File
@@ -16,6 +16,7 @@ use std::collections::BTreeMap;
use std::future::Future;
#[cfg(test)]
use std::sync::Mutex as StdMutex;
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::{Arc, LazyLock, RwLock};
use self::heal_info::{BackgroundHealInfoReadStatus, read_background_heal_info_with_epoch, save_background_heal_info_for_epoch};
@@ -63,7 +64,6 @@ use tokio_util::sync::CancellationToken;
use tokio_util::task::AbortOnDropHandle;
use tracing::{debug, error, info, instrument, warn};
use crate::storage_api::owner::SCANNER_PUBLICATION_LEASE_TTL_MS;
use crate::storage_api::scan::{
BucketOperations, BucketOptions, NamespaceLocking as _, SCANNER_ACTIVITY_LEGACY_PROTOCOL_VERSION,
SCANNER_ACTIVITY_PREVIOUS_PROTOCOL_VERSION, SCANNER_ACTIVITY_PROTOCOL_VERSION,
@@ -1205,10 +1205,6 @@ fn data_usage_persist_timeout() -> Duration {
DataUsageCache::persistence_timeout()
}
fn scanner_publication_lease_budget_allows_persistence(timeout: Duration) -> bool {
timeout < Duration::from_millis(SCANNER_PUBLICATION_LEASE_TTL_MS)
}
#[cfg(not(test))]
const SCANNER_CYCLE_EPOCH_FENCE_TIMEOUT: Duration = Duration::from_secs(30);
#[cfg(test)]
@@ -1493,11 +1489,6 @@ async fn run_data_scanner_cycle_with_budget(
let mut remote_publication_leases = None;
let remote_lease_defer_reason = if remote_publication_lease_targets.is_empty() {
None
} else if !scanner_publication_lease_budget_allows_persistence(usage_persist_timeout) {
// The lease is intentionally fixed-duration and has no renewal path.
// Refuse a persistence budget that could outlive it instead of
// allowing the peer to admit movement while a local PUT is in flight.
Some(ScannerCycleDeferReason::PublicationLeaseBudgetExceeded)
} else if let Some(notification_system) = storeapi.notification_system() {
match notification_system
.acquire_scanner_publication_leases(remote_publication_lease_targets.clone())
@@ -1548,26 +1539,16 @@ async fn run_data_scanner_cycle_with_budget(
remote_lease_fence.is_some(),
))
.then_some(ScannerCycleDeferReason::ActivityBaselineUnavailable);
let remote_lease_covers_persistence = remote_lease_deadline.is_none_or(|deadline| {
std::time::Instant::now()
.checked_add(usage_persist_timeout)
.is_some_and(|latest_finish| latest_finish < deadline)
});
let publication_defer_reason = publication_defer_reason
.or(remote_lease_defer_reason)
.or(remote_lease_fence_defer_reason);
let publication_defer_reason = (!remote_lease_covers_persistence)
.then_some(ScannerCycleDeferReason::PublicationLeaseDeadlineExceeded)
.or(publication_defer_reason);
// Include reasons discovered while acquiring or validating remote leases.
// In particular, the static budget gate above is reached after the scan
// result is classified, so computing this flag earlier would suppress its
// deferred metric.
let publication_deferred = publication_defer_reason.is_some();
let budget_elapsed = cycle_budget.budget_elapsed() && !ctx.is_cancelled();
let remote_lease_probe = remote_publication_leases
.as_ref()
.map(|(notification_system, grants)| (Arc::clone(notification_system), grants.clone()));
let remote_lease_release_safe = Arc::new(AtomicBool::new(true));
let mut usage_persist_outcome = match publication_defer_reason {
Some(reason) => {
drop(receiver);
@@ -1581,6 +1562,11 @@ async fn run_data_scanner_cycle_with_budget(
let ctx_clone = ctx.clone();
let route_probe_store = storeapi.clone();
let remote_lease_fence = remote_lease_fence.clone();
let remote_lease_release_safe_for_task = Arc::clone(&remote_lease_release_safe);
let remote_lease_tokens = remote_publication_leases
.as_ref()
.map(|(_, grants)| grants.iter().map(|grant| grant.lease.token).collect())
.unwrap_or_default();
let mut usage_persist_task = AbortOnDropHandle::new(tokio::spawn(async move {
store_data_usage_in_backend_with_outcome_for_epoch_and_baseline_and_route_probe_for_publication_epoch_and_lease_fence(
ctx_clone,
@@ -1592,7 +1578,9 @@ async fn run_data_scanner_cycle_with_budget(
publication_epoch,
remote_lease_deadline,
remote_lease_fence,
),
)
.with_remote_lease_tokens(remote_lease_tokens)
.with_lease_release_flag(remote_lease_release_safe_for_task),
move || {
let storeapi = route_probe_store.clone();
let remote_lease_probe = remote_lease_probe.clone();
@@ -1657,7 +1645,16 @@ async fn run_data_scanner_cycle_with_budget(
let lease_expired = remote_publication_leases
.as_ref()
.is_some_and(|(_, grants)| grants.iter().any(|grant| !grant.lease.is_valid()));
if let Some((notification_system, grants)) = remote_publication_leases.take() {
if !remote_lease_release_safe.load(Ordering::Acquire) {
// A cancelled or detached storage mutation did not report a safe
// terminal state. Keep remote grants until their own expiry rather
// than releasing movement admission while a commit may be unknown.
usage_persist_outcome = if usage_persist_outcome == DataUsagePersistOutcome::Failed {
DataUsagePersistOutcome::Failed
} else {
DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::PublicationLeaseDeadlineExceeded)
};
} else if let Some((notification_system, grants)) = remote_publication_leases.take() {
let release_result = notification_system.release_scanner_publication_leases(grants).await;
let lease_release_failed = release_result.is_err();
if lease_expired || lease_release_failed {
+1 -13
View File
@@ -4628,7 +4628,6 @@ fn scanner_cycle_cache_floor_stays_pending_during_deferred_usage_publication() {
for reason in [
ScannerCycleDeferReason::DataMovement,
ScannerCycleDeferReason::ActivityBaselineUnavailable,
ScannerCycleDeferReason::PublicationLeaseBudgetExceeded,
ScannerCycleDeferReason::PublicationLeaseDeadlineExceeded,
ScannerCycleDeferReason::PublicationLeaseReleaseFailed,
] {
@@ -4801,18 +4800,7 @@ fn data_usage_persist_wait_covers_cache_retries_and_backup() {
}
#[test]
fn scanner_publication_lease_budget_has_a_strict_ttl_boundary() {
let ttl = Duration::from_millis(SCANNER_PUBLICATION_LEASE_TTL_MS);
assert!(scanner_publication_lease_budget_allows_persistence(
ttl.saturating_sub(Duration::from_millis(1))
));
assert!(!scanner_publication_lease_budget_allows_persistence(ttl));
assert!(!scanner_publication_lease_budget_allows_persistence(ttl + Duration::from_millis(1)));
assert_eq!(
ScannerCycleDeferReason::PublicationLeaseBudgetExceeded.as_str(),
"publication_lease_budget_exceeded"
);
fn scanner_publication_lease_deadline_reason_remains_distinct() {
assert_eq!(
ScannerCycleDeferReason::PublicationLeaseDeadlineExceeded.as_str(),
"publication_lease_deadline_exceeded"
+65 -7
View File
@@ -13,7 +13,10 @@
// limitations under the License.
/// Data-usage snapshot persistence: CAS store pipeline, epoch baselines, and observed-snapshot cleanup.
use super::*;
use crate::storage_api::owner::ScannerPublicationCommitState;
use std::collections::HashMap;
use std::sync::atomic::AtomicBool;
use uuid::Uuid;
#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
pub(super) enum DataUsagePersistOutcome {
@@ -34,6 +37,16 @@ fn remote_lease_expired(deadline: Option<std::time::Instant>) -> bool {
deadline.is_some_and(|deadline| std::time::Instant::now() >= deadline)
}
fn scanner_publication_scope_deadline(
persist_timeout: Duration,
remote_lease_deadline: Option<std::time::Instant>,
) -> tokio::time::Instant {
let configured_deadline = tokio::time::Instant::now() + persist_timeout;
remote_lease_deadline
.map(tokio::time::Instant::from_std)
.map_or(configured_deadline, |lease_deadline| configured_deadline.min(lease_deadline))
}
#[derive(Clone, Debug)]
pub(super) struct DataUsagePersistBaseline {
pub(super) data: Option<Bytes>,
@@ -126,6 +139,8 @@ pub(super) struct ScannerPublicationFence {
pub(super) expected_publication_epoch: Option<u64>,
pub(super) remote_lease_deadline: Option<std::time::Instant>,
pub(super) scanner_publication_lease_fence: Option<String>,
pub(super) remote_lease_tokens: Vec<Uuid>,
pub(super) lease_release_safe: Arc<AtomicBool>,
}
impl ScannerPublicationFence {
@@ -138,8 +153,20 @@ impl ScannerPublicationFence {
expected_publication_epoch,
remote_lease_deadline,
scanner_publication_lease_fence,
remote_lease_tokens: Vec::new(),
lease_release_safe: Arc::new(AtomicBool::new(true)),
}
}
pub(super) fn with_remote_lease_tokens(mut self, remote_lease_tokens: Vec<Uuid>) -> Self {
self.remote_lease_tokens = remote_lease_tokens;
self
}
pub(super) fn with_lease_release_flag(mut self, lease_release_safe: Arc<AtomicBool>) -> Self {
self.lease_release_safe = lease_release_safe;
self
}
}
#[derive(Debug)]
@@ -290,6 +317,8 @@ where
expected_publication_epoch,
remote_lease_deadline,
scanner_publication_lease_fence,
remote_lease_tokens,
lease_release_safe,
} = publication_fence;
let mut outcome = DataUsagePersistOutcome::NoUpdate;
let mut next_baseline = initial_baseline;
@@ -580,25 +609,54 @@ where
let done_save = Metrics::time(Metric::SaveUsage);
let save_result = {
let Some(_publication_admission) =
scanner_publication_admission_for_epoch(storeapi.clone(), publication_epoch_for_save).await
else {
done_save();
break DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement);
let publication_scope = storeapi
.scanner_data_usage_publication_commit_scope_with_release_flag(
publication_epoch_for_save,
scanner_publication_scope_deadline(data_usage_persist_timeout(), remote_lease_deadline),
remote_lease_tokens.clone(),
Arc::clone(&lease_release_safe),
)
.await;
let legacy_publication_admission = if publication_scope.is_none() {
let Some(admission) =
scanner_publication_admission_for_epoch(storeapi.clone(), publication_epoch_for_save).await
else {
done_save();
break DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement);
};
Some(admission)
} else {
None
};
if remote_lease_expired(remote_lease_deadline) {
done_save();
break DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::PublicationLeaseDeadlineExceeded);
}
save_config_shared_with_preconditions_and_lease_fence(
let save_result = crate::save_config_shared_with_preconditions_and_lease_fence_and_scope(
storeapi.clone(),
target_path,
data.clone(),
sha256hex.clone(),
revision.preconditions(),
scanner_publication_lease_fence.as_deref(),
publication_scope.clone(),
)
.await
.await;
drop(legacy_publication_admission);
if let Some(scope) = publication_scope {
match scope.wait_for_completion().await {
ScannerPublicationCommitState::Committed | ScannerPublicationCommitState::AbortedBeforeCommit => {
save_result
}
ScannerPublicationCommitState::Indeterminate
| ScannerPublicationCommitState::Admitted
| ScannerPublicationCommitState::InFlight => Err(EcstoreError::other(
"scanner publication commit scope did not reach a safe terminal state",
)),
}
} else {
save_result
}
};
done_save();
-5
View File
@@ -574,10 +574,6 @@ pub(crate) async fn scanner_set_disk_inventory(set: &SetDisks) -> Vec<Arc<Disk>>
pub(crate) enum ScannerCycleDeferReason {
ActivityBaselineUnavailable,
DataMovement,
/// The configured persistence budget cannot fit within the fixed remote
/// publication-lease TTL. This is a deterministic configuration/contract
/// mismatch, not evidence that a peer activity probe failed.
PublicationLeaseBudgetExceeded,
/// A granted lease's absolute deadline cannot cover the persistence
/// operation. This can occur even when the configured budget fits the
/// nominal TTL because lease acquisition consumed part of the window.
@@ -592,7 +588,6 @@ impl ScannerCycleDeferReason {
match self {
Self::ActivityBaselineUnavailable => "activity_baseline_unavailable",
Self::DataMovement => "data_movement",
Self::PublicationLeaseBudgetExceeded => "publication_lease_budget_exceeded",
Self::PublicationLeaseDeadlineExceeded => "publication_lease_deadline_exceeded",
Self::PublicationLeaseReleaseFailed => "publication_lease_release_failed",
}
+5 -3
View File
@@ -92,7 +92,9 @@ pub(crate) use rustfs_ecstore::api::event::{EventArgs as EcstoreEventArgs, send_
pub(crate) use rustfs_ecstore::api::layout::{
EndpointServerPools as EcstoreEndpointServerPools, Endpoints as EcstoreEndpoints, PoolEndpoints as EcstorePoolEndpoints,
};
pub(crate) use rustfs_ecstore::api::object::SCANNER_PUBLICATION_LEASE_FENCE_METADATA_KEY;
pub(crate) use rustfs_ecstore::api::object::{
SCANNER_PUBLICATION_LEASE_FENCE_METADATA_KEY, ScannerPublicationCommitScope, ScannerPublicationCommitState,
};
#[cfg(test)]
pub(crate) use rustfs_ecstore::api::rebalance::{
RebalStatus as EcstoreRebalStatus, RebalanceInfo as EcstoreRebalanceInfo, RebalanceMeta as EcstoreRebalanceMeta,
@@ -106,9 +108,9 @@ pub(crate) use rustfs_ecstore::api::runtime::{
setup_is_erasure_sd as ecstore_is_erasure_sd,
};
pub(crate) use rustfs_ecstore::api::set_disk::SetDisks as EcstoreSetDisks;
pub(crate) use rustfs_ecstore::api::storage::ECStore as EcstoreStore;
#[cfg(test)]
pub(crate) use rustfs_ecstore::api::storage::init_local_disks_with_instance_ctx as ecstore_init_local_disks_with_instance_ctx;
pub(crate) use rustfs_ecstore::api::storage::{ECStore as EcstoreStore, SCANNER_PUBLICATION_LEASE_TTL_MS};
use rustfs_storage_api as storage_contracts;
pub(crate) mod owner {
@@ -125,7 +127,7 @@ pub(crate) mod owner {
EcstoreNsScannerOpenRequest, EcstoreObjectLockConfiguration, EcstoreObjectOpts, EcstoreReplicationConfigurationExt,
EcstoreReplicationScannerBridge, EcstoreResultType, EcstoreScanGuard, EcstoreSetDisks, EcstoreStorageError, EcstoreStore,
EcstoreVersioningApi, EcstoreVersioningConfiguration, SCANNER_PUBLICATION_LEASE_FENCE_METADATA_KEY,
SCANNER_PUBLICATION_LEASE_TTL_MS, ScannerReplicationHealObject, ScannerReplicationHealResult,
ScannerPublicationCommitScope, ScannerPublicationCommitState, ScannerReplicationHealObject, ScannerReplicationHealResult,
ScannerReplicationQueueAdmission, ecstore_apply_expiry_rule, ecstore_apply_transition_rule, ecstore_expiry_state_handle,
ecstore_get_global_tier_config_mgr, ecstore_get_lifecycle_config, ecstore_get_object_lock_config,
ecstore_get_replication_config, ecstore_invalidate_admin_data_usage_snapshot_cache,