fix(storage): harden ODM and scanner publication (#7187)

* fix(storage): harden ODM and scanner publication

* fix(app): simplify absent SSE configuration matching

* test(heal): settle PUT rename tails before disk-wipe fixtures

* fix(ecstore): remove duplicate local rename implementation

Keep the canonical commit module after concurrent storage changes merged.
The control-write and rollback changes are already present there.

Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>

* fix(ci): satisfy new clippy lints

* style(scanner): order merged test imports

* fix(scanner): invalidate bucket work after namespace completion

* fix(scanner): fence cached snapshots by scan execution

---------

Co-authored-by: houseme <housemecn@gmail.com>
Co-authored-by: heihutu <heihutu@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
This commit is contained in:
Zhengchao An
2026-09-05 21:47:12 +08:00
committed by GitHub
parent 447f3c704b
commit e2a921bc16
35 changed files with 2252 additions and 274 deletions
+112
View File
@@ -78,6 +78,21 @@ pub(crate) struct ScannerPublicationLeaseEntry {
pub(crate) _operation_guard: OwnedRwLockReadGuard<()>,
}
pub(crate) struct NamespaceCommitGuard {
ctx: Arc<InstanceContext>,
counted: bool,
}
impl Drop for NamespaceCommitGuard {
fn drop(&mut self) {
if self.counted {
// Publish the new generation before a zero-pending publication probe.
self.ctx.advance_namespace_commit_generation();
self.ctx.namespace_commits.fetch_sub(1, Ordering::AcqRel);
}
}
}
/// Runtime state owned by a single `ECStore` instance.
///
/// This is intentionally minimal in the first migration slice; subsequent
@@ -209,9 +224,13 @@ pub struct InstanceContext {
/// Last storage-owned movement snapshot observed under the operation
/// gate. SetDisks cache writers fail closed until ECStore refreshes it.
scanner_publication_state: AtomicU8,
namespace_commits: AtomicU64,
namespace_commit_generation: AtomicU64,
/// Resolves object-encryption material at the application boundary.
object_encryption_resolver: OnceLock<Arc<dyn ObjectEncryptionResolver>>,
tier_delete_journal_recovery_stores: std::sync::Mutex<HashSet<Uuid>>,
#[cfg(test)]
suppress_tier_delete_journal_recovery: bool,
transition_transaction_recovery_stores: std::sync::Mutex<HashSet<Uuid>>,
tier_delete_journal_recovery_wakeup: tokio::sync::Notify,
}
@@ -256,8 +275,12 @@ impl InstanceContext {
data_movement_generation_exhausted: AtomicBool::new(false),
data_movement_generation_notify: Arc::new(Notify::new()),
scanner_publication_state: AtomicU8::new(SCANNER_PUBLICATION_STATE_UNKNOWN),
namespace_commits: AtomicU64::new(0),
namespace_commit_generation: AtomicU64::new(0),
object_encryption_resolver: OnceLock::new(),
tier_delete_journal_recovery_stores: std::sync::Mutex::new(HashSet::new()),
#[cfg(test)]
suppress_tier_delete_journal_recovery: false,
transition_transaction_recovery_stores: std::sync::Mutex::new(HashSet::new()),
tier_delete_journal_recovery_wakeup: tokio::sync::Notify::new(),
}
@@ -385,6 +408,36 @@ impl InstanceContext {
&& self.scanner_publication_state.load(Ordering::Acquire) == SCANNER_PUBLICATION_STATE_ALLOWED
}
pub(crate) fn begin_namespace_commit(self: &Arc<Self>) -> Arc<NamespaceCommitGuard> {
let counted = self
.namespace_commits
.fetch_update(Ordering::AcqRel, Ordering::Acquire, |count| count.checked_add(1))
.is_ok();
if counted {
self.advance_namespace_commit_generation();
} else {
self.namespace_commit_generation.store(u64::MAX, Ordering::Release);
}
Arc::new(NamespaceCommitGuard {
ctx: Arc::clone(self),
counted,
})
}
fn advance_namespace_commit_generation(&self) {
let _ = self
.namespace_commit_generation
.fetch_update(Ordering::AcqRel, Ordering::Acquire, |generation| Some(generation.saturating_add(1)));
}
pub(crate) fn namespace_commit_generation(&self) -> u64 {
self.namespace_commit_generation.load(Ordering::Acquire)
}
pub(crate) fn namespace_commits_pending(&self) -> bool {
self.namespace_commits.load(Ordering::Acquire) != 0 || self.namespace_commit_generation() == u64::MAX
}
pub(crate) fn set_scanner_publication_state(&self, blocked: bool) {
self.scanner_publication_state.store(
if blocked {
@@ -640,12 +693,21 @@ impl InstanceContext {
}
pub(crate) fn mark_tier_delete_journal_recovery_started(&self, store_id: Uuid) -> bool {
#[cfg(test)]
if self.suppress_tier_delete_journal_recovery {
return false;
}
self.tier_delete_journal_recovery_stores
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner)
.insert(store_id)
}
#[cfg(test)]
pub(crate) fn suppress_tier_delete_journal_recovery_for_test(&mut self) {
self.suppress_tier_delete_journal_recovery = true;
}
pub(crate) fn mark_transition_transaction_recovery_started(&self, store_id: Uuid) -> bool {
self.transition_transaction_recovery_stores
.lock()
@@ -756,6 +818,50 @@ pub fn bootstrap_ctx() -> Arc<InstanceContext> {
mod tests {
use super::*;
#[test]
fn namespace_commit_guards_are_instance_local_and_count_until_last_owner() {
let first = Arc::new(InstanceContext::new());
let other = Arc::new(InstanceContext::new());
first.set_scanner_publication_state(false);
other.set_scanner_publication_state(false);
assert!(first.scanner_publication_state_allowed());
let one = first.begin_namespace_commit();
let shared_owner = Arc::clone(&one);
let two = first.begin_namespace_commit();
assert!(first.namespace_commits_pending());
assert!(first.scanner_publication_state_allowed(), "pending writes must not block scan admission");
assert_eq!(first.namespace_commit_generation(), 2);
assert!(!other.namespace_commits_pending());
assert_eq!(other.namespace_commit_generation(), 0);
assert!(other.scanner_publication_state_allowed());
drop(one);
assert_eq!(first.namespace_commit_generation(), 2);
drop(shared_owner);
assert!(first.namespace_commits_pending());
assert_eq!(first.namespace_commit_generation(), 3);
drop(two);
assert!(!first.namespace_commits_pending());
assert_eq!(first.namespace_commit_generation(), 4);
assert!(first.scanner_publication_state_allowed());
}
#[test]
fn namespace_commit_counter_exhaustion_keeps_publication_blocked() {
for (count, generation) in [(0, u64::MAX - 1), (u64::MAX, 0)] {
let ctx = Arc::new(InstanceContext::new());
ctx.set_scanner_publication_state(false);
ctx.namespace_commits.store(count, Ordering::Release);
ctx.namespace_commit_generation.store(generation, Ordering::Release);
let guard = ctx.begin_namespace_commit();
assert!(ctx.namespace_commits_pending());
assert_eq!(ctx.namespace_commit_generation(), u64::MAX);
drop(guard);
assert!(ctx.namespace_commits_pending());
assert_eq!(ctx.namespace_commit_generation(), u64::MAX);
assert_eq!(ctx.namespace_commits.load(Ordering::Acquire), count);
}
}
// The SetupType inputs must derive the exact (is_erasure,
// is_dist_erasure, is_erasure_sd) triples that the original three
// process-global erasure bools produced via update_erasure_type().
@@ -1073,6 +1179,12 @@ mod tests {
assert!(!ctx_a.mark_tier_delete_journal_recovery_started(store_a));
assert!(ctx_a.mark_tier_delete_journal_recovery_started(store_b));
assert!(ctx_b.mark_tier_delete_journal_recovery_started(store_a));
let mut manual_ctx = InstanceContext::new();
manual_ctx.suppress_tier_delete_journal_recovery_for_test();
assert!(!manual_ctx.mark_tier_delete_journal_recovery_started(store_a));
assert!(!manual_ctx.mark_tier_delete_journal_recovery_started(store_b));
assert!(ctx_b.mark_tier_delete_journal_recovery_started(store_b));
}
#[test]