mirror of
https://github.com/rustfs/rustfs.git
synced 2026-08-25 21:46:50 +00:00
fix(scanner): fence unknown tier accounting (#6396)
This commit is contained in:
@@ -16,8 +16,8 @@
|
||||
use crate::RUSTFS_META_BUCKET;
|
||||
use crate::scanner_budget::{ScannerCycleBudget, ScannerCycleBudgetConfig};
|
||||
use crate::scanner_io::{
|
||||
DataUsageCacheScanState, ScannerDiskScanOutcome, ScannerIODisk, acquire_scanner_cache_locks, cache_root_entry_info,
|
||||
current_cache_root_or_prepare, scanner_set_disk_inventory,
|
||||
DataUsageCacheReuseOptions, DataUsageCacheScanState, ScannerDiskScanOutcome, ScannerIODisk, acquire_scanner_cache_locks,
|
||||
cache_root_entry_info, current_cache_root_or_prepare_with_generation, scanner_set_disk_inventory,
|
||||
};
|
||||
use crate::storage_api::owner::NS_SCANNER_PROTOCOL_VERSION;
|
||||
use crate::{
|
||||
@@ -213,6 +213,7 @@ 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,
|
||||
}
|
||||
@@ -223,6 +224,7 @@ struct RemoteScannerResponseExpectation<'a> {
|
||||
source: DataUsageCacheSource,
|
||||
next_cycle: u64,
|
||||
scan_plan_digest: DataUsageScanPlanDigest,
|
||||
tier_registry_generation: u64,
|
||||
}
|
||||
|
||||
#[derive(Debug)]
|
||||
@@ -669,6 +671,11 @@ 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())
|
||||
@@ -704,7 +711,25 @@ 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}"))
|
||||
})?;
|
||||
let scan_state = current_cache_root_or_prepare(&mut cache, &bucket, source, next_cycle, leader_epoch, scan_plan_digest, true);
|
||||
// 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),
|
||||
},
|
||||
);
|
||||
match scan_state {
|
||||
DataUsageCacheScanState::Current(usage) => {
|
||||
if guard.is_lock_lost() {
|
||||
@@ -869,6 +894,7 @@ pub(crate) async fn scan_remote_bucket(
|
||||
session_id,
|
||||
session_sequence,
|
||||
scan_plan_digest,
|
||||
tier_registry_generation,
|
||||
skip_healing,
|
||||
scan_mode,
|
||||
} = spec;
|
||||
@@ -957,6 +983,7 @@ pub(crate) async fn scan_remote_bucket(
|
||||
source: expected_source,
|
||||
next_cycle,
|
||||
scan_plan_digest,
|
||||
tier_registry_generation,
|
||||
},
|
||||
authenticator,
|
||||
rpc_deadline,
|
||||
@@ -1012,6 +1039,7 @@ 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,
|
||||
@@ -1111,6 +1139,11 @@ 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",
|
||||
|
||||
@@ -195,6 +195,7 @@ 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),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -751,6 +752,43 @@ 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();
|
||||
|
||||
Reference in New Issue
Block a user