mirror of
https://github.com/rustfs/rustfs.git
synced 2026-08-11 15:46:53 +00:00
fix(ilm): harden durable transition admission (#5235)
Co-authored-by: heihutu <heihutu@gmail.com>
This commit is contained in:
@@ -4158,10 +4158,11 @@ pub async fn apply_lifecycle_action(event: &lifecycle::Event, src: &LcEventSrc,
|
||||
mod tests {
|
||||
use super::{
|
||||
DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS, DEFAULT_TRANSITION_QUEUE_CAPACITY, DEFAULT_TRANSITION_WORKERS_ABSOLUTE_MAX,
|
||||
DEFAULT_TRANSITION_WORKERS_CAP, ExpiryState, FreeVersionTask, ManualTransitionRunOptions, ManualTransitionRunReport,
|
||||
StaleMultipartUploadCandidate, TIER_FREE_VERSION_RECOVERY_BASE_INTERVAL, TIER_FREE_VERSION_RECOVERY_MAX_IDLE_INTERVAL,
|
||||
TRANSITION_COMPLETE, TierFreeVersionRecoverySchedule, TransitionEnqueueOutcome, TransitionState, TransitionedObject,
|
||||
VersionReplicationScan, cleanup_empty_multipart_sha_dirs_on_local_disks, cleanup_stale_multipart_uploads_once_at,
|
||||
DEFAULT_TRANSITION_WORKERS_CAP, ExpiryState, FreeVersionTask, ManualTransitionQueueSnapshot, ManualTransitionRunOptions,
|
||||
ManualTransitionRunReport, StaleMultipartUploadCandidate, TIER_FREE_VERSION_RECOVERY_BASE_INTERVAL,
|
||||
TIER_FREE_VERSION_RECOVERY_MAX_IDLE_INTERVAL, TRANSITION_COMPLETE, TierFreeVersionRecoverySchedule,
|
||||
TransitionEnqueueOutcome, TransitionState, TransitionedObject, VersionReplicationScan,
|
||||
cleanup_empty_multipart_sha_dirs_on_local_disks, cleanup_stale_multipart_uploads_once_at,
|
||||
enqueue_recovered_free_version_with_state, enqueue_transition_with_lifecycle, enqueue_transition_with_lifecycle_report,
|
||||
eval_action_from_lifecycle, jitter_tier_free_version_recovery_delay, lifecycle_action_blocked_by_replication,
|
||||
lifecycle_delete_all_versions_replication_scan, lifecycle_deleted_object, lifecycle_replication_blocks_action,
|
||||
@@ -4182,8 +4183,9 @@ mod tests {
|
||||
decode_manual_transition_continuation_token, encode_manual_transition_continuation_token,
|
||||
};
|
||||
use crate::bucket::lifecycle::manual_transition_job::{
|
||||
ManualTransitionJobRecord, ManualTransitionJobState, ManualTransitionScopeAdmission, load_manual_transition_job_record,
|
||||
load_manual_transition_scope_admission, save_manual_transition_job_record,
|
||||
ManualTransitionJobRecord, ManualTransitionJobState, ManualTransitionScopeAdmission, ManualTransitionScopeAdmissionClaim,
|
||||
claim_manual_transition_scope_admission, legacy_manual_transition_scope_key, load_manual_transition_job_record,
|
||||
load_manual_transition_scope_admission, request_manual_transition_job_cancel, save_manual_transition_job_record,
|
||||
save_manual_transition_scope_admission_if_absent,
|
||||
};
|
||||
use crate::bucket::lifecycle::replication_sink::{
|
||||
@@ -7031,7 +7033,7 @@ mod tests {
|
||||
.await
|
||||
.expect("manual transition recovery should run");
|
||||
|
||||
assert_eq!(stats.scanned, 1);
|
||||
assert!(stats.scanned >= 1);
|
||||
assert_eq!(stats.resumed, 1);
|
||||
assert_eq!(stats.failed, 0);
|
||||
let recovered = load_manual_transition_job_record(ecstore.clone(), job_id)
|
||||
@@ -7064,6 +7066,106 @@ mod tests {
|
||||
assert!(!report.has_partial_enqueue());
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
#[serial]
|
||||
async fn manual_transition_job_cancel_marks_running_record_only() {
|
||||
let (_paths, ecstore) = setup_test_env().await;
|
||||
let running_id = Uuid::new_v4();
|
||||
let running = ManualTransitionJobRecord::new(
|
||||
running_id,
|
||||
"manual-cancel-running-bucket",
|
||||
&ManualTransitionRunOptions::default(),
|
||||
"owner-a",
|
||||
);
|
||||
save_manual_transition_job_record(ecstore.clone(), &running)
|
||||
.await
|
||||
.expect("running job record should save");
|
||||
|
||||
let cancelled = request_manual_transition_job_cancel(ecstore.clone(), running_id)
|
||||
.await
|
||||
.expect("running job cancel request should persist");
|
||||
|
||||
assert_eq!(cancelled.state, ManualTransitionJobState::Running);
|
||||
assert!(cancelled.cancel_requested);
|
||||
let loaded = load_manual_transition_job_record(ecstore.clone(), running_id)
|
||||
.await
|
||||
.expect("cancelled running job should reload");
|
||||
assert!(loaded.cancel_requested);
|
||||
|
||||
let terminal_id = Uuid::new_v4();
|
||||
let mut terminal = ManualTransitionJobRecord::new(
|
||||
terminal_id,
|
||||
"manual-cancel-terminal-bucket",
|
||||
&ManualTransitionRunOptions::default(),
|
||||
"owner-a",
|
||||
);
|
||||
terminal.complete(
|
||||
ManualTransitionRunReport {
|
||||
bucket: "manual-cancel-terminal-bucket".to_string(),
|
||||
..Default::default()
|
||||
},
|
||||
ManualTransitionQueueSnapshot::default(),
|
||||
);
|
||||
save_manual_transition_job_record(ecstore.clone(), &terminal)
|
||||
.await
|
||||
.expect("terminal job record should save");
|
||||
|
||||
let after_terminal_cancel = request_manual_transition_job_cancel(ecstore, terminal_id)
|
||||
.await
|
||||
.expect("terminal job cancel should be idempotent");
|
||||
|
||||
assert_eq!(after_terminal_cancel.state, ManualTransitionJobState::Completed);
|
||||
assert!(!after_terminal_cancel.cancel_requested);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
#[serial]
|
||||
async fn manual_transition_admission_blocks_active_legacy_scope_record() {
|
||||
let (_paths, ecstore) = setup_test_env().await;
|
||||
let legacy_options = ManualTransitionRunOptions {
|
||||
prefix: "logs/".to_string(),
|
||||
tier: Some("warm".to_string()),
|
||||
..Default::default()
|
||||
};
|
||||
let legacy_id = Uuid::new_v4();
|
||||
let mut legacy = ManualTransitionJobRecord::new(legacy_id, "manual-legacy-scope-bucket", &legacy_options, "old-owner");
|
||||
legacy.scope_key = legacy_manual_transition_scope_key(&legacy.bucket, &legacy_options);
|
||||
save_manual_transition_job_record(ecstore.clone(), &legacy)
|
||||
.await
|
||||
.expect("legacy job record should save");
|
||||
save_manual_transition_scope_admission_if_absent(ecstore.clone(), &ManualTransitionScopeAdmission::from_job(&legacy))
|
||||
.await
|
||||
.expect("legacy scope admission should save");
|
||||
|
||||
let new_options = ManualTransitionRunOptions {
|
||||
prefix: "archive/".to_string(),
|
||||
tier: Some("cold".to_string()),
|
||||
..Default::default()
|
||||
};
|
||||
let new_record = ManualTransitionJobRecord::new(Uuid::new_v4(), "manual-legacy-scope-bucket", &new_options, "new-owner");
|
||||
save_manual_transition_job_record(ecstore.clone(), &new_record)
|
||||
.await
|
||||
.expect("new job record should save");
|
||||
|
||||
let claim =
|
||||
claim_manual_transition_scope_admission(ecstore.clone(), &ManualTransitionScopeAdmission::from_job(&new_record))
|
||||
.await
|
||||
.expect("new bucket-level admission claim should resolve");
|
||||
|
||||
let ManualTransitionScopeAdmissionClaim::Conflict(active) = claim else {
|
||||
panic!("active legacy job must block new bucket-level admission");
|
||||
};
|
||||
assert_eq!(active.job_id, legacy_id);
|
||||
assert_eq!(active.scope_key, legacy.scope_key);
|
||||
assert!(
|
||||
matches!(
|
||||
load_manual_transition_scope_admission(ecstore, &new_record.scope_key).await,
|
||||
Err(Error::ConfigNotFound)
|
||||
),
|
||||
"conflicted bucket-level admission must be released"
|
||||
);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn existing_object_lifecycle_allows_expired_marker_after_replication_completed() {
|
||||
let lc = expired_delete_marker_lifecycle();
|
||||
|
||||
@@ -23,8 +23,10 @@ use crate::bucket::lifecycle::bucket_lifecycle_ops::{
|
||||
ManualTransitionQueueSnapshot, ManualTransitionRunOptions, ManualTransitionRunReport,
|
||||
};
|
||||
use crate::bucket::lifecycle::config_boundary;
|
||||
use crate::disk::RUSTFS_META_BUCKET;
|
||||
use crate::error::{Error, Result as EcstoreResult};
|
||||
use crate::object_api::ObjectOptions;
|
||||
use crate::storage_api_contracts::list::ListOperations as _;
|
||||
use crate::storage_api_contracts::object::HTTPPreconditions;
|
||||
use crate::store::ECStore;
|
||||
|
||||
@@ -33,6 +35,7 @@ pub const MANUAL_TRANSITION_JOB_RECORD_PREFIX: &str = "ilm/manual-transition/job
|
||||
pub const MANUAL_TRANSITION_SCOPE_RECORD_PREFIX: &str = "ilm/manual-transition/scopes";
|
||||
pub const MAX_MANUAL_TRANSITION_JOB_RECORD_SIZE: usize = 64 * 1024;
|
||||
const MANUAL_TRANSITION_JOB_LEASE_SECONDS: i128 = 60;
|
||||
const MANUAL_TRANSITION_LEGACY_SCOPE_SCAN_LIMIT: i32 = 1000;
|
||||
|
||||
#[derive(Debug, thiserror::Error)]
|
||||
pub enum ManualTransitionJobError {
|
||||
@@ -351,6 +354,16 @@ pub enum ManualTransitionScopeAdmissionClaim {
|
||||
}
|
||||
|
||||
pub fn manual_transition_scope_key(bucket: &str, options: &ManualTransitionRunOptions) -> String {
|
||||
let mut scope = String::new();
|
||||
scope.push_str(bucket);
|
||||
scope.push('\0');
|
||||
// Durable admission v1 is bucket-level: a single CAS alias must cover prefix
|
||||
// overlap and wildcard-tier conflicts without a non-atomic scope scan.
|
||||
scope.push_str(if options.dry_run { "dry_run" } else { "run" });
|
||||
hex_sha256(scope.as_bytes(), ToOwned::to_owned)
|
||||
}
|
||||
|
||||
pub(crate) fn legacy_manual_transition_scope_key(bucket: &str, options: &ManualTransitionRunOptions) -> String {
|
||||
let mut scope = String::new();
|
||||
scope.push_str(bucket);
|
||||
scope.push('\0');
|
||||
@@ -553,7 +566,7 @@ pub async fn claim_manual_transition_scope_admission(
|
||||
admission: &ManualTransitionScopeAdmission,
|
||||
) -> EcstoreResult<ManualTransitionScopeAdmissionClaim> {
|
||||
match save_manual_transition_scope_admission_if_absent(api.clone(), admission).await {
|
||||
Ok(()) => return Ok(ManualTransitionScopeAdmissionClaim::Claimed),
|
||||
Ok(()) => return finish_manual_transition_scope_admission_claim(api, admission).await,
|
||||
Err(Error::PreconditionFailed) => {}
|
||||
Err(err) => return Err(err),
|
||||
}
|
||||
@@ -572,8 +585,8 @@ pub async fn claim_manual_transition_scope_admission(
|
||||
}
|
||||
};
|
||||
if active_job_reclaimable {
|
||||
return match save_manual_transition_scope_admission_if_current(api, admission, &etag).await {
|
||||
Ok(()) => Ok(ManualTransitionScopeAdmissionClaim::Claimed),
|
||||
return match save_manual_transition_scope_admission_if_current(api.clone(), admission, &etag).await {
|
||||
Ok(()) => finish_manual_transition_scope_admission_claim(api, admission).await,
|
||||
Err(Error::PreconditionFailed) => Ok(ManualTransitionScopeAdmissionClaim::Conflict(Box::new(active))),
|
||||
Err(err) => Err(err),
|
||||
};
|
||||
@@ -582,6 +595,78 @@ pub async fn claim_manual_transition_scope_admission(
|
||||
Ok(ManualTransitionScopeAdmissionClaim::Conflict(Box::new(active)))
|
||||
}
|
||||
|
||||
async fn finish_manual_transition_scope_admission_claim(
|
||||
api: Arc<ECStore>,
|
||||
admission: &ManualTransitionScopeAdmission,
|
||||
) -> EcstoreResult<ManualTransitionScopeAdmissionClaim> {
|
||||
if let Some(active) = find_active_legacy_manual_transition_scope_conflict(api.clone(), admission).await? {
|
||||
delete_manual_transition_scope_admission_if_current(api, &admission.scope_key, admission.job_id, admission.lease_id)
|
||||
.await?;
|
||||
return Ok(ManualTransitionScopeAdmissionClaim::Conflict(Box::new(active)));
|
||||
}
|
||||
Ok(ManualTransitionScopeAdmissionClaim::Claimed)
|
||||
}
|
||||
|
||||
async fn find_active_legacy_manual_transition_scope_conflict(
|
||||
api: Arc<ECStore>,
|
||||
admission: &ManualTransitionScopeAdmission,
|
||||
) -> EcstoreResult<Option<ManualTransitionScopeAdmission>> {
|
||||
let mut marker = None;
|
||||
loop {
|
||||
let page = api
|
||||
.clone()
|
||||
.list_objects_v2(
|
||||
RUSTFS_META_BUCKET,
|
||||
MANUAL_TRANSITION_JOB_RECORD_PREFIX,
|
||||
marker,
|
||||
None,
|
||||
MANUAL_TRANSITION_LEGACY_SCOPE_SCAN_LIMIT,
|
||||
false,
|
||||
None,
|
||||
false,
|
||||
)
|
||||
.await?;
|
||||
for object in page.objects {
|
||||
let job_id =
|
||||
manual_transition_job_id_from_record_object_name(&object.name).map_err(manual_transition_job_store_error)?;
|
||||
if job_id == admission.job_id {
|
||||
continue;
|
||||
}
|
||||
let record = match load_manual_transition_job_record(api.clone(), job_id).await {
|
||||
Ok(record) => record,
|
||||
Err(Error::ConfigNotFound) => continue,
|
||||
Err(err) => return Err(err),
|
||||
};
|
||||
if record.scope_key == admission.scope_key {
|
||||
continue;
|
||||
}
|
||||
let legacy_scope_key = legacy_manual_transition_scope_key(
|
||||
&record.bucket,
|
||||
&ManualTransitionRunOptions {
|
||||
prefix: record.prefix.clone(),
|
||||
tier: record.tier.clone(),
|
||||
dry_run: record.dry_run,
|
||||
..Default::default()
|
||||
},
|
||||
);
|
||||
if record.scope_key != legacy_scope_key {
|
||||
continue;
|
||||
}
|
||||
if record.bucket == admission.bucket
|
||||
&& record.dry_run == admission.dry_run
|
||||
&& !record.is_terminal()
|
||||
&& !manual_transition_job_lease_expired(&record)
|
||||
{
|
||||
return Ok(Some(ManualTransitionScopeAdmission::from_job(&record)));
|
||||
}
|
||||
}
|
||||
if !page.is_truncated {
|
||||
return Ok(None);
|
||||
}
|
||||
marker = page.next_continuation_token;
|
||||
}
|
||||
}
|
||||
|
||||
pub async fn request_manual_transition_job_cancel(api: Arc<ECStore>, job_id: Uuid) -> EcstoreResult<ManualTransitionJobRecord> {
|
||||
for _ in 0..4 {
|
||||
let (mut record, etag) = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
|
||||
@@ -876,6 +961,46 @@ mod tests {
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn manual_transition_scope_key_uses_bucket_level_admission_for_durable_v1() {
|
||||
let broad = manual_transition_scope_key(
|
||||
"bucket",
|
||||
&ManualTransitionRunOptions {
|
||||
prefix: "logs/".to_string(),
|
||||
tier: None,
|
||||
..Default::default()
|
||||
},
|
||||
);
|
||||
let nested = manual_transition_scope_key(
|
||||
"bucket",
|
||||
&ManualTransitionRunOptions {
|
||||
prefix: "logs/2026/".to_string(),
|
||||
tier: Some("warm".to_string()),
|
||||
..Default::default()
|
||||
},
|
||||
);
|
||||
let disjoint = manual_transition_scope_key(
|
||||
"bucket",
|
||||
&ManualTransitionRunOptions {
|
||||
prefix: "archive/".to_string(),
|
||||
tier: Some("cold".to_string()),
|
||||
..Default::default()
|
||||
},
|
||||
);
|
||||
let dry_run = manual_transition_scope_key(
|
||||
"bucket",
|
||||
&ManualTransitionRunOptions {
|
||||
prefix: "logs/".to_string(),
|
||||
dry_run: true,
|
||||
..Default::default()
|
||||
},
|
||||
);
|
||||
|
||||
assert_eq!(broad, nested);
|
||||
assert_eq!(broad, disjoint);
|
||||
assert_ne!(broad, dry_run);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn manual_transition_scope_admission_carries_job_lease_fence() {
|
||||
let options = ManualTransitionRunOptions::default();
|
||||
|
||||
Reference in New Issue
Block a user