gsadmin/rustfs
1
0
Fork 0
mirror of https://github.com/rustfs/rustfs.git synced 2026-08-24 05:06:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(replication): snapshot existing object admission targets (#5634)

Browse Source
This commit is contained in:
cxymds
2026-08-03 00:13:32 +08:00
committed by GitHub
parent ec67884f8d
commit a918f1a48a
5 changed files with 125 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/ecstore/src/bucket/replication/replication_filemeta_boundary.rs
+1 -1
View File
@@ -17,7 +17,7 @@ pub use rustfs_replication::{MrfOpKind, MrfReplicateEntry};
pub(crate) use rustfs_replication::{
REPLICATE_EXISTING, REPLICATE_EXISTING_DELETE, REPLICATE_HEAL_DELETE, ReplicateTargetDecision, ReplicatedInfos,
ReplicatedTargetInfo, ReplicationAction, ReplicationWorkerOperation, ResyncDecision, get_replication_state,
parse_replicate_decision, target_reset_header, version_purge_statuses_map,
parse_replicate_decision, replicate_decision_for_admitted_targets, target_reset_header, version_purge_statuses_map,
};
pub use rustfs_replication::{
REPLICATE_INCOMING_DELETE, ReplicateDecision, ReplicateObjectInfo, ReplicationState, ReplicationStatusType, ReplicationType,
crates/ecstore/src/bucket/replication/replication_pool.rs
+32 -21
View File
@@ -16,8 +16,8 @@ use super::replication_config_store::ReplicationConfigStore;
use super::replication_error_boundary::Error as EcstoreError;
use super::replication_filemeta_boundary::{
MrfOpKind, MrfReplicateEntry, REPLICATE_HEAL_DELETE, ReplicateDecision, ReplicateObjectInfo, ReplicatedTargetInfo,
ReplicationStatusType, ReplicationType, ReplicationWorkerOperation, ResyncDecision, replication_statuses_map,
version_purge_statuses_map,
ReplicationStatusType, ReplicationType, ReplicationWorkerOperation, ResyncDecision, replicate_decision_for_admitted_targets,
replication_statuses_map, version_purge_statuses_map,
};
use super::replication_lock_boundary::ReplicationLockTiming;
use super::replication_logging::{EVENT_REPLICATION_CONFIG_LOOKUP_SKIPPED, LOG_COMPONENT_ECSTORE, LOG_SUBSYSTEM_REPLICATION};
@@ -990,21 +990,25 @@ impl<S: ReplicationStorage> ReplicationPool<S> {
delete_marker: entry.delete_marker,
..Default::default()
};
let dsc = check_replicate_delete(
&entry.bucket,
&ObjectToDelete {
object_name: entry.object.clone(),
version_id: entry.version_id,
..Default::default()
},
&oi,
&ObjectOptions {
versioned,
..Default::default()
},
None,
)
.await;
let dsc = if entry.target_arns.is_empty() {
check_replicate_delete(
&entry.bucket,
&ObjectToDelete {
object_name: entry.object.clone(),
version_id: entry.version_id,
..Default::default()
},
&oi,
&ObjectOptions {
versioned,
..Default::default()
},
None,
)
.await
} else {
replicate_decision_for_admitted_targets(&entry.target_arns)
};
let mut rstate = oi.replication_state();
rstate.replicate_decision_str = dsc.to_string();
@@ -1036,7 +1040,7 @@ impl<S: ReplicationStorage> ReplicationPool<S> {
schedule_replication_delete(dv).await;
queued_count += 1;
}
MrfOpKind::Object => {
MrfOpKind::Object | MrfOpKind::Heal | MrfOpKind::ExistingObject => {
let opts = ObjectOptions {
version_id: entry.version_id.map(|u| u.to_string()),
..Default::default()
@@ -1055,9 +1059,16 @@ impl<S: ReplicationStorage> ReplicationPool<S> {
continue;
}
};
// Route through queue_replication_heal so the replication decision (dsc)
// is computed from the live config — required for replicate_object.
queue_replication_heal(&entry.bucket, oi, entry.retry_count as u32).await;
if entry.target_arns.is_empty() {
// Legacy entries predate target admission persistence. They cannot
// be safely attributed, so retain the old live-config fallback.
queue_replication_heal(&entry.bucket, oi, entry.retry_count as u32).await;
} else if let Some(pool) = runtime_sources::replication_pool() {
let dsc = replicate_decision_for_admitted_targets(&entry.target_arns);
let mut roi = replicate_object_info_from_object_info(oi, dsc, entry.op.replication_type());
roi.retry_count = entry.retry_count.max(0) as u32;
let _ = pool.queue_replica_task(roi).await;
}
queued_count += 1;
}
MrfOpKind::Metadata => {
crates/ecstore/src/bucket/replication/replication_resyncer.rs
+3 -4
View File
@@ -792,6 +792,7 @@ impl ReplicationResyncer {
let storage = storage.clone();
let results_tx = results_tx.clone();
let bucket_name = opts.bucket.clone();
let target_arn = opts.arn.clone();
let f = tokio::spawn(async move {
while let Some(mut roi) = rx.recv().await {
@@ -819,6 +820,7 @@ impl ReplicationResyncer {
bucket: roi.bucket.clone(),
event_type: REPLICATE_EXISTING_DELETE.to_string(),
op_type: ReplicationType::ExistingObject,
target_arn: target_arn.clone(),
..Default::default()
};
replicate_delete(doi, storage.clone()).await;
@@ -2009,10 +2011,7 @@ pub async fn replicate_object<S: ReplicationStorage>(roi: ReplicateObjectInfo, s
let bucket = roi.bucket.clone();
let object = roi.name.clone();
// The admission decision is the target-granular contract. Re-evaluating the
// live config here could fan a synchronous request out to targets that were
// not admitted, or promote an async target after a mixed-mode split.
let tgt_arns = roi.dsc.replicate_target_arns();
let tgt_arns = roi.admitted_target_arns();
// Acquire a per-object namespace lock so that at most one worker (across all cluster
// nodes and MRF retry goroutines) replicates this object version at a time.