gsadmin/rustfs
1
0
Fork 0
mirror of https://github.com/rustfs/rustfs.git synced 2026-08-10 23:26:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(ecstore): roll back delete on disks that staged then errored (#4676)

Browse Source 
fix(ecstore): roll back delete on disks that staged then errored (backlog#1158)

#4300 rolls back a failed delete only on disks that returned Ok, skipping any
disk that staged its rollback backup, applied the delete, and then errored --
leaving that disk deleted while its peers are restored. On rollback, fan the
undo out to every online disk instead; the disk-side restore_delete_rollback
is already idempotent (Ok no-op when nothing was staged), so unstaged disks
are unaffected. The err.is_some() skip now applies only to the success/cleanup
path. Covers both single-object and batch delete.

Refs backlog#1158.
This commit is contained in:
Zhengchao An
2026-07-10 20:34:19 +08:00
committed by GitHub
parent b604074217
commit 80ddd8fa7e
2 changed files with 73 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/ecstore/src/set_disk/ops/object.rs
+13 -2
View File
@@ -1409,7 +1409,14 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
let should_rollback = quorum_result.is_err();
let mut rollback_futures = Vec::new();
for (index, err) in errs.iter().enumerate() {
if err.is_some() {
// backlog#1158: when rolling back, fan the idempotent undo out to every
// online disk (each self-decides from its staged backup: restore if the
// rollback dir is present, no-op otherwise). This covers a disk that
// staged + applied the delete and *then* errored, which the plain
// `err.is_some()` skip would leave deleted while its peers were restored.
// On success only the successful disks' backup dirs need cleaning; errored
// disks' residue is reclaimed by heal/scanner.
if !should_rollback && err.is_some() {
continue;
}
@@ -1764,7 +1771,11 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
// delete_versions commits one xl.meta per object group, so rollback must use the same boundary.
let should_rollback = fi_vers.versions.iter().any(|fi| del_errs[fi.idx].is_some());
for (disk_idx, disk) in disks.iter().enumerate() {
if fi_vers.versions.iter().any(|fi| del_obj_errs[disk_idx][fi.idx].is_some()) {
// backlog#1158: on rollback, include every online disk so a disk that
// staged + applied the delete and then errored is still restored (the
// disk-side undo is idempotent, no-op when nothing was staged). On
// success, skip the errored disks and only clean up successful ones.
if !should_rollback && fi_vers.versions.iter().any(|fi| del_obj_errs[disk_idx][fi.idx].is_some()) {
continue;
}