fix(ecstore): drain multipart uploads before decommission (#6414)

* fix(ecstore): drain multipart uploads before decommission

* fix(ecstore): prioritize active multipart pools

---------

Co-authored-by: houseme <housemecn@gmail.com>
This commit is contained in:
Zhengchao An
2026-08-23 23:04:35 +08:00
committed by GitHub
parent 201c653dcd
commit 8c3aeaebed
4 changed files with 311 additions and 71 deletions
+45
View File
@@ -7152,6 +7152,9 @@ impl ECStore {
let buckets = self.get_buckets_to_decommission().await?;
let pool = self.pools[idx].clone();
self.ensure_decommission_multipart_uploads_drained(idx, &pool, &buckets)
.await?;
for (set_index, set) in pool.disk_set.iter().enumerate() {
for bucket_info in &buckets {
let mut lifecycle_config = None;
@@ -7320,6 +7323,48 @@ impl ECStore {
Ok(())
}
async fn ensure_decommission_multipart_uploads_drained(
&self,
idx: usize,
pool: &Sets,
buckets: &[DecomBucketInfo],
) -> Result<()> {
let mut bucket_names = buckets
.iter()
.filter(|bucket| bucket.name != RUSTFS_META_BUCKET)
.map(|bucket| bucket.name.as_str())
.collect::<Vec<_>>();
bucket_names.sort_unstable();
bucket_names.dedup();
// Take one bucket fence at a time so cross-bucket COPY cannot form an
// ABBA cycle. Suspension prevents new source uploads after each fence.
for bucket in bucket_names {
let lifecycle_guard = self.acquire_bucket_lifecycle_write_lock(bucket).await?;
if lifecycle_guard.is_lock_lost() {
return Err(Error::other(format!(
"decommission multipart drain lost the bucket lifecycle fence for `{bucket}`"
)));
}
for set in &pool.disk_set {
if let Some(upload_path) = set.first_multipart_upload_path_for_decommission(bucket).await? {
return Err(Error::other(format!(
"pool {idx} still contains multipart upload `{upload_path}` for bucket `{bucket}`; resolve it before retrying decommission"
)));
}
}
}
Ok(())
}
#[cfg(test)]
pub(crate) async fn ensure_decommission_multipart_uploads_drained_for_test(self: &Arc<Self>, idx: usize) -> Result<()> {
let buckets = self.get_buckets_to_decommission().await?;
let pool = self.pools[idx].clone();
self.ensure_decommission_multipart_uploads_drained(idx, pool.as_ref(), &buckets)
.await
}
#[cfg(test)]
pub(crate) async fn check_after_decommission_for_test(self: &Arc<Self>, idx: usize) -> Result<()> {
let generation = self.active_decommission_generation(idx).await?;