rgw: fix for bucket delete racing with mdlog sync #11648
Conversation
| @@ -2481,6 +2481,12 @@ void RGWDeleteBucket::execute() | |||
| } | |||
|
|
|||
| op_ret = store->delete_bucket(s->bucket, ot); | |||
|
|
|||
| if (op_ret == -ECANCELED && !store->is_meta_master()) { | |||
There was a problem hiding this comment.
@cbodley can't this happen with multiple deletes going to the same bucket on the master zone also?
There was a problem hiding this comment.
probably, yeah. should we map ECANCELED to ENOENT in that case, or return success there as well?
There was a problem hiding this comment.
currently the ECANCELED is returned by cls_version, which actually reraises a ENOENT from version check as a ECANCELED as the version check fails through, should we be just raising the ENOENT as it is, if there is no one else consuming the cls_version check that is...
Fixes: http://tracker.ceph.com/issues/17698 Signed-off-by: Casey Bodley <cbodley@redhat.com>
|
updated to handle races on the metadata master as well. also added a return, because we've already called |
|
the latest change worked for both the cases, ie. races in master zone & slave zone for bucket removal |
proposed fix for bucket deletion on a remote racing with mdlog sync
Fixes: http://tracker.ceph.com/issues/17698