Bug #2527
closed
RGW may return 409 Conflict when deleting a bucket
Added by Jeremy Hanmer almost 12 years ago.
Updated over 11 years ago.
Description
If a bucket delete call occurs immediately after running a delete operation on the final remaining object in that bucket, there's a chance of hitting a 409 Conflict error. I assume this is because the delete operation hasn't been executed on the backend yet. The problem is exacerbated if the cluster is below 100% health (even with 2 out of ~90 OSDs out, I hit it almost every time).
- Project changed from Ceph to rgw
Logs:ubuntu@teuthology:/a/teuthology-2012-08-29_19:00:06-regression-master-testing-gcov/11434
Recent logs: ubuntu@teuthology:/a/teuthology-2012-08-30_00:00:06-regression-next-testing-basic/11588
- Priority changed from Normal to Urgent
- Target version set to v0.54a
The bucket index update when doing object removal is async. That means that when we get to remove the bucket, the index might not have been cleared yet.
- Priority changed from Urgent to Normal
Might be related to the messenger issues that were fixed in 0.52. We'll wait a bit, see if it still happens.
Recent log: ubuntu@teuthology:/a/teuthology-2012-10-10_02:00:02-regression-testing-master-basic/5604
- Status changed from New to Resolved
The problem here was that the bucket was not really empty, there was an issue with cache not being updated (so user suspend didn't get through to the gateway).
Fixed by commit:5d27f3da654b2d1b3f90a5758fc47b89f012a8d2 and commit:e0e33d2c99fe7c056bb91d5cbbe1ccfb6bbcfb1b.
Also available in: Atom
PDF