Description of problem: When using RGW to delete large buckets with 50-100 or more million objects the process is slow and appears to be sequential in nature. Version-Release number of selected component (if applicable):1.3.2 How reproducible:100% Steps to Reproduce: 1.Create a Ceph cluster on 1.3.2 2.Load up a bucket with many millions of objects 3.Delete objects in the bucket Actual results: Slow process takes time to delete. Expected results: RGW should be able to delete multiple objects in parallel and speed up the process when doing large deletes Additional info:
Jewel backport ongoing at https://github.com/ceph/ceph/pull/10661 for v10.2.4 (currently fails to build there, looks like the RGW team will need to resolve the failure)
(In reply to Ken Dreyer (Red Hat) from comment #21) > Jewel backport ongoing at https://github.com/ceph/ceph/pull/10661 for > v10.2.4 (currently fails to build there, looks like the RGW team will need > to resolve the failure) PR #10661 updated with a proposed compile fix and motivation.
To reproduce, create an rgw bucket and put any substantial number of objects, then measure the time required to sequentially delete each object and then the bucket. After this change, that time can be compared with the result of "time radosgw-admin bucket rm --bucket=<bucket name> --bypass-gc --purge-objects" having first recreated the bucket and objects.
Hi, I simultaneously tried purging a bucket and deleting all objects in another bucket . Each of those bucket had 5 million objects and buckets were of same size. When one bucket got purged, another bucket had more than 1.8 million objects. Purging took 64 hours and deletion took 110+ hours. As purging is faster than sequential deletion as per above observation I'm closing this bug. Please let me know if there any concerns or issues. Regards, Vasishta
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://rhn.redhat.com/errata/RHSA-2016-2815.html