Ceph » rgw

When we use command "radosgw-admin bucket rm --bucket=<bucket name> --bypass-gc --purge-objects" to delete a very large bucket.
It might take very long time to finish since it will bypass the gc, so when the deleting process is terminated because of some reason (eg. machine reboot, administrator just killed it.)

That is highly possible to leave a partial deleted rgw object (only deleted some shadow object, and the head object is still there).

When the next time, the administrator want to continue to run "radosgw-admin bucket rm --bucket=<bucket name> --bypass-gc --purge-objects" to delete the rest of the object in this bucket.
The process will error out like below:
2019-06-28 11:55:23.587429 7f9ab8265e40 -1 ERROR: could not drain handles as aio completion returned with -2
2019-06-28 11:55:23.588152 7f9ab8265e40 -1 ERROR: unable to remove bucket(2) No such file or director

This happened because radosgw-admin will try to delete the partial deleted rgw object that left from last time.
And then it do that, it will try to delete all its shadow objects from beginning again,
Since those front shadow objects were already deleted from last time, it will certainly got -ENOENT when it calls drain_handles which will make the whole process exit.

Above is the scenario we met which makes bypass-gc deletion for large bucket not working at all.

Even in general, I think we should skip the -ENOENT return value here, because "-ENOENT" just mean the object is already deleted
And the function rgw_remove_bucket_bypass_gc is just doing the deletion, there should be no difference for rgw_remove_bucket_bypass_gc treating return values between "0" and "-ENOENT"