Re: In the ceph multisite master-zone, read ,write,delete objects, and the master-zone has data remaining.

[WeiGuo Ren <rwg1335252904@gmail.com>]

Do we need to solve this problem?

WeiGuo Ren <rwg1335252904@gmail.com> 于2021年3月10日周三 下午4:34写道：
>
> In my test environment, the ceph version is v14.2.5, and there are two
> rgws, which are instances of two zones, respectively rgwA
> (master-zone) and rgwB (slave-zone). Cosbench reads, writes, and
> deletes to rgwA. , The final result rgwA has data residue, but rgwB
> has no residue.
>
> Looking at the log later, I found that this happened:
> 1. When rgwA deletes the object, the rgwA instance has not yet started
> datasync (or the process is slow) to synchronize the object in the
> slave-zone.
> 2. When rgwA starts data synchronization, rgwB has not deleted the object.
> In process 2, rgwA will retrieve the object from the slave-zone, and
> then rgwA will enter the incremental synchronization state to
> synchronize the bilog, but the bilog about the del object will be
> filtered out, because syncs_trace has  master zone.
>
> Below I did a similar reproducing operation (both in the master
> version and ceph 14.2.5)
> rgwA and rgwB are two zones of the same zonegroup .rgwA and rgwB is
> running ( set rgw_run_sync_thread=true)
> rgwA and rgwB are two zones of the same zonegroup .rgwA and rgwB is
> running ( set rgw_run_sync_thread=true)
> t1: rgwA set rgw_run_sync_thread=false and restart it for it to take
> effect. We use s3cmd to create a bucket in rgwA. And upload an object1
> in rgwA. We use s3cmd to observe whether object1 has been synchronized
> in rgwB. or  look radosgw-admin bucket sync status is cauht up it. If
> the synchronization has passed, proceed to the next step.
> t2:rgwB set rgw_run_sync_thread=false and restart it for it to take
> effect. rgwA delete object1 .
> t3:rgwA set rgw_run_sync_thread=true and restart it for it to take
> effect. LOOK radosgw-admin bucket sync status is cauht up it.
> t4: rgwB set rgw_run_sync_thread=true and restart it for it to take
> effect. LOOK radosgw-admin bucket sync status is cauht up it .
> The reslut: rgwA has object1,rgwB dosen't have object1.
> This URL mentioned this problem  https://tracker.ceph.com/issues/47555
>
> Could someone can help me? or If the bucket about the rgwA instance is
> not in the incremental synchronization state, can we prohibit rgwA
> from deleting object1?