RGW: Multipart upload may double the quota

[gaosibei]

When user upload object using multipart api, they may upload failed. The reason is that the object size is added twice to quota stats.

Fixes: http://tracker.ceph.com/issues/21586

Signed-off-by: Sibei Gao gaosb@inspur.com

[mattbenjamin]

@gaosibei is it possible accounted_size could be corrected by the call path leading to _do_write_meta?

[gaosibei]

I think that the accounted_size could not be corected outside of the _do_write_meta. Because, the accounted_size is both used in quota status and index_op.

at the rgw_rados.cc:6940 (_do_write_meta)
r = index_op->complete(poolid, epoch, size, accounted_size,...)

[joscollin]

Fixes: missing.
https://github.com/ceph/ceph/blob/master/SubmittingPatches.rst#tag-the-commit

[joscollin]

@gaosibei No, that's not the tracker for this PR.

[gaosibei]

@joscollin

sorry

The tracker issue is http://tracker.ceph.com/issues/21586

[gaosibei]

@mattbenjamin @mdw-at-linuxbox

I think that the accounted_size could not be corected outside of the _do_write_meta.

Because, the accounted_size is both used in quota status and index_op.

at the rgw_rados.cc (_do_write_meta)

1. r = index_op->complete(poolid, epoch, size, accounted_size,...)
2. store->quota_handler->update_stats(meta.owner, obj.bucket, (orig_exists ? 0 : 1),
   accounted_size, orig_size);

Now, I give a new commit which seems better than before

[mdw-at-linuxbox]

Looks like the tracker issue is fixed. I tried this out and verified it works.

Tracker issue matches now - so I think your change was made.

[tchaikov]

@mattbenjamin @mdw-at-linuxbox i am marking the tracker ticket with "backport=jewel,luminous" after checking jewel and luminous branches. does it make sense to you guys?