<div dir="ltr">Mohit, <div><br></div><div>Can you set up a call/meeting where you can explain the current patch? That would make it easier to review.</div><div><br></div><div>Thanks,</div><div>Nithya<br><div class="gmail_extra"><br><div class="gmail_quote">On 13 July 2017 at 10:32, Raghavendra Gowdappa <span dir="ltr"><<a href="mailto:rgowdapp@redhat.com" target="_blank">rgowdapp@redhat.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">All,<br>
<br>
Patch [1] is getting more complex day by day. We had to extend permission healing to xattr healing as Mohith saw some regressions in uid/gid healing codepath. So, I would request you to take a look and comment.<br>
<br>
Current behavior of the patch:<br>
* MDS (Meta data server) is the source of truth (among other subvols containing directory) for user, posix-acls, some other xattrs, uid/gid on a directory<br>
* If mds is not there, choose existing hashed-subvol as mds. Once chosen mds remains constant till that brick is removed.<br>
* non-availability of mds leads to failure of setxattr/setattr operation. Some volume level operations like add-brick/remove-brick maintain metadata on root, it is necessary that these operations don't fail. So, no mds is chosen for root and existing behavior (doing ops on all subvols simultaneously) is retained. Also, note that root won't have any hashed-subvol (we could've chose first child of dht as mds, but didn't for reasons mentioned earlier)<br>
* uid/gid healing logic is changed from having subvol with latest ctime as source of truth to mds being source of truth.<br>
<br>
[1] <a href="https://review.gluster.org/15468" rel="noreferrer" target="_blank">https://review.gluster.org/<wbr>15468</a><br>
<br>
regards,<br>
Raghavendra<br>
</blockquote></div><br></div></div></div>