<br><br><div class="gmail_quote">On Thu, Nov 29, 2012 at 10:20 AM, Jeff Darcy <span dir="ltr"><<a href="mailto:jdarcy@redhat.com" target="_blank">jdarcy@redhat.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div class="im">On 11/29/2012 10:14 AM, John Mark Walker wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
from @saiweb :<br>
<br>
"@rackerhacker be warned unless it has been fixed NFS in gluster does not distribute locks, multi node writes == split brain"<br>
<br>
<a href="https://twitter.com/Saiweb/status/274114952963432449" target="_blank">https://twitter.com/Saiweb/<u></u>status/274114952963432449</a><br>
</blockquote>
<br></div>
Half true, or perhaps a matter of terminology. AFAIK we don't have distributed locking, but the result of concurrent access would be inconsistency rather than (what we call) split brain.<div class="HOEnZb"><div class="h5">
<br><br></div></div></blockquote><div><br></div><div>Gluster NFS's locks (NLM) are just a "pass-through" translation to posix-locks in the (appropriate) brick of the file. There is no need for typical "distributed locking" as NFS is completely stateless. We should not even be seeing even inconsistency if app is holding locks (even if app instances are running on different nfs clients mounting off different NFS heads).</div>
<div><br></div><div>BTW, NLM is available only in 3.3.x. If using 3.2.x there is no NFS locking (clients forced to mount with -onolock) and there WILL be inconsistency (but never split brain).</div><div><br></div><div>Avati</div>
<div><br></div></div>