<div>See below for what I'm currently running. I'm testing without readahead at the moment - I haven't seen it fail with it commented out so far, but it's too early to tell.</div><div><br></div><div>Before I added the trace translator the error rate was much higher. After I added it, it took several hours before it started failing, even though I've excluded almost all the calls.</div>
<div><br></div><div>I can only reproduce it when I add the server in question to our production cluster, unfortunately - I've tried repeatedly to reproduce it with a test environment, but I'm clearly missing out on some element of the access pattern we see in production. I should probably try to replay the requests from our access logs or something.</div>
<div><br></div><div>Vidar</div><div><br></div><div><div>volume webcluster1</div><div> type protocol/client</div><div> option transport-type tcp/client</div><div> option remote-host 192.168.165.13</div><div> option remote-subvolume webcluster</div>
<div>end-volume</div><div><br></div><div>volume webcluster2</div><div> type protocol/client</div><div> option transport-type tcp/client</div><div> option remote-host 192.168.165.11</div><div> option remote-subvolume webcluster</div>
<div>end-volume</div><div><br></div><div>### Replication</div><div><br></div><div>volume afr</div><div> type cluster/afr</div><div> subvolumes webcluster1 webcluster2</div><div>end-volume</div><div><br></div><div>### Performance</div>
<div><br></div><div>#volume readahead</div><div># type performance/read-ahead</div><div># option page-count 2 # cache per file = (page-count x page-size)</div><div># subvolumes afr</div><div>#end-volume</div><div>
<br></div><div>### Add IO-Cache feature</div><div>volume iocache</div><div> type performance/io-cache</div><div> option page-size 256KB</div><div> subvolumes afr #readahead</div><div>end-volume</div><div><br></div><div>
<div>### Add writeback feature</div><div>volume writeback</div><div> type performance/write-behind</div><div> option flush-behind off</div><div> subvolumes iocache</div><div>end-volume</div><div><br></div><div>volume trace</div>
<div> type debug/trace</div><div> subvolumes writeback</div><div># option include open,close,create,readdir,opendir,closedir</div><div> option exclude lookup,read,write,open,close,create,readdir,opendir,closedir,flush,readlink,fstat,stat</div>
<div>end-volume</div><div><br></div></div><div class="gmail_quote">On Wed, Jun 24, 2009 at 10:15 AM, Vikas Gorur <span dir="ltr"><<a href="mailto:vikas@gluster.com">vikas@gluster.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<br>
----- "Vidar Hokstad" <<a href="mailto:vidar@hokstad.com">vidar@hokstad.com</a>> wrote:<br>
<br>
What is your client configuration? Do you have a way to reproduce this?<br>
<br>
Vikas<br>
--<br>
Engineer - <a href="http://gluster.com/" target="_blank">http://gluster.com/</a><br>
<br>
A: Because it messes up the way people read text.<br>
Q: Why is a top-posting such a bad thing?<br>
<font color="#888888">--<br>
</font></blockquote></div><br></div>