|
Hi everyone! After a bit of an ordeal with our Gluster servers last week, I discovered some very important coincidences that can very badly affect Gluster performance when they occur. Should the Mlocate updater start when Gluster is going through the self-heal process, especially one that is very long (ours can take a few days sometimes), Gluster's performance can become extremely poor with symptoms of high server load, moderate to high CPU usage, high disk IO activity, and occasional to lengthy times where there is 100% disk utilization. IO Wait times on any clients can be high to extreme. Memory usage might also be higher than normal, and on servers with less than say, 8 GB of RAM, this might cause RAM to be swapped to disk, exacerbating the problem further. The Mlocate update program "updatedb" is installed by default on most Linux implementations, and is run nightly by default in the daily cron jobs. I have added a new file to the Gluster installation documentation (named "post-installation") with the recommendation that this daily cron job be removed. Removing this job would have little to no impact on the operation of Gluster's operating system. If anyone would like to test and verify this in a non-production
environment, be my guest. |
_______________________________________________ Gluster-users mailing list Gluster-users@xxxxxxxxxxx http://lists.gluster.org/mailman/listinfo/gluster-users
