Recently we had a power outage that affected several hosts and our fiber switches. After the smoke settled (literly,) we had 2 AIX boxes that had numbers in the Queued I/O column (on one of the 2 hba's) of powermt watch that would not go away. On one unit, iostats indicated an almost 100% disk activity on one of the SAN attached disks but the Connectrix trafic was 0. Other unit seemed to be working. But, the Q-I0 field never cleared. We had to reboot the hosts to clear the problem. The host with the high disk activity didn't want to unmount the file systems and we had to resort to a unit reset.
My question is, is there a more graceful way to flush, restart, reset, whatever, the hba rather than the rather drastic step of booting the host.
Thank you for the response. By manage, do you mean powermt remove? The docs state "On all platforms except AIX, powermt_remove does remove a path...if open." Since these are AIX hosts, does that mean if a path is open, it will be removed? Also, and more important, how do I restart, powermt config?
i was looking at "powermt manage" and "powermt unmanage" commands ..unfortunately i don't see an option to "unmange" a particular path ..you have to unmanage particular device.
Thanks again for the responses. When all else fails, RTFM. The docs on PP 5.1 reveal a new (from 4.5 docs) command for AIX and Solaris; powermt disable/enable. This sounds like it may be the silver bullet I was looking for. Since my suspect hosts were at 4.5.1, it wouldn't have helped in this case, but it will spur me to upgrade in case it will do what I want in the future. Until then, it looks like a stuck Q IO means a reboot.
dynamox
11 Legend
•
20419 Posts
•
87439 Points
784
0
Posted August 8th, 2007 19:00