i/o problem

[marotori]

This thread...

http://forum.proxmox.com/threads/8661-Proxmox-soft-lockups-on-megaraid-controller?p=49095#post49095

Seems to have an answer! At least in my initial tests!

In summary....

# this option stops the disk soft lockups!
echo 0 > /sys/block/sda/queue/iosched/slice_idle

 

[drd]

In summary....

# this option stops the disk soft lockups!
echo 0 > /sys/block/sda/queue/iosched/slice_idle

So, after the intial tests... was this the solution? (Judging by more recent posts it seems that this didn't help either?)

Thanks,
David

 

[marotori]

Sadly.. I have no 100% fix. Still searching.

 

[drd]

Currently I'm thinking it might have something to do with the fact that my VMs are on a shared LVM volume group (shared across cluster nodes), and that one node starts the backups while the other hasn't finished yet.

Is that somehting that may happen in your environment as well?

 

[marotori]

Nope. Mine are openvz boxes on a single server.

I can actually generate the same issues by hammering a clean server - no vm's with loads of random io.

 

[drd]

Good point. On second thought, my nodes do share the same I/O controllers (Intel Modular Server), so overlapping backup windows can also create higher I/O than normal. If it's not a concurrency issue, it can still be a load issue. In any case, I'll make sure the backup windows don't overlap, it sure won 't hurt...

 

[drd]

Well, I can confirm: also with non-overlapping backup windows I got the crash last night...

 

[bazzi]

We also have a non overlapping backups and crashes from time to time.

 

[bfal]

Hello,

i have the same Problem as described here. On my Boxes the Problem is
that the Logical Volume switch to suspend State and resumed only
if i resume the volume manual by "dmsetup resume $LV".

At the moment i have workaround the problem by the following
script which checks every 5 minutes (by Crontab) if the volume ist suspended
and resumed the volume if needed.

Maybe someone can help this simple script.

#!/bin/sh

LV="<<<insert volume name>>"
LVSTATE=$(dmsetup info $LV |awk -F: 'NR == 2 {gsub(/ /,"");print $2}')

if [ "$LVSTATE" = "SUSPENDED" ]; then.
    /sbin/dmsetup resume $LV
fi
exit 0