[Linux-cluster] Starter Cluster / GFS
gordan at bobich.net
Fri Nov 19 15:03:33 UTC 2010
Nicolas Ross wrote:
>> I see. In that case, I dare say that the writes hitting one disk at a time is normal. It's a volume group, not RAID1 per se.
> Ok, that's what I assumed. Is it possible to have best of both world;-) ? That is, strip the data on both disk, using vg (semi raid 0), without using dm, and without using striping at the vg level (vgcreate -i) ?
I'm not sure LVM can do that, but I'm not a big fan of LVM so perhaps
somebody else can provide a definitive answer?
I keep thinking it'd be really nice if MD finally got cluster awareness,
but that feature has been coming for years and is till nowhere to be
seen, so I wouldn't hold my breath for it.
>> Interesting. If that really is the same GFS2 bug you're tripping that I bumped into 3 years ago and you're on RHEL6b, that is quite concerning. Doubly so if GFS1 support has been completely removed.
> How can I confirm that is what is happening ? Can you provide me with a test case ?
Back when I tried using it, it was reproducible by having > 1 dovecot
IMAP client running in the cluster with maildirs stored on GFS2.
Accessing mailboxes with only 2-3 users used to reliably deadlock it in
under 10 seconds. The only cure was a full cluster reboot. IIRC I was
told at the time here that the bug was one of the known ones, so I never
bothered investigating further.
More information about the Linux-cluster