On Tue, 2009-01-13 at 19:20 +0100, Jan-Frode Myklebust wrote:
On Tue, Jan 13, 2009 at 12:46:04PM -0500, Timo Sirainen wrote:
Reading your old mails: are you still using GPFS? This crash just shouldn't be happening, so perhaps something randomly breaks with it. Are you using mmap_disable=yes? Multiple servers can access the same user's mails at the same time?
Yes, we're using GPFS, but haven't turned on mmap_disable=yes. I believe GPFS should handle mmap across the cluster, and it's been working fine without this setting with the old v1.0.0+.
Do you have any idea what the performance impact should be of turning on mmap_disable=yes ?
I don't think there would be much of a performance impact. With GPFS it might actually even be a positive impact (especially with v1.0).
I don't really have ideas why it was working with a previous v1.0 version but not with v1.0.15. Although perhaps I have just added more error checks and the previous version just handled the error condition wrong..