On Tue, 25 Jan 2011 01:11:53 +0200, Timo Sirainen tss@iki.fi wrote:
On 25.1.2011, at 1.06, list@airstreamcomm.net wrote:
Multi-server setup that tries to flush NFS caches:
dotlock_use_excl = no # only needed with NFSv2, NFSv3+ supports O_EXCL and it's faster
You're probably using NFSv3, right? Then this isn't needed.
Also not the "tries" word. It doesn't work perfectly, although in your case it seems to be working than expected. Still, these NFS problems are the reason I created director: http://wiki2.dovecot.org/Director
We are using NFSv3, and for five months the system worked with four dovecot servers and three postfix servers all accessing the same NFS server simultaneously. We cannot pick out a change in our network or on the virtual environment that our machines resides that would have impacted the system this drastically. We have also confirmed that our clocks on all systems accessing the NFS server and the NFS server itself are within 1 second of each other. It's confounding us why the logs show such strange time stamps:
Jan 25 11:30:11 1295976611 POP3(4eagles): Warning: Created dotlock file's timestamp is different than current time (1295976643 vs 1295976607): /mail/4/e/4eagles/Maildir/dovecot-uidlist
We added the epoch time to the log_timestamp setting to compare it to the dotlock error, and as you can see the "Created dotlock" epoch time is 32 seconds in the future compared to the epoch time of the log event. At this point I hope you can help us understand where a timestamp from the future might be generated from.
As for the director, we will be considering the option after doing some heavy testing.
Thanks,
Michael