Errors with count:User quota and NFS

[Tom Sommer]

On 2016-11-01 09:47, Tom Sommer wrote:
> On 2016-10-31 22:04, Timo Sirainen wrote:
> 
>>> Oct 31 10:52:37 imap(xxx at xxxx.xx): Warning: Locking transaction log 
>>> file /mnt/nfs/xxx.xx/xxx/indexes/dovecot.list.index.log took 31 
>>> seconds (syncing)
>>> Oct 31 10:52:37 imap(xxx at xxx.xx): Warning: Locking transaction log 
>>> file /mnt/nfs/xxx.xx/xxx/indexes/dovecot.list.index.log took 31 
>>> seconds (syncing)
>> 
>> This just means something is being slow. Not necessarily a problem.
>> Although it could also indicate a deadlock. Is this Maildir? Did you
>> say you were using lock_method=dotlock?
> 
> I removed dotlock some time ago (using director) and switched to:
> 
> lock_method = fcntl
> mail_fsync = always
> 
> With "count" as quota backend I get a lot of these errors on the 
> director:
> 
> imap-login: Error: proxy(xxx at xxxx.xxx): Login for xxx.xxx.xxx.xxx:143
> timed out in state=2 (after 30 secs, local=xxxxx:58478):
> user=<xxxx at xxxx.dk>, method=CRAM-MD5, rip=xxxx, lip=xxxxx, TLS,
> session=<pTl0UTlAV8VWNF5+>
> 
> on the server:
> 
> imap(xxxx at xxxx.xxx): Warning: Locking transaction log file
> /mnt/nfs/xxxx.dk/xxxxx/indexes/.INBOX/dovecot.index.log took 32
> seconds (appending)

Actually this is worse than I thought. I don't know if it's because of 
the lock, or it's a general bug in 'count', but whenever I switch quota 
backend to 'count', customers complain that they aren't receiving any 
mails.
Mails are stored correctly on the server, but it seems Dovecot doesn't 
show them, perhaps due to corrupt/hanging/locked indexes.

So either 'count' is very I/O sensitive by design, and so useless on 
NFS, or there is some bug in there that breaks indexing.

Using maildir and lmtp.

The only errors I see in the logs, are the ones pasted here.