[Dovecot] 2.0.3 Bug - hanging issue

[Alan Brown]

 > From: Timo Sirainen <tss at iki.fi>
> Subject: Re: [Dovecot] 2.0.3 Bug - hanging issue
>
> On 25.10.2010, at 14.13, Alan Brown wrote:
>
>> >  Oct 25 12:36:34 msslat dovecot: imap(keh2): Error: user keh2: Initialization failed: Namespace '#mbox/': mkdir(/stage/mail/imap1/keh2/mail) failed: Permission denied (euid=2291(keh2) egid=101(luci) missing +w perm: /stage/mail/imap1, euid is not dir owner)
>> >  
>> >  Because the parent directory was rwxr-xr-x root.root
> Right..
>
>> >  The problem is that for some reason dovecot then wedged and a lot of other stuff piled up behind it, causing people to complain their mail wasn't coming through. Load average climbed to over 100.
>> >  
>> >  Creating the user directories freed things up...
> So the problem is that if there's a misconfiguration, load goes up? I don't really know if there's much to do about this. I guess it could wait a second or so before disconnecting client to avoid it reconnecting back too fast..
>

The client wasn't reconnecting as far as I can tell.

It looked to me like the server process wedged after this error - we 
only got the error message logged once.

The problem is that after it occurred no new clients could complete 
connections, but existing clients worked ok. Incoming connections were 
accepted, but then nothing would happen.

As far as I can tell the parent would fork new child processes but they 
couldn't complete initialisation. Mail clients simply complained of a 
timeout during connection. This was affecting all users.


With regard to error disconnections - yes a delay is a good idea.