[imp] runnaway processes

[Uffe Pensar]

Michael Yingbull wrote:

> Uffe Pensar wrote:
>
>> Hi,
>>
>> We are running  horde 2.2.7/imp-3.2.6 on a Debian woody server with 
>> the latest stable releases of php, up-imapproxy, uw-imapd-ssl, mysql, 
>> sendmail, clam and apache-ssl . With about little over 100 concurrent 
>> users we have a load of 3-10. About 60 % or more use the 
>> webinterface. The problem is that sometimes the server load rises 
>> rapidly and and we can't see anything strange in the logs (it stops 
>> logging). In the end we have to reboot the hard way.
>> A couple of time I have seen inetd generating root owned imapd 
>> processes (one or two /second) and it wan't stop until inetd is 
>> killed. After a reboot  its up and working again.
>
>
> Those root owned IMAP processes are uw starting up - once there has 
> been a login, it will chown the process to the user that logged in.    
> If you are seeing lots of them building up, I'd suspect a problem with 
> your authentication.

I suspected that too, but its nice to get it verified.

>   Most of the time I've had a problem with a sudden load average spike 
> (to rediculous levels) it's usually been a network problem (do you use 
> ldap for authentication?  nfs mount parts of your mail system? etc?).

We authenticated from three servers but from this morning its only two.:-)
(and both servers are now in the local net).
Not ldap but radius and thats was perhaps the critical part.

>
> Check for any problems on any other hosts that tie into that machine 
> in any way during that time.    You may want to try having Ethereal 
> handy and use it when the problem is happening to see what you see on 
> the network.    

The problem is/was that it happened so quickly. Our nagios -server 
checks the email server every
three minutes but it was always too late. Its was only when I was 
sitting working at the console that
I could see what was happening.

How about compiling and tuning uw-imapd-ssl  or  another package ?