On 12/30/2010 11:07 AM, Mark Martinec wrote:
> Clay and Christian,
>
>> With some of my preliminary testing of -pre12 upon issuing the "amavisd
>> reload" command I see the old child processes lingering around in a
>> "<defunct>" state.
>> The amavisd-nanny process only shows the newly created children.
>>  From a debug console everything appears to be functioning normally:
>> Subsequent reloads will generate another set of zombies.
>
> Thanks for the report. I haven't observed such behaviour so far
> (tried even on Linux), but it definitely needs investigating.
>
>> Could this be due in part to the new warm restart and how the parent is
>> handling the HUP handling the child processes? Just throwing that out as
>> an idea...
>>
>> Here is the output showing the state of the children after the reload
>> command and you can see older child processes still indicating a PPID of
>> the master:
>
> Looks like the HUPed master process after a rebirth (exec) fails to reclaim
> a process exit status of its previous set of child processes. Handling
> of child processes and reclaiming their exit status is entirely in hands
> of Net::Server, amavisd is not dealing with these by itself.
>
> Luckily these defunct processes no longer consume any resources
> on the host, except for a process slot - they no longer exist, it's only
> their exit status value that is still hanging around, waiting to be read
> by their parent process. If you do a full (cold) restart of amavisd
> eventually, all their zombies will be reclaimed by the init process.
> So yes, it is ugly, needs investigating, but is not burdening the
> host's memory or cpu resources.
>

Agreed.

> I can see the perl is 5.008008, but which version of Net::Server is that?
> And what OS?
>

# cat /etc/redhat-release
CentOS release 5.5 (Final)

Kernel: 2.6.18-194.26.1.el5

Net::Server package is an rpm provided by RPMFORGE:
perl-Net-Server-0.99-1.el5.rf.noarch

As an added note I did not see the zombies in 2.5.4 or 2.6.4.

Let me know if you need further testing or info.

Clay

------------------------------------------------------------------------------
Learn how Oracle Real Application Clusters (RAC) One Node allows customers
to consolidate database storage, standardize their database environment, and, 
should the need arise, upgrade to a full multi-node Oracle RAC database 
without downtime or disruption
http://p.sf.net/sfu/oracle-sfdevnl
_______________________________________________
AMaViS-user mailing list
AMaViS-user@lists.sourceforge.net 
https://lists.sourceforge.net/lists/listinfo/amavis-user 
 Please visit http://www.ijs.si/software/amavisd/ regularly
 For administrativa requests please send email to rainer at openantivirus dot 
org

Reply via email to