Amavis process pinning CPU at 100% returning after 5 to 30 minutes

[Arrigo Triulzi]

On 29 Aug 2022, at 16:51, Damian <amavis at arcsin.de> wrote:
> 
>> I was wondering if anyone is seeing similar behaviour or has any recommendations to debug this further.
> Depending on your mail throughput, you could increase $log_level up to 5 and see what is being logged right before the lockup. Additionally a mail which resides 30-40 minutes in processing is easy to grab from $TEMPBASE. Maybe you can then reproduce the issue.

As I have SSDs I think I will have to resort to upping the $log_level as you suggest - I did try figuring out what was special about the emails but injecting them into another process they take literally seconds to get through, this is why it is so frustrating.  What I do is that I kill the pinned process, Postfix queues the mail for processing again as 127.0.0.1:10024 stops responding, and then when you postqueue -i on the queue-id it goes through immediately (this is actually what my horrible hack is doing automatically).

I was hoping that someone else had encountered this issue.

One suspicion could be DNS (because when you re-inject the DNS result will obviously be cached) but both resolv.conf and /var/spool/postfix/etc/resolv.conf both point to the local resolver… 

Arrigo