[NGINX] cache process manager consume a lot of CPU at reload and has to be restarted

[Maxim Dounin]

Hello!

On Fri, Apr 01, 2011 at 11:42:44AM +0200, Jérôme Loyet wrote:

> Hello,
> 
> I had an issue last night with a 0.8.54 nginx instance. Around 1AM, a HUP
> signal is sent to log rotation. This nginx instance is set up as a reverse
> proxy to an apache/php and it does some cache. Here is the proxy_cache
> related conf I've set up.
> 
> proxy_cache_path /CACHE levels=1:2:2 keys_zone=cache:256m inactive=1d
> max_size=20g;
> proxy_cache cache;
> 
> Last night, the cache process manager has restarted well, but it has
> consumed all CPU available on the server. Moreover, the SAR command
> indicates that it did not seek, read or write on the disks (as the disk
> usage remains almost null). Our exploitation team had to restart nginx 40
> minutes later and it solves the problem.
> 
> Has someone already experience such an issue ? Do you know why this could
> happened ?

Yesterday similar issue was discussed on russian mailing list, 
here:

http://nginx.org/pipermail/nginx-ru/2011-March/040396.html

Cache manger was spinning in ngx_spinlock() waiting for cache lock 
to be released.  Cache lock was never released as it was hold by 
worker process which caught SIGSEGV while holding cache lock.

Maxim Dounin