rsyslogd crashing on Debian 6.0 and freezing Apache

Is Apache configured to use syslog for logging? Normally, Apache handles writing to its own separate log files and shouldn't be affected by the presence or absence of syslog.

Typically a HUP signal is sent to a daemon to tell it to reload its configuration files, or to close and re-open any log files it is writing to. If this happens at a predictable time, I would guess that a log rotation script is sending the HUP signal to rsyslogd, but I don't know why that would cause it to freeze.

It could be that Apache2 is freezing because logrotate is kicking in to rotate your Apache2 logs as well. This would require logrotate to issue a reload(?) to Apache2 to tell it to close and reopen log files.

This would make the most sense if it's happening around the time system logs are being rotated out.

-Tim

I've done the following test:

# time logrotate -f /etc/logrotate.conf 
real    0m7.945s
user    0m3.007s
sys    0m0.529s

# tail /var/log/messages
Jul 10 14:37:27 server01 rsyslogd: [origin software="rsyslogd" swVersion="4.6.4" x-pid="5785" x-info="http://www.rsyslog.com"] rsyslogd was HUPed, type 'lightweight'.

And, interestingly, Apache was not affected by the log rotation. So it seems that I've been erroneously thinking that the HUP signal being sent to rsyslogd and Apache's behavior were somewhat related.

I guess I'll need to wait for Apache to freeze again to know what's really happening…

Thanks for your help anyway.

Ok. I've identified the problem:

In fact, Apache was freezing because of APC. When the APC cache becomes full, PHP apps, if wrongly written (and a lot of them are) tend to freeze.

My solution was to increase the APC cache from the default 32Mb to 128Mb and Apache has been working fantastically well for the past week now. It hasn't frozen once.