Hi all,
I have been asked to configure logrotate for a specific customer on our server. The server hosts many virtual hosts and most dont have logging enabled (dont ask me why - I am just doing as I am told).
Anyway, the one that does, happens to generate large logs which is causing a problem.

I have looked at the logrotate.d/httpd file and think this is where I need to start. I am however a bit lost and wondered if anyone can help.

From what I can gather, adding the following should be correct:
/home/ovpw/logs/* {
missingok
notifempty
sharedscripts
monthly
rotate 4
create
compress

postrotate
/usr/sbin/apachectl graceful
endscript

}

My questions are: is the /usr/sbin/apachectl graceful correct ? The other entry in there uses: /bin/kill -HUP `cat /var/run/httpd.pid 2>/dev/null` 2> /dev/null || true
This looks a bit cumbersome to me.
Which of the two should I use ?

This should (If I have it right) keep 4 months of logs which will be compressed, rotating them monthly. It should also restart apache to ensure the logs are re-created..

Can I do this, and just add more blocks for each directory that I want to rotate (for example, if I need to do this for more Vhosts, do they go in this same file?) ?

Do I need to restart anything to make this lot work, or will it all just work as expected when cron.daily next kicks off logrotate ?

Thanks for any help

Rich

If you use a wildcard this way, make sure you rotate the logs into a different directory with the olddir option. Otherwise, you'll end up re-rotating old logs. Example:


Yes, use apachectl graceful. You do not need to send SIGHUP manually using kill. Apachectl graceful will properly send SIGUSR1, which will (among other things) cause httpd to re-open its log files.

Correct. You will have, e.g.:

You can add more blocks to the same file, but if you are going to specify all the same options, you can just add more log files to the existing block:


As long as cron.daily kicks off logrotate, you're done. However, I would test your logrotate config first to make sure it runs as expected. Use debug mode to see what will happen without actually rotating any files or modifying internal logrotate state. E.g. If the script you made is /etc/logrotate.d/apache:

Wow.
Thanks for such a comprehensive, and meaningful reply.
Everything is clear except one point which has me a bit lost.

Where you say I need the olddir entry to prevent re-rotating old logs.
If I look at the default config from /etc/logrotate.d/httpd, it has the following entry:

/var/log/httpd/*log {
missingok
notifempty
sharedscripts
postrotate
/bin/kill -HUP `cat /var/run/httpd.pid 2>/dev/null` 2> /dev/null || true
endscript
}

This results in /var/log/httpd/ containing:
access_log
access_log.1
access_log.2
access_log.3

This setup works (as I would expect) without using "olddir"

What I am hoping for, is to have (after 4 months of rotations)
/home/ovpw/logs/access_log
/home/ovpw/logs/access_log.1.gz
/home/ovpw/logs/access_log.2.gz
/home/ovpw/logs/access_log.3.gz
Then after the next month, to have .1.gz moved to .2.gz, .2.gz moved to .3.gz and .3.gz removed.


If I am being stupid here, please point out where I am wrong.

Thanks for all input

Yes, this will work for standard apache log names, without 'olddir'.

The reason is that '*log' will not match your old log names, which end in .1, .2, .gz, and so on. However, the '*' that you specified in you original post will match your old log names, so you need to put your old logs in some other directory to prevent re-rotation.

So:
If '*log' sufficiently matches all log names in the log directory, go ahead and use it without 'olddir'.

However, if you need to use '*' to sufficiently match all log names (e.g. because some of them have names that end in something other than 'log'), use 'olddir' to put old logs into some other directory.

Right that makes sense.

Thanks for the replies - all is as I wanted now.

Cheers

Rich