nginx log rotationEdit

Overview

So what do you need to know to handle log rotation for nginx? The official answer from Igor Sysoev (the nginx lead developer) can be found here. In short, the sequence you must follow is:

  1. Move (or otherwise process) the old log file
  2. Send the nginx master process a USR1 signal

logrotate setup

There are many ways that this kind of processing can be automated but I use logrotate. On RHEL 5.1 its configuration files can be found in /etc/logrotate.d/. For example, consider the file, /etc/logrotate.d/httpd, which handles the rotation of Apache log files:

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

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

Using this as a model, and consulting the logrotate man page, we come up with a file like this for nginx, /etc/logrotate.d/nginx. Obviously the paths on your local system will most likely vary. Here I specify log rotation for two separate Rails applications, each of which has its own nginx master process (note the different PID files).

/home/path_to_rails_app_1/deploy/shared/log/*_log {
    missingok
    notifempty
    sharedscripts
    postrotate
        test ! -f /var/run/nginx.pid || kill -USR1 `cat /var/run/nginx.pid`
    endscript
}

/home/path_to_rails_app_2/deploy/shared/log/*_log {
    missingok
    notifempty
    sharedscripts
    postrotate
        test ! -f /var/run/nginx-staging.pid || kill -USR1 `cat /var/run/nginx-staging.pid`
    endscript
}

I’m actually quite glad that I looked into this because in doing so I noticed that having nginx write its log files to the same location as Apache wouldn’t be a good thing (this is the way I initially had it set up). As you can see, these log files would be rotated according to the rules in the /etc/logrotate.d/httpd file and nginx would never receive the requiredUSR1 signal.

Testing

# -d turns on debug mode: output is verbose, no actual changes are made to the log files
sudo logrotate -d /etc/logrotate.conf /etc/logrotate.d/nginx

# if the debug output looks good, proceed with a real rotation (-v turns on verbose output)
sudo logrotate -v /etc/logrotate.conf /etc/logrotate.d/nginx

In my initial testing this log rotation did not work as expected; the verbose output ended with lines like the following:

renaming /home/path_to_app/deploy/shared/log/access_log to /home/path_to_app/deploy/shared/log/access_log.1
disposeName will be /home/path_to_app/deploy/shared/log/access_log.1
running postrotate script
removing old log /home/path_to_app/deploy/shared/log/access_log.1

In other words:

  1. Rename access_log to access_log.1 (correct)
  2. Run the postrotate script (correct)
  3. Remove oldest log file, access_log.1 (incorrect!)

That last action is the unexpected one, because in my /etc/logrotate.conf file (where global settings are stored) I have:

# keep 4 weeks worth of backlogs
rotate 4

Troubleshooting permutations attempted:

  • Remove nocreate line from /etc/logrotate.d/nginx (no visible effect)
  • Add explicit rotate 4 to /etc/logrotate.d/nginx (works)

For each permutation is is necessary to hit the website (to make sure the log is non-empty) and also pass the -f (force) flag to logrotate, otherwise it will most likely refuse to rotate the log file because of its size. I finally realized that my initial runs did not behave as expected because I wasn’t passing in /etc/logwatch.conf as the first parameter (it is not pulled in automatically); in other words, to test out a custom file in /etc/logwatch.d as it would really be executed in normal use you must precede it with /etc/logwatch.conf on the command line. Once I realized this I removed the explicit rotate 4 from the /etc/logwatch.d/nginx configuration file.

Rotating other Rails log files

The nginx log files are only part of the situation. When you run mongrel_cluster you’ll get another log file mongrel.12345.log (where 12345 is the port number the mongrel instance is listening on) for each mongrel instance in the cluster, as well as the traditional production.log, development.log, test.log and any others that your application might produce.

See: