Mike Bailey

Getting monit to restart mongrel after a crash

An annoying aspect of the Mongrel webserver is that it refuses to start if it detects a stale pidfile. This causes real problems when you’re trying to use something like Monit to automatically restart mongrel after a crash.

Most daemons check whether the process_id in the pidfile is running. Ezra has indicated that a future release of mongrel will do this but in the meantime, we can use mongrel_cluster with the —clean option to remove stale pidfiles before starting mongrel.

Update /etc/init.d/mongrel_cluster to include the —clean option in the start and restart commands.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
  start)
    # Create pid directory
    mkdir -p $PID_DIR
    # chown $USER:$USER $PID_DIR

    mongrel_cluster_ctl start --clean -c $CONF_DIR
    RETVAL=$?
;;
  stop)
    mongrel_cluster_ctl stop -c $CONF_DIR
    RETVAL=$?
;;
  restart)
    mongrel_cluster_ctl restart --clean -c $CONF_DIR
    RETVAL=$?

Update your monit config to use mongrel_cluster. Note that monit sets a restricted path (PATH=/bin:/usr/bin:/sbin:/usr/sbin) and the internals of the mongrel_cluster gem call mongrel_rails without specifying the path. mongrel maintainers [http://mongrel.rubyforge.org/ticket/31#comment:1 suggest] using env in the monit command and said this is already fixed in a svn. I’ve found creating a symlink from /usr/local/bin/mongrel_rails to /usr/bin/mongrel_rails does the trick. Then update your monit config to look something like this:

1
2
3
4
check process mongrel-8010 with pidfile /var/www/apps/tubemarks/shared/pids/mongrel.8010.pid
group mongrel
start program = "/usr/local/bin/ruby /usr/local/bin/mongrel_rails cluster::start --only 8010 -C /etc/mongrel_cluster/tubemarks.yml"
start program = "/usr/local/bin/ruby /usr/local/bin/mongrel_rails cluster::stop --only 8010 -C /etc/mongrel_cluster/tubemarks.yml"

The nice part here is the —only option which allows you to restrict the command to a single mongrel process (as defined in the config file).

I’ve deprec-1.99.16 has been updated to use mongrel_cluster as described above to clean up stale pidfiles before starting mongrel. As a side note, I was glad to see mongrel has a new website and is being fed and cared for by it’s new owners.