Dead Simple Ruby Daemons Using Looper

2009-01-20 19:00:00 -0500


There seems to be a bit of interest in the Ruby community lately in putting together handy libraries for creating Ruby daemons, so I thought I’d throw our own particular solution into the mix. Our solution intentionally avoids a lot of the common daemonizing tasks (like detaching from the terminal) so it might not be quite appropriate to call it daemonizing, but it fills our requirements nicely. (As an aside, check out Kenneth Kalmer’s new library daemon-kit for a nice actual-daemon implementation in Ruby.)

We originally tried coding things by hand using Daemons, and then modularizing our oft-repeated code to stay DRY, but we kept running into stability issues. Also, when you fork a Ruby daemon, particularly one that’s loaded up a large stack in memory, you end up at least briefly using twice as much memory. So, memory hogging and crashing needed to be avoided.

None of our daemons required advanced process handling, either. They were simple, single instance workers that had very particular tasks (like schedulers and message handlers for PingMe). We found ourselves asking, “do we really need to be forking child processes for this?”

The answer is no. Instead we are using a module we just pushed to Github that we call Looper. It allows us to take any bit of code or a class and run it as a kind of daemon. Or, more specifically, the class uses Looper to take advantage of its simple signal trapping, loop handling (including sleeping), and the class can rely on Looper to catch any unhandled exception, report it, and keep on keepin’ on.

The loopme method takes a block that it runs for you, handles sleeping between runs, and will catch any unhandled exceptions that bubble up. Thus, if you want to exit on a particular exception, you’ve got to rescue it in your code and set @run to false. It also sets up signal trapping so that signals TERM, INT and HUP will all cause the loop to end (I’d be open to changing this behavior or adding additional signal responses if anybody has an interest).

Here’s an example “daemon” that uses looper to kick it:



require 'looper'
require 'twitter'

class DoSomething
include Looper

def initialize(config)
@run = true
# do config stuff, etc...
@sleep = config[:sleep].nil? ? 60 : config[:sleep]
end # initialize

def run
loopme(@sleep) do
begin
# this is where the meat of your code goes...
messages = twitter.direct_messages({:since => Time.now.strftime("%a, %d %b %Y %H:%M:%S %Z")})
rescue Twitter::EpicFailure => e
puts "bailing out, dude!"
# set run to false to put the kabosh on the next run
@run = false
end
end
end
end

# and here's how we kick it off:
DoSomething.new( { :sleep => 10 } ).run

Pretty simple, right? No coding up the signals, the sleeping, the global exception handler, etc. You can take a look at looper.rb to see exactly what it does for you. From here, starting and stopping our script is really easy, and can be re-used for each of our daemons. It boils down to starting it up with nohup (and script/runner because we like having our rails stack), and then killing it later with the PID:

$ nohup script/runner -e RAILS_ENV /path/to/DoSomething.rb

There’s no need to use Rails’ script/runner, you could just call ruby itself and save yourself a lot of memory ;-)

The following is an init script that we use to start and stop any of our Looper daemons:



#!/bin/bash
#
# chkconfig: 345 70 30
# description: control script for daemons
#

usage(){
echo $"Usage: $0 {start|stop|restart} {daemon1|godzilla|voltron}"
}

# daemon name is required as second param
if [ "$2" = "" ]
then
usage
exit 5
fi

# RAILS_ENV?
if [ "$RAILS_ENV" = "" ]; then
RAILS_ENV='development'
fi

# RAILS_DIR?
if [ "$RAILS_ENV" = "production" ]; then
RAILS_DIR='/www/app/current'
else # assume development
RAILS_DIR='/www/dev.app/current'
fi

PROGRAM="${RAILS_DIR}/daemons/${2}/looper.rb"
RUNNER="${RAILS_DIR}/script/runner"

LOG="/var/log/app/daemon_${RAILS_ENV}.log"

# if the daemon is missing, well, the daemon is missing!
test -f $PROGRAM || exit 5

# See how we were called
case "$1" in
start)
nohup $RUNNER -e $RAILS_ENV $PROGRAM >> $LOG 2>&1 &
;;
stop)
PID=`ps auxw | grep ${PROGRAM} | grep -v grep | awk '{ print $2 }'`
echo "Stopping PID: ${PID}"
kill $PID
sleep 2
;;
restart)
$0 stop
$0 start
;;
*)
usage
exit 1
;;
esac

exit 0
Zetetic is the creator of the super-flexible Tempo Time Tracking system.

blog comments powered by Disqus