ganglia-monitor will not start on boot

I checked the init script and it does contain:

# Required-Start: $network $named $remote_fs $syslog

Here is a workaround:

I added a tiny script to /etc/network/if-up.d/ that runs "/etc/init.d/ganglia-monitor restart". Scripts in this folder are run whenever a interface is brought up, and it seems to do the trick.

Not sure what's wrong with the package. I did not have this problem with ganglia-monitor 3.1.2 on Lucid.

Hi Mark,

I may be wrong, but I believe $network doesn't exist anymore as an init.d service, so the Required-Start: $network won't do anything.

The issue is that sysvinit scripts are started as soon as lo is up:

start on filesystem and net-device-up IFACE=lo

The simplest workaround is to change IFACE=lo to IFACE=ethX, but if there are any sysvinit scripts required to bring up ethX this may be problematic.

The eventual way to solve this is to make this into an upstart controlled service, and then you can just do

start on net-device-up IFACE=ethX

in just the upstart job.

Sorry, to clarify, the workaround of

start on filesystem and net-device-up IFACE=ethX

 would go in

/etc/init/rc-sysinit.conf

Removed the init scripts for gmond (ganglia-monitor) and gmetad and replaced them with upstart scripts (provided by Bernard Li of the Ganglia project).

This version checks to make sure you are building for Ubuntu 9.10 and above. If so, it creates upstart jobs for ganglia-monitor and gmetad. Otherwise it will use the original init scripts. This should make the change Debian-friendly.

Mark, IIRC, if there is a .init and a .upstart file, Ubuntu's dh_init will use the .upstart file, but Debian and older ubuntu's will just ignore it and use the .init file.

So I'm not sure you actually need all of that code. Did you try just putting both in and building it on Debian and Hardy?

I ask because this is how the current MongoDB packages are managed and it seems to work fine.

Oh first, Mark, thanks so much for the patch!

Also, 'start on started network' will not solve the issue that you reported, as it will just start gmond as soon as the interfaces have been started, which doesn't mean they're up. You will need to do

start on net-device-up

Plus for your particular issue, this wouldn't work because lo would fire the 'net-device-up' event, so you'd probably need to add IFACE=eth0. However that is too specific to your setup, so it probably wouldn't be acceptable in packages.

I would wait on doing any changes though, as we'll be defining some best practices for upstart jobs, and possibly defining more generic events (like 'multicast-ready' or something like that) during the upcoming UDS (which starts this Monday)

Up to date debdiff file with necessary changes.

First off, thanks for the great work on this Jonathan. We discussed in a private chat, but just to document the debdiff:

* start on net-device-up IFACE!=lo is not really a valid option. Servers often have *many* devices to bring up so starting on the first non-loopback is really not a valid solution. Since 11.10, runlevel 2 waits for all interfaces in /etc/network/interfaces (or 2 minutes to pass trying to bring up said interfaces). Given that, you want

start on runlevel [2345]
stop on runlevel [^2345]

* Just as a general convention, I'd expect respawn up in the block with 'expect fork'. Usually the primary process exec is the last thing.

* Its not clear at all why the debdiff adds a sysvinit script. Please explain.

This bug was fixed in the package ganglia - 3.3.8-1ubuntu1

---------------
ganglia (3.3.8-1ubuntu1) raring; urgency=low

  * Upstartify init scripts (LP: #656427, LP: #1078711).
 -- Jonathan Davies <email address hidden> Wed, 14 Nov 2012 15:23:46 +0000