Puppet – How to Use Custom Status Command for a Service

debiandebian-squeezepostgresqlpuppet

I am using debian squeeze with PostgreSQL 9.1 from backports. Puppet has version 2.7.14.
Unfortunatly the init script returns the wrong exit code for status. Therefore I wrote a custom status command to detect whether postgresql is running or not.

service { 'postgresql':
  ensure => running,
  enable => true,
  hasstatus  => false,
  hasrestart => true,
  status => "pg_lsclusters -h | awk 'BEGIN {rc=0} {if ($4 != \"online\") rc=3} END { exit rc }'",
  provider => debian,
}

My command works like a charme, but puppet seems to have a problem. I always get notice: /Stage[main]/Postgresql/Service[postgresql]/ensure: ensure changed 'stopped' to 'running' although it is already running.

So tried the following:

service { 'postgresql':
  ensure => running,
  enable => true,
  hasstatus  => false,
  hasrestart => true,
  status => "exit 0",
  provider => debian,
}

As I understood this custom status command, puppet should always think that postgresql is running. Nevertheless puppet tries to start postgresql – every time.

What is my fault? Or is it a bug in puppet?

Best Answer

My best guesses are that the $4 in your command is getting swallowed up by puppet's own interpolation and that exit 0 doesn't quite work right due to shell interaction issues.

I would try a few things.

If the problem is puppet's interpolation on $4 in your command escape the $ like so: status => "pg_lsclusters -h | awk 'BEGIN {rc=0} {if (\$4 != \"online\") rc=3} END { exit rc }'" (sometimes more backslashes are required, but I'm pretty sure 1 is enough here).
Make sure the test command is really working right. exit is a shell internal and I'm not sure how puppet will treat that. So use the canonical "return success" command instead: status => "/bin/true"

Maybe status is being overridden by provider => debian (which would be a puppet bug), so instead specify all the commands and use the base provider (this won't enable properly, however):

service { 'postgresql':
  provider => base,
  ensure   => 'running',
  start    => '/etc/init.d/postgresql start',
  restart  => '/etc/init.d/postgresql restart',
  stop     => '/etc/init.d/postgresql stop',
  status   => "pg_lsclusters -h | awk 'BEGIN {rc=0} {if (\$4 != \"online\") rc=3} END { exit rc }'",
}

Related Solutions

Specifying prerequisites for Puppet custom facts

As far as I know, you can't. Either let it fail, or detect and fail gracefully. I have a number of plugins that only work on Debian that fail on Red Hat without consequence.

Also, please note that it is IMPOSSIBLE to have a fact evaluated AFTER some configuration went in. The architecture just doesn't support it:

Client                               Server
Compute facts
Ask for catalog passing facts =>     Receive Catalog request
                                     Compute catalog using facts
                              <=     Return Catalog
Based on the dependency tree,
  For each configuration with satisfied dependencies
    Apply configuration
    Mark (or not) dependency as satisfied
Send report, if configured so

So, you see, the configuration is only applied long, long after the facts were processed, and there's no way of going back. What may happen is that the next run will now be able to generate that fact.

See also the trick used by the Common module to handle the lack of lsbrelease on Debian without producing fatal errors.

Postgresql – How to ensure a service is running, using Chef

By default, Chef will check if the service is running, and starts it if the service isn't running.

How it determines that the service is running depends.

By default, Chef will attempt to match the name of the service (postgresql here) in the process table using ps.

ps -ef | grep postgresql

Essentially. The name of the service will be used for the pattern match when inspecting the process table. This may or may not be what you want/need, especially depending on the platform and how it names the "postgresql" service.

However, you can tell Chef that the service supports a "status" command, which means Chef will generally do something like,

/etc/init.d/postgresql status

And use the return code to determine if it's running or not (non-zero is not running).

Chef doesn't do this by default because not all service scripts support a status command (frustratingly), and Chef doesn't innately know what the right thing to do is. It tries to do the sane default thing, but sometimes naïve. Thus, you can tell Chef that the resource has a status command and not be so naïve.

service "postgresql" do
  supports :status => true
  action :start
end

Now, if the service is not actually named "postgresql" but is instead "postgresql-92" or similar, you can do this as:

service "postgresql-92" do
  supports :status => true
  action :start
end

service "postgresql" do
  service_name "postgresql-92"
  supports :status => true
  action :start
end

You can find out what's going on in more detail by running chef with debug output, too:

chef-client -l debug

Best Answer

Related Solutions

Specifying prerequisites for Puppet custom facts

Postgresql – How to ensure a service is running, using Chef

Related Topic