[monit] nova-api-metadata not monitored on CentOS and Ubuntu

[jameswthorne]

In RPC v4.2.2, on CentOS, the openstack-nova-metadata-api and, on Ubuntu, the nova-api-metadata services are not being monitored by monit.

[claco]

@jameswthorne Can you tell me a little about this environment? I just did a fresh install of v4.2.2 on Ubuntu, and I looks like it is being monitoring in my install on controller1. Will verify on CentOS shortly.

Process 'nova-api-metadata'
  status                            Running
  monitoring status                 Monitored
  pid                               32511
  parent pid                        1
  uptime                            15m
  children                          2
  memory kilobytes                  38740
  memory kilobytes total            110400
  memory percent                    1.8%
  memory percent total              5.3%
  cpu percent                       0.2%
  cpu percent total                 0.2%
  data collected                    Mon, 26 May 2014 14:13:19

[jameswthorne]

Hey @claco

I use the following Vagrant files to spin up an environment (these Vagrantfiles always pull from the specific branch and not master):

vagrantfile-rpcv422-ubuntu-neutron-networking.txt

vagrantfile-rpcv422-centos-neutron-networking.txt

I created a fresh environment from the Ubuntu Vagrantfile on May 23rd.

In /etc/init.d I see nova-api-ec2, nova-api-metadata, and nova-api-os-compute, but when running monit summary, I only see nova-api-ec2 and nova-api-os-compute being monitored.

root@controller1:~# monit status | grep metadata
Process 'neutron-metadata-agent'

I also just checked the RPC SE Lab, which is currently running RPC v4.2.2 on top of CentOS 6.5 and was build on May 9th, and openstack-nova-metadata-api is also not shown when running monit summary.

[root@controller01 ~]# monit status | grep metadata
Process 'neutron-metadata-agent'

[claco]

@jameswthorne Is the RPC SE Lab also using the single controller role setup? I doing the full ha-controller1/2 that i is the official way, and it is there. I suspect this is just a matter of things missing from the run_list when not doing ha-controller1/2.

[jameswthorne]

@claco

The RPC SE Lab is using the ha-controller roles. Let me re-kick the lab with Ubuntu and the ha-controller roles to see what I encounter.

[claco]

Specifically... ha-controller1/2 roles have "role[nova-api-metadata]", but single-controller nor single-network-node doesn't, so the monitoring wouldn't trigger a drop of the conf.d monit files if it doesn't see the nova-api-metadata in the run_list.

[claco]

@jameswthorne Ok, I want to say that everything is expected given the code/commit messages. The recommend install is to use hacontroller1 instead of single-controller.

On Ubuntu, there is indeed a nova metadata config in monit on the controller when using the ha roles.

Under Centos, this commit (https://github.com/rcbops-cookbooks/openstack-monitoring/commit/efe5bcc5cf6c9d1fda9779a8faa7d958200f33f7) leads me to believe that the metadata api is run within another service, so we would not drop a monit config for it.

However, on a fresh Centos 6.5 kick, I do see /etc/init.d/openstack-nova-metadata.api (in a stopped state). Maybe @invsblduck can shed some light on this. Maybe it's some leftover from Grizzly?

If this is not the case, then just removing that block that removes the metadata conf should fix it.