No aggregator service running when the toolchain is deployed without controller

[Simon Pasquier]

In a so-called dedicated StackLight environment (where you deploy only the StackLight backends), there's no aggregator service because there's no controller. As a consequence, there's no GSE metric for the StackLight backends.

[guillaume thouvenin (guillaume-thouvenin)]

I proposed to fix this issue in two steps.

1) We quickly fix it without deploying the aggregator client. So this dedicated environment won't be monitored. It answers to the use case in which we want to deploy only backends and those backends will be used by other environments.

2) It is a better solution to also monitor the monitoring environment so in this step we will deploy the aggregator server. We will do this in a second step because it requires more work and currently the deployment in dedicated env is broken. So that's why I propose to fix it and then improve it.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/312050

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/312050
Committed: https://git.openstack.org/cgit/openstack/fuel-plugin-lma-collector/commit/?id=bdd104eb04233d98d8e0138a8c18fb602904a0d5
Submitter: Jenkins
Branch: master

commit bdd104eb04233d98d8e0138a8c18fb602904a0d5
Author: Guillaume Thouvenin <email address hidden>
Date: Tue May 3 15:02:22 2016 +0200

    Don't deploy aggregator on dedicated environment

    On a dedicated environment, without controllers, there are no
    servers to aggregate metrics. So this patch prevents the deployment
    of aggregator clients.

    Change-Id: Ib4d49a8f4d255b1a65c2e7f3b7f18a7b7e420108
    Partial-bug: #1574609

[Simon Pasquier (simon-pasquier)]

A few ideas on how to fix this:
- Hiera-ify the aggregator configuration so it can be decided upfront where the aggregator servers will be running, what is the aggregator VIP, etc.
- Decouple the aggregator" election from Pacemaker. Right now the information is hardcoded in the GSE filter (gse_cluster_filter.lua). Ideally there should be another filter that consumes the Pacemaker metric and spits out well-known messages that define whether or not the instance is the current aggregator.
- Decouple collectd from Pacemaker too.