Drop the monitoring stack from the up phase

[msimonin]

Kolla can deploy influxdb + grafana + collectd + telegraf. The agents are configurable through the node_custom_config parameter.

@rcherrueau, should we get rid of the corresponding roles in the up phase ?

Pros:

• less code to maintain
• an up-to-date code
• it would be easy to use a dedicated network for all the monitoring traffic (this requires a small fix in Kolla)
• (no regression on cadvisor) Telegraf as a docker input to monitor the running dockers
  https://github.com/influxdata/telegraf/tree/release-1.3/plugins/inputs/docker

Pros or Cons (depends)

• configuration will be done through node_custom_config mecanism

Cons

• if Enos as a general purpose framework aka reusable up phase #200 we should probably keep our own monitoring stack (shared accross the applications supported by EnOS)

[rcherrueau]

Not sure how to do that. In the actual implementation activating monitoring is as simple as enable_monitoring: yes and I would appreciate that it stay like that (leveraging kolla or not). Btw, the keynote at Cloud Day's this morning was from the CEO of CloudScreen and he gave really good feedback on prometheus.

[msimonin]

I had a quick look :

kolla:
  enable_telegraf: yes
  enable_collectd: yes
  enable_grafana: yes
  enable_influxdb: yes

This is obviously not as simple (at least in the current status of Kolla and EnOS). But we could keep enable_monitoring: yes|no and generate those variables at runtime if needed.

I see some advantages nevertheless :

• less code to maintain
• an up-to-date code
• it would be easy to use a dedicated network for all the monitoring traffic (this requires a small fix in Kolla)

Let's iterate on the pros and cons :)

[rcherrueau]

This is a nice list of pros 👍. Let's do this!

[msimonin]

I've put the list at the top level

[msimonin]

just added a cons