rewrite service.sls to be able to use it with a cluster manager like Pacemaker

[hoonetorg]

See this commit and comment: https://github.com/hoonetorg/haproxy-formula/commit/cc242d799ce0201e91d15153f41e6d4fda1db35a

Very short the current situation: enable: True -> haproxy starts at boot and runs enable: False -> haproxy doesn't start at boot and will be stopped at salt run if started.

Clustermanager needs:

• service disabled at boot and
• control by itself if started or stopped

[hoonetorg]

I could create a PR if you want.

[johnkeates]

Looks a bit convoluted to me, and doesn't seem to be taking systemd in to account?

[hoonetorg]

service.(running|dead|enabled): also works with systemd.

Regarding convoluted: using a default.yml

  service:
    name: template
    state: running
    enable: True

and then use variables instead of that long salt['pillar.get']... doesn't look that bad anymore :).

# -*- coding: utf-8 -*-
# vim: ft=sls

{% from "template/map.jinja" import template with context %}

template_service__service:
  service.{{ template.service.state }}:
    - name: {{ template.service.name }}
{% if template.service.state in [ 'running', 'dead' ] %}
    - enable: {{ template.service.enable }}
{% endif %}

What do you think?

I could first assign all service related stuff to variables and then it would look similar.

[hoonetorg]

Also service.dead doesn't support reload option. I would fix that before creating a PR.

[johnkeates]

It's the whole file.replace deal I was targeting. Anyway, best way would be:

• Create a default that starts + enables
• Lookup service:autostart(bool) and service:run (bool)

Then service can be used to always reload, set autostart to the value of service:autostart and start or kill the service depending on service:run.

This way, the terminology is somewhat more explicit, which is important since we are deviating from the normal process (which is: let the OS decide and start+enable by default).

I think I'll patch in map.jina and defaults.yaml so this can be overridden in a more standardised way using the lookup sub-pillar.

[hoonetorg]

these are the 2 rendered sls snippets which would make 99% of people happy.

non-cluster (that is already working):

haproxy.service:
  service.running:
    - name: haproxy

    - enable: True
    - reload: True

    - require:
      - pkg: haproxy

cluster(not possible yet):

haproxy.service:
  service.disabled:
    - name: haproxy

    - require:
      - pkg: haproxy

The changes to the file.replace /etc/default/haproxy part were only done to make the regex more failsafe (If somebody puts blanks/tabs before/after ENABLED

[hoonetorg]

@johnkeates: Will you prepare something on that?

Remark: I can imagine what you want to achieve, but I'd not recommend to rename enable to autostart (because it would destroy backward compatibility): pillar:

haproxy:
  enable: (which defaults to True, and only means: do start on boot/autostart)
  running: (which defaults also to True and in case of a change in haproxy.cfg triggers a reload).

[johnkeates]

Good point, I guess it stays ;-)