notify systemd when consul daemon is really started

[ymartin-ovh]

Have the following issue on my environment:

puppet runs:

# Notice: /Stage[main]/Consul::Run_service/Service[consul]: Triggered 'refresh' from 1 event
# Warning: Cannot retrieve ACL token list: Failed to open TCP connection to localhost:8500 (Address family not supported by protocol - socket(2) for "localhost" port 8500)

I suspect that systemd returns to fast and assume that daemon is ready but it is not the case.
Since consul can notify systemd, I change type from simple (default when type is absent) to notify.

External links:
https://www.hashicorp.com/resources/systemd-the-good-parts

[solarkennedy]

You are right, and it is set like this in an "upstream" version of this unit file:
https://learn.hashicorp.com/tutorials/consul/deployment-guide#configure-systemd

That's cool!

[ymartin-ovh]

Thanks for the link. It seems, we need to keep type=exec for standalone deployment.
Don't know if consul::config_hash['retry_join'] > 1 is sufficient to select the right server type.

Regards

[chrisboulton]

I was about to go implement support for Type=notify in this same module before I came across this change and realised we were just using an out of date version.

Our environment relies on auto-discovery support in Consul, which means our configuration for retry_join only has one entry:

  "retry_join": [
    "provider=gce tag_value=.. zone_pattern=.."
  ],

I'm wondering if it makes sense to submit a PR that changes this to Type=notify always, and what the downside of that change would be? (alternatively maybe we move this into a configurable param and remove the determining logic?)