Relating to multiple principals on the same machine

[simskij]

Enhancement Proposal

Currently, it's not possible to relate the grafana agent subordinate to multiple principals on the same machine. This is something we need to support as it's common in many solutions that deploy to machine substrates.

The suggested way of solving this, is by having each principal contribute a configuration fragment, and then having the snap entrypoint/wrapper concatenate them on the fly at startup.

[lucabello]

We can relate to multiple principals now; however, we still we need to deal with the multiple instances of grafana-agent overwriting each other.

[skatsaounis]

If a breaking change like the one below was introduced, would that solve the problem in an easier way?

• grafana-agent is changing from subordinate charm to machine charm.
• grafana-agent can be related with many machine charm apps through cos-agent relation, same cos_agent interface
• on any relation change from any related app, it is re-calculating the metrics, logs, traces since those are the different configuration bits per relation.

Rest of /etc/grafana-agent.yaml configuration options are grafana-agent specific so they can be managed by the existing charm configuration options of grafana-agent-operator.

[skatsaounis]

On second thoughts, there is also a need for a check on colocation which is solved when being subordinate. So I do not know but maybe you have a better idea.

[lucabello]

Related to #32

[sed-i]

Related:

• subordinate charms should have a limit in metadata juju/juju#18665
• https://discourse.charmhub.io/t/one-grafana-agent-charm-to-rule-them-all/16014