Support for correlations on datasources

[ClementGautier]

Is your feature request related to a problem? Please describe.
I would like to configure our metrics/logs/traces correlations as code using the operator but I believe the feature isn't implemented yet because there is no GrafanaDatasource.spec.datasource.correlations field on the schema.

(If applicable)If your feature request solves a bug please provide a link to the community issue
N/A (I think I'm the first to raise the issue?)

Describe the solution you'd like
Ideally mapping all the fields available for provisioning described here, using the correlation API

Describe alternatives you've considered
The only workaround today is to set the datasource as editable and edit it in the UI, which isn't optimal

Additional context
N/A

Existing solutions
N/A

[theSuess]

Thanks for reporting this! It should be pretty straightforward to support this by adding the fields to the spec

Would you be up to implementing this? We're happy to review this and help you along the way

[ClementGautier]

I might be able to work on this next week yes, as you said it looks quite straightforward. I might need a little guidance about the reconciliation of the objects: each correlation need to be POSTed independently so what happens if you add 2 correlations, one gets created but the second fails? Shouldn't we implement a new resource (let say GrafanaDatasourceCorrelation) that takes a datasourceRef and get its own reconciliation cycle?

[Baarsgaard]

I'd consider a dedicated controller overkill when it's "One Datasource to Many Correlations" and that correlations do not introduce any security implications not already present in Datasources themselves.

Implementation wise, we have a similar case with GrafanaAlertRuleGroups where each rule has to be created individually.
But once created, they can be updated and or deleted as a group.

The implementation for AlertRuleGroups is pretty straightforward, return the first encountered error and continue with the next Grafana instance.
Assuming the error is either unique or the same across all Grafana instances it makes it easy to reason about the errors when they all appear in the status message .status.conditions[].message

If you're open for implementing the feature, I would recommend just implementing the fields as is to begin with, then we can follow-up with a .spec.datasource.correlations[].targetRef field to reference other GrafanaDatasources in the same namespace and resolve their UIDs.

It seems like it would be somewhat straightforward to track state with individual instances as it's possible to list all Correlations on a datasource and compare it with the list in the .spec.datasource.correlations

If you have more questions, feel free to ask or open a draft PR with questions 😃