Component label + local service definition

[mr-miles]

Most importantly (and trivially), this PR adds a component label to the pods that are part of the daemonset. Without this label it is impossible to target the agent pods with any selector (the other parts of the chart have component labels).

Secondly, it defines an (optional) service resource targetting the pods in the daemonset and using a "local" internal traffic policy (This tells kube-proxy to only use node local endpoints for cluster internal traffic). This is useful because pods that have no way to be configured with the node ip can use the entry.

[dmitryax]

Hi, @mr-miles. Thanks for your contribution. The change itself LGTM.

The problem is that it changes immutable fields (daemonset selector labels), which breaks the helm upgrade, and users will have to reinstall their daemonsets.

We will need to migrate all the labels to the recommended naming conventions anyway, so I think if we can go through that only once and avoid doing it this time additionally.

What if we just add the component: otel-agent pod label only and keep the selector as is? The service selector should work well, right?

[mr-miles]

Thanks @jinja2 @dmitryax - have done that and checks looks more healthy. Didn't know about the daemonset selectors being immutable so learnt something too :-)

[dmitryax]

Please run make render to generate examples

[mr-miles]

Make render worked but the cri-o test is still complaining, not sure how to sort that though

[dmitryax]

Looks like a flaky test. I restarted it

[dmitryax]

Thank you, @mr-miles. LGTM. just one nit

[mr-miles]

success - all checks have passed!

[mr-miles]

Hi @jinja2 @dmitryax - wondering if it is possible to revive this PR? I think it just fell between the cracks.

I have just freshened it up by rebased it onto main and re-rendered the examples.

Alternatively please let me know if its no longer useful.

[jinja2]

I think we can still go ahead with this PR. But I am wondering if we should make settinghostNetwork for the daemonset optionally false (default will still be true) possible with the PR. I can see users running non-hostmetrics receivers leveraging this service option and wanting to not run in the hostnetwork. @dmitryax wdyt?

[jinja2]

@mr-miles can you run make render to regenerate examples?

[mr-miles]

hi @jinja2 - thanks for your help! Not sure why it closed the PR, but ...

I have rebased and "make render"-ed and corrected line endings and fixed up the github-actions robot's changes so should be good now

(The git robot is quite determined but I think I got the better of it)