Migrate event recorder to fluxcd/pkg/runtime/events.Recorder

[adri1197]

Migrate the controller's event recorder to fluxcd/pkg/runtime/events.Recorder and the Kubernetes event API from v1beta1 to v1, so events carry structured metadata (source object reference and action type).

• Replace kuberecorder.EventRecorder with events.Recorder, constructed via events.NewRecorder(...,events.WithManager(mgr)).
• Introduce a typed Action in api/v1 describing the observable stages of the reconcile loop, and emit one on every event call: Reconcile, CheckDependencies,ResolveSource, etc.
• Thread the source reference through reconcile, apply, prune, checkHealth, and finalize so each event records the originating source.
• Bump fluxcd/pkg/runtime to v0.115.0 and fluxcd/pkg/apis/event to v0.30.0.
• Update the controller test suite to the events.Recorder and events.k8s.io/v1 event API.

Part of - fluxcd/flux2#5761

[matheuscscp]

Hey @adri1197 please rebase all your PRs, we have upgraded to k8s 1.37 across all controllers and bumped most fluxcd/pkg packages. The only remaining dep for you to bump across your PRs is fluxcd/pkg/runtime. Please start with kustomize-controller as there's another PR depending on the runtime bump 🙏

[matheuscscp]

LGTM! 🚀

Thanks @adri1197!

[stefanprodan]

Make the success event message stable across reconciliations of the same revision (drop the duration from the message; log it instead) and stamp each event with a unique token annotation, so Kubernetes can deduplicate

This is not true, deduplication will never happen now because ResourceVersion will always differ (due to status patching) and even if we mutate and set that to nil, the new client-go broadcaster hardcodes 6m for aggregation, basically this PR reverts #1743

[matheuscscp]

We will do what's described here now: #1742 (comment)

Please rebase after that! 🙏

[matheuscscp]

@adri1197 All done, please rebase! 🙏

[matheuscscp]

LGTM! 🚀

Thanks @adri1197!

[stefanprodan]

Found some major blockers:

• Messages are rejected over 1024 bytes which makes all our events to fail https://github.com/kubernetes/kubernetes/blob/v1.37.0/pkg/apis/core/validation/events.go#L37 and https://github.com/kubernetes/kubernetes/blob/f54c212e3a2f75d674b717a9b29052b20b60aefc/pkg/apis/core/validation/events.go#L183

• The Series grouping does not contain the message so in the 6m window all out events for a revision will be dropped, you only see the first message https://github.com/kubernetes/client-go/blob/v0.37.0/tools/events/event_broadcaster.go#L297-L310

[stefanprodan]

Given the 1024 limit of the event message, we can not ship this as is, all of Flux observability is built on the assumption that the events cary the desired state changes and errors. With 1024 limit, KC can no longer issue events with the list of applied changes, dry-run errors will no longer be surfaced to users, HC can no longer report helm upgrade errors, etc.

Found major blockers

[matheuscscp]

@adri1197 Please regenerate the api docs, they are having a diff in CI 🙏

[adri1197]

@adri1197 Please regenerate the api docs, they are having a diff in CI 🙏

Done! 😃

[matheuscscp]

Last thing! 🙏

[matheuscscp]

LGTM! 🚀

Thanks @adri1197!

[matheuscscp]

Hey @adri1197 Stefan just found an issue, we are defining lots of Action constants that arent used anywhere, we cant do that, lets remove them pls 🙏 Also, if you 're using an LLM please disclose it as described in our AI guidelines 🙏

The Action constants live in the new api/v1/action.go added by PR 1621 (the event-recorder migration). Comparing the 14 constants defined there against every kustomizev1.Action* reference in the PR, five are defined but never used:
Constant	Value	Status
ActionFetchSource	"FetchSource"	unused
ActionGenerate	"Generate"	unused
ActionBuild	"Build"	unused
ActionDetectDrift	"DetectDrift"	unused
ActionImpersonate	"Impersonate"	unused
The nine that are actually wired into event calls:
- ActionReconcile (deferred success event + reconcile failure)
- ActionCheckDependencies
- ActionResolveSource
- ActionDecrypt
- ActionApply
- ActionPrune
- ActionHealthCheck
- ActionFinalize
- ActionWaitForTermination

[adri1197]

Hey @adri1197 Stefan just found an issue, we are defining lots of Action constants that arent used anywhere, we cant do that, lets remove them pls 🙏 Also, if you 're using an LLM please disclose it as described in our AI guidelines 🙏

The Action constants live in the new api/v1/action.go added by PR 1621 (the event-recorder migration). Comparing the 14 constants defined there against every kustomizev1.Action* reference in the PR, five are defined but never used:
Constant	Value	Status
ActionFetchSource	"FetchSource"	unused
ActionGenerate	"Generate"	unused
ActionBuild	"Build"	unused
ActionDetectDrift	"DetectDrift"	unused
ActionImpersonate	"Impersonate"	unused
The nine that are actually wired into event calls:
- ActionReconcile (deferred success event + reconcile failure)
- ActionCheckDependencies
- ActionResolveSource
- ActionDecrypt
- ActionApply
- ActionPrune
- ActionHealthCheck
- ActionFinalize
- ActionWaitForTermination

Sorry, my bad.. Just used it to summarize the actions within the existing events. Working on it. Will add the trailing message about AI in the commit msg, as well.

[matheuscscp]

Thanks @adri1197, next please do source-controller and image-automation-controller as the pkg/runtime upgrade is blocking these PRs:

• Add githubAppClientID as an alternative to githubAppID source-controller#2168
• Add githubAppClientID as an alternative to githubAppID image-automation-controller#1097

[stefanprodan]

LGTM

Thanks @adri1197