Profiling server not starting in the adapter framework

[theakshaypant]

Describe the bug

Components built on the eventing receive adapter framework (pkg/adapter/v2) never start the pprof HTTP server, making runtime profiling a no-op even when runtime-profiling: "enabled" is set in the observability ConfigMap.

MainWithInformers creates a ProfilingServer and correctly passes it to SetupObservabilityOrDie (which sets the enabled flag and logs "Profiling enabled: true"), but never calls ListenAndServe. Port 8008 is never bound.

Expected behavior

Setting runtime-profiling: "enabled" in the observability ConfigMap should cause the pprof endpoints (/debug/pprof/) to become accessible on port 8008 for adapter-based components, matching the behaviour of sharedmain-based components.

To Reproduce

1. Deployed receive adapter server (This is one I am using)
2. Set runtime-profiling: "enabled" in the config-observability ConfigMap
3. The server logs

{"level":"info","ts":"2026-04-01T12:58:53.568Z","logger":"pipelinesascode.pprof","caller":"k8s/pprof.go:51","msg":"Profiling enabled: true",xxx}

4. Attempt to curl http://<adapter-pod>:8008/debug/pprof/ — connection refused

Knative release version: tag v0.48.1

[theakshaypant]

My understanding is that MainWithInformers in pkg/adapter/v2/main.go has this flow:

• Creates a ProfilingServer via k8sruntime.NewProfilingServer
• Passes it to SetupObservabilityOrDie, which reads the observability and calls pprof.UpdateFromConfig — this sets an atomic.Bool enabled and logs "Profiling enabled: true"
• Starts the adapter, health probes, and controller goroutines
• Never calls pprof.ListenAndServe()

Would it be the right solution to add a goroutine in this method after the health probe block that starts the profiling server based on the configuration

--- a/pkg/adapter/v2/main.go
+++ b/pkg/adapter/v2/main.go
@@ -279,6 +279,17 @@ func MainWithInformers(...) {
        if !HealthProbesDisabled(ctx) {
                wg.Add(1)
                go func() {
                        defer wg.Done()
                        injection.ServeHealthProbes(ctx, injection.HealthCheckDefaultPort)
                }()
        }

+       if cfg, err := env.GetObservabilityConfig(); err == nil && cfg.Runtime.ProfilingEnabled() {
+               wg.Add(1)
+               go func() {
+                       defer wg.Done()
+                       if err := pprof.ListenAndServe(); err != nil {
+                               logger.Warnw("Profiling server shut down", zap.Error(err))
+                       }
+               }()
+       }
+
        // Finally start the adapter (blocking)
        if err := adapter.Start(ctx); err != nil {

Happy to raise a PR if this is the right approach or know if there is something that I have done/described incorrectly or if more info is needed.

[creydr]

Hello @theakshaypant,
thanks for bringing this up and checking on a solution for this. I am happy to check on a PR with something like #9007 (comment)

[theakshaypant]

Hi @creydr , thanks for the quick response! I have created and linked a PR, did cover some testing scenarios that I have listed on the PR.