job: Log an error if a job fails to stop

job/job.go

@@ -95,7 +95,7 @@ func (c *registry) Stop(ctx cell.HookContext) error {
 		return nil
 	}
 	c.started = false
-	return c.runtimeLifecycle.stop(ctx)
+	return c.runtimeLifecycle.stop(ctx, c.logger)
 }
 
 // PreStopHookMarker tells [cell.DefaultLifecycle] that this

job/lifecycle.go

@@ -5,6 +5,7 @@ package job
 
 import (
 	"context"
+	"log/slog"
 	"sync"
 
 	"github.com/cilium/hive/cell"
@@ -56,7 +57,7 @@ func (r *jobLifecycle) remove(qj *queuedJob) {
 	qj.next = nil
 }
 
-func (r *jobLifecycle) stop(ctx cell.HookContext) error {
+func (r *jobLifecycle) stop(ctx cell.HookContext, log *slog.Logger) error {
 	// Collect jobs to stop and unlink them. We must stop them without holding the
 	// lock as [queuedJob.Stop] will try to call [jobLifecycle.remove].
 	var jobsToStop []*queuedJob
@@ -73,6 +74,9 @@ func (r *jobLifecycle) stop(ctx cell.HookContext) error {
 	for _, job := range jobsToStop {
 		job.Stop(ctx)
 		if ctx.Err() != nil {
+			log.Error("Stop cancelled while waiting for job to stop",
+				"job",
+				job.job.info())
 			break
 		}
 	}

job/lifecycle_test.go

@@ -14,6 +14,7 @@ import (
 	"github.com/stretchr/testify/require"
 
 	"github.com/cilium/hive/cell"
+	"github.com/cilium/hive/hivetest"
 )
 
 func TestJobLifecycleInsertAndRemove(t *testing.T) {
@@ -30,15 +31,15 @@ func TestJobLifecycleInsertAndRemove(t *testing.T) {
 
 	requireRuntimeLifecycleList(t, r, third, second, first)
 
-	removeRuntimeLifecycleJob(r, second)
+	r.runtimeLifecycle.remove(second)
 	requireRuntimeLifecycleList(t, r, third, first)
 	requireRuntimeLifecycleJobUnlinked(t, r, second)
 
-	removeRuntimeLifecycleJob(r, third)
+	r.runtimeLifecycle.remove(third)
 	requireRuntimeLifecycleList(t, r, first)
 	requireRuntimeLifecycleJobUnlinked(t, r, third)
 
-	removeRuntimeLifecycleJob(r, first)
+	r.runtimeLifecycle.remove(first)
 	requireRuntimeLifecycleList(t, r)
 	requireRuntimeLifecycleJobUnlinked(t, r, first)
 }
@@ -86,7 +87,7 @@ func TestJobLifecycleStopStopsJobsInReverseStartOrder(t *testing.T) {
 	insertRuntimeLifecycleJobs(r, first, second, third)
 	waitForLifecycleJobs(t, first, second, third)
 
-	require.NoError(t, stopRuntimeLifecycle(r, context.Background()))
+	require.NoError(t, r.runtimeLifecycle.stop(context.Background(), hivetest.Logger(t)))
 
 	requireRuntimeLifecycleList(t, r)
 
@@ -102,7 +103,7 @@ func TestJobLifecycleStopReturnsContextError(t *testing.T) {
 	cancel()
 
 	var lifecycle jobLifecycle
-	assert.ErrorIs(t, lifecycle.stop(ctx), context.Canceled)
+	assert.ErrorIs(t, lifecycle.stop(ctx, hivetest.Logger(t)), context.Canceled)
 }
 
 type blockingLifecycleJob struct {
@@ -180,14 +181,6 @@ func insertRuntimeLifecycleJobs(r *registry, jobs ...*queuedJob) {
 	}
 }
 
-func removeRuntimeLifecycleJob(r *registry, qj *queuedJob) {
-	r.runtimeLifecycle.remove(qj)
-}
-
-func stopRuntimeLifecycle(r *registry, ctx context.Context) error {
-	return r.runtimeLifecycle.stop(ctx)
-}
-
 func requireRuntimeLifecycleList(t *testing.T, r *registry, want ...*queuedJob) {
 	t.Helper()