From 4a3ae69e7e4002a7b06364bdc5984a44cd640872 Mon Sep 17 00:00:00 2001
From: frankvollenweider <frank.vollenweider@wolkenreich.ch>
Date: Thu, 24 Sep 2026 20:26:47 +0200
Subject: [PATCH] fix: delete a garbage collected Job's Pods with it

gc jobs deleted Jobs with empty DeleteOptions. For batch/v1 Jobs the API
server then orphans the Job's Pods, and says so in a warning on every
delete. The orphaned Pods have no ownerReferences, so nothing collects
them afterwards. With the default jx-gcjobs CronJob this leaves one
completed jx-boot Pod in jx-git-operator per boot Job, and they
accumulate until the nodes are replaced.

Delete with background propagation, as kubectl delete job does, so the
garbage collector removes the Pods together with the Job.

Fixes #1103

Signed-off-by: frankvollenweider <frank.vollenweider@wolkenreich.ch>
---
 pkg/cmd/gc/jobs/jobs.go      |  4 ++-
 pkg/cmd/gc/jobs/jobs_test.go | 57 ++++++++++++++++++++++++++++++++++++
 2 files changed, 60 insertions(+), 1 deletion(-)
 create mode 100644 pkg/cmd/gc/jobs/jobs_test.go

diff --git a/pkg/cmd/gc/jobs/jobs.go b/pkg/cmd/gc/jobs/jobs.go
index ae9f6e34..d9b44d96 100644
--- a/pkg/cmd/gc/jobs/jobs.go
+++ b/pkg/cmd/gc/jobs/jobs.go
@@ -93,7 +93,9 @@ func (o *Options) Run() error {
 		return err
 	}
 
-	deleteOptions := metav1.DeleteOptions{}
+	// without a propagation policy the API server keeps a deleted Job's Pods
+	propagation := metav1.DeletePropagationBackground
+	deleteOptions := metav1.DeleteOptions{PropagationPolicy: &propagation}
 	errors := []error{}
 
 	// we need to keep all jobs, don't waste time sorting or iterating
diff --git a/pkg/cmd/gc/jobs/jobs_test.go b/pkg/cmd/gc/jobs/jobs_test.go
new file mode 100644
index 00000000..af73a9aa
--- /dev/null
+++ b/pkg/cmd/gc/jobs/jobs_test.go
@@ -0,0 +1,57 @@
+package jobs
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	batchv1 "k8s.io/api/batch/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/client-go/kubernetes/fake"
+	k8stesting "k8s.io/client-go/testing"
+)
+
+func TestGCJobsDeletesPodsWithJob(t *testing.T) {
+	t.Parallel()
+
+	ns := "jx-git-operator"
+	oldStart := metav1.NewTime(time.Now().Add(-2 * time.Hour))
+	newStart := metav1.NewTime(time.Now().Add(-10 * time.Minute))
+
+	kubeClient := fake.NewClientset(
+		&batchv1.Job{
+			ObjectMeta: metav1.ObjectMeta{Name: "old", Namespace: ns},
+			Status:     batchv1.JobStatus{StartTime: &oldStart, Succeeded: 1},
+		},
+		&batchv1.Job{
+			ObjectMeta: metav1.ObjectMeta{Name: "new", Namespace: ns},
+			Status:     batchv1.JobStatus{StartTime: &newStart, Succeeded: 1},
+		},
+	)
+
+	o := &Options{
+		Namespace:  ns,
+		Age:        time.Hour,
+		Keep:       0,
+		KubeClient: kubeClient,
+	}
+	require.NoError(t, o.Run())
+
+	var deletes []k8stesting.DeleteActionImpl
+	for _, action := range kubeClient.Actions() {
+		if d, ok := action.(k8stesting.DeleteActionImpl); ok {
+			deletes = append(deletes, d)
+		}
+	}
+	require.Len(t, deletes, 1)
+	assert.Equal(t, "old", deletes[0].GetName())
+	require.NotNil(t, deletes[0].DeleteOptions.PropagationPolicy, "a Job deleted without a propagation policy leaves its Pods behind")
+	assert.Equal(t, metav1.DeletePropagationBackground, *deletes[0].DeleteOptions.PropagationPolicy)
+
+	remaining, err := kubeClient.BatchV1().Jobs(ns).List(context.TODO(), metav1.ListOptions{})
+	require.NoError(t, err)
+	require.Len(t, remaining.Items, 1)
+	assert.Equal(t, "new", remaining.Items[0].Name)
+}