pkg/leader: require the podname to be set explicitly

We can't trust that the hostname is always the pod name.

Author: squeed

pkg/leader/doc.go

@@ -41,6 +41,14 @@ The lock record in this case is a ConfigMap whose OwnerReference is set to the
 Pod that is the leader. When the leader is destroyed, the ConfigMap gets
 garbage-collected, enabling a different candidate Pod to become the leader.
 
-Leader for Life requires that all candidate Pods be in the same Namespace.
+Leader for Life requires that all candidate Pods be in the same Namespace. It
+uses the downwards API to determine the pod name, as hostname is not reliable.
+You should run it configured with:
+
+env:
+  - name: POD_NAME
+    valueFrom:
+      fieldRef:
+        fieldPath: metadata.name
 */
 package leader

pkg/leader/leader.go

@@ -17,6 +17,7 @@ package leader
 import (
 	"context"
 	"errors"
+	"fmt"
 	"io/ioutil"
 	"os"
 	"strings"
@@ -39,6 +40,8 @@ var errNoNS = errors.New("namespace not found for current environment")
 // attempts to become the leader.
 const maxBackoffInterval = time.Second * 16
 
+const PodNameEnv = "POD_NAME"
+
 // Become ensures that the current pod is the leader within its namespace. If
 // run outside a cluster, it will skip leader election and return nil. It
 // continuously tries to create a ConfigMap with the provided name and the
@@ -156,12 +159,14 @@ func myNS() (string, error) {
 
 // myOwnerRef returns an OwnerReference that corresponds to the pod in which
 // this code is currently running.
+// It expects the environment variable POD_NAME to be set by the downwards API
 func myOwnerRef(ctx context.Context, client crclient.Client, ns string) (*metav1.OwnerReference, error) {
-	hostname, err := os.Hostname()
-	if err != nil {
-		return nil, err
+	podName := os.Getenv(PodNameEnv)
+	if podName == "" {
+		return nil, fmt.Errorf("required env %s not set, please configure downward API", PodNameEnv)
 	}
-	logrus.Debugf("found hostname: %s", hostname)
+
+	logrus.Debugf("found podname: %s", podName)
 
 	myPod := &corev1.Pod{
 		TypeMeta: metav1.TypeMeta{
@@ -170,8 +175,8 @@ func myOwnerRef(ctx context.Context, client crclient.Client, ns string) (*metav1
 		},
 	}
 
-	key := crclient.ObjectKey{Namespace: ns, Name: hostname}
-	err = client.Get(ctx, key, myPod)
+	key := crclient.ObjectKey{Namespace: ns, Name: podName}
+	err := client.Get(ctx, key, myPod)
 	if err != nil {
 		logrus.Errorf("failed to get pod: %v", err)
 		return nil, err

pkg/scaffold/operator.go

@@ -62,6 +62,10 @@ spec:
               valueFrom:
                 fieldRef:
                   fieldPath: metadata.namespace
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
             - name: OPERATOR_NAME
               value: "{{.ProjectName}}"
 `

pkg/scaffold/operator_test.go

@@ -61,6 +61,10 @@ spec:
               valueFrom:
                 fieldRef:
                   fieldPath: metadata.namespace
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
             - name: OPERATOR_NAME
               value: "app-operator"
 `