Fix k8s service registration case where Vault fails to unlabel itself as a leader (#21642)

changelog/21642.txt

@@ -0,0 +1,3 @@
+```release-note:bug
+serviceregistration: Fix bug where multiple nodes in a secondary cluster could be labelled active after updating the cluster's primary
+```

vault/ha.go

@@ -710,6 +710,13 @@ func (c *Core) waitForLeadership(newLeaderCh chan func(), manualStepDownCh, stop
 				c.heldHALock = nil
 			}
 
+			// Advertise ourselves as a standby.
+			if c.serviceRegistration != nil {
+				if err := c.serviceRegistration.NotifyActiveStateChange(false); err != nil {
+					c.logger.Warn("failed to notify standby status", "error", err)
+				}
+			}
+
 			// If we are stopped return, otherwise unlock the statelock
 			if stopped {
 				return
@@ -1133,18 +1140,7 @@ func (c *Core) cleanLeaderPrefix(ctx context.Context, uuid string, leaderLostCh
 // clearLeader is used to clear our leadership entry
 func (c *Core) clearLeader(uuid string) error {
 	key := coreLeaderPrefix + uuid
-	err := c.barrier.Delete(context.Background(), key)
-
-	// Advertise ourselves as a standby
-	if c.serviceRegistration != nil {
-		if err := c.serviceRegistration.NotifyActiveStateChange(false); err != nil {
-			if c.logger.IsWarn() {
-				c.logger.Warn("failed to notify standby status", "error", err)
-			}
-		}
-	}
-
-	return err
+	return c.barrier.Delete(context.Background(), key)
 }
 
 func (c *Core) SetNeverBecomeActive(on bool) {