From 0332745eef041f098953bed4caba16da54db469f Mon Sep 17 00:00:00 2001
From: Andy Zhang <87735571+Andyz26@users.noreply.github.com>
Date: Tue, 9 Apr 2024 14:56:56 -0700
Subject: [PATCH] fix rcActor supervisor (#652)

---
 .../resourcecluster/ResourceClusterActor.java | 21 +++++++++++++++++++
 1 file changed, 21 insertions(+)

diff --git a/mantis-control-plane/mantis-control-plane-server/src/main/java/io/mantisrx/master/resourcecluster/ResourceClusterActor.java b/mantis-control-plane/mantis-control-plane-server/src/main/java/io/mantisrx/master/resourcecluster/ResourceClusterActor.java
index 266aa2a0c..7a9a699f9 100644
--- a/mantis-control-plane/mantis-control-plane-server/src/main/java/io/mantisrx/master/resourcecluster/ResourceClusterActor.java
+++ b/mantis-control-plane/mantis-control-plane-server/src/main/java/io/mantisrx/master/resourcecluster/ResourceClusterActor.java
@@ -20,8 +20,11 @@
 
 import akka.actor.AbstractActorWithTimers;
 import akka.actor.ActorRef;
+import akka.actor.OneForOneStrategy;
 import akka.actor.Props;
 import akka.actor.Status;
+import akka.actor.SupervisorStrategy;
+import akka.japi.pf.DeciderBuilder;
 import akka.japi.pf.ReceiveBuilder;
 import com.netflix.spectator.api.Tag;
 import com.netflix.spectator.api.TagList;
@@ -95,6 +98,24 @@
 @ToString(of = {"clusterID"})
 @Slf4j
 class ResourceClusterActor extends AbstractActorWithTimers {
+    /**
+     * For ResourceClusterActor instances, we need to ensure they are always running after encountering error so that
+     * TaskExecutors can still remain connected. If there is a fatal error that needs to be escalated to terminate the
+     * whole system/leader you can define a fatal exception type and override its behavior to
+     * SupervisorStrategy.escalate() instead.
+     */
+    private static SupervisorStrategy resourceClusterActorStrategy =
+        new OneForOneStrategy(
+            3,
+            Duration.ofSeconds(60),
+            DeciderBuilder
+                .match(Exception.class, e -> SupervisorStrategy.restart())
+                .build());
+
+    @Override
+    public SupervisorStrategy supervisorStrategy() {
+        return resourceClusterActorStrategy;
+    }
 
     private final Duration heartbeatTimeout;
     private final Duration assignmentTimeout;