Stop auto-followers on shutdown (#40124)

When shutting down a node, auto-followers will keep trying to run. This
is happening even as transport services and other components are being
closed. In some cases, this can lead to a stack overflow as we rapidly
try to check the license state of the remote cluster, can not because
the transport service is shutdown, and then immeidately retry
again. This can happen faster than the shutdown, and we die with stack
overflow. This commit adds a stop command to auto-followers so that this
retry loop occurs at most once on shutdown.

Author: jasontedor

x-pack/plugin/ccr/src/main/java/org/elasticsearch/xpack/ccr/Ccr.java

@@ -163,8 +163,7 @@ public Collection<Object> createComponents(
 
         return Arrays.asList(
             ccrLicenseChecker,
-            new AutoFollowCoordinator(settings, client, clusterService, ccrLicenseChecker, threadPool::relativeTimeInMillis)
-        );
+            new AutoFollowCoordinator(settings, client, clusterService, ccrLicenseChecker, threadPool::relativeTimeInMillis));
     }
 
     @Override

x-pack/plugin/ccr/src/main/java/org/elasticsearch/xpack/ccr/action/AutoFollowCoordinator.java

@@ -25,6 +25,8 @@
 import org.elasticsearch.cluster.service.ClusterService;
 import org.elasticsearch.common.collect.CopyOnWriteHashMap;
 import org.elasticsearch.common.collect.Tuple;
+import org.elasticsearch.common.component.AbstractLifecycleComponent;
+import org.elasticsearch.common.component.Lifecycle;
 import org.elasticsearch.common.settings.Settings;
 import org.elasticsearch.common.unit.TimeValue;
 import org.elasticsearch.common.util.concurrent.AtomicArray;
@@ -65,7 +67,7 @@
  * A component that runs only on the elected master node and follows leader indices automatically
  * if they match with a auto follow pattern that is defined in {@link AutoFollowMetadata}.
  */
-public class AutoFollowCoordinator implements ClusterStateListener {
+public class AutoFollowCoordinator extends AbstractLifecycleComponent implements ClusterStateListener {
 
     private static final Logger LOGGER = LogManager.getLogger(AutoFollowCoordinator.class);
     private static final int MAX_AUTO_FOLLOW_ERRORS = 256;
@@ -90,7 +92,7 @@ public AutoFollowCoordinator(
             ClusterService clusterService,
             CcrLicenseChecker ccrLicenseChecker,
             LongSupplier relativeMillisTimeProvider) {
-
+        super(settings);
         this.client = client;
         this.clusterService = clusterService;
         this.ccrLicenseChecker = Objects.requireNonNull(ccrLicenseChecker, "ccrLicenseChecker");
@@ -113,6 +115,26 @@ protected boolean removeEldestEntry(final Map.Entry<String, ElasticsearchExcepti
         waitForMetadataTimeOut = CcrSettings.CCR_AUTO_FOLLOW_WAIT_FOR_METADATA_TIMEOUT.get(settings);
     }
 
+    @Override
+    protected void doStart() {
+
+    }
+
+    @Override
+    protected void doStop() {
+        LOGGER.trace("stopping all auto-followers");
+        /*
+         * Synchronization is not necessary here; the field is volatile and the map is a copy-on-write map, any new auto-followers will not
+         * start since we check started status of the coordinator before starting them.
+         */
+        autoFollowers.values().forEach(AutoFollower::stop);
+    }
+
+    @Override
+    protected void doClose() {
+
+    }
+
     public synchronized AutoFollowStats getStats() {
         final Map<String, AutoFollower> autoFollowers = this.autoFollowers;
         final TreeMap<String, AutoFollowedCluster> timesSinceLastAutoFollowPerRemoteCluster = new TreeMap<>();
@@ -241,8 +263,10 @@ public void clusterStateProcessed(String source, ClusterState oldState, ClusterS
 
             };
             newAutoFollowers.put(remoteCluster, autoFollower);
-            LOGGER.info("starting auto follower for remote cluster [{}]", remoteCluster);
-            autoFollower.start();
+            LOGGER.info("starting auto-follower for remote cluster [{}]", remoteCluster);
+            if (lifecycleState() == Lifecycle.State.STARTED) {
+                autoFollower.start();
+            }
         }
 
         List<String> removedRemoteClusters = new ArrayList<>();
@@ -252,13 +276,15 @@ public void clusterStateProcessed(String source, ClusterState oldState, ClusterS
             boolean exist = autoFollowMetadata.getPatterns().values().stream()
                 .anyMatch(pattern -> pattern.getRemoteCluster().equals(remoteCluster));
             if (exist == false) {
-                LOGGER.info("removing auto follower for remote cluster [{}]", remoteCluster);
+                LOGGER.info("removing auto-follower for remote cluster [{}]", remoteCluster);
                 autoFollower.removed = true;
                 removedRemoteClusters.add(remoteCluster);
             } else if (autoFollower.remoteClusterConnectionMissing) {
-                LOGGER.info("retrying auto follower [{}] after remote cluster connection was missing", remoteCluster);
+                LOGGER.info("retrying auto-follower for remote cluster [{}] after remote cluster connection was missing", remoteCluster);
                 autoFollower.remoteClusterConnectionMissing = false;
-                autoFollower.start();
+                if (lifecycleState() == Lifecycle.State.STARTED) {
+                    autoFollower.start();
+                }
             }
         }
         assert assertNoOtherActiveAutoFollower(newAutoFollowers);
@@ -308,6 +334,7 @@ abstract static class AutoFollower {
         volatile boolean removed = false;
         private volatile CountDown autoFollowPatternsCountDown;
         private volatile AtomicArray<AutoFollowResult> autoFollowResults;
+        private volatile boolean stop;
 
         AutoFollower(final String remoteCluster,
                      final Consumer<List<AutoFollowResult>> statsUpdater,
@@ -320,6 +347,10 @@ abstract static class AutoFollower {
         }
 
         void start() {
+            if (stop) {
+                LOGGER.trace("auto-follower is stopped for remote cluster [{}]", remoteCluster);
+                return;
+            }
             if (removed) {
                 // This check exists to avoid two AutoFollower instances a single remote cluster.
                 // (If an auto follow pattern is deleted and then added back quickly enough then
@@ -385,6 +416,11 @@ void start() {
             });
         }
 
+        void stop() {
+            LOGGER.trace("stopping auto-follower for remote cluster [{}]", remoteCluster);
+            stop = true;
+        }
+
         private void autoFollowIndices(final AutoFollowMetadata autoFollowMetadata,
                                        final ClusterState clusterState,
                                        final ClusterState remoteClusterState,