[ML] Stop the ML memory tracker before closing node (#39111)

The ML memory tracker does searches against ML results
and config indices.  These searches can be asynchronous,
and if they are running while the node is closing then
they can cause problems for other components.

This change adds a stop() method to the MlMemoryTracker
that waits for in-flight searches to complete.  Once
stop() has returned the MlMemoryTracker will not kick
off any new searches.

The MlLifeCycleService now calls MlMemoryTracker.stop()
before stopping stopping the node.

Fixes #37117

Author: David Roberts

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/MachineLearning.java

@@ -435,10 +435,10 @@ public Collection<Object> createComponents(Client client, ClusterService cluster
         DatafeedManager datafeedManager = new DatafeedManager(threadPool, client, clusterService, datafeedJobBuilder,
                 System::currentTimeMillis, auditor, autodetectProcessManager);
         this.datafeedManager.set(datafeedManager);
-        MlLifeCycleService mlLifeCycleService = new MlLifeCycleService(environment, clusterService, datafeedManager,
-                autodetectProcessManager);
         MlMemoryTracker memoryTracker = new MlMemoryTracker(settings, clusterService, threadPool, jobManager, jobResultsProvider);
         this.memoryTracker.set(memoryTracker);
+        MlLifeCycleService mlLifeCycleService = new MlLifeCycleService(environment, clusterService, datafeedManager,
+                autodetectProcessManager, memoryTracker);
 
         // This object's constructor attaches to the license state, so there's no need to retain another reference to it
         new InvalidLicenseEnforcer(getLicenseState(), threadPool, datafeedManager, autodetectProcessManager);

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/MlLifeCycleService.java

@@ -9,6 +9,7 @@
 import org.elasticsearch.common.component.LifecycleListener;
 import org.elasticsearch.env.Environment;
 import org.elasticsearch.xpack.ml.datafeed.DatafeedManager;
+import org.elasticsearch.xpack.ml.process.MlMemoryTracker;
 import org.elasticsearch.xpack.ml.process.NativeController;
 import org.elasticsearch.xpack.ml.process.NativeControllerHolder;
 import org.elasticsearch.xpack.ml.job.process.autodetect.AutodetectProcessManager;
@@ -20,16 +21,14 @@ public class MlLifeCycleService {
     private final Environment environment;
     private final DatafeedManager datafeedManager;
     private final AutodetectProcessManager autodetectProcessManager;
-
-    public MlLifeCycleService(Environment environment, ClusterService clusterService) {
-        this(environment, clusterService, null, null);
-    }
+    private final MlMemoryTracker memoryTracker;
 
     public MlLifeCycleService(Environment environment, ClusterService clusterService, DatafeedManager datafeedManager,
-                              AutodetectProcessManager autodetectProcessManager) {
+                              AutodetectProcessManager autodetectProcessManager, MlMemoryTracker memoryTracker) {
         this.environment = environment;
         this.datafeedManager = datafeedManager;
         this.autodetectProcessManager = autodetectProcessManager;
+        this.memoryTracker = memoryTracker;
         clusterService.addLifecycleListener(new LifecycleListener() {
             @Override
             public void beforeStop() {
@@ -59,5 +58,8 @@ public synchronized void stop() {
         } catch (IOException e) {
             // We're stopping anyway, so don't let this complicate the shutdown sequence
         }
+        if (memoryTracker != null) {
+            memoryTracker.stop();
+        }
     }
 }

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/process/MlMemoryTracker.java

@@ -32,6 +32,7 @@
 import java.util.Iterator;
 import java.util.List;
 import java.util.concurrent.ConcurrentHashMap;
+import java.util.concurrent.Phaser;
 import java.util.stream.Collectors;
 
 /**
@@ -55,6 +56,7 @@ public class MlMemoryTracker implements LocalNodeMasterListener {
     private final ClusterService clusterService;
     private final JobManager jobManager;
     private final JobResultsProvider jobResultsProvider;
+    private final Phaser stopPhaser;
     private volatile boolean isMaster;
     private volatile Instant lastUpdateTime;
     private volatile Duration reassignmentRecheckInterval;
@@ -65,6 +67,7 @@ public MlMemoryTracker(Settings settings, ClusterService clusterService, ThreadP
         this.clusterService = clusterService;
         this.jobManager = jobManager;
         this.jobResultsProvider = jobResultsProvider;
+        this.stopPhaser = new Phaser(1);
         setReassignmentRecheckInterval(PersistentTasksClusterService.CLUSTER_TASKS_ALLOCATION_RECHECK_INTERVAL_SETTING.get(settings));
         clusterService.addLocalNodeMasterListener(this);
         clusterService.getClusterSettings().addSettingsUpdateConsumer(
@@ -89,6 +92,23 @@ public void offMaster() {
         lastUpdateTime = null;
     }
 
+    /**
+     * Wait for all outstanding searches to complete.
+     * After returning, no new searches can be started.
+     */
+    public void stop() {
+        logger.trace("ML memory tracker stop called");
+        // We never terminate the phaser
+        assert stopPhaser.isTerminated() == false;
+        // If there are no registered parties or no unarrived parties then there is a flaw
+        // in the register/arrive/unregister logic in another method that uses the phaser
+        assert stopPhaser.getRegisteredParties() > 0;
+        assert stopPhaser.getUnarrivedParties() > 0;
+        stopPhaser.arriveAndAwaitAdvance();
+        assert stopPhaser.getPhase() > 0;
+        logger.debug("ML memory tracker stopped");
+    }
+
     @Override
     public String executorName() {
         return MachineLearning.UTILITY_THREAD_POOL_NAME;
@@ -146,13 +166,13 @@ public boolean asyncRefresh() {
             try {
                 ActionListener<Void> listener = ActionListener.wrap(
                     aVoid -> logger.trace("Job memory requirement refresh request completed successfully"),
-                    e -> logger.error("Failed to refresh job memory requirements", e)
+                    e -> logger.warn("Failed to refresh job memory requirements", e)
                 );
                 threadPool.executor(executorName()).execute(
                     () -> refresh(clusterService.state().getMetaData().custom(PersistentTasksCustomMetaData.TYPE), listener));
                 return true;
             } catch (EsRejectedExecutionException e) {
-                logger.debug("Couldn't schedule ML memory update - node might be shutting down", e);
+                logger.warn("Couldn't schedule ML memory update - node might be shutting down", e);
             }
         }
 
@@ -246,25 +266,43 @@ public void refreshJobMemory(String jobId, ActionListener<Long> listener) {
             return;
         }
 
+        // The phaser prevents searches being started after the memory tracker's stop() method has returned
+        if (stopPhaser.register() != 0) {
+            // Phases above 0 mean we've been stopped, so don't do any operations that involve external interaction
+            stopPhaser.arriveAndDeregister();
+            listener.onFailure(new EsRejectedExecutionException("Couldn't run ML memory update - node is shutting down"));
+            return;
+        }
+        ActionListener<Long> phaserListener = ActionListener.wrap(
+            r -> {
+                stopPhaser.arriveAndDeregister();
+                listener.onResponse(r);
+            },
+            e -> {
+                stopPhaser.arriveAndDeregister();
+                listener.onFailure(e);
+            }
+        );
+
         try {
             jobResultsProvider.getEstablishedMemoryUsage(jobId, null, null,
                 establishedModelMemoryBytes -> {
                     if (establishedModelMemoryBytes <= 0L) {
-                        setJobMemoryToLimit(jobId, listener);
+                        setJobMemoryToLimit(jobId, phaserListener);
                     } else {
                         Long memoryRequirementBytes = establishedModelMemoryBytes + Job.PROCESS_MEMORY_OVERHEAD.getBytes();
                         memoryRequirementByJob.put(jobId, memoryRequirementBytes);
-                        listener.onResponse(memoryRequirementBytes);
+                        phaserListener.onResponse(memoryRequirementBytes);
                     }
                 },
                 e -> {
                     logger.error("[" + jobId + "] failed to calculate job established model memory requirement", e);
-                    setJobMemoryToLimit(jobId, listener);
+                    setJobMemoryToLimit(jobId, phaserListener);
                 }
             );
         } catch (Exception e) {
             logger.error("[" + jobId + "] failed to calculate job established model memory requirement", e);
-            setJobMemoryToLimit(jobId, listener);
+            setJobMemoryToLimit(jobId, phaserListener);
         }
     }
 

x-pack/plugin/ml/src/test/java/org/elasticsearch/xpack/ml/process/MlMemoryTrackerTests.java

@@ -10,6 +10,7 @@
 import org.elasticsearch.common.settings.ClusterSettings;
 import org.elasticsearch.common.settings.Settings;
 import org.elasticsearch.common.unit.ByteSizeUnit;
+import org.elasticsearch.common.util.concurrent.EsRejectedExecutionException;
 import org.elasticsearch.persistent.PersistentTasksClusterService;
 import org.elasticsearch.persistent.PersistentTasksCustomMetaData;
 import org.elasticsearch.test.ESTestCase;
@@ -29,6 +30,7 @@
 import java.util.concurrent.atomic.AtomicReference;
 import java.util.function.Consumer;
 
+import static org.hamcrest.CoreMatchers.instanceOf;
 import static org.mockito.Matchers.any;
 import static org.mockito.Matchers.eq;
 import static org.mockito.Mockito.anyString;
@@ -157,6 +159,19 @@ public void testRefreshOne() {
         assertNull(memoryTracker.getJobMemoryRequirement(jobId));
     }
 
+    public void testStop() {
+
+        memoryTracker.onMaster();
+        memoryTracker.stop();
+
+        AtomicReference<Exception> exception = new AtomicReference<>();
+        memoryTracker.refreshJobMemory("job", ActionListener.wrap(ESTestCase::assertNull, exception::set));
+
+        assertNotNull(exception.get());
+        assertThat(exception.get(), instanceOf(EsRejectedExecutionException.class));
+        assertEquals("Couldn't run ML memory update - node is shutting down", exception.get().getMessage());
+    }
+
     private PersistentTasksCustomMetaData.PersistentTask<OpenJobAction.JobParams> makeTestTask(String jobId) {
         return new PersistentTasksCustomMetaData.PersistentTask<>("job-" + jobId, MlTasks.JOB_TASK_NAME, new OpenJobAction.JobParams(jobId),
             0, PersistentTasksCustomMetaData.INITIAL_ASSIGNMENT);