[7.17] Handling exceptions on watcher reload (#105442) (#106209)

Author: sakurai-youhei

docs/changelog/105442.yaml

@@ -0,0 +1,6 @@
+pr: 105442
+summary: Handling exceptions on watcher reload
+area: Watcher
+type: bug
+issues:
+ - 69842

x-pack/plugin/watcher/src/main/java/org/elasticsearch/xpack/watcher/WatcherLifeCycleService.java

@@ -166,7 +166,9 @@ public void clusterChanged(ClusterChangedEvent event) {
             if (watcherService.validate(event.state())) {
                 previousShardRoutings.set(localAffectedShardRoutings);
                 if (state.get() == WatcherState.STARTED) {
-                    watcherService.reload(event.state(), "new local watcher shard allocation ids");
+                    watcherService.reload(event.state(), "new local watcher shard allocation ids", (exception) -> {
+                        clearAllocationIds(); // will cause reload again
+                    });
                 } else if (isStoppedOrStopping) {
                     this.state.set(WatcherState.STARTING);
                     watcherService.start(event.state(), () -> this.state.set(WatcherState.STARTED), (exception) -> {

x-pack/plugin/watcher/src/main/java/org/elasticsearch/xpack/watcher/WatcherService.java

@@ -202,7 +202,7 @@ void stopExecutor() {
      * Reload the watcher service, does not switch the state from stopped to started, just keep going
      * @param state cluster state, which is needed to find out about local shards
      */
-    void reload(ClusterState state, String reason) {
+    void reload(ClusterState state, String reason, Consumer<Exception> exceptionConsumer) {
         boolean hasValidWatcherTemplates = WatcherIndexTemplateRegistry.validate(state);
         if (hasValidWatcherTemplates == false) {
             logger.warn("missing watcher index templates");
@@ -222,7 +222,10 @@ void reload(ClusterState state, String reason) {
         int cancelledTaskCount = executionService.clearExecutionsAndQueue(() -> {});
         logger.info("reloading watcher, reason [{}], cancelled [{}] queued tasks", reason, cancelledTaskCount);
 
-        executor.execute(wrapWatcherService(() -> reloadInner(state, reason, false), e -> logger.error("error reloading watcher", e)));
+        executor.execute(wrapWatcherService(() -> reloadInner(state, reason, false), e -> {
+            logger.error("error reloading watcher", e);
+            exceptionConsumer.accept(e);
+        }));
     }
 
     /**

x-pack/plugin/watcher/src/test/java/org/elasticsearch/xpack/watcher/WatcherLifeCycleServiceTests.java

@@ -252,6 +252,91 @@ public void testExceptionOnStart() {
         assertThat(lifeCycleService.getState().get(), equalTo(WatcherState.STARTED));
     }
 
+    public void testReloadWithIdenticalRoutingTable() {
+        /*
+         * This tests that the identical routing table causes reload only once.
+         */
+        startWatcher();
+
+        ClusterChangedEvent[] events = masterChangeScenario();
+        assertThat(events[1].previousState(), equalTo(events[0].state()));
+        assertFalse(events[1].routingTableChanged());
+
+        for (ClusterChangedEvent event : events) {
+            when(watcherService.validate(event.state())).thenReturn(true);
+            lifeCycleService.clusterChanged(event);
+        }
+        // reload should occur on the first event
+        verify(watcherService).reload(eq(events[0].state()), anyString(), any());
+        // but it shouldn't on the second event unless routing table changes
+        verify(watcherService, never()).reload(eq(events[1].state()), anyString(), any());
+    }
+
+    public void testReloadWithIdenticalRoutingTableAfterException() {
+        /*
+         * This tests that even the identical routing table causes reload again if some exception (for example a timeout while loading
+         * watches) interrupted the previous one.
+         */
+        startWatcher();
+
+        ClusterChangedEvent[] events = masterChangeScenario();
+        assertThat(events[1].previousState(), equalTo(events[0].state()));
+        assertFalse(events[1].routingTableChanged());
+
+        // simulate exception on the first event
+        doAnswer(invocation -> {
+            Consumer<Exception> exceptionConsumer = invocation.getArgument(2);
+            exceptionConsumer.accept(new ElasticsearchTimeoutException(new TimeoutException("Artificial timeout")));
+            return null;
+        }).when(watcherService).reload(eq(events[0].state()), anyString(), any());
+
+        for (ClusterChangedEvent event : events) {
+            when(watcherService.validate(event.state())).thenReturn(true);
+            lifeCycleService.clusterChanged(event);
+        }
+        // reload should occur on the first event but it fails
+        verify(watcherService).reload(eq(events[0].state()), anyString(), any());
+        // reload should occur again on the second event because the previous one failed
+        verify(watcherService).reload(eq(events[1].state()), anyString(), any());
+    }
+
+    private ClusterChangedEvent[] masterChangeScenario() {
+        DiscoveryNodes nodes = new DiscoveryNodes.Builder().localNodeId("node_1").add(newNode("node_1")).add(newNode("node_2")).build();
+
+        Index index = new Index(Watch.INDEX, "uuid");
+        IndexRoutingTable.Builder indexRoutingTableBuilder = IndexRoutingTable.builder(index);
+        indexRoutingTableBuilder.addShard(
+            TestShardRouting.newShardRouting(new ShardId(index, 0), "node_1", true, ShardRoutingState.STARTED)
+        );
+        RoutingTable routingTable = RoutingTable.builder().add(indexRoutingTableBuilder.build()).build();
+
+        IndexMetadata.Builder indexMetadataBuilder = IndexMetadata.builder(Watch.INDEX)
+            .settings(settings(Version.CURRENT).put(IndexMetadata.INDEX_FORMAT_SETTING.getKey(), 6)) // the internal index format,
+            // required
+            .numberOfShards(1)
+            .numberOfReplicas(0);
+        Metadata metadata = Metadata.builder()
+            .put(IndexTemplateMetadata.builder(HISTORY_TEMPLATE_NAME).patterns(randomIndexPatterns()))
+            .put(indexMetadataBuilder)
+            .build();
+
+        ClusterState emptyState = ClusterState.builder(new ClusterName("my-cluster")).nodes(nodes).metadata(metadata).build();
+        ClusterState stateWithMasterNode1 = ClusterState.builder(new ClusterName("my-cluster"))
+            .nodes(DiscoveryNodes.builder(nodes).masterNodeId("node_1"))
+            .metadata(metadata)
+            .routingTable(routingTable)
+            .build();
+        ClusterState stateWithMasterNode2 = ClusterState.builder(new ClusterName("my-cluster"))
+            .nodes(DiscoveryNodes.builder(nodes).masterNodeId("node_2"))
+            .metadata(metadata)
+            .routingTable(routingTable)
+            .build();
+
+        return new ClusterChangedEvent[] {
+            new ClusterChangedEvent("any", stateWithMasterNode1, emptyState),
+            new ClusterChangedEvent("any", stateWithMasterNode2, stateWithMasterNode1) };
+    }
+
     public void testNoLocalShards() {
         Index watchIndex = new Index(Watch.INDEX, "foo");
         ShardId shardId = new ShardId(watchIndex, 0);
@@ -301,7 +386,7 @@ public void testNoLocalShards() {
         when(watcherService.validate(eq(clusterStateWithLocalShards))).thenReturn(true);
         when(watcherService.validate(eq(clusterStateWithoutLocalShards))).thenReturn(false);
         lifeCycleService.clusterChanged(new ClusterChangedEvent("any", clusterStateWithLocalShards, clusterStateWithoutLocalShards));
-        verify(watcherService, times(1)).reload(eq(clusterStateWithLocalShards), eq("new local watcher shard allocation ids"));
+        verify(watcherService, times(1)).reload(eq(clusterStateWithLocalShards), eq("new local watcher shard allocation ids"), any());
         verify(watcherService, times(1)).validate(eq(clusterStateWithLocalShards));
         verifyNoMoreInteractions(watcherService);
 
@@ -386,12 +471,12 @@ public void testReplicaWasAddedOrRemoved() {
 
         when(watcherService.validate(eq(firstEvent.state()))).thenReturn(true);
         lifeCycleService.clusterChanged(firstEvent);
-        verify(watcherService).reload(eq(firstEvent.state()), anyString());
+        verify(watcherService).reload(eq(firstEvent.state()), anyString(), any());
 
         reset(watcherService);
         when(watcherService.validate(eq(secondEvent.state()))).thenReturn(true);
         lifeCycleService.clusterChanged(secondEvent);
-        verify(watcherService).reload(eq(secondEvent.state()), anyString());
+        verify(watcherService).reload(eq(secondEvent.state()), anyString(), any());
     }
 
     // make sure that cluster state changes can be processed on nodes that do not hold data
@@ -457,7 +542,7 @@ public void testNonDataNode() {
 
         lifeCycleService.clusterChanged(new ClusterChangedEvent("any", currentState, previousState));
         verify(watcherService, times(0)).pauseExecution(any());
-        verify(watcherService, times(0)).reload(any(), any());
+        verify(watcherService, times(0)).reload(any(), any(), any());
     }
 
     public void testThatMissingWatcherIndexMetadataOnlyResetsOnce() {
@@ -490,7 +575,7 @@ public void testThatMissingWatcherIndexMetadataOnlyResetsOnce() {
 
         // first add the shard allocation ids, by going from empty cs to CS with watcher index
         lifeCycleService.clusterChanged(new ClusterChangedEvent("any", clusterStateWithWatcherIndex, clusterStateWithoutWatcherIndex));
-        verify(watcherService).reload(eq(clusterStateWithWatcherIndex), anyString());
+        verify(watcherService).reload(eq(clusterStateWithWatcherIndex), anyString(), any());
 
         // now remove watches index, and ensure that pausing is only called once, no matter how often called (i.e. each CS update)
         lifeCycleService.clusterChanged(new ClusterChangedEvent("any", clusterStateWithoutWatcherIndex, clusterStateWithWatcherIndex));
@@ -629,7 +714,7 @@ public void testWatcherReloadsOnNodeOutageWithWatcherShard() {
         when(watcherService.validate(any())).thenReturn(true);
         ClusterChangedEvent event = new ClusterChangedEvent("whatever", currentState, previousState);
         lifeCycleService.clusterChanged(event);
-        verify(watcherService).reload(eq(event.state()), anyString());
+        verify(watcherService).reload(eq(event.state()), anyString(), any());
     }
 
     private void startWatcher() {
@@ -658,7 +743,7 @@ private void startWatcher() {
 
         lifeCycleService.clusterChanged(new ClusterChangedEvent("foo", state, emptyState));
         assertThat(lifeCycleService.getState().get(), is(WatcherState.STARTED));
-        verify(watcherService, times(1)).reload(eq(state), anyString());
+        verify(watcherService, times(1)).reload(eq(state), anyString(), any());
         assertThat(lifeCycleService.shardRoutings(), hasSize(1));
 
         // reset the mock, the user has to mock everything themselves again

x-pack/plugin/watcher/src/test/java/org/elasticsearch/xpack/watcher/WatcherServiceTests.java

@@ -78,6 +78,7 @@
 import static org.mockito.Mockito.doAnswer;
 import static org.mockito.Mockito.mock;
 import static org.mockito.Mockito.never;
+import static org.mockito.Mockito.spy;
 import static org.mockito.Mockito.verify;
 import static org.mockito.Mockito.when;
 
@@ -348,12 +349,38 @@ void stopExecutor() {}
         ClusterState.Builder csBuilder = new ClusterState.Builder(new ClusterName("_name"));
         csBuilder.metadata(Metadata.builder());
 
-        service.reload(csBuilder.build(), "whatever");
+        service.reload(csBuilder.build(), "whatever", exception -> {});
         verify(executionService).clearExecutionsAndQueue(any());
         verify(executionService, never()).pause(any());
         verify(triggerService).pauseExecution();
     }
 
+    // the trigger service should not start unless watches are loaded successfully
+    public void testReloadingWatcherDoesNotStartTriggerServiceIfFailingToLoadWatches() {
+        ExecutionService executionService = mock(ExecutionService.class);
+        TriggerService triggerService = mock(TriggerService.class);
+        WatcherService service = new WatcherService(
+            Settings.EMPTY,
+            triggerService,
+            mock(TriggeredWatchStore.class),
+            executionService,
+            mock(WatchParser.class),
+            client,
+            EsExecutors.DIRECT_EXECUTOR_SERVICE
+        ) {
+            @Override
+            void stopExecutor() {}
+        };
+
+        ClusterState.Builder csBuilder = new ClusterState.Builder(new ClusterName("_name"));
+        Metadata metadata = spy(Metadata.builder().build());
+        when(metadata.getIndicesLookup()).thenThrow(RuntimeException.class); // simulate exception in WatcherService's private loadWatches()
+
+        service.reload(csBuilder.metadata(metadata).build(), "whatever", exception -> {});
+        verify(triggerService).pauseExecution();
+        verify(triggerService, never()).start(any());
+    }
+
     private static DiscoveryNode newNode() {
         return new DiscoveryNode(
             "node",