Pass only ElasticsearchTimeoutException through the response and

others as top-level errors

Author: pawankartik-elastic

server/src/internalClusterTest/java/org/elasticsearch/indices/cluster/RemoteFieldCapsForceConnectTimeoutIT.java

@@ -75,7 +75,7 @@ public void testTimeoutSetting() {
             MockTransportService mts = (MockTransportService) cluster(LOCAL_CLUSTER).getInstance(TransportService.class, nodeName);
 
             mts.addConnectBehavior(
-                cluster(REMOTE_CLUSTER_1).getInstance(TransportService.class, cluster(REMOTE_CLUSTER_1).getNodeNames()[0]),
+                cluster(REMOTE_CLUSTER_1).getInstance(TransportService.class, (String) null),
                 ((transport, discoveryNode, profile, listener) -> {
                     try {
                         latch.await();
@@ -103,7 +103,13 @@ public void testTimeoutSetting() {
             throw new AssertionError(e);
         } finally {
             var fieldCapsRequest = new FieldCapabilitiesRequest();
-            fieldCapsRequest.indices("*", "*:*");
+            /*
+             * We have a local and 2 remote clusters but will target only the remote that we stalled.
+             * This is because when the timeout kicks in, and we move on from the stalled remote, we do not want
+             * the error to be a top-level error. Rather, it must be present in the response object under "failures".
+             * All other errors are free to be top-level errors though.
+             */
+            fieldCapsRequest.indices(REMOTE_CLUSTER_1 + ":*");
             fieldCapsRequest.fields("foo", "bar", "baz");
             var result = safeGet(client().execute(TransportFieldCapabilitiesAction.TYPE, fieldCapsRequest));
 

server/src/main/java/org/elasticsearch/action/fieldcaps/TransportFieldCapabilitiesAction.java

@@ -11,6 +11,7 @@
 
 import org.apache.lucene.util.ArrayUtil;
 import org.apache.lucene.util.automaton.TooComplexToDeterminizeException;
+import org.elasticsearch.ElasticsearchTimeoutException;
 import org.elasticsearch.ExceptionsHelper;
 import org.elasticsearch.action.ActionListener;
 import org.elasticsearch.action.ActionListenerResponseHandler;
@@ -398,7 +399,21 @@ private static void mergeIndexResponses(
                 listener.onResponse(new FieldCapabilitiesResponse(new ArrayList<>(indexResponses.values()), failures));
             }
         } else {
-            listener.onResponse(new FieldCapabilitiesResponse(Collections.emptyList(), failures));
+            // we have no responses at all, maybe because of errors
+            if (indexFailures.isEmpty() == false) {
+                /*
+                 * Under no circumstances are we to pass timeout errors originating from SubscribableListener as top-level errors.
+                 * Instead, they should always be passed through the response object, as part of "failures".
+                 */
+                if (failures.stream().anyMatch(failure -> failure.getException() instanceof ElasticsearchTimeoutException)) {
+                    listener.onResponse(new FieldCapabilitiesResponse(Collections.emptyList(), failures));
+                } else {
+                    // throw back the first exception
+                    listener.onFailure(failures.get(0).getException());
+                }
+            } else {
+                listener.onResponse(new FieldCapabilitiesResponse(Collections.emptyList(), Collections.emptyList()));
+            }
         }
     }