Use all profiling events on startup

With this commit we use `profiling-events-all` to query profiling events
if the appropriate index for a sample count is not present yet. This can
happen on cluster startup when only a few events have been accumulated
because additional indices are only created when there are enough
events. This aligns behavior of the Elasticsearch plugin with the Kibana
plugin.

Relates elastic#91640

x-pack/plugin/profiler/src/main/java/org/elasticsearch/xpack/profiler/TransportGetProfilingAction.java

@@ -6,6 +6,8 @@
  */
 package org.elasticsearch.xpack.profiler;
 
+import org.apache.logging.log4j.LogManager;
+import org.apache.logging.log4j.Logger;
 import org.elasticsearch.action.ActionListener;
 import org.elasticsearch.action.get.MultiGetItemResponse;
 import org.elasticsearch.action.get.MultiGetResponse;
@@ -17,6 +19,7 @@
 import org.elasticsearch.client.internal.node.NodeClient;
 import org.elasticsearch.common.inject.Inject;
 import org.elasticsearch.common.util.Maps;
+import org.elasticsearch.index.IndexNotFoundException;
 import org.elasticsearch.search.aggregations.bucket.terms.StringTerms;
 import org.elasticsearch.search.aggregations.bucket.terms.TermsAggregationBuilder;
 import org.elasticsearch.search.aggregations.metrics.Sum;
@@ -33,6 +36,7 @@
 import java.util.Set;
 
 public class TransportGetProfilingAction extends HandledTransportAction<GetProfilingRequest, GetProfilingResponse> {
+    private static final Logger log = LogManager.getLogger(TransportGetProfilingAction.class);
     private final NodeClient nodeClient;
     private final TransportService transportService;
 
@@ -59,11 +63,21 @@ public void onResponse(SearchResponse searchResponse) {
                     searchEventGroupByStackTrace(client, request, resampledIndex, submitListener);
                 }
 
-                @Override
-                public void onFailure(Exception e) {
+            @Override
+            public void onFailure(Exception e) {
+                // Apart from profiling-events-all, indices are created lazily. In a relatively empty cluster it can happen
+                // that there are so few data that we need to resort to the full index. As this is an edge case we'd rather
+                // fail instead of prematurely checking for existence in all cases.
+                if (e instanceof IndexNotFoundException) {
+                    String missingIndex = ((IndexNotFoundException) e).getIndex().getName();
+                    EventsIndex fullIndex = EventsIndex.FULL_INDEX;
+                    log.debug("Index [{}] does not exist. Using [{}] instead.", missingIndex, fullIndex.getName());
+                    searchEventGroupByStackTrace(client, request, fullIndex, submitListener);
+                } else {
                     submitListener.onFailure(e);
                 }
-            });
+            }
+        });
     }
 
     private void searchEventGroupByStackTrace(