Introduce batched query execution and data-node side reduce (#121885) (#126563)

* Introduce batched query execution and data-node side reduce (#121885)

This change moves the query phase a single roundtrip per node just like can_match or field_caps work already.
A a result of executing multiple shard queries from a single request we can also partially reduce each node's query results on the data node side before responding to the coordinating node.

As a result this change significantly reduces the impact of network latencies on the end-to-end query performance, reduces the amount of work done (memory and cpu) on the coordinating node and the network traffic by factors of up to the number of shards per data node!

Benchmarking shows up to orders of magnitude improvements in heap and network traffic dimensions in querying across a larger number of shards.

* Filter out empty top docs results before merging (#126385)

`Lucene.EMPTY_TOP_DOCS` to identify empty to docs results. These were previously
null results, but did not need to be send over transport as incremental reduction
was performed only on the data node.

Now it can happen that the coord node received a merge result with empty top docs,
which has nothing interesting for merging, but that can lead to an exception because
the type of the empty array does not match the type of other shards results, for
instance if the query was sorted by field. To resolve this, we filter out empty
top docs results before merging.

Closes #126118

---------

Co-authored-by: Luca Cavanna <[email protected]>

Author: original-brownbear

docs/changelog/121885.yaml

@@ -0,0 +1,5 @@
+pr: 121885
+summary: Introduce batched query execution and data-node side reduce
+area: Search
+type: enhancement
+issues: []

docs/changelog/126385.yaml

@@ -0,0 +1,6 @@
+pr: 126385
+summary: Filter out empty top docs results before merging
+area: Search
+type: bug
+issues:
+ - 126118

qa/smoke-test-http/src/internalClusterTest/java/org/elasticsearch/http/SearchErrorTraceIT.java

@@ -16,6 +16,7 @@
 import org.elasticsearch.action.search.MultiSearchRequest;
 import org.elasticsearch.action.search.SearchRequest;
 import org.elasticsearch.client.Request;
+import org.elasticsearch.common.settings.Settings;
 import org.elasticsearch.common.util.CollectionUtils;
 import org.elasticsearch.plugins.Plugin;
 import org.elasticsearch.search.ErrorTraceHelper;
@@ -24,6 +25,7 @@
 import org.elasticsearch.test.MockLog;
 import org.elasticsearch.test.transport.MockTransportService;
 import org.elasticsearch.xcontent.XContentType;
+import org.junit.After;
 import org.junit.Before;
 import org.junit.BeforeClass;
 
@@ -50,6 +52,13 @@ public static void setDebugLogLevel() {
     @Before
     public void setupMessageListener() {
         hasStackTrace = ErrorTraceHelper.setupErrorTraceListener(internalCluster());
+        // TODO: make this test work with batched query execution by enhancing ErrorTraceHelper.setupErrorTraceListener
+        updateClusterSettings(Settings.builder().put(SearchService.BATCHED_QUERY_PHASE.getKey(), false));
+    }
+
+    @After
+    public void resetSettings() {
+        updateClusterSettings(Settings.builder().putNull(SearchService.BATCHED_QUERY_PHASE.getKey()));
     }
 
     private void setupIndexWithDocs() {

rest-api-spec/src/yamlRestTest/resources/rest-api-spec/test/search/120_batch_reduce_size.yml

@@ -1,4 +1,7 @@
 setup:
+  - skip:
+      awaits_fix: "TODO fix this test, the response with batched execution is not deterministic enough for the available matchers"
+
   - do:
       indices.create:
           index: test_1
@@ -48,7 +51,6 @@ setup:
         batched_reduce_size: 2
         body: { "size" : 0, "aggs" : { "str_terms" : { "terms" : { "field" : "str" } } } }
 
-  - match: { num_reduce_phases: 4 }
   - match: { hits.total: 3 }
   - length: { aggregations.str_terms.buckets: 2 }
   - match: { aggregations.str_terms.buckets.0.key: "abc" }

server/src/internalClusterTest/java/org/elasticsearch/action/IndicesRequestIT.java

@@ -574,11 +574,8 @@ public void testSearchQueryThenFetch() throws Exception {
         );
 
         clearInterceptedActions();
-        assertIndicesSubset(
-            Arrays.asList(searchRequest.indices()),
-            SearchTransportService.QUERY_ACTION_NAME,
-            SearchTransportService.FETCH_ID_ACTION_NAME
-        );
+        assertIndicesSubset(Arrays.asList(searchRequest.indices()), true, SearchTransportService.QUERY_ACTION_NAME);
+        assertIndicesSubset(Arrays.asList(searchRequest.indices()), SearchTransportService.FETCH_ID_ACTION_NAME);
     }
 
     public void testSearchDfsQueryThenFetch() throws Exception {
@@ -631,10 +628,6 @@ private static void assertIndicesSubset(List<String> indices, String... actions)
         assertIndicesSubset(indices, false, actions);
     }
 
-    private static void assertIndicesSubsetOptionalRequests(List<String> indices, String... actions) {
-        assertIndicesSubset(indices, true, actions);
-    }
-
     private static void assertIndicesSubset(List<String> indices, boolean optional, String... actions) {
         // indices returned by each bulk shard request need to be a subset of the original indices
         for (String action : actions) {

server/src/internalClusterTest/java/org/elasticsearch/action/admin/cluster/node/tasks/TasksIT.java

@@ -41,6 +41,7 @@
 import org.elasticsearch.index.query.QueryBuilders;
 import org.elasticsearch.persistent.PersistentTasksCustomMetadata;
 import org.elasticsearch.plugins.Plugin;
+import org.elasticsearch.search.SearchService;
 import org.elasticsearch.search.builder.SearchSourceBuilder;
 import org.elasticsearch.tasks.RemovedTaskListener;
 import org.elasticsearch.tasks.Task;
@@ -352,6 +353,8 @@ public void testTransportBulkTasks() {
     }
 
     public void testSearchTaskDescriptions() {
+        // TODO: enhance this test to also check the tasks created by batched query execution
+        updateClusterSettings(Settings.builder().put(SearchService.BATCHED_QUERY_PHASE.getKey(), false));
         registerTaskManagerListeners(TransportSearchAction.TYPE.name());  // main task
         registerTaskManagerListeners(TransportSearchAction.TYPE.name() + "[*]");  // shard task
         createIndex("test");
@@ -398,7 +401,7 @@ public void testSearchTaskDescriptions() {
             // assert that all task descriptions have non-zero length
             assertThat(taskInfo.description().length(), greaterThan(0));
         }
-
+        updateClusterSettings(Settings.builder().putNull(SearchService.BATCHED_QUERY_PHASE.getKey()));
     }
 
     public void testSearchTaskHeaderLimit() {

server/src/internalClusterTest/java/org/elasticsearch/action/search/TransportSearchIT.java

@@ -40,6 +40,7 @@
 import org.elasticsearch.rest.RestStatus;
 import org.elasticsearch.search.DocValueFormat;
 import org.elasticsearch.search.SearchHit;
+import org.elasticsearch.search.SearchService;
 import org.elasticsearch.search.aggregations.AbstractAggregationBuilder;
 import org.elasticsearch.search.aggregations.AggregationBuilder;
 import org.elasticsearch.search.aggregations.AggregationExecutionContext;
@@ -445,6 +446,7 @@ public void testSearchIdle() throws Exception {
     }
 
     public void testCircuitBreakerReduceFail() throws Exception {
+        updateClusterSettings(Settings.builder().put(SearchService.BATCHED_QUERY_PHASE.getKey(), false));
         int numShards = randomIntBetween(1, 10);
         indexSomeDocs("test", numShards, numShards * 3);
 
@@ -518,7 +520,9 @@ public void onFailure(Exception exc) {
             }
             assertBusy(() -> assertThat(requestBreakerUsed(), equalTo(0L)));
         } finally {
-            updateClusterSettings(Settings.builder().putNull("indices.breaker.request.limit"));
+            updateClusterSettings(
+                Settings.builder().putNull("indices.breaker.request.limit").putNull(SearchService.BATCHED_QUERY_PHASE.getKey())
+            );
         }
     }
 

server/src/internalClusterTest/java/org/elasticsearch/search/SearchCancellationIT.java

@@ -23,6 +23,7 @@
 import org.elasticsearch.action.search.TransportSearchAction;
 import org.elasticsearch.action.search.TransportSearchScrollAction;
 import org.elasticsearch.common.Strings;
+import org.elasticsearch.common.settings.Settings;
 import org.elasticsearch.core.TimeValue;
 import org.elasticsearch.script.Script;
 import org.elasticsearch.script.ScriptType;
@@ -239,6 +240,8 @@ public void testCancelMultiSearch() throws Exception {
     }
 
     public void testCancelFailedSearchWhenPartialResultDisallowed() throws Exception {
+        // TODO: make this test compatible with batched execution, currently the exceptions are slightly different with batched
+        updateClusterSettings(Settings.builder().put(SearchService.BATCHED_QUERY_PHASE.getKey(), false));
         // Have at least two nodes so that we have parallel execution of two request guaranteed even if max concurrent requests per node
         // are limited to 1
         internalCluster().ensureAtLeastNumDataNodes(2);

server/src/internalClusterTest/java/org/elasticsearch/search/aggregations/bucket/TermsDocCountErrorIT.java

@@ -13,12 +13,15 @@
 import org.elasticsearch.action.search.SearchResponse;
 import org.elasticsearch.cluster.metadata.IndexMetadata;
 import org.elasticsearch.common.settings.Settings;
+import org.elasticsearch.search.SearchService;
 import org.elasticsearch.search.aggregations.Aggregator.SubAggCollectionMode;
 import org.elasticsearch.search.aggregations.BucketOrder;
 import org.elasticsearch.search.aggregations.bucket.terms.Terms;
 import org.elasticsearch.search.aggregations.bucket.terms.Terms.Bucket;
 import org.elasticsearch.search.aggregations.bucket.terms.TermsAggregatorFactory.ExecutionMode;
 import org.elasticsearch.test.ESIntegTestCase;
+import org.junit.After;
+import org.junit.Before;
 
 import java.io.IOException;
 import java.util.ArrayList;
@@ -50,6 +53,18 @@ public static String randomExecutionHint() {
 
     private static int numRoutingValues;
 
+    @Before
+    public void disableBatchedExecution() {
+        // TODO: it's practically impossible to get a 100% deterministic test with batched execution unfortunately, adjust this test to
+        // still do something useful with batched execution (i.e. use somewhat relaxed assertions)
+        updateClusterSettings(Settings.builder().put(SearchService.BATCHED_QUERY_PHASE.getKey(), false));
+    }
+
+    @After
+    public void resetSettings() {
+        updateClusterSettings(Settings.builder().putNull(SearchService.BATCHED_QUERY_PHASE.getKey()));
+    }
+
     @Override
     public void setupSuiteScopeCluster() throws Exception {
         assertAcked(indicesAdmin().prepareCreate("idx").setMapping(STRING_FIELD_NAME, "type=keyword").get());

server/src/main/java/org/elasticsearch/TransportVersions.java

@@ -203,6 +203,7 @@ static TransportVersion def(int id) {
     public static final TransportVersion REMOTE_EXCEPTION_8_19 = def(8_841_0_16);
     public static final TransportVersion AMAZON_BEDROCK_TASK_SETTINGS_8_19 = def(8_841_0_17);
     public static final TransportVersion SEMANTIC_TEXT_CHUNKING_CONFIG_8_19 = def(8_841_0_18);
+    public static final TransportVersion BATCHED_QUERY_PHASE_VERSION_BACKPORT_8_X = def(8_841_0_19);
 
     /*
      * STOP! READ THIS FIRST! No, really,