[Backport 2.17] [RW Separation] Add routing preference to route requests only to search replicas.

server/src/internalClusterTest/java/org/opensearch/indices/settings/SearchOnlyReplicaIT.java

@@ -8,14 +8,17 @@
 
 package org.opensearch.indices.settings;
 
+import org.opensearch.action.search.SearchResponse;
 import org.opensearch.action.support.WriteRequest;
 import org.opensearch.cluster.ClusterState;
 import org.opensearch.cluster.metadata.IndexMetadata;
 import org.opensearch.cluster.metadata.Metadata;
 import org.opensearch.cluster.routing.IndexShardRoutingTable;
+import org.opensearch.cluster.routing.Preference;
 import org.opensearch.cluster.routing.ShardRouting;
 import org.opensearch.common.settings.Settings;
 import org.opensearch.common.util.FeatureFlags;
+import org.opensearch.index.query.QueryBuilders;
 import org.opensearch.indices.replication.common.ReplicationType;
 import org.opensearch.test.InternalTestCluster;
 import org.opensearch.test.OpenSearchIntegTestCase;
@@ -110,7 +113,6 @@ public void testFailoverWithSearchReplica_WithWriterReplicas() throws IOExceptio
         // add back a node
         internalCluster().startDataOnlyNode();
         ensureGreen(TEST_INDEX);
-
     }
 
     public void testFailoverWithSearchReplica_WithoutWriterReplicas() throws IOException {
@@ -175,6 +177,39 @@ public void testSearchReplicaScaling() {
         assertActiveSearchShards(0);
     }
 
+    public void testSearchReplicaRoutingPreference() throws IOException {
+        int numSearchReplicas = 1;
+        int numWriterReplicas = 1;
+        internalCluster().startClusterManagerOnlyNode();
+        String primaryNodeName = internalCluster().startDataOnlyNode();
+        createIndex(
+            TEST_INDEX,
+            Settings.builder()
+                .put(indexSettings())
+                .put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, numWriterReplicas)
+                .put(IndexMetadata.SETTING_NUMBER_OF_SEARCH_REPLICAS, numSearchReplicas)
+                .build()
+        );
+        ensureYellow(TEST_INDEX);
+        client().prepareIndex(TEST_INDEX).setId("1").setSource("foo", "bar").setRefreshPolicy(WriteRequest.RefreshPolicy.IMMEDIATE).get();
+        // add 2 nodes for the replicas
+        internalCluster().startDataOnlyNodes(2);
+        ensureGreen(TEST_INDEX);
+
+        assertActiveShardCounts(numSearchReplicas, numWriterReplicas);
+
+        // set preference to search replica here - we default to this when there are
+        // search replicas but tests will randomize this value if unset
+        SearchResponse response = client().prepareSearch(TEST_INDEX)
+            .setPreference(Preference.SEARCH_REPLICA.type())
+            .setQuery(QueryBuilders.matchAllQuery())
+            .get();
+
+        String nodeId = response.getHits().getAt(0).getShard().getNodeId();
+        IndexShardRoutingTable indexShardRoutingTable = getIndexShardRoutingTable();
+        assertEquals(nodeId, indexShardRoutingTable.searchOnlyReplicas().get(0).currentNodeId());
+    }
+
     /**
      * Helper to assert counts of active shards for each type.
      */

server/src/main/java/org/opensearch/cluster/routing/IndexShardRoutingTable.java

@@ -648,15 +648,11 @@ public ShardIterator replicaActiveInitializingShardIt() {
             return new PlainShardIterator(shardId, Collections.emptyList());
         }
 
-        LinkedList<ShardRouting> ordered = new LinkedList<>();
-        for (ShardRouting replica : shuffler.shuffle(replicas)) {
-            if (replica.active()) {
-                ordered.addFirst(replica);
-            } else if (replica.initializing()) {
-                ordered.addLast(replica);
-            }
-        }
-        return new PlainShardIterator(shardId, ordered);
+        return filterAndOrderShards(replica -> true);
+    }
+
+    public ShardIterator searchReplicaActiveInitializingShardIt() {
+        return filterAndOrderShards(ShardRouting::isSearchOnly);
     }
 
     /**
@@ -687,6 +683,20 @@ public ShardIterator replicaFirstActiveInitializingShardsIt() {
         return new PlainShardIterator(shardId, ordered);
     }
 
+    private ShardIterator filterAndOrderShards(Predicate<ShardRouting> filter) {
+        LinkedList<ShardRouting> ordered = new LinkedList<>();
+        for (ShardRouting replica : shuffler.shuffle(replicas)) {
+            if (filter.test(replica)) {
+                if (replica.active()) {
+                    ordered.addFirst(replica);
+                } else if (replica.initializing()) {
+                    ordered.addLast(replica);
+                }
+            }
+        }
+        return new PlainShardIterator(shardId, ordered);
+    }
+
     /**
      * Returns an iterator on active and initializing shards residing on the provided nodeId.
      */

server/src/main/java/org/opensearch/cluster/routing/OperationRouting.java

@@ -121,6 +121,7 @@ public class OperationRouting {
     private volatile boolean isFailOpenEnabled;
     private volatile boolean isStrictWeightedShardRouting;
     private volatile boolean ignoreWeightedRouting;
+    private final boolean isReaderWriterSplitEnabled;
 
     public OperationRouting(Settings settings, ClusterSettings clusterSettings) {
         // whether to ignore awareness attributes when routing requests
@@ -141,6 +142,7 @@ public OperationRouting(Settings settings, ClusterSettings clusterSettings) {
         clusterSettings.addSettingsUpdateConsumer(WEIGHTED_ROUTING_FAILOPEN_ENABLED, this::setFailOpenEnabled);
         clusterSettings.addSettingsUpdateConsumer(STRICT_WEIGHTED_SHARD_ROUTING_ENABLED, this::setStrictWeightedShardRouting);
         clusterSettings.addSettingsUpdateConsumer(IGNORE_WEIGHTED_SHARD_ROUTING, this::setIgnoreWeightedRouting);
+        this.isReaderWriterSplitEnabled = FeatureFlags.READER_WRITER_SPLIT_EXPERIMENTAL_SETTING.get(settings);
     }
 
     void setUseAdaptiveReplicaSelection(boolean useAdaptiveReplicaSelection) {
@@ -254,6 +256,14 @@ public GroupShardsIterator<ShardIterator> searchShards(
                 preference = Preference.PRIMARY_FIRST.type();
             }
 
+            if (isReaderWriterSplitEnabled) {
+                if (preference == null || preference.isEmpty()) {
+                    if (indexMetadataForShard.getNumberOfSearchOnlyReplicas() > 0) {
+                        preference = Preference.SEARCH_REPLICA.type();
+                    }
+                }
+            }
+
             ShardIterator iterator = preferenceActiveShardIterator(
                 shard,
                 clusterState.nodes().getLocalNodeId(),
@@ -365,6 +375,8 @@ private ShardIterator preferenceActiveShardIterator(
                     return indexShard.primaryFirstActiveInitializingShardsIt();
                 case REPLICA_FIRST:
                     return indexShard.replicaFirstActiveInitializingShardsIt();
+                case SEARCH_REPLICA:
+                    return indexShard.searchReplicaActiveInitializingShardIt();
                 case ONLY_LOCAL:
                     return indexShard.onlyNodeActiveInitializingShardsIt(localNodeId);
                 case ONLY_NODES:

server/src/main/java/org/opensearch/cluster/routing/Preference.java

@@ -73,6 +73,11 @@ public enum Preference {
      */
     REPLICA_FIRST("_replica_first"),
 
+    /**
+     * Route to search replica shards
+     */
+    SEARCH_REPLICA("_search_replica"),
+
     /**
      * Route to the local shard only
      */
@@ -127,6 +132,8 @@ public static Preference parse(String preference) {
                 return ONLY_LOCAL;
             case "_only_nodes":
                 return ONLY_NODES;
+            case "_search_replica":
+                return SEARCH_REPLICA;
             default:
                 throw new IllegalArgumentException("no Preference for [" + preferenceType + "]");
         }

server/src/test/java/org/opensearch/cluster/routing/OperationRoutingTests.java

@@ -1118,6 +1118,82 @@ public void testPartialIndexPrimaryDefault() throws Exception {
         }
     }
 
+    public void testSearchReplicaDefaultRouting() throws Exception {
+        final int numShards = 1;
+        final int numReplicas = 2;
+        final int numSearchReplicas = 2;
+        final String indexName = "test";
+        final String[] indexNames = new String[] { indexName };
+
+        ClusterService clusterService = null;
+        ThreadPool threadPool = null;
+
+        try {
+            OperationRouting opRouting = new OperationRouting(
+                Settings.builder().put(FeatureFlags.READER_WRITER_SPLIT_EXPERIMENTAL, "true").build(),
+                new ClusterSettings(Settings.EMPTY, ClusterSettings.BUILT_IN_CLUSTER_SETTINGS)
+            );
+
+            ClusterState state = ClusterStateCreationUtils.stateWithAssignedPrimariesAndReplicas(
+                indexNames,
+                numShards,
+                numReplicas,
+                numSearchReplicas
+            );
+            IndexShardRoutingTable indexShardRoutingTable = state.getRoutingTable().index(indexName).getShards().get(0);
+            ShardId shardId = indexShardRoutingTable.searchOnlyReplicas().get(0).shardId();
+
+            threadPool = new TestThreadPool("testSearchReplicaDefaultRouting");
+            clusterService = ClusterServiceUtils.createClusterService(threadPool);
+
+            // add a search replica in initializing state:
+            DiscoveryNode node = new DiscoveryNode(
+                "node_initializing",
+                OpenSearchTestCase.buildNewFakeTransportAddress(),
+                Collections.emptyMap(),
+                new HashSet<>(DiscoveryNodeRole.BUILT_IN_ROLES),
+                Version.CURRENT
+            );
+
+            IndexMetadata indexMetadata = IndexMetadata.builder(indexName)
+                .settings(Settings.builder().put(state.metadata().index(indexName).getSettings()).build())
+                .numberOfSearchReplicas(3)
+                .numberOfReplicas(2)
+                .build();
+            Metadata.Builder metadataBuilder = Metadata.builder(state.metadata()).put(indexMetadata, false).generateClusterUuidIfNeeded();
+            IndexRoutingTable.Builder indexShardRoutingBuilder = IndexRoutingTable.builder(indexMetadata.getIndex());
+            indexShardRoutingBuilder.addIndexShard(indexShardRoutingTable);
+            indexShardRoutingBuilder.addShard(
+                TestShardRouting.newShardRouting(shardId, node.getId(), null, false, true, ShardRoutingState.INITIALIZING, null)
+            );
+            state = ClusterState.builder(state)
+                .routingTable(RoutingTable.builder().add(indexShardRoutingBuilder).build())
+                .metadata(metadataBuilder.build())
+                .build();
+
+            // Verify default preference is primary only
+            GroupShardsIterator<ShardIterator> groupIterator = opRouting.searchShards(state, indexNames, null, null);
+            assertThat("one group per shard", groupIterator.size(), equalTo(numShards));
+            for (ShardIterator shardIterator : groupIterator) {
+                assertEquals("We should have 3 shards returned", shardIterator.size(), 3);
+                int i = 0;
+                for (ShardRouting shardRouting : shardIterator) {
+                    assertTrue(
+                        "Only search replicas should exist with preference SEARCH_REPLICA",
+                        shardIterator.nextOrNull().isSearchOnly()
+                    );
+                    if (i == shardIterator.size()) {
+                        assertTrue("Initializing shard should appear last", shardRouting.initializing());
+                        assertFalse("Initializing shard should appear last", shardRouting.active());
+                    }
+                }
+            }
+        } finally {
+            IOUtils.close(clusterService);
+            terminate(threadPool);
+        }
+    }
+
     private DiscoveryNode[] setupNodes() {
         // Sets up two data nodes in zone-a and one data node in zone-b
         List<String> zones = Arrays.asList("a", "a", "b");

test/framework/src/main/java/org/opensearch/action/support/replication/ClusterStateCreationUtils.java

@@ -63,6 +63,7 @@
 
 import static org.opensearch.cluster.metadata.IndexMetadata.SETTING_CREATION_DATE;
 import static org.opensearch.cluster.metadata.IndexMetadata.SETTING_NUMBER_OF_REPLICAS;
+import static org.opensearch.cluster.metadata.IndexMetadata.SETTING_NUMBER_OF_SEARCH_REPLICAS;
 import static org.opensearch.cluster.metadata.IndexMetadata.SETTING_NUMBER_OF_SHARDS;
 import static org.opensearch.cluster.metadata.IndexMetadata.SETTING_VERSION_CREATED;
 import static org.opensearch.test.OpenSearchTestCase.randomFrom;
@@ -325,7 +326,18 @@ public static ClusterState stateWithAssignedPrimariesAndOneReplica(String index,
      * Creates cluster state with several indexes, shards and replicas and all shards STARTED.
      */
     public static ClusterState stateWithAssignedPrimariesAndReplicas(String[] indices, int numberOfShards, int numberOfReplicas) {
+        return stateWithAssignedPrimariesAndReplicas(indices, numberOfShards, numberOfReplicas, 0);
+    }
 
+    /**
+     * Creates cluster state with several indexes, shards and replicas and all shards STARTED.
+     */
+    public static ClusterState stateWithAssignedPrimariesAndReplicas(
+        String[] indices,
+        int numberOfShards,
+        int numberOfReplicas,
+        int numberOfSearchReplicas
+    ) {
         int numberOfDataNodes = numberOfReplicas + 1;
         DiscoveryNodes.Builder discoBuilder = DiscoveryNodes.builder();
         for (int i = 0; i < numberOfDataNodes + 1; i++) {
@@ -347,6 +359,7 @@ public static ClusterState stateWithAssignedPrimariesAndReplicas(String[] indice
                         .put(SETTING_VERSION_CREATED, Version.CURRENT)
                         .put(SETTING_NUMBER_OF_SHARDS, numberOfShards)
                         .put(SETTING_NUMBER_OF_REPLICAS, numberOfReplicas)
+                        .put(SETTING_NUMBER_OF_SEARCH_REPLICAS, numberOfSearchReplicas)
                         .put(SETTING_CREATION_DATE, System.currentTimeMillis())
                 )
                 .build();
@@ -363,6 +376,19 @@ public static ClusterState stateWithAssignedPrimariesAndReplicas(String[] indice
                         TestShardRouting.newShardRouting(index, i, newNode(replica + 1).getId(), null, false, ShardRoutingState.STARTED)
                     );
                 }
+                for (int replica = numberOfReplicas; replica < numberOfSearchReplicas + numberOfReplicas; replica++) {
+                    indexShardRoutingBuilder.addShard(
+                        TestShardRouting.newShardRouting(
+                            new ShardId(index, IndexMetadata.INDEX_UUID_NA_VALUE, i),
+                            newNode(replica + 1).getId(),
+                            null,
+                            false,
+                            true,
+                            ShardRoutingState.STARTED,
+                            null
+                        )
+                    );
+                }
                 indexRoutingTableBuilder.addIndexShard(indexShardRoutingBuilder.build());
             }
             routingTableBuilder.add(indexRoutingTableBuilder.build());