[Batch Fetch] Fix for hasInitiatedFetching to fix allocation explain and manual reroute APIs (opensearch-project#14972)

rahulkarajgikar · web-flow · commit d08c4253e189 · 2024-07-29T18:18:44.000+05:30
* Fix for hasInitiatedFetching() in batch mode

Signed-off-by: Rahul Karajgikar &lt;karajgik@amazon.com&gt;
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -5,6 +5,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 
 ## [Unreleased 2.x]
 ### Added
+- Fix for hasInitiatedFetching to fix allocation explain and manual reroute APIs (([#14972](https://github.com/opensearch-project/OpenSearch/pull/14972))
 
 ### Dependencies
 - Bump `org.apache.commons:commons-lang3` from 3.14.0 to 3.15.0 ([#14861](https://github.com/opensearch-project/OpenSearch/pull/14861))
diff --git a/server/src/internalClusterTest/java/org/opensearch/gateway/RecoveryFromGatewayIT.java b/server/src/internalClusterTest/java/org/opensearch/gateway/RecoveryFromGatewayIT.java
@@ -57,6 +57,7 @@
 import org.opensearch.cluster.routing.ShardRouting;
 import org.opensearch.cluster.routing.ShardRoutingState;
 import org.opensearch.cluster.routing.UnassignedInfo;
+import org.opensearch.cluster.routing.allocation.AllocateUnassignedDecision;
 import org.opensearch.cluster.routing.allocation.AllocationDecision;
 import org.opensearch.cluster.routing.allocation.ExistingShardsAllocator;
 import org.opensearch.cluster.service.ClusterService;
@@ -797,11 +798,26 @@ public void testBatchModeEnabledWithoutTimeout() throws Exception {
         );
         assertTrue(ExistingShardsAllocator.EXISTING_SHARDS_ALLOCATOR_BATCH_MODE.get(internalCluster().clusterService().getSettings()));
         assertEquals(1, gatewayAllocator.getNumberOfStartedShardBatches());
-        assertEquals(1, gatewayAllocator.getNumberOfStoreShardBatches());
+        // Replica shard would be marked ineligible since there are no data nodes.
+        // It would then be removed from any batch and batches would get deleted, so we would have 0 replica batches
+        assertEquals(0, gatewayAllocator.getNumberOfStoreShardBatches());
 
-        // Now start both data nodes and ensure batch mode is working
-        logger.info("--> restarting the stopped nodes");
+        // Now start one data node
+        logger.info("--> restarting the first stopped node");
         internalCluster().startDataOnlyNode(Settings.builder().put("node.name", dataOnlyNodes.get(0)).put(node0DataPathSettings).build());
+        ensureStableCluster(2);
+        ensureYellow("test");
+        assertEquals(0, gatewayAllocator.getNumberOfStartedShardBatches());
+        assertEquals(0, gatewayAllocator.getNumberOfStoreShardBatches());
+        assertEquals(0, gatewayAllocator.getNumberOfInFlightFetches());
+
+        // calling reroute and asserting on reroute response
+        logger.info("--> calling reroute while cluster is yellow");
+        clusterRerouteResponse = client().admin().cluster().prepareReroute().setRetryFailed(true).get();
+        assertTrue(clusterRerouteResponse.isAcknowledged());
+
+        // Now start last data node and ensure batch mode is working and cluster goes green
+        logger.info("--> restarting the second stopped node");
         internalCluster().startDataOnlyNode(Settings.builder().put("node.name", dataOnlyNodes.get(1)).put(node1DataPathSettings).build());
         ensureStableCluster(3);
         ensureGreen("test");
@@ -842,11 +858,26 @@ public void testBatchModeEnabledWithSufficientTimeoutAndClusterGreen() throws Ex
         );
         assertTrue(ExistingShardsAllocator.EXISTING_SHARDS_ALLOCATOR_BATCH_MODE.get(internalCluster().clusterService().getSettings()));
         assertEquals(1, gatewayAllocator.getNumberOfStartedShardBatches());
-        assertEquals(1, gatewayAllocator.getNumberOfStoreShardBatches());
+        // Replica shard would be marked ineligible since there are no data nodes.
+        // It would then be removed from any batch and batches would get deleted, so we would have 0 replica batches
+        assertEquals(0, gatewayAllocator.getNumberOfStoreShardBatches());
 
-        // Now start both data nodes and ensure batch mode is working
-        logger.info("--> restarting the stopped nodes");
+        // Now start one data nodes and ensure batch mode is working
+        logger.info("--> restarting the first stopped node");
         internalCluster().startDataOnlyNode(Settings.builder().put("node.name", dataOnlyNodes.get(0)).put(node0DataPathSettings).build());
+        ensureStableCluster(2);
+        ensureYellow("test");
+        assertEquals(0, gatewayAllocator.getNumberOfStartedShardBatches());
+        assertEquals(0, gatewayAllocator.getNumberOfStoreShardBatches());
+        assertEquals(0, gatewayAllocator.getNumberOfInFlightFetches());
+
+        // calling reroute and asserting on reroute response
+        logger.info("--> calling reroute while cluster is yellow");
+        clusterRerouteResponse = client().admin().cluster().prepareReroute().setRetryFailed(true).get();
+        assertTrue(clusterRerouteResponse.isAcknowledged());
+
+        // Now start last data node and ensure batch mode is working and cluster goes green
+        logger.info("--> restarting the second stopped node");
         internalCluster().startDataOnlyNode(Settings.builder().put("node.name", dataOnlyNodes.get(1)).put(node1DataPathSettings).build());
         ensureStableCluster(3);
         ensureGreen("test");
@@ -907,7 +938,9 @@ public void testBatchModeEnabledWithInSufficientTimeoutButClusterGreen() throws
 
         assertTrue(ExistingShardsAllocator.EXISTING_SHARDS_ALLOCATOR_BATCH_MODE.get(internalCluster().clusterService().getSettings()));
         assertEquals(10, gatewayAllocator.getNumberOfStartedShardBatches());
-        assertEquals(10, gatewayAllocator.getNumberOfStoreShardBatches());
+        // All replica shards would be marked ineligible since there are no data nodes.
+        // They would then be removed from any batch and batches would get deleted, so we would have 0 replica batches
+        assertEquals(0, gatewayAllocator.getNumberOfStoreShardBatches());
         health = client(internalCluster().getClusterManagerName()).admin().cluster().health(Requests.clusterHealthRequest()).actionGet();
         assertFalse(health.isTimedOut());
         assertEquals(RED, health.getStatus());
@@ -1051,6 +1084,18 @@ public void testMultipleReplicaShardAssignmentWithDelayedAllocationAndDifferentN
         ensureGreen("test");
     }
 
+    public void testAllocationExplainReturnsNoWhenExtraReplicaShardInNonBatchMode() throws Exception {
+        // Non batch mode - This test is to validate that we don't return AWAITING_INFO in allocation explain API when the deciders are
+        // returning NO
+        this.allocationExplainReturnsNoWhenExtraReplicaShard(false);
+    }
+
+    public void testAllocationExplainReturnsNoWhenExtraReplicaShardInBatchMode() throws Exception {
+        // Batch mode - This test is to validate that we don't return AWAITING_INFO in allocation explain API when the deciders are
+        // returning NO
+        this.allocationExplainReturnsNoWhenExtraReplicaShard(true);
+    }
+
     public void testNBatchesCreationAndAssignment() throws Exception {
         // we will reduce batch size to 5 to make sure we have enough batches to test assignment
         // Total number of primary shards = 50 (50 indices*1)
@@ -1104,7 +1149,9 @@ public void testNBatchesCreationAndAssignment() throws Exception {
         );
         assertTrue(ExistingShardsAllocator.EXISTING_SHARDS_ALLOCATOR_BATCH_MODE.get(internalCluster().clusterService().getSettings()));
         assertEquals(10, gatewayAllocator.getNumberOfStartedShardBatches());
-        assertEquals(10, gatewayAllocator.getNumberOfStoreShardBatches());
+        // All replica shards would be marked ineligible since there are no data nodes.
+        // They would then be removed from any batch and batches would get deleted, so we would have 0 replica batches
+        assertEquals(0, gatewayAllocator.getNumberOfStoreShardBatches());
         health = client(internalCluster().getClusterManagerName()).admin().cluster().health(Requests.clusterHealthRequest()).actionGet();
         assertFalse(health.isTimedOut());
         assertEquals(RED, health.getStatus());
@@ -1193,7 +1240,9 @@ public void testCulpritShardInBatch() throws Exception {
         );
         assertTrue(ExistingShardsAllocator.EXISTING_SHARDS_ALLOCATOR_BATCH_MODE.get(internalCluster().clusterService().getSettings()));
         assertEquals(1, gatewayAllocator.getNumberOfStartedShardBatches());
-        assertEquals(1, gatewayAllocator.getNumberOfStoreShardBatches());
+        // Replica shard would be marked ineligible since there are no data nodes.
+        // It would then be removed from any batch and batches would get deleted, so we would have 0 replica batches
+        assertEquals(0, gatewayAllocator.getNumberOfStoreShardBatches());
         assertTrue(clusterRerouteResponse.isAcknowledged());
         health = client(internalCluster().getClusterManagerName()).admin().cluster().health(Requests.clusterHealthRequest()).actionGet();
         assertFalse(health.isTimedOut());
@@ -1511,4 +1560,97 @@ private List<String> findNodesWithShard(final boolean primary) {
         Collections.shuffle(requiredStartedShards, random());
         return requiredStartedShards.stream().map(shard -> state.nodes().get(shard.currentNodeId()).getName()).collect(Collectors.toList());
     }
+
+    private void allocationExplainReturnsNoWhenExtraReplicaShard(boolean batchModeEnabled) throws Exception {
+        internalCluster().startClusterManagerOnlyNodes(
+            1,
+            Settings.builder().put(ExistingShardsAllocator.EXISTING_SHARDS_ALLOCATOR_BATCH_MODE.getKey(), batchModeEnabled).build()
+        );
+        internalCluster().startDataOnlyNodes(5);
+        createIndex(
+            "test",
+            Settings.builder().put(IndexMetadata.SETTING_NUMBER_OF_SHARDS, 1).put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 4).build()
+        );
+        ensureGreen("test");
+        ensureStableCluster(6);
+
+        // Stop one of the nodes to make the cluster yellow
+        // We cannot directly create an index with replica = data node count because then the whole flow will get skipped due to
+        // INDEX_CREATED
+        List<String> nodesWithReplicaShards = findNodesWithShard(false);
+        Settings replicaNodeDataPathSettings = internalCluster().dataPathSettings(nodesWithReplicaShards.get(0));
+        internalCluster().stopRandomNode(InternalTestCluster.nameFilter(nodesWithReplicaShards.get(0)));
+
+        ensureStableCluster(5);
+        ensureYellow("test");
+
+        logger.info("--> calling allocation explain API");
+        // shard should have decision NO because there is no valid node for the extra replica to go to
+        AllocateUnassignedDecision aud = client().admin()
+            .cluster()
+            .prepareAllocationExplain()
+            .setIndex("test")
+            .setShard(0)
+            .setPrimary(false)
+            .get()
+            .getExplanation()
+            .getShardAllocationDecision()
+            .getAllocateDecision();
+
+        assertEquals(AllocationDecision.NO, aud.getAllocationDecision());
+        assertEquals("cannot allocate because allocation is not permitted to any of the nodes", aud.getExplanation());
+
+        // Now creating a new index with too many replicas and trying again
+        createIndex(
+            "test2",
+            Settings.builder().put(IndexMetadata.SETTING_NUMBER_OF_SHARDS, 1).put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 5).build()
+        );
+
+        ensureYellowAndNoInitializingShards("test2");
+
+        logger.info("--> calling allocation explain API again");
+        // shard should have decision NO because there are 6 replicas and 4 data nodes
+        aud = client().admin()
+            .cluster()
+            .prepareAllocationExplain()
+            .setIndex("test2")
+            .setShard(0)
+            .setPrimary(false)
+            .get()
+            .getExplanation()
+            .getShardAllocationDecision()
+            .getAllocateDecision();
+
+        assertEquals(AllocationDecision.NO, aud.getAllocationDecision());
+        assertEquals("cannot allocate because allocation is not permitted to any of the nodes", aud.getExplanation());
+
+        logger.info("--> restarting the stopped node");
+        internalCluster().startDataOnlyNode(
+            Settings.builder().put("node.name", nodesWithReplicaShards.get(0)).put(replicaNodeDataPathSettings).build()
+        );
+
+        ensureStableCluster(6);
+        ensureGreen("test");
+
+        logger.info("--> calling allocation explain API 3rd time");
+        // shard should still have decision NO because there are 6 replicas and 5 data nodes
+        aud = client().admin()
+            .cluster()
+            .prepareAllocationExplain()
+            .setIndex("test2")
+            .setShard(0)
+            .setPrimary(false)
+            .get()
+            .getExplanation()
+            .getShardAllocationDecision()
+            .getAllocateDecision();
+
+        assertEquals(AllocationDecision.NO, aud.getAllocationDecision());
+        assertEquals("cannot allocate because allocation is not permitted to any of the nodes", aud.getExplanation());
+
+        internalCluster().startDataOnlyNodes(1);
+
+        ensureStableCluster(7);
+        ensureGreen("test2");
+    }
 }
diff --git a/server/src/main/java/org/opensearch/gateway/AsyncShardBatchFetch.java b/server/src/main/java/org/opensearch/gateway/AsyncShardBatchFetch.java
@@ -80,6 +80,14 @@ public synchronized void clearShard(ShardId shardId) {
         this.cache.deleteShard(shardId);
     }
 
+    public boolean hasEmptyCache() {
+        return this.cache.getCache().isEmpty();
+    }
+
+    public AsyncShardFetchCache<T> getCache() {
+        return this.cache;
+    }
+
     /**
      * Cache implementation of transport actions returning batch of shards related data in the response.
      * Store node level responses of transport actions like {@link TransportNodesListGatewayStartedShardsBatch} or
diff --git a/server/src/main/java/org/opensearch/gateway/ReplicaShardBatchAllocator.java b/server/src/main/java/org/opensearch/gateway/ReplicaShardBatchAllocator.java
@@ -183,7 +183,7 @@ private AllocateUnassignedDecision getUnassignedShardAllocationDecision(
         if (allocationDecision.type() != Decision.Type.YES && (!explain || !hasInitiatedFetching(shardRouting))) {
             // only return early if we are not in explain mode, or we are in explain mode but we have not
             // yet attempted to fetch any shard data
-            logger.trace("{}: ignoring allocation, can't be allocated on any node", shardRouting);
+            logger.trace("{}: ignoring allocation, can't be allocated on any node. Decision: {}", shardRouting, allocationDecision.type());
             return AllocateUnassignedDecision.no(
                 UnassignedInfo.AllocationStatus.fromDecision(allocationDecision.type()),
                 result.v2() != null ? new ArrayList<>(result.v2().values()) : null
diff --git a/server/src/main/java/org/opensearch/gateway/ShardsBatchGatewayAllocator.java b/server/src/main/java/org/opensearch/gateway/ShardsBatchGatewayAllocator.java
@@ -576,8 +576,37 @@ protected AsyncShardFetch.FetchResult<TransportNodesListShardStoreMetadataBatch.
 
         @Override
         protected boolean hasInitiatedFetching(ShardRouting shard) {
+            /**
+             * This function is to check if asyncFetch has happened before for this shard batch, or is ongoing.
+             * It should return false if there has never been a fetch for this batch.
+             * This function is currently only used in the case of replica shards when all deciders returned NO/THROTTLE, and explain mode is ON.
+             * Allocation explain and manual reroute APIs try to append shard store information (matching bytes) to the allocation decision.
+             * However, these APIs do not want to trigger a new asyncFetch for these ineligible shards, unless the data from nodes is already there.
+             * This function is used to see if a fetch has happened to decide if it is possible to append shard store info without a new async fetch.
+             * In the case when shard has a batch but no fetch has happened before, it would be because it is a new batch.
+             * In the case when shard has a batch, and a fetch has happened before, and no fetch is ongoing, it would be because we have already completed fetch for all nodes.
+             *
+             * In order to check if a fetch has ever happened, we check 2 things:
+             * 1. If the shard batch cache is empty, we know that fetch has never happened so we return false.
+             * 2. If we see that the list of nodes to fetch from is empty, we know that all nodes have data or are ongoing a fetch. So we return true.
+             * 3. Otherwise we return false.
+             *
+             * see {@link AsyncShardFetchCache#findNodesToFetch()}
+             */
             String batchId = getBatchId(shard, shard.primary());
-            return batchId != null;
+            if (batchId == null) {
+                return false;
+            }
+            logger.trace("Checking if fetching done for batch id {}", batchId);
+            ShardsBatch shardsBatch = shard.primary() ? batchIdToStartedShardBatch.get(batchId) : batchIdToStoreShardBatch.get(batchId);
+            // if fetchData has never been called, the per node cache will be empty and have no nodes
+            // this is because cache.fillShardCacheWithDataNodes(nodes) initialises this map and is called in AsyncShardFetch.fetchData
+            if (shardsBatch == null || shardsBatch.getAsyncFetcher().hasEmptyCache()) {
+                logger.trace("Batch cache is empty for batch {} ", batchId);
+                return false;
+            }
+            // this check below is to make sure we already have all the data and that we wouldn't create a new async fetchData call
+            return shardsBatch.getAsyncFetcher().getCache().findNodesToFetch().isEmpty();
         }
     }
 
