Implement Segment replication Backpressure (#6563)

* Add Segment Replication backpressure. This PR introduces new mechanisms to keep track of the current replicas within a replication group and apply backpressure if they fall too far behind. Writes will be rejected under the following conditions: 1. More than half (default setting) of the replication group is 'stale'. Defined by setting MAX_ALLOWED_STALE_SHARDS. 2. A replica is stale if it is behind more than MAX_INDEXING_CHECKPOINTS, default 4 AND its current replication lag is over MAX_REPLICATION_TIME_SETTING, default 5 minutes. This PR intentionally implements rejections only for index operations, allowing other TransportWriteActions to succeed, TransportResyncReplicationAction and RetentionLeaseSyncAction. Blocking these requests will fail recoveries as new nodes are added. Signed-off-by: Marc Handalian <handalm@amazon.com> * Add changelog Signed-off-by: Marc Handalian <handalm@amazon.com> * Fix test class to match naming conventions. Signed-off-by: Marc Handalian <handalm@amazon.com> * PR feedback. Signed-off-by: Marc Handalian <handalm@amazon.com> * Change setting keys to remove index scope. Signed-off-by: Marc Handalian <handalm@amazon.com> --------- Signed-off-by: Marc Handalian <handalm@amazon.com>
opensearch-project · Mar 15, 2023 · 21cdb72 · 21cdb72
1 parent bb901f5
commit 21cdb72
Show file tree

Hide file tree

Showing 27 changed files with 1,396 additions and 23 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -12,6 +12,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 - Add 'base_path' setting to File System Repository ([#6558](https://github.com/opensearch-project/OpenSearch/pull/6558))
 - Return success on DeletePits when no PITs exist. ([#6544](https://github.com/opensearch-project/OpenSearch/pull/6544))
 - Add node repurpose command for search nodes ([#6517](https://github.com/opensearch-project/OpenSearch/pull/6517))
+- [Segment Replication] Apply backpressure when replicas fall behind ([#6563](https://github.com/opensearch-project/OpenSearch/pull/6563))
 
 ### Dependencies
 - Bump `org.apache.logging.log4j:log4j-core` from 2.18.0 to 2.20.0 ([#6490](https://github.com/opensearch-project/OpenSearch/pull/6490))

diff --git a/server/src/internalClusterTest/java/org/opensearch/index/SegmentReplicationPressureIT.java b/server/src/internalClusterTest/java/org/opensearch/index/SegmentReplicationPressureIT.java
@@ -0,0 +1,271 @@
+/*
+ * Copyright OpenSearch Contributors.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package org.opensearch.index;
+
+import org.opensearch.action.bulk.BulkItemResponse;
+import org.opensearch.action.bulk.BulkRequest;
+import org.opensearch.action.bulk.BulkResponse;
+import org.opensearch.action.index.IndexRequest;
+import org.opensearch.cluster.metadata.IndexMetadata;
+import org.opensearch.common.UUIDs;
+import org.opensearch.common.lease.Releasable;
+import org.opensearch.common.settings.Settings;
+import org.opensearch.common.unit.TimeValue;
+import org.opensearch.common.util.concurrent.OpenSearchRejectedExecutionException;
+import org.opensearch.index.shard.IndexShard;
+import org.opensearch.indices.replication.SegmentReplicationBaseIT;
+import org.opensearch.plugins.Plugin;
+import org.opensearch.rest.RestStatus;
+import org.opensearch.test.OpenSearchIntegTestCase;
+import org.opensearch.test.transport.MockTransportService;
+
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.Collection;
+import java.util.Collections;
+import java.util.List;
+import java.util.concurrent.CountDownLatch;
+import java.util.concurrent.atomic.AtomicInteger;
+
+import static java.util.Arrays.asList;
+import static org.opensearch.cluster.metadata.IndexMetadata.SETTING_NUMBER_OF_REPLICAS;
+import static org.opensearch.index.SegmentReplicationPressureService.MAX_INDEXING_CHECKPOINTS;
+import static org.opensearch.index.SegmentReplicationPressureService.MAX_REPLICATION_TIME_SETTING;
+import static org.opensearch.index.SegmentReplicationPressureService.SEGMENT_REPLICATION_INDEXING_PRESSURE_ENABLED;
+import static org.opensearch.test.hamcrest.OpenSearchAssertions.assertAcked;
+
+@OpenSearchIntegTestCase.ClusterScope(scope = OpenSearchIntegTestCase.Scope.TEST, numDataNodes = 0)
+public class SegmentReplicationPressureIT extends SegmentReplicationBaseIT {
+
+    private static final int MAX_CHECKPOINTS_BEHIND = 2;
+
+    @Override
+    protected Settings nodeSettings(int nodeOrdinal) {
+        return Settings.builder()
+            .put(super.nodeSettings(nodeOrdinal))
+            .put(SEGMENT_REPLICATION_INDEXING_PRESSURE_ENABLED.getKey(), true)
+            .put(MAX_REPLICATION_TIME_SETTING.getKey(), TimeValue.timeValueSeconds(1))
+            .put(MAX_INDEXING_CHECKPOINTS.getKey(), MAX_CHECKPOINTS_BEHIND)
+            .build();
+    }
+
+    @Override
+    protected Collection<Class<? extends Plugin>> nodePlugins() {
+        return asList(MockTransportService.TestPlugin.class);
+    }
+
+    public void testWritesRejected() throws Exception {
+        final String primaryNode = internalCluster().startNode();
+        createIndex(INDEX_NAME);
+        ensureYellowAndNoInitializingShards(INDEX_NAME);
+        final String replicaNode = internalCluster().startNode();
+        ensureGreen(INDEX_NAME);
+
+        final IndexShard primaryShard = getIndexShard(primaryNode, INDEX_NAME);
+        final List<String> replicaNodes = asList(replicaNode);
+        assertEqualSegmentInfosVersion(replicaNodes, primaryShard);
+
+        final CountDownLatch latch = new CountDownLatch(1);
+        final AtomicInteger totalDocs = new AtomicInteger(0);
+        try (final Releasable ignored = blockReplication(replicaNodes, latch)) {
+            Thread indexingThread = new Thread(() -> { totalDocs.getAndSet(indexUntilCheckpointCount()); });
+            indexingThread.start();
+            indexingThread.join();
+            latch.await();
+            // index again while we are stale.
+            assertBusy(() -> {
+                expectThrows(OpenSearchRejectedExecutionException.class, () -> {
+                    indexDoc();
+                    totalDocs.incrementAndGet();
+                });
+            });
+        }
+        refresh(INDEX_NAME);
+        // wait for the replicas to catch up after block is released.
+        waitForSearchableDocs(totalDocs.get(), replicaNodes.toArray(new String[] {}));
+
+        // index another doc showing there is no pressure enforced.
+        indexDoc();
+        waitForSearchableDocs(totalDocs.incrementAndGet(), replicaNodes.toArray(new String[] {}));
+        verifyStoreContent();
+    }
+
+    /**
+     * This test ensures that a replica can be added while the index is under write block.
+     * Ensuring that only write requests are blocked.
+     */
+    public void testAddReplicaWhileWritesBlocked() throws Exception {
+        final String primaryNode = internalCluster().startNode();
+        createIndex(INDEX_NAME);
+        ensureYellowAndNoInitializingShards(INDEX_NAME);
+        final String replicaNode = internalCluster().startNode();
+        ensureGreen(INDEX_NAME);
+
+        final IndexShard primaryShard = getIndexShard(primaryNode, INDEX_NAME);
+        final List<String> replicaNodes = new ArrayList<>();
+        replicaNodes.add(replicaNode);
+        assertEqualSegmentInfosVersion(replicaNodes, primaryShard);
+
+        final CountDownLatch latch = new CountDownLatch(1);
+        final AtomicInteger totalDocs = new AtomicInteger(0);
+        try (final Releasable ignored = blockReplication(replicaNodes, latch)) {
+            Thread indexingThread = new Thread(() -> { totalDocs.getAndSet(indexUntilCheckpointCount()); });
+            indexingThread.start();
+            indexingThread.join();
+            latch.await();
+            // index again while we are stale.
+            assertBusy(() -> {
+                expectThrows(OpenSearchRejectedExecutionException.class, () -> {
+                    indexDoc();
+                    totalDocs.incrementAndGet();
+                });
+            });
+            final String replica_2 = internalCluster().startNode();
+            assertAcked(
+                client().admin()
+                    .indices()
+                    .prepareUpdateSettings(INDEX_NAME)
+                    .setSettings(Settings.builder().put(SETTING_NUMBER_OF_REPLICAS, 2))
+            );
+            ensureGreen(INDEX_NAME);
+            replicaNodes.add(replica_2);
+            waitForSearchableDocs(totalDocs.get(), replica_2);
+        }
+        refresh(INDEX_NAME);
+        // wait for the replicas to catch up after block is released.
+        waitForSearchableDocs(totalDocs.get(), replicaNodes.toArray(new String[] {}));
+
+        // index another doc showing there is no pressure enforced.
+        indexDoc();
+        waitForSearchableDocs(totalDocs.incrementAndGet(), replicaNodes.toArray(new String[] {}));
+        verifyStoreContent();
+    }
+
+    public void testBelowReplicaLimit() throws Exception {
+        final Settings settings = Settings.builder().put(indexSettings()).put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 3).build();
+        final String primaryNode = internalCluster().startNode();
+        createIndex(INDEX_NAME, settings);
+        ensureYellowAndNoInitializingShards(INDEX_NAME);
+        List<String> replicaNodes = new ArrayList<>();
+        for (int i = 0; i < 3; i++) {
+            replicaNodes.add(internalCluster().startNode());
+        }
+        ensureGreen(INDEX_NAME);
+
+        final IndexShard primaryShard = getIndexShard(primaryNode, INDEX_NAME);
+        assertEqualSegmentInfosVersion(replicaNodes, primaryShard);
+
+        final CountDownLatch latch = new CountDownLatch(1);
+        final AtomicInteger totalDocs = new AtomicInteger(0);
+        // only block a single replica, pressure should not get applied.
+        try (final Releasable ignored = blockReplication(replicaNodes.subList(0, 1), latch)) {
+            Thread indexingThread = new Thread(() -> totalDocs.getAndSet(indexUntilCheckpointCount()));
+            indexingThread.start();
+            indexingThread.join();
+            latch.await();
+            indexDoc();
+            totalDocs.incrementAndGet();
+            refresh(INDEX_NAME);
+        }
+        // index another doc showing there is no pressure enforced.
+        indexDoc();
+        refresh(INDEX_NAME);
+        waitForSearchableDocs(totalDocs.incrementAndGet(), replicaNodes.toArray(new String[] {}));
+        verifyStoreContent();
+    }
+
+    public void testBulkWritesRejected() throws Exception {
+        final String primaryNode = internalCluster().startNode();
+        createIndex(INDEX_NAME);
+        ensureYellowAndNoInitializingShards(INDEX_NAME);
+        final String replicaNode = internalCluster().startNode();
+        final String coordinator = internalCluster().startCoordinatingOnlyNode(Settings.EMPTY);
+        ensureGreen(INDEX_NAME);
+
+        final IndexShard primaryShard = getIndexShard(primaryNode, INDEX_NAME);
+        final List<String> replicaNodes = asList(replicaNode);
+        assertEqualSegmentInfosVersion(replicaNodes, primaryShard);
+
+        final CountDownLatch latch = new CountDownLatch(1);
+        List<String> nodes = List.of(primaryNode, replicaNode, coordinator);
+
+        int docsPerBatch = randomIntBetween(1, 200);
+        int totalDocs = docsPerBatch * MAX_CHECKPOINTS_BEHIND;
+        try (final Releasable ignored = blockReplication(replicaNodes, latch)) {
+            Thread indexingThread = new Thread(() -> {
+                for (int i = 0; i < MAX_CHECKPOINTS_BEHIND + 1; i++) {
+                    executeBulkRequest(nodes, docsPerBatch);
+                    refresh(INDEX_NAME);
+                }
+            });
+            indexingThread.start();
+            indexingThread.join();
+            latch.await();
+            // try and index again while we are stale.
+            assertBusy(() -> { assertFailedRequests(executeBulkRequest(nodes, randomIntBetween(1, 200))); });
+        }
+        refresh(INDEX_NAME);
+        // wait for the replicas to catch up after block is released.
+        waitForSearchableDocs(totalDocs, replicaNodes.toArray(new String[] {}));
+
+        // index another doc showing there is no pressure enforced.
+        executeBulkRequest(nodes, totalDocs);
+        waitForSearchableDocs(totalDocs * 2L, replicaNodes.toArray(new String[] {}));
+        verifyStoreContent();
+    }
+
+    private BulkResponse executeBulkRequest(List<String> nodes, int docsPerBatch) {
+        final BulkRequest bulkRequest = new BulkRequest();
+        for (int j = 0; j < docsPerBatch; ++j) {
+            IndexRequest request = new IndexRequest(INDEX_NAME).id(UUIDs.base64UUID())
+                .source(Collections.singletonMap("key", randomAlphaOfLength(50)));
+            bulkRequest.add(request);
+        }
+        final BulkResponse bulkItemResponses = client(randomFrom(nodes)).bulk(bulkRequest).actionGet();
+        refresh(INDEX_NAME);
+        return bulkItemResponses;
+    }
+
+    /**
+     * Index and Refresh in batches to force checkpoints behind.
+     * Asserts that there are no stale replicas according to the primary until cp count is reached.
+     */
+    private int indexUntilCheckpointCount() {
+        int total = 0;
+        for (int i = 0; i < MAX_CHECKPOINTS_BEHIND; i++) {
+            final int numDocs = randomIntBetween(1, 100);
+            for (int j = 0; j < numDocs; ++j) {
+                indexDoc();
+            }
+            total += numDocs;
+            refresh(INDEX_NAME);
+        }
+        return total;
+    }
+
+    private void assertFailedRequests(BulkResponse response) {
+        assertTrue(Arrays.stream(response.getItems()).allMatch(BulkItemResponse::isFailed));
+        assertTrue(
+            Arrays.stream(response.getItems())
+                .map(BulkItemResponse::getFailure)
+                .allMatch((failure) -> failure.getStatus() == RestStatus.TOO_MANY_REQUESTS)
+        );
+    }
+
+    private void indexDoc() {
+        client().prepareIndex(INDEX_NAME).setId(UUIDs.base64UUID()).setSource("{}", "{}").get();
+    }
+
+    private void assertEqualSegmentInfosVersion(List<String> replicaNames, IndexShard primaryShard) {
+        for (String replicaName : replicaNames) {
+            final IndexShard replicaShard = getIndexShard(replicaName, INDEX_NAME);
+            assertEquals(
+                primaryShard.getLatestReplicationCheckpoint().getSegmentInfosVersion(),
+                replicaShard.getLatestReplicationCheckpoint().getSegmentInfosVersion()
+            );
+        }
+    }
+}
diff --git a/...internalClusterTest/java/org/opensearch/indices/replication/SegmentReplicationBaseIT.java b/...internalClusterTest/java/org/opensearch/indices/replication/SegmentReplicationBaseIT.java
@@ -17,6 +17,7 @@
 import org.opensearch.cluster.routing.IndexShardRoutingTable;
 import org.opensearch.cluster.routing.ShardRouting;
 import org.opensearch.common.Nullable;
+import org.opensearch.common.lease.Releasable;
 import org.opensearch.common.settings.Settings;
 import org.opensearch.common.util.FeatureFlags;
 import org.opensearch.index.Index;
@@ -30,12 +31,14 @@
 import org.opensearch.plugins.Plugin;
 import org.opensearch.test.OpenSearchIntegTestCase;
 import org.opensearch.test.transport.MockTransportService;
+import org.opensearch.transport.TransportService;
 
 import java.util.Arrays;
 import java.util.Collection;
 import java.util.List;
 import java.util.Map;
 import java.util.Optional;
+import java.util.concurrent.CountDownLatch;
 import java.util.concurrent.TimeUnit;
 import java.util.stream.Collectors;
 
@@ -197,4 +200,31 @@ protected IndexShard getIndexShard(String node, String indexName) {
         return indexService.getShard(shardId.get());
     }
 
+    protected Releasable blockReplication(List<String> nodes, CountDownLatch latch) {
+        CountDownLatch pauseReplicationLatch = new CountDownLatch(nodes.size());
+        for (String node : nodes) {
+
+            MockTransportService mockTargetTransportService = ((MockTransportService) internalCluster().getInstance(
+                TransportService.class,
+                node
+            ));
+            mockTargetTransportService.addSendBehavior((connection, requestId, action, request, options) -> {
+                if (action.equals(SegmentReplicationSourceService.Actions.GET_SEGMENT_FILES)) {
+                    try {
+                        latch.countDown();
+                        pauseReplicationLatch.await();
+                    } catch (InterruptedException e) {
+                        throw new RuntimeException(e);
+                    }
+                }
+                connection.sendRequest(requestId, action, request, options);
+            });
+        }
+        return () -> {
+            while (pauseReplicationLatch.getCount() > 0) {
+                pauseReplicationLatch.countDown();
+            }
+        };
+    }
+
 }