Examples with ReplicationTracker org.opensearch.index.seqno.ReplicationTracker used on opensource projects

Example 1 with ReplicationTracker

use of org.opensearch.index.seqno.ReplicationTracker in project OpenSearch by opensearch-project.

the class InternalEngineTests method testSeqNoAndCheckpoints.

public void testSeqNoAndCheckpoints() throws IOException, InterruptedException {
    final int opCount = randomIntBetween(1, 256);
    long primarySeqNo = SequenceNumbers.NO_OPS_PERFORMED;
    final String[] ids = new String[] { "1", "2", "3" };
    final Set<String> indexedIds = new HashSet<>();
    long localCheckpoint = SequenceNumbers.NO_OPS_PERFORMED;
    long replicaLocalCheckpoint = SequenceNumbers.NO_OPS_PERFORMED;
    final long globalCheckpoint;
    long maxSeqNo = SequenceNumbers.NO_OPS_PERFORMED;
    IOUtils.close(store, engine);
    store = createStore();
    InternalEngine initialEngine = null;
    try {
        initialEngine = createEngine(defaultSettings, store, createTempDir(), newLogMergePolicy(), null);
        final ShardRouting primary = TestShardRouting.newShardRouting("test", shardId.id(), "node1", null, true, ShardRoutingState.STARTED, allocationId);
        final ShardRouting initializingReplica = TestShardRouting.newShardRouting(shardId, "node2", false, ShardRoutingState.INITIALIZING);
        ReplicationTracker gcpTracker = (ReplicationTracker) initialEngine.config().getGlobalCheckpointSupplier();
        gcpTracker.updateFromMaster(1L, new HashSet<>(Collections.singletonList(primary.allocationId().getId())), new IndexShardRoutingTable.Builder(shardId).addShard(primary).build());
        gcpTracker.activatePrimaryMode(primarySeqNo);
        if (defaultSettings.isSoftDeleteEnabled()) {
            final CountDownLatch countDownLatch = new CountDownLatch(1);
            gcpTracker.addPeerRecoveryRetentionLease(initializingReplica.currentNodeId(), SequenceNumbers.NO_OPS_PERFORMED, ActionListener.wrap(countDownLatch::countDown));
            countDownLatch.await();
        }
        gcpTracker.updateFromMaster(2L, new HashSet<>(Collections.singletonList(primary.allocationId().getId())), new IndexShardRoutingTable.Builder(shardId).addShard(primary).addShard(initializingReplica).build());
        gcpTracker.initiateTracking(initializingReplica.allocationId().getId());
        gcpTracker.markAllocationIdAsInSync(initializingReplica.allocationId().getId(), replicaLocalCheckpoint);
        final ShardRouting replica = initializingReplica.moveToStarted();
        gcpTracker.updateFromMaster(3L, new HashSet<>(Arrays.asList(primary.allocationId().getId(), replica.allocationId().getId())), new IndexShardRoutingTable.Builder(shardId).addShard(primary).addShard(replica).build());
        for (int op = 0; op < opCount; op++) {
            final String id;
            // mostly index, sometimes delete
            if (rarely() && indexedIds.isEmpty() == false) {
                // we have some docs indexed, so delete one of them
                id = randomFrom(indexedIds);
                final Engine.Delete delete = new Engine.Delete("test", id, newUid(id), UNASSIGNED_SEQ_NO, primaryTerm.get(), rarely() ? 100 : Versions.MATCH_ANY, VersionType.INTERNAL, PRIMARY, System.nanoTime(), UNASSIGNED_SEQ_NO, 0);
                final Engine.DeleteResult result = initialEngine.delete(delete);
                if (result.getResultType() == Engine.Result.Type.SUCCESS) {
                    assertThat(result.getSeqNo(), equalTo(primarySeqNo + 1));
                    assertThat(initialEngine.getSeqNoStats(-1).getMaxSeqNo(), equalTo(primarySeqNo + 1));
                    indexedIds.remove(id);
                    primarySeqNo++;
                } else {
                    assertThat(result.getSeqNo(), equalTo(UNASSIGNED_SEQ_NO));
                    assertThat(initialEngine.getSeqNoStats(-1).getMaxSeqNo(), equalTo(primarySeqNo));
                }
            } else {
                // index a document
                id = randomFrom(ids);
                ParsedDocument doc = testParsedDocument(id, null, testDocumentWithTextField(), SOURCE, null);
                final Engine.Index index = new Engine.Index(newUid(doc), doc, UNASSIGNED_SEQ_NO, primaryTerm.get(), rarely() ? 100 : Versions.MATCH_ANY, VersionType.INTERNAL, PRIMARY, System.nanoTime(), -1, false, UNASSIGNED_SEQ_NO, 0);
                final Engine.IndexResult result = initialEngine.index(index);
                if (result.getResultType() == Engine.Result.Type.SUCCESS) {
                    assertThat(result.getSeqNo(), equalTo(primarySeqNo + 1));
                    assertThat(initialEngine.getSeqNoStats(-1).getMaxSeqNo(), equalTo(primarySeqNo + 1));
                    indexedIds.add(id);
                    primarySeqNo++;
                } else {
                    assertThat(result.getSeqNo(), equalTo(UNASSIGNED_SEQ_NO));
                    assertThat(initialEngine.getSeqNoStats(-1).getMaxSeqNo(), equalTo(primarySeqNo));
                }
            }
            // to advance persisted local checkpoint
            initialEngine.syncTranslog();
            if (randomInt(10) < 3) {
                // only update rarely as we do it every doc
                replicaLocalCheckpoint = randomIntBetween(Math.toIntExact(replicaLocalCheckpoint), Math.toIntExact(primarySeqNo));
            }
            gcpTracker.updateLocalCheckpoint(primary.allocationId().getId(), initialEngine.getPersistedLocalCheckpoint());
            gcpTracker.updateLocalCheckpoint(replica.allocationId().getId(), replicaLocalCheckpoint);
            if (rarely()) {
                localCheckpoint = primarySeqNo;
                maxSeqNo = primarySeqNo;
                initialEngine.flush(true, true);
            }
        }
        logger.info("localcheckpoint {}, global {}", replicaLocalCheckpoint, primarySeqNo);
        globalCheckpoint = gcpTracker.getGlobalCheckpoint();
        assertEquals(primarySeqNo, initialEngine.getSeqNoStats(-1).getMaxSeqNo());
        assertEquals(primarySeqNo, initialEngine.getPersistedLocalCheckpoint());
        assertThat(globalCheckpoint, equalTo(replicaLocalCheckpoint));
        assertThat(Long.parseLong(initialEngine.commitStats().getUserData().get(SequenceNumbers.LOCAL_CHECKPOINT_KEY)), equalTo(localCheckpoint));
        // to guarantee the global checkpoint is written to the translog checkpoint
        initialEngine.getTranslog().sync();
        assertThat(initialEngine.getTranslog().getLastSyncedGlobalCheckpoint(), equalTo(globalCheckpoint));
        assertThat(Long.parseLong(initialEngine.commitStats().getUserData().get(SequenceNumbers.MAX_SEQ_NO)), equalTo(maxSeqNo));
    } finally {
        IOUtils.close(initialEngine);
    }
    try (InternalEngine recoveringEngine = new InternalEngine(initialEngine.config())) {
        recoveringEngine.recoverFromTranslog(translogHandler, Long.MAX_VALUE);
        assertEquals(primarySeqNo, recoveringEngine.getSeqNoStats(-1).getMaxSeqNo());
        assertThat(Long.parseLong(recoveringEngine.commitStats().getUserData().get(SequenceNumbers.LOCAL_CHECKPOINT_KEY)), equalTo(primarySeqNo));
        assertThat(recoveringEngine.getTranslog().getLastSyncedGlobalCheckpoint(), equalTo(globalCheckpoint));
        assertThat(Long.parseLong(recoveringEngine.commitStats().getUserData().get(SequenceNumbers.MAX_SEQ_NO)), // we have assigned sequence numbers to should be in the commit
        equalTo(primarySeqNo));
        assertThat(recoveringEngine.getProcessedLocalCheckpoint(), equalTo(primarySeqNo));
        assertThat(recoveringEngine.getPersistedLocalCheckpoint(), equalTo(primarySeqNo));
        assertThat(recoveringEngine.getSeqNoStats(-1).getMaxSeqNo(), equalTo(primarySeqNo));
        assertThat(generateNewSeqNo(recoveringEngine), equalTo(primarySeqNo + 1));
    }
}

Example 2 with ReplicationTracker

use of org.opensearch.index.seqno.ReplicationTracker in project OpenSearch by opensearch-project.

the class NoOpEngineTests method testNoopAfterRegularEngine.

public void testNoopAfterRegularEngine() throws IOException {
    int docs = randomIntBetween(1, 10);
    ReplicationTracker tracker = (ReplicationTracker) engine.config().getGlobalCheckpointSupplier();
    ShardRouting routing = TestShardRouting.newShardRouting("test", shardId.id(), "node", null, true, ShardRoutingState.STARTED, allocationId);
    IndexShardRoutingTable table = new IndexShardRoutingTable.Builder(shardId).addShard(routing).build();
    tracker.updateFromMaster(1L, Collections.singleton(allocationId.getId()), table);
    tracker.activatePrimaryMode(SequenceNumbers.NO_OPS_PERFORMED);
    for (int i = 0; i < docs; i++) {
        ParsedDocument doc = testParsedDocument("" + i, null, testDocumentWithTextField(), B_1, null);
        engine.index(indexForDoc(doc));
        tracker.updateLocalCheckpoint(allocationId.getId(), i);
    }
    engine.flush(true, true);
    long localCheckpoint = engine.getPersistedLocalCheckpoint();
    long maxSeqNo = engine.getSeqNoStats(100L).getMaxSeqNo();
    engine.close();
    final NoOpEngine noOpEngine = new NoOpEngine(noOpConfig(INDEX_SETTINGS, store, primaryTranslogDir, tracker));
    assertThat(noOpEngine.getPersistedLocalCheckpoint(), equalTo(localCheckpoint));
    assertThat(noOpEngine.getSeqNoStats(100L).getMaxSeqNo(), equalTo(maxSeqNo));
    try (Engine.IndexCommitRef ref = noOpEngine.acquireLastIndexCommit(false)) {
        try (IndexReader reader = DirectoryReader.open(ref.getIndexCommit())) {
            assertThat(reader.numDocs(), equalTo(docs));
        }
    }
    noOpEngine.close();
}

Example 3 with ReplicationTracker

use of org.opensearch.index.seqno.ReplicationTracker in project OpenSearch by opensearch-project.

the class IndexShard method resetEngineToGlobalCheckpoint.

/**
 * Rollback the current engine to the safe commit, then replay local translog up to the global checkpoint.
 */
void resetEngineToGlobalCheckpoint() throws IOException {
    assert Thread.holdsLock(mutex) == false : "resetting engine under mutex";
    assert getActiveOperationsCount() == OPERATIONS_BLOCKED : "resetting engine without blocking operations; active operations are [" + getActiveOperations() + ']';
    // persist the global checkpoint to disk
    sync();
    final SeqNoStats seqNoStats = seqNoStats();
    final TranslogStats translogStats = translogStats();
    // flush to make sure the latest commit, which will be opened by the read-only engine, includes all operations.
    flush(new FlushRequest().waitIfOngoing(true));
    SetOnce<Engine> newEngineReference = new SetOnce<>();
    final long globalCheckpoint = getLastKnownGlobalCheckpoint();
    assert globalCheckpoint == getLastSyncedGlobalCheckpoint();
    synchronized (engineMutex) {
        verifyNotClosed();
        // we must create both new read-only engine and new read-write engine under engineMutex to ensure snapshotStoreMetadata,
        // acquireXXXCommit and close works.
        final Engine readOnlyEngine = new ReadOnlyEngine(newEngineConfig(replicationTracker), seqNoStats, translogStats, false, Function.identity(), true) {

            @Override
            public IndexCommitRef acquireLastIndexCommit(boolean flushFirst) {
                synchronized (engineMutex) {
                    if (newEngineReference.get() == null) {
                        throw new AlreadyClosedException("engine was closed");
                    }
                    // ignore flushFirst since we flushed above and we do not want to interfere with ongoing translog replay
                    return newEngineReference.get().acquireLastIndexCommit(false);
                }
            }

            @Override
            public IndexCommitRef acquireSafeIndexCommit() {
                synchronized (engineMutex) {
                    if (newEngineReference.get() == null) {
                        throw new AlreadyClosedException("engine was closed");
                    }
                    return newEngineReference.get().acquireSafeIndexCommit();
                }
            }

            @Override
            public void close() throws IOException {
                assert Thread.holdsLock(engineMutex);
                Engine newEngine = newEngineReference.get();
                if (newEngine == currentEngineReference.get()) {
                    // we successfully installed the new engine so do not close it.
                    newEngine = null;
                }
                IOUtils.close(super::close, newEngine);
            }
        };
        IOUtils.close(currentEngineReference.getAndSet(readOnlyEngine));
        newEngineReference.set(engineFactory.newReadWriteEngine(newEngineConfig(replicationTracker)));
        onNewEngine(newEngineReference.get());
    }
    final Engine.TranslogRecoveryRunner translogRunner = (engine, snapshot) -> runTranslogRecovery(engine, snapshot, Engine.Operation.Origin.LOCAL_RESET, () -> {
    // TODO: add a dedicate recovery stats for the reset translog
    });
    newEngineReference.get().recoverFromTranslog(translogRunner, globalCheckpoint);
    newEngineReference.get().refresh("reset_engine");
    synchronized (engineMutex) {
        verifyNotClosed();
        IOUtils.close(currentEngineReference.getAndSet(newEngineReference.get()));
        // We set active because we are now writing operations to the engine; this way,
        // if we go idle after some time and become inactive, we still give sync'd flush a chance to run.
        active.set(true);
    }
    // time elapses after the engine is created above (pulling the config settings) until we set the engine reference, during
    // which settings changes could possibly have happened, so here we forcefully push any config changes to the new engine.
    onSettingsChanged();
}

Example 4 with ReplicationTracker

use of org.opensearch.index.seqno.ReplicationTracker in project OpenSearch by opensearch-project.

the class IndexShard method newEngineConfig.

private EngineConfig newEngineConfig(LongSupplier globalCheckpointSupplier) {
    final Sort indexSort = indexSortSupplier.get();
    final Engine.Warmer warmer = reader -> {
        assert Thread.holdsLock(mutex) == false : "warming engine under mutex";
        assert reader != null;
        if (this.warmer != null) {
            this.warmer.warm(reader);
        }
    };
    return this.engineConfigFactory.newEngineConfig(shardId, threadPool, indexSettings, warmer, store, indexSettings.getMergePolicy(), mapperService != null ? mapperService.indexAnalyzer() : null, similarityService.similarity(mapperService), engineConfigFactory.newCodecServiceOrDefault(indexSettings, mapperService, logger, codecService), shardEventListener, indexCache != null ? indexCache.query() : null, cachingPolicy, translogConfig, IndexingMemoryController.SHARD_INACTIVE_TIME_SETTING.get(indexSettings.getSettings()), Arrays.asList(refreshListeners, refreshPendingLocationListener), Collections.singletonList(new RefreshMetricUpdater(refreshMetric)), indexSort, circuitBreakerService, globalCheckpointSupplier, replicationTracker::getRetentionLeases, () -> getOperationPrimaryTerm(), tombstoneDocSupplier());
}

Example 5 with ReplicationTracker

use of org.opensearch.index.seqno.ReplicationTracker in project OpenSearch by opensearch-project.

the class EngineTestCase method config.

public EngineConfig config(final IndexSettings indexSettings, final Store store, final Path translogPath, final MergePolicy mergePolicy, final ReferenceManager.RefreshListener externalRefreshListener, final ReferenceManager.RefreshListener internalRefreshListener, final Sort indexSort, @Nullable final LongSupplier maybeGlobalCheckpointSupplier, @Nullable final Supplier<RetentionLeases> maybeRetentionLeasesSupplier, final CircuitBreakerService breakerService) {
    final IndexWriterConfig iwc = newIndexWriterConfig();
    final TranslogConfig translogConfig = new TranslogConfig(shardId, translogPath, indexSettings, BigArrays.NON_RECYCLING_INSTANCE);
    final Engine.EventListener eventListener = new Engine.EventListener() {
    };
    // we don't need to notify anybody in this test
    final List<ReferenceManager.RefreshListener> extRefreshListenerList = externalRefreshListener == null ? emptyList() : Collections.singletonList(externalRefreshListener);
    final List<ReferenceManager.RefreshListener> intRefreshListenerList = internalRefreshListener == null ? emptyList() : Collections.singletonList(internalRefreshListener);
    final LongSupplier globalCheckpointSupplier;
    final Supplier<RetentionLeases> retentionLeasesSupplier;
    if (maybeGlobalCheckpointSupplier == null) {
        assert maybeRetentionLeasesSupplier == null;
        final ReplicationTracker replicationTracker = new ReplicationTracker(shardId, allocationId.getId(), indexSettings, randomNonNegativeLong(), SequenceNumbers.NO_OPS_PERFORMED, update -> {
        }, () -> 0L, (leases, listener) -> listener.onResponse(new ReplicationResponse()), () -> SafeCommitInfo.EMPTY);
        globalCheckpointSupplier = replicationTracker;
        retentionLeasesSupplier = replicationTracker::getRetentionLeases;
    } else {
        assert maybeRetentionLeasesSupplier != null;
        globalCheckpointSupplier = maybeGlobalCheckpointSupplier;
        retentionLeasesSupplier = maybeRetentionLeasesSupplier;
    }
    return new EngineConfig(shardId, threadPool, indexSettings, null, store, mergePolicy, iwc.getAnalyzer(), iwc.getSimilarity(), new CodecService(null, logger), eventListener, IndexSearcher.getDefaultQueryCache(), IndexSearcher.getDefaultQueryCachingPolicy(), translogConfig, TimeValue.timeValueMinutes(5), extRefreshListenerList, intRefreshListenerList, indexSort, breakerService, globalCheckpointSupplier, retentionLeasesSupplier, primaryTerm, tombstoneDocSupplier());
}