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

Example 11 with RetentionLease

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

the class RecoverySourceHandlerTests method testCancelRecoveryDuringPhase1.

public void testCancelRecoveryDuringPhase1() throws Exception {
    Store store = newStore(createTempDir("source"), false);
    IndexShard shard = mock(IndexShard.class);
    when(shard.store()).thenReturn(store);
    Directory dir = store.directory();
    RandomIndexWriter writer = new RandomIndexWriter(random(), dir, newIndexWriterConfig());
    int numDocs = randomIntBetween(10, 100);
    for (int i = 0; i < numDocs; i++) {
        Document document = new Document();
        document.add(new StringField("id", Integer.toString(i), Field.Store.YES));
        document.add(newField("field", randomUnicodeOfCodepointLengthBetween(1, 10), TextField.TYPE_STORED));
        writer.addDocument(document);
    }
    writer.commit();
    writer.close();
    AtomicBoolean wasCancelled = new AtomicBoolean();
    SetOnce<Runnable> cancelRecovery = new SetOnce<>();
    final TestRecoveryTargetHandler recoveryTarget = new TestRecoveryTargetHandler() {

        @Override
        public void receiveFileInfo(List<String> phase1FileNames, List<Long> phase1FileSizes, List<String> phase1ExistingFileNames, List<Long> phase1ExistingFileSizes, int totalTranslogOps, ActionListener<Void> listener) {
            recoveryExecutor.execute(() -> listener.onResponse(null));
            if (randomBoolean()) {
                wasCancelled.set(true);
                cancelRecovery.get().run();
            }
        }

        @Override
        public void writeFileChunk(StoreFileMetadata md, long position, BytesReference content, boolean lastChunk, int totalTranslogOps, ActionListener<Void> listener) {
            recoveryExecutor.execute(() -> listener.onResponse(null));
            if (rarely()) {
                wasCancelled.set(true);
                cancelRecovery.get().run();
            }
        }

        @Override
        public void cleanFiles(int totalTranslogOps, long globalCheckpoint, Store.MetadataSnapshot sourceMetadata, ActionListener<Void> listener) {
            recoveryExecutor.execute(() -> listener.onResponse(null));
            if (randomBoolean()) {
                wasCancelled.set(true);
                cancelRecovery.get().run();
            }
        }
    };
    final StartRecoveryRequest startRecoveryRequest = getStartRecoveryRequest();
    final RecoverySourceHandler handler = new RecoverySourceHandler(shard, recoveryTarget, threadPool, startRecoveryRequest, between(1, 16), between(1, 4), between(1, 4)) {

        @Override
        void createRetentionLease(long startingSeqNo, ActionListener<RetentionLease> listener) {
            final String leaseId = ReplicationTracker.getPeerRecoveryRetentionLeaseId(startRecoveryRequest.targetNode().getId());
            listener.onResponse(new RetentionLease(leaseId, startingSeqNo, threadPool.absoluteTimeInMillis(), ReplicationTracker.PEER_RECOVERY_RETENTION_LEASE_SOURCE));
        }
    };
    cancelRecovery.set(() -> handler.cancel("test"));
    final StepListener<RecoverySourceHandler.SendFileResult> phase1Listener = new StepListener<>();
    try {
        final CountDownLatch latch = new CountDownLatch(1);
        handler.phase1(DirectoryReader.listCommits(dir).get(0), 0, () -> 0, new LatchedActionListener<>(phase1Listener, latch));
        latch.await();
        phase1Listener.result();
    } catch (Exception e) {
        assertTrue(wasCancelled.get());
        assertNotNull(ExceptionsHelper.unwrap(e, CancellableThreads.ExecutionCancelledException.class));
    }
    store.close();
}

Example 12 with RetentionLease

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

the class RecoverySourceHandler method createRetentionLease.

void createRetentionLease(final long startingSeqNo, ActionListener<RetentionLease> listener) {
    runUnderPrimaryPermit(() -> {
        // Clone the peer recovery retention lease belonging to the source shard. We are retaining history between the the local
        // checkpoint of the safe commit we're creating and this lease's retained seqno with the retention lock, and by cloning an
        // existing lease we (approximately) know that all our peers are also retaining history as requested by the cloned lease. If
        // the recovery now fails before copying enough history over then a subsequent attempt will find this lease, determine it is
        // not enough, and fall back to a file-based recovery.
        // 
        // (approximately) because we do not guarantee to be able to satisfy every lease on every peer.
        logger.trace("cloning primary's retention lease");
        try {
            final StepListener<ReplicationResponse> cloneRetentionLeaseStep = new StepListener<>();
            final RetentionLease clonedLease = shard.cloneLocalPeerRecoveryRetentionLease(request.targetNode().getId(), new ThreadedActionListener<>(logger, shard.getThreadPool(), ThreadPool.Names.GENERIC, cloneRetentionLeaseStep, false));
            logger.trace("cloned primary's retention lease as [{}]", clonedLease);
            cloneRetentionLeaseStep.whenComplete(rr -> listener.onResponse(clonedLease), listener::onFailure);
        } catch (RetentionLeaseNotFoundException e) {
            // recovery as a conservative estimate for the global checkpoint.
            assert shard.indexSettings().getIndexVersionCreated().before(LegacyESVersion.V_7_4_0) || shard.indexSettings().isSoftDeleteEnabled() == false;
            final StepListener<ReplicationResponse> addRetentionLeaseStep = new StepListener<>();
            final long estimatedGlobalCheckpoint = startingSeqNo - 1;
            final RetentionLease newLease = shard.addPeerRecoveryRetentionLease(request.targetNode().getId(), estimatedGlobalCheckpoint, new ThreadedActionListener<>(logger, shard.getThreadPool(), ThreadPool.Names.GENERIC, addRetentionLeaseStep, false));
            addRetentionLeaseStep.whenComplete(rr -> listener.onResponse(newLease), listener::onFailure);
            logger.trace("created retention lease with estimated checkpoint of [{}]", estimatedGlobalCheckpoint);
        }
    }, shardId + " establishing retention lease for [" + request.targetAllocationId() + "]", shard, cancellableThreads, logger);
}

Example 13 with RetentionLease

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

the class RecoverySourceHandler method recoverToTarget.

/**
 * performs the recovery from the local engine to the target
 */
public void recoverToTarget(ActionListener<RecoveryResponse> listener) {
    addListener(listener);
    final Closeable releaseResources = () -> IOUtils.close(resources);
    try {
        cancellableThreads.setOnCancel((reason, beforeCancelEx) -> {
            final RuntimeException e;
            if (shard.state() == IndexShardState.CLOSED) {
                // check if the shard got closed on us
                e = new IndexShardClosedException(shard.shardId(), "shard is closed and recovery was canceled reason [" + reason + "]");
            } else {
                e = new CancellableThreads.ExecutionCancelledException("recovery was canceled reason [" + reason + "]");
            }
            if (beforeCancelEx != null) {
                e.addSuppressed(beforeCancelEx);
            }
            IOUtils.closeWhileHandlingException(releaseResources, () -> future.onFailure(e));
            throw e;
        });
        final Consumer<Exception> onFailure = e -> {
            assert Transports.assertNotTransportThread(RecoverySourceHandler.this + "[onFailure]");
            IOUtils.closeWhileHandlingException(releaseResources, () -> future.onFailure(e));
        };
        final SetOnce<RetentionLease> retentionLeaseRef = new SetOnce<>();
        runUnderPrimaryPermit(() -> {
            final IndexShardRoutingTable routingTable = shard.getReplicationGroup().getRoutingTable();
            ShardRouting targetShardRouting = routingTable.getByAllocationId(request.targetAllocationId());
            if (targetShardRouting == null) {
                logger.debug("delaying recovery of {} as it is not listed as assigned to target node {}", request.shardId(), request.targetNode());
                throw new DelayRecoveryException("source node does not have the shard listed in its state as allocated on the node");
            }
            assert targetShardRouting.initializing() : "expected recovery target to be initializing but was " + targetShardRouting;
            retentionLeaseRef.set(shard.getRetentionLeases().get(ReplicationTracker.getPeerRecoveryRetentionLeaseId(targetShardRouting)));
        }, shardId + " validating recovery target [" + request.targetAllocationId() + "] registered ", shard, cancellableThreads, logger);
        final Closeable retentionLock = shard.acquireHistoryRetentionLock();
        resources.add(retentionLock);
        final long startingSeqNo;
        final boolean isSequenceNumberBasedRecovery = request.startingSeqNo() != SequenceNumbers.UNASSIGNED_SEQ_NO && isTargetSameHistory() && shard.hasCompleteHistoryOperations(PEER_RECOVERY_NAME, request.startingSeqNo()) && ((retentionLeaseRef.get() == null && shard.useRetentionLeasesInPeerRecovery() == false) || (retentionLeaseRef.get() != null && retentionLeaseRef.get().retainingSequenceNumber() <= request.startingSeqNo()));
        if (isSequenceNumberBasedRecovery && retentionLeaseRef.get() != null) {
            // all the history we need is retained by an existing retention lease, so we do not need a separate retention lock
            retentionLock.close();
            logger.trace("history is retained by {}", retentionLeaseRef.get());
        } else {
            // all the history we need is retained by the retention lock, obtained before calling shard.hasCompleteHistoryOperations()
            // and before acquiring the safe commit we'll be using, so we can be certain that all operations after the safe commit's
            // local checkpoint will be retained for the duration of this recovery.
            logger.trace("history is retained by retention lock");
        }
        final StepListener<SendFileResult> sendFileStep = new StepListener<>();
        final StepListener<TimeValue> prepareEngineStep = new StepListener<>();
        final StepListener<SendSnapshotResult> sendSnapshotStep = new StepListener<>();
        final StepListener<Void> finalizeStep = new StepListener<>();
        if (isSequenceNumberBasedRecovery) {
            logger.trace("performing sequence numbers based recovery. starting at [{}]", request.startingSeqNo());
            startingSeqNo = request.startingSeqNo();
            if (retentionLeaseRef.get() == null) {
                createRetentionLease(startingSeqNo, ActionListener.map(sendFileStep, ignored -> SendFileResult.EMPTY));
            } else {
                sendFileStep.onResponse(SendFileResult.EMPTY);
            }
        } else {
            final Engine.IndexCommitRef safeCommitRef;
            try {
                safeCommitRef = acquireSafeCommit(shard);
                resources.add(safeCommitRef);
            } catch (final Exception e) {
                throw new RecoveryEngineException(shard.shardId(), 1, "snapshot failed", e);
            }
            // Try and copy enough operations to the recovering peer so that if it is promoted to primary then it has a chance of being
            // able to recover other replicas using operations-based recoveries. If we are not using retention leases then we
            // conservatively copy all available operations. If we are using retention leases then "enough operations" is just the
            // operations from the local checkpoint of the safe commit onwards, because when using soft deletes the safe commit retains
            // at least as much history as anything else. The safe commit will often contain all the history retained by the current set
            // of retention leases, but this is not guaranteed: an earlier peer recovery from a different primary might have created a
            // retention lease for some history that this primary already discarded, since we discard history when the global checkpoint
            // advances and not when creating a new safe commit. In any case this is a best-effort thing since future recoveries can
            // always fall back to file-based ones, and only really presents a problem if this primary fails before things have settled
            // down.
            startingSeqNo = Long.parseLong(safeCommitRef.getIndexCommit().getUserData().get(SequenceNumbers.LOCAL_CHECKPOINT_KEY)) + 1L;
            logger.trace("performing file-based recovery followed by history replay starting at [{}]", startingSeqNo);
            try {
                final int estimateNumOps = estimateNumberOfHistoryOperations(startingSeqNo);
                final Releasable releaseStore = acquireStore(shard.store());
                resources.add(releaseStore);
                sendFileStep.whenComplete(r -> IOUtils.close(safeCommitRef, releaseStore), e -> {
                    try {
                        IOUtils.close(safeCommitRef, releaseStore);
                    } catch (final IOException ex) {
                        logger.warn("releasing snapshot caused exception", ex);
                    }
                });
                final StepListener<ReplicationResponse> deleteRetentionLeaseStep = new StepListener<>();
                runUnderPrimaryPermit(() -> {
                    try {
                        // If the target previously had a copy of this shard then a file-based recovery might move its global
                        // checkpoint backwards. We must therefore remove any existing retention lease so that we can create a
                        // new one later on in the recovery.
                        shard.removePeerRecoveryRetentionLease(request.targetNode().getId(), new ThreadedActionListener<>(logger, shard.getThreadPool(), ThreadPool.Names.GENERIC, deleteRetentionLeaseStep, false));
                    } catch (RetentionLeaseNotFoundException e) {
                        logger.debug("no peer-recovery retention lease for " + request.targetAllocationId());
                        deleteRetentionLeaseStep.onResponse(null);
                    }
                }, shardId + " removing retention lease for [" + request.targetAllocationId() + "]", shard, cancellableThreads, logger);
                deleteRetentionLeaseStep.whenComplete(ignored -> {
                    assert Transports.assertNotTransportThread(RecoverySourceHandler.this + "[phase1]");
                    phase1(safeCommitRef.getIndexCommit(), startingSeqNo, () -> estimateNumOps, sendFileStep);
                }, onFailure);
            } catch (final Exception e) {
                throw new RecoveryEngineException(shard.shardId(), 1, "sendFileStep failed", e);
            }
        }
        assert startingSeqNo >= 0 : "startingSeqNo must be non negative. got: " + startingSeqNo;
        sendFileStep.whenComplete(r -> {
            assert Transports.assertNotTransportThread(RecoverySourceHandler.this + "[prepareTargetForTranslog]");
            // For a sequence based recovery, the target can keep its local translog
            prepareTargetForTranslog(estimateNumberOfHistoryOperations(startingSeqNo), prepareEngineStep);
        }, onFailure);
        prepareEngineStep.whenComplete(prepareEngineTime -> {
            assert Transports.assertNotTransportThread(RecoverySourceHandler.this + "[phase2]");
            /*
                 * add shard to replication group (shard will receive replication requests from this point on) now that engine is open.
                 * This means that any document indexed into the primary after this will be replicated to this replica as well
                 * make sure to do this before sampling the max sequence number in the next step, to ensure that we send
                 * all documents up to maxSeqNo in phase2.
                 */
            runUnderPrimaryPermit(() -> shard.initiateTracking(request.targetAllocationId()), shardId + " initiating tracking of " + request.targetAllocationId(), shard, cancellableThreads, logger);
            final long endingSeqNo = shard.seqNoStats().getMaxSeqNo();
            if (logger.isTraceEnabled()) {
                logger.trace("snapshot translog for recovery; current size is [{}]", estimateNumberOfHistoryOperations(startingSeqNo));
            }
            final Translog.Snapshot phase2Snapshot = shard.newChangesSnapshot(PEER_RECOVERY_NAME, startingSeqNo, Long.MAX_VALUE, false);
            resources.add(phase2Snapshot);
            retentionLock.close();
            // we have to capture the max_seen_auto_id_timestamp and the max_seq_no_of_updates to make sure that these values
            // are at least as high as the corresponding values on the primary when any of these operations were executed on it.
            final long maxSeenAutoIdTimestamp = shard.getMaxSeenAutoIdTimestamp();
            final long maxSeqNoOfUpdatesOrDeletes = shard.getMaxSeqNoOfUpdatesOrDeletes();
            final RetentionLeases retentionLeases = shard.getRetentionLeases();
            final long mappingVersionOnPrimary = shard.indexSettings().getIndexMetadata().getMappingVersion();
            phase2(startingSeqNo, endingSeqNo, phase2Snapshot, maxSeenAutoIdTimestamp, maxSeqNoOfUpdatesOrDeletes, retentionLeases, mappingVersionOnPrimary, sendSnapshotStep);
        }, onFailure);
        // Recovery target can trim all operations >= startingSeqNo as we have sent all these operations in the phase 2
        final long trimAboveSeqNo = startingSeqNo - 1;
        sendSnapshotStep.whenComplete(r -> finalizeRecovery(r.targetLocalCheckpoint, trimAboveSeqNo, finalizeStep), onFailure);
        finalizeStep.whenComplete(r -> {
            // TODO: return the actual throttle time
            final long phase1ThrottlingWaitTime = 0L;
            final SendSnapshotResult sendSnapshotResult = sendSnapshotStep.result();
            final SendFileResult sendFileResult = sendFileStep.result();
            final RecoveryResponse response = new RecoveryResponse(sendFileResult.phase1FileNames, sendFileResult.phase1FileSizes, sendFileResult.phase1ExistingFileNames, sendFileResult.phase1ExistingFileSizes, sendFileResult.totalSize, sendFileResult.existingTotalSize, sendFileResult.took.millis(), phase1ThrottlingWaitTime, prepareEngineStep.result().millis(), sendSnapshotResult.sentOperations, sendSnapshotResult.tookTime.millis());
            try {
                future.onResponse(response);
            } finally {
                IOUtils.close(resources);
            }
        }, onFailure);
    } catch (Exception e) {
        IOUtils.closeWhileHandlingException(releaseResources, () -> future.onFailure(e));
    }
}

Example 14 with RetentionLease

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

the class ReplicaShardAllocatorTests method testNotCancellingRecoveryIfCurrentRecoveryHasRetentionLease.

public void testNotCancellingRecoveryIfCurrentRecoveryHasRetentionLease() {
    RoutingAllocation allocation = onePrimaryOnNode1And1ReplicaRecovering(yesAllocationDeciders());
    List<RetentionLease> peerRecoveryRetentionLeasesOnPrimary = new ArrayList<>();
    long retainingSeqNo = randomLongBetween(1, Long.MAX_VALUE);
    peerRecoveryRetentionLeasesOnPrimary.add(newRetentionLease(node1, retainingSeqNo));
    peerRecoveryRetentionLeasesOnPrimary.add(newRetentionLease(node2, randomLongBetween(1, retainingSeqNo)));
    if (randomBoolean()) {
        peerRecoveryRetentionLeasesOnPrimary.add(newRetentionLease(node3, randomLongBetween(0, retainingSeqNo)));
    }
    testAllocator.addData(node1, peerRecoveryRetentionLeasesOnPrimary, "MATCH", new StoreFileMetadata("file1", 10, "MATCH_CHECKSUM", MIN_SUPPORTED_LUCENE_VERSION));
    testAllocator.addData(node2, randomSyncId(), new StoreFileMetadata("file1", 10, "MATCH_CHECKSUM", MIN_SUPPORTED_LUCENE_VERSION));
    testAllocator.addData(node3, randomSyncId(), new StoreFileMetadata("file1", 10, "MATCH_CHECKSUM", MIN_SUPPORTED_LUCENE_VERSION));
    testAllocator.processExistingRecoveries(allocation);
    assertThat(allocation.routingNodesChanged(), equalTo(false));
    assertThat(allocation.routingNodes().shardsWithState(ShardRoutingState.UNASSIGNED).size(), equalTo(0));
}

Example 15 with RetentionLease

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

the class ReplicaShardAllocatorTests method testIgnoreRetentionLeaseIfCopyIsEmpty.

public void testIgnoreRetentionLeaseIfCopyIsEmpty() {
    RoutingAllocation allocation = onePrimaryOnNode1And1Replica(yesAllocationDeciders());
    long retainingSeqNo = randomLongBetween(1, Long.MAX_VALUE);
    List<RetentionLease> retentionLeases = new ArrayList<>();
    retentionLeases.add(newRetentionLease(node1, retainingSeqNo));
    retentionLeases.add(newRetentionLease(node2, randomLongBetween(0, retainingSeqNo)));
    if (randomBoolean()) {
        retentionLeases.add(newRetentionLease(node3, randomLongBetween(0, retainingSeqNo)));
    }
    testAllocator.addData(node1, retentionLeases, randomSyncId(), new StoreFileMetadata("file1", 10, "MATCH_CHECKSUM", MIN_SUPPORTED_LUCENE_VERSION));
    // has retention lease but store is empty
    testAllocator.addData(node2, null);
    testAllocator.addData(node3, randomSyncId(), new StoreFileMetadata("file1", 10, "MATCH_CHECKSUM", MIN_SUPPORTED_LUCENE_VERSION));
    allocateAllUnassigned(allocation);
    assertThat(allocation.routingNodes().shardsWithState(ShardRoutingState.INITIALIZING).size(), equalTo(1));
    assertThat(allocation.routingNodes().shardsWithState(ShardRoutingState.INITIALIZING).get(0).currentNodeId(), equalTo(node3.getId()));
}