Examples with ShardGenerations org.opensearch.repositories.ShardGenerations used on opensource projects

Example 1 with ShardGenerations

use of org.opensearch.repositories.ShardGenerations in project OpenSearch by opensearch-project.

the class SnapshotsService method shards.

/**
 * Calculates the assignment of shards to data nodes for a new snapshot based on the given cluster state and the
 * indices that should be included in the snapshot.
 *
 * @param indices             Indices to snapshot
 * @param useShardGenerations whether to write {@link ShardGenerations} during the snapshot
 * @return list of shard to be included into current snapshot
 */
private static ImmutableOpenMap<ShardId, SnapshotsInProgress.ShardSnapshotStatus> shards(SnapshotsInProgress snapshotsInProgress, @Nullable SnapshotDeletionsInProgress deletionsInProgress, Metadata metadata, RoutingTable routingTable, List<IndexId> indices, boolean useShardGenerations, RepositoryData repositoryData, String repoName) {
    ImmutableOpenMap.Builder<ShardId, SnapshotsInProgress.ShardSnapshotStatus> builder = ImmutableOpenMap.builder();
    final ShardGenerations shardGenerations = repositoryData.shardGenerations();
    final InFlightShardSnapshotStates inFlightShardStates = InFlightShardSnapshotStates.forRepo(repoName, snapshotsInProgress.entries());
    final boolean readyToExecute = deletionsInProgress == null || deletionsInProgress.getEntries().stream().noneMatch(entry -> entry.repository().equals(repoName) && entry.state() == SnapshotDeletionsInProgress.State.STARTED);
    for (IndexId index : indices) {
        final String indexName = index.getName();
        final boolean isNewIndex = repositoryData.getIndices().containsKey(indexName) == false;
        IndexMetadata indexMetadata = metadata.index(indexName);
        if (indexMetadata == null) {
            // The index was deleted before we managed to start the snapshot - mark it as missing.
            builder.put(new ShardId(indexName, IndexMetadata.INDEX_UUID_NA_VALUE, 0), ShardSnapshotStatus.MISSING);
        } else {
            final IndexRoutingTable indexRoutingTable = routingTable.index(indexName);
            for (int i = 0; i < indexMetadata.getNumberOfShards(); i++) {
                final ShardId shardId = indexRoutingTable.shard(i).shardId();
                final String shardRepoGeneration;
                if (useShardGenerations) {
                    final String inFlightGeneration = inFlightShardStates.generationForShard(index, shardId.id(), shardGenerations);
                    if (inFlightGeneration == null && isNewIndex) {
                        assert shardGenerations.getShardGen(index, shardId.getId()) == null : "Found shard generation for new index [" + index + "]";
                        shardRepoGeneration = ShardGenerations.NEW_SHARD_GEN;
                    } else {
                        shardRepoGeneration = inFlightGeneration;
                    }
                } else {
                    shardRepoGeneration = null;
                }
                final ShardSnapshotStatus shardSnapshotStatus;
                if (indexRoutingTable == null) {
                    shardSnapshotStatus = new SnapshotsInProgress.ShardSnapshotStatus(null, ShardState.MISSING, "missing routing table", shardRepoGeneration);
                } else {
                    ShardRouting primary = indexRoutingTable.shard(i).primaryShard();
                    if (readyToExecute == false || inFlightShardStates.isActive(indexName, i)) {
                        shardSnapshotStatus = ShardSnapshotStatus.UNASSIGNED_QUEUED;
                    } else if (primary == null || !primary.assignedToNode()) {
                        shardSnapshotStatus = new ShardSnapshotStatus(null, ShardState.MISSING, "primary shard is not allocated", shardRepoGeneration);
                    } else if (primary.relocating() || primary.initializing()) {
                        shardSnapshotStatus = new ShardSnapshotStatus(primary.currentNodeId(), ShardState.WAITING, shardRepoGeneration);
                    } else if (!primary.started()) {
                        shardSnapshotStatus = new ShardSnapshotStatus(primary.currentNodeId(), ShardState.MISSING, "primary shard hasn't been started yet", shardRepoGeneration);
                    } else {
                        shardSnapshotStatus = new ShardSnapshotStatus(primary.currentNodeId(), shardRepoGeneration);
                    }
                }
                builder.put(shardId, shardSnapshotStatus);
            }
        }
    }
    return builder.build();
}

Example 2 with ShardGenerations

use of org.opensearch.repositories.ShardGenerations in project OpenSearch by opensearch-project.

the class SnapshotsService method buildGenerations.

private static ShardGenerations buildGenerations(SnapshotsInProgress.Entry snapshot, Metadata metadata) {
    ShardGenerations.Builder builder = ShardGenerations.builder();
    final Map<String, IndexId> indexLookup = new HashMap<>();
    snapshot.indices().forEach(idx -> indexLookup.put(idx.getName(), idx));
    if (snapshot.isClone()) {
        snapshot.clones().forEach(c -> {
            final IndexId indexId = indexLookup.get(c.key.indexName());
            builder.put(indexId, c.key.shardId(), c.value.generation());
        });
    } else {
        snapshot.shards().forEach(c -> {
            if (metadata.index(c.key.getIndex()) == null) {
                assert snapshot.partial() : "Index [" + c.key.getIndex() + "] was deleted during a snapshot but snapshot was not partial.";
                return;
            }
            final IndexId indexId = indexLookup.get(c.key.getIndexName());
            if (indexId != null) {
                builder.put(indexId, c.key.id(), c.value.generation());
            }
        });
    }
    return builder.build();
}

Example 3 with ShardGenerations

use of org.opensearch.repositories.ShardGenerations in project OpenSearch by opensearch-project.

the class SnapshotsService method startCloning.

/**
 * Determine the number of shards in each index of a clone operation and update the cluster state accordingly.
 *
 * @param repository     repository to run operation on
 * @param cloneEntry     clone operation in the cluster state
 */
private void startCloning(Repository repository, SnapshotsInProgress.Entry cloneEntry) {
    final List<IndexId> indices = cloneEntry.indices();
    final SnapshotId sourceSnapshot = cloneEntry.source();
    final Snapshot targetSnapshot = cloneEntry.snapshot();
    final Executor executor = threadPool.executor(ThreadPool.Names.SNAPSHOT);
    // Exception handler for IO exceptions with loading index and repo metadata
    final Consumer<Exception> onFailure = e -> {
        initializingClones.remove(targetSnapshot);
        logger.info(() -> new ParameterizedMessage("Failed to start snapshot clone [{}]", cloneEntry), e);
        removeFailedSnapshotFromClusterState(targetSnapshot, e, null, null);
    };
    // 1. step, load SnapshotInfo to make sure that source snapshot was successful for the indices we want to clone
    // TODO: we could skip this step for snapshots with state SUCCESS
    final StepListener<SnapshotInfo> snapshotInfoListener = new StepListener<>();
    executor.execute(ActionRunnable.supply(snapshotInfoListener, () -> repository.getSnapshotInfo(sourceSnapshot)));
    final StepListener<Collection<Tuple<IndexId, Integer>>> allShardCountsListener = new StepListener<>();
    final GroupedActionListener<Tuple<IndexId, Integer>> shardCountListener = new GroupedActionListener<>(allShardCountsListener, indices.size());
    snapshotInfoListener.whenComplete(snapshotInfo -> {
        for (IndexId indexId : indices) {
            if (RestoreService.failed(snapshotInfo, indexId.getName())) {
                throw new SnapshotException(targetSnapshot, "Can't clone index [" + indexId + "] because its snapshot was not successful.");
            }
        }
        // 2. step, load the number of shards we have in each index to be cloned from the index metadata.
        repository.getRepositoryData(ActionListener.wrap(repositoryData -> {
            for (IndexId index : indices) {
                executor.execute(ActionRunnable.supply(shardCountListener, () -> {
                    final IndexMetadata metadata = repository.getSnapshotIndexMetaData(repositoryData, sourceSnapshot, index);
                    return Tuple.tuple(index, metadata.getNumberOfShards());
                }));
            }
        }, onFailure));
    }, onFailure);
    // 3. step, we have all the shard counts, now update the cluster state to have clone jobs in the snap entry
    allShardCountsListener.whenComplete(counts -> repository.executeConsistentStateUpdate(repoData -> new ClusterStateUpdateTask() {

        private SnapshotsInProgress.Entry updatedEntry;

        @Override
        public ClusterState execute(ClusterState currentState) {
            final SnapshotsInProgress snapshotsInProgress = currentState.custom(SnapshotsInProgress.TYPE, SnapshotsInProgress.EMPTY);
            final List<SnapshotsInProgress.Entry> updatedEntries = new ArrayList<>(snapshotsInProgress.entries());
            boolean changed = false;
            final String localNodeId = currentState.nodes().getLocalNodeId();
            final String repoName = cloneEntry.repository();
            final ShardGenerations shardGenerations = repoData.shardGenerations();
            for (int i = 0; i < updatedEntries.size(); i++) {
                if (cloneEntry.snapshot().equals(updatedEntries.get(i).snapshot())) {
                    final ImmutableOpenMap.Builder<RepositoryShardId, ShardSnapshotStatus> clonesBuilder = ImmutableOpenMap.builder();
                    final InFlightShardSnapshotStates inFlightShardStates = InFlightShardSnapshotStates.forRepo(repoName, snapshotsInProgress.entries());
                    for (Tuple<IndexId, Integer> count : counts) {
                        for (int shardId = 0; shardId < count.v2(); shardId++) {
                            final RepositoryShardId repoShardId = new RepositoryShardId(count.v1(), shardId);
                            final String indexName = repoShardId.indexName();
                            if (inFlightShardStates.isActive(indexName, shardId)) {
                                clonesBuilder.put(repoShardId, ShardSnapshotStatus.UNASSIGNED_QUEUED);
                            } else {
                                clonesBuilder.put(repoShardId, new ShardSnapshotStatus(localNodeId, inFlightShardStates.generationForShard(repoShardId.index(), shardId, shardGenerations)));
                            }
                        }
                    }
                    updatedEntry = cloneEntry.withClones(clonesBuilder.build());
                    updatedEntries.set(i, updatedEntry);
                    changed = true;
                    break;
                }
            }
            return updateWithSnapshots(currentState, changed ? SnapshotsInProgress.of(updatedEntries) : null, null);
        }

        @Override
        public void onFailure(String source, Exception e) {
            initializingClones.remove(targetSnapshot);
            logger.info(() -> new ParameterizedMessage("Failed to start snapshot clone [{}]", cloneEntry), e);
            failAllListenersOnMasterFailOver(e);
        }

        @Override
        public void clusterStateProcessed(String source, ClusterState oldState, ClusterState newState) {
            initializingClones.remove(targetSnapshot);
            if (updatedEntry != null) {
                final Snapshot target = updatedEntry.snapshot();
                final SnapshotId sourceSnapshot = updatedEntry.source();
                for (ObjectObjectCursor<RepositoryShardId, ShardSnapshotStatus> indexClone : updatedEntry.clones()) {
                    final ShardSnapshotStatus shardStatusBefore = indexClone.value;
                    if (shardStatusBefore.state() != ShardState.INIT) {
                        continue;
                    }
                    final RepositoryShardId repoShardId = indexClone.key;
                    runReadyClone(target, sourceSnapshot, shardStatusBefore, repoShardId, repository);
                }
            } else {
                // Extremely unlikely corner case of master failing over between between starting the clone and
                // starting shard clones.
                logger.warn("Did not find expected entry [{}] in the cluster state", cloneEntry);
            }
        }
    }, "start snapshot clone", onFailure), onFailure);
}

Example 4 with ShardGenerations

use of org.opensearch.repositories.ShardGenerations in project OpenSearch by opensearch-project.

the class SnapshotsService method finalizeSnapshotEntry.

private void finalizeSnapshotEntry(SnapshotsInProgress.Entry entry, Metadata metadata, RepositoryData repositoryData) {
    assert currentlyFinalizing.contains(entry.repository());
    try {
        final String failure = entry.failure();
        final Snapshot snapshot = entry.snapshot();
        logger.trace("[{}] finalizing snapshot in repository, state: [{}], failure[{}]", snapshot, entry.state(), failure);
        ArrayList<SnapshotShardFailure> shardFailures = new ArrayList<>();
        for (ObjectObjectCursor<ShardId, ShardSnapshotStatus> shardStatus : entry.shards()) {
            ShardId shardId = shardStatus.key;
            ShardSnapshotStatus status = shardStatus.value;
            final ShardState state = status.state();
            if (state.failed()) {
                shardFailures.add(new SnapshotShardFailure(status.nodeId(), shardId, status.reason()));
            } else if (state.completed() == false) {
                shardFailures.add(new SnapshotShardFailure(status.nodeId(), shardId, "skipped"));
            } else {
                assert state == ShardState.SUCCESS;
            }
        }
        final ShardGenerations shardGenerations = buildGenerations(entry, metadata);
        final String repository = snapshot.getRepository();
        final SnapshotInfo snapshotInfo = new SnapshotInfo(snapshot.getSnapshotId(), shardGenerations.indices().stream().map(IndexId::getName).collect(Collectors.toList()), entry.dataStreams(), entry.startTime(), failure, threadPool.absoluteTimeInMillis(), entry.partial() ? shardGenerations.totalShards() : entry.shards().size(), shardFailures, entry.includeGlobalState(), entry.userMetadata());
        final StepListener<Metadata> metadataListener = new StepListener<>();
        final Repository repo = repositoriesService.repository(snapshot.getRepository());
        if (entry.isClone()) {
            threadPool.executor(ThreadPool.Names.SNAPSHOT).execute(ActionRunnable.supply(metadataListener, () -> {
                final Metadata.Builder metaBuilder = Metadata.builder(repo.getSnapshotGlobalMetadata(entry.source()));
                for (IndexId index : entry.indices()) {
                    metaBuilder.put(repo.getSnapshotIndexMetaData(repositoryData, entry.source(), index), false);
                }
                return metaBuilder.build();
            }));
        } else {
            metadataListener.onResponse(metadata);
        }
        metadataListener.whenComplete(meta -> repo.finalizeSnapshot(shardGenerations, repositoryData.getGenId(), metadataForSnapshot(entry, meta), snapshotInfo, entry.version(), state -> stateWithoutSnapshot(state, snapshot), ActionListener.wrap(newRepoData -> {
            completeListenersIgnoringException(endAndGetListenersToResolve(snapshot), Tuple.tuple(newRepoData, snapshotInfo));
            logger.info("snapshot [{}] completed with state [{}]", snapshot, snapshotInfo.state());
            runNextQueuedOperation(newRepoData, repository, true);
        }, e -> handleFinalizationFailure(e, entry, repositoryData))), e -> handleFinalizationFailure(e, entry, repositoryData));
    } catch (Exception e) {
        assert false : new AssertionError(e);
        handleFinalizationFailure(e, entry, repositoryData);
    }
}

Example 5 with ShardGenerations

use of org.opensearch.repositories.ShardGenerations in project OpenSearch by opensearch-project.

the class BlobStoreRepository method finalizeSnapshot.

@Override
public void finalizeSnapshot(final ShardGenerations shardGenerations, final long repositoryStateId, final Metadata clusterMetadata, SnapshotInfo snapshotInfo, Version repositoryMetaVersion, Function<ClusterState, ClusterState> stateTransformer, final ActionListener<RepositoryData> listener) {
    assert repositoryStateId > RepositoryData.UNKNOWN_REPO_GEN : "Must finalize based on a valid repository generation but received [" + repositoryStateId + "]";
    final Collection<IndexId> indices = shardGenerations.indices();
    final SnapshotId snapshotId = snapshotInfo.snapshotId();
    // Once we are done writing the updated index-N blob we remove the now unreferenced index-${uuid} blobs in each shard
    // directory if all nodes are at least at version SnapshotsService#SHARD_GEN_IN_REPO_DATA_VERSION
    // If there are older version nodes in the cluster, we don't need to run this cleanup as it will have already happened
    // when writing the index-${N} to each shard directory.
    final boolean writeShardGens = SnapshotsService.useShardGenerations(repositoryMetaVersion);
    final Consumer<Exception> onUpdateFailure = e -> listener.onFailure(new SnapshotException(metadata.name(), snapshotId, "failed to update snapshot in repository", e));
    final Executor executor = threadPool.executor(ThreadPool.Names.SNAPSHOT);
    final boolean writeIndexGens = SnapshotsService.useIndexGenerations(repositoryMetaVersion);
    final StepListener<RepositoryData> repoDataListener = new StepListener<>();
    getRepositoryData(repoDataListener);
    repoDataListener.whenComplete(existingRepositoryData -> {
        final Map<IndexId, String> indexMetas;
        final Map<String, String> indexMetaIdentifiers;
        if (writeIndexGens) {
            indexMetaIdentifiers = ConcurrentCollections.newConcurrentMap();
            indexMetas = ConcurrentCollections.newConcurrentMap();
        } else {
            indexMetas = null;
            indexMetaIdentifiers = null;
        }
        final ActionListener<Void> allMetaListener = new GroupedActionListener<>(ActionListener.wrap(v -> {
            final RepositoryData updatedRepositoryData = existingRepositoryData.addSnapshot(snapshotId, snapshotInfo.state(), Version.CURRENT, shardGenerations, indexMetas, indexMetaIdentifiers);
            writeIndexGen(updatedRepositoryData, repositoryStateId, repositoryMetaVersion, stateTransformer, ActionListener.wrap(newRepoData -> {
                if (writeShardGens) {
                    cleanupOldShardGens(existingRepositoryData, updatedRepositoryData);
                }
                listener.onResponse(newRepoData);
            }, onUpdateFailure));
        }, onUpdateFailure), 2 + indices.size());
        // We ignore all FileAlreadyExistsException when writing metadata since otherwise a master failover while in this method will
        // mean that no snap-${uuid}.dat blob is ever written for this snapshot. This is safe because any updated version of the
        // index or global metadata will be compatible with the segments written in this snapshot as well.
        // Failing on an already existing index-${repoGeneration} below ensures that the index.latest blob is not updated in a way
        // that decrements the generation it points at
        // Write Global MetaData
        executor.execute(ActionRunnable.run(allMetaListener, () -> GLOBAL_METADATA_FORMAT.write(clusterMetadata, blobContainer(), snapshotId.getUUID(), compress)));
        // write the index metadata for each index in the snapshot
        for (IndexId index : indices) {
            executor.execute(ActionRunnable.run(allMetaListener, () -> {
                final IndexMetadata indexMetaData = clusterMetadata.index(index.getName());
                if (writeIndexGens) {
                    final String identifiers = IndexMetaDataGenerations.buildUniqueIdentifier(indexMetaData);
                    String metaUUID = existingRepositoryData.indexMetaDataGenerations().getIndexMetaBlobId(identifiers);
                    if (metaUUID == null) {
                        // We don't yet have this version of the metadata so we write it
                        metaUUID = UUIDs.base64UUID();
                        INDEX_METADATA_FORMAT.write(indexMetaData, indexContainer(index), metaUUID, compress);
                        indexMetaIdentifiers.put(identifiers, metaUUID);
                    }
                    indexMetas.put(index, identifiers);
                } else {
                    INDEX_METADATA_FORMAT.write(clusterMetadata.index(index.getName()), indexContainer(index), snapshotId.getUUID(), compress);
                }
            }));
        }
        executor.execute(ActionRunnable.run(allMetaListener, () -> SNAPSHOT_FORMAT.write(snapshotInfo, blobContainer(), snapshotId.getUUID(), compress)));
    }, onUpdateFailure);
}