Search in sources :

Example 11 with SnapshotsInProgress

use of org.elasticsearch.cluster.SnapshotsInProgress in project elasticsearch by elastic.

the class SnapshotShardsService method processIndexShardSnapshots.

/**
     * Checks if any new shards should be snapshotted on this node
     *
     * @param event cluster state changed event
     */
private void processIndexShardSnapshots(ClusterChangedEvent event) {
    SnapshotsInProgress snapshotsInProgress = event.state().custom(SnapshotsInProgress.TYPE);
    Map<Snapshot, SnapshotShards> survivors = new HashMap<>();
    // First, remove snapshots that are no longer there
    for (Map.Entry<Snapshot, SnapshotShards> entry : shardSnapshots.entrySet()) {
        final Snapshot snapshot = entry.getKey();
        if (snapshotsInProgress != null && snapshotsInProgress.snapshot(snapshot) != null) {
            survivors.put(entry.getKey(), entry.getValue());
        } else {
            // state update, which is being processed here
            for (IndexShardSnapshotStatus snapshotStatus : entry.getValue().shards.values()) {
                if (snapshotStatus.stage() == Stage.INIT || snapshotStatus.stage() == Stage.STARTED) {
                    snapshotStatus.abort();
                }
            }
        }
    }
    // For now we will be mostly dealing with a single snapshot at a time but might have multiple simultaneously running
    // snapshots in the future
    Map<Snapshot, Map<ShardId, IndexShardSnapshotStatus>> newSnapshots = new HashMap<>();
    // Now go through all snapshots and update existing or create missing
    final String localNodeId = event.state().nodes().getLocalNodeId();
    final DiscoveryNode masterNode = event.state().nodes().getMasterNode();
    final Map<Snapshot, Map<String, IndexId>> snapshotIndices = new HashMap<>();
    if (snapshotsInProgress != null) {
        for (SnapshotsInProgress.Entry entry : snapshotsInProgress.entries()) {
            snapshotIndices.put(entry.snapshot(), entry.indices().stream().collect(Collectors.toMap(IndexId::getName, Function.identity())));
            if (entry.state() == State.STARTED) {
                Map<ShardId, IndexShardSnapshotStatus> startedShards = new HashMap<>();
                SnapshotShards snapshotShards = shardSnapshots.get(entry.snapshot());
                for (ObjectObjectCursor<ShardId, ShardSnapshotStatus> shard : entry.shards()) {
                    // Add all new shards to start processing on
                    if (localNodeId.equals(shard.value.nodeId())) {
                        if (shard.value.state() == State.INIT && (snapshotShards == null || !snapshotShards.shards.containsKey(shard.key))) {
                            logger.trace("[{}] - Adding shard to the queue", shard.key);
                            startedShards.put(shard.key, new IndexShardSnapshotStatus());
                        }
                    }
                }
                if (!startedShards.isEmpty()) {
                    newSnapshots.put(entry.snapshot(), startedShards);
                    if (snapshotShards != null) {
                        // We already saw this snapshot but we need to add more started shards
                        Map<ShardId, IndexShardSnapshotStatus> shards = new HashMap<>();
                        // Put all shards that were already running on this node
                        shards.putAll(snapshotShards.shards);
                        // Put all newly started shards
                        shards.putAll(startedShards);
                        survivors.put(entry.snapshot(), new SnapshotShards(unmodifiableMap(shards)));
                    } else {
                        // Brand new snapshot that we haven't seen before
                        survivors.put(entry.snapshot(), new SnapshotShards(unmodifiableMap(startedShards)));
                    }
                }
            } else if (entry.state() == State.ABORTED) {
                // Abort all running shards for this snapshot
                SnapshotShards snapshotShards = shardSnapshots.get(entry.snapshot());
                if (snapshotShards != null) {
                    for (ObjectObjectCursor<ShardId, ShardSnapshotStatus> shard : entry.shards()) {
                        IndexShardSnapshotStatus snapshotStatus = snapshotShards.shards.get(shard.key);
                        if (snapshotStatus != null) {
                            switch(snapshotStatus.stage()) {
                                case INIT:
                                case STARTED:
                                    snapshotStatus.abort();
                                    break;
                                case FINALIZE:
                                    logger.debug("[{}] trying to cancel snapshot on shard [{}] that is finalizing, letting it finish", entry.snapshot(), shard.key);
                                    break;
                                case DONE:
                                    logger.debug("[{}] trying to cancel snapshot on the shard [{}] that is already done, updating status on the master", entry.snapshot(), shard.key);
                                    updateIndexShardSnapshotStatus(entry.snapshot(), shard.key, new ShardSnapshotStatus(localNodeId, State.SUCCESS), masterNode);
                                    break;
                                case FAILURE:
                                    logger.debug("[{}] trying to cancel snapshot on the shard [{}] that has already failed, updating status on the master", entry.snapshot(), shard.key);
                                    updateIndexShardSnapshotStatus(entry.snapshot(), shard.key, new ShardSnapshotStatus(localNodeId, State.FAILED, snapshotStatus.failure()), masterNode);
                                    break;
                                default:
                                    throw new IllegalStateException("Unknown snapshot shard stage " + snapshotStatus.stage());
                            }
                        }
                    }
                }
            }
        }
    }
    // Update the list of snapshots that we saw and tried to started
    // If startup of these shards fails later, we don't want to try starting these shards again
    shutdownLock.lock();
    try {
        shardSnapshots = unmodifiableMap(survivors);
        if (shardSnapshots.isEmpty()) {
            // Notify all waiting threads that no more snapshots
            shutdownCondition.signalAll();
        }
    } finally {
        shutdownLock.unlock();
    }
    // We have new shards to starts
    if (newSnapshots.isEmpty() == false) {
        Executor executor = threadPool.executor(ThreadPool.Names.SNAPSHOT);
        for (final Map.Entry<Snapshot, Map<ShardId, IndexShardSnapshotStatus>> entry : newSnapshots.entrySet()) {
            Map<String, IndexId> indicesMap = snapshotIndices.get(entry.getKey());
            assert indicesMap != null;
            for (final Map.Entry<ShardId, IndexShardSnapshotStatus> shardEntry : entry.getValue().entrySet()) {
                final ShardId shardId = shardEntry.getKey();
                try {
                    final IndexShard indexShard = indicesService.indexServiceSafe(shardId.getIndex()).getShardOrNull(shardId.id());
                    final IndexId indexId = indicesMap.get(shardId.getIndexName());
                    assert indexId != null;
                    executor.execute(new AbstractRunnable() {

                        @Override
                        public void doRun() {
                            snapshot(indexShard, entry.getKey(), indexId, shardEntry.getValue());
                            updateIndexShardSnapshotStatus(entry.getKey(), shardId, new ShardSnapshotStatus(localNodeId, State.SUCCESS), masterNode);
                        }

                        @Override
                        public void onFailure(Exception e) {
                            logger.warn((Supplier<?>) () -> new ParameterizedMessage("[{}] [{}] failed to create snapshot", shardId, entry.getKey()), e);
                            updateIndexShardSnapshotStatus(entry.getKey(), shardId, new ShardSnapshotStatus(localNodeId, State.FAILED, ExceptionsHelper.detailedMessage(e)), masterNode);
                        }
                    });
                } catch (Exception e) {
                    updateIndexShardSnapshotStatus(entry.getKey(), shardId, new ShardSnapshotStatus(localNodeId, State.FAILED, ExceptionsHelper.detailedMessage(e)), masterNode);
                }
            }
        }
    }
}
Also used : IndexShardSnapshotStatus(org.elasticsearch.index.snapshots.IndexShardSnapshotStatus) AbstractRunnable(org.elasticsearch.common.util.concurrent.AbstractRunnable) DiscoveryNode(org.elasticsearch.cluster.node.DiscoveryNode) HashMap(java.util.HashMap) ShardId(org.elasticsearch.index.shard.ShardId) Executor(java.util.concurrent.Executor) ClusterStateTaskExecutor(org.elasticsearch.cluster.ClusterStateTaskExecutor) Supplier(org.apache.logging.log4j.util.Supplier) IndexId(org.elasticsearch.repositories.IndexId) IndexShard(org.elasticsearch.index.shard.IndexShard) IndexShardSnapshotFailedException(org.elasticsearch.index.snapshots.IndexShardSnapshotFailedException) SnapshotFailedEngineException(org.elasticsearch.index.engine.SnapshotFailedEngineException) IOException(java.io.IOException) SnapshotsInProgress(org.elasticsearch.cluster.SnapshotsInProgress) ShardSnapshotStatus(org.elasticsearch.cluster.SnapshotsInProgress.ShardSnapshotStatus) IndexShardSnapshotStatus(org.elasticsearch.index.snapshots.IndexShardSnapshotStatus) ParameterizedMessage(org.apache.logging.log4j.message.ParameterizedMessage) ObjectObjectCursor(com.carrotsearch.hppc.cursors.ObjectObjectCursor) Map(java.util.Map) ImmutableOpenMap(org.elasticsearch.common.collect.ImmutableOpenMap) HashMap(java.util.HashMap) Collections.emptyMap(java.util.Collections.emptyMap) Collections.unmodifiableMap(java.util.Collections.unmodifiableMap)

Example 12 with SnapshotsInProgress

use of org.elasticsearch.cluster.SnapshotsInProgress in project elasticsearch by elastic.

the class SnapshotsService method isRepositoryInUse.

/**
     * Checks if a repository is currently in use by one of the snapshots
     *
     * @param clusterState cluster state
     * @param repository   repository id
     * @return true if repository is currently in use by one of the running snapshots
     */
public static boolean isRepositoryInUse(ClusterState clusterState, String repository) {
    SnapshotsInProgress snapshots = clusterState.custom(SnapshotsInProgress.TYPE);
    if (snapshots != null) {
        for (SnapshotsInProgress.Entry snapshot : snapshots.entries()) {
            if (repository.equals(snapshot.snapshot().getRepository())) {
                return true;
            }
        }
    }
    SnapshotDeletionsInProgress deletionsInProgress = clusterState.custom(SnapshotDeletionsInProgress.TYPE);
    if (deletionsInProgress != null) {
        for (SnapshotDeletionsInProgress.Entry entry : deletionsInProgress.getEntries()) {
            if (entry.getSnapshot().getRepository().equals(repository)) {
                return true;
            }
        }
    }
    return false;
}
Also used : SnapshotsInProgress(org.elasticsearch.cluster.SnapshotsInProgress) SnapshotDeletionsInProgress(org.elasticsearch.cluster.SnapshotDeletionsInProgress)

Example 13 with SnapshotsInProgress

use of org.elasticsearch.cluster.SnapshotsInProgress in project elasticsearch by elastic.

the class SnapshotsService method indicesToFailForCloseOrDeletion.

private static Set<Index> indicesToFailForCloseOrDeletion(ClusterState currentState, Set<IndexMetaData> indices) {
    SnapshotsInProgress snapshots = currentState.custom(SnapshotsInProgress.TYPE);
    Set<Index> indicesToFail = null;
    if (snapshots != null) {
        for (final SnapshotsInProgress.Entry entry : snapshots.entries()) {
            if (entry.partial() == false) {
                if (entry.state() == State.INIT) {
                    for (IndexId index : entry.indices()) {
                        IndexMetaData indexMetaData = currentState.metaData().index(index.getName());
                        if (indexMetaData != null && indices.contains(indexMetaData)) {
                            if (indicesToFail == null) {
                                indicesToFail = new HashSet<>();
                            }
                            indicesToFail.add(indexMetaData.getIndex());
                        }
                    }
                } else {
                    for (ObjectObjectCursor<ShardId, SnapshotsInProgress.ShardSnapshotStatus> shard : entry.shards()) {
                        if (!shard.value.state().completed()) {
                            IndexMetaData indexMetaData = currentState.metaData().index(shard.key.getIndex());
                            if (indexMetaData != null && indices.contains(indexMetaData)) {
                                if (indicesToFail == null) {
                                    indicesToFail = new HashSet<>();
                                }
                                indicesToFail.add(shard.key.getIndex());
                            }
                        }
                    }
                }
            }
        }
    }
    return indicesToFail;
}
Also used : ShardId(org.elasticsearch.index.shard.ShardId) IndexId(org.elasticsearch.repositories.IndexId) SnapshotsInProgress(org.elasticsearch.cluster.SnapshotsInProgress) Index(org.elasticsearch.index.Index) ShardSnapshotStatus(org.elasticsearch.cluster.SnapshotsInProgress.ShardSnapshotStatus) IndexShardSnapshotStatus(org.elasticsearch.index.snapshots.IndexShardSnapshotStatus) IndexMetaData(org.elasticsearch.cluster.metadata.IndexMetaData)

Example 14 with SnapshotsInProgress

use of org.elasticsearch.cluster.SnapshotsInProgress in project elasticsearch by elastic.

the class SnapshotsService method removedNodesCleanupNeeded.

private boolean removedNodesCleanupNeeded(ClusterChangedEvent event) {
    SnapshotsInProgress snapshotsInProgress = event.state().custom(SnapshotsInProgress.TYPE);
    if (snapshotsInProgress == null) {
        return false;
    }
    // Check if we just became the master
    boolean newMaster = !event.previousState().nodes().isLocalNodeElectedMaster();
    for (SnapshotsInProgress.Entry snapshot : snapshotsInProgress.entries()) {
        if (newMaster && (snapshot.state() == State.SUCCESS || snapshot.state() == State.INIT)) {
            // We just replaced old master and snapshots in intermediate states needs to be cleaned
            return true;
        }
        for (DiscoveryNode node : event.nodesDelta().removedNodes()) {
            for (ObjectCursor<ShardSnapshotStatus> shardStatus : snapshot.shards().values()) {
                if (!shardStatus.value.state().completed() && node.getId().equals(shardStatus.value.nodeId())) {
                    // At least one shard was running on the removed node - we need to fail it
                    return true;
                }
            }
        }
    }
    return false;
}
Also used : DiscoveryNode(org.elasticsearch.cluster.node.DiscoveryNode) SnapshotsInProgress(org.elasticsearch.cluster.SnapshotsInProgress) ShardSnapshotStatus(org.elasticsearch.cluster.SnapshotsInProgress.ShardSnapshotStatus) IndexShardSnapshotStatus(org.elasticsearch.index.snapshots.IndexShardSnapshotStatus)

Example 15 with SnapshotsInProgress

use of org.elasticsearch.cluster.SnapshotsInProgress in project elasticsearch by elastic.

the class MetaDataDeleteIndexServiceTests method testDeleteSnapshotting.

public void testDeleteSnapshotting() {
    String index = randomAsciiOfLength(5);
    Snapshot snapshot = new Snapshot("doesn't matter", new SnapshotId("snapshot name", "snapshot uuid"));
    SnapshotsInProgress snaps = new SnapshotsInProgress(new SnapshotsInProgress.Entry(snapshot, true, false, SnapshotsInProgress.State.INIT, singletonList(new IndexId(index, "doesn't matter")), System.currentTimeMillis(), (long) randomIntBetween(0, 1000), ImmutableOpenMap.of()));
    ClusterState state = ClusterState.builder(clusterState(index)).putCustom(SnapshotsInProgress.TYPE, snaps).build();
    Exception e = expectThrows(IllegalArgumentException.class, () -> service.deleteIndices(state, singleton(state.metaData().getIndices().get(index).getIndex())));
    assertEquals("Cannot delete indices that are being snapshotted: [[" + index + "]]. Try again after snapshot finishes " + "or cancel the currently running snapshot.", e.getMessage());
}
Also used : Snapshot(org.elasticsearch.snapshots.Snapshot) SnapshotId(org.elasticsearch.snapshots.SnapshotId) IndexId(org.elasticsearch.repositories.IndexId) ClusterState(org.elasticsearch.cluster.ClusterState) SnapshotsInProgress(org.elasticsearch.cluster.SnapshotsInProgress) IndexNotFoundException(org.elasticsearch.index.IndexNotFoundException)

Aggregations

SnapshotsInProgress (org.elasticsearch.cluster.SnapshotsInProgress)16 ShardSnapshotStatus (org.elasticsearch.cluster.SnapshotsInProgress.ShardSnapshotStatus)7 IOException (java.io.IOException)6 ArrayList (java.util.ArrayList)6 ClusterState (org.elasticsearch.cluster.ClusterState)6 IndexShardSnapshotStatus (org.elasticsearch.index.snapshots.IndexShardSnapshotStatus)6 ClusterStateUpdateTask (org.elasticsearch.cluster.ClusterStateUpdateTask)5 ImmutableOpenMap (org.elasticsearch.common.collect.ImmutableOpenMap)5 IndexId (org.elasticsearch.repositories.IndexId)5 CopyOnWriteArrayList (java.util.concurrent.CopyOnWriteArrayList)4 ShardId (org.elasticsearch.index.shard.ShardId)4 RepositoryMissingException (org.elasticsearch.repositories.RepositoryMissingException)4 ObjectObjectCursor (com.carrotsearch.hppc.cursors.ObjectObjectCursor)3 List (java.util.List)3 ParameterizedMessage (org.apache.logging.log4j.message.ParameterizedMessage)3 Supplier (org.apache.logging.log4j.util.Supplier)3 SnapshotDeletionsInProgress (org.elasticsearch.cluster.SnapshotDeletionsInProgress)3 DiscoveryNode (org.elasticsearch.cluster.node.DiscoveryNode)3 Collections.emptyMap (java.util.Collections.emptyMap)2 Collections.unmodifiableMap (java.util.Collections.unmodifiableMap)2