Examples with RoutingTable org.elasticsearch.cluster.routing.RoutingTable used on opensource projects

Example 31 with RoutingTable

use of org.elasticsearch.cluster.routing.RoutingTable in project elasticsearch by elastic.

the class AllocationBenchmark method setUp.

@Setup
public void setUp() throws Exception {
    final String[] params = indicesShardsReplicasNodes.split("\\|");
    int numIndices = toInt(params[0]);
    int numShards = toInt(params[1]);
    int numReplicas = toInt(params[2]);
    int numNodes = toInt(params[3]);
    strategy = Allocators.createAllocationService(Settings.builder().put("cluster.routing.allocation.awareness.attributes", "tag").build());
    MetaData.Builder mb = MetaData.builder();
    for (int i = 1; i <= numIndices; i++) {
        mb.put(IndexMetaData.builder("test_" + i).settings(Settings.builder().put("index.version.created", Version.CURRENT)).numberOfShards(numShards).numberOfReplicas(numReplicas));
    }
    MetaData metaData = mb.build();
    RoutingTable.Builder rb = RoutingTable.builder();
    for (int i = 1; i <= numIndices; i++) {
        rb.addAsNew(metaData.index("test_" + i));
    }
    RoutingTable routingTable = rb.build();
    DiscoveryNodes.Builder nb = DiscoveryNodes.builder();
    for (int i = 1; i <= numNodes; i++) {
        nb.add(Allocators.newNode("node" + i, Collections.singletonMap("tag", "tag_" + (i % numTags))));
    }
    initialClusterState = ClusterState.builder(ClusterName.CLUSTER_NAME_SETTING.getDefault(Settings.EMPTY)).metaData(metaData).routingTable(routingTable).nodes(nb).build();
}

Example 32 with RoutingTable

use of org.elasticsearch.cluster.routing.RoutingTable in project elasticsearch by elastic.

the class TransportIndicesShardStoresAction method masterOperation.

@Override
protected void masterOperation(IndicesShardStoresRequest request, ClusterState state, ActionListener<IndicesShardStoresResponse> listener) {
    final RoutingTable routingTables = state.routingTable();
    final RoutingNodes routingNodes = state.getRoutingNodes();
    final String[] concreteIndices = indexNameExpressionResolver.concreteIndexNames(state, request);
    final Set<ShardId> shardIdsToFetch = new HashSet<>();
    logger.trace("using cluster state version [{}] to determine shards", state.version());
    // collect relevant shard ids of the requested indices for fetching store infos
    for (String index : concreteIndices) {
        IndexRoutingTable indexShardRoutingTables = routingTables.index(index);
        if (indexShardRoutingTables == null) {
            continue;
        }
        for (IndexShardRoutingTable routing : indexShardRoutingTables) {
            final int shardId = routing.shardId().id();
            ClusterShardHealth shardHealth = new ClusterShardHealth(shardId, routing);
            if (request.shardStatuses().contains(shardHealth.getStatus())) {
                shardIdsToFetch.add(routing.shardId());
            }
        }
    }
    // async fetch store infos from all the nodes
    // NOTE: instead of fetching shard store info one by one from every node (nShards * nNodes requests)
    // we could fetch all shard store info from every node once (nNodes requests)
    // we have to implement a TransportNodesAction instead of using TransportNodesListGatewayStartedShards
    // for fetching shard stores info, that operates on a list of shards instead of a single shard
    new AsyncShardStoresInfoFetches(state.nodes(), routingNodes, shardIdsToFetch, listener).start();
}

Example 33 with RoutingTable

use of org.elasticsearch.cluster.routing.RoutingTable in project elasticsearch by elastic.

the class MetaDataDeleteIndexService method deleteIndices.

/**
     * Delete some indices from the cluster state.
     */
public ClusterState deleteIndices(ClusterState currentState, Set<Index> indices) {
    final MetaData meta = currentState.metaData();
    final Set<IndexMetaData> metaDatas = indices.stream().map(i -> meta.getIndexSafe(i)).collect(toSet());
    // Check if index deletion conflicts with any running snapshots
    SnapshotsService.checkIndexDeletion(currentState, metaDatas);
    RoutingTable.Builder routingTableBuilder = RoutingTable.builder(currentState.routingTable());
    MetaData.Builder metaDataBuilder = MetaData.builder(meta);
    ClusterBlocks.Builder clusterBlocksBuilder = ClusterBlocks.builder().blocks(currentState.blocks());
    final IndexGraveyard.Builder graveyardBuilder = IndexGraveyard.builder(metaDataBuilder.indexGraveyard());
    final int previousGraveyardSize = graveyardBuilder.tombstones().size();
    for (final Index index : indices) {
        String indexName = index.getName();
        logger.info("{} deleting index", index);
        routingTableBuilder.remove(indexName);
        clusterBlocksBuilder.removeIndexBlocks(indexName);
        metaDataBuilder.remove(indexName);
    }
    // add tombstones to the cluster state for each deleted index
    final IndexGraveyard currentGraveyard = graveyardBuilder.addTombstones(indices).build(settings);
    // the new graveyard set on the metadata
    metaDataBuilder.indexGraveyard(currentGraveyard);
    logger.trace("{} tombstones purged from the cluster state. Previous tombstone size: {}. Current tombstone size: {}.", graveyardBuilder.getNumPurged(), previousGraveyardSize, currentGraveyard.getTombstones().size());
    MetaData newMetaData = metaDataBuilder.build();
    ClusterBlocks blocks = clusterBlocksBuilder.build();
    // update snapshot restore entries
    ImmutableOpenMap<String, ClusterState.Custom> customs = currentState.getCustoms();
    final RestoreInProgress restoreInProgress = currentState.custom(RestoreInProgress.TYPE);
    if (restoreInProgress != null) {
        RestoreInProgress updatedRestoreInProgress = RestoreService.updateRestoreStateWithDeletedIndices(restoreInProgress, indices);
        if (updatedRestoreInProgress != restoreInProgress) {
            ImmutableOpenMap.Builder<String, ClusterState.Custom> builder = ImmutableOpenMap.builder(customs);
            builder.put(RestoreInProgress.TYPE, updatedRestoreInProgress);
            customs = builder.build();
        }
    }
    return allocationService.reroute(ClusterState.builder(currentState).routingTable(routingTableBuilder.build()).metaData(newMetaData).blocks(blocks).customs(customs).build(), "deleted indices [" + indices + "]");
}

Example 34 with RoutingTable

use of org.elasticsearch.cluster.routing.RoutingTable in project elasticsearch by elastic.

the class IndexMetaDataUpdater method updateInSyncAllocations.

/**
     * Updates in-sync allocations with routing changes that were made to the routing table.
     */
private IndexMetaData.Builder updateInSyncAllocations(RoutingTable newRoutingTable, IndexMetaData oldIndexMetaData, IndexMetaData.Builder indexMetaDataBuilder, ShardId shardId, Updates updates) {
    assert Sets.haveEmptyIntersection(updates.addedAllocationIds, updates.removedAllocationIds) : "allocation ids cannot be both added and removed in the same allocation round, added ids: " + updates.addedAllocationIds + ", removed ids: " + updates.removedAllocationIds;
    Set<String> oldInSyncAllocationIds = oldIndexMetaData.inSyncAllocationIds(shardId.id());
    // check if we have been force-initializing an empty primary or a stale primary
    if (updates.initializedPrimary != null && oldInSyncAllocationIds.isEmpty() == false && oldInSyncAllocationIds.contains(updates.initializedPrimary.allocationId().getId()) == false) {
        // we're not reusing an existing in-sync allocation id to initialize a primary, which means that we're either force-allocating
        // an empty or a stale primary (see AllocateEmptyPrimaryAllocationCommand or AllocateStalePrimaryAllocationCommand).
        RecoverySource.Type recoverySourceType = updates.initializedPrimary.recoverySource().getType();
        boolean emptyPrimary = recoverySourceType == RecoverySource.Type.EMPTY_STORE;
        assert updates.addedAllocationIds.isEmpty() : (emptyPrimary ? "empty" : "stale") + " primary is not force-initialized in same allocation round where shards are started";
        if (indexMetaDataBuilder == null) {
            indexMetaDataBuilder = IndexMetaData.builder(oldIndexMetaData);
        }
        if (emptyPrimary) {
            // forcing an empty primary resets the in-sync allocations to the empty set (ShardRouting.allocatedPostIndexCreate)
            indexMetaDataBuilder.putInSyncAllocationIds(shardId.id(), Collections.emptySet());
        } else {
            // forcing a stale primary resets the in-sync allocations to the singleton set with the stale id
            indexMetaDataBuilder.putInSyncAllocationIds(shardId.id(), Collections.singleton(updates.initializedPrimary.allocationId().getId()));
        }
    } else {
        // standard path for updating in-sync ids
        Set<String> inSyncAllocationIds = new HashSet<>(oldInSyncAllocationIds);
        inSyncAllocationIds.addAll(updates.addedAllocationIds);
        inSyncAllocationIds.removeAll(updates.removedAllocationIds);
        // Prevent set of inSyncAllocationIds to grow unboundedly. This can happen for example if we don't write to a primary
        // but repeatedly shut down nodes that have active replicas.
        // We use number_of_replicas + 1 (= possible active shard copies) to bound the inSyncAllocationIds set
        // Only trim the set of allocation ids when it grows, otherwise we might trim too eagerly when the number
        // of replicas was decreased while shards were unassigned.
        // +1 for the primary
        int maxActiveShards = oldIndexMetaData.getNumberOfReplicas() + 1;
        IndexShardRoutingTable newShardRoutingTable = newRoutingTable.shardRoutingTable(shardId);
        if (inSyncAllocationIds.size() > oldInSyncAllocationIds.size() && inSyncAllocationIds.size() > maxActiveShards) {
            // trim entries that have no corresponding shard routing in the cluster state (i.e. trim unavailable copies)
            List<ShardRouting> assignedShards = newShardRoutingTable.assignedShards();
            assert assignedShards.size() <= maxActiveShards : "cannot have more assigned shards " + assignedShards + " than maximum possible active shards " + maxActiveShards;
            Set<String> assignedAllocations = assignedShards.stream().map(s -> s.allocationId().getId()).collect(Collectors.toSet());
            inSyncAllocationIds = inSyncAllocationIds.stream().sorted(// values with routing entries first
            Comparator.comparing(assignedAllocations::contains).reversed()).limit(maxActiveShards).collect(Collectors.toSet());
        }
        // in-sync set, this could create an empty primary on the next allocation.
        if (newShardRoutingTable.activeShards().isEmpty() && updates.firstFailedPrimary != null) {
            // add back allocation id of failed primary
            inSyncAllocationIds.add(updates.firstFailedPrimary.allocationId().getId());
        }
        assert inSyncAllocationIds.isEmpty() == false || oldInSyncAllocationIds.isEmpty() : "in-sync allocations cannot become empty after they have been non-empty: " + oldInSyncAllocationIds;
        // be extra safe here and only update in-sync set if it is non-empty
        if (inSyncAllocationIds.isEmpty() == false) {
            if (indexMetaDataBuilder == null) {
                indexMetaDataBuilder = IndexMetaData.builder(oldIndexMetaData);
            }
            indexMetaDataBuilder.putInSyncAllocationIds(shardId.id(), inSyncAllocationIds);
        }
    }
    return indexMetaDataBuilder;
}

Example 35 with RoutingTable

use of org.elasticsearch.cluster.routing.RoutingTable in project elasticsearch by elastic.

the class IndexMetaDataUpdater method removeStaleIdsWithoutRoutings.

/**
     * Removes allocation ids from the in-sync set for shard copies for which there is no routing entries in the routing table.
     * This method is called in AllocationService before any changes to the routing table are made.
     */
public static ClusterState removeStaleIdsWithoutRoutings(ClusterState clusterState, List<StaleShard> staleShards) {
    MetaData oldMetaData = clusterState.metaData();
    RoutingTable oldRoutingTable = clusterState.routingTable();
    MetaData.Builder metaDataBuilder = null;
    // group staleShards entries by index
    for (Map.Entry<Index, List<StaleShard>> indexEntry : staleShards.stream().collect(Collectors.groupingBy(fs -> fs.getShardId().getIndex())).entrySet()) {
        final IndexMetaData oldIndexMetaData = oldMetaData.getIndexSafe(indexEntry.getKey());
        IndexMetaData.Builder indexMetaDataBuilder = null;
        // group staleShards entries by shard id
        for (Map.Entry<ShardId, List<StaleShard>> shardEntry : indexEntry.getValue().stream().collect(Collectors.groupingBy(staleShard -> staleShard.getShardId())).entrySet()) {
            int shardNumber = shardEntry.getKey().getId();
            Set<String> oldInSyncAllocations = oldIndexMetaData.inSyncAllocationIds(shardNumber);
            Set<String> idsToRemove = shardEntry.getValue().stream().map(e -> e.getAllocationId()).collect(Collectors.toSet());
            assert idsToRemove.stream().allMatch(id -> oldRoutingTable.getByAllocationId(shardEntry.getKey(), id) == null) : "removing stale ids: " + idsToRemove + ", some of which have still a routing entry: " + oldRoutingTable;
            Set<String> remainingInSyncAllocations = Sets.difference(oldInSyncAllocations, idsToRemove);
            assert remainingInSyncAllocations.isEmpty() == false : "Set of in-sync ids cannot become empty for shard " + shardEntry.getKey() + " (before: " + oldInSyncAllocations + ", ids to remove: " + idsToRemove + ")";
            // (see ShardRouting#allocatedPostIndexCreate)
            if (remainingInSyncAllocations.isEmpty() == false) {
                if (indexMetaDataBuilder == null) {
                    indexMetaDataBuilder = IndexMetaData.builder(oldIndexMetaData);
                }
                indexMetaDataBuilder.putInSyncAllocationIds(shardNumber, remainingInSyncAllocations);
            }
        }
        if (indexMetaDataBuilder != null) {
            if (metaDataBuilder == null) {
                metaDataBuilder = MetaData.builder(oldMetaData);
            }
            metaDataBuilder.put(indexMetaDataBuilder);
        }
    }
    if (metaDataBuilder != null) {
        return ClusterState.builder(clusterState).metaData(metaDataBuilder).build();
    } else {
        return clusterState;
    }
}