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

Example 56 with UnassignedInfo

use of org.elasticsearch.cluster.routing.UnassignedInfo in project crate by crate.

the class CancelAllocationCommand method execute.

@Override
public RerouteExplanation execute(RoutingAllocation allocation, boolean explain) {
    DiscoveryNode discoNode = allocation.nodes().resolveNode(node);
    ShardRouting shardRouting = null;
    RoutingNodes routingNodes = allocation.routingNodes();
    RoutingNode routingNode = routingNodes.node(discoNode.getId());
    IndexMetadata indexMetadata = null;
    if (routingNode != null) {
        indexMetadata = allocation.metadata().index(index());
        if (indexMetadata == null) {
            throw new IndexNotFoundException(index());
        }
        ShardId shardId = new ShardId(indexMetadata.getIndex(), shardId());
        shardRouting = routingNode.getByShardId(shardId);
    }
    if (shardRouting == null) {
        if (explain) {
            return new RerouteExplanation(this, allocation.decision(Decision.NO, "cancel_allocation_command", "can't cancel " + shardId + ", failed to find it on node " + discoNode));
        }
        throw new IllegalArgumentException("[cancel_allocation] can't cancel " + shardId + ", failed to find it on node " + discoNode);
    }
    if (shardRouting.primary() && allowPrimary == false) {
        if ((shardRouting.initializing() && shardRouting.relocatingNodeId() != null) == false) {
            // only allow cancelling initializing shard of primary relocation without allowPrimary flag
            if (explain) {
                return new RerouteExplanation(this, allocation.decision(Decision.NO, "cancel_allocation_command", "can't cancel " + shardId + " on node " + discoNode + ", shard is primary and " + shardRouting.state().name().toLowerCase(Locale.ROOT)));
            }
            throw new IllegalArgumentException("[cancel_allocation] can't cancel " + shardId + " on node " + discoNode + ", shard is primary and " + shardRouting.state().name().toLowerCase(Locale.ROOT));
        }
    }
    routingNodes.failShard(LogManager.getLogger(CancelAllocationCommand.class), shardRouting, new UnassignedInfo(UnassignedInfo.Reason.REROUTE_CANCELLED, null), indexMetadata, allocation.changes());
    // TODO: We don't have to remove a cancelled shard from in-sync set once we have a strict resync implementation.
    allocation.removeAllocationId(shardRouting);
    return new RerouteExplanation(this, allocation.decision(Decision.YES, "cancel_allocation_command", "shard " + shardId + " on node " + discoNode + " can be cancelled"));
}

Example 57 with UnassignedInfo

use of org.elasticsearch.cluster.routing.UnassignedInfo in project crate by crate.

the class MaxRetryAllocationDecider method canAllocate.

@Override
public Decision canAllocate(ShardRouting shardRouting, RoutingAllocation allocation) {
    final UnassignedInfo unassignedInfo = shardRouting.unassignedInfo();
    final Decision decision;
    if (unassignedInfo != null && unassignedInfo.getNumFailedAllocations() > 0) {
        final IndexMetadata indexMetadata = allocation.metadata().getIndexSafe(shardRouting.index());
        final int maxRetry = SETTING_ALLOCATION_MAX_RETRY.get(indexMetadata.getSettings());
        if (unassignedInfo.getNumFailedAllocations() >= maxRetry) {
            decision = allocation.decision(Decision.NO, NAME, "shard has exceeded the maximum number of retries [%d] on " + "failed allocation attempts - manually execute 'ALTER CLUSTER REROUTE RETRY FAILED' to retry, [%s]", maxRetry, unassignedInfo.toString());
        } else {
            decision = allocation.decision(Decision.YES, NAME, "shard has failed allocating [%d] times but [%d] retries are allowed", unassignedInfo.getNumFailedAllocations(), maxRetry);
        }
    } else {
        decision = allocation.decision(Decision.YES, NAME, "shard has no previous failures");
    }
    return decision;
}

Example 58 with UnassignedInfo

use of org.elasticsearch.cluster.routing.UnassignedInfo in project crate by crate.

the class AllocateEmptyPrimaryAllocationCommand method execute.

@Override
public RerouteExplanation execute(RoutingAllocation allocation, boolean explain) {
    final DiscoveryNode discoNode;
    try {
        discoNode = allocation.nodes().resolveNode(node);
    } catch (IllegalArgumentException e) {
        return explainOrThrowRejectedCommand(explain, allocation, e);
    }
    final RoutingNodes routingNodes = allocation.routingNodes();
    RoutingNode routingNode = routingNodes.node(discoNode.getId());
    if (routingNode == null) {
        return explainOrThrowMissingRoutingNode(allocation, explain, discoNode);
    }
    try {
        allocation.routingTable().shardRoutingTable(index, shardId).primaryShard();
    } catch (IndexNotFoundException | ShardNotFoundException e) {
        return explainOrThrowRejectedCommand(explain, allocation, e);
    }
    ShardRouting shardRouting = null;
    for (ShardRouting shard : allocation.routingNodes().unassigned()) {
        if (shard.getIndexName().equals(index) && shard.getId() == shardId && shard.primary()) {
            shardRouting = shard;
            break;
        }
    }
    if (shardRouting == null) {
        return explainOrThrowRejectedCommand(explain, allocation, "primary [" + index + "][" + shardId + "] is already assigned");
    }
    if (shardRouting.recoverySource().getType() != RecoverySource.Type.EMPTY_STORE && acceptDataLoss == false) {
        String dataLossWarning = "allocating an empty primary for [" + index + "][" + shardId + "] can result in data loss. Please confirm by setting the accept_data_loss parameter to true";
        return explainOrThrowRejectedCommand(explain, allocation, dataLossWarning);
    }
    UnassignedInfo unassignedInfoToUpdate = null;
    if (shardRouting.unassignedInfo().getReason() != UnassignedInfo.Reason.FORCED_EMPTY_PRIMARY) {
        String unassignedInfoMessage = "force empty allocation from previous reason " + shardRouting.unassignedInfo().getReason() + ", " + shardRouting.unassignedInfo().getMessage();
        unassignedInfoToUpdate = new UnassignedInfo(UnassignedInfo.Reason.FORCED_EMPTY_PRIMARY, unassignedInfoMessage, shardRouting.unassignedInfo().getFailure(), 0, System.nanoTime(), System.currentTimeMillis(), false, shardRouting.unassignedInfo().getLastAllocationStatus(), Collections.emptySet());
    }
    initializeUnassignedShard(allocation, routingNodes, routingNode, shardRouting, unassignedInfoToUpdate, EmptyStoreRecoverySource.INSTANCE);
    return new RerouteExplanation(this, allocation.decision(Decision.YES, name() + " (allocation command)", "ignore deciders"));
}

Example 59 with UnassignedInfo

use of org.elasticsearch.cluster.routing.UnassignedInfo in project crate by crate.

the class ReplicaShardAllocator method makeAllocationDecision.

@Override
public AllocateUnassignedDecision makeAllocationDecision(final ShardRouting unassignedShard, final RoutingAllocation allocation, final Logger logger) {
    if (isResponsibleFor(unassignedShard) == false) {
        // this allocator is not responsible for deciding on this shard
        return AllocateUnassignedDecision.NOT_TAKEN;
    }
    final RoutingNodes routingNodes = allocation.routingNodes();
    final boolean explain = allocation.debugDecision();
    // pre-check if it can be allocated to any node that currently exists, so we won't list the store for it for nothing
    Tuple<Decision, Map<String, NodeAllocationResult>> result = canBeAllocatedToAtLeastOneNode(unassignedShard, allocation);
    Decision allocateDecision = result.v1();
    if (allocateDecision.type() != Decision.Type.YES && (explain == false || hasInitiatedFetching(unassignedShard) == false)) {
        // only return early if we are not in explain mode, or we are in explain mode but we have not
        // yet attempted to fetch any shard data
        logger.trace("{}: ignoring allocation, can't be allocated on any node", unassignedShard);
        return AllocateUnassignedDecision.no(UnassignedInfo.AllocationStatus.fromDecision(allocateDecision.type()), result.v2() != null ? new ArrayList<>(result.v2().values()) : null);
    }
    AsyncShardFetch.FetchResult<NodeStoreFilesMetadata> shardStores = fetchData(unassignedShard, allocation);
    if (shardStores.hasData() == false) {
        logger.trace("{}: ignoring allocation, still fetching shard stores", unassignedShard);
        allocation.setHasPendingAsyncFetch();
        List<NodeAllocationResult> nodeDecisions = null;
        if (explain) {
            nodeDecisions = buildDecisionsForAllNodes(unassignedShard, allocation);
        }
        return AllocateUnassignedDecision.no(AllocationStatus.FETCHING_SHARD_DATA, nodeDecisions);
    }
    ShardRouting primaryShard = routingNodes.activePrimary(unassignedShard.shardId());
    if (primaryShard == null) {
        assert explain : "primary should only be null here if we are in explain mode, so we didn't " + "exit early when canBeAllocatedToAtLeastOneNode didn't return a YES decision";
        return AllocateUnassignedDecision.no(UnassignedInfo.AllocationStatus.fromDecision(allocateDecision.type()), new ArrayList<>(result.v2().values()));
    }
    assert primaryShard.currentNodeId() != null;
    final DiscoveryNode primaryNode = allocation.nodes().get(primaryShard.currentNodeId());
    final TransportNodesListShardStoreMetadata.StoreFilesMetadata primaryStore = findStore(primaryNode, shardStores);
    if (primaryStore == null) {
        // if we can't find the primary data, it is probably because the primary shard is corrupted (and listing failed)
        // we want to let the replica be allocated in order to expose the actual problem with the primary that the replica
        // will try and recover from
        // Note, this is the existing behavior, as exposed in running CorruptFileTest#testNoPrimaryData
        logger.trace("{}: no primary shard store found or allocated, letting actual allocation figure it out", unassignedShard);
        return AllocateUnassignedDecision.NOT_TAKEN;
    }
    MatchingNodes matchingNodes = findMatchingNodes(unassignedShard, allocation, false, primaryNode, primaryStore, shardStores, explain);
    assert explain == false || matchingNodes.nodeDecisions != null : "in explain mode, we must have individual node decisions";
    List<NodeAllocationResult> nodeDecisions = augmentExplanationsWithStoreInfo(result.v2(), matchingNodes.nodeDecisions);
    if (allocateDecision.type() != Decision.Type.YES) {
        return AllocateUnassignedDecision.no(UnassignedInfo.AllocationStatus.fromDecision(allocateDecision.type()), nodeDecisions);
    } else if (matchingNodes.getNodeWithHighestMatch() != null) {
        RoutingNode nodeWithHighestMatch = allocation.routingNodes().node(matchingNodes.getNodeWithHighestMatch().getId());
        // we only check on THROTTLE since we checked before before on NO
        Decision decision = allocation.deciders().canAllocate(unassignedShard, nodeWithHighestMatch, allocation);
        if (decision.type() == Decision.Type.THROTTLE) {
            logger.debug("[{}][{}]: throttling allocation [{}] to [{}] in order to reuse its unallocated persistent store", unassignedShard.index(), unassignedShard.id(), unassignedShard, nodeWithHighestMatch.node());
            // we are throttling this, as we have enough other shards to allocate to this node, so ignore it for now
            return AllocateUnassignedDecision.throttle(nodeDecisions);
        } else {
            logger.debug("[{}][{}]: allocating [{}] to [{}] in order to reuse its unallocated persistent store", unassignedShard.index(), unassignedShard.id(), unassignedShard, nodeWithHighestMatch.node());
            // we found a match
            return AllocateUnassignedDecision.yes(nodeWithHighestMatch.node(), null, nodeDecisions, true);
        }
    } else if (matchingNodes.hasAnyData() == false && unassignedShard.unassignedInfo().isDelayed()) {
        // if we didn't manage to find *any* data (regardless of matching sizes), and the replica is
        // unassigned due to a node leaving, so we delay allocation of this replica to see if the
        // node with the shard copy will rejoin so we can re-use the copy it has
        logger.debug("{}: allocation of [{}] is delayed", unassignedShard.shardId(), unassignedShard);
        long remainingDelayMillis = 0L;
        long totalDelayMillis = 0L;
        if (explain) {
            UnassignedInfo unassignedInfo = unassignedShard.unassignedInfo();
            Metadata metadata = allocation.metadata();
            IndexMetadata indexMetadata = metadata.index(unassignedShard.index());
            totalDelayMillis = INDEX_DELAYED_NODE_LEFT_TIMEOUT_SETTING.get(indexMetadata.getSettings()).getMillis();
            long remainingDelayNanos = unassignedInfo.getRemainingDelay(System.nanoTime(), indexMetadata.getSettings());
            remainingDelayMillis = TimeValue.timeValueNanos(remainingDelayNanos).millis();
        }
        return AllocateUnassignedDecision.delayed(remainingDelayMillis, totalDelayMillis, nodeDecisions);
    }
    return AllocateUnassignedDecision.NOT_TAKEN;
}

Example 60 with UnassignedInfo

use of org.elasticsearch.cluster.routing.UnassignedInfo in project crate by crate.

the class ReplicaShardAllocator method processExistingRecoveries.

/**
 * Process existing recoveries of replicas and see if we need to cancel them if we find a better
 * match. Today, a better match is one that can perform a no-op recovery while the previous recovery
 * has to copy segment files.
 */
public void processExistingRecoveries(RoutingAllocation allocation) {
    Metadata metadata = allocation.metadata();
    RoutingNodes routingNodes = allocation.routingNodes();
    List<Runnable> shardCancellationActions = new ArrayList<>();
    for (RoutingNode routingNode : routingNodes) {
        for (ShardRouting shard : routingNode) {
            if (shard.primary()) {
                continue;
            }
            if (shard.initializing() == false) {
                continue;
            }
            if (shard.relocatingNodeId() != null) {
                continue;
            }
            // if we are allocating a replica because of index creation, no need to go and find a copy, there isn't one...
            if (shard.unassignedInfo() != null && shard.unassignedInfo().getReason() == UnassignedInfo.Reason.INDEX_CREATED) {
                continue;
            }
            AsyncShardFetch.FetchResult<NodeStoreFilesMetadata> shardStores = fetchData(shard, allocation);
            if (shardStores.hasData() == false) {
                logger.trace("{}: fetching new stores for initializing shard", shard);
                // still fetching
                continue;
            }
            ShardRouting primaryShard = allocation.routingNodes().activePrimary(shard.shardId());
            assert primaryShard != null : "the replica shard can be allocated on at least one node, so there must be an active primary";
            assert primaryShard.currentNodeId() != null;
            final DiscoveryNode primaryNode = allocation.nodes().get(primaryShard.currentNodeId());
            final TransportNodesListShardStoreMetadata.StoreFilesMetadata primaryStore = findStore(primaryNode, shardStores);
            if (primaryStore == null) {
                // if we can't find the primary data, it is probably because the primary shard is corrupted (and listing failed)
                // just let the recovery find it out, no need to do anything about it for the initializing shard
                logger.trace("{}: no primary shard store found or allocated, letting actual allocation figure it out", shard);
                continue;
            }
            MatchingNodes matchingNodes = findMatchingNodes(shard, allocation, true, primaryNode, primaryStore, shardStores, false);
            if (matchingNodes.getNodeWithHighestMatch() != null) {
                DiscoveryNode currentNode = allocation.nodes().get(shard.currentNodeId());
                DiscoveryNode nodeWithHighestMatch = matchingNodes.getNodeWithHighestMatch();
                // current node will not be in matchingNodes as it is filtered away by SameShardAllocationDecider
                if (currentNode.equals(nodeWithHighestMatch) == false && matchingNodes.canPerformNoopRecovery(nodeWithHighestMatch) && canPerformOperationBasedRecovery(primaryStore, shardStores, currentNode) == false) {
                    // we found a better match that can perform noop recovery, cancel the existing allocation.
                    logger.debug("cancelling allocation of replica on [{}], can perform a noop recovery on node [{}]", currentNode, nodeWithHighestMatch);
                    final Set<String> failedNodeIds = shard.unassignedInfo() == null ? Collections.emptySet() : shard.unassignedInfo().getFailedNodeIds();
                    UnassignedInfo unassignedInfo = new UnassignedInfo(UnassignedInfo.Reason.REALLOCATED_REPLICA, "existing allocation of replica to [" + currentNode + "] cancelled, can perform a noop recovery on [" + nodeWithHighestMatch + "]", null, 0, allocation.getCurrentNanoTime(), System.currentTimeMillis(), false, UnassignedInfo.AllocationStatus.NO_ATTEMPT, failedNodeIds);
                    // don't cancel shard in the loop as it will cause a ConcurrentModificationException
                    shardCancellationActions.add(() -> routingNodes.failShard(logger, shard, unassignedInfo, metadata.getIndexSafe(shard.index()), allocation.changes()));
                }
            }
        }
    }
    for (Runnable action : shardCancellationActions) {
        action.run();
    }
}