Examples with RerouteExplanation org.elasticsearch.cluster.routing.allocation.RerouteExplanation used on opensource projects

Example 6 with RerouteExplanation

use of org.elasticsearch.cluster.routing.allocation.RerouteExplanation in project crate by crate.

the class AllocateReplicaAllocationCommand 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 primaryShardRouting = null;
    for (RoutingNode node : allocation.routingNodes()) {
        for (ShardRouting shard : node) {
            if (shard.getIndexName().equals(index) && shard.getId() == shardId && shard.primary()) {
                primaryShardRouting = shard;
                break;
            }
        }
    }
    if (primaryShardRouting == null) {
        return explainOrThrowRejectedCommand(explain, allocation, "trying to allocate a replica shard [" + index + "][" + shardId + "], while corresponding primary shard is still unassigned");
    }
    List<ShardRouting> replicaShardRoutings = new ArrayList<>();
    for (ShardRouting shard : allocation.routingNodes().unassigned()) {
        if (shard.getIndexName().equals(index) && shard.getId() == shardId && shard.primary() == false) {
            replicaShardRoutings.add(shard);
        }
    }
    ShardRouting shardRouting;
    if (replicaShardRoutings.isEmpty()) {
        return explainOrThrowRejectedCommand(explain, allocation, "all copies of [" + index + "][" + shardId + "] are already assigned. Use the move allocation command instead");
    } else {
        shardRouting = replicaShardRoutings.get(0);
    }
    Decision decision = allocation.deciders().canAllocate(shardRouting, routingNode, allocation);
    if (decision.type() == Decision.Type.NO) {
        // don't use explainOrThrowRejectedCommand to keep the original "NO" decision
        if (explain) {
            return new RerouteExplanation(this, decision);
        }
        throw new IllegalArgumentException("[" + name() + "] allocation of [" + index + "][" + shardId + "] on node " + discoNode + " is not allowed, reason: " + decision);
    }
    initializeUnassignedShard(allocation, routingNodes, routingNode, shardRouting);
    return new RerouteExplanation(this, decision);
}

Example 7 with RerouteExplanation

use of org.elasticsearch.cluster.routing.allocation.RerouteExplanation in project crate by crate.

the class MoveAllocationCommand method execute.

@Override
public RerouteExplanation execute(RoutingAllocation allocation, boolean explain) {
    DiscoveryNode fromDiscoNode = allocation.nodes().resolveNode(fromNode);
    DiscoveryNode toDiscoNode = allocation.nodes().resolveNode(toNode);
    Decision decision = null;
    boolean found = false;
    RoutingNode fromRoutingNode = allocation.routingNodes().node(fromDiscoNode.getId());
    if (fromRoutingNode == null && !fromDiscoNode.isDataNode()) {
        throw new IllegalArgumentException("[move_allocation] can't move [" + index + "][" + shardId + "] from " + fromDiscoNode + " to " + toDiscoNode + ": source [" + fromDiscoNode.getName() + "] is not a data node.");
    }
    RoutingNode toRoutingNode = allocation.routingNodes().node(toDiscoNode.getId());
    if (toRoutingNode == null && !toDiscoNode.isDataNode()) {
        throw new IllegalArgumentException("[move_allocation] can't move [" + index + "][" + shardId + "] from " + fromDiscoNode + " to " + toDiscoNode + ": source [" + toDiscoNode.getName() + "] is not a data node.");
    }
    for (ShardRouting shardRouting : fromRoutingNode) {
        if (!shardRouting.shardId().getIndexName().equals(index)) {
            continue;
        }
        if (shardRouting.shardId().id() != shardId) {
            continue;
        }
        found = true;
        // TODO we can possibly support also relocating cases, where we cancel relocation and move...
        if (!shardRouting.started()) {
            if (explain) {
                return new RerouteExplanation(this, allocation.decision(Decision.NO, "move_allocation_command", "shard " + shardId + " has not been started"));
            }
            throw new IllegalArgumentException("[move_allocation] can't move " + shardId + ", shard is not started (state = " + shardRouting.state() + "]");
        }
        decision = allocation.deciders().canAllocate(shardRouting, toRoutingNode, allocation);
        if (decision.type() == Decision.Type.NO) {
            if (explain) {
                return new RerouteExplanation(this, decision);
            }
            throw new IllegalArgumentException("[move_allocation] can't move " + shardId + ", from " + fromDiscoNode + ", to " + toDiscoNode + ", since its not allowed, reason: " + decision);
        }
        if (decision.type() == Decision.Type.THROTTLE) {
        // its being throttled, maybe have a flag to take it into account and fail? for now, just do it since the "user" wants it...
        }
        allocation.routingNodes().relocateShard(shardRouting, toRoutingNode.nodeId(), allocation.clusterInfo().getShardSize(shardRouting, ShardRouting.UNAVAILABLE_EXPECTED_SHARD_SIZE), allocation.changes());
    }
    if (!found) {
        if (explain) {
            return new RerouteExplanation(this, allocation.decision(Decision.NO, "move_allocation_command", "shard " + shardId + " not found"));
        }
        throw new IllegalArgumentException("[move_allocation] can't move " + shardId + ", failed to find it on node " + fromDiscoNode);
    }
    return new RerouteExplanation(this, decision);
}

Example 8 with RerouteExplanation

use of org.elasticsearch.cluster.routing.allocation.RerouteExplanation in project elasticsearch by elastic.

the class ClusterRerouteIT method testRerouteExplain.

public void testRerouteExplain() {
    Settings commonSettings = Settings.builder().build();
    logger.info("--> starting a node");
    String node_1 = internalCluster().startNode(commonSettings);
    assertThat(cluster().size(), equalTo(1));
    ClusterHealthResponse healthResponse = client().admin().cluster().prepareHealth().setWaitForNodes("1").execute().actionGet();
    assertThat(healthResponse.isTimedOut(), equalTo(false));
    logger.info("--> create an index with 1 shard");
    client().admin().indices().prepareCreate("test").setSettings(Settings.builder().put("index.number_of_shards", 1).put("index.number_of_replicas", 0)).execute().actionGet();
    ensureGreen("test");
    logger.info("--> disable allocation");
    Settings newSettings = Settings.builder().put(CLUSTER_ROUTING_ALLOCATION_ENABLE_SETTING.getKey(), Allocation.NONE.name()).build();
    client().admin().cluster().prepareUpdateSettings().setTransientSettings(newSettings).execute().actionGet();
    logger.info("--> starting a second node");
    String node_2 = internalCluster().startNode(commonSettings);
    assertThat(cluster().size(), equalTo(2));
    healthResponse = client().admin().cluster().prepareHealth().setWaitForNodes("2").execute().actionGet();
    assertThat(healthResponse.isTimedOut(), equalTo(false));
    logger.info("--> try to move the shard from node1 to node2");
    MoveAllocationCommand cmd = new MoveAllocationCommand("test", 0, node_1, node_2);
    ClusterRerouteResponse resp = client().admin().cluster().prepareReroute().add(cmd).setExplain(true).execute().actionGet();
    RoutingExplanations e = resp.getExplanations();
    assertThat(e.explanations().size(), equalTo(1));
    RerouteExplanation explanation = e.explanations().get(0);
    assertThat(explanation.command().name(), equalTo(cmd.name()));
    assertThat(((MoveAllocationCommand) explanation.command()).shardId(), equalTo(cmd.shardId()));
    assertThat(((MoveAllocationCommand) explanation.command()).fromNode(), equalTo(cmd.fromNode()));
    assertThat(((MoveAllocationCommand) explanation.command()).toNode(), equalTo(cmd.toNode()));
    assertThat(explanation.decisions().type(), equalTo(Decision.Type.YES));
}

Example 9 with RerouteExplanation

use of org.elasticsearch.cluster.routing.allocation.RerouteExplanation in project crate by crate.

the class AllocateStalePrimaryAllocationCommand 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 (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);
    }
    if (shardRouting.recoverySource().getType() != RecoverySource.Type.EXISTING_STORE) {
        return explainOrThrowRejectedCommand(explain, allocation, "trying to allocate an existing primary shard [" + index + "][" + shardId + "], while no such shard has ever been active");
    }
    initializeUnassignedShard(allocation, routingNodes, routingNode, shardRouting, null, RecoverySource.ExistingStoreRecoverySource.FORCE_STALE_PRIMARY_INSTANCE);
    return new RerouteExplanation(this, allocation.decision(Decision.YES, name() + " (allocation command)", "ignore deciders"));
}

Example 10 with RerouteExplanation

use of org.elasticsearch.cluster.routing.allocation.RerouteExplanation 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"));
}