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

Example 16 with AllocationCommands

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

the class NodeVersionAllocationDeciderTests method testRebalanceDoesNotAllocatePrimaryAndReplicasOnDifferentVersionNodes.

public void testRebalanceDoesNotAllocatePrimaryAndReplicasOnDifferentVersionNodes() {
    ShardId shard1 = new ShardId("test1", "_na_", 0);
    ShardId shard2 = new ShardId("test2", "_na_", 0);
    final DiscoveryNode newNode = new DiscoveryNode("newNode", buildNewFakeTransportAddress(), emptyMap(), MASTER_DATA_ROLES, Version.CURRENT);
    final DiscoveryNode oldNode1 = new DiscoveryNode("oldNode1", buildNewFakeTransportAddress(), emptyMap(), MASTER_DATA_ROLES, VersionUtils.getPreviousVersion());
    final DiscoveryNode oldNode2 = new DiscoveryNode("oldNode2", buildNewFakeTransportAddress(), emptyMap(), MASTER_DATA_ROLES, VersionUtils.getPreviousVersion());
    AllocationId allocationId1P = AllocationId.newInitializing();
    AllocationId allocationId1R = AllocationId.newInitializing();
    AllocationId allocationId2P = AllocationId.newInitializing();
    AllocationId allocationId2R = AllocationId.newInitializing();
    Metadata metadata = Metadata.builder().put(IndexMetadata.builder(shard1.getIndexName()).settings(settings(Version.CURRENT).put(Settings.EMPTY)).numberOfShards(1).numberOfReplicas(1).putInSyncAllocationIds(0, Set.of(allocationId1P.getId(), allocationId1R.getId()))).put(IndexMetadata.builder(shard2.getIndexName()).settings(settings(Version.CURRENT).put(Settings.EMPTY)).numberOfShards(1).numberOfReplicas(1).putInSyncAllocationIds(0, Set.of(allocationId2P.getId(), allocationId2R.getId()))).build();
    RoutingTable routingTable = RoutingTable.builder().add(IndexRoutingTable.builder(shard1.getIndex()).addIndexShard(new IndexShardRoutingTable.Builder(shard1).addShard(TestShardRouting.newShardRouting(shard1.getIndexName(), shard1.getId(), newNode.getId(), null, true, ShardRoutingState.STARTED, allocationId1P)).addShard(TestShardRouting.newShardRouting(shard1.getIndexName(), shard1.getId(), oldNode1.getId(), null, false, ShardRoutingState.STARTED, allocationId1R)).build())).add(IndexRoutingTable.builder(shard2.getIndex()).addIndexShard(new IndexShardRoutingTable.Builder(shard2).addShard(TestShardRouting.newShardRouting(shard2.getIndexName(), shard2.getId(), newNode.getId(), null, true, ShardRoutingState.STARTED, allocationId2P)).addShard(TestShardRouting.newShardRouting(shard2.getIndexName(), shard2.getId(), oldNode1.getId(), null, false, ShardRoutingState.STARTED, allocationId2R)).build())).build();
    ClusterState state = ClusterState.builder(org.elasticsearch.cluster.ClusterName.CLUSTER_NAME_SETTING.getDefault(Settings.EMPTY)).metadata(metadata).routingTable(routingTable).nodes(DiscoveryNodes.builder().add(newNode).add(oldNode1).add(oldNode2)).build();
    AllocationDeciders allocationDeciders = new AllocationDeciders(Collections.singleton(new NodeVersionAllocationDecider()));
    AllocationService strategy = new MockAllocationService(allocationDeciders, new TestGatewayAllocator(), new BalancedShardsAllocator(Settings.EMPTY), EmptyClusterInfoService.INSTANCE);
    state = strategy.reroute(state, new AllocationCommands(), true, false).getClusterState();
    // the two indices must stay as is, the replicas cannot move to oldNode2 because versions don't match
    assertThat(state.routingTable().index(shard2.getIndex()).shardsWithState(ShardRoutingState.RELOCATING).size(), equalTo(0));
    assertThat(state.routingTable().index(shard1.getIndex()).shardsWithState(ShardRoutingState.RELOCATING).size(), equalTo(0));
}

Example 17 with AllocationCommands

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

the class TrackFailedAllocationNodesTests method testTrackFailedNodes.

public void testTrackFailedNodes() {
    int maxRetries = MaxRetryAllocationDecider.SETTING_ALLOCATION_MAX_RETRY.get(Settings.EMPTY);
    AllocationService allocationService = createAllocationService();
    Metadata metadata = Metadata.builder().put(IndexMetadata.builder("idx").settings(Settings.builder().put(IndexMetadata.SETTING_VERSION_CREATED, Version.CURRENT).put(IndexMetadata.SETTING_AUTO_EXPAND_REPLICAS, false).build()).numberOfShards(1).numberOfReplicas(1)).build();
    DiscoveryNodes.Builder discoNodes = DiscoveryNodes.builder();
    for (int i = 0; i < 5; i++) {
        discoNodes.add(newNode("node-" + i));
    }
    ClusterState clusterState = ClusterState.builder(ClusterName.CLUSTER_NAME_SETTING.getDefault(Settings.EMPTY)).nodes(discoNodes).metadata(metadata).routingTable(RoutingTable.builder().addAsNew(metadata.index("idx")).build()).build();
    clusterState = allocationService.reroute(clusterState, "reroute");
    Set<String> failedNodeIds = new HashSet<>();
    // track the failed nodes if shard is not started
    for (int i = 0; i < maxRetries; i++) {
        failedNodeIds.add(clusterState.routingTable().index("idx").shard(0).shards().get(0).currentNodeId());
        clusterState = allocationService.applyFailedShard(clusterState, clusterState.routingTable().index("idx").shard(0).shards().get(0), randomBoolean());
        assertThat(clusterState.routingTable().index("idx").shard(0).shards().get(0).unassignedInfo().getFailedNodeIds(), equalTo(failedNodeIds));
    }
    // reroute with retryFailed=true should discard the failedNodes
    assertThat(clusterState.routingTable().index("idx").shard(0).shards().get(0).state(), equalTo(ShardRoutingState.UNASSIGNED));
    clusterState = allocationService.reroute(clusterState, new AllocationCommands(), false, true).getClusterState();
    assertThat(clusterState.routingTable().index("idx").shard(0).shards().get(0).unassignedInfo().getFailedNodeIds(), empty());
    // do not track the failed nodes while shard is started
    clusterState = startInitializingShardsAndReroute(allocationService, clusterState);
    assertThat(clusterState.routingTable().index("idx").shard(0).shards().get(0).state(), equalTo(ShardRoutingState.STARTED));
    clusterState = allocationService.applyFailedShard(clusterState, clusterState.routingTable().index("idx").shard(0).shards().get(0), false);
    assertThat(clusterState.routingTable().index("idx").shard(0).shards().get(0).unassignedInfo().getFailedNodeIds(), empty());
}

Example 18 with AllocationCommands

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

the class RetryFailedAllocationTests method testRetryFailedResetForAllocationCommands.

@Test
public void testRetryFailedResetForAllocationCommands() {
    final int retries = MaxRetryAllocationDecider.SETTING_ALLOCATION_MAX_RETRY.get(Settings.EMPTY);
    clusterState = strategy.reroute(clusterState, "initial allocation");
    clusterState = startShardsAndReroute(strategy, clusterState, getPrimary());
    // Exhaust all replica allocation attempts with shard failures
    for (int i = 0; i < retries; i++) {
        List<FailedShard> failedShards = Collections.singletonList(new FailedShard(getReplica(), "failing-shard::attempt-" + i, new ElasticsearchException("simulated"), randomBoolean()));
        clusterState = strategy.applyFailedShards(clusterState, failedShards, List.of());
        clusterState = strategy.reroute(clusterState, "allocation retry attempt-" + i);
    }
    assertThat("replica should not be assigned", getReplica().state(), equalTo(ShardRoutingState.UNASSIGNED));
    assertThat("reroute should be a no-op", strategy.reroute(clusterState, "test"), sameInstance(clusterState));
    // Now allocate replica with retry_failed flag set
    AllocationService.CommandsResult result = strategy.reroute(clusterState, new AllocationCommands(new AllocateReplicaAllocationCommand(INDEX_NAME, 0, getPrimary().currentNodeId().equals("node1") ? "node2" : "node1")), false, true);
    clusterState = result.getClusterState();
    assertEquals(ShardRoutingState.INITIALIZING, getReplica().state());
    clusterState = startShardsAndReroute(strategy, clusterState, getReplica());
    assertEquals(ShardRoutingState.STARTED, getReplica().state());
    assertFalse(clusterState.getRoutingNodes().hasUnassignedShards());
}

Example 19 with AllocationCommands

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

the class MaxRetryAllocationDeciderTests method testSingleRetryOnIgnore.

public void testSingleRetryOnIgnore() {
    ClusterState clusterState = createInitialClusterState();
    RoutingTable routingTable = clusterState.routingTable();
    final int retries = MaxRetryAllocationDecider.SETTING_ALLOCATION_MAX_RETRY.get(Settings.EMPTY);
    // now fail it N-1 times
    for (int i = 0; i < retries - 1; i++) {
        List<FailedShard> failedShards = Collections.singletonList(new FailedShard(routingTable.index("idx").shard(0).shards().get(0), "boom" + i, new UnsupportedOperationException()));
        ClusterState newState = strategy.applyFailedShards(clusterState, failedShards);
        assertThat(newState, not(equalTo(clusterState)));
        clusterState = newState;
        routingTable = newState.routingTable();
        assertEquals(routingTable.index("idx").shards().size(), 1);
        assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), INITIALIZING);
        assertEquals(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getNumFailedAllocations(), i + 1);
        assertEquals(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getMessage(), "boom" + i);
    }
    // now we go and check that we are actually stick to unassigned on the next failure
    List<FailedShard> failedShards = Collections.singletonList(new FailedShard(routingTable.index("idx").shard(0).shards().get(0), "boom", new UnsupportedOperationException()));
    ClusterState newState = strategy.applyFailedShards(clusterState, failedShards);
    assertThat(newState, not(equalTo(clusterState)));
    clusterState = newState;
    routingTable = newState.routingTable();
    assertEquals(routingTable.index("idx").shards().size(), 1);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getNumFailedAllocations(), retries);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), UNASSIGNED);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getMessage(), "boom");
    // manual reroute should retry once
    newState = strategy.reroute(clusterState, new AllocationCommands(), false, true).getClusterState();
    assertThat(newState, not(equalTo(clusterState)));
    clusterState = newState;
    routingTable = newState.routingTable();
    clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
    assertEquals(routingTable.index("idx").shards().size(), 1);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getNumFailedAllocations(), retries);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), INITIALIZING);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getMessage(), "boom");
    // now we go and check that we are actually stick to unassigned on the next failure ie. no retry
    failedShards = Collections.singletonList(new FailedShard(routingTable.index("idx").shard(0).shards().get(0), "boom", new UnsupportedOperationException()));
    newState = strategy.applyFailedShards(clusterState, failedShards);
    assertThat(newState, not(equalTo(clusterState)));
    clusterState = newState;
    routingTable = newState.routingTable();
    assertEquals(routingTable.index("idx").shards().size(), 1);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getNumFailedAllocations(), retries + 1);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), UNASSIGNED);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getMessage(), "boom");
}

Example 20 with AllocationCommands

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

the class InSyncAllocationIdTests method testInSyncAllocationIdsUpdated.

public void testInSyncAllocationIdsUpdated() {
    logger.info("creating an index with 1 shard, 2 replicas");
    MetaData metaData = MetaData.builder().put(IndexMetaData.builder("test").settings(settings(Version.CURRENT)).numberOfShards(1).numberOfReplicas(2)).put(IndexMetaData.builder("test-old").settings(settings(Version.CURRENT)).numberOfShards(1).numberOfReplicas(2).putInSyncAllocationIds(0, new HashSet<>(Arrays.asList("x", "y")))).build();
    RoutingTable routingTable = RoutingTable.builder().addAsNew(metaData.index("test")).build();
    ClusterState clusterState = ClusterState.builder(org.elasticsearch.cluster.ClusterName.CLUSTER_NAME_SETTING.getDefault(Settings.EMPTY)).metaData(metaData).routingTable(routingTable).build();
    logger.info("adding three nodes and performing rerouting");
    clusterState = ClusterState.builder(clusterState).nodes(DiscoveryNodes.builder().add(newNode("node1")).add(newNode("node2")).add(newNode("node3"))).build();
    clusterState = allocation.reroute(clusterState, "reroute");
    assertThat(clusterState.metaData().index("test").inSyncAllocationIds(0).size(), equalTo(0));
    assertThat(clusterState.metaData().index("test-old").inSyncAllocationIds(0), equalTo(new HashSet<>(Arrays.asList("x", "y"))));
    logger.info("start primary shard");
    clusterState = allocation.applyStartedShards(clusterState, clusterState.getRoutingNodes().shardsWithState(INITIALIZING));
    assertThat(clusterState.getRoutingTable().shardsWithState(STARTED).size(), equalTo(1));
    assertThat(clusterState.metaData().index("test").inSyncAllocationIds(0).size(), equalTo(1));
    assertThat(clusterState.getRoutingTable().shardsWithState(STARTED).get(0).allocationId().getId(), equalTo(clusterState.metaData().index("test").inSyncAllocationIds(0).iterator().next()));
    assertThat(clusterState.metaData().index("test-old").inSyncAllocationIds(0), equalTo(new HashSet<>(Arrays.asList("x", "y"))));
    logger.info("start replica shards");
    clusterState = allocation.applyStartedShards(clusterState, clusterState.getRoutingNodes().shardsWithState(INITIALIZING));
    assertThat(clusterState.metaData().index("test").inSyncAllocationIds(0).size(), equalTo(3));
    logger.info("remove a node");
    clusterState = ClusterState.builder(clusterState).nodes(DiscoveryNodes.builder(clusterState.nodes()).remove("node1")).build();
    clusterState = allocation.deassociateDeadNodes(clusterState, true, "reroute");
    // in-sync allocation ids should not be updated
    assertThat(clusterState.metaData().index("test").inSyncAllocationIds(0).size(), equalTo(3));
    logger.info("remove all remaining nodes");
    clusterState = ClusterState.builder(clusterState).nodes(DiscoveryNodes.builder(clusterState.nodes()).remove("node2").remove("node3")).build();
    clusterState = allocation.deassociateDeadNodes(clusterState, true, "reroute");
    // in-sync allocation ids should not be updated
    assertThat(clusterState.getRoutingTable().shardsWithState(UNASSIGNED).size(), equalTo(3));
    assertThat(clusterState.metaData().index("test").inSyncAllocationIds(0).size(), equalTo(3));
    // force empty primary
    clusterState = ClusterState.builder(clusterState).nodes(DiscoveryNodes.builder(clusterState.nodes()).add(newNode("node1"))).build();
    clusterState = allocation.reroute(clusterState, new AllocationCommands(new AllocateEmptyPrimaryAllocationCommand("test", 0, "node1", true)), false, false).getClusterState();
    // check that in-sync allocation ids are reset by forcing an empty primary
    assertThat(clusterState.metaData().index("test").inSyncAllocationIds(0).size(), equalTo(0));
    logger.info("start primary shard");
    clusterState = allocation.applyStartedShards(clusterState, clusterState.getRoutingNodes().shardsWithState(INITIALIZING));
    assertThat(clusterState.getRoutingTable().shardsWithState(STARTED).size(), equalTo(1));
    assertThat(clusterState.metaData().index("test").inSyncAllocationIds(0).size(), equalTo(1));
    logger.info("fail primary shard");
    ShardRouting startedPrimary = clusterState.getRoutingNodes().shardsWithState(STARTED).get(0);
    clusterState = allocation.applyFailedShard(clusterState, startedPrimary);
    assertThat(clusterState.getRoutingTable().shardsWithState(STARTED).size(), equalTo(0));
    assertEquals(Collections.singleton(startedPrimary.allocationId().getId()), clusterState.metaData().index("test").inSyncAllocationIds(0));
}