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

Example 1 with RoutingTable

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

the class ShrinkIndexIT method testCreateShrinkIndexFails.

/**
     * Tests that we can manually recover from a failed allocation due to shards being moved away etc.
     */
public void testCreateShrinkIndexFails() throws Exception {
    internalCluster().ensureAtLeastNumDataNodes(2);
    prepareCreate("source").setSettings(Settings.builder().put(indexSettings()).put("number_of_shards", randomIntBetween(2, 7)).put("number_of_replicas", 0)).get();
    for (int i = 0; i < 20; i++) {
        client().prepareIndex("source", randomFrom("t1", "t2", "t3")).setSource("{\"foo\" : \"bar\", \"i\" : " + i + "}", XContentType.JSON).get();
    }
    ImmutableOpenMap<String, DiscoveryNode> dataNodes = client().admin().cluster().prepareState().get().getState().nodes().getDataNodes();
    assertTrue("at least 2 nodes but was: " + dataNodes.size(), dataNodes.size() >= 2);
    DiscoveryNode[] discoveryNodes = dataNodes.values().toArray(DiscoveryNode.class);
    String spareNode = discoveryNodes[0].getName();
    String mergeNode = discoveryNodes[1].getName();
    // ensure all shards are allocated otherwise the ensure green below might not succeed since we require the merge node
    // if we change the setting too quickly we will end up with one replica unassigned which can't be assigned anymore due
    // to the require._name below.
    ensureGreen();
    // relocate all shards to one node such that we can merge it.
    client().admin().indices().prepareUpdateSettings("source").setSettings(Settings.builder().put("index.routing.allocation.require._name", mergeNode).put("index.blocks.write", true)).get();
    ensureGreen();
    // now merge source into a single shard index
    client().admin().indices().prepareShrinkIndex("source", "target").setWaitForActiveShards(ActiveShardCount.NONE).setSettings(Settings.builder().put("index.routing.allocation.exclude._name", // we manually exclude the merge node to forcefully fuck it up
    mergeNode).put("index.number_of_replicas", 0).put("index.allocation.max_retries", 1).build()).get();
    client().admin().cluster().prepareHealth("target").setWaitForEvents(Priority.LANGUID).get();
    // now we move all shards away from the merge node
    client().admin().indices().prepareUpdateSettings("source").setSettings(Settings.builder().put("index.routing.allocation.require._name", spareNode).put("index.blocks.write", true)).get();
    ensureGreen("source");
    // erase the forcefully fuckup!
    client().admin().indices().prepareUpdateSettings("target").setSettings(Settings.builder().putNull("index.routing.allocation.exclude._name")).get();
    // wait until it fails
    assertBusy(() -> {
        ClusterStateResponse clusterStateResponse = client().admin().cluster().prepareState().get();
        RoutingTable routingTables = clusterStateResponse.getState().routingTable();
        assertTrue(routingTables.index("target").shard(0).getShards().get(0).unassigned());
        assertEquals(UnassignedInfo.Reason.ALLOCATION_FAILED, routingTables.index("target").shard(0).getShards().get(0).unassignedInfo().getReason());
        assertEquals(1, routingTables.index("target").shard(0).getShards().get(0).unassignedInfo().getNumFailedAllocations());
    });
    // now relocate them all to the right node
    client().admin().indices().prepareUpdateSettings("source").setSettings(Settings.builder().put("index.routing.allocation.require._name", mergeNode)).get();
    ensureGreen("source");
    final InternalClusterInfoService infoService = (InternalClusterInfoService) internalCluster().getInstance(ClusterInfoService.class, internalCluster().getMasterName());
    infoService.refresh();
    // kick off a retry and wait until it's done!
    ClusterRerouteResponse clusterRerouteResponse = client().admin().cluster().prepareReroute().setRetryFailed(true).get();
    long expectedShardSize = clusterRerouteResponse.getState().routingTable().index("target").shard(0).getShards().get(0).getExpectedShardSize();
    // we support the expected shard size in the allocator to sum up over the source index shards
    assertTrue("expected shard size must be set but wasn't: " + expectedShardSize, expectedShardSize > 0);
    ensureGreen();
    assertHitCount(client().prepareSearch("target").setSize(100).setQuery(new TermsQueryBuilder("foo", "bar")).get(), 20);
}

Example 2 with RoutingTable

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

the class ClusterRerouteTests method testClusterStateUpdateTask.

public void testClusterStateUpdateTask() {
    AllocationService allocationService = new AllocationService(Settings.builder().build(), new AllocationDeciders(Settings.EMPTY, Collections.singleton(new MaxRetryAllocationDecider(Settings.EMPTY))), new TestGatewayAllocator(), new BalancedShardsAllocator(Settings.EMPTY), EmptyClusterInfoService.INSTANCE);
    ClusterState clusterState = createInitialClusterState(allocationService);
    ClusterRerouteRequest req = new ClusterRerouteRequest();
    req.dryRun(true);
    AtomicReference<ClusterRerouteResponse> responseRef = new AtomicReference<>();
    ActionListener<ClusterRerouteResponse> responseActionListener = new ActionListener<ClusterRerouteResponse>() {

        @Override
        public void onResponse(ClusterRerouteResponse clusterRerouteResponse) {
            responseRef.set(clusterRerouteResponse);
        }

        @Override
        public void onFailure(Exception e) {
        }
    };
    TransportClusterRerouteAction.ClusterRerouteResponseAckedClusterStateUpdateTask task = new TransportClusterRerouteAction.ClusterRerouteResponseAckedClusterStateUpdateTask(logger, allocationService, req, responseActionListener);
    ClusterState execute = task.execute(clusterState);
    // dry-run
    assertSame(execute, clusterState);
    task.onAllNodesAcked(null);
    assertNotSame(responseRef.get().getState(), execute);
    // now we allocate
    req.dryRun(false);
    final int retries = MaxRetryAllocationDecider.SETTING_ALLOCATION_MAX_RETRY.get(Settings.EMPTY);
    // now fail it N-1 times
    for (int i = 0; i < retries; i++) {
        ClusterState newState = task.execute(clusterState);
        // dry-run=false
        assertNotSame(newState, clusterState);
        clusterState = newState;
        RoutingTable routingTable = clusterState.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);
        List<FailedShard> failedShards = Collections.singletonList(new FailedShard(routingTable.index("idx").shard(0).shards().get(0), "boom" + i, new UnsupportedOperationException()));
        newState = allocationService.applyFailedShards(clusterState, failedShards);
        assertThat(newState, not(equalTo(clusterState)));
        clusterState = newState;
        routingTable = clusterState.routingTable();
        assertEquals(routingTable.index("idx").shards().size(), 1);
        if (i == retries - 1) {
            assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), UNASSIGNED);
        } else {
            assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), INITIALIZING);
        }
        assertEquals(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getNumFailedAllocations(), i + 1);
    }
    // without retry_failed we won't allocate that shard
    ClusterState newState = task.execute(clusterState);
    // dry-run=false
    assertNotSame(newState, clusterState);
    task.onAllNodesAcked(null);
    assertSame(responseRef.get().getState(), newState);
    RoutingTable routingTable = clusterState.routingTable();
    assertEquals(routingTable.index("idx").shards().size(), 1);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), UNASSIGNED);
    assertEquals(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getNumFailedAllocations(), retries);
    // now we manually retry and get the shard back into initializing
    req.setRetryFailed(true);
    newState = task.execute(clusterState);
    // dry-run=false
    assertNotSame(newState, clusterState);
    clusterState = newState;
    routingTable = clusterState.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(), retries);
}

Example 3 with RoutingTable

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

the class ClusterStateCreationUtils method state.

/**
     * Creates cluster state with the given indices, each index containing #(numberOfPrimaries)
     * started primary shards and no replicas.  The cluster state contains #(numberOfNodes) nodes
     * and assigns primaries to those nodes.
     */
public static ClusterState state(final int numberOfNodes, final String[] indices, final int numberOfPrimaries) {
    DiscoveryNodes.Builder discoBuilder = DiscoveryNodes.builder();
    Set<String> nodes = new HashSet<>();
    for (int i = 0; i < numberOfNodes; i++) {
        final DiscoveryNode node = newNode(i);
        discoBuilder = discoBuilder.add(node);
        nodes.add(node.getId());
    }
    discoBuilder.localNodeId(newNode(0).getId());
    discoBuilder.masterNodeId(newNode(0).getId());
    MetaData.Builder metaData = MetaData.builder();
    RoutingTable.Builder routingTable = RoutingTable.builder();
    List<String> nodesList = new ArrayList<>(nodes);
    int currentNodeToAssign = 0;
    for (String index : indices) {
        IndexMetaData indexMetaData = IndexMetaData.builder(index).settings(Settings.builder().put(SETTING_VERSION_CREATED, Version.CURRENT).put(SETTING_NUMBER_OF_SHARDS, numberOfPrimaries).put(SETTING_NUMBER_OF_REPLICAS, 0).put(SETTING_CREATION_DATE, System.currentTimeMillis())).build();
        RoutingTable.Builder routing = new RoutingTable.Builder();
        routing.addAsNew(indexMetaData);
        IndexRoutingTable.Builder indexRoutingTable = IndexRoutingTable.builder(indexMetaData.getIndex());
        for (int i = 0; i < numberOfPrimaries; i++) {
            ShardId shardId = new ShardId(indexMetaData.getIndex(), i);
            IndexShardRoutingTable.Builder indexShardRoutingBuilder = new IndexShardRoutingTable.Builder(shardId);
            indexShardRoutingBuilder.addShard(TestShardRouting.newShardRouting(shardId, nodesList.get(currentNodeToAssign++), true, ShardRoutingState.STARTED));
            if (currentNodeToAssign == nodesList.size()) {
                currentNodeToAssign = 0;
            }
            indexRoutingTable.addIndexShard(indexShardRoutingBuilder.build());
        }
        metaData.put(indexMetaData, false);
        routingTable.add(indexRoutingTable);
    }
    ClusterState.Builder state = ClusterState.builder(new ClusterName("test"));
    state.nodes(discoBuilder);
    state.metaData(metaData.generateClusterUuidIfNeeded().build());
    state.routingTable(routingTable.build());
    return state.build();
}

Example 4 with RoutingTable

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

the class AddIncrementallyTests method initCluster.

private ClusterState initCluster(AllocationService service, int numberOfNodes, int numberOfIndices, int numberOfShards, int numberOfReplicas) {
    MetaData.Builder metaDataBuilder = MetaData.builder();
    RoutingTable.Builder routingTableBuilder = RoutingTable.builder();
    for (int i = 0; i < numberOfIndices; i++) {
        IndexMetaData.Builder index = IndexMetaData.builder("test" + i).settings(settings(Version.CURRENT)).numberOfShards(numberOfShards).numberOfReplicas(numberOfReplicas);
        metaDataBuilder = metaDataBuilder.put(index);
    }
    MetaData metaData = metaDataBuilder.build();
    for (ObjectCursor<IndexMetaData> cursor : metaData.indices().values()) {
        routingTableBuilder.addAsNew(cursor.value);
    }
    RoutingTable initialRoutingTable = routingTableBuilder.build();
    logger.info("start {} nodes", numberOfNodes);
    DiscoveryNodes.Builder nodes = DiscoveryNodes.builder();
    for (int i = 0; i < numberOfNodes; i++) {
        nodes.add(newNode("node" + i));
    }
    ClusterState clusterState = ClusterState.builder(org.elasticsearch.cluster.ClusterName.CLUSTER_NAME_SETTING.getDefault(Settings.EMPTY)).nodes(nodes).metaData(metaData).routingTable(initialRoutingTable).build();
    clusterState = service.reroute(clusterState, "reroute");
    logger.info("restart all the primary shards, replicas will start initializing");
    RoutingNodes routingNodes = clusterState.getRoutingNodes();
    clusterState = service.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING));
    logger.info("start the replica shards");
    routingNodes = clusterState.getRoutingNodes();
    clusterState = service.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING));
    routingNodes = clusterState.getRoutingNodes();
    logger.info("complete rebalancing");
    return applyStartedShardsUntilNoChange(clusterState, service);
}

Example 5 with RoutingTable

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

the class AllocationCommandsTests method testAllocateCommand.

public void testAllocateCommand() {
    AllocationService allocation = createAllocationService(Settings.builder().put(EnableAllocationDecider.CLUSTER_ROUTING_ALLOCATION_ENABLE_SETTING.getKey(), "none").put(EnableAllocationDecider.CLUSTER_ROUTING_REBALANCE_ENABLE_SETTING.getKey(), "none").build());
    final String index = "test";
    logger.info("--> building initial routing table");
    MetaData metaData = MetaData.builder().put(IndexMetaData.builder(index).settings(settings(Version.CURRENT)).numberOfShards(1).numberOfReplicas(1).putInSyncAllocationIds(0, Collections.singleton("asdf")).putInSyncAllocationIds(1, Collections.singleton("qwertz"))).build();
    // shard routing is added as "from recovery" instead of "new index creation" so that we can test below that allocating an empty
    // primary with accept_data_loss flag set to false fails
    RoutingTable routingTable = RoutingTable.builder().addAsRecovery(metaData.index(index)).build();
    ClusterState clusterState = ClusterState.builder(org.elasticsearch.cluster.ClusterName.CLUSTER_NAME_SETTING.getDefault(Settings.EMPTY)).metaData(metaData).routingTable(routingTable).build();
    final ShardId shardId = new ShardId(metaData.index(index).getIndex(), 0);
    logger.info("--> adding 3 nodes on same rack and do rerouting");
    clusterState = ClusterState.builder(clusterState).nodes(DiscoveryNodes.builder().add(newNode("node1")).add(newNode("node2")).add(newNode("node3")).add(newNode("node4", singleton(DiscoveryNode.Role.MASTER)))).build();
    clusterState = allocation.reroute(clusterState, "reroute");
    assertThat(clusterState.getRoutingNodes().shardsWithState(INITIALIZING).size(), equalTo(0));
    logger.info("--> allocating to non-existent node, should fail");
    try {
        allocation.reroute(clusterState, new AllocationCommands(randomAllocateCommand(index, shardId.id(), "node42")), false, false);
        fail("expected IllegalArgumentException when allocating to non-existing node");
    } catch (IllegalArgumentException e) {
        assertThat(e.getMessage(), containsString("failed to resolve [node42], no matching nodes"));
    }
    logger.info("--> allocating to non-data node, should fail");
    try {
        allocation.reroute(clusterState, new AllocationCommands(randomAllocateCommand(index, shardId.id(), "node4")), false, false);
        fail("expected IllegalArgumentException when allocating to non-data node");
    } catch (IllegalArgumentException e) {
        assertThat(e.getMessage(), containsString("allocation can only be done on data nodes"));
    }
    logger.info("--> allocating non-existing shard, should fail");
    try {
        allocation.reroute(clusterState, new AllocationCommands(randomAllocateCommand("test", 1, "node2")), false, false);
        fail("expected ShardNotFoundException when allocating non-existing shard");
    } catch (ShardNotFoundException e) {
        assertThat(e.getMessage(), containsString("no such shard"));
    }
    logger.info("--> allocating non-existing index, should fail");
    try {
        allocation.reroute(clusterState, new AllocationCommands(randomAllocateCommand("test2", 0, "node2")), false, false);
        fail("expected ShardNotFoundException when allocating non-existing index");
    } catch (IndexNotFoundException e) {
        assertThat(e.getMessage(), containsString("no such index"));
    }
    logger.info("--> allocating empty primary with acceptDataLoss flag set to false");
    try {
        allocation.reroute(clusterState, new AllocationCommands(new AllocateEmptyPrimaryAllocationCommand("test", 0, "node1", false)), false, false);
        fail("expected IllegalArgumentException when allocating empty primary with acceptDataLoss flag set to false");
    } catch (IllegalArgumentException e) {
        assertThat(e.getMessage(), containsString("allocating an empty primary for " + shardId + " can result in data loss. Please confirm by setting the accept_data_loss parameter to true"));
    }
    logger.info("--> allocating stale primary with acceptDataLoss flag set to false");
    try {
        allocation.reroute(clusterState, new AllocationCommands(new AllocateStalePrimaryAllocationCommand(index, shardId.id(), "node1", false)), false, false);
        fail("expected IllegalArgumentException when allocating stale primary with acceptDataLoss flag set to false");
    } catch (IllegalArgumentException e) {
        assertThat(e.getMessage(), containsString("allocating an empty primary for " + shardId + " can result in data loss. Please confirm by setting the accept_data_loss parameter to true"));
    }
    logger.info("--> allocating empty primary with acceptDataLoss flag set to true");
    ClusterState newState = allocation.reroute(clusterState, new AllocationCommands(new AllocateEmptyPrimaryAllocationCommand("test", 0, "node1", true)), false, false).getClusterState();
    assertThat(newState, not(equalTo(clusterState)));
    clusterState = newState;
    assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(1));
    assertThat(clusterState.getRoutingNodes().node("node1").shardsWithState(INITIALIZING).size(), equalTo(1));
    assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(0));
    logger.info("--> start the primary shard");
    clusterState = allocation.applyStartedShards(clusterState, clusterState.getRoutingNodes().shardsWithState(INITIALIZING));
    assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(1));
    assertThat(clusterState.getRoutingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
    assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(0));
    logger.info("--> allocate the replica shard on the primary shard node, should fail");
    try {
        allocation.reroute(clusterState, new AllocationCommands(new AllocateReplicaAllocationCommand("test", 0, "node1")), false, false);
        fail("expected IllegalArgumentException when allocating replica shard on the primary shard node");
    } catch (IllegalArgumentException e) {
    }
    logger.info("--> allocate the replica shard on on the second node");
    newState = allocation.reroute(clusterState, new AllocationCommands(new AllocateReplicaAllocationCommand("test", 0, "node2")), false, false).getClusterState();
    assertThat(newState, not(equalTo(clusterState)));
    clusterState = newState;
    assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(1));
    assertThat(clusterState.getRoutingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
    assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(1));
    assertThat(clusterState.getRoutingNodes().node("node2").shardsWithState(INITIALIZING).size(), equalTo(1));
    logger.info("--> start the replica shard");
    clusterState = allocation.applyStartedShards(clusterState, clusterState.getRoutingNodes().shardsWithState(INITIALIZING));
    assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(1));
    assertThat(clusterState.getRoutingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
    assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(1));
    assertThat(clusterState.getRoutingNodes().node("node2").shardsWithState(STARTED).size(), equalTo(1));
    logger.info("--> verify that we fail when there are no unassigned shards");
    try {
        allocation.reroute(clusterState, new AllocationCommands(randomAllocateCommand("test", 0, "node3")), false, false);
        fail("expected IllegalArgumentException when allocating shard while no unassigned shard available");
    } catch (IllegalArgumentException e) {
    }
}