@Test
public void testRoutingTableSerialization() throws Exception {
MetaData metaData =
MetaData.builder()
.put(IndexMetaData.builder("test").numberOfShards(10).numberOfReplicas(1))
.build();
RoutingTable routingTable = RoutingTable.builder().addAsNew(metaData.index("test")).build();
DiscoveryNodes nodes =
DiscoveryNodes.builder()
.put(newNode("node1"))
.put(newNode("node2"))
.put(newNode("node3"))
.build();
ClusterState clusterState =
ClusterState.builder().nodes(nodes).metaData(metaData).routingTable(routingTable).build();
AllocationService strategy = createAllocationService();
RoutingTable source = strategy.reroute(clusterState).routingTable();
BytesStreamOutput outStream = new BytesStreamOutput();
RoutingTable.Builder.writeTo(source, outStream);
BytesStreamInput inStream = new BytesStreamInput(outStream.bytes().toBytes(), false);
RoutingTable target = RoutingTable.Builder.readFrom(inStream);
assertThat(target.prettyPrint(), equalTo(source.prettyPrint()));
}
private ClusterState createInitialClusterState(AllocationService service) {
MetaData.Builder metaBuilder = MetaData.builder();
metaBuilder.put(
IndexMetaData.builder("idx")
.settings(settings(Version.CURRENT))
.numberOfShards(1)
.numberOfReplicas(0));
MetaData metaData = metaBuilder.build();
RoutingTable.Builder routingTableBuilder = RoutingTable.builder();
routingTableBuilder.addAsNew(metaData.index("idx"));
RoutingTable routingTable = routingTableBuilder.build();
ClusterState clusterState =
ClusterState.builder(
org.elasticsearch.cluster.ClusterName.CLUSTER_NAME_SETTING.getDefault(
Settings.EMPTY))
.metaData(metaData)
.routingTable(routingTable)
.build();
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder().add(newNode("node1")).add(newNode("node2")))
.build();
RoutingTable prevRoutingTable = routingTable;
routingTable = service.reroute(clusterState, "reroute").routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertEquals(prevRoutingTable.index("idx").shards().size(), 1);
assertEquals(prevRoutingTable.index("idx").shard(0).shards().get(0).state(), UNASSIGNED);
assertEquals(routingTable.index("idx").shards().size(), 1);
assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), INITIALIZING);
return clusterState;
}
public static void main(String[] args) {
final int numberOfRuns = 1;
final int numIndices = 5 * 365; // five years
final int numShards = 6;
final int numReplicas = 2;
final int numberOfNodes = 30;
final int numberOfTags = 2;
AllocationService strategy =
ElasticsearchAllocationTestCase.createAllocationService(
ImmutableSettings.EMPTY, new Random(1));
MetaData.Builder mb = MetaData.builder();
for (int i = 1; i <= numIndices; i++) {
mb.put(
IndexMetaData.builder("test_" + i)
.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 <= numberOfNodes; i++) {
nb.put(newNode("node" + i, ImmutableMap.of("tag", "tag_" + (i % numberOfTags))));
}
ClusterState initialClusterState =
ClusterState.builder().metaData(metaData).routingTable(routingTable).nodes(nb).build();
long start = System.currentTimeMillis();
for (int i = 0; i < numberOfRuns; i++) {
logger.info("[{}] starting... ", i);
long runStart = System.currentTimeMillis();
ClusterState clusterState = initialClusterState;
while (clusterState.readOnlyRoutingNodes().hasUnassignedShards()) {
logger.info(
"[{}] remaining unassigned {}",
i,
clusterState.readOnlyRoutingNodes().unassigned().size());
RoutingAllocation.Result result =
strategy.applyStartedShards(
clusterState, clusterState.readOnlyRoutingNodes().shardsWithState(INITIALIZING));
clusterState = ClusterState.builder(clusterState).routingResult(result).build();
result = strategy.reroute(clusterState);
clusterState = ClusterState.builder(clusterState).routingResult(result).build();
}
logger.info(
"[{}] took {}", i, TimeValue.timeValueMillis(System.currentTimeMillis() - runStart));
}
long took = System.currentTimeMillis() - start;
logger.info(
"total took {}, AVG {}",
TimeValue.timeValueMillis(took),
TimeValue.timeValueMillis(took / numberOfRuns));
}
@Override
protected void masterOperation(
final ClusterStateRequest request,
final ClusterState state,
ActionListener<ClusterStateResponse> listener)
throws ElasticsearchException {
ClusterState currentState = clusterService.state();
logger.trace("Serving cluster state request using version {}", currentState.version());
ClusterState.Builder builder = ClusterState.builder(currentState.getClusterName());
builder.version(currentState.version());
if (request.nodes()) {
builder.nodes(currentState.nodes());
}
if (request.routingTable()) {
if (request.indices().length > 0) {
RoutingTable.Builder routingTableBuilder = RoutingTable.builder();
for (String filteredIndex : request.indices()) {
if (currentState.routingTable().getIndicesRouting().containsKey(filteredIndex)) {
routingTableBuilder.add(
currentState.routingTable().getIndicesRouting().get(filteredIndex));
}
}
builder.routingTable(routingTableBuilder);
} else {
builder.routingTable(currentState.routingTable());
}
}
if (request.blocks()) {
builder.blocks(currentState.blocks());
}
if (request.metaData()) {
MetaData.Builder mdBuilder;
if (request.indices().length == 0) {
mdBuilder = MetaData.builder(currentState.metaData());
} else {
mdBuilder = MetaData.builder();
}
if (request.indices().length > 0) {
String[] indices =
currentState
.metaData()
.concreteIndices(IndicesOptions.lenientExpandOpen(), request.indices());
for (String filteredIndex : indices) {
IndexMetaData indexMetaData = currentState.metaData().index(filteredIndex);
if (indexMetaData != null) {
mdBuilder.put(indexMetaData, false);
}
}
}
builder.metaData(mdBuilder);
}
listener.onResponse(new ClusterStateResponse(clusterName, builder.build()));
}
public static void writeTo(ClusterState state, StreamOutput out) throws IOException {
out.writeLong(state.version());
MetaData.Builder.writeTo(state.metaData(), out);
RoutingTable.Builder.writeTo(state.routingTable(), out);
DiscoveryNodes.Builder.writeTo(state.nodes(), out);
ClusterBlocks.Builder.writeClusterBlocks(state.blocks(), out);
state.allocationExplanation().writeTo(out);
}
private void removeIndex(
ClusterBlocks.Builder blocks,
MetaData.Builder metaData,
RoutingTable.Builder routingTable,
IndexMetaData index) {
metaData.remove(index.getIndex());
routingTable.remove(index.getIndex());
blocks.removeIndexBlocks(index.getIndex());
}
/** Creates cluster state with several shards and one replica and all shards STARTED. */
public static ClusterState stateWithAssignedPrimariesAndOneReplica(
String index, int numberOfShards) {
int numberOfNodes = 2; // we need a non-local master to test shard failures
DiscoveryNodes.Builder discoBuilder = DiscoveryNodes.builder();
for (int i = 0; i < numberOfNodes + 1; i++) {
final DiscoveryNode node = newNode(i);
discoBuilder = discoBuilder.put(node);
}
discoBuilder.localNodeId(newNode(0).getId());
discoBuilder.masterNodeId(
newNode(1).getId()); // we need a non-local master to test shard failures
IndexMetaData indexMetaData =
IndexMetaData.builder(index)
.settings(
Settings.builder()
.put(SETTING_VERSION_CREATED, Version.CURRENT)
.put(SETTING_NUMBER_OF_SHARDS, numberOfShards)
.put(SETTING_NUMBER_OF_REPLICAS, 1)
.put(SETTING_CREATION_DATE, System.currentTimeMillis()))
.build();
ClusterState.Builder state = ClusterState.builder(new ClusterName("test"));
state.nodes(discoBuilder);
state.metaData(MetaData.builder().put(indexMetaData, false).generateClusterUuidIfNeeded());
IndexRoutingTable.Builder indexRoutingTableBuilder =
IndexRoutingTable.builder(indexMetaData.getIndex());
for (int i = 0; i < numberOfShards; i++) {
RoutingTable.Builder routing = new RoutingTable.Builder();
routing.addAsNew(indexMetaData);
final ShardId shardId = new ShardId(index, "_na_", i);
IndexShardRoutingTable.Builder indexShardRoutingBuilder =
new IndexShardRoutingTable.Builder(shardId);
indexShardRoutingBuilder.addShard(
TestShardRouting.newShardRouting(
index, i, newNode(0).getId(), null, null, true, ShardRoutingState.STARTED, null));
indexShardRoutingBuilder.addShard(
TestShardRouting.newShardRouting(
index, i, newNode(1).getId(), null, null, false, ShardRoutingState.STARTED, null));
indexRoutingTableBuilder.addIndexShard(indexShardRoutingBuilder.build());
}
state.routingTable(RoutingTable.builder().add(indexRoutingTableBuilder.build()).build());
return state.build();
}
public static ClusterState readFrom(
StreamInput in, @Nullable Settings globalSettings, @Nullable DiscoveryNode localNode)
throws IOException {
Builder builder = new Builder();
builder.version = in.readLong();
builder.metaData = MetaData.Builder.readFrom(in, globalSettings);
builder.routingTable = RoutingTable.Builder.readFrom(in);
builder.nodes = DiscoveryNodes.Builder.readFrom(in, localNode);
builder.blocks = ClusterBlocks.Builder.readClusterBlocks(in);
builder.allocationExplanation = AllocationExplanation.readAllocationExplanation(in);
return builder.build();
}
private void addNewIndex(
ClusterState tribeState,
ClusterBlocks.Builder blocks,
MetaData.Builder metaData,
RoutingTable.Builder routingTable,
IndexMetaData tribeIndex) {
Settings tribeSettings =
Settings.builder().put(tribeIndex.getSettings()).put(TRIBE_NAME, tribeName).build();
metaData.put(IndexMetaData.builder(tribeIndex).settings(tribeSettings));
routingTable.add(tribeState.routingTable().index(tribeIndex.getIndex()));
if (Regex.simpleMatch(blockIndicesMetadata, tribeIndex.getIndex())) {
blocks.addIndexBlock(tribeIndex.getIndex(), IndexMetaData.INDEX_METADATA_BLOCK);
}
if (Regex.simpleMatch(blockIndicesRead, tribeIndex.getIndex())) {
blocks.addIndexBlock(tribeIndex.getIndex(), IndexMetaData.INDEX_READ_BLOCK);
}
if (Regex.simpleMatch(blockIndicesWrite, tribeIndex.getIndex())) {
blocks.addIndexBlock(tribeIndex.getIndex(), IndexMetaData.INDEX_WRITE_BLOCK);
}
}
@Test
public void testMultiIndexUnevenNodes() {
AllocationService strategy =
createAllocationService(
settingsBuilder()
.put("cluster.routing.allocation.concurrent_recoveries", 10)
.put(
ClusterRebalanceAllocationDecider.CLUSTER_ROUTING_ALLOCATION_ALLOW_REBALANCE,
"always")
.put("cluster.routing.allocation.cluster_concurrent_rebalance", -1)
.build());
final int numberOfIndices = 10;
logger.info("Building initial routing table with " + numberOfIndices + " indices");
MetaData.Builder metaDataBuilder = MetaData.builder();
for (int i = 0; i < numberOfIndices; i++) {
metaDataBuilder.put(
IndexMetaData.builder("test" + i)
.settings(settings(Version.CURRENT))
.numberOfShards(1)
.numberOfReplicas(0));
}
MetaData metaData = metaDataBuilder.build();
RoutingTable.Builder routingTableBuilder = RoutingTable.builder();
for (int i = 0; i < numberOfIndices; i++) {
routingTableBuilder.addAsNew(metaData.index("test" + i));
}
RoutingTable routingTable = routingTableBuilder.build();
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.metaData(metaData)
.routingTable(routingTable)
.build();
assertThat(routingTable.indicesRouting().size(), equalTo(numberOfIndices));
logger.info("Starting 3 nodes and rerouting");
clusterState =
ClusterState.builder(clusterState)
.nodes(
DiscoveryNodes.builder()
.put(newNode("node1"))
.put(newNode("node2"))
.put(newNode("node3")))
.build();
RoutingTable prevRoutingTable = routingTable;
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(prevRoutingTable != routingTable, equalTo(true));
for (int i = 0; i < numberOfIndices; i++) {
assertThat(routingTable.index("test" + i).shards().size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).shards().size(), equalTo(1));
assertThat(
routingTable.index("test" + i).shard(0).shards().get(0).state(), equalTo(INITIALIZING));
}
RoutingNodes routingNodes = clusterState.routingNodes();
assertThat(numberOfShardsOfType(routingNodes, INITIALIZING), equalTo(numberOfIndices));
assertThat(
routingNodes.node("node1").numberOfShardsWithState(INITIALIZING),
anyOf(equalTo(3), equalTo(4)));
assertThat(
routingNodes.node("node2").numberOfShardsWithState(INITIALIZING),
anyOf(equalTo(3), equalTo(4)));
assertThat(
routingNodes.node("node2").numberOfShardsWithState(INITIALIZING),
anyOf(equalTo(3), equalTo(4)));
logger.info("Start two more nodes, things should remain the same");
clusterState =
ClusterState.builder(clusterState)
.nodes(
DiscoveryNodes.builder(clusterState.nodes())
.put(newNode("node4"))
.put(newNode("node5")))
.build();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
prevRoutingTable = routingTable;
routingTable = strategy.reroute(clusterState).routingTable();
assertThat(prevRoutingTable == routingTable, equalTo(true));
routingNodes = clusterState.routingNodes();
prevRoutingTable = routingTable;
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(prevRoutingTable != routingTable, equalTo(true));
for (int i = 0; i < numberOfIndices; i++) {
assertThat(routingTable.index("test" + i).shards().size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).shards().size(), equalTo(1));
assertThat(
routingTable.index("test" + i).shard(0).shards().get(0).state(),
anyOf(equalTo(RELOCATING), equalTo(STARTED)));
}
routingNodes = clusterState.routingNodes();
assertThat(
"4 source shard routing are relocating",
numberOfShardsOfType(routingNodes, RELOCATING),
equalTo(4));
assertThat(
"4 target shard routing are initializing",
numberOfShardsOfType(routingNodes, INITIALIZING),
equalTo(4));
logger.info("Now, mark the relocated as started");
prevRoutingTable = routingTable;
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
// routingTable = strategy.reroute(new RoutingStrategyInfo(metaData, routingTable),
// nodes);
assertThat(prevRoutingTable != routingTable, equalTo(true));
for (int i = 0; i < numberOfIndices; i++) {
assertThat(routingTable.index("test" + i).shards().size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).shards().size(), equalTo(1));
assertThat(
routingTable.index("test" + i).shard(0).shards().get(0).state(),
anyOf(equalTo(RELOCATING), equalTo(STARTED)));
}
routingNodes = clusterState.routingNodes();
assertThat(numberOfShardsOfType(routingNodes, STARTED), equalTo(numberOfIndices));
for (RoutingNode routingNode : routingNodes) {
assertThat(routingNode.numberOfShardsWithState(STARTED), equalTo(2));
}
}
@Test
public void testMultiIndexEvenDistribution() {
AllocationService strategy =
createAllocationService(
settingsBuilder()
.put("cluster.routing.allocation.concurrent_recoveries", 10)
.put(
ClusterRebalanceAllocationDecider.CLUSTER_ROUTING_ALLOCATION_ALLOW_REBALANCE,
"always")
.put("cluster.routing.allocation.cluster_concurrent_rebalance", -1)
.build());
final int numberOfIndices = 50;
logger.info("Building initial routing table with " + numberOfIndices + " indices");
MetaData.Builder metaDataBuilder = MetaData.builder();
for (int i = 0; i < numberOfIndices; i++) {
metaDataBuilder.put(
IndexMetaData.builder("test" + i)
.settings(settings(Version.CURRENT))
.numberOfShards(1)
.numberOfReplicas(0));
}
MetaData metaData = metaDataBuilder.build();
RoutingTable.Builder routingTableBuilder = RoutingTable.builder();
for (int i = 0; i < numberOfIndices; i++) {
routingTableBuilder.addAsNew(metaData.index("test" + i));
}
RoutingTable routingTable = routingTableBuilder.build();
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.metaData(metaData)
.routingTable(routingTable)
.build();
assertThat(routingTable.indicesRouting().size(), equalTo(numberOfIndices));
for (int i = 0; i < numberOfIndices; i++) {
assertThat(routingTable.index("test" + i).shards().size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).shards().size(), equalTo(1));
assertThat(
routingTable.index("test" + i).shard(0).shards().get(0).state(), equalTo(UNASSIGNED));
assertThat(
routingTable.index("test" + i).shard(0).shards().get(0).currentNodeId(), nullValue());
}
logger.info("Adding " + (numberOfIndices / 2) + " nodes");
DiscoveryNodes.Builder nodesBuilder = DiscoveryNodes.builder();
List<DiscoveryNode> nodes = newArrayList();
for (int i = 0; i < (numberOfIndices / 2); i++) {
nodesBuilder.put(newNode("node" + i));
}
RoutingTable prevRoutingTable = routingTable;
clusterState = ClusterState.builder(clusterState).nodes(nodesBuilder).build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(prevRoutingTable != routingTable, equalTo(true));
for (int i = 0; i < numberOfIndices; i++) {
assertThat(routingTable.index("test" + i).shards().size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).shards().size(), equalTo(1));
assertThat(
routingTable.index("test" + i).shard(0).shards().get(0).unassigned(), equalTo(false));
assertThat(
routingTable.index("test" + i).shard(0).shards().get(0).state(), equalTo(INITIALIZING));
assertThat(routingTable.index("test" + i).shard(0).shards().get(0).primary(), equalTo(true));
// make sure we still have 2 shards initializing per node on the first 25 nodes
String nodeId = routingTable.index("test" + i).shard(0).shards().get(0).currentNodeId();
int nodeIndex = Integer.parseInt(nodeId.substring("node".length()));
assertThat(nodeIndex, lessThan(25));
}
RoutingNodes routingNodes = clusterState.routingNodes();
Set<String> encounteredIndices = newHashSet();
for (RoutingNode routingNode : routingNodes) {
assertThat(routingNode.numberOfShardsWithState(STARTED), equalTo(0));
assertThat(routingNode.size(), equalTo(2));
// make sure we still have 2 shards initializing per node on the only 25 nodes
int nodeIndex = Integer.parseInt(routingNode.nodeId().substring("node".length()));
assertThat(nodeIndex, lessThan(25));
// check that we don't have a shard associated with a node with the same index name (we have a
// single shard)
for (ShardRouting shardRoutingEntry : routingNode) {
assertThat(encounteredIndices, not(hasItem(shardRoutingEntry.index())));
encounteredIndices.add(shardRoutingEntry.index());
}
}
logger.info("Adding additional " + (numberOfIndices / 2) + " nodes, nothing should change");
nodesBuilder = DiscoveryNodes.builder(clusterState.nodes());
for (int i = (numberOfIndices / 2); i < numberOfIndices; i++) {
nodesBuilder.put(newNode("node" + i));
}
prevRoutingTable = routingTable;
clusterState = ClusterState.builder(clusterState).nodes(nodesBuilder).build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(prevRoutingTable != routingTable, equalTo(false));
logger.info("Marking the shard as started");
prevRoutingTable = routingTable;
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(prevRoutingTable != routingTable, equalTo(true));
int numberOfRelocatingShards = 0;
int numberOfStartedShards = 0;
for (int i = 0; i < numberOfIndices; i++) {
assertThat(routingTable.index("test" + i).shards().size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).size(), equalTo(1));
assertThat(routingTable.index("test" + i).shard(0).shards().size(), equalTo(1));
assertThat(
routingTable.index("test" + i).shard(0).shards().get(0).unassigned(), equalTo(false));
assertThat(
routingTable.index("test" + i).shard(0).shards().get(0).state(),
anyOf(equalTo(STARTED), equalTo(RELOCATING)));
if (routingTable.index("test" + i).shard(0).shards().get(0).state() == STARTED) {
numberOfStartedShards++;
} else if (routingTable.index("test" + i).shard(0).shards().get(0).state() == RELOCATING) {
numberOfRelocatingShards++;
}
assertThat(routingTable.index("test" + i).shard(0).shards().get(0).primary(), equalTo(true));
// make sure we still have 2 shards either relocating or started on the first 25 nodes (still)
String nodeId = routingTable.index("test" + i).shard(0).shards().get(0).currentNodeId();
int nodeIndex = Integer.parseInt(nodeId.substring("node".length()));
assertThat(nodeIndex, lessThan(25));
}
assertThat(numberOfRelocatingShards, equalTo(25));
assertThat(numberOfStartedShards, equalTo(25));
}
/**
* Creates cluster state with and index that has one shard and #(replicaStates) replicas
*
* @param index name of the index
* @param activePrimaryLocal if active primary should coincide with the local node in the cluster
* state
* @param primaryState state of primary
* @param replicaStates states of the replicas. length of this array determines also the number of
* replicas
*/
public static ClusterState state(
String index,
boolean activePrimaryLocal,
ShardRoutingState primaryState,
ShardRoutingState... replicaStates) {
final int numberOfReplicas = replicaStates.length;
int numberOfNodes = numberOfReplicas + 1;
if (primaryState == ShardRoutingState.RELOCATING) {
numberOfNodes++;
}
for (ShardRoutingState state : replicaStates) {
if (state == ShardRoutingState.RELOCATING) {
numberOfNodes++;
}
}
numberOfNodes = Math.max(2, numberOfNodes); // we need a non-local master to test shard failures
final ShardId shardId = new ShardId(index, "_na_", 0);
DiscoveryNodes.Builder discoBuilder = DiscoveryNodes.builder();
Set<String> unassignedNodes = new HashSet<>();
for (int i = 0; i < numberOfNodes + 1; i++) {
final DiscoveryNode node = newNode(i);
discoBuilder = discoBuilder.put(node);
unassignedNodes.add(node.getId());
}
discoBuilder.localNodeId(newNode(0).getId());
discoBuilder.masterNodeId(
newNode(1).getId()); // we need a non-local master to test shard failures
final int primaryTerm = 1 + randomInt(200);
IndexMetaData indexMetaData =
IndexMetaData.builder(index)
.settings(
Settings.builder()
.put(SETTING_VERSION_CREATED, Version.CURRENT)
.put(SETTING_NUMBER_OF_SHARDS, 1)
.put(SETTING_NUMBER_OF_REPLICAS, numberOfReplicas)
.put(SETTING_CREATION_DATE, System.currentTimeMillis()))
.primaryTerm(0, primaryTerm)
.build();
RoutingTable.Builder routing = new RoutingTable.Builder();
routing.addAsNew(indexMetaData);
IndexShardRoutingTable.Builder indexShardRoutingBuilder =
new IndexShardRoutingTable.Builder(shardId);
String primaryNode = null;
String relocatingNode = null;
UnassignedInfo unassignedInfo = null;
if (primaryState != ShardRoutingState.UNASSIGNED) {
if (activePrimaryLocal) {
primaryNode = newNode(0).getId();
unassignedNodes.remove(primaryNode);
} else {
Set<String> unassignedNodesExecludingPrimary = new HashSet<>(unassignedNodes);
unassignedNodesExecludingPrimary.remove(newNode(0).getId());
primaryNode = selectAndRemove(unassignedNodesExecludingPrimary);
}
if (primaryState == ShardRoutingState.RELOCATING) {
relocatingNode = selectAndRemove(unassignedNodes);
}
} else {
unassignedInfo = new UnassignedInfo(UnassignedInfo.Reason.INDEX_CREATED, null);
}
indexShardRoutingBuilder.addShard(
TestShardRouting.newShardRouting(
index, 0, primaryNode, relocatingNode, null, true, primaryState, unassignedInfo));
for (ShardRoutingState replicaState : replicaStates) {
String replicaNode = null;
relocatingNode = null;
unassignedInfo = null;
if (replicaState != ShardRoutingState.UNASSIGNED) {
assert primaryNode != null : "a replica is assigned but the primary isn't";
replicaNode = selectAndRemove(unassignedNodes);
if (replicaState == ShardRoutingState.RELOCATING) {
relocatingNode = selectAndRemove(unassignedNodes);
}
} else {
unassignedInfo = new UnassignedInfo(UnassignedInfo.Reason.INDEX_CREATED, null);
}
indexShardRoutingBuilder.addShard(
TestShardRouting.newShardRouting(
index,
shardId.id(),
replicaNode,
relocatingNode,
null,
false,
replicaState,
unassignedInfo));
}
ClusterState.Builder state = ClusterState.builder(new ClusterName("test"));
state.nodes(discoBuilder);
state.metaData(MetaData.builder().put(indexMetaData, false).generateClusterUuidIfNeeded());
state.routingTable(
RoutingTable.builder()
.add(
IndexRoutingTable.builder(indexMetaData.getIndex())
.addIndexShard(indexShardRoutingBuilder.build()))
.build());
return state.build();
}
private ClusterState applyUpdate(ClusterState currentState, ClusterChangedEvent task) {
boolean clusterStateChanged = false;
ClusterState tribeState = task.state();
DiscoveryNodes.Builder nodes = DiscoveryNodes.builder(currentState.nodes());
// -- merge nodes
// go over existing nodes, and see if they need to be removed
for (DiscoveryNode discoNode : currentState.nodes()) {
String markedTribeName = discoNode.attributes().get(TRIBE_NAME);
if (markedTribeName != null && markedTribeName.equals(tribeName)) {
if (tribeState.nodes().get(discoNode.id()) == null) {
clusterStateChanged = true;
logger.info("[{}] removing node [{}]", tribeName, discoNode);
nodes.remove(discoNode.id());
}
}
}
// go over tribe nodes, and see if they need to be added
for (DiscoveryNode tribe : tribeState.nodes()) {
if (currentState.nodes().get(tribe.id()) == null) {
// a new node, add it, but also add the tribe name to the attributes
Map<String, String> tribeAttr = new HashMap<>();
for (ObjectObjectCursor<String, String> attr : tribe.attributes()) {
tribeAttr.put(attr.key, attr.value);
}
tribeAttr.put(TRIBE_NAME, tribeName);
DiscoveryNode discoNode =
new DiscoveryNode(
tribe.name(),
tribe.id(),
tribe.getHostName(),
tribe.getHostAddress(),
tribe.address(),
unmodifiableMap(tribeAttr),
tribe.version());
clusterStateChanged = true;
logger.info("[{}] adding node [{}]", tribeName, discoNode);
nodes.put(discoNode);
}
}
// -- merge metadata
ClusterBlocks.Builder blocks = ClusterBlocks.builder().blocks(currentState.blocks());
MetaData.Builder metaData = MetaData.builder(currentState.metaData());
RoutingTable.Builder routingTable = RoutingTable.builder(currentState.routingTable());
// go over existing indices, and see if they need to be removed
for (IndexMetaData index : currentState.metaData()) {
String markedTribeName = index.getSettings().get(TRIBE_NAME);
if (markedTribeName != null && markedTribeName.equals(tribeName)) {
IndexMetaData tribeIndex = tribeState.metaData().index(index.getIndex());
clusterStateChanged = true;
if (tribeIndex == null || tribeIndex.getState() == IndexMetaData.State.CLOSE) {
logger.info("[{}] removing index [{}]", tribeName, index.getIndex());
removeIndex(blocks, metaData, routingTable, index);
} else {
// always make sure to update the metadata and routing table, in case
// there are changes in them (new mapping, shards moving from initializing to started)
routingTable.add(tribeState.routingTable().index(index.getIndex()));
Settings tribeSettings =
Settings.builder().put(tribeIndex.getSettings()).put(TRIBE_NAME, tribeName).build();
metaData.put(IndexMetaData.builder(tribeIndex).settings(tribeSettings));
}
}
}
// go over tribe one, and see if they need to be added
for (IndexMetaData tribeIndex : tribeState.metaData()) {
// if there is no routing table yet, do nothing with it...
IndexRoutingTable table = tribeState.routingTable().index(tribeIndex.getIndex());
if (table == null) {
continue;
}
final IndexMetaData indexMetaData = currentState.metaData().index(tribeIndex.getIndex());
if (indexMetaData == null) {
if (!droppedIndices.contains(tribeIndex.getIndex())) {
// a new index, add it, and add the tribe name as a setting
clusterStateChanged = true;
logger.info("[{}] adding index [{}]", tribeName, tribeIndex.getIndex());
addNewIndex(tribeState, blocks, metaData, routingTable, tribeIndex);
}
} else {
String existingFromTribe = indexMetaData.getSettings().get(TRIBE_NAME);
if (!tribeName.equals(existingFromTribe)) {
// we have a potential conflict on index names, decide what to do...
if (ON_CONFLICT_ANY.equals(onConflict)) {
// we chose any tribe, carry on
} else if (ON_CONFLICT_DROP.equals(onConflict)) {
// drop the indices, there is a conflict
clusterStateChanged = true;
logger.info(
"[{}] dropping index [{}] due to conflict with [{}]",
tribeName,
tribeIndex.getIndex(),
existingFromTribe);
removeIndex(blocks, metaData, routingTable, tribeIndex);
droppedIndices.add(tribeIndex.getIndex());
} else if (onConflict.startsWith(ON_CONFLICT_PREFER)) {
// on conflict, prefer a tribe...
String preferredTribeName = onConflict.substring(ON_CONFLICT_PREFER.length());
if (tribeName.equals(preferredTribeName)) {
// the new one is hte preferred one, replace...
clusterStateChanged = true;
logger.info(
"[{}] adding index [{}], preferred over [{}]",
tribeName,
tribeIndex.getIndex(),
existingFromTribe);
removeIndex(blocks, metaData, routingTable, tribeIndex);
addNewIndex(tribeState, blocks, metaData, routingTable, tribeIndex);
} // else: either the existing one is the preferred one, or we haven't seen one, carry
// on
}
}
}
}
if (!clusterStateChanged) {
return currentState;
} else {
return ClusterState.builder(currentState)
.incrementVersion()
.blocks(blocks)
.nodes(nodes)
.metaData(metaData)
.routingTable(routingTable.build())
.build();
}
}
public Builder routingTable(RoutingTable.Builder routingTable) {
return routingTable(routingTable.build());
}
@Test
public void testNoRebalanceOnPrimaryOverload() {
Settings.Builder settings = settingsBuilder();
AllocationService strategy =
new AllocationService(
settings.build(),
randomAllocationDeciders(
settings.build(),
new NodeSettingsService(Settings.Builder.EMPTY_SETTINGS),
getRandom()),
new ShardsAllocators(
settings.build(),
NoopGatewayAllocator.INSTANCE,
new ShardsAllocator() {
@Override
public boolean rebalance(RoutingAllocation allocation) {
return false;
}
@Override
public boolean move(
ShardRouting shardRouting, RoutingNode node, RoutingAllocation allocation) {
return false;
}
@Override
public void applyStartedShards(StartedRerouteAllocation allocation) {}
@Override
public void applyFailedShards(FailedRerouteAllocation allocation) {}
/*
* // this allocator tries to rebuild this scenario where a rebalance is
* // triggered solely by the primary overload on node [1] where a shard
* // is rebalanced to node 0
routing_nodes:
-----node_id[0][V]
--------[test][0], node[0], [R], s[STARTED]
--------[test][4], node[0], [R], s[STARTED]
-----node_id[1][V]
--------[test][0], node[1], [P], s[STARTED]
--------[test][1], node[1], [P], s[STARTED]
--------[test][3], node[1], [R], s[STARTED]
-----node_id[2][V]
--------[test][1], node[2], [R], s[STARTED]
--------[test][2], node[2], [R], s[STARTED]
--------[test][4], node[2], [P], s[STARTED]
-----node_id[3][V]
--------[test][2], node[3], [P], s[STARTED]
--------[test][3], node[3], [P], s[STARTED]
---- unassigned
*/
@Override
public boolean allocateUnassigned(RoutingAllocation allocation) {
RoutingNodes.UnassignedShards unassigned =
allocation.routingNodes().unassigned();
boolean changed = !unassigned.isEmpty();
for (ShardRouting sr : unassigned) {
switch (sr.id()) {
case 0:
if (sr.primary()) {
allocation.routingNodes().initialize(sr, "node1");
} else {
allocation.routingNodes().initialize(sr, "node0");
}
break;
case 1:
if (sr.primary()) {
allocation.routingNodes().initialize(sr, "node1");
} else {
allocation.routingNodes().initialize(sr, "node2");
}
break;
case 2:
if (sr.primary()) {
allocation.routingNodes().initialize(sr, "node3");
} else {
allocation.routingNodes().initialize(sr, "node2");
}
break;
case 3:
if (sr.primary()) {
allocation.routingNodes().initialize(sr, "node3");
} else {
allocation.routingNodes().initialize(sr, "node1");
}
break;
case 4:
if (sr.primary()) {
allocation.routingNodes().initialize(sr, "node2");
} else {
allocation.routingNodes().initialize(sr, "node0");
}
break;
}
}
unassigned.clear();
return changed;
}
}),
EmptyClusterInfoService.INSTANCE);
MetaData.Builder metaDataBuilder = MetaData.builder();
RoutingTable.Builder routingTableBuilder = RoutingTable.builder();
IndexMetaData.Builder indexMeta =
IndexMetaData.builder("test")
.settings(settings(Version.CURRENT))
.numberOfShards(5)
.numberOfReplicas(1);
metaDataBuilder = metaDataBuilder.put(indexMeta);
MetaData metaData = metaDataBuilder.build();
for (ObjectCursor<IndexMetaData> cursor : metaData.indices().values()) {
routingTableBuilder.addAsNew(cursor.value);
}
RoutingTable routingTable = routingTableBuilder.build();
DiscoveryNodes.Builder nodes = DiscoveryNodes.builder();
for (int i = 0; i < 4; i++) {
DiscoveryNode node = newNode("node" + i);
nodes.put(node);
}
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.nodes(nodes)
.metaData(metaData)
.routingTable(routingTable)
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
RoutingNodes routingNodes = clusterState.getRoutingNodes();
for (RoutingNode routingNode : routingNodes) {
for (ShardRouting shardRouting : routingNode) {
assertThat(shardRouting.state(), Matchers.equalTo(ShardRoutingState.INITIALIZING));
}
}
strategy = createAllocationService(settings.build());
logger.info("use the new allocator and check if it moves shards");
routingNodes = clusterState.getRoutingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
for (RoutingNode routingNode : routingNodes) {
for (ShardRouting shardRouting : routingNode) {
assertThat(shardRouting.state(), Matchers.equalTo(ShardRoutingState.STARTED));
}
}
logger.info("start the replica shards");
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
for (RoutingNode routingNode : routingNodes) {
for (ShardRouting shardRouting : routingNode) {
assertThat(shardRouting.state(), Matchers.equalTo(ShardRoutingState.STARTED));
}
}
logger.info("rebalancing");
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
for (RoutingNode routingNode : routingNodes) {
for (ShardRouting shardRouting : routingNode) {
assertThat(shardRouting.state(), Matchers.equalTo(ShardRoutingState.STARTED));
}
}
}
private ClusterState initCluster(AllocationService strategy) {
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 routingTable = routingTableBuilder.build();
logger.info("start " + numberOfNodes + " nodes");
DiscoveryNodes.Builder nodes = DiscoveryNodes.builder();
for (int i = 0; i < numberOfNodes; i++) {
nodes.put(newNode("node" + i));
}
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.nodes(nodes)
.metaData(metaData)
.routingTable(routingTable)
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
RoutingNodes routingNodes = clusterState.getRoutingNodes();
logger.info("restart all the primary shards, replicas will start initializing");
routingNodes = clusterState.getRoutingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
logger.info("start the replica shards");
routingNodes = clusterState.getRoutingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
logger.info("complete rebalancing");
RoutingTable prev = routingTable;
while (true) {
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
if (routingTable == prev) break;
prev = routingTable;
}
return clusterState;
}