@Test
public void testClusterStateSerialization() 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"))
.localNodeId("node1")
.masterNodeId("node2")
.build();
ClusterState clusterState =
ClusterState.builder().nodes(nodes).metaData(metaData).routingTable(routingTable).build();
AllocationService strategy = createAllocationService();
clusterState =
ClusterState.builder(clusterState)
.routingTable(strategy.reroute(clusterState).routingTable())
.build();
ClusterState serializedClusterState =
ClusterState.Builder.fromBytes(
ClusterState.Builder.toBytes(clusterState), newNode("node1"));
assertThat(
serializedClusterState.routingTable().prettyPrint(),
equalTo(clusterState.routingTable().prettyPrint()));
}
private ClusterState addNode(ClusterState clusterState, AllocationService strategy) {
logger.info(
"now, start 1 more node, check that rebalancing will happen because we set it to always");
clusterState =
ClusterState.builder(clusterState)
.nodes(
DiscoveryNodes.builder(clusterState.nodes()).put(newNode("node" + numberOfNodes)))
.build();
RoutingTable routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
RoutingNodes routingNodes = clusterState.getRoutingNodes();
// move initializing to started
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;
}
@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()));
}
@Test
public void testSingleIndexShardFailed() {
AllocationService strategy =
createAllocationService(
settingsBuilder().put("cluster.routing.allocation.concurrent_recoveries", 10).build());
logger.info("Building initial routing table");
MetaData metaData =
MetaData.builder()
.put(IndexMetaData.builder("test").numberOfShards(1).numberOfReplicas(0))
.build();
RoutingTable routingTable = RoutingTable.builder().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
ClusterState.builder().metaData(metaData).routingTable(routingTable).build();
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().get(0).state(), equalTo(UNASSIGNED));
assertThat(routingTable.index("test").shard(0).shards().get(0).currentNodeId(), nullValue());
logger.info("Adding one node and rerouting");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder().put(newNode("node1")))
.build();
RoutingTable prevRoutingTable = routingTable;
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(prevRoutingTable != routingTable, equalTo(true));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().get(0).unassigned(), equalTo(false));
assertThat(routingTable.index("test").shard(0).shards().get(0).state(), equalTo(INITIALIZING));
assertThat(
routingTable.index("test").shard(0).shards().get(0).currentNodeId(), equalTo("node1"));
logger.info("Marking the shard as failed");
RoutingNodes routingNodes = clusterState.routingNodes();
prevRoutingTable = routingTable;
routingTable =
strategy
.applyFailedShard(
clusterState, routingNodes.node("node1").shardsWithState(INITIALIZING).get(0))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(prevRoutingTable != routingTable, equalTo(true));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().get(0).state(), equalTo(UNASSIGNED));
assertThat(routingTable.index("test").shard(0).shards().get(0).currentNodeId(), nullValue());
}
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));
}
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;
}
private ClusterState removeNodes(ClusterState clusterState, AllocationService strategy) {
logger.info("Removing half the nodes (" + (numberOfNodes + 1) / 2 + ")");
DiscoveryNodes.Builder nodes = DiscoveryNodes.builder(clusterState.nodes());
for (int i = (numberOfNodes + 1) / 2; i <= numberOfNodes; i++) {
nodes.remove("node" + i);
}
clusterState = ClusterState.builder(clusterState).nodes(nodes.build()).build();
RoutingNodes routingNodes = clusterState.getRoutingNodes();
logger.info("start all the primary shards, replicas will start initializing");
RoutingTable routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
logger.info("start the replica shards");
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
logger.info("rebalancing");
routingTable = strategy.reroute(clusterState).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;
}
@Test
public void simpleFlagTests() {
AllocationService allocation =
new AllocationService(
settingsBuilder().put("cluster.routing.allocation.concurrent_recoveries", 10).build());
logger.info("creating an index with 1 shard, no replica");
MetaData metaData =
newMetaDataBuilder()
.put(newIndexMetaDataBuilder("test").numberOfShards(1).numberOfReplicas(0))
.build();
RoutingTable routingTable = routingTable().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
newClusterStateBuilder().metaData(metaData).routingTable(routingTable).build();
assertThat(
clusterState.routingTable().index("test").shard(0).primaryAllocatedPostApi(),
equalTo(false));
logger.info("adding two nodes and performing rerouting");
clusterState =
newClusterStateBuilder()
.state(clusterState)
.nodes(newNodesBuilder().put(newNode("node1")).put(newNode("node2")))
.build();
RoutingAllocation.Result rerouteResult = allocation.reroute(clusterState);
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(
clusterState.routingTable().index("test").shard(0).primaryAllocatedPostApi(),
equalTo(false));
logger.info("start primary shard");
rerouteResult =
allocation.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(
clusterState.routingTable().index("test").shard(0).primaryAllocatedPostApi(),
equalTo(true));
}
@Test
public void sameHost() {
AllocationService strategy =
new AllocationService(
settingsBuilder().put(SameShardAllocationDecider.SAME_HOST_SETTING, true).build());
MetaData metaData =
newMetaDataBuilder()
.put(newIndexMetaDataBuilder("test").numberOfShards(2).numberOfReplicas(1))
.build();
RoutingTable routingTable = routingTable().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
newClusterStateBuilder().metaData(metaData).routingTable(routingTable).build();
logger.info("--> adding two nodes with the same host");
clusterState =
newClusterStateBuilder()
.state(clusterState)
.nodes(
newNodesBuilder()
.put(newNode("node1", new InetSocketTransportAddress("test1", 80)))
.put(newNode("node2", new InetSocketTransportAddress("test1", 80))))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(
clusterState.readOnlyRoutingNodes().numberOfShardsOfType(ShardRoutingState.INITIALIZING),
equalTo(2));
logger.info(
"--> start all primary shards, no replica will be started since its on the same host");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.readOnlyRoutingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(
clusterState.readOnlyRoutingNodes().numberOfShardsOfType(ShardRoutingState.STARTED),
equalTo(2));
assertThat(
clusterState.readOnlyRoutingNodes().numberOfShardsOfType(ShardRoutingState.INITIALIZING),
equalTo(0));
logger.info("--> add another node, with a different host, replicas will be allocating");
clusterState =
newClusterStateBuilder()
.state(clusterState)
.nodes(
newNodesBuilder()
.putAll(clusterState.nodes())
.put(newNode("node3", new InetSocketTransportAddress("test2", 80))))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(
clusterState.readOnlyRoutingNodes().numberOfShardsOfType(ShardRoutingState.STARTED),
equalTo(2));
assertThat(
clusterState.readOnlyRoutingNodes().numberOfShardsOfType(ShardRoutingState.INITIALIZING),
equalTo(2));
for (MutableShardRouting shardRouting :
clusterState.readOnlyRoutingNodes().shardsWithState(INITIALIZING)) {
assertThat(shardRouting.currentNodeId(), equalTo("node3"));
}
}
public void testSimple() {
AllocationService strategy =
createAllocationService(
Settings.builder()
.put(
ClusterRebalanceAllocationDecider
.CLUSTER_ROUTING_ALLOCATION_ALLOW_REBALANCE_SETTING.getKey(),
ClusterRebalanceAllocationDecider.ClusterRebalanceType.ALWAYS.toString())
.build());
MetaData metaData =
MetaData.builder()
.put(
IndexMetaData.builder("test1")
.settings(settings(Version.CURRENT))
.numberOfShards(1)
.numberOfReplicas(1))
.put(
IndexMetaData.builder("test2")
.settings(settings(Version.CURRENT))
.numberOfShards(1)
.numberOfReplicas(1))
.build();
RoutingTable routingTable =
RoutingTable.builder()
.addAsNew(metaData.index("test1"))
.addAsNew(metaData.index("test2"))
.build();
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.metaData(metaData)
.routingTable(routingTable)
.build();
logger.info("start two nodes");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder().put(newNode("node1")).put(newNode("node2")))
.build();
routingTable = strategy.reroute(clusterState, "reroute").routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
for (int i = 0; i < routingTable.index("test1").shards().size(); i++) {
assertThat(routingTable.index("test1").shard(i).shards().size(), equalTo(2));
assertThat(
routingTable.index("test1").shard(i).primaryShard().state(), equalTo(INITIALIZING));
assertThat(
routingTable.index("test1").shard(i).replicaShards().get(0).state(), equalTo(UNASSIGNED));
}
for (int i = 0; i < routingTable.index("test2").shards().size(); i++) {
assertThat(routingTable.index("test2").shard(i).shards().size(), equalTo(2));
assertThat(
routingTable.index("test2").shard(i).primaryShard().state(), equalTo(INITIALIZING));
assertThat(
routingTable.index("test2").shard(i).replicaShards().get(0).state(), equalTo(UNASSIGNED));
}
logger.info("start all the primary shards for test1, replicas will start initializing");
RoutingNodes routingNodes = clusterState.getRoutingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState("test1", INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
for (int i = 0; i < routingTable.index("test1").shards().size(); i++) {
assertThat(routingTable.index("test1").shard(i).shards().size(), equalTo(2));
assertThat(routingTable.index("test1").shard(i).primaryShard().state(), equalTo(STARTED));
assertThat(
routingTable.index("test1").shard(i).replicaShards().get(0).state(),
equalTo(INITIALIZING));
}
for (int i = 0; i < routingTable.index("test2").shards().size(); i++) {
assertThat(routingTable.index("test2").shard(i).shards().size(), equalTo(2));
assertThat(
routingTable.index("test2").shard(i).primaryShard().state(), equalTo(INITIALIZING));
assertThat(
routingTable.index("test2").shard(i).replicaShards().get(0).state(), equalTo(UNASSIGNED));
}
}
/**
* Tests that higher prioritized primaries and replicas are allocated first even on the balanced
* shard allocator See https://github.com/elastic/elasticsearch/issues/13249 for details
*/
public void testPrioritizedIndicesAllocatedFirst() {
AllocationService allocation =
createAllocationService(
Settings.builder()
.put(
ThrottlingAllocationDecider
.CLUSTER_ROUTING_ALLOCATION_NODE_CONCURRENT_RECOVERIES_SETTING.getKey(),
1)
.put(
ThrottlingAllocationDecider
.CLUSTER_ROUTING_ALLOCATION_NODE_CONCURRENT_OUTGOING_RECOVERIES_SETTING
.getKey(),
10)
.put(
ThrottlingAllocationDecider
.CLUSTER_ROUTING_ALLOCATION_NODE_INITIAL_PRIMARIES_RECOVERIES_SETTING
.getKey(),
1)
.put(
ThrottlingAllocationDecider
.CLUSTER_ROUTING_ALLOCATION_NODE_CONCURRENT_INCOMING_RECOVERIES_SETTING
.getKey(),
1)
.build());
final String highPriorityName;
final String lowPriorityName;
final int priorityFirst;
final int prioritySecond;
if (randomBoolean()) {
highPriorityName = "first";
lowPriorityName = "second";
prioritySecond = 1;
priorityFirst = 100;
} else {
lowPriorityName = "first";
highPriorityName = "second";
prioritySecond = 100;
priorityFirst = 1;
}
MetaData metaData =
MetaData.builder()
.put(
IndexMetaData.builder("first")
.settings(
settings(Version.CURRENT)
.put(IndexMetaData.SETTING_PRIORITY, priorityFirst))
.numberOfShards(2)
.numberOfReplicas(1))
.put(
IndexMetaData.builder("second")
.settings(
settings(Version.CURRENT)
.put(IndexMetaData.SETTING_PRIORITY, prioritySecond))
.numberOfShards(2)
.numberOfReplicas(1))
.build();
RoutingTable routingTable =
RoutingTable.builder()
.addAsNew(metaData.index("first"))
.addAsNew(metaData.index("second"))
.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().put(newNode("node1")).put(newNode("node2")))
.build();
RoutingAllocation.Result rerouteResult = allocation.reroute(clusterState, "reroute");
clusterState =
ClusterState.builder(clusterState).routingTable(rerouteResult.routingTable()).build();
routingTable = allocation.reroute(clusterState, "reroute").routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertEquals(2, clusterState.getRoutingNodes().shardsWithState(INITIALIZING).size());
assertEquals(
highPriorityName,
clusterState.getRoutingNodes().shardsWithState(INITIALIZING).get(0).getIndexName());
assertEquals(
highPriorityName,
clusterState.getRoutingNodes().shardsWithState(INITIALIZING).get(1).getIndexName());
routingTable =
allocation
.applyStartedShards(
clusterState, clusterState.getRoutingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertEquals(2, clusterState.getRoutingNodes().shardsWithState(INITIALIZING).size());
assertEquals(
lowPriorityName,
clusterState.getRoutingNodes().shardsWithState(INITIALIZING).get(0).getIndexName());
assertEquals(
lowPriorityName,
clusterState.getRoutingNodes().shardsWithState(INITIALIZING).get(1).getIndexName());
routingTable =
allocation
.applyStartedShards(
clusterState, clusterState.getRoutingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertEquals(
clusterState.getRoutingNodes().shardsWithState(INITIALIZING).toString(),
2,
clusterState.getRoutingNodes().shardsWithState(INITIALIZING).size());
assertEquals(
highPriorityName,
clusterState.getRoutingNodes().shardsWithState(INITIALIZING).get(0).getIndexName());
assertEquals(
highPriorityName,
clusterState.getRoutingNodes().shardsWithState(INITIALIZING).get(1).getIndexName());
routingTable =
allocation
.applyStartedShards(
clusterState, clusterState.getRoutingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertEquals(2, clusterState.getRoutingNodes().shardsWithState(INITIALIZING).size());
assertEquals(
lowPriorityName,
clusterState.getRoutingNodes().shardsWithState(INITIALIZING).get(0).getIndexName());
assertEquals(
lowPriorityName,
clusterState.getRoutingNodes().shardsWithState(INITIALIZING).get(1).getIndexName());
}
@Test
public void testPrimaryRecoveryThrottling() {
AllocationService strategy =
new AllocationService(
settingsBuilder()
.put("cluster.routing.allocation.node_concurrent_recoveries", 3)
.put("cluster.routing.allocation.node_initial_primaries_recoveries", 3)
.build());
logger.info("Building initial routing table");
MetaData metaData =
newMetaDataBuilder()
.put(newIndexMetaDataBuilder("test").numberOfShards(10).numberOfReplicas(1))
.build();
RoutingTable routingTable =
routingTable()
.add(indexRoutingTable("test").initializeEmpty(metaData.index("test")))
.build();
ClusterState clusterState =
newClusterStateBuilder().metaData(metaData).routingTable(routingTable).build();
logger.info("start one node, do reroute, only 3 should initialize");
clusterState =
newClusterStateBuilder()
.state(clusterState)
.nodes(newNodesBuilder().put(RoutingAllocationTests.newNode("node1")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(routingTable.shardsWithState(STARTED).size(), equalTo(0));
assertThat(routingTable.shardsWithState(INITIALIZING).size(), equalTo(3));
assertThat(routingTable.shardsWithState(UNASSIGNED).size(), equalTo(17));
logger.info("start initializing, another 3 should initialize");
routingTable =
strategy
.applyStartedShards(clusterState, routingTable.shardsWithState(INITIALIZING))
.routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(routingTable.shardsWithState(STARTED).size(), equalTo(3));
assertThat(routingTable.shardsWithState(INITIALIZING).size(), equalTo(3));
assertThat(routingTable.shardsWithState(UNASSIGNED).size(), equalTo(14));
logger.info("start initializing, another 3 should initialize");
routingTable =
strategy
.applyStartedShards(clusterState, routingTable.shardsWithState(INITIALIZING))
.routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(routingTable.shardsWithState(STARTED).size(), equalTo(6));
assertThat(routingTable.shardsWithState(INITIALIZING).size(), equalTo(3));
assertThat(routingTable.shardsWithState(UNASSIGNED).size(), equalTo(11));
logger.info("start initializing, another 1 should initialize");
routingTable =
strategy
.applyStartedShards(clusterState, routingTable.shardsWithState(INITIALIZING))
.routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(routingTable.shardsWithState(STARTED).size(), equalTo(9));
assertThat(routingTable.shardsWithState(INITIALIZING).size(), equalTo(1));
assertThat(routingTable.shardsWithState(UNASSIGNED).size(), equalTo(10));
logger.info("start initializing, all primaries should be started");
routingTable =
strategy
.applyStartedShards(clusterState, routingTable.shardsWithState(INITIALIZING))
.routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(routingTable.shardsWithState(STARTED).size(), equalTo(10));
assertThat(routingTable.shardsWithState(INITIALIZING).size(), equalTo(0));
assertThat(routingTable.shardsWithState(UNASSIGNED).size(), equalTo(10));
}
@Test
public void indexLevelShardsLimitRemain() {
AllocationService strategy =
new AllocationService(
settingsBuilder()
.put("cluster.routing.allocation.concurrent_recoveries", 10)
.put("cluster.routing.allocation.node_initial_primaries_recoveries", 10)
.put("cluster.routing.allocation.cluster_concurrent_rebalance", -1)
.put("cluster.routing.allocation.balance.index", 0.0f)
.put("cluster.routing.allocation.balance.replica", 1.0f)
.put("cluster.routing.allocation.balance.primary", 0.0f)
.build());
logger.info("Building initial routing table");
MetaData metaData =
newMetaDataBuilder()
.put(
newIndexMetaDataBuilder("test")
.settings(
ImmutableSettings.settingsBuilder()
.put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, 5)
.put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 0)))
.build();
RoutingTable routingTable = routingTable().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
newClusterStateBuilder().metaData(metaData).routingTable(routingTable).build();
logger.info("Adding one node and reroute");
clusterState =
newClusterStateBuilder()
.state(clusterState)
.nodes(newNodesBuilder().put(newNode("node1")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
logger.info("Start the primary shards");
RoutingNodes routingNodes = clusterState.routingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(clusterState.readOnlyRoutingNodes().numberOfShardsOfType(STARTED), equalTo(5));
logger.info("add another index with 5 shards");
metaData =
newMetaDataBuilder()
.metaData(metaData)
.put(
newIndexMetaDataBuilder("test1")
.settings(
ImmutableSettings.settingsBuilder()
.put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, 5)
.put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 0)))
.build();
routingTable =
routingTable().routingTable(routingTable).addAsNew(metaData.index("test1")).build();
clusterState =
newClusterStateBuilder()
.state(clusterState)
.metaData(metaData)
.routingTable(routingTable)
.build();
logger.info("Add another one node and reroute");
clusterState =
newClusterStateBuilder()
.state(clusterState)
.nodes(newNodesBuilder().putAll(clusterState.nodes()).put(newNode("node2")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.routingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(clusterState.readOnlyRoutingNodes().numberOfShardsOfType(STARTED), equalTo(10));
for (MutableShardRouting shardRouting : clusterState.readOnlyRoutingNodes().node("node1")) {
assertThat(shardRouting.index(), equalTo("test"));
}
for (MutableShardRouting shardRouting : clusterState.readOnlyRoutingNodes().node("node2")) {
assertThat(shardRouting.index(), equalTo("test1"));
}
logger.info(
"update "
+ ShardsLimitAllocationDecider.INDEX_TOTAL_SHARDS_PER_NODE
+ " for test, see that things move");
metaData =
newMetaDataBuilder()
.metaData(metaData)
.put(
newIndexMetaDataBuilder("test")
.settings(
ImmutableSettings.settingsBuilder()
.put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, 5)
.put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 0)
.put(ShardsLimitAllocationDecider.INDEX_TOTAL_SHARDS_PER_NODE, 3)))
.build();
clusterState = newClusterStateBuilder().state(clusterState).metaData(metaData).build();
logger.info("reroute after setting");
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(
clusterState.readOnlyRoutingNodes().node("node1").numberOfShardsWithState(STARTED),
equalTo(3));
assertThat(
clusterState.readOnlyRoutingNodes().node("node1").numberOfShardsWithState(RELOCATING),
equalTo(2));
assertThat(
clusterState.readOnlyRoutingNodes().node("node2").numberOfShardsWithState(RELOCATING),
equalTo(2));
assertThat(
clusterState.readOnlyRoutingNodes().node("node2").numberOfShardsWithState(STARTED),
equalTo(3));
// the first move will destroy the balance and the balancer will move 2 shards from node2 to
// node one right after
// moving the nodes to node2 since we consider INITIALIZING nodes during rebalance
routingNodes = clusterState.routingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
// now we are done compared to EvenShardCountAllocator since the Balancer is not soely based on
// the average
assertThat(
clusterState.readOnlyRoutingNodes().node("node1").numberOfShardsWithState(STARTED),
equalTo(5));
assertThat(
clusterState.readOnlyRoutingNodes().node("node2").numberOfShardsWithState(STARTED),
equalTo(5));
}
public void testBalanceAllNodesStartedAddIndex() {
AllocationService strategy =
createAllocationService(
Settings.builder()
.put("cluster.routing.allocation.node_concurrent_recoveries", 10)
.put("cluster.routing.allocation.node_initial_primaries_recoveries", 10)
.put(
ClusterRebalanceAllocationDecider
.CLUSTER_ROUTING_ALLOCATION_ALLOW_REBALANCE_SETTING.getKey(),
"always")
.put("cluster.routing.allocation.cluster_concurrent_rebalance", -1)
.build());
logger.info("Building initial routing table");
MetaData metaData =
MetaData.builder()
.put(
IndexMetaData.builder("test")
.settings(settings(Version.CURRENT))
.numberOfShards(3)
.numberOfReplicas(1))
.build();
RoutingTable initialRoutingTable =
RoutingTable.builder().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
ClusterState.builder(
org.elasticsearch.cluster.ClusterName.CLUSTER_NAME_SETTING.getDefault(
Settings.EMPTY))
.metaData(metaData)
.routingTable(initialRoutingTable)
.build();
assertThat(clusterState.routingTable().index("test").shards().size(), equalTo(3));
for (int i = 0; i < clusterState.routingTable().index("test").shards().size(); i++) {
assertThat(clusterState.routingTable().index("test").shard(i).size(), equalTo(2));
assertThat(clusterState.routingTable().index("test").shard(i).shards().size(), equalTo(2));
assertThat(
clusterState.routingTable().index("test").shard(i).shards().get(0).state(),
equalTo(UNASSIGNED));
assertThat(
clusterState.routingTable().index("test").shard(i).shards().get(1).state(),
equalTo(UNASSIGNED));
assertThat(
clusterState.routingTable().index("test").shard(i).shards().get(0).currentNodeId(),
nullValue());
assertThat(
clusterState.routingTable().index("test").shard(i).shards().get(1).currentNodeId(),
nullValue());
}
logger.info("Adding three node and performing rerouting");
clusterState =
ClusterState.builder(clusterState)
.nodes(
DiscoveryNodes.builder()
.add(newNode("node1"))
.add(newNode("node2"))
.add(newNode("node3")))
.build();
ClusterState newState = strategy.reroute(clusterState, "reroute");
assertThat(newState, not(equalTo(clusterState)));
clusterState = newState;
assertThat(clusterState.routingTable().index("test").shards().size(), equalTo(3));
for (int i = 0; i < clusterState.routingTable().index("test").shards().size(); i++) {
assertThat(clusterState.routingTable().index("test").shard(i).size(), equalTo(2));
assertThat(clusterState.routingTable().index("test").shard(i).shards().size(), equalTo(2));
assertThat(
clusterState.routingTable().index("test").shard(i).primaryShard().state(),
equalTo(INITIALIZING));
assertThat(
clusterState.routingTable().index("test").shard(i).replicaShards().size(), equalTo(1));
assertThat(
clusterState.routingTable().index("test").shard(i).replicaShards().get(0).state(),
equalTo(UNASSIGNED));
assertThat(
clusterState.routingTable().index("test").shard(i).replicaShards().get(0).currentNodeId(),
nullValue());
}
logger.info("Another round of rebalancing");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()))
.build();
newState = strategy.reroute(clusterState, "reroute");
assertThat(newState, equalTo(clusterState));
RoutingNodes routingNodes = clusterState.getRoutingNodes();
newState =
strategy.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING));
assertThat(newState, not(equalTo(clusterState)));
clusterState = newState;
assertThat(clusterState.routingTable().index("test").shards().size(), equalTo(3));
for (int i = 0; i < clusterState.routingTable().index("test").shards().size(); i++) {
assertThat(clusterState.routingTable().index("test").shard(i).size(), equalTo(2));
assertThat(clusterState.routingTable().index("test").shard(i).shards().size(), equalTo(2));
assertThat(
clusterState.routingTable().index("test").shard(i).primaryShard().state(),
equalTo(STARTED));
assertThat(
clusterState.routingTable().index("test").shard(i).replicaShards().size(), equalTo(1));
// backup shards are initializing as well, we make sure that they
// recover from primary *started* shards in the
// IndicesClusterStateService
assertThat(
clusterState.routingTable().index("test").shard(i).replicaShards().get(0).state(),
equalTo(INITIALIZING));
}
logger.info("Reroute, nothing should change");
newState = strategy.reroute(clusterState, "reroute");
assertThat(newState, equalTo(clusterState));
logger.info("Start the more shards");
routingNodes = clusterState.getRoutingNodes();
newState =
strategy.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING));
assertThat(newState, not(equalTo(clusterState)));
clusterState = newState;
routingNodes = clusterState.getRoutingNodes();
assertThat(clusterState.routingTable().index("test").shards().size(), equalTo(3));
for (int i = 0; i < clusterState.routingTable().index("test").shards().size(); i++) {
assertThat(clusterState.routingTable().index("test").shard(i).size(), equalTo(2));
assertThat(clusterState.routingTable().index("test").shard(i).shards().size(), equalTo(2));
assertThat(
clusterState.routingTable().index("test").shard(i).primaryShard().state(),
equalTo(STARTED));
assertThat(
clusterState.routingTable().index("test").shard(i).replicaShards().size(), equalTo(1));
}
assertThat(routingNodes.node("node1").numberOfShardsWithState(STARTED), equalTo(2));
assertThat(routingNodes.node("node2").numberOfShardsWithState(STARTED), equalTo(2));
assertThat(routingNodes.node("node3").numberOfShardsWithState(STARTED), equalTo(2));
assertThat(routingNodes.node("node1").shardsWithState("test", STARTED).size(), equalTo(2));
assertThat(routingNodes.node("node2").shardsWithState("test", STARTED).size(), equalTo(2));
assertThat(routingNodes.node("node3").shardsWithState("test", STARTED).size(), equalTo(2));
logger.info("Add new index 3 shards 1 replica");
MetaData updatedMetaData =
MetaData.builder(clusterState.metaData())
.put(
IndexMetaData.builder("test1")
.settings(
settings(Version.CURRENT)
.put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, 3)
.put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 1)))
.build();
RoutingTable updatedRoutingTable =
RoutingTable.builder(clusterState.routingTable())
.addAsNew(updatedMetaData.index("test1"))
.build();
clusterState =
ClusterState.builder(clusterState)
.metaData(updatedMetaData)
.routingTable(updatedRoutingTable)
.build();
assertThat(clusterState.routingTable().index("test1").shards().size(), equalTo(3));
newState = strategy.reroute(clusterState, "reroute");
assertThat(newState, not(equalTo(clusterState)));
clusterState = newState;
assertThat(clusterState.routingTable().index("test1").shards().size(), equalTo(3));
for (int i = 0; i < clusterState.routingTable().index("test1").shards().size(); i++) {
assertThat(clusterState.routingTable().index("test1").shard(i).size(), equalTo(2));
assertThat(clusterState.routingTable().index("test1").shard(i).shards().size(), equalTo(2));
assertThat(
clusterState.routingTable().index("test1").shard(i).primaryShard().state(),
equalTo(INITIALIZING));
assertThat(
clusterState.routingTable().index("test1").shard(i).replicaShards().size(), equalTo(1));
assertThat(
clusterState.routingTable().index("test1").shard(i).replicaShards().get(0).state(),
equalTo(UNASSIGNED));
assertThat(
clusterState
.routingTable()
.index("test1")
.shard(i)
.replicaShards()
.get(0)
.currentNodeId(),
nullValue());
}
logger.info("Another round of rebalancing");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()))
.build();
newState = strategy.reroute(clusterState, "reroute");
assertThat(newState, equalTo(clusterState));
routingNodes = clusterState.getRoutingNodes();
newState =
strategy.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING));
assertThat(newState, not(equalTo(clusterState)));
clusterState = newState;
assertThat(clusterState.routingTable().index("test1").shards().size(), equalTo(3));
for (int i = 0; i < clusterState.routingTable().index("test1").shards().size(); i++) {
assertThat(clusterState.routingTable().index("test1").shard(i).size(), equalTo(2));
assertThat(clusterState.routingTable().index("test1").shard(i).shards().size(), equalTo(2));
assertThat(
clusterState.routingTable().index("test1").shard(i).primaryShard().state(),
equalTo(STARTED));
assertThat(
clusterState.routingTable().index("test1").shard(i).replicaShards().size(), equalTo(1));
// backup shards are initializing as well, we make sure that they
// recover from primary *started* shards in the
// IndicesClusterStateService
assertThat(
clusterState.routingTable().index("test1").shard(i).replicaShards().get(0).state(),
equalTo(INITIALIZING));
}
logger.info("Reroute, nothing should change");
newState = strategy.reroute(clusterState, "reroute");
assertThat(newState, equalTo(clusterState));
logger.info("Start the more shards");
routingNodes = clusterState.getRoutingNodes();
newState =
strategy.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING));
assertThat(newState, not(equalTo(clusterState)));
clusterState = newState;
routingNodes = clusterState.getRoutingNodes();
assertThat(clusterState.routingTable().index("test1").shards().size(), equalTo(3));
for (int i = 0; i < clusterState.routingTable().index("test1").shards().size(); i++) {
assertThat(clusterState.routingTable().index("test1").shard(i).size(), equalTo(2));
assertThat(clusterState.routingTable().index("test1").shard(i).shards().size(), equalTo(2));
assertThat(
clusterState.routingTable().index("test1").shard(i).primaryShard().state(),
equalTo(STARTED));
assertThat(
clusterState.routingTable().index("test1").shard(i).replicaShards().size(), equalTo(1));
}
assertThat(routingNodes.node("node1").numberOfShardsWithState(STARTED), equalTo(4));
assertThat(routingNodes.node("node2").numberOfShardsWithState(STARTED), equalTo(4));
assertThat(routingNodes.node("node3").numberOfShardsWithState(STARTED), equalTo(4));
assertThat(routingNodes.node("node1").shardsWithState("test1", STARTED).size(), equalTo(2));
assertThat(routingNodes.node("node2").shardsWithState("test1", STARTED).size(), equalTo(2));
assertThat(routingNodes.node("node3").shardsWithState("test1", STARTED).size(), equalTo(2));
}
@Test
@TestLogging("cluster.routing.allocation.decider:TRACE")
public void diskThresholdWithAbsoluteSizesTest() {
Settings diskSettings =
settingsBuilder()
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED, true)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK, "30b")
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK, "9b")
.build();
Map<String, DiskUsage> usages = new HashMap<>();
usages.put("node1", new DiskUsage("node1", 100, 10)); // 90% used
usages.put("node2", new DiskUsage("node2", 100, 10)); // 90% used
usages.put("node3", new DiskUsage("node3", 100, 60)); // 40% used
usages.put("node4", new DiskUsage("node4", 100, 80)); // 20% used
usages.put("node5", new DiskUsage("node5", 100, 85)); // 15% used
Map<String, Long> shardSizes = new HashMap<>();
shardSizes.put("[test][0][p]", 10L); // 10 bytes
shardSizes.put("[test][0][r]", 10L);
final ClusterInfo clusterInfo =
new ClusterInfo(ImmutableMap.copyOf(usages), ImmutableMap.copyOf(shardSizes));
AllocationDeciders deciders =
new AllocationDeciders(
ImmutableSettings.EMPTY,
new HashSet<>(
Arrays.asList(
new SameShardAllocationDecider(ImmutableSettings.EMPTY),
new DiskThresholdDecider(diskSettings))));
ClusterInfoService cis =
new ClusterInfoService() {
@Override
public ClusterInfo getClusterInfo() {
logger.info("--> calling fake getClusterInfo");
return clusterInfo;
}
};
AllocationService strategy =
new AllocationService(
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(),
deciders,
new ShardsAllocators(),
cis);
MetaData metaData =
MetaData.builder()
.put(IndexMetaData.builder("test").numberOfShards(1).numberOfReplicas(2))
.build();
RoutingTable routingTable = RoutingTable.builder().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.metaData(metaData)
.routingTable(routingTable)
.build();
logger.info("--> adding node1 and node2 node");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder().put(newNode("node1")).put(newNode("node2")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Primary should initialize, even though both nodes are over the limit initialize
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(1));
String nodeWithPrimary, nodeWithoutPrimary;
if (clusterState.getRoutingNodes().node("node1").size() == 1) {
nodeWithPrimary = "node1";
nodeWithoutPrimary = "node2";
} else {
nodeWithPrimary = "node2";
nodeWithoutPrimary = "node1";
}
logger.info("--> nodeWithPrimary: {}", nodeWithPrimary);
logger.info("--> nodeWithoutPrimary: {}", nodeWithoutPrimary);
// Make node without the primary now habitable to replicas
usages.put(nodeWithoutPrimary, new DiskUsage(nodeWithoutPrimary, 100, 35)); // 65% used
final ClusterInfo clusterInfo2 =
new ClusterInfo(ImmutableMap.copyOf(usages), ImmutableMap.copyOf(shardSizes));
cis =
new ClusterInfoService() {
@Override
public ClusterInfo getClusterInfo() {
logger.info("--> calling fake getClusterInfo");
return clusterInfo2;
}
};
strategy =
new AllocationService(
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(),
deciders,
new ShardsAllocators(),
cis);
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Now the replica should be able to initialize
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(2));
logger.info("--> start the shards (primaries)");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Assert that we're able to start the primary and replica, since they were both initializing
assertThat(
clusterState.routingNodes().shardsWithState(ShardRoutingState.STARTED).size(), equalTo(2));
// Assert that node1 got a single shard (the primary), even though its disk usage is too high
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(1));
// Assert that node2 got a single shard (a replica)
assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(1));
// Assert that one replica is still unassigned
// assertThat(clusterState.routingNodes().shardsWithState(ShardRoutingState.UNASSIGNED).size(),
// equalTo(1));
logger.info("--> adding node3");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()).put(newNode("node3")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Assert that the replica is initialized now that node3 is available with enough space
assertThat(
clusterState.routingNodes().shardsWithState(ShardRoutingState.STARTED).size(), equalTo(2));
assertThat(
clusterState.routingNodes().shardsWithState(ShardRoutingState.INITIALIZING).size(),
equalTo(1));
logger.info("--> start the shards (replicas)");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Assert that all replicas could be started
assertThat(
clusterState.routingNodes().shardsWithState(ShardRoutingState.STARTED).size(), equalTo(3));
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node3").size(), equalTo(1));
logger.info("--> changing decider settings");
// Set the low threshold to 60 instead of 70
// Set the high threshold to 70 instead of 80
// node2 now should not have new shards allocated to it, but shards can remain
diskSettings =
settingsBuilder()
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED, true)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK, "40b")
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK, "30b")
.build();
deciders =
new AllocationDeciders(
ImmutableSettings.EMPTY,
new HashSet<>(
Arrays.asList(
new SameShardAllocationDecider(ImmutableSettings.EMPTY),
new DiskThresholdDecider(diskSettings))));
strategy =
new AllocationService(
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(),
deciders,
new ShardsAllocators(),
cis);
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Shards remain started
assertThat(clusterState.routingNodes().shardsWithState(STARTED).size(), equalTo(3));
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node3").size(), equalTo(1));
logger.info("--> changing settings again");
// Set the low threshold to 50 instead of 60
// Set the high threshold to 60 instead of 70
// node2 now should not have new shards allocated to it, and shards cannot remain
diskSettings =
settingsBuilder()
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED, true)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK, "50b")
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK, "40b")
.build();
deciders =
new AllocationDeciders(
ImmutableSettings.EMPTY,
new HashSet<>(
Arrays.asList(
new SameShardAllocationDecider(ImmutableSettings.EMPTY),
new DiskThresholdDecider(diskSettings))));
strategy =
new AllocationService(
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(),
deciders,
new ShardsAllocators(),
cis);
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Shards remain started
assertThat(clusterState.routingNodes().shardsWithState(STARTED).size(), equalTo(3));
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(1));
// Shard hasn't been moved off of node2 yet because there's nowhere for it to go
assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node3").size(), equalTo(1));
logger.info("--> adding node4");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()).put(newNode("node4")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Shards remain started
assertThat(clusterState.routingNodes().shardsWithState(STARTED).size(), equalTo(2));
// One shard is relocating off of node1
assertThat(clusterState.routingNodes().shardsWithState(RELOCATING).size(), equalTo(1));
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> apply INITIALIZING shards");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// primary shard already has been relocated away
assertThat(clusterState.getRoutingNodes().node(nodeWithPrimary).size(), equalTo(0));
// node with increased space still has its shard
assertThat(clusterState.getRoutingNodes().node(nodeWithoutPrimary).size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node3").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node4").size(), equalTo(1));
logger.info("--> adding node5");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()).put(newNode("node5")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Shards remain started on node3 and node4
assertThat(clusterState.routingNodes().shardsWithState(STARTED).size(), equalTo(2));
// One shard is relocating off of node2 now
assertThat(clusterState.routingNodes().shardsWithState(RELOCATING).size(), equalTo(1));
// Initializing on node5
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> apply INITIALIZING shards");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logger.info("--> final cluster state:");
logShardStates(clusterState);
// Node1 still has no shards because it has no space for them
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(0));
// Node5 is available now, so the shard is moved off of node2
assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(0));
assertThat(clusterState.getRoutingNodes().node("node3").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node4").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node5").size(), equalTo(1));
}
@Test
public void tesStartedShardsMatching() {
AllocationService allocation = createAllocationService();
logger.info("--> building initial cluster state");
final IndexMetaData indexMetaData =
IndexMetaData.builder("test").numberOfShards(3).numberOfReplicas(0).build();
ClusterState.Builder stateBuilder =
ClusterState.builder(ClusterName.DEFAULT)
.nodes(DiscoveryNodes.builder().put(newNode("node1")).put(newNode("node2")))
.metaData(MetaData.builder().put(indexMetaData, false));
final ImmutableShardRouting initShard =
new ImmutableShardRouting(
"test", 0, "node1", randomBoolean(), ShardRoutingState.INITIALIZING, 1);
final ImmutableShardRouting startedShard =
new ImmutableShardRouting(
"test", 1, "node2", randomBoolean(), ShardRoutingState.STARTED, 1);
final ImmutableShardRouting relocatingShard =
new ImmutableShardRouting(
"test", 2, "node1", "node2", randomBoolean(), ShardRoutingState.RELOCATING, 1);
stateBuilder.routingTable(
RoutingTable.builder()
.add(
IndexRoutingTable.builder("test")
.addIndexShard(
new IndexShardRoutingTable.Builder(initShard.shardId(), true)
.addShard(initShard)
.build())
.addIndexShard(
new IndexShardRoutingTable.Builder(startedShard.shardId(), true)
.addShard(startedShard)
.build())
.addIndexShard(
new IndexShardRoutingTable.Builder(relocatingShard.shardId(), true)
.addShard(relocatingShard)
.build())));
ClusterState state = stateBuilder.build();
logger.info("--> test starting of shard");
RoutingAllocation.Result result =
allocation.applyStartedShards(
state,
Arrays.asList(
new ImmutableShardRouting(
initShard.index(),
initShard.id(),
initShard.currentNodeId(),
initShard.relocatingNodeId(),
initShard.primary(),
ShardRoutingState.INITIALIZING,
randomInt())),
false);
assertTrue(
"failed to start "
+ initShard
+ "\ncurrent routing table:"
+ result.routingTable().prettyPrint(),
result.changed());
assertTrue(
initShard + "isn't started \ncurrent routing table:" + result.routingTable().prettyPrint(),
result
.routingTable()
.index("test")
.shard(initShard.id())
.countWithState(ShardRoutingState.STARTED)
== 1);
logger.info("--> testing shard variants that shouldn't match the started shard");
result =
allocation.applyStartedShards(
state,
Arrays.asList(
new ImmutableShardRouting(
initShard.index(),
initShard.id(),
initShard.currentNodeId(),
initShard.relocatingNodeId(),
!initShard.primary(),
ShardRoutingState.INITIALIZING,
1)),
false);
assertFalse(
"wrong primary flag shouldn't start shard "
+ initShard
+ "\ncurrent routing table:"
+ result.routingTable().prettyPrint(),
result.changed());
result =
allocation.applyStartedShards(
state,
Arrays.asList(
new ImmutableShardRouting(
initShard.index(),
initShard.id(),
"some_node",
initShard.currentNodeId(),
initShard.primary(),
ShardRoutingState.INITIALIZING,
1)),
false);
assertFalse(
"relocating shard from node shouldn't start shard "
+ initShard
+ "\ncurrent routing table:"
+ result.routingTable().prettyPrint(),
result.changed());
result =
allocation.applyStartedShards(
state,
Arrays.asList(
new ImmutableShardRouting(
initShard.index(),
initShard.id(),
initShard.currentNodeId(),
"some_node",
initShard.primary(),
ShardRoutingState.INITIALIZING,
1)),
false);
assertFalse(
"relocating shard to node shouldn't start shard "
+ initShard
+ "\ncurrent routing table:"
+ result.routingTable().prettyPrint(),
result.changed());
logger.info("--> testing double starting");
result =
allocation.applyStartedShards(
state,
Arrays.asList(
new ImmutableShardRouting(
startedShard.index(),
startedShard.id(),
startedShard.currentNodeId(),
startedShard.relocatingNodeId(),
startedShard.primary(),
ShardRoutingState.INITIALIZING,
1)),
false);
assertFalse(
"duplicate starting of the same shard should be ignored \ncurrent routing table:"
+ result.routingTable().prettyPrint(),
result.changed());
logger.info("--> testing starting of relocating shards");
result =
allocation.applyStartedShards(
state,
Arrays.asList(
new ImmutableShardRouting(
relocatingShard.index(),
relocatingShard.id(),
relocatingShard.relocatingNodeId(),
relocatingShard.currentNodeId(),
relocatingShard.primary(),
ShardRoutingState.INITIALIZING,
randomInt())),
false);
assertTrue(
"failed to start "
+ relocatingShard
+ "\ncurrent routing table:"
+ result.routingTable().prettyPrint(),
result.changed());
ShardRouting shardRouting =
result.routingTable().index("test").shard(relocatingShard.id()).getShards().get(0);
assertThat(shardRouting.state(), equalTo(ShardRoutingState.STARTED));
assertThat(shardRouting.currentNodeId(), equalTo("node2"));
assertThat(shardRouting.relocatingNodeId(), nullValue());
logger.info("--> testing shard variants that shouldn't match the relocating shard");
result =
allocation.applyStartedShards(
state,
Arrays.asList(
new ImmutableShardRouting(
relocatingShard.index(),
relocatingShard.id(),
relocatingShard.relocatingNodeId(),
relocatingShard.currentNodeId(),
!relocatingShard.primary(),
ShardRoutingState.INITIALIZING,
1)),
false);
assertFalse(
"wrong primary flag shouldn't start shard "
+ relocatingShard
+ "\ncurrent routing table:"
+ result.routingTable().prettyPrint(),
result.changed());
result =
allocation.applyStartedShards(
state,
Arrays.asList(
new ImmutableShardRouting(
relocatingShard.index(),
relocatingShard.id(),
"some_node",
relocatingShard.currentNodeId(),
relocatingShard.primary(),
ShardRoutingState.INITIALIZING,
1)),
false);
assertFalse(
"relocating shard to a different node shouldn't start shard "
+ relocatingShard
+ "\ncurrent routing table:"
+ result.routingTable().prettyPrint(),
result.changed());
result =
allocation.applyStartedShards(
state,
Arrays.asList(
new ImmutableShardRouting(
relocatingShard.index(),
relocatingShard.id(),
relocatingShard.relocatingNodeId(),
"some_node",
relocatingShard.primary(),
ShardRoutingState.INITIALIZING,
1)),
false);
assertFalse(
"relocating shard from a different node shouldn't start shard "
+ relocatingShard
+ "\ncurrent routing table:"
+ result.routingTable().prettyPrint(),
result.changed());
result =
allocation.applyStartedShards(
state,
Arrays.asList(
new ImmutableShardRouting(
relocatingShard.index(),
relocatingShard.id(),
relocatingShard.relocatingNodeId(),
relocatingShard.primary(),
ShardRoutingState.INITIALIZING,
1)),
false);
assertFalse(
"non-relocating shard shouldn't start shard"
+ relocatingShard
+ "\ncurrent routing table:"
+ result.routingTable().prettyPrint(),
result.changed());
}
@Test
public void moveShardCommand() {
AllocationService allocation =
new AllocationService(
settingsBuilder().put("cluster.routing.allocation.concurrent_recoveries", 10).build());
logger.info("creating an index with 1 shard, no replica");
MetaData metaData =
newMetaDataBuilder()
.put(newIndexMetaDataBuilder("test").numberOfShards(1).numberOfReplicas(0))
.build();
RoutingTable routingTable = routingTable().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
newClusterStateBuilder().metaData(metaData).routingTable(routingTable).build();
logger.info("adding two nodes and performing rerouting");
clusterState =
newClusterStateBuilder()
.state(clusterState)
.nodes(newNodesBuilder().put(newNode("node1")).put(newNode("node2")))
.build();
RoutingAllocationResult rerouteResult = allocation.reroute(clusterState);
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
logger.info("start primary shard");
rerouteResult =
allocation.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
logger.info("move the shard");
String existingNodeId =
clusterState.routingTable().index("test").shard(0).primaryShard().currentNodeId();
String toNodeId;
if ("node1".equals(existingNodeId)) {
toNodeId = "node2";
} else {
toNodeId = "node1";
}
rerouteResult =
allocation.reroute(
clusterState,
new AllocationCommands(
new MoveAllocationCommand(new ShardId("test", 0), existingNodeId, toNodeId)));
assertThat(rerouteResult.changed(), equalTo(true));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(
clusterState.routingNodes().node(existingNodeId).shards().get(0).state(),
equalTo(ShardRoutingState.RELOCATING));
assertThat(
clusterState.routingNodes().node(toNodeId).shards().get(0).state(),
equalTo(ShardRoutingState.INITIALIZING));
logger.info("finish moving the shard");
rerouteResult =
allocation.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node(existingNodeId).shards().isEmpty(), equalTo(true));
assertThat(
clusterState.routingNodes().node(toNodeId).shards().get(0).state(),
equalTo(ShardRoutingState.STARTED));
}
@Test
public void cancelCommand() {
AllocationService allocation =
new AllocationService(
settingsBuilder()
.put("cluster.routing.allocation.disable_new_allocation", true)
.put("cluster.routing.allocation.disable_allocation", true)
.build());
logger.info("--> building initial routing table");
MetaData metaData =
newMetaDataBuilder()
.put(newIndexMetaDataBuilder("test").numberOfShards(1).numberOfReplicas(1))
.build();
RoutingTable routingTable = routingTable().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
newClusterStateBuilder().metaData(metaData).routingTable(routingTable).build();
logger.info("--> adding 3 nodes");
clusterState =
newClusterStateBuilder()
.state(clusterState)
.nodes(
newNodesBuilder().put(newNode("node1")).put(newNode("node2")).put(newNode("node3")))
.build();
RoutingAllocationResult rerouteResult = allocation.reroute(clusterState);
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(0));
logger.info("--> allocating with primary flag set to true");
rerouteResult =
allocation.reroute(
clusterState,
new AllocationCommands(
new AllocateAllocationCommand(new ShardId("test", 0), "node1", true)));
assertThat(rerouteResult.changed(), equalTo(true));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(INITIALIZING).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(0));
logger.info("--> cancel primary allocation, make sure it fails...");
try {
allocation.reroute(
clusterState,
new AllocationCommands(
new CancelAllocationCommand(new ShardId("test", 0), "node1", false)));
assert false;
} catch (ElasticSearchIllegalArgumentException e) {
}
logger.info("--> start the primary shard");
rerouteResult =
allocation.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(0));
logger.info("--> cancel primary allocation, make sure it fails...");
try {
allocation.reroute(
clusterState,
new AllocationCommands(
new CancelAllocationCommand(new ShardId("test", 0), "node1", false)));
assert false;
} catch (ElasticSearchIllegalArgumentException e) {
}
logger.info("--> allocate the replica shard on on the second node");
rerouteResult =
allocation.reroute(
clusterState,
new AllocationCommands(
new AllocateAllocationCommand(new ShardId("test", 0), "node2", false)));
assertThat(rerouteResult.changed(), equalTo(true));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node2").shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> cancel the relocation allocation");
rerouteResult =
allocation.reroute(
clusterState,
new AllocationCommands(
new CancelAllocationCommand(new ShardId("test", 0), "node2", false)));
assertThat(rerouteResult.changed(), equalTo(true));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(0));
assertThat(clusterState.routingNodes().node("node3").shards().size(), equalTo(0));
logger.info("--> allocate the replica shard on on the second node");
rerouteResult =
allocation.reroute(
clusterState,
new AllocationCommands(
new AllocateAllocationCommand(new ShardId("test", 0), "node2", false)));
assertThat(rerouteResult.changed(), equalTo(true));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node2").shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> cancel the primary being replicated, make sure it fails");
try {
allocation.reroute(
clusterState,
new AllocationCommands(
new CancelAllocationCommand(new ShardId("test", 0), "node1", false)));
assert false;
} catch (ElasticSearchIllegalArgumentException e) {
}
logger.info("--> start the replica shard");
rerouteResult =
allocation.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node2").shardsWithState(STARTED).size(), equalTo(1));
logger.info("--> cancel allocation of the replica shard");
rerouteResult =
allocation.reroute(
clusterState,
new AllocationCommands(
new CancelAllocationCommand(new ShardId("test", 0), "node2", false)));
assertThat(rerouteResult.changed(), equalTo(true));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(0));
assertThat(clusterState.routingNodes().node("node3").shards().size(), equalTo(0));
logger.info("--> allocate the replica shard on on the second node");
rerouteResult =
allocation.reroute(
clusterState,
new AllocationCommands(
new AllocateAllocationCommand(new ShardId("test", 0), "node2", false)));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(rerouteResult.changed(), equalTo(true));
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node2").shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> start the replica shard");
rerouteResult =
allocation.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node2").shardsWithState(STARTED).size(), equalTo(1));
logger.info("--> cancel the primary allocation (with allow_primary set to true)");
rerouteResult =
allocation.reroute(
clusterState,
new AllocationCommands(
new CancelAllocationCommand(new ShardId("test", 0), "node1", true)));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(rerouteResult.changed(), equalTo(true));
assertThat(
clusterState.routingNodes().node("node2").shardsWithState(STARTED).get(0).primary(),
equalTo(true));
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(0));
assertThat(clusterState.routingNodes().node("node3").shards().size(), equalTo(0));
}
@Test
public void allocateCommand() {
AllocationService allocation =
new AllocationService(
settingsBuilder()
.put("cluster.routing.allocation.disable_new_allocation", true)
.put("cluster.routing.allocation.disable_allocation", true)
.build());
logger.info("--> building initial routing table");
MetaData metaData =
newMetaDataBuilder()
.put(newIndexMetaDataBuilder("test").numberOfShards(1).numberOfReplicas(1))
.build();
RoutingTable routingTable = routingTable().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
newClusterStateBuilder().metaData(metaData).routingTable(routingTable).build();
logger.info("--> adding 3 nodes on same rack and do rerouting");
clusterState =
newClusterStateBuilder()
.state(clusterState)
.nodes(
newNodesBuilder().put(newNode("node1")).put(newNode("node2")).put(newNode("node3")))
.build();
RoutingAllocationResult rerouteResult = allocation.reroute(clusterState);
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(0));
logger.info("--> allocating with primary flag set to false, should fail");
try {
allocation.reroute(
clusterState,
new AllocationCommands(
new AllocateAllocationCommand(new ShardId("test", 0), "node1", false)));
assert false;
} catch (ElasticSearchIllegalArgumentException e) {
}
logger.info("--> allocating with primary flag set to true");
rerouteResult =
allocation.reroute(
clusterState,
new AllocationCommands(
new AllocateAllocationCommand(new ShardId("test", 0), "node1", true)));
assertThat(rerouteResult.changed(), equalTo(true));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(INITIALIZING).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(0));
logger.info("--> start the primary shard");
rerouteResult =
allocation.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(0));
logger.info("--> allocate the replica shard on the primary shard node, should fail");
try {
allocation.reroute(
clusterState,
new AllocationCommands(
new AllocateAllocationCommand(new ShardId("test", 0), "node1", false)));
assert false;
} catch (ElasticSearchIllegalArgumentException e) {
}
logger.info("--> allocate the replica shard on on the second node");
rerouteResult =
allocation.reroute(
clusterState,
new AllocationCommands(
new AllocateAllocationCommand(new ShardId("test", 0), "node2", false)));
assertThat(rerouteResult.changed(), equalTo(true));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node2").shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> start the replica shard");
rerouteResult =
allocation.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING));
clusterState =
newClusterStateBuilder()
.state(clusterState)
.routingTable(rerouteResult.routingTable())
.build();
assertThat(clusterState.routingNodes().node("node1").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().node("node1").shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().node("node2").shards().size(), equalTo(1));
assertThat(
clusterState.routingNodes().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(
new AllocateAllocationCommand(new ShardId("test", 0), "node3", false)));
assert false;
} catch (ElasticSearchIllegalArgumentException e) {
}
}
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<FailedRerouteAllocation.FailedShard> failedShards =
Collections.singletonList(
new FailedRerouteAllocation.FailedShard(
routingTable.index("idx").shard(0).shards().get(0),
"boom" + i,
new UnsupportedOperationException()));
RoutingAllocation.Result result = strategy.applyFailedShards(clusterState, failedShards);
assertTrue(result.changed());
routingTable = result.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
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<FailedRerouteAllocation.FailedShard> failedShards =
Collections.singletonList(
new FailedRerouteAllocation.FailedShard(
routingTable.index("idx").shard(0).shards().get(0),
"boom",
new UnsupportedOperationException()));
RoutingAllocation.Result result = strategy.applyFailedShards(clusterState, failedShards);
assertTrue(result.changed());
routingTable = result.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(), UNASSIGNED);
assertEquals(
routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getMessage(), "boom");
result =
strategy.reroute(
clusterState,
new AllocationCommands(),
false,
true); // manual reroute should retry once
assertTrue(result.changed());
routingTable = result.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 FailedRerouteAllocation.FailedShard(
routingTable.index("idx").shard(0).shards().get(0),
"boom",
new UnsupportedOperationException()));
result = strategy.applyFailedShards(clusterState, failedShards);
assertTrue(result.changed());
routingTable = result.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 + 1);
assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), UNASSIGNED);
assertEquals(
routingTable.index("idx").shard(0).shards().get(0).unassignedInfo().getMessage(), "boom");
}
public void testClusterStateUpdateTask() {
AllocationService allocationService =
new AllocationService(
Settings.builder().build(),
new AllocationDeciders(
Settings.EMPTY,
Collections.singleton(new MaxRetryAllocationDecider(Settings.EMPTY))),
NoopGatewayAllocator.INSTANCE,
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);
assertSame(execute, clusterState); // dry-run
task.onAllNodesAcked(null);
assertNotSame(responseRef.get().getState(), execute);
req.dryRun(false); // now we allocate
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);
assertNotSame(newState, clusterState); // dry-run=false
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<FailedRerouteAllocation.FailedShard> failedShards =
Collections.singletonList(
new FailedRerouteAllocation.FailedShard(
routingTable.index("idx").shard(0).shards().get(0),
"boom" + i,
new UnsupportedOperationException()));
RoutingAllocation.Result result =
allocationService.applyFailedShards(clusterState, failedShards);
assertTrue(result.changed());
clusterState = ClusterState.builder(clusterState).routingTable(result.routingTable()).build();
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);
assertNotSame(newState, clusterState); // dry-run=false
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);
req.setRetryFailed(true); // now we manually retry and get the shard back into initializing
newState = task.execute(clusterState);
assertNotSame(newState, clusterState); // dry-run=false
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);
}
@Test
public void testUpdateNumberOfReplicas() {
AllocationService strategy =
createAllocationService(
settingsBuilder().put("cluster.routing.allocation.concurrent_recoveries", 10).build());
logger.info("Building initial routing table");
MetaData metaData =
MetaData.builder()
.put(IndexMetaData.builder("test").numberOfShards(1).numberOfReplicas(1))
.build();
RoutingTable routingTable = RoutingTable.builder().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
ClusterState.builder().metaData(metaData).routingTable(routingTable).build();
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(2));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(2));
assertThat(routingTable.index("test").shard(0).shards().get(0).state(), equalTo(UNASSIGNED));
assertThat(routingTable.index("test").shard(0).shards().get(1).state(), equalTo(UNASSIGNED));
assertThat(routingTable.index("test").shard(0).shards().get(0).currentNodeId(), nullValue());
assertThat(routingTable.index("test").shard(0).shards().get(1).currentNodeId(), nullValue());
logger.info("Adding two nodes and performing rerouting");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder().put(newNode("node1")).put(newNode("node2")))
.build();
RoutingTable prevRoutingTable = routingTable;
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logger.info("Start all the primary shards");
RoutingNodes routingNodes = clusterState.routingNodes();
prevRoutingTable = routingTable;
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logger.info("Start all the replica shards");
routingNodes = clusterState.routingNodes();
prevRoutingTable = routingTable;
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
final String nodeHoldingPrimary =
routingTable.index("test").shard(0).primaryShard().currentNodeId();
final String nodeHoldingReplica =
routingTable.index("test").shard(0).replicaShards().get(0).currentNodeId();
assertThat(nodeHoldingPrimary, not(equalTo(nodeHoldingReplica)));
assertThat(prevRoutingTable != routingTable, equalTo(true));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(2));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(2));
assertThat(routingTable.index("test").shard(0).primaryShard().state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).primaryShard().currentNodeId(),
equalTo(nodeHoldingPrimary));
assertThat(routingTable.index("test").shard(0).replicaShards().size(), equalTo(1));
assertThat(
routingTable.index("test").shard(0).replicaShards().get(0).state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).replicaShards().get(0).currentNodeId(),
equalTo(nodeHoldingReplica));
logger.info("add another replica");
routingNodes = clusterState.routingNodes();
prevRoutingTable = routingTable;
routingTable = RoutingTable.builder(routingTable).updateNumberOfReplicas(2).build();
metaData = MetaData.builder(clusterState.metaData()).updateNumberOfReplicas(2).build();
clusterState =
ClusterState.builder(clusterState).routingTable(routingTable).metaData(metaData).build();
assertThat(clusterState.metaData().index("test").numberOfReplicas(), equalTo(2));
assertThat(prevRoutingTable != routingTable, equalTo(true));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(3));
assertThat(routingTable.index("test").shard(0).primaryShard().state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).primaryShard().currentNodeId(),
equalTo(nodeHoldingPrimary));
assertThat(routingTable.index("test").shard(0).replicaShards().size(), equalTo(2));
assertThat(
routingTable.index("test").shard(0).replicaShards().get(0).state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).replicaShards().get(0).currentNodeId(),
equalTo(nodeHoldingReplica));
assertThat(
routingTable.index("test").shard(0).replicaShards().get(1).state(), equalTo(UNASSIGNED));
logger.info("Add another node and start the added replica");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()).put(newNode("node3")))
.build();
prevRoutingTable = routingTable;
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(prevRoutingTable != routingTable, equalTo(true));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(3));
assertThat(routingTable.index("test").shard(0).primaryShard().state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).primaryShard().currentNodeId(),
equalTo(nodeHoldingPrimary));
assertThat(routingTable.index("test").shard(0).replicaShards().size(), equalTo(2));
assertThat(
routingTable.index("test").shard(0).replicaShardsWithState(STARTED).size(), equalTo(1));
assertThat(
routingTable.index("test").shard(0).replicaShardsWithState(STARTED).get(0).currentNodeId(),
equalTo(nodeHoldingReplica));
assertThat(
routingTable.index("test").shard(0).replicaShardsWithState(INITIALIZING).size(),
equalTo(1));
assertThat(
routingTable
.index("test")
.shard(0)
.replicaShardsWithState(INITIALIZING)
.get(0)
.currentNodeId(),
equalTo("node3"));
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));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(3));
assertThat(routingTable.index("test").shard(0).primaryShard().state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).primaryShard().currentNodeId(),
equalTo(nodeHoldingPrimary));
assertThat(routingTable.index("test").shard(0).replicaShards().size(), equalTo(2));
assertThat(
routingTable.index("test").shard(0).replicaShardsWithState(STARTED).size(), equalTo(2));
assertThat(
routingTable.index("test").shard(0).replicaShardsWithState(STARTED).get(0).currentNodeId(),
anyOf(equalTo(nodeHoldingReplica), equalTo("node3")));
assertThat(
routingTable.index("test").shard(0).replicaShardsWithState(STARTED).get(1).currentNodeId(),
anyOf(equalTo(nodeHoldingReplica), equalTo("node3")));
logger.info("now remove a replica");
routingNodes = clusterState.routingNodes();
prevRoutingTable = routingTable;
routingTable = RoutingTable.builder(routingTable).updateNumberOfReplicas(1).build();
metaData = MetaData.builder(clusterState.metaData()).updateNumberOfReplicas(1).build();
clusterState =
ClusterState.builder(clusterState).routingTable(routingTable).metaData(metaData).build();
assertThat(clusterState.metaData().index("test").numberOfReplicas(), equalTo(1));
assertThat(prevRoutingTable != routingTable, equalTo(true));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(2));
assertThat(routingTable.index("test").shard(0).primaryShard().state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).primaryShard().currentNodeId(),
equalTo(nodeHoldingPrimary));
assertThat(routingTable.index("test").shard(0).replicaShards().size(), equalTo(1));
assertThat(
routingTable.index("test").shard(0).replicaShards().get(0).state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).replicaShards().get(0).currentNodeId(),
anyOf(equalTo(nodeHoldingReplica), equalTo("node3")));
logger.info("do a reroute, should remain the same");
prevRoutingTable = routingTable;
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(prevRoutingTable != routingTable, equalTo(false));
}
@Test
public void diskThresholdTest() {
Settings diskSettings =
settingsBuilder()
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED, true)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK, 0.7)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK, 0.8)
.build();
Map<String, DiskUsage> usages = new HashMap<>();
usages.put("node1", new DiskUsage("node1", 100, 10)); // 90% used
usages.put("node2", new DiskUsage("node2", 100, 35)); // 65% used
usages.put("node3", new DiskUsage("node3", 100, 60)); // 40% used
usages.put("node4", new DiskUsage("node4", 100, 80)); // 20% used
Map<String, Long> shardSizes = new HashMap<>();
shardSizes.put("[test][0][p]", 10L); // 10 bytes
shardSizes.put("[test][0][r]", 10L);
final ClusterInfo clusterInfo =
new ClusterInfo(ImmutableMap.copyOf(usages), ImmutableMap.copyOf(shardSizes));
AllocationDeciders deciders =
new AllocationDeciders(
ImmutableSettings.EMPTY,
new HashSet<>(
Arrays.asList(
new SameShardAllocationDecider(ImmutableSettings.EMPTY),
new DiskThresholdDecider(diskSettings))));
ClusterInfoService cis =
new ClusterInfoService() {
@Override
public ClusterInfo getClusterInfo() {
logger.info("--> calling fake getClusterInfo");
return clusterInfo;
}
};
AllocationService strategy =
new AllocationService(
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(),
deciders,
new ShardsAllocators(),
cis);
MetaData metaData =
MetaData.builder()
.put(IndexMetaData.builder("test").numberOfShards(1).numberOfReplicas(1))
.build();
RoutingTable routingTable = RoutingTable.builder().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.metaData(metaData)
.routingTable(routingTable)
.build();
logger.info("--> adding two nodes");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder().put(newNode("node1")).put(newNode("node2")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Primary shard should be initializing, replica should not
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> start the shards (primaries)");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Assert that we're able to start the primary
assertThat(
clusterState.routingNodes().shardsWithState(ShardRoutingState.STARTED).size(), equalTo(1));
// Assert that node1 didn't get any shards because its disk usage is too high
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(0));
logger.info("--> start the shards (replicas)");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Assert that the replica couldn't be started since node1 doesn't have enough space
assertThat(
clusterState.routingNodes().shardsWithState(ShardRoutingState.STARTED).size(), equalTo(1));
logger.info("--> adding node3");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()).put(newNode("node3")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Assert that the replica is initialized now that node3 is available with enough space
assertThat(
clusterState.routingNodes().shardsWithState(ShardRoutingState.STARTED).size(), equalTo(1));
assertThat(
clusterState.routingNodes().shardsWithState(ShardRoutingState.INITIALIZING).size(),
equalTo(1));
logger.info("--> start the shards (replicas)");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Assert that the replica couldn't be started since node1 doesn't have enough space
assertThat(
clusterState.routingNodes().shardsWithState(ShardRoutingState.STARTED).size(), equalTo(2));
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(0));
assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node3").size(), equalTo(1));
logger.info("--> changing decider settings");
// Set the low threshold to 60 instead of 70
// Set the high threshold to 70 instead of 80
// node2 now should not have new shards allocated to it, but shards can remain
diskSettings =
settingsBuilder()
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED, true)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK, "60%")
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK, 0.7)
.build();
deciders =
new AllocationDeciders(
ImmutableSettings.EMPTY,
new HashSet<>(
Arrays.asList(
new SameShardAllocationDecider(ImmutableSettings.EMPTY),
new DiskThresholdDecider(diskSettings))));
strategy =
new AllocationService(
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(),
deciders,
new ShardsAllocators(),
cis);
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Shards remain started
assertThat(clusterState.routingNodes().shardsWithState(STARTED).size(), equalTo(2));
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(0));
assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node3").size(), equalTo(1));
logger.info("--> changing settings again");
// Set the low threshold to 50 instead of 60
// Set the high threshold to 60 instead of 70
// node2 now should not have new shards allocated to it, and shards cannot remain
diskSettings =
settingsBuilder()
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED, true)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK, 0.5)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK, 0.6)
.build();
deciders =
new AllocationDeciders(
ImmutableSettings.EMPTY,
new HashSet<>(
Arrays.asList(
new SameShardAllocationDecider(ImmutableSettings.EMPTY),
new DiskThresholdDecider(diskSettings))));
strategy =
new AllocationService(
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(),
deciders,
new ShardsAllocators(),
cis);
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Shards remain started
assertThat(clusterState.routingNodes().shardsWithState(STARTED).size(), equalTo(2));
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(0));
// Shard hasn't been moved off of node2 yet because there's nowhere for it to go
assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node3").size(), equalTo(1));
logger.info("--> adding node4");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()).put(newNode("node4")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Shards remain started
assertThat(clusterState.routingNodes().shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> apply INITIALIZING shards");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(0));
// Node4 is available now, so the shard is moved off of node2
assertThat(clusterState.getRoutingNodes().node("node2").size(), equalTo(0));
assertThat(clusterState.getRoutingNodes().node("node3").size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node4").size(), equalTo(1));
}
@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));
}
@Test
public void diskThresholdWithShardSizes() {
Settings diskSettings =
settingsBuilder()
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED, true)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK, 0.7)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK, "71%")
.build();
Map<String, DiskUsage> usages = new HashMap<>();
usages.put("node1", new DiskUsage("node1", 100, 31)); // 69% used
usages.put("node2", new DiskUsage("node2", 100, 1)); // 99% used
Map<String, Long> shardSizes = new HashMap<>();
shardSizes.put("[test][0][p]", 10L); // 10 bytes
final ClusterInfo clusterInfo =
new ClusterInfo(ImmutableMap.copyOf(usages), ImmutableMap.copyOf(shardSizes));
AllocationDeciders deciders =
new AllocationDeciders(
ImmutableSettings.EMPTY,
new HashSet<>(
Arrays.asList(
new SameShardAllocationDecider(ImmutableSettings.EMPTY),
new DiskThresholdDecider(diskSettings))));
ClusterInfoService cis =
new ClusterInfoService() {
@Override
public ClusterInfo getClusterInfo() {
logger.info("--> calling fake getClusterInfo");
return clusterInfo;
}
};
AllocationService strategy =
new AllocationService(
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(),
deciders,
new ShardsAllocators(),
cis);
MetaData metaData =
MetaData.builder()
.put(IndexMetaData.builder("test").numberOfShards(1).numberOfReplicas(0))
.build();
RoutingTable routingTable = RoutingTable.builder().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.metaData(metaData)
.routingTable(routingTable)
.build();
logger.info("--> adding node1");
clusterState =
ClusterState.builder(clusterState)
.nodes(
DiscoveryNodes.builder()
.put(newNode("node1"))
.put(
newNode(
"node2")) // node2 is added because DiskThresholdDecider automatically
// ignore single-node clusters
)
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logger.info("--> start the shards (primaries)");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Shard can't be allocated to node1 (or node2) because it would cause too much usage
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(0));
// No shards are started, no nodes have enough disk for allocation
assertThat(clusterState.routingNodes().shardsWithState(STARTED).size(), equalTo(0));
}
@Test
public void testSingleIndexStartedShard() {
AllocationService strategy =
createAllocationService(
settingsBuilder().put("cluster.routing.allocation.concurrent_recoveries", 10).build());
logger.info("Building initial routing table");
MetaData metaData =
MetaData.builder()
.put(
IndexMetaData.builder("test")
.settings(settings(Version.CURRENT))
.numberOfShards(1)
.numberOfReplicas(0))
.build();
RoutingTable routingTable = RoutingTable.builder().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.metaData(metaData)
.routingTable(routingTable)
.build();
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().get(0).state(), equalTo(UNASSIGNED));
assertThat(routingTable.index("test").shard(0).shards().get(0).currentNodeId(), nullValue());
logger.info("Adding one node and performing rerouting");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder().put(newNode("node1")))
.build();
RoutingTable prevRoutingTable = routingTable;
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().get(0).state(), equalTo(INITIALIZING));
assertThat(
routingTable.index("test").shard(0).shards().get(0).currentNodeId(), equalTo("node1"));
logger.info("Rerouting again, nothing should change");
prevRoutingTable = routingTable;
clusterState = ClusterState.builder(clusterState).build();
routingTable = strategy.reroute(clusterState).routingTable();
assertThat(routingTable == prevRoutingTable, equalTo(true));
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logger.info("Marking the shard as started");
RoutingNodes routingNodes = clusterState.routingNodes();
prevRoutingTable = routingTable;
routingTable =
strategy
.applyStartedShards(
clusterState, routingNodes.node("node1").shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(routingTable != prevRoutingTable, equalTo(true));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().get(0).state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).shards().get(0).currentNodeId(), equalTo("node1"));
logger.info("Starting another node and making sure nothing changed");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()).put(newNode("node2")))
.build();
prevRoutingTable = routingTable;
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(routingTable == prevRoutingTable, equalTo(true));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().get(0).state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).shards().get(0).currentNodeId(), equalTo("node1"));
logger.info("Killing node1 where the shard is, checking the shard is relocated");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()).remove("node1"))
.build();
prevRoutingTable = routingTable;
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(routingTable != prevRoutingTable, equalTo(true));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().get(0).state(), equalTo(INITIALIZING));
assertThat(
routingTable.index("test").shard(0).shards().get(0).currentNodeId(), equalTo("node2"));
logger.info(
"Start another node, make sure that things remain the same (shard is in node2 and initializing)");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()).put(newNode("node3")))
.build();
prevRoutingTable = routingTable;
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(routingTable == prevRoutingTable, equalTo(true));
logger.info("Start the shard on node 2");
routingNodes = clusterState.routingNodes();
prevRoutingTable = routingTable;
routingTable =
strategy
.applyStartedShards(
clusterState, routingNodes.node("node2").shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
assertThat(routingTable != prevRoutingTable, equalTo(true));
assertThat(routingTable.index("test").shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().size(), equalTo(1));
assertThat(routingTable.index("test").shard(0).shards().get(0).state(), equalTo(STARTED));
assertThat(
routingTable.index("test").shard(0).shards().get(0).currentNodeId(), equalTo("node2"));
}
@Test
public void unknownDiskUsageTest() {
Settings diskSettings =
settingsBuilder()
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED, true)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK, 0.7)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK, 0.85)
.build();
Map<String, DiskUsage> usages = new HashMap<>();
usages.put("node2", new DiskUsage("node2", 100, 50)); // 50% used
usages.put("node3", new DiskUsage("node3", 100, 0)); // 100% used
Map<String, Long> shardSizes = new HashMap<>();
shardSizes.put("[test][0][p]", 10L); // 10 bytes
shardSizes.put("[test][0][r]", 10L); // 10 bytes
final ClusterInfo clusterInfo =
new ClusterInfo(ImmutableMap.copyOf(usages), ImmutableMap.copyOf(shardSizes));
AllocationDeciders deciders =
new AllocationDeciders(
ImmutableSettings.EMPTY,
new HashSet<>(
Arrays.asList(
new SameShardAllocationDecider(ImmutableSettings.EMPTY),
new DiskThresholdDecider(diskSettings))));
ClusterInfoService cis =
new ClusterInfoService() {
@Override
public ClusterInfo getClusterInfo() {
logger.info("--> calling fake getClusterInfo");
return clusterInfo;
}
};
AllocationService strategy =
new AllocationService(
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(),
deciders,
new ShardsAllocators(),
cis);
MetaData metaData =
MetaData.builder()
.put(IndexMetaData.builder("test").numberOfShards(1).numberOfReplicas(0))
.build();
RoutingTable routingTable = RoutingTable.builder().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.metaData(metaData)
.routingTable(routingTable)
.build();
logger.info("--> adding node1");
clusterState =
ClusterState.builder(clusterState)
.nodes(
DiscoveryNodes.builder()
.put(newNode("node1"))
.put(
newNode(
"node3")) // node3 is added because DiskThresholdDecider automatically
// ignore single-node clusters
)
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
// Shard can be allocated to node1, even though it only has 25% free,
// because it's a primary that's never been allocated before
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> start the shards (primaries)");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// A single shard is started on node1, even though it normally would not
// be allowed, because it's a primary that hasn't been allocated, and node1
// is still below the high watermark (unlike node3)
assertThat(clusterState.routingNodes().shardsWithState(STARTED).size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().node("node1").size(), equalTo(1));
}
@Test
public void indexLevelShardsLimitAllocate() {
AllocationService strategy =
new AllocationService(
settingsBuilder().put("cluster.routing.allocation.concurrent_recoveries", 10).build());
logger.info("Building initial routing table");
MetaData metaData =
newMetaDataBuilder()
.put(
newIndexMetaDataBuilder("test")
.settings(
ImmutableSettings.settingsBuilder()
.put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, 4)
.put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 1)
.put(ShardsLimitAllocationDecider.INDEX_TOTAL_SHARDS_PER_NODE, 2)))
.build();
RoutingTable routingTable = routingTable().addAsNew(metaData.index("test")).build();
ClusterState clusterState =
newClusterStateBuilder().metaData(metaData).routingTable(routingTable).build();
logger.info("Adding two nodes and performing rerouting");
clusterState =
newClusterStateBuilder()
.state(clusterState)
.nodes(newNodesBuilder().put(newNode("node1")).put(newNode("node2")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(
clusterState
.readOnlyRoutingNodes()
.node("node1")
.numberOfShardsWithState(ShardRoutingState.INITIALIZING),
equalTo(2));
assertThat(
clusterState
.readOnlyRoutingNodes()
.node("node2")
.numberOfShardsWithState(ShardRoutingState.INITIALIZING),
equalTo(2));
logger.info("Start the primary shards");
RoutingNodes routingNodes = clusterState.routingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
assertThat(
clusterState
.readOnlyRoutingNodes()
.node("node1")
.numberOfShardsWithState(ShardRoutingState.STARTED),
equalTo(2));
assertThat(
clusterState
.readOnlyRoutingNodes()
.node("node1")
.numberOfShardsWithState(ShardRoutingState.INITIALIZING),
equalTo(0));
assertThat(
clusterState
.readOnlyRoutingNodes()
.node("node2")
.numberOfShardsWithState(ShardRoutingState.STARTED),
equalTo(2));
assertThat(
clusterState
.readOnlyRoutingNodes()
.node("node2")
.numberOfShardsWithState(ShardRoutingState.INITIALIZING),
equalTo(0));
assertThat(clusterState.readOnlyRoutingNodes().unassigned().size(), equalTo(4));
logger.info("Do another reroute, make sure its still not allocated");
routingNodes = clusterState.routingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = newClusterStateBuilder().state(clusterState).routingTable(routingTable).build();
}
@Test
public void testShardRelocationsTakenIntoAccount() {
Settings diskSettings =
settingsBuilder()
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED, true)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_INCLUDE_RELOCATIONS, true)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK, 0.7)
.put(DiskThresholdDecider.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK, 0.8)
.build();
Map<String, DiskUsage> usages = new HashMap<>();
usages.put("node1", new DiskUsage("node1", 100, 40)); // 60% used
usages.put("node2", new DiskUsage("node2", 100, 40)); // 60% used
usages.put("node2", new DiskUsage("node3", 100, 40)); // 60% used
Map<String, Long> shardSizes = new HashMap<>();
shardSizes.put("[test][0][p]", 14L); // 14 bytes
shardSizes.put("[test][0][r]", 14L);
shardSizes.put("[test2][0][p]", 1L); // 1 bytes
shardSizes.put("[test2][0][r]", 1L);
final ClusterInfo clusterInfo =
new ClusterInfo(ImmutableMap.copyOf(usages), ImmutableMap.copyOf(shardSizes));
AllocationDeciders deciders =
new AllocationDeciders(
ImmutableSettings.EMPTY,
new HashSet<>(
Arrays.asList(
new SameShardAllocationDecider(ImmutableSettings.EMPTY),
new DiskThresholdDecider(diskSettings))));
ClusterInfoService cis =
new ClusterInfoService() {
@Override
public ClusterInfo getClusterInfo() {
logger.info("--> calling fake getClusterInfo");
return clusterInfo;
}
};
AllocationService strategy =
new AllocationService(
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(),
deciders,
new ShardsAllocators(),
cis);
MetaData metaData =
MetaData.builder()
.put(IndexMetaData.builder("test").numberOfShards(1).numberOfReplicas(1))
.put(IndexMetaData.builder("test2").numberOfShards(1).numberOfReplicas(1))
.build();
RoutingTable routingTable =
RoutingTable.builder()
.addAsNew(metaData.index("test"))
.addAsNew(metaData.index("test2"))
.build();
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.metaData(metaData)
.routingTable(routingTable)
.build();
logger.info("--> adding two nodes");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder().put(newNode("node1")).put(newNode("node2")))
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// shards should be initializing
assertThat(clusterState.routingNodes().shardsWithState(INITIALIZING).size(), equalTo(4));
logger.info("--> start the shards");
routingTable =
strategy
.applyStartedShards(
clusterState, clusterState.routingNodes().shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
// Assert that we're able to start the primary and replicas
assertThat(
clusterState.routingNodes().shardsWithState(ShardRoutingState.STARTED).size(), equalTo(4));
logger.info("--> adding node3");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder(clusterState.nodes()).put(newNode("node3")))
.build();
AllocationCommand relocate1 =
new MoveAllocationCommand(new ShardId("test", 0), "node2", "node3");
AllocationCommands cmds = new AllocationCommands(relocate1);
routingTable = strategy.reroute(clusterState, cmds).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
logShardStates(clusterState);
AllocationCommand relocate2 =
new MoveAllocationCommand(new ShardId("test2", 0), "node2", "node3");
cmds = new AllocationCommands(relocate2);
try {
// The shard for the "test" index is already being relocated to
// node3, which will put it over the low watermark when it
// completes, with shard relocations taken into account this should
// throw an exception about not being able to complete
strategy.reroute(clusterState, cmds).routingTable();
fail("should not have been able to reroute the shard");
} catch (ElasticsearchIllegalArgumentException e) {
assertThat(
"can't allocated because there isn't enough room: " + e.getMessage(),
e.getMessage().contains("less than required [30.0%] free disk on node, free: [26.0%]"),
equalTo(true));
}
}
