@Test
public void testMultiIndexUnevenNodes() {
AllocationService strategy =
createAllocationService(
settingsBuilder()
.put("cluster.routing.allocation.concurrent_recoveries", 10)
.put("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).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().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 testNoRebalanceOnPrimaryOverload() {
Settings.Builder settings = settingsBuilder();
AllocationService strategy =
new AllocationService(
settings.build(),
randomAllocationDeciders(
settings.build(),
new NodeSettingsService(Settings.Builder.EMPTY_SETTINGS),
getRandom()),
new ShardsAllocators(
settings.build(),
NoopGatewayAllocator.INSTANCE,
new ShardsAllocator() {
@Override
public boolean rebalance(RoutingAllocation allocation) {
return false;
}
@Override
public boolean move(
ShardRouting shardRouting, RoutingNode node, RoutingAllocation allocation) {
return false;
}
@Override
public void applyStartedShards(StartedRerouteAllocation allocation) {}
@Override
public void applyFailedShards(FailedRerouteAllocation allocation) {}
/*
* // this allocator tries to rebuild this scenario where a rebalance is
* // triggered solely by the primary overload on node [1] where a shard
* // is rebalanced to node 0
routing_nodes:
-----node_id[0][V]
--------[test][0], node[0], [R], s[STARTED]
--------[test][4], node[0], [R], s[STARTED]
-----node_id[1][V]
--------[test][0], node[1], [P], s[STARTED]
--------[test][1], node[1], [P], s[STARTED]
--------[test][3], node[1], [R], s[STARTED]
-----node_id[2][V]
--------[test][1], node[2], [R], s[STARTED]
--------[test][2], node[2], [R], s[STARTED]
--------[test][4], node[2], [P], s[STARTED]
-----node_id[3][V]
--------[test][2], node[3], [P], s[STARTED]
--------[test][3], node[3], [P], s[STARTED]
---- unassigned
*/
@Override
public boolean allocateUnassigned(RoutingAllocation allocation) {
RoutingNodes.UnassignedShards unassigned =
allocation.routingNodes().unassigned();
boolean changed = !unassigned.isEmpty();
for (ShardRouting sr : unassigned) {
switch (sr.id()) {
case 0:
if (sr.primary()) {
allocation.routingNodes().initialize(sr, "node1");
} else {
allocation.routingNodes().initialize(sr, "node0");
}
break;
case 1:
if (sr.primary()) {
allocation.routingNodes().initialize(sr, "node1");
} else {
allocation.routingNodes().initialize(sr, "node2");
}
break;
case 2:
if (sr.primary()) {
allocation.routingNodes().initialize(sr, "node3");
} else {
allocation.routingNodes().initialize(sr, "node2");
}
break;
case 3:
if (sr.primary()) {
allocation.routingNodes().initialize(sr, "node3");
} else {
allocation.routingNodes().initialize(sr, "node1");
}
break;
case 4:
if (sr.primary()) {
allocation.routingNodes().initialize(sr, "node2");
} else {
allocation.routingNodes().initialize(sr, "node0");
}
break;
}
}
unassigned.clear();
return changed;
}
}),
EmptyClusterInfoService.INSTANCE);
MetaData.Builder metaDataBuilder = MetaData.builder();
RoutingTable.Builder routingTableBuilder = RoutingTable.builder();
IndexMetaData.Builder indexMeta =
IndexMetaData.builder("test")
.settings(settings(Version.CURRENT))
.numberOfShards(5)
.numberOfReplicas(1);
metaDataBuilder = metaDataBuilder.put(indexMeta);
MetaData metaData = metaDataBuilder.build();
for (ObjectCursor<IndexMetaData> cursor : metaData.indices().values()) {
routingTableBuilder.addAsNew(cursor.value);
}
RoutingTable routingTable = routingTableBuilder.build();
DiscoveryNodes.Builder nodes = DiscoveryNodes.builder();
for (int i = 0; i < 4; i++) {
DiscoveryNode node = newNode("node" + i);
nodes.put(node);
}
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.nodes(nodes)
.metaData(metaData)
.routingTable(routingTable)
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
RoutingNodes routingNodes = clusterState.getRoutingNodes();
for (RoutingNode routingNode : routingNodes) {
for (ShardRouting shardRouting : routingNode) {
assertThat(shardRouting.state(), Matchers.equalTo(ShardRoutingState.INITIALIZING));
}
}
strategy = createAllocationService(settings.build());
logger.info("use the new allocator and check if it moves shards");
routingNodes = clusterState.getRoutingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
for (RoutingNode routingNode : routingNodes) {
for (ShardRouting shardRouting : routingNode) {
assertThat(shardRouting.state(), Matchers.equalTo(ShardRoutingState.STARTED));
}
}
logger.info("start the replica shards");
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
for (RoutingNode routingNode : routingNodes) {
for (ShardRouting shardRouting : routingNode) {
assertThat(shardRouting.state(), Matchers.equalTo(ShardRoutingState.STARTED));
}
}
logger.info("rebalancing");
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
for (RoutingNode routingNode : routingNodes) {
for (ShardRouting shardRouting : routingNode) {
assertThat(shardRouting.state(), Matchers.equalTo(ShardRoutingState.STARTED));
}
}
}
@Test
public void testMultiIndexEvenDistribution() {
AllocationService strategy =
createAllocationService(
settingsBuilder()
.put("cluster.routing.allocation.concurrent_recoveries", 10)
.put("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).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().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 (MutableShardRouting 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));
}
private ClusterState initCluster(AllocationService strategy) {
MetaData.Builder metaDataBuilder = MetaData.builder();
RoutingTable.Builder routingTableBuilder = RoutingTable.builder();
for (int i = 0; i < numberOfIndices; i++) {
IndexMetaData.Builder index =
IndexMetaData.builder("test" + i)
.settings(settings(Version.CURRENT))
.numberOfShards(numberOfShards)
.numberOfReplicas(numberOfReplicas);
metaDataBuilder = metaDataBuilder.put(index);
}
MetaData metaData = metaDataBuilder.build();
for (ObjectCursor<IndexMetaData> cursor : metaData.indices().values()) {
routingTableBuilder.addAsNew(cursor.value);
}
RoutingTable routingTable = routingTableBuilder.build();
logger.info("start " + numberOfNodes + " nodes");
DiscoveryNodes.Builder nodes = DiscoveryNodes.builder();
for (int i = 0; i < numberOfNodes; i++) {
nodes.put(newNode("node" + i));
}
ClusterState clusterState =
ClusterState.builder(org.elasticsearch.cluster.ClusterName.DEFAULT)
.nodes(nodes)
.metaData(metaData)
.routingTable(routingTable)
.build();
routingTable = strategy.reroute(clusterState).routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
RoutingNodes routingNodes = clusterState.getRoutingNodes();
logger.info("restart all the primary shards, replicas will start initializing");
routingNodes = clusterState.getRoutingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
logger.info("start the replica shards");
routingNodes = clusterState.getRoutingNodes();
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
logger.info("complete rebalancing");
RoutingTable prev = routingTable;
while (true) {
routingTable =
strategy
.applyStartedShards(clusterState, routingNodes.shardsWithState(INITIALIZING))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
routingNodes = clusterState.getRoutingNodes();
if (routingTable == prev) break;
prev = routingTable;
}
return clusterState;
}
