@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 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));
}
}
@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 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
@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 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").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 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 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 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));
}
@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());
}