@Test
public void simpleFlagTests() {
AllocationService allocation =
createAllocationService(
settingsBuilder().put("cluster.routing.allocation.concurrent_recoveries", 10).build());
logger.info("creating an index with 1 shard, no replica");
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(
clusterState.routingTable().index("test").shard(0).primaryAllocatedPostApi(),
equalTo(false));
logger.info("adding two nodes and performing rerouting");
clusterState =
ClusterState.builder(clusterState)
.nodes(DiscoveryNodes.builder().put(newNode("node1")).put(newNode("node2")))
.build();
RoutingAllocation.Result rerouteResult = allocation.reroute(clusterState);
clusterState =
ClusterState.builder(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 =
ClusterState.builder(clusterState).routingTable(rerouteResult.routingTable()).build();
assertThat(
clusterState.routingTable().index("test").shard(0).primaryAllocatedPostApi(),
equalTo(true));
}
/**
* 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 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());
}
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 testFailedAllocation() {
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);
ShardRouting unassignedPrimary = routingTable.index("idx").shard(0).shards().get(0);
assertEquals(unassignedPrimary.state(), INITIALIZING);
assertEquals(unassignedPrimary.unassignedInfo().getNumFailedAllocations(), i + 1);
assertEquals(unassignedPrimary.unassignedInfo().getMessage(), "boom" + i);
// MaxRetryAllocationDecider#canForceAllocatePrimary should return YES decisions because
// canAllocate returns YES here
assertEquals(
Decision.YES,
new MaxRetryAllocationDecider(Settings.EMPTY)
.canForceAllocatePrimary(
unassignedPrimary,
null,
new RoutingAllocation(null, null, clusterState, null, 0, false)));
}
// 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);
ShardRouting unassignedPrimary = routingTable.index("idx").shard(0).shards().get(0);
assertEquals(unassignedPrimary.unassignedInfo().getNumFailedAllocations(), retries);
assertEquals(unassignedPrimary.state(), UNASSIGNED);
assertEquals(unassignedPrimary.unassignedInfo().getMessage(), "boom");
// MaxRetryAllocationDecider#canForceAllocatePrimary should return a NO decision because
// canAllocate returns NO here
assertEquals(
Decision.NO,
new MaxRetryAllocationDecider(Settings.EMPTY)
.canForceAllocatePrimary(
unassignedPrimary,
null,
new RoutingAllocation(null, null, clusterState, null, 0, false)));
}
// change the settings and ensure we can do another round of allocation for that index.
clusterState =
ClusterState.builder(clusterState)
.routingTable(routingTable)
.metaData(
MetaData.builder(clusterState.metaData())
.put(
IndexMetaData.builder(clusterState.metaData().index("idx"))
.settings(
Settings.builder()
.put(clusterState.metaData().index("idx").getSettings())
.put("index.allocation.max_retries", retries + 1)
.build())
.build(),
true)
.build())
.build();
RoutingAllocation.Result result = strategy.reroute(clusterState, "settings changed", false);
assertTrue(result.changed());
routingTable = result.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
// good we are initializing and we are maintaining failure information
assertEquals(routingTable.index("idx").shards().size(), 1);
ShardRouting unassignedPrimary = routingTable.index("idx").shard(0).shards().get(0);
assertEquals(unassignedPrimary.unassignedInfo().getNumFailedAllocations(), retries);
assertEquals(unassignedPrimary.state(), INITIALIZING);
assertEquals(unassignedPrimary.unassignedInfo().getMessage(), "boom");
// bumped up the max retry count, so canForceAllocatePrimary should return a YES decision
assertEquals(
Decision.YES,
new MaxRetryAllocationDecider(Settings.EMPTY)
.canForceAllocatePrimary(
routingTable.index("idx").shard(0).shards().get(0),
null,
new RoutingAllocation(null, null, clusterState, null, 0, false)));
// now we start the shard
routingTable =
strategy
.applyStartedShards(
clusterState,
Collections.singletonList(routingTable.index("idx").shard(0).shards().get(0)))
.routingTable();
clusterState = ClusterState.builder(clusterState).routingTable(routingTable).build();
// all counters have been reset to 0 ie. no unassigned info
assertEquals(routingTable.index("idx").shards().size(), 1);
assertNull(routingTable.index("idx").shard(0).shards().get(0).unassignedInfo());
assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), STARTED);
// now fail again and see if it has a new counter
List<FailedRerouteAllocation.FailedShard> failedShards =
Collections.singletonList(
new FailedRerouteAllocation.FailedShard(
routingTable.index("idx").shard(0).shards().get(0),
"ZOOOMG",
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);
unassignedPrimary = routingTable.index("idx").shard(0).shards().get(0);
assertEquals(unassignedPrimary.unassignedInfo().getNumFailedAllocations(), 1);
assertEquals(unassignedPrimary.state(), INITIALIZING);
assertEquals(unassignedPrimary.unassignedInfo().getMessage(), "ZOOOMG");
// Counter reset, so MaxRetryAllocationDecider#canForceAllocatePrimary should return a YES
// decision
assertEquals(
Decision.YES,
new MaxRetryAllocationDecider(Settings.EMPTY)
.canForceAllocatePrimary(
unassignedPrimary,
null,
new RoutingAllocation(null, null, clusterState, null, 0, false)));
}
