use of org.opensearch.cluster.ClusterInfoService in project OpenSearch by opensearch-project.
the class DiskThresholdDeciderIT method refreshDiskUsage.
private void refreshDiskUsage() {
final ClusterInfoService clusterInfoService = internalCluster().getCurrentMasterNodeInstance(ClusterInfoService.class);
((InternalClusterInfoService) clusterInfoService).refresh();
// even though it's now possible to achieve better balance, so we have to do an explicit reroute. TODO fix this?
if (StreamSupport.stream(clusterInfoService.getClusterInfo().getNodeMostAvailableDiskUsages().values().spliterator(), false).allMatch(cur -> cur.value.getFreeBytes() > WATERMARK_BYTES)) {
assertAcked(client().admin().cluster().prepareReroute());
}
assertFalse(client().admin().cluster().prepareHealth().setWaitForEvents(Priority.LANGUID).setWaitForNoRelocatingShards(true).setWaitForNoInitializingShards(true).get().isTimedOut());
}
use of org.opensearch.cluster.ClusterInfoService in project OpenSearch by opensearch-project.
the class DiskThresholdDeciderIT method testHighWatermarkNotExceeded.
public void testHighWatermarkNotExceeded() throws Exception {
internalCluster().startMasterOnlyNode();
internalCluster().startDataOnlyNode();
final String dataNodeName = internalCluster().startDataOnlyNode();
ensureStableCluster(3);
final InternalClusterInfoService clusterInfoService = (InternalClusterInfoService) internalCluster().getCurrentMasterNodeInstance(ClusterInfoService.class);
internalCluster().getCurrentMasterNodeInstance(ClusterService.class).addListener(event -> clusterInfoService.refresh());
final String dataNode0Id = internalCluster().getInstance(NodeEnvironment.class, dataNodeName).nodeId();
final Path dataNode0Path = internalCluster().getInstance(Environment.class, dataNodeName).dataFiles()[0];
final String indexName = randomAlphaOfLength(10).toLowerCase(Locale.ROOT);
createIndex(indexName, Settings.builder().put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 0).put(IndexMetadata.SETTING_NUMBER_OF_SHARDS, 6).put(INDEX_STORE_STATS_REFRESH_INTERVAL_SETTING.getKey(), "0ms").build());
final long minShardSize = createReasonableSizedShards(indexName);
// reduce disk size of node 0 so that no shards fit below the high watermark, forcing all shards onto the other data node
// (subtract the translog size since the disk threshold decider ignores this and may therefore move the shard back again)
fileSystemProvider.getTestFileStore(dataNode0Path).setTotalSpace(minShardSize + WATERMARK_BYTES - 1L);
assertBusyWithDiskUsageRefresh(dataNode0Id, indexName, empty());
// increase disk size of node 0 to allow just enough room for one shard, and check that it's rebalanced back
fileSystemProvider.getTestFileStore(dataNode0Path).setTotalSpace(minShardSize + WATERMARK_BYTES + 1L);
assertBusyWithDiskUsageRefresh(dataNode0Id, indexName, hasSize(1));
}
use of org.opensearch.cluster.ClusterInfoService in project OpenSearch by opensearch-project.
the class DiskThresholdDeciderTests method testDiskThresholdWithAbsoluteSizes.
public void testDiskThresholdWithAbsoluteSizes() {
Settings diskSettings = Settings.builder().put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED_SETTING.getKey(), true).put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK_SETTING.getKey(), "30b").put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK_SETTING.getKey(), "9b").put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_FLOOD_STAGE_WATERMARK_SETTING.getKey(), "5b").build();
ImmutableOpenMap.Builder<String, DiskUsage> usagesBuilder = ImmutableOpenMap.builder();
// 90% used
usagesBuilder.put("node1", new DiskUsage("node1", "n1", "/dev/null", 100, 10));
// 90% used
usagesBuilder.put("node2", new DiskUsage("node2", "n2", "/dev/null", 100, 10));
// 40% used
usagesBuilder.put("node3", new DiskUsage("node3", "n3", "/dev/null", 100, 60));
// 20% used
usagesBuilder.put("node4", new DiskUsage("node4", "n4", "/dev/null", 100, 80));
// 15% used
usagesBuilder.put("node5", new DiskUsage("node5", "n5", "/dev/null", 100, 85));
ImmutableOpenMap<String, DiskUsage> usages = usagesBuilder.build();
ImmutableOpenMap.Builder<String, Long> shardSizesBuilder = ImmutableOpenMap.builder();
// 10 bytes
shardSizesBuilder.put("[test][0][p]", 10L);
shardSizesBuilder.put("[test][0][r]", 10L);
ImmutableOpenMap<String, Long> shardSizes = shardSizesBuilder.build();
final ClusterInfo clusterInfo = new DevNullClusterInfo(usages, usages, shardSizes);
ClusterSettings clusterSettings = new ClusterSettings(Settings.EMPTY, ClusterSettings.BUILT_IN_CLUSTER_SETTINGS);
AllocationDeciders deciders = new AllocationDeciders(new HashSet<>(Arrays.asList(new SameShardAllocationDecider(Settings.EMPTY, clusterSettings), makeDecider(diskSettings))));
ClusterInfoService cis = () -> {
logger.info("--> calling fake getClusterInfo");
return clusterInfo;
};
AllocationService strategy = new AllocationService(deciders, new TestGatewayAllocator(), new BalancedShardsAllocator(Settings.EMPTY), cis, EmptySnapshotsInfoService.INSTANCE);
Metadata metadata = Metadata.builder().put(IndexMetadata.builder("test").settings(settings(Version.CURRENT)).numberOfShards(1).numberOfReplicas(2)).build();
RoutingTable initialRoutingTable = RoutingTable.builder().addAsNew(metadata.index("test")).build();
ClusterState clusterState = ClusterState.builder(ClusterName.CLUSTER_NAME_SETTING.getDefault(Settings.EMPTY)).metadata(metadata).routingTable(initialRoutingTable).build();
logger.info("--> adding node1 and node2 node");
clusterState = ClusterState.builder(clusterState).nodes(DiscoveryNodes.builder().add(newNode("node1")).add(newNode("node2"))).build();
clusterState = strategy.reroute(clusterState, "reroute");
logShardStates(clusterState);
// Primary should initialize, even though both nodes are over the limit initialize
assertThat(clusterState.getRoutingNodes().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
usagesBuilder = ImmutableOpenMap.builder(usages);
// 65% used
usagesBuilder.put(nodeWithoutPrimary, new DiskUsage(nodeWithoutPrimary, "", "/dev/null", 100, 35));
usages = usagesBuilder.build();
final ClusterInfo clusterInfo2 = new DevNullClusterInfo(usages, usages, shardSizes);
cis = () -> {
logger.info("--> calling fake getClusterInfo");
return clusterInfo2;
};
strategy = new AllocationService(deciders, new TestGatewayAllocator(), new BalancedShardsAllocator(Settings.EMPTY), cis, EmptySnapshotsInfoService.INSTANCE);
clusterState = strategy.reroute(clusterState, "reroute");
logShardStates(clusterState);
// Now the replica should be able to initialize
assertThat(clusterState.getRoutingNodes().shardsWithState(INITIALIZING).size(), equalTo(2));
logger.info("--> start the shards (primaries)");
clusterState = startInitializingShardsAndReroute(strategy, clusterState);
logShardStates(clusterState);
// Assert that we're able to start the primary and replica, since they were both initializing
assertThat(clusterState.getRoutingNodes().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()).add(newNode("node3"))).build();
clusterState = strategy.reroute(clusterState, "reroute");
logShardStates(clusterState);
// Assert that the replica is initialized now that node3 is available with enough space
assertThat(clusterState.getRoutingNodes().shardsWithState(ShardRoutingState.STARTED).size(), equalTo(2));
assertThat(clusterState.getRoutingNodes().shardsWithState(ShardRoutingState.INITIALIZING).size(), equalTo(1));
logger.info("--> start the shards (replicas)");
clusterState = startInitializingShardsAndReroute(strategy, clusterState);
logShardStates(clusterState);
// Assert that all replicas could be started
assertThat(clusterState.getRoutingNodes().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 = Settings.builder().put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED_SETTING.getKey(), true).put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK_SETTING.getKey(), "40b").put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK_SETTING.getKey(), "30b").put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_FLOOD_STAGE_WATERMARK_SETTING.getKey(), "20b").build();
deciders = new AllocationDeciders(new HashSet<>(Arrays.asList(new SameShardAllocationDecider(Settings.EMPTY, clusterSettings), makeDecider(diskSettings))));
strategy = new AllocationService(deciders, new TestGatewayAllocator(), new BalancedShardsAllocator(Settings.EMPTY), cis, EmptySnapshotsInfoService.INSTANCE);
clusterState = strategy.reroute(clusterState, "reroute");
logShardStates(clusterState);
// Shards remain started
assertThat(clusterState.getRoutingNodes().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 = Settings.builder().put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED_SETTING.getKey(), true).put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK_SETTING.getKey(), "50b").put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK_SETTING.getKey(), "40b").put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_FLOOD_STAGE_WATERMARK_SETTING.getKey(), "30b").build();
deciders = new AllocationDeciders(new HashSet<>(Arrays.asList(new SameShardAllocationDecider(Settings.EMPTY, clusterSettings), makeDecider(diskSettings))));
strategy = new AllocationService(deciders, new TestGatewayAllocator(), new BalancedShardsAllocator(Settings.EMPTY), cis, EmptySnapshotsInfoService.INSTANCE);
clusterState = strategy.reroute(clusterState, "reroute");
logShardStates(clusterState);
// Shards remain started
assertThat(clusterState.getRoutingNodes().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()).add(newNode("node4"))).build();
clusterState = strategy.reroute(clusterState, "reroute");
logShardStates(clusterState);
// Shards remain started
assertThat(clusterState.getRoutingNodes().shardsWithState(STARTED).size(), equalTo(2));
// One shard is relocating off of node1
assertThat(clusterState.getRoutingNodes().shardsWithState(RELOCATING).size(), equalTo(1));
assertThat(clusterState.getRoutingNodes().shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> apply INITIALIZING shards");
clusterState = startInitializingShardsAndReroute(strategy, clusterState);
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()).add(newNode("node5"))).build();
clusterState = strategy.reroute(clusterState, "reroute");
logShardStates(clusterState);
// Shards remain started on node3 and node4
assertThat(clusterState.getRoutingNodes().shardsWithState(STARTED).size(), equalTo(2));
// One shard is relocating off of node2 now
assertThat(clusterState.getRoutingNodes().shardsWithState(RELOCATING).size(), equalTo(1));
// Initializing on node5
assertThat(clusterState.getRoutingNodes().shardsWithState(INITIALIZING).size(), equalTo(1));
logger.info("--> apply INITIALIZING shards");
clusterState = startInitializingShardsAndReroute(strategy, clusterState);
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));
}
use of org.opensearch.cluster.ClusterInfoService in project OpenSearch by opensearch-project.
the class DiskThresholdDeciderTests method testDiskThresholdWithSnapshotShardSizes.
public void testDiskThresholdWithSnapshotShardSizes() {
final long shardSizeInBytes = randomBoolean() ? 10L : 50L;
logger.info("--> using shard size [{}]", shardSizeInBytes);
final Settings diskSettings = Settings.builder().put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED_SETTING.getKey(), true).put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK_SETTING.getKey(), "90%").put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK_SETTING.getKey(), "95%").build();
final ImmutableOpenMap.Builder<String, DiskUsage> usagesBuilder = ImmutableOpenMap.builder();
// 79% used
usagesBuilder.put("node1", new DiskUsage("node1", "n1", "/dev/null", 100, 21));
// 99% used
usagesBuilder.put("node2", new DiskUsage("node2", "n2", "/dev/null", 100, 1));
final ImmutableOpenMap<String, DiskUsage> usages = usagesBuilder.build();
final ClusterInfoService clusterInfoService = () -> new DevNullClusterInfo(usages, usages, ImmutableOpenMap.of());
final AllocationDeciders deciders = new AllocationDeciders(new HashSet<>(Arrays.asList(new RestoreInProgressAllocationDecider(), new SameShardAllocationDecider(Settings.EMPTY, new ClusterSettings(Settings.EMPTY, ClusterSettings.BUILT_IN_CLUSTER_SETTINGS)), makeDecider(diskSettings))));
final Snapshot snapshot = new Snapshot("_repository", new SnapshotId("_snapshot_name", UUIDs.randomBase64UUID(random())));
final IndexId indexId = new IndexId("_indexid_name", UUIDs.randomBase64UUID(random()));
final ShardId shardId = new ShardId(new Index("test", IndexMetadata.INDEX_UUID_NA_VALUE), 0);
final Metadata metadata = Metadata.builder().put(IndexMetadata.builder("test").settings(settings(Version.CURRENT)).numberOfShards(1).numberOfReplicas(0).putInSyncAllocationIds(0, singleton(AllocationId.newInitializing().getId()))).build();
final RoutingTable routingTable = RoutingTable.builder().addAsNewRestore(metadata.index("test"), new RecoverySource.SnapshotRecoverySource("_restore_uuid", snapshot, Version.CURRENT, indexId), new IntHashSet()).build();
final ImmutableOpenMap.Builder<ShardId, RestoreInProgress.ShardRestoreStatus> shards = ImmutableOpenMap.builder();
shards.put(shardId, new RestoreInProgress.ShardRestoreStatus("node1"));
final RestoreInProgress.Builder restores = new RestoreInProgress.Builder().add(new RestoreInProgress.Entry("_restore_uuid", snapshot, RestoreInProgress.State.INIT, singletonList("test"), shards.build()));
ClusterState clusterState = ClusterState.builder(new ClusterName(getTestName())).metadata(metadata).routingTable(routingTable).putCustom(RestoreInProgress.TYPE, restores.build()).nodes(// node2 is added because DiskThresholdDecider
DiscoveryNodes.builder().add(newNode("node1")).add(newNode("node2"))).build();
assertThat(clusterState.getRoutingNodes().shardsWithState(UNASSIGNED).stream().map(ShardRouting::unassignedInfo).allMatch(unassignedInfo -> Reason.NEW_INDEX_RESTORED.equals(unassignedInfo.getReason())), is(true));
assertThat(clusterState.getRoutingNodes().shardsWithState(UNASSIGNED).stream().map(ShardRouting::unassignedInfo).allMatch(unassignedInfo -> AllocationStatus.NO_ATTEMPT.equals(unassignedInfo.getLastAllocationStatus())), is(true));
assertThat(clusterState.getRoutingNodes().shardsWithState(UNASSIGNED).size(), equalTo(1));
final AtomicReference<SnapshotShardSizeInfo> snapshotShardSizeInfoRef = new AtomicReference<>(SnapshotShardSizeInfo.EMPTY);
final AllocationService strategy = new AllocationService(deciders, new TestGatewayAllocator(), new BalancedShardsAllocator(Settings.EMPTY), clusterInfoService, snapshotShardSizeInfoRef::get);
// reroute triggers snapshot shard size fetching
clusterState = strategy.reroute(clusterState, "reroute");
logShardStates(clusterState);
// shard cannot be assigned yet as the snapshot shard size is unknown
assertThat(clusterState.getRoutingNodes().shardsWithState(UNASSIGNED).stream().map(ShardRouting::unassignedInfo).allMatch(unassignedInfo -> AllocationStatus.FETCHING_SHARD_DATA.equals(unassignedInfo.getLastAllocationStatus())), is(true));
assertThat(clusterState.getRoutingNodes().shardsWithState(UNASSIGNED).size(), equalTo(1));
final SnapshotShard snapshotShard = new SnapshotShard(snapshot, indexId, shardId);
final ImmutableOpenMap.Builder<SnapshotShard, Long> snapshotShardSizes = ImmutableOpenMap.builder();
final boolean shouldAllocate;
if (randomBoolean()) {
logger.info("--> simulating snapshot shards size retrieval success");
snapshotShardSizes.put(snapshotShard, shardSizeInBytes);
logger.info("--> shard allocation depends on its size");
shouldAllocate = shardSizeInBytes < usages.get("node1").getFreeBytes();
} else {
logger.info("--> simulating snapshot shards size retrieval failure");
snapshotShardSizes.put(snapshotShard, ShardRouting.UNAVAILABLE_EXPECTED_SHARD_SIZE);
logger.info("--> shard is always allocated when its size could not be retrieved");
shouldAllocate = true;
}
snapshotShardSizeInfoRef.set(new SnapshotShardSizeInfo(snapshotShardSizes.build()));
// reroute uses the previous snapshot shard size
clusterState = startInitializingShardsAndReroute(strategy, clusterState);
logShardStates(clusterState);
assertThat(clusterState.getRoutingNodes().shardsWithState(UNASSIGNED).size(), equalTo(shouldAllocate ? 0 : 1));
assertThat(clusterState.getRoutingNodes().shardsWithState("test", INITIALIZING, STARTED).size(), equalTo(shouldAllocate ? 1 : 0));
}
use of org.opensearch.cluster.ClusterInfoService in project OpenSearch by opensearch-project.
the class DiskThresholdDeciderTests method testCanRemainWithShardRelocatingAway.
public void testCanRemainWithShardRelocatingAway() {
Settings diskSettings = Settings.builder().put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED_SETTING.getKey(), true).put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_LOW_DISK_WATERMARK_SETTING.getKey(), "60%").put(DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_HIGH_DISK_WATERMARK_SETTING.getKey(), "70%").build();
// We have an index with 2 primary shards each taking 40 bytes. Each node has 100 bytes available
ImmutableOpenMap.Builder<String, DiskUsage> usagesBuilder = ImmutableOpenMap.builder();
// 80% used
usagesBuilder.put("node1", new DiskUsage("node1", "n1", "/dev/null", 100, 20));
// 0% used
usagesBuilder.put("node2", new DiskUsage("node2", "n2", "/dev/null", 100, 100));
ImmutableOpenMap<String, DiskUsage> usages = usagesBuilder.build();
ImmutableOpenMap.Builder<String, Long> shardSizesBuilder = ImmutableOpenMap.builder();
shardSizesBuilder.put("[test][0][p]", 40L);
shardSizesBuilder.put("[test][1][p]", 40L);
shardSizesBuilder.put("[foo][0][p]", 10L);
ImmutableOpenMap<String, Long> shardSizes = shardSizesBuilder.build();
final ClusterInfo clusterInfo = new DevNullClusterInfo(usages, usages, shardSizes);
DiskThresholdDecider diskThresholdDecider = makeDecider(diskSettings);
Metadata metadata = Metadata.builder().put(IndexMetadata.builder("test").settings(settings(Version.CURRENT)).numberOfShards(2).numberOfReplicas(0)).put(IndexMetadata.builder("foo").settings(settings(Version.CURRENT)).numberOfShards(1).numberOfReplicas(0)).build();
RoutingTable initialRoutingTable = RoutingTable.builder().addAsNew(metadata.index("test")).addAsNew(metadata.index("foo")).build();
DiscoveryNode discoveryNode1 = new DiscoveryNode("node1", buildNewFakeTransportAddress(), emptyMap(), MASTER_DATA_ROLES, Version.CURRENT);
DiscoveryNode discoveryNode2 = new DiscoveryNode("node2", buildNewFakeTransportAddress(), emptyMap(), MASTER_DATA_ROLES, Version.CURRENT);
DiscoveryNodes discoveryNodes = DiscoveryNodes.builder().add(discoveryNode1).add(discoveryNode2).build();
ClusterState baseClusterState = ClusterState.builder(ClusterName.CLUSTER_NAME_SETTING.getDefault(Settings.EMPTY)).metadata(metadata).routingTable(initialRoutingTable).nodes(discoveryNodes).build();
// Two shards consuming each 80% of disk space while 70% is allowed, so shard 0 isn't allowed here
ShardRouting firstRouting = TestShardRouting.newShardRouting("test", 0, "node1", null, true, ShardRoutingState.STARTED);
ShardRouting secondRouting = TestShardRouting.newShardRouting("test", 1, "node1", null, true, ShardRoutingState.STARTED);
RoutingNode firstRoutingNode = new RoutingNode("node1", discoveryNode1, firstRouting, secondRouting);
RoutingTable.Builder builder = RoutingTable.builder().add(IndexRoutingTable.builder(firstRouting.index()).addIndexShard(new IndexShardRoutingTable.Builder(firstRouting.shardId()).addShard(firstRouting).build()).addIndexShard(new IndexShardRoutingTable.Builder(secondRouting.shardId()).addShard(secondRouting).build()));
ClusterState clusterState = ClusterState.builder(baseClusterState).routingTable(builder.build()).build();
RoutingAllocation routingAllocation = new RoutingAllocation(null, new RoutingNodes(clusterState), clusterState, clusterInfo, null, System.nanoTime());
routingAllocation.debugDecision(true);
Decision decision = diskThresholdDecider.canRemain(firstRouting, firstRoutingNode, routingAllocation);
assertThat(decision.type(), equalTo(Decision.Type.NO));
assertThat(((Decision.Single) decision).getExplanation(), containsString("the shard cannot remain on this node because it is above the high watermark cluster setting " + "[cluster.routing.allocation.disk.watermark.high=70%] and there is less than the required [30.0%] free disk on node, " + "actual free: [20.0%]"));
// Two shards consuming each 80% of disk space while 70% is allowed, but one is relocating, so shard 0 can stay
firstRouting = TestShardRouting.newShardRouting("test", 0, "node1", null, true, ShardRoutingState.STARTED);
secondRouting = TestShardRouting.newShardRouting("test", 1, "node1", "node2", true, ShardRoutingState.RELOCATING);
ShardRouting fooRouting = TestShardRouting.newShardRouting("foo", 0, null, true, ShardRoutingState.UNASSIGNED);
firstRoutingNode = new RoutingNode("node1", discoveryNode1, firstRouting, secondRouting);
builder = RoutingTable.builder().add(IndexRoutingTable.builder(firstRouting.index()).addIndexShard(new IndexShardRoutingTable.Builder(firstRouting.shardId()).addShard(firstRouting).build()).addIndexShard(new IndexShardRoutingTable.Builder(secondRouting.shardId()).addShard(secondRouting).build()));
clusterState = ClusterState.builder(baseClusterState).routingTable(builder.build()).build();
routingAllocation = new RoutingAllocation(null, new RoutingNodes(clusterState), clusterState, clusterInfo, null, System.nanoTime());
routingAllocation.debugDecision(true);
decision = diskThresholdDecider.canRemain(firstRouting, firstRoutingNode, routingAllocation);
assertThat(decision.type(), equalTo(Decision.Type.YES));
assertEquals("there is enough disk on this node for the shard to remain, free: [60b]", ((Decision.Single) decision).getExplanation());
decision = diskThresholdDecider.canAllocate(fooRouting, firstRoutingNode, routingAllocation);
assertThat(decision.type(), equalTo(Decision.Type.NO));
if (fooRouting.recoverySource().getType() == RecoverySource.Type.EMPTY_STORE) {
assertThat(((Decision.Single) decision).getExplanation(), containsString("the node is above the high watermark cluster setting [cluster.routing.allocation.disk.watermark.high=70%], using " + "more disk space than the maximum allowed [70.0%], actual free: [20.0%]"));
} else {
assertThat(((Decision.Single) decision).getExplanation(), containsString("the node is above the low watermark cluster setting [cluster.routing.allocation.disk.watermark.low=60%], using more " + "disk space than the maximum allowed [60.0%], actual free: [20.0%]"));
}
// Creating AllocationService instance and the services it depends on...
ClusterInfoService cis = () -> {
logger.info("--> calling fake getClusterInfo");
return clusterInfo;
};
AllocationDeciders deciders = new AllocationDeciders(new HashSet<>(Arrays.asList(new SameShardAllocationDecider(Settings.EMPTY, new ClusterSettings(Settings.EMPTY, ClusterSettings.BUILT_IN_CLUSTER_SETTINGS)), diskThresholdDecider)));
AllocationService strategy = new AllocationService(deciders, new TestGatewayAllocator(), new BalancedShardsAllocator(Settings.EMPTY), cis, EmptySnapshotsInfoService.INSTANCE);
// Ensure that the reroute call doesn't alter the routing table, since the first primary is relocating away
// and therefor we will have sufficient disk space on node1.
ClusterState result = strategy.reroute(clusterState, "reroute");
assertThat(result, equalTo(clusterState));
assertThat(result.routingTable().index("test").getShards().get(0).primaryShard().state(), equalTo(STARTED));
assertThat(result.routingTable().index("test").getShards().get(0).primaryShard().currentNodeId(), equalTo("node1"));
assertThat(result.routingTable().index("test").getShards().get(0).primaryShard().relocatingNodeId(), nullValue());
assertThat(result.routingTable().index("test").getShards().get(1).primaryShard().state(), equalTo(RELOCATING));
assertThat(result.routingTable().index("test").getShards().get(1).primaryShard().currentNodeId(), equalTo("node1"));
assertThat(result.routingTable().index("test").getShards().get(1).primaryShard().relocatingNodeId(), equalTo("node2"));
}
Aggregations