use of org.opensearch.cluster.service.ClusterService in project OpenSearch by opensearch-project.
the class TransportAddVotingConfigExclusionsActionTests method testExcludeByNodeNameSucceedsEvenIfAllExclusionsAlreadyAdded.
public void testExcludeByNodeNameSucceedsEvenIfAllExclusionsAlreadyAdded() throws InterruptedException {
final ClusterState state = clusterService.state();
final ClusterState.Builder builder = builder(state);
builder.metadata(Metadata.builder(state.metadata()).coordinationMetadata(CoordinationMetadata.builder(state.coordinationMetadata()).addVotingConfigExclusion(otherNode1Exclusion).build()));
setState(clusterService, builder);
final CountDownLatch countDownLatch = new CountDownLatch(1);
transportService.sendRequest(localNode, AddVotingConfigExclusionsAction.NAME, new AddVotingConfigExclusionsRequest("other1"), expectSuccess(r -> {
assertNotNull(r);
countDownLatch.countDown();
}));
assertTrue(countDownLatch.await(30, TimeUnit.SECONDS));
assertThat(clusterService.getClusterApplierService().state().getVotingConfigExclusions(), contains(otherNode1Exclusion));
}
use of org.opensearch.cluster.service.ClusterService in project OpenSearch by opensearch-project.
the class TransportClearVotingConfigExclusionsActionTests method setupForTest.
@Before
public void setupForTest() {
final MockTransport transport = new MockTransport();
transportService = transport.createTransportService(Settings.EMPTY, threadPool, TransportService.NOOP_TRANSPORT_INTERCEPTOR, boundTransportAddress -> localNode, null, emptySet());
new TransportClearVotingConfigExclusionsAction(transportService, clusterService, threadPool, new ActionFilters(emptySet()), new IndexNameExpressionResolver(new ThreadContext(Settings.EMPTY)));
// registers action
transportService.start();
transportService.acceptIncomingRequests();
final ClusterState.Builder builder = builder(new ClusterName("cluster")).nodes(new Builder().add(localNode).add(otherNode1).add(otherNode2).localNodeId(localNode.getId()).masterNodeId(localNode.getId()));
builder.metadata(Metadata.builder().coordinationMetadata(CoordinationMetadata.builder().addVotingConfigExclusion(otherNode1Exclusion).addVotingConfigExclusion(otherNode2Exclusion).build()));
setState(clusterService, builder);
}
use of org.opensearch.cluster.service.ClusterService in project OpenSearch by opensearch-project.
the class IndexRecoveryIT method testUsesFileBasedRecoveryIfRetentionLeaseAheadOfGlobalCheckpoint.
public void testUsesFileBasedRecoveryIfRetentionLeaseAheadOfGlobalCheckpoint() throws Exception {
internalCluster().ensureAtLeastNumDataNodes(2);
String indexName = "test-index";
createIndex(indexName, Settings.builder().put(IndexMetadata.SETTING_NUMBER_OF_SHARDS, 1).put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 1).put(IndexSettings.INDEX_SOFT_DELETES_SETTING.getKey(), true).put(UnassignedInfo.INDEX_DELAYED_NODE_LEFT_TIMEOUT_SETTING.getKey(), "12h").build());
indexRandom(randomBoolean(), randomBoolean(), randomBoolean(), IntStream.range(0, between(0, 100)).mapToObj(n -> client().prepareIndex(indexName).setSource("num", n)).collect(toList()));
ensureGreen(indexName);
final ShardId shardId = new ShardId(resolveIndex(indexName), 0);
final DiscoveryNodes discoveryNodes = clusterService().state().nodes();
final IndexShardRoutingTable indexShardRoutingTable = clusterService().state().routingTable().shardRoutingTable(shardId);
final IndexShard primary = internalCluster().getInstance(IndicesService.class, discoveryNodes.get(indexShardRoutingTable.primaryShard().currentNodeId()).getName()).getShardOrNull(shardId);
final ShardRouting replicaShardRouting = indexShardRoutingTable.replicaShards().get(0);
internalCluster().restartNode(discoveryNodes.get(replicaShardRouting.currentNodeId()).getName(), new InternalTestCluster.RestartCallback() {
@Override
public Settings onNodeStopped(String nodeName) throws Exception {
assertFalse(client().admin().cluster().prepareHealth().setWaitForNodes(Integer.toString(discoveryNodes.getSize() - 1)).setWaitForEvents(Priority.LANGUID).get().isTimedOut());
indexRandom(randomBoolean(), randomBoolean(), randomBoolean(), IntStream.range(0, between(1, 100)).mapToObj(n -> client().prepareIndex(indexName).setSource("num", n)).collect(toList()));
// We do not guarantee that the replica can recover locally all the way to its own global checkpoint before starting
// to recover from the primary, so we must be careful not to perform an operations-based recovery if this would require
// some operations that are not being retained. Emulate this by advancing the lease ahead of the replica's GCP:
primary.renewRetentionLease(ReplicationTracker.getPeerRecoveryRetentionLeaseId(replicaShardRouting), primary.seqNoStats().getMaxSeqNo() + 1, ReplicationTracker.PEER_RECOVERY_RETENTION_LEASE_SOURCE);
return super.onNodeStopped(nodeName);
}
});
ensureGreen(indexName);
// noinspection OptionalGetWithoutIsPresent because it fails the test if absent
final RecoveryState recoveryState = client().admin().indices().prepareRecoveries(indexName).get().shardRecoveryStates().get(indexName).stream().filter(rs -> rs.getPrimary() == false).findFirst().get();
assertThat(recoveryState.getIndex().totalFileCount(), greaterThan(0));
}
use of org.opensearch.cluster.service.ClusterService in project OpenSearch by opensearch-project.
the class IndexShardIT method testLockTryingToDelete.
public void testLockTryingToDelete() throws Exception {
createIndex("test");
ensureGreen();
NodeEnvironment env = getInstanceFromNode(NodeEnvironment.class);
ClusterService cs = getInstanceFromNode(ClusterService.class);
final Index index = cs.state().metadata().index("test").getIndex();
Path[] shardPaths = env.availableShardPaths(new ShardId(index, 0));
logger.info("--> paths: [{}]", (Object) shardPaths);
// Should not be able to acquire the lock because it's already open
try {
NodeEnvironment.acquireFSLockForPaths(IndexSettingsModule.newIndexSettings("test", Settings.EMPTY), shardPaths);
fail("should not have been able to acquire the lock");
} catch (LockObtainFailedException e) {
assertTrue("msg: " + e.getMessage(), e.getMessage().contains("unable to acquire write.lock"));
}
// Test without the regular shard lock to assume we can acquire it
// (worst case, meaning that the shard lock could be acquired and
// we're green to delete the shard's directory)
ShardLock sLock = new DummyShardLock(new ShardId(index, 0));
try {
env.deleteShardDirectoryUnderLock(sLock, IndexSettingsModule.newIndexSettings("test", Settings.EMPTY));
fail("should not have been able to delete the directory");
} catch (LockObtainFailedException e) {
assertTrue("msg: " + e.getMessage(), e.getMessage().contains("unable to acquire write.lock"));
}
}
use of org.opensearch.cluster.service.ClusterService in project OpenSearch by opensearch-project.
the class IndexRecoveryIT method testUsesFileBasedRecoveryIfOperationsBasedRecoveryWouldBeUnreasonable.
public void testUsesFileBasedRecoveryIfOperationsBasedRecoveryWouldBeUnreasonable() throws Exception {
internalCluster().ensureAtLeastNumDataNodes(2);
String indexName = "test-index";
final Settings.Builder settings = Settings.builder().put(IndexMetadata.SETTING_NUMBER_OF_SHARDS, 1).put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 1).put(IndexSettings.INDEX_SOFT_DELETES_SETTING.getKey(), true).put(UnassignedInfo.INDEX_DELAYED_NODE_LEFT_TIMEOUT_SETTING.getKey(), "12h").put(IndexService.RETENTION_LEASE_SYNC_INTERVAL_SETTING.getKey(), "100ms");
final double reasonableOperationsBasedRecoveryProportion;
if (randomBoolean()) {
reasonableOperationsBasedRecoveryProportion = randomDoubleBetween(0.05, 0.99, true);
settings.put(IndexSettings.FILE_BASED_RECOVERY_THRESHOLD_SETTING.getKey(), reasonableOperationsBasedRecoveryProportion);
} else {
reasonableOperationsBasedRecoveryProportion = IndexSettings.FILE_BASED_RECOVERY_THRESHOLD_SETTING.get(Settings.EMPTY);
}
logger.info("--> performing ops-based recoveries up to [{}%] of docs", reasonableOperationsBasedRecoveryProportion * 100.0);
createIndex(indexName, settings.build());
indexRandom(randomBoolean(), false, randomBoolean(), IntStream.range(0, between(0, 100)).mapToObj(n -> client().prepareIndex(indexName).setSource("num", n)).collect(toList()));
ensureGreen(indexName);
flush(indexName);
// wait for all history to be discarded
assertBusy(() -> {
for (ShardStats shardStats : client().admin().indices().prepareStats(indexName).get().getShards()) {
final long maxSeqNo = shardStats.getSeqNoStats().getMaxSeqNo();
assertTrue(shardStats.getRetentionLeaseStats().retentionLeases() + " should discard history up to " + maxSeqNo, shardStats.getRetentionLeaseStats().retentionLeases().leases().stream().allMatch(l -> l.retainingSequenceNumber() == maxSeqNo + 1));
}
});
// ensure that all operations are in the safe commit
flush(indexName);
final ShardStats shardStats = client().admin().indices().prepareStats(indexName).get().getShards()[0];
final long docCount = shardStats.getStats().docs.getCount();
assertThat(shardStats.getStats().docs.getDeleted(), equalTo(0L));
assertThat(shardStats.getSeqNoStats().getMaxSeqNo() + 1, equalTo(docCount));
final ShardId shardId = new ShardId(resolveIndex(indexName), 0);
final DiscoveryNodes discoveryNodes = clusterService().state().nodes();
final IndexShardRoutingTable indexShardRoutingTable = clusterService().state().routingTable().shardRoutingTable(shardId);
final ShardRouting replicaShardRouting = indexShardRoutingTable.replicaShards().get(0);
assertTrue("should have lease for " + replicaShardRouting, client().admin().indices().prepareStats(indexName).get().getShards()[0].getRetentionLeaseStats().retentionLeases().contains(ReplicationTracker.getPeerRecoveryRetentionLeaseId(replicaShardRouting)));
internalCluster().restartNode(discoveryNodes.get(replicaShardRouting.currentNodeId()).getName(), new InternalTestCluster.RestartCallback() {
@Override
public Settings onNodeStopped(String nodeName) throws Exception {
assertFalse(client().admin().cluster().prepareHealth().setWaitForNodes(Integer.toString(discoveryNodes.getSize() - 1)).setWaitForEvents(Priority.LANGUID).get().isTimedOut());
final int newDocCount = Math.toIntExact(Math.round(Math.ceil((1 + Math.ceil(docCount * reasonableOperationsBasedRecoveryProportion)) / (1 - reasonableOperationsBasedRecoveryProportion))));
/*
* newDocCount >= (ceil(docCount * p) + 1) / (1-p)
*
* ==> 0 <= newDocCount * (1-p) - ceil(docCount * p) - 1
* = newDocCount - (newDocCount * p + ceil(docCount * p) + 1)
* < newDocCount - (ceil(newDocCount * p) + ceil(docCount * p))
* <= newDocCount - ceil(newDocCount * p + docCount * p)
*
* ==> docCount < newDocCount + docCount - ceil((newDocCount + docCount) * p)
* == localCheckpoint + 1 - ceil((newDocCount + docCount) * p)
* == firstReasonableSeqNo
*
* The replica has docCount docs, i.e. has operations with seqnos [0..docCount-1], so a seqno-based recovery will start
* from docCount < firstReasonableSeqNo
*
* ==> it is unreasonable to recover the replica using a seqno-based recovery
*/
indexRandom(randomBoolean(), randomBoolean(), randomBoolean(), IntStream.range(0, newDocCount).mapToObj(n -> client().prepareIndex(indexName).setSource("num", n)).collect(toList()));
flush(indexName);
assertBusy(() -> assertFalse("should no longer have lease for " + replicaShardRouting, client().admin().indices().prepareStats(indexName).get().getShards()[0].getRetentionLeaseStats().retentionLeases().contains(ReplicationTracker.getPeerRecoveryRetentionLeaseId(replicaShardRouting))));
return super.onNodeStopped(nodeName);
}
});
ensureGreen(indexName);
// noinspection OptionalGetWithoutIsPresent because it fails the test if absent
final RecoveryState recoveryState = client().admin().indices().prepareRecoveries(indexName).get().shardRecoveryStates().get(indexName).stream().filter(rs -> rs.getPrimary() == false).findFirst().get();
assertThat(recoveryState.getIndex().totalFileCount(), greaterThan(0));
}
Aggregations