Examples with RecoveryState org.opensearch.indices.recovery.RecoveryState used on opensource projects

Example 6 with RecoveryState

use of org.opensearch.indices.recovery.RecoveryState in project OpenSearch by opensearch-project.

the class RemoveCorruptedShardDataCommandIT method testCorruptTranslogTruncationOfReplica.

public void testCorruptTranslogTruncationOfReplica() throws Exception {
    internalCluster().startClusterManagerOnlyNode();
    final String node1 = internalCluster().startDataOnlyNode();
    final String node2 = internalCluster().startDataOnlyNode();
    logger.info("--> nodes name: {}, {}", node1, node2);
    final String indexName = "test";
    assertAcked(prepareCreate(indexName).setSettings(Settings.builder().put(IndexMetadata.SETTING_NUMBER_OF_SHARDS, 1).put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 1).put(IndexSettings.INDEX_REFRESH_INTERVAL_SETTING.getKey(), "-1").put(MockEngineSupport.DISABLE_FLUSH_ON_CLOSE.getKey(), // never flush - always recover from translog
    true).put("index.routing.allocation.exclude._name", node2)));
    ensureYellow();
    assertAcked(client().admin().indices().prepareUpdateSettings(indexName).setSettings(Settings.builder().put("index.routing.allocation.exclude._name", (String) null)));
    ensureGreen();
    // Index some documents
    int numDocsToKeep = randomIntBetween(0, 100);
    logger.info("--> indexing [{}] docs to be kept", numDocsToKeep);
    IndexRequestBuilder[] builders = new IndexRequestBuilder[numDocsToKeep];
    for (int i = 0; i < builders.length; i++) {
        builders[i] = client().prepareIndex(indexName).setSource("foo", "bar");
    }
    indexRandom(false, false, false, Arrays.asList(builders));
    flush(indexName);
    disableTranslogFlush(indexName);
    // having no extra docs is an interesting case for seq no based recoveries - test it more often
    int numDocsToTruncate = randomBoolean() ? 0 : randomIntBetween(0, 100);
    logger.info("--> indexing [{}] more docs to be truncated", numDocsToTruncate);
    builders = new IndexRequestBuilder[numDocsToTruncate];
    for (int i = 0; i < builders.length; i++) {
        builders[i] = client().prepareIndex(indexName).setSource("foo", "bar");
    }
    indexRandom(false, false, false, Arrays.asList(builders));
    final int totalDocs = numDocsToKeep + numDocsToTruncate;
    // sample the replica node translog dirs
    final ShardId shardId = new ShardId(resolveIndex(indexName), 0);
    final Path translogDir = getPathToShardData(node2, shardId, ShardPath.TRANSLOG_FOLDER_NAME);
    final Settings node1PathSettings = internalCluster().dataPathSettings(node1);
    final Settings node2PathSettings = internalCluster().dataPathSettings(node2);
    assertBusy(() -> internalCluster().getInstances(GatewayMetaState.class).forEach(gw -> assertTrue(gw.allPendingAsyncStatesWritten())));
    // stop data nodes
    internalCluster().stopRandomDataNode();
    internalCluster().stopRandomDataNode();
    // Corrupt the translog file(s) on the replica
    logger.info("--> corrupting translog");
    TestTranslog.corruptRandomTranslogFile(logger, random(), translogDir);
    // Start the node with the non-corrupted data path
    logger.info("--> starting node");
    internalCluster().startNode(node1PathSettings);
    ensureYellow();
    // Run a search and make sure it succeeds
    assertHitCount(client().prepareSearch(indexName).setQuery(matchAllQuery()).get(), totalDocs);
    // check replica corruption
    final RemoveCorruptedShardDataCommand command = new RemoveCorruptedShardDataCommand();
    final MockTerminal terminal = new MockTerminal();
    final OptionParser parser = command.getParser();
    final Environment environment = TestEnvironment.newEnvironment(Settings.builder().put(internalCluster().getDefaultSettings()).put(node2PathSettings).build());
    terminal.addTextInput("y");
    OptionSet options = parser.parse("-d", translogDir.toAbsolutePath().toString());
    logger.info("--> running command for [{}]", translogDir.toAbsolutePath());
    command.execute(terminal, options, environment);
    logger.info("--> output:\n{}", terminal.getOutput());
    logger.info("--> starting the replica node to test recovery");
    internalCluster().startNode(node2PathSettings);
    ensureGreen(indexName);
    for (String node : internalCluster().nodesInclude(indexName)) {
        assertHitCount(client().prepareSearch(indexName).setPreference("_only_nodes:" + node).setQuery(matchAllQuery()).get(), totalDocs);
    }
    final RecoveryResponse recoveryResponse = client().admin().indices().prepareRecoveries(indexName).setActiveOnly(false).get();
    final RecoveryState replicaRecoveryState = recoveryResponse.shardRecoveryStates().get(indexName).stream().filter(recoveryState -> recoveryState.getPrimary() == false).findFirst().get();
    // the replica translog was disabled so it doesn't know what hte global checkpoint is and thus can't do ops based recovery
    assertThat(replicaRecoveryState.getIndex().toString(), replicaRecoveryState.getIndex().recoveredFileCount(), greaterThan(0));
    // Ensure that the global checkpoint and local checkpoint are restored from the max seqno of the last commit.
    final SeqNoStats seqNoStats = getSeqNoStats(indexName, 0);
    assertThat(seqNoStats.getGlobalCheckpoint(), equalTo(seqNoStats.getMaxSeqNo()));
    assertThat(seqNoStats.getLocalCheckpoint(), equalTo(seqNoStats.getMaxSeqNo()));
}

Example 7 with RecoveryState

use of org.opensearch.indices.recovery.RecoveryState in project OpenSearch by opensearch-project.

the class CloseIndexIT method testRecoverExistingReplica.

/**
 * Ensures that if a replica of a closed index does not have the same content as the primary, then a file-based recovery will occur.
 */
public void testRecoverExistingReplica() throws Exception {
    final String indexName = "test-recover-existing-replica";
    internalCluster().ensureAtLeastNumDataNodes(2);
    List<String> dataNodes = randomSubsetOf(2, Sets.newHashSet(clusterService().state().nodes().getDataNodes().valuesIt()).stream().map(DiscoveryNode::getName).collect(Collectors.toSet()));
    createIndex(indexName, Settings.builder().put(IndexMetadata.SETTING_NUMBER_OF_SHARDS, 1).put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 1).put("index.routing.allocation.include._name", String.join(",", dataNodes)).build());
    indexRandom(randomBoolean(), randomBoolean(), randomBoolean(), IntStream.range(0, randomIntBetween(0, 50)).mapToObj(n -> client().prepareIndex(indexName).setSource("num", n)).collect(toList()));
    ensureGreen(indexName);
    client().admin().indices().prepareFlush(indexName).get();
    // index more documents while one shard copy is offline
    internalCluster().restartNode(dataNodes.get(1), new InternalTestCluster.RestartCallback() {

        @Override
        public Settings onNodeStopped(String nodeName) throws Exception {
            Client client = client(dataNodes.get(0));
            int moreDocs = randomIntBetween(1, 50);
            for (int i = 0; i < moreDocs; i++) {
                client.prepareIndex(indexName).setSource("num", i).get();
            }
            assertAcked(client.admin().indices().prepareClose(indexName));
            return super.onNodeStopped(nodeName);
        }
    });
    assertIndexIsClosed(indexName);
    ensureGreen(indexName);
    internalCluster().assertSameDocIdsOnShards();
    for (RecoveryState recovery : client().admin().indices().prepareRecoveries(indexName).get().shardRecoveryStates().get(indexName)) {
        if (recovery.getPrimary() == false) {
            assertThat(recovery.getIndex().fileDetails(), not(empty()));
        }
    }
}

Example 8 with RecoveryState

use of org.opensearch.indices.recovery.RecoveryState in project OpenSearch by opensearch-project.

the class AbortedRestoreIT method testAbortedRestoreAlsoAbortFileRestores.

public void testAbortedRestoreAlsoAbortFileRestores() throws Exception {
    internalCluster().startClusterManagerOnlyNode();
    final String dataNode = internalCluster().startDataOnlyNode();
    final String indexName = "test-abort-restore";
    createIndex(indexName, indexSettingsNoReplicas(1).build());
    indexRandomDocs(indexName, scaledRandomIntBetween(10, 1_000));
    ensureGreen();
    forceMerge();
    final String repositoryName = "repository";
    createRepository(repositoryName, "mock");
    final String snapshotName = "snapshot";
    createFullSnapshot(repositoryName, snapshotName);
    assertAcked(client().admin().indices().prepareDelete(indexName));
    logger.info("--> blocking all data nodes for repository [{}]", repositoryName);
    blockAllDataNodes(repositoryName);
    failReadsAllDataNodes(repositoryName);
    logger.info("--> starting restore");
    final ActionFuture<RestoreSnapshotResponse> future = client().admin().cluster().prepareRestoreSnapshot(repositoryName, snapshotName).setWaitForCompletion(true).setIndices(indexName).execute();
    assertBusy(() -> {
        final RecoveryResponse recoveries = client().admin().indices().prepareRecoveries(indexName).setIndicesOptions(IndicesOptions.LENIENT_EXPAND_OPEN).setActiveOnly(true).get();
        assertThat(recoveries.hasRecoveries(), is(true));
        final List<RecoveryState> shardRecoveries = recoveries.shardRecoveryStates().get(indexName);
        assertThat(shardRecoveries, hasSize(1));
        assertThat(future.isDone(), is(false));
        for (RecoveryState shardRecovery : shardRecoveries) {
            assertThat(shardRecovery.getRecoverySource().getType(), equalTo(RecoverySource.Type.SNAPSHOT));
            assertThat(shardRecovery.getStage(), equalTo(RecoveryState.Stage.INDEX));
        }
    });
    final ThreadPool.Info snapshotThreadPoolInfo = threadPool(dataNode).info(ThreadPool.Names.SNAPSHOT);
    assertThat(snapshotThreadPoolInfo.getMax(), greaterThan(0));
    logger.info("--> waiting for snapshot thread [max={}] pool to be full", snapshotThreadPoolInfo.getMax());
    waitForMaxActiveSnapshotThreads(dataNode, equalTo(snapshotThreadPoolInfo.getMax()));
    logger.info("--> aborting restore by deleting the index");
    assertAcked(client().admin().indices().prepareDelete(indexName));
    logger.info("--> unblocking repository [{}]", repositoryName);
    unblockAllDataNodes(repositoryName);
    logger.info("--> restore should have failed");
    final RestoreSnapshotResponse restoreSnapshotResponse = future.get();
    assertThat(restoreSnapshotResponse.getRestoreInfo().failedShards(), equalTo(1));
    assertThat(restoreSnapshotResponse.getRestoreInfo().successfulShards(), equalTo(0));
    logger.info("--> waiting for snapshot thread pool to be empty");
    waitForMaxActiveSnapshotThreads(dataNode, equalTo(0));
}

Example 9 with RecoveryState

use of org.opensearch.indices.recovery.RecoveryState in project OpenSearch by opensearch-project.

the class IndexShard method startRecovery.

public void startRecovery(RecoveryState recoveryState, PeerRecoveryTargetService recoveryTargetService, RecoveryListener recoveryListener, RepositoriesService repositoriesService, Consumer<MappingMetadata> mappingUpdateConsumer, IndicesService indicesService) {
    // }
    assert recoveryState.getRecoverySource().equals(shardRouting.recoverySource());
    switch(recoveryState.getRecoverySource().getType()) {
        case EMPTY_STORE:
        case EXISTING_STORE:
            executeRecovery("from store", recoveryState, recoveryListener, this::recoverFromStore);
            break;
        case PEER:
            try {
                markAsRecovering("from " + recoveryState.getSourceNode(), recoveryState);
                recoveryTargetService.startRecovery(this, recoveryState.getSourceNode(), recoveryListener);
            } catch (Exception e) {
                failShard("corrupted preexisting index", e);
                recoveryListener.onFailure(recoveryState, new RecoveryFailedException(recoveryState, null, e), true);
            }
            break;
        case SNAPSHOT:
            final String repo = ((SnapshotRecoverySource) recoveryState.getRecoverySource()).snapshot().getRepository();
            executeRecovery("from snapshot", recoveryState, recoveryListener, l -> restoreFromRepository(repositoriesService.repository(repo), l));
            break;
        case LOCAL_SHARDS:
            final IndexMetadata indexMetadata = indexSettings().getIndexMetadata();
            final Index resizeSourceIndex = indexMetadata.getResizeSourceIndex();
            final List<IndexShard> startedShards = new ArrayList<>();
            final IndexService sourceIndexService = indicesService.indexService(resizeSourceIndex);
            final Set<ShardId> requiredShards;
            final int numShards;
            if (sourceIndexService != null) {
                requiredShards = IndexMetadata.selectRecoverFromShards(shardId().id(), sourceIndexService.getMetadata(), indexMetadata.getNumberOfShards());
                for (IndexShard shard : sourceIndexService) {
                    if (shard.state() == IndexShardState.STARTED && requiredShards.contains(shard.shardId())) {
                        startedShards.add(shard);
                    }
                }
                numShards = requiredShards.size();
            } else {
                numShards = -1;
                requiredShards = Collections.emptySet();
            }
            if (numShards == startedShards.size()) {
                assert requiredShards.isEmpty() == false;
                executeRecovery("from local shards", recoveryState, recoveryListener, l -> recoverFromLocalShards(mappingUpdateConsumer, startedShards.stream().filter((s) -> requiredShards.contains(s.shardId())).collect(Collectors.toList()), l));
            } else {
                final RuntimeException e;
                if (numShards == -1) {
                    e = new IndexNotFoundException(resizeSourceIndex);
                } else {
                    e = new IllegalStateException("not all required shards of index " + resizeSourceIndex + " are started yet, expected " + numShards + " found " + startedShards.size() + " can't recover shard " + shardId());
                }
                throw e;
            }
            break;
        default:
            throw new IllegalArgumentException("Unknown recovery source " + recoveryState.getRecoverySource());
    }
}

Example 10 with RecoveryState

use of org.opensearch.indices.recovery.RecoveryState in project OpenSearch by opensearch-project.

the class IndexShard method storeStats.

public StoreStats storeStats() {
    try {
        final RecoveryState recoveryState = this.recoveryState;
        final long bytesStillToRecover = recoveryState == null ? -1L : recoveryState.getIndex().bytesStillToRecover();
        return store.stats(bytesStillToRecover == -1 ? StoreStats.UNKNOWN_RESERVED_BYTES : bytesStillToRecover);
    } catch (IOException e) {
        failShard("Failing shard because of exception during storeStats", e);
        throw new OpenSearchException("io exception while building 'store stats'", e);
    }
}