Examples with IndexShardRoutingTable org.opensearch.cluster.routing.IndexShardRoutingTable used on opensource projects

Example 6 with IndexShardRoutingTable

use of org.opensearch.cluster.routing.IndexShardRoutingTable in project OpenSearch by opensearch-project.

the class IndexRecoveryIT method testUsesFileBasedRecoveryIfRetentionLeaseMissing.

public void testUsesFileBasedRecoveryIfRetentionLeaseMissing() 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());
            final PlainActionFuture<ReplicationResponse> future = new PlainActionFuture<>();
            primary.removeRetentionLease(ReplicationTracker.getPeerRecoveryRetentionLeaseId(replicaShardRouting), future);
            future.get();
            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));
}

Example 7 with IndexShardRoutingTable

use of org.opensearch.cluster.routing.IndexShardRoutingTable 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));
}

Example 8 with IndexShardRoutingTable

use of org.opensearch.cluster.routing.IndexShardRoutingTable in project OpenSearch by opensearch-project.

the class ReplicaToPrimaryPromotionIT method testPromoteReplicaToPrimary.

public void testPromoteReplicaToPrimary() throws Exception {
    final String indexName = randomAlphaOfLength(5).toLowerCase(Locale.ROOT);
    createIndex(indexName);
    final int numOfDocs = scaledRandomIntBetween(0, 200);
    if (numOfDocs > 0) {
        try (BackgroundIndexer indexer = new BackgroundIndexer(indexName, "_doc", client(), numOfDocs)) {
            waitForDocs(numOfDocs, indexer);
        }
        refresh(indexName);
    }
    assertHitCount(client().prepareSearch(indexName).setSize(0).get(), numOfDocs);
    ensureGreen(indexName);
    // sometimes test with a closed index
    final IndexMetadata.State indexState = randomFrom(IndexMetadata.State.OPEN, IndexMetadata.State.CLOSE);
    if (indexState == IndexMetadata.State.CLOSE) {
        CloseIndexResponse closeIndexResponse = client().admin().indices().prepareClose(indexName).get();
        assertThat("close index not acked - " + closeIndexResponse, closeIndexResponse.isAcknowledged(), equalTo(true));
        ensureGreen(indexName);
    }
    // pick up a data node that contains a random primary shard
    ClusterState state = client(internalCluster().getMasterName()).admin().cluster().prepareState().get().getState();
    final int numShards = state.metadata().index(indexName).getNumberOfShards();
    final ShardRouting primaryShard = state.routingTable().index(indexName).shard(randomIntBetween(0, numShards - 1)).primaryShard();
    final DiscoveryNode randomNode = state.nodes().resolveNode(primaryShard.currentNodeId());
    // stop the random data node, all remaining shards are promoted to primaries
    internalCluster().stopRandomNode(InternalTestCluster.nameFilter(randomNode.getName()));
    ensureYellowAndNoInitializingShards(indexName);
    state = client(internalCluster().getMasterName()).admin().cluster().prepareState().get().getState();
    for (IndexShardRoutingTable shardRoutingTable : state.routingTable().index(indexName)) {
        for (ShardRouting shardRouting : shardRoutingTable.activeShards()) {
            assertThat(shardRouting + " should be promoted as a primary", shardRouting.primary(), is(true));
        }
    }
    if (indexState == IndexMetadata.State.CLOSE) {
        assertAcked(client().admin().indices().prepareOpen(indexName));
        ensureYellowAndNoInitializingShards(indexName);
    }
    assertHitCount(client().prepareSearch(indexName).setSize(0).get(), numOfDocs);
}

Example 9 with IndexShardRoutingTable

use of org.opensearch.cluster.routing.IndexShardRoutingTable in project OpenSearch by opensearch-project.

the class CorruptedFileIT method testCorruptionOnNetworkLayer.

/**
 * Tests corruption that happens on the network layer and that the primary does not get affected by corruption that happens on the way
 * to the replica. The file on disk stays uncorrupted
 */
public void testCorruptionOnNetworkLayer() throws ExecutionException, InterruptedException {
    int numDocs = scaledRandomIntBetween(100, 1000);
    internalCluster().ensureAtLeastNumDataNodes(2);
    if (cluster().numDataNodes() < 3) {
        internalCluster().startDataOnlyNode();
    }
    NodesStatsResponse nodeStats = client().admin().cluster().prepareNodesStats().get();
    List<NodeStats> dataNodeStats = new ArrayList<>();
    for (NodeStats stat : nodeStats.getNodes()) {
        if (stat.getNode().isDataNode()) {
            dataNodeStats.add(stat);
        }
    }
    assertThat(dataNodeStats.size(), greaterThanOrEqualTo(2));
    Collections.shuffle(dataNodeStats, random());
    NodeStats primariesNode = dataNodeStats.get(0);
    NodeStats unluckyNode = dataNodeStats.get(1);
    assertAcked(prepareCreate("test").setSettings(Settings.builder().put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, "0").put(IndexMetadata.SETTING_NUMBER_OF_SHARDS, // don't go crazy here it must recovery fast
    between(1, 4)).put(MockFSIndexStore.INDEX_CHECK_INDEX_ON_CLOSE_SETTING.getKey(), false).put("index.routing.allocation.include._name", primariesNode.getNode().getName()).put(EnableAllocationDecider.INDEX_ROUTING_REBALANCE_ENABLE_SETTING.getKey(), EnableAllocationDecider.Rebalance.NONE)));
    ensureGreen();
    IndexRequestBuilder[] builders = new IndexRequestBuilder[numDocs];
    for (int i = 0; i < builders.length; i++) {
        builders[i] = client().prepareIndex("test").setSource("field", "value");
    }
    indexRandom(true, builders);
    ensureGreen();
    assertAllSuccessful(client().admin().indices().prepareFlush().setForce(true).execute().actionGet());
    // we have to flush at least once here since we don't corrupt the translog
    SearchResponse countResponse = client().prepareSearch().setSize(0).get();
    assertHitCount(countResponse, numDocs);
    final boolean truncate = randomBoolean();
    for (NodeStats dataNode : dataNodeStats) {
        MockTransportService mockTransportService = ((MockTransportService) internalCluster().getInstance(TransportService.class, dataNode.getNode().getName()));
        mockTransportService.addSendBehavior(internalCluster().getInstance(TransportService.class, unluckyNode.getNode().getName()), (connection, requestId, action, request, options) -> {
            if (action.equals(PeerRecoveryTargetService.Actions.FILE_CHUNK)) {
                FileChunkRequest req = (FileChunkRequest) request;
                if (truncate && req.length() > 1) {
                    BytesRef bytesRef = req.content().toBytesRef();
                    BytesArray array = new BytesArray(bytesRef.bytes, bytesRef.offset, (int) req.length() - 1);
                    request = new FileChunkRequest(req.recoveryId(), req.requestSeqNo(), req.shardId(), req.metadata(), req.position(), array, req.lastChunk(), req.totalTranslogOps(), req.sourceThrottleTimeInNanos());
                } else {
                    assert req.content().toBytesRef().bytes == req.content().toBytesRef().bytes : "no internal reference!!";
                    final byte[] array = req.content().toBytesRef().bytes;
                    int i = randomIntBetween(0, req.content().length() - 1);
                    // flip one byte in the content
                    array[i] = (byte) ~array[i];
                }
            }
            connection.sendRequest(requestId, action, request, options);
        });
    }
    Settings build = Settings.builder().put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, "1").put("index.routing.allocation.include._name", "*").build();
    client().admin().indices().prepareUpdateSettings("test").setSettings(build).get();
    client().admin().cluster().prepareReroute().get();
    ClusterHealthResponse actionGet = client().admin().cluster().health(Requests.clusterHealthRequest("test").waitForGreenStatus()).actionGet();
    if (actionGet.isTimedOut()) {
        logger.info("ensureGreen timed out, cluster state:\n{}\n{}", client().admin().cluster().prepareState().get().getState(), client().admin().cluster().preparePendingClusterTasks().get());
        assertThat("timed out waiting for green state", actionGet.isTimedOut(), equalTo(false));
    }
    // we are green so primaries got not corrupted.
    // ensure that no shard is actually allocated on the unlucky node
    ClusterStateResponse clusterStateResponse = client().admin().cluster().prepareState().get();
    for (IndexShardRoutingTable table : clusterStateResponse.getState().getRoutingTable().index("test")) {
        for (ShardRouting routing : table) {
            if (unluckyNode.getNode().getId().equals(routing.currentNodeId())) {
                assertThat(routing.state(), not(equalTo(ShardRoutingState.STARTED)));
                assertThat(routing.state(), not(equalTo(ShardRoutingState.RELOCATING)));
            }
        }
    }
    final int numIterations = scaledRandomIntBetween(5, 20);
    for (int i = 0; i < numIterations; i++) {
        SearchResponse response = client().prepareSearch().setSize(numDocs).get();
        assertHitCount(response, numDocs);
    }
}

Example 10 with IndexShardRoutingTable

use of org.opensearch.cluster.routing.IndexShardRoutingTable in project OpenSearch by opensearch-project.

the class GeoShapeIntegrationIT method findNodeName.

private String findNodeName(String index) {
    ClusterState state = client().admin().cluster().prepareState().get().getState();
    IndexShardRoutingTable shard = state.getRoutingTable().index(index).shard(0);
    String nodeId = shard.assignedShards().get(0).currentNodeId();
    return state.getNodes().get(nodeId).getName();
}