Examples with ShardAllocationDecision org.opensearch.cluster.routing.allocation.ShardAllocationDecision used on opensource projects

Example 1 with ShardAllocationDecision

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

the class AllocationIdIT method checkNoValidShardCopy.

private void checkNoValidShardCopy(String indexName, ShardId shardId) throws Exception {
    assertBusy(() -> {
        final ClusterAllocationExplanation explanation = client().admin().cluster().prepareAllocationExplain().setIndex(indexName).setShard(shardId.id()).setPrimary(true).get().getExplanation();
        final ShardAllocationDecision shardAllocationDecision = explanation.getShardAllocationDecision();
        assertThat(shardAllocationDecision.isDecisionTaken(), equalTo(true));
        assertThat(shardAllocationDecision.getAllocateDecision().getAllocationDecision(), equalTo(AllocationDecision.NO_VALID_SHARD_COPY));
    });
}

Example 2 with ShardAllocationDecision

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

the class BalancedShardsAllocator method decideShardAllocation.

@Override
public ShardAllocationDecision decideShardAllocation(final ShardRouting shard, final RoutingAllocation allocation) {
    Balancer balancer = new Balancer(logger, allocation, movePrimaryFirst, weightFunction, threshold);
    AllocateUnassignedDecision allocateUnassignedDecision = AllocateUnassignedDecision.NOT_TAKEN;
    MoveDecision moveDecision = MoveDecision.NOT_TAKEN;
    if (shard.unassigned()) {
        allocateUnassignedDecision = balancer.decideAllocateUnassigned(shard);
    } else {
        moveDecision = balancer.decideMove(shard);
        if (moveDecision.isDecisionTaken() && moveDecision.canRemain()) {
            MoveDecision rebalanceDecision = balancer.decideRebalance(shard);
            moveDecision = rebalanceDecision.withRemainDecision(moveDecision.getCanRemainDecision());
        }
    }
    return new ShardAllocationDecision(allocateUnassignedDecision, moveDecision);
}

Example 3 with ShardAllocationDecision

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

the class ClusterAllocationExplanationTests method randomClusterAllocationExplanation.

private static ClusterAllocationExplanation randomClusterAllocationExplanation(boolean assignedShard) {
    ShardRouting shardRouting = TestShardRouting.newShardRouting(new ShardId(new Index("idx", "123"), 0), assignedShard ? "node-0" : null, true, assignedShard ? ShardRoutingState.STARTED : ShardRoutingState.UNASSIGNED);
    DiscoveryNode node = assignedShard ? new DiscoveryNode("node-0", buildNewFakeTransportAddress(), emptyMap(), emptySet(), Version.CURRENT) : null;
    ShardAllocationDecision shardAllocationDecision;
    if (assignedShard) {
        MoveDecision moveDecision = MoveDecision.cannotRebalance(Decision.YES, AllocationDecision.NO, 3, null).withRemainDecision(Decision.YES);
        shardAllocationDecision = new ShardAllocationDecision(AllocateUnassignedDecision.NOT_TAKEN, moveDecision);
    } else {
        AllocateUnassignedDecision allocateDecision = AllocateUnassignedDecision.no(UnassignedInfo.AllocationStatus.DECIDERS_NO, null);
        shardAllocationDecision = new ShardAllocationDecision(allocateDecision, MoveDecision.NOT_TAKEN);
    }
    return new ClusterAllocationExplanation(shardRouting, node, null, null, shardAllocationDecision);
}

Example 4 with ShardAllocationDecision

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

the class RemoveCorruptedShardDataCommandIT method testCorruptTranslogTruncation.

public void testCorruptTranslogTruncation() throws Exception {
    internalCluster().startNodes(2);
    final String node1 = internalCluster().getNodeNames()[0];
    final String node2 = internalCluster().getNodeNames()[1];
    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().putNull("index.routing.allocation.exclude._name")));
    ensureGreen();
    // Index some documents
    int numDocsToKeep = randomIntBetween(10, 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 doc 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));
    RemoveCorruptedShardDataCommand command = new RemoveCorruptedShardDataCommand();
    MockTerminal terminal = new MockTerminal();
    OptionParser parser = command.getParser();
    if (randomBoolean() && numDocsToTruncate > 0) {
        // flush the replica, so it will have more docs than what the primary will have
        Index index = resolveIndex(indexName);
        IndexShard replica = internalCluster().getInstance(IndicesService.class, node2).getShardOrNull(new ShardId(index, 0));
        replica.flush(new FlushRequest());
        logger.info("--> performed extra flushing on replica");
    }
    final Settings node1PathSettings = internalCluster().dataPathSettings(node1);
    final Settings node2PathSettings = internalCluster().dataPathSettings(node2);
    // shut down the replica node to be tested later
    internalCluster().stopRandomNode(InternalTestCluster.nameFilter(node2));
    final Path translogDir = getPathToShardData(indexName, ShardPath.TRANSLOG_FOLDER_NAME);
    final Path indexDir = getPathToShardData(indexName, ShardPath.INDEX_FOLDER_NAME);
    // Restart the single node
    logger.info("--> restarting node");
    internalCluster().restartRandomDataNode(new InternalTestCluster.RestartCallback() {

        @Override
        public Settings onNodeStopped(String nodeName) throws Exception {
            logger.info("--> corrupting translog on node {}", nodeName);
            TestTranslog.corruptRandomTranslogFile(logger, random(), translogDir);
            return super.onNodeStopped(nodeName);
        }
    });
    // all shards should be failed due to a corrupted translog
    assertBusy(() -> {
        final UnassignedInfo unassignedInfo = client().admin().cluster().prepareAllocationExplain().setIndex(indexName).setShard(0).setPrimary(true).get().getExplanation().getUnassignedInfo();
        assertThat(unassignedInfo.getReason(), equalTo(UnassignedInfo.Reason.ALLOCATION_FAILED));
        assertThat(ExceptionsHelper.unwrap(unassignedInfo.getFailure(), TranslogCorruptedException.class), not(nullValue()));
    });
    // have to shut down primary node - otherwise node lock is present
    internalCluster().restartNode(node1, new InternalTestCluster.RestartCallback() {

        @Override
        public Settings onNodeStopped(String nodeName) throws Exception {
            assertBusy(() -> {
                logger.info("--> checking that lock has been released for {}", indexDir);
                // noinspection EmptyTryBlock since we're just trying to obtain the lock
                try (Directory dir = FSDirectory.open(indexDir, NativeFSLockFactory.INSTANCE);
                    Lock ignored = dir.obtainLock(IndexWriter.WRITE_LOCK_NAME)) {
                } catch (LockObtainFailedException lofe) {
                    logger.info("--> failed acquiring lock for {}", indexDir);
                    throw new AssertionError("still waiting for lock release at [" + indexDir + "]", lofe);
                } catch (IOException ioe) {
                    throw new AssertionError("unexpected IOException [" + indexDir + "]", ioe);
                }
            });
            final Environment environment = TestEnvironment.newEnvironment(Settings.builder().put(internalCluster().getDefaultSettings()).put(node1PathSettings).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());
            return super.onNodeStopped(nodeName);
        }
    });
    String primaryNodeId = null;
    final ClusterState state = client().admin().cluster().prepareState().get().getState();
    final DiscoveryNodes nodes = state.nodes();
    for (ObjectObjectCursor<String, DiscoveryNode> cursor : nodes.getNodes()) {
        final String name = cursor.value.getName();
        if (name.equals(node1)) {
            primaryNodeId = cursor.key;
            break;
        }
    }
    assertThat(primaryNodeId, notNullValue());
    assertThat(terminal.getOutput(), containsString("allocate_stale_primary"));
    assertThat(terminal.getOutput(), containsString("\"node\" : \"" + primaryNodeId + "\""));
    // there is only _stale_ primary (due to new allocation id)
    assertBusy(() -> {
        final ClusterAllocationExplanation explanation = client().admin().cluster().prepareAllocationExplain().setIndex(indexName).setShard(0).setPrimary(true).get().getExplanation();
        final ShardAllocationDecision shardAllocationDecision = explanation.getShardAllocationDecision();
        assertThat(shardAllocationDecision.isDecisionTaken(), equalTo(true));
        assertThat(shardAllocationDecision.getAllocateDecision().getAllocationDecision(), equalTo(AllocationDecision.NO_VALID_SHARD_COPY));
    });
    client().admin().cluster().prepareReroute().add(new AllocateStalePrimaryAllocationCommand(indexName, 0, primaryNodeId, true)).get();
    assertBusy(() -> {
        final ClusterAllocationExplanation explanation = client().admin().cluster().prepareAllocationExplain().setIndex(indexName).setShard(0).setPrimary(true).get().getExplanation();
        assertThat(explanation.getCurrentNode(), notNullValue());
        assertThat(explanation.getShardState(), equalTo(ShardRoutingState.STARTED));
    });
    ensureYellow(indexName);
    // Run a search and make sure it succeeds
    assertHitCount(client().prepareSearch(indexName).setQuery(matchAllQuery()).get(), numDocsToKeep);
    logger.info("--> starting the replica node to test recovery");
    internalCluster().startNode(node2PathSettings);
    ensureGreen(indexName);
    for (String node : internalCluster().nodesInclude(indexName)) {
        SearchRequestBuilder q = client().prepareSearch(indexName).setPreference("_only_nodes:" + node).setQuery(matchAllQuery());
        assertHitCount(q.get(), numDocsToKeep);
    }
    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();
    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 5 with ShardAllocationDecision

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

the class RemoveCorruptedShardDataCommandIT method testCorruptIndex.

public void testCorruptIndex() throws Exception {
    final String node = internalCluster().startNode();
    final String indexName = "index42";
    assertAcked(prepareCreate(indexName).setSettings(Settings.builder().put(IndexMetadata.SETTING_NUMBER_OF_SHARDS, 1).put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 0).put(MergePolicyConfig.INDEX_MERGE_ENABLED, false).put(IndexSettings.INDEX_REFRESH_INTERVAL_SETTING.getKey(), "-1").put(MockEngineSupport.DISABLE_FLUSH_ON_CLOSE.getKey(), true).put(IndexSettings.INDEX_CHECK_ON_STARTUP.getKey(), "checksum")));
    // index some docs in several segments
    int numDocs = 0;
    for (int k = 0, attempts = randomIntBetween(5, 10); k < attempts; k++) {
        final int numExtraDocs = between(10, 100);
        IndexRequestBuilder[] builders = new IndexRequestBuilder[numExtraDocs];
        for (int i = 0; i < builders.length; i++) {
            builders[i] = client().prepareIndex(indexName).setSource("foo", "bar");
        }
        numDocs += numExtraDocs;
        indexRandom(false, false, false, Arrays.asList(builders));
        flush(indexName);
    }
    logger.info("--> indexed {} docs", numDocs);
    final RemoveCorruptedShardDataCommand command = new RemoveCorruptedShardDataCommand();
    final MockTerminal terminal = new MockTerminal();
    final OptionParser parser = command.getParser();
    final Settings nodePathSettings = internalCluster().dataPathSettings(node);
    final Environment environment = TestEnvironment.newEnvironment(Settings.builder().put(internalCluster().getDefaultSettings()).put(nodePathSettings).build());
    final OptionSet options = parser.parse("-index", indexName, "-shard-id", "0");
    // Try running it before the node is stopped (and shard is closed)
    try {
        command.execute(terminal, options, environment);
        fail("expected the command to fail as node is locked");
    } catch (Exception e) {
        assertThat(e.getMessage(), allOf(containsString("failed to lock node's directory"), containsString("is OpenSearch still running?")));
    }
    final Path indexDir = getPathToShardData(indexName, ShardPath.INDEX_FOLDER_NAME);
    internalCluster().restartNode(node, new InternalTestCluster.RestartCallback() {

        @Override
        public Settings onNodeStopped(String nodeName) throws Exception {
            // Try running it before the shard is corrupted, it should flip out because there is no corruption file marker
            try {
                command.execute(terminal, options, environment);
                fail("expected the command to fail as there is no corruption file marker");
            } catch (Exception e) {
                assertThat(e.getMessage(), startsWith("Shard does not seem to be corrupted at"));
            }
            CorruptionUtils.corruptIndex(random(), indexDir, false);
            return super.onNodeStopped(nodeName);
        }
    });
    // shard should be failed due to a corrupted index
    assertBusy(() -> {
        final ClusterAllocationExplanation explanation = client().admin().cluster().prepareAllocationExplain().setIndex(indexName).setShard(0).setPrimary(true).get().getExplanation();
        final ShardAllocationDecision shardAllocationDecision = explanation.getShardAllocationDecision();
        assertThat(shardAllocationDecision.isDecisionTaken(), equalTo(true));
        assertThat(shardAllocationDecision.getAllocateDecision().getAllocationDecision(), equalTo(AllocationDecision.NO_VALID_SHARD_COPY));
    });
    internalCluster().restartNode(node, new InternalTestCluster.RestartCallback() {

        @Override
        public Settings onNodeStopped(String nodeName) throws Exception {
            terminal.addTextInput("y");
            command.execute(terminal, options, environment);
            return super.onNodeStopped(nodeName);
        }
    });
    waitNoPendingTasksOnAll();
    String nodeId = null;
    final ClusterState state = client().admin().cluster().prepareState().get().getState();
    final DiscoveryNodes nodes = state.nodes();
    for (ObjectObjectCursor<String, DiscoveryNode> cursor : nodes.getNodes()) {
        final String name = cursor.value.getName();
        if (name.equals(node)) {
            nodeId = cursor.key;
            break;
        }
    }
    assertThat(nodeId, notNullValue());
    logger.info("--> output:\n{}", terminal.getOutput());
    assertThat(terminal.getOutput(), containsString("allocate_stale_primary"));
    assertThat(terminal.getOutput(), containsString("\"node\" : \"" + nodeId + "\""));
    // there is only _stale_ primary (due to new allocation id)
    assertBusy(() -> {
        final ClusterAllocationExplanation explanation = client().admin().cluster().prepareAllocationExplain().setIndex(indexName).setShard(0).setPrimary(true).get().getExplanation();
        final ShardAllocationDecision shardAllocationDecision = explanation.getShardAllocationDecision();
        assertThat(shardAllocationDecision.isDecisionTaken(), equalTo(true));
        assertThat(shardAllocationDecision.getAllocateDecision().getAllocationDecision(), equalTo(AllocationDecision.NO_VALID_SHARD_COPY));
    });
    client().admin().cluster().prepareReroute().add(new AllocateStalePrimaryAllocationCommand(indexName, 0, nodeId, true)).get();
    assertBusy(() -> {
        final ClusterAllocationExplanation explanation = client().admin().cluster().prepareAllocationExplain().setIndex(indexName).setShard(0).setPrimary(true).get().getExplanation();
        assertThat(explanation.getCurrentNode(), notNullValue());
        assertThat(explanation.getShardState(), equalTo(ShardRoutingState.STARTED));
    });
    final Pattern pattern = Pattern.compile("Corrupted Lucene index segments found -\\s+(?<docs>\\d+) documents will be lost.");
    final Matcher matcher = pattern.matcher(terminal.getOutput());
    assertThat(matcher.find(), equalTo(true));
    final int expectedNumDocs = numDocs - Integer.parseInt(matcher.group("docs"));
    ensureGreen(indexName);
    assertHitCount(client().prepareSearch(indexName).setQuery(matchAllQuery()).get(), expectedNumDocs);
}