Examples with TransportRequest org.elasticsearch.transport.TransportRequest used on opensource projects

Example 1 with TransportRequest

use of org.elasticsearch.transport.TransportRequest in project elasticsearch by elastic.

the class IndexWithShadowReplicasIT method testPrimaryRelocationWhereRecoveryFails.

public void testPrimaryRelocationWhereRecoveryFails() throws Exception {
    Path dataPath = createTempDir();
    Settings nodeSettings = Settings.builder().put("node.add_lock_id_to_custom_path", false).put(Environment.PATH_SHARED_DATA_SETTING.getKey(), dataPath).build();
    String node1 = internalCluster().startNode(nodeSettings);
    final String IDX = "test";
    Settings idxSettings = Settings.builder().put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, 1).put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 1).put(IndexMetaData.SETTING_DATA_PATH, dataPath.toAbsolutePath().toString()).put(IndexMetaData.SETTING_SHADOW_REPLICAS, true).put(IndexMetaData.SETTING_SHARED_FILESYSTEM, true).build();
    prepareCreate(IDX).setSettings(idxSettings).addMapping("doc", "foo", "type=text").get();
    // Node1 has the primary, now node2 has the replica
    String node2 = internalCluster().startNode(nodeSettings);
    ensureGreen(IDX);
    flushAndRefresh(IDX);
    String node3 = internalCluster().startNode(nodeSettings);
    final AtomicInteger counter = new AtomicInteger(0);
    final CountDownLatch started = new CountDownLatch(1);
    final int numPhase1Docs = scaledRandomIntBetween(25, 200);
    final int numPhase2Docs = scaledRandomIntBetween(25, 200);
    final int numPhase3Docs = scaledRandomIntBetween(25, 200);
    final CountDownLatch phase1finished = new CountDownLatch(1);
    final CountDownLatch phase2finished = new CountDownLatch(1);
    final CountDownLatch phase3finished = new CountDownLatch(1);
    final AtomicBoolean keepFailing = new AtomicBoolean(true);
    MockTransportService mockTransportService = ((MockTransportService) internalCluster().getInstance(TransportService.class, node1));
    mockTransportService.addDelegate(internalCluster().getInstance(TransportService.class, node3), new MockTransportService.DelegateTransport(mockTransportService.original()) {

        @Override
        protected void sendRequest(Connection connection, long requestId, String action, TransportRequest request, TransportRequestOptions options) throws IOException {
            if (keepFailing.get() && action.equals(PeerRecoveryTargetService.Actions.TRANSLOG_OPS)) {
                logger.info("--> failing translog ops");
                throw new ElasticsearchException("failing on purpose");
            }
            super.sendRequest(connection, requestId, action, request, options);
        }
    });
    Thread thread = new Thread() {

        @Override
        public void run() {
            started.countDown();
            while (counter.get() < (numPhase1Docs + numPhase2Docs + numPhase3Docs)) {
                final IndexResponse indexResponse = client().prepareIndex(IDX, "doc", Integer.toString(counter.incrementAndGet())).setSource("foo", "bar").get();
                assertEquals(DocWriteResponse.Result.CREATED, indexResponse.getResult());
                final int docCount = counter.get();
                if (docCount == numPhase1Docs) {
                    phase1finished.countDown();
                } else if (docCount == (numPhase1Docs + numPhase2Docs)) {
                    phase2finished.countDown();
                }
            }
            logger.info("--> stopping indexing thread");
            phase3finished.countDown();
        }
    };
    thread.start();
    started.await();
    // wait for a certain number of documents to be indexed
    phase1finished.await();
    logger.info("--> excluding {} from allocation", node1);
    // now prevent primary from being allocated on node 1 move to node_3
    Settings build = Settings.builder().put("index.routing.allocation.exclude._name", node1).build();
    client().admin().indices().prepareUpdateSettings(IDX).setSettings(build).execute().actionGet();
    // wait for more documents to be indexed post-recovery, also waits for
    // indexing thread to stop
    phase2finished.await();
    // stop failing
    keepFailing.set(false);
    // wait for more docs to be indexed
    phase3finished.await();
    ensureGreen(IDX);
    thread.join();
    logger.info("--> performing query");
    flushAndRefresh();
    SearchResponse resp = client().prepareSearch(IDX).setQuery(matchAllQuery()).get();
    assertHitCount(resp, counter.get());
}

Example 2 with TransportRequest

use of org.elasticsearch.transport.TransportRequest in project elasticsearch by elastic.

the class ExceptionRetryIT method testRetryDueToExceptionOnNetworkLayer.

/**
     * Tests retry mechanism when indexing. If an exception occurs when indexing then the indexing request is tried again before finally
     * failing. If auto generated ids are used this must not lead to duplicate ids
     * see https://github.com/elastic/elasticsearch/issues/8788
     */
public void testRetryDueToExceptionOnNetworkLayer() throws ExecutionException, InterruptedException, IOException {
    final AtomicBoolean exceptionThrown = new AtomicBoolean(false);
    int numDocs = scaledRandomIntBetween(100, 1000);
    Client client = internalCluster().coordOnlyNodeClient();
    NodesStatsResponse nodeStats = client().admin().cluster().prepareNodesStats().get();
    NodeStats unluckyNode = randomFrom(nodeStats.getNodes().stream().filter((s) -> s.getNode().isDataNode()).collect(Collectors.toList()));
    assertAcked(client().admin().indices().prepareCreate("index").setSettings(Settings.builder().put("index.number_of_replicas", 1).put("index.number_of_shards", 5)));
    ensureGreen("index");
    logger.info("unlucky node: {}", unluckyNode.getNode());
    //create a transport service that throws a ConnectTransportException for one bulk request and therefore triggers a retry.
    for (NodeStats dataNode : nodeStats.getNodes()) {
        MockTransportService mockTransportService = ((MockTransportService) internalCluster().getInstance(TransportService.class, dataNode.getNode().getName()));
        mockTransportService.addDelegate(internalCluster().getInstance(TransportService.class, unluckyNode.getNode().getName()), new MockTransportService.DelegateTransport(mockTransportService.original()) {

            @Override
            protected void sendRequest(Connection connection, long requestId, String action, TransportRequest request, TransportRequestOptions options) throws IOException {
                super.sendRequest(connection, requestId, action, request, options);
                if (action.equals(TransportShardBulkAction.ACTION_NAME) && exceptionThrown.compareAndSet(false, true)) {
                    logger.debug("Throw ConnectTransportException");
                    throw new ConnectTransportException(connection.getNode(), action);
                }
            }
        });
    }
    BulkRequestBuilder bulkBuilder = client.prepareBulk();
    for (int i = 0; i < numDocs; i++) {
        XContentBuilder doc = null;
        doc = jsonBuilder().startObject().field("foo", "bar").endObject();
        bulkBuilder.add(client.prepareIndex("index", "type").setSource(doc));
    }
    BulkResponse response = bulkBuilder.get();
    if (response.hasFailures()) {
        for (BulkItemResponse singleIndexRespons : response.getItems()) {
            if (singleIndexRespons.isFailed()) {
                fail("None of the bulk items should fail but got " + singleIndexRespons.getFailureMessage());
            }
        }
    }
    refresh();
    SearchResponse searchResponse = client().prepareSearch("index").setSize(numDocs * 2).addStoredField("_id").get();
    Set<String> uniqueIds = new HashSet();
    long dupCounter = 0;
    boolean found_duplicate_already = false;
    for (int i = 0; i < searchResponse.getHits().getHits().length; i++) {
        if (!uniqueIds.add(searchResponse.getHits().getHits()[i].getId())) {
            if (!found_duplicate_already) {
                SearchResponse dupIdResponse = client().prepareSearch("index").setQuery(termQuery("_id", searchResponse.getHits().getHits()[i].getId())).setExplain(true).get();
                assertThat(dupIdResponse.getHits().getTotalHits(), greaterThan(1L));
                logger.info("found a duplicate id:");
                for (SearchHit hit : dupIdResponse.getHits()) {
                    logger.info("Doc {} was found on shard {}", hit.getId(), hit.getShard().getShardId());
                }
                logger.info("will not print anymore in case more duplicates are found.");
                found_duplicate_already = true;
            }
            dupCounter++;
        }
    }
    assertSearchResponse(searchResponse);
    assertThat(dupCounter, equalTo(0L));
    assertHitCount(searchResponse, numDocs);
    IndicesStatsResponse index = client().admin().indices().prepareStats("index").clear().setSegments(true).get();
    IndexStats indexStats = index.getIndex("index");
    long maxUnsafeAutoIdTimestamp = Long.MIN_VALUE;
    for (IndexShardStats indexShardStats : indexStats) {
        for (ShardStats shardStats : indexShardStats) {
            SegmentsStats segments = shardStats.getStats().getSegments();
            maxUnsafeAutoIdTimestamp = Math.max(maxUnsafeAutoIdTimestamp, segments.getMaxUnsafeAutoIdTimestamp());
        }
    }
    assertTrue("exception must have been thrown otherwise setup is broken", exceptionThrown.get());
    assertTrue("maxUnsafeAutoIdTimestamp must be > than 0 we have at least one retry", maxUnsafeAutoIdTimestamp > -1);
}

Example 3 with TransportRequest

use of org.elasticsearch.transport.TransportRequest in project elasticsearch by elastic.

the class RemoteClusterConnection method getConnection.

/**
     * Returns a connection to the remote cluster. This connection might be a proxy connection that redirects internally to the
     * given node.
     */
Transport.Connection getConnection(DiscoveryNode remoteClusterNode) {
    DiscoveryNode discoveryNode = nodeSupplier.get();
    Transport.Connection connection = transportService.getConnection(discoveryNode);
    return new Transport.Connection() {

        @Override
        public DiscoveryNode getNode() {
            return remoteClusterNode;
        }

        @Override
        public void sendRequest(long requestId, String action, TransportRequest request, TransportRequestOptions options) throws IOException, TransportException {
            connection.sendRequest(requestId, TransportActionProxy.getProxyAction(action), TransportActionProxy.wrapRequest(remoteClusterNode, request), options);
        }

        @Override
        public void close() throws IOException {
            assert false : "proxy connections must not be closed";
        }
    };
}

Example 4 with TransportRequest

use of org.elasticsearch.transport.TransportRequest in project elasticsearch by elastic.

the class IndexRecoveryIT method testDisconnectsDuringRecovery.

/**
     * Tests scenario where recovery target successfully sends recovery request to source but then the channel gets closed while
     * the source is working on the recovery process.
     */
@TestLogging("_root:DEBUG,org.elasticsearch.indices.recovery:TRACE")
public void testDisconnectsDuringRecovery() throws Exception {
    boolean primaryRelocation = randomBoolean();
    final String indexName = "test";
    final Settings nodeSettings = Settings.builder().put(RecoverySettings.INDICES_RECOVERY_RETRY_DELAY_NETWORK_SETTING.getKey(), TimeValue.timeValueMillis(randomIntBetween(0, 100))).build();
    TimeValue disconnectAfterDelay = TimeValue.timeValueMillis(randomIntBetween(0, 100));
    // start a master node
    String masterNodeName = internalCluster().startMasterOnlyNode(nodeSettings);
    final String blueNodeName = internalCluster().startNode(Settings.builder().put("node.attr.color", "blue").put(nodeSettings).build());
    final String redNodeName = internalCluster().startNode(Settings.builder().put("node.attr.color", "red").put(nodeSettings).build());
    client().admin().indices().prepareCreate(indexName).setSettings(Settings.builder().put(IndexMetaData.INDEX_ROUTING_INCLUDE_GROUP_SETTING.getKey() + "color", "blue").put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, 1).put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 0)).get();
    List<IndexRequestBuilder> requests = new ArrayList<>();
    int numDocs = scaledRandomIntBetween(25, 250);
    for (int i = 0; i < numDocs; i++) {
        requests.add(client().prepareIndex(indexName, "type").setSource("{}", XContentType.JSON));
    }
    indexRandom(true, requests);
    ensureSearchable(indexName);
    assertHitCount(client().prepareSearch(indexName).get(), numDocs);
    MockTransportService masterTransportService = (MockTransportService) internalCluster().getInstance(TransportService.class, masterNodeName);
    MockTransportService blueMockTransportService = (MockTransportService) internalCluster().getInstance(TransportService.class, blueNodeName);
    MockTransportService redMockTransportService = (MockTransportService) internalCluster().getInstance(TransportService.class, redNodeName);
    redMockTransportService.addDelegate(blueMockTransportService, new MockTransportService.DelegateTransport(redMockTransportService.original()) {

        private final AtomicInteger count = new AtomicInteger();

        @Override
        protected void sendRequest(Connection connection, long requestId, String action, TransportRequest request, TransportRequestOptions options) throws IOException {
            logger.info("--> sending request {} on {}", action, connection.getNode());
            if (PeerRecoverySourceService.Actions.START_RECOVERY.equals(action) && count.incrementAndGet() == 1) {
                // ensures that it's considered as valid recovery attempt by source
                try {
                    awaitBusy(() -> client(blueNodeName).admin().cluster().prepareState().setLocal(true).get().getState().getRoutingTable().index("test").shard(0).getAllInitializingShards().isEmpty() == false);
                } catch (InterruptedException e) {
                    throw new RuntimeException(e);
                }
                super.sendRequest(connection, requestId, action, request, options);
                try {
                    Thread.sleep(disconnectAfterDelay.millis());
                } catch (InterruptedException e) {
                    throw new RuntimeException(e);
                }
                throw new ConnectTransportException(connection.getNode(), "DISCONNECT: simulation disconnect after successfully sending " + action + " request");
            } else {
                super.sendRequest(connection, requestId, action, request, options);
            }
        }
    });
    final AtomicBoolean finalized = new AtomicBoolean();
    blueMockTransportService.addDelegate(redMockTransportService, new MockTransportService.DelegateTransport(blueMockTransportService.original()) {

        @Override
        protected void sendRequest(Connection connection, long requestId, String action, TransportRequest request, TransportRequestOptions options) throws IOException {
            logger.info("--> sending request {} on {}", action, connection.getNode());
            if (action.equals(PeerRecoveryTargetService.Actions.FINALIZE)) {
                finalized.set(true);
            }
            super.sendRequest(connection, requestId, action, request, options);
        }
    });
    for (MockTransportService mockTransportService : Arrays.asList(redMockTransportService, blueMockTransportService)) {
        mockTransportService.addDelegate(masterTransportService, new MockTransportService.DelegateTransport(mockTransportService.original()) {

            @Override
            protected void sendRequest(Connection connection, long requestId, String action, TransportRequest request, TransportRequestOptions options) throws IOException {
                logger.info("--> sending request {} on {}", action, connection.getNode());
                if ((primaryRelocation && finalized.get()) == false) {
                    assertNotEquals(action, ShardStateAction.SHARD_FAILED_ACTION_NAME);
                }
                super.sendRequest(connection, requestId, action, request, options);
            }
        });
    }
    if (primaryRelocation) {
        logger.info("--> starting primary relocation recovery from blue to red");
        client().admin().indices().prepareUpdateSettings(indexName).setSettings(Settings.builder().put(IndexMetaData.INDEX_ROUTING_INCLUDE_GROUP_SETTING.getKey() + "color", "red")).get();
        // also waits for relocation / recovery to complete
        ensureGreen();
        // if a primary relocation fails after the source shard has been marked as relocated, both source and target are failed. If the
        // source shard is moved back to started because the target fails first, it's possible that there is a cluster state where the
        // shard is marked as started again (and ensureGreen returns), but while applying the cluster state the primary is failed and
        // will be reallocated. The cluster will thus become green, then red, then green again. Triggering a refresh here before
        // searching helps, as in contrast to search actions, refresh waits for the closed shard to be reallocated.
        client().admin().indices().prepareRefresh(indexName).get();
    } else {
        logger.info("--> starting replica recovery from blue to red");
        client().admin().indices().prepareUpdateSettings(indexName).setSettings(Settings.builder().put(IndexMetaData.INDEX_ROUTING_INCLUDE_GROUP_SETTING.getKey() + "color", "red,blue").put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 1)).get();
        ensureGreen();
    }
    for (int i = 0; i < 10; i++) {
        assertHitCount(client().prepareSearch(indexName).get(), numDocs);
    }
}

Example 5 with TransportRequest

use of org.elasticsearch.transport.TransportRequest in project elasticsearch by elastic.

the class TruncatedRecoveryIT method testCancelRecoveryAndResume.

/**
     * This test tries to truncate some of larger files in the index to trigger leftovers on the recovery
     * target. This happens during recovery when the last chunk of the file is transferred to the replica
     * we just throw an exception to make sure the recovery fails and we leave some half baked files on the target.
     * Later we allow full recovery to ensure we can still recover and don't run into corruptions.
     */
public void testCancelRecoveryAndResume() throws Exception {
    assertTrue(client().admin().cluster().prepareUpdateSettings().setTransientSettings(Settings.builder().put(CHUNK_SIZE_SETTING.getKey(), new ByteSizeValue(randomIntBetween(50, 300), ByteSizeUnit.BYTES))).get().isAcknowledged());
    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());
    // we use 2 nodes a lucky and unlucky one
    // the lucky one holds the primary
    // the unlucky one gets the replica and the truncated leftovers
    NodeStats primariesNode = dataNodeStats.get(0);
    NodeStats unluckyNode = dataNodeStats.get(1);
    // create the index and prevent allocation on any other nodes than the lucky one
    // we have no replicas so far and make sure that we allocate the primary on the lucky node
    assertAcked(prepareCreate("test").addMapping("type1", "field1", "type=text", "the_id", "type=text").setSettings(Settings.builder().put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 0).put(IndexMetaData.SETTING_NUMBER_OF_SHARDS, numberOfShards()).put("index.routing.allocation.include._name", // only allocate on the lucky node
    primariesNode.getNode().getName())));
    // index some docs and check if they are coming back
    int numDocs = randomIntBetween(100, 200);
    List<IndexRequestBuilder> builder = new ArrayList<>();
    for (int i = 0; i < numDocs; i++) {
        String id = Integer.toString(i);
        builder.add(client().prepareIndex("test", "type1", id).setSource("field1", English.intToEnglish(i), "the_id", id));
    }
    indexRandom(true, builder);
    for (int i = 0; i < numDocs; i++) {
        String id = Integer.toString(i);
        assertHitCount(client().prepareSearch().setQuery(QueryBuilders.termQuery("the_id", id)).get(), 1);
    }
    ensureGreen();
    // ensure we have flushed segments and make them a big one via optimize
    client().admin().indices().prepareFlush().setForce(true).get();
    client().admin().indices().prepareForceMerge().setMaxNumSegments(1).setFlush(true).get();
    final CountDownLatch latch = new CountDownLatch(1);
    final AtomicBoolean truncate = new AtomicBoolean(true);
    for (NodeStats dataNode : dataNodeStats) {
        MockTransportService mockTransportService = ((MockTransportService) internalCluster().getInstance(TransportService.class, dataNode.getNode().getName()));
        mockTransportService.addDelegate(internalCluster().getInstance(TransportService.class, unluckyNode.getNode().getName()), new MockTransportService.DelegateTransport(mockTransportService.original()) {

            @Override
            protected void sendRequest(Connection connection, long requestId, String action, TransportRequest request, TransportRequestOptions options) throws IOException {
                if (action.equals(PeerRecoveryTargetService.Actions.FILE_CHUNK)) {
                    RecoveryFileChunkRequest req = (RecoveryFileChunkRequest) request;
                    logger.debug("file chunk [{}] lastChunk: {}", req, req.lastChunk());
                    if ((req.name().endsWith("cfs") || req.name().endsWith("fdt")) && req.lastChunk() && truncate.get()) {
                        latch.countDown();
                        throw new RuntimeException("Caused some truncated files for fun and profit");
                    }
                }
                super.sendRequest(connection, requestId, action, request, options);
            }
        });
    }
    //
    logger.info("--> bumping replicas to 1");
    client().admin().indices().prepareUpdateSettings("test").setSettings(Settings.builder().put(IndexMetaData.SETTING_NUMBER_OF_REPLICAS, 1).put(// now allow allocation on all nodes
    "index.routing.allocation.include._name", primariesNode.getNode().getName() + "," + unluckyNode.getNode().getName())).get();
    latch.await();
    // at this point we got some truncated left overs on the replica on the unlucky node
    // now we are allowing the recovery to allocate again and finish to see if we wipe the truncated files
    truncate.compareAndSet(true, false);
    ensureGreen("test");
    for (int i = 0; i < numDocs; i++) {
        String id = Integer.toString(i);
        assertHitCount(client().prepareSearch().setQuery(QueryBuilders.termQuery("the_id", id)).get(), 1);
    }
}