Search in sources :

Example 1 with DuplicateKeyException

use of io.crate.exceptions.DuplicateKeyException in project crate by crate.

the class ClusterDisruptionIT method testAckedIndexing.

/**
 * Test that we do not loose document whose indexing request was successful, under a randomly selected disruption scheme
 * We also collect & report the type of indexing failures that occur.
 * <p>
 * This test is a superset of tests run in the Jepsen test suite, with the exception of versioned updates
 */
@TestLogging("_root:DEBUG,org.elasticsearch.action.bulk:TRACE,org.elasticsearch.action.get:TRACE," + "org.elasticsearch.discovery:TRACE,org.elasticsearch.action.support.replication:TRACE," + "org.elasticsearch.cluster.service:TRACE,org.elasticsearch.indices.recovery:TRACE," + "org.elasticsearch.indices.cluster:TRACE,org.elasticsearch.index.shard:TRACE")
@Test
public void testAckedIndexing() throws Exception {
    final List<String> nodes = startCluster(3);
    int numberOfShards = 1 + randomInt(2);
    int replicas = randomInt(2);
    logger.info("creating table t clustered into {} shards with {} replicas", numberOfShards, replicas);
    execute("create table t (id int primary key, x string) clustered into " + numberOfShards + " shards " + "with (number_of_replicas = " + replicas + ", \"write.wait_for_active_shards\" = 1, \"global_checkpoint_sync.interval\"='1s')");
    ensureGreen();
    ServiceDisruptionScheme disruptionScheme = addRandomDisruptionScheme();
    logger.info("disruption scheme [{}] added", disruptionScheme);
    // id -> node sent.
    final ConcurrentHashMap<String, String> ackedDocs = new ConcurrentHashMap<>();
    final AtomicBoolean stop = new AtomicBoolean(false);
    List<Thread> indexers = new ArrayList<>(nodes.size());
    List<Semaphore> semaphores = new ArrayList<>(nodes.size());
    final AtomicInteger idGenerator = new AtomicInteger(0);
    final AtomicReference<CountDownLatch> countDownLatchRef = new AtomicReference<>();
    final List<Exception> exceptedExceptions = new CopyOnWriteArrayList<>();
    logger.info("starting indexers");
    try {
        for (final String node : nodes) {
            final Semaphore semaphore = new Semaphore(0);
            semaphores.add(semaphore);
            final String name = "indexer_" + indexers.size();
            Thread thread = new Thread(() -> {
                while (!stop.get()) {
                    String id = null;
                    try {
                        if (!semaphore.tryAcquire(10, TimeUnit.SECONDS)) {
                            continue;
                        }
                        logger.info("[{}] Acquired semaphore and it has {} permits left", name, semaphore.availablePermits());
                        try {
                            id = String.valueOf(idGenerator.incrementAndGet());
                            int shard = Math.floorMod(Murmur3HashFunction.hash(id), numberOfShards);
                            logger.trace("[{}] indexing id [{}] through node [{}] targeting shard [{}]", name, id, node, shard);
                            execute("insert into t (id, x) values (?, ?)", new Object[] { id, randomInt(100) }, node, TimeValue.timeValueSeconds(1L));
                            ackedDocs.put(id, node);
                            logger.trace("[{}] indexed id [{}] through node [{}], response [{}]", name, id, node, response);
                        } catch (ElasticsearchException | DuplicateKeyException e) {
                            exceptedExceptions.add(e);
                            final String rowId = id;
                            logger.trace(() -> new ParameterizedMessage("[{}] failed id [{}] through node [{}]", name, rowId, node), e);
                        } finally {
                            countDownLatchRef.get().countDown();
                            logger.trace("[{}] decreased counter : {}", name, countDownLatchRef.get().getCount());
                        }
                    } catch (InterruptedException e) {
                    // fine - semaphore interrupt
                    } catch (AssertionError | Exception e) {
                        logger.trace(() -> new ParameterizedMessage("unexpected exception in background thread of [{}]", node), e);
                    }
                }
            });
            thread.setName(name);
            thread.start();
            indexers.add(thread);
        }
        int docsPerIndexer = randomInt(3);
        logger.info("indexing {} docs per indexer before partition", docsPerIndexer);
        countDownLatchRef.set(new CountDownLatch(docsPerIndexer * indexers.size()));
        for (Semaphore semaphore : semaphores) {
            semaphore.release(docsPerIndexer);
        }
        assertTrue(countDownLatchRef.get().await(1, TimeUnit.MINUTES));
        for (int iter = 1 + randomInt(1); iter > 0; iter--) {
            logger.info("starting disruptions & indexing (iteration [{}])", iter);
            disruptionScheme.startDisrupting();
            docsPerIndexer = randomIntBetween(1, 4);
            logger.info("indexing {} docs per indexer during partition", docsPerIndexer);
            countDownLatchRef.set(new CountDownLatch(docsPerIndexer * indexers.size()));
            Collections.shuffle(semaphores, random());
            for (Semaphore semaphore : semaphores) {
                assertThat(semaphore.availablePermits(), equalTo(0));
                semaphore.release(docsPerIndexer);
            }
            logger.info("waiting for indexing requests to complete");
            assertThat("indexing requests must complete", countDownLatchRef.get().await(20, TimeUnit.SECONDS), is(true));
            logger.info("stopping disruption");
            disruptionScheme.stopDisrupting();
            for (String node : internalCluster().getNodeNames()) {
                ensureStableCluster(nodes.size(), TimeValue.timeValueMillis(disruptionScheme.expectedTimeToHeal().millis() + DISRUPTION_HEALING_OVERHEAD.millis()), true, node);
            }
            // is the super-connected node and recovery source and target are on opposite sides of the bridge
            if (disruptionScheme instanceof NetworkDisruption && ((NetworkDisruption) disruptionScheme).getDisruptedLinks() instanceof Bridge) {
                logger.warn("retrying failed allocations in case of a bridge partition");
                execute("ALTER CLUSTER REROUTE RETRY FAILED");
            }
            ensureGreen();
            logger.info("validating successful docs");
            assertBusy(() -> {
                for (String node : nodes) {
                    try {
                        logger.debug("validating through node [{}] ([{}] acked docs)", node, ackedDocs.size());
                        for (String id : ackedDocs.keySet()) {
                            execute("select * from t where id = ?", new Object[] { id }, node);
                            assertThat("doc [" + id + "] indexed via node [" + ackedDocs.get(id) + "] not found", response.rowCount(), is(1L));
                        }
                    } catch (AssertionError | NoShardAvailableActionException e) {
                        throw new AssertionError(e.getMessage() + " (checked via node [" + node + "]", e);
                    }
                }
            }, 30, TimeUnit.SECONDS);
            logger.info("done validating (iteration [{}])", iter);
        }
    } finally {
        logger.info("shutting down indexers");
        stop.set(true);
        for (Thread indexer : indexers) {
            indexer.interrupt();
            indexer.join(60000);
        }
        if (exceptedExceptions.size() > 0) {
            StringBuilder sb = new StringBuilder();
            for (Exception e : exceptedExceptions) {
                sb.append("\n").append(e.getMessage());
            }
            logger.debug("Indexing exceptions during disruption: {}", sb);
        }
    }
}
Also used : ArrayList(java.util.ArrayList) CopyOnWriteArrayList(java.util.concurrent.CopyOnWriteArrayList) ServiceDisruptionScheme(org.elasticsearch.test.disruption.ServiceDisruptionScheme) Semaphore(java.util.concurrent.Semaphore) ElasticsearchException(org.elasticsearch.ElasticsearchException) ConcurrentHashMap(java.util.concurrent.ConcurrentHashMap) NetworkDisruption(org.elasticsearch.test.disruption.NetworkDisruption) AtomicReference(java.util.concurrent.atomic.AtomicReference) CountDownLatch(java.util.concurrent.CountDownLatch) ElasticsearchException(org.elasticsearch.ElasticsearchException) CorruptIndexException(org.apache.lucene.index.CorruptIndexException) DuplicateKeyException(io.crate.exceptions.DuplicateKeyException) NoShardAvailableActionException(org.elasticsearch.action.NoShardAvailableActionException) DuplicateKeyException(io.crate.exceptions.DuplicateKeyException) AtomicBoolean(java.util.concurrent.atomic.AtomicBoolean) NoShardAvailableActionException(org.elasticsearch.action.NoShardAvailableActionException) AtomicInteger(java.util.concurrent.atomic.AtomicInteger) ParameterizedMessage(org.apache.logging.log4j.message.ParameterizedMessage) Bridge(org.elasticsearch.test.disruption.NetworkDisruption.Bridge) CopyOnWriteArrayList(java.util.concurrent.CopyOnWriteArrayList) TestLogging(org.elasticsearch.test.junit.annotations.TestLogging) Test(org.junit.Test)

Aggregations

DuplicateKeyException (io.crate.exceptions.DuplicateKeyException)1 ArrayList (java.util.ArrayList)1 ConcurrentHashMap (java.util.concurrent.ConcurrentHashMap)1 CopyOnWriteArrayList (java.util.concurrent.CopyOnWriteArrayList)1 CountDownLatch (java.util.concurrent.CountDownLatch)1 Semaphore (java.util.concurrent.Semaphore)1 AtomicBoolean (java.util.concurrent.atomic.AtomicBoolean)1 AtomicInteger (java.util.concurrent.atomic.AtomicInteger)1 AtomicReference (java.util.concurrent.atomic.AtomicReference)1 ParameterizedMessage (org.apache.logging.log4j.message.ParameterizedMessage)1 CorruptIndexException (org.apache.lucene.index.CorruptIndexException)1 ElasticsearchException (org.elasticsearch.ElasticsearchException)1 NoShardAvailableActionException (org.elasticsearch.action.NoShardAvailableActionException)1 NetworkDisruption (org.elasticsearch.test.disruption.NetworkDisruption)1 Bridge (org.elasticsearch.test.disruption.NetworkDisruption.Bridge)1 ServiceDisruptionScheme (org.elasticsearch.test.disruption.ServiceDisruptionScheme)1 TestLogging (org.elasticsearch.test.junit.annotations.TestLogging)1 Test (org.junit.Test)1