Search in sources :

Example 1 with TransactionAware

use of org.apache.tephra.TransactionAware in project cdap by caskdata.

the class DatasetBasedStreamSizeScheduleStoreTest method testDeletion.

private void testDeletion(final ProgramId programId) throws Exception {
    final boolean defaultVersion = programId.getVersion().equals(ApplicationId.DEFAULT_VERSION);
    DatasetId storeTable = NamespaceId.SYSTEM.dataset(ScheduleStoreTableUtil.SCHEDULE_STORE_DATASET_NAME);
    final Table table = datasetFramework.getDataset(storeTable, ImmutableMap.<String, String>of(), null);
    Assert.assertNotNull(table);
    TransactionExecutor txnl = txExecutorFactory.createExecutor(ImmutableList.of((TransactionAware) table));
    final byte[] startKey = Bytes.toBytes(DatasetBasedStreamSizeScheduleStore.KEY_PREFIX);
    final byte[] stopKey = Bytes.stopKeyForPrefix(startKey);
    txnl.execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            Scanner scanner = table.scan(startKey, stopKey);
            Assert.assertNull(scanner.next());
            scanner.close();
        }
    });
    // Create one stream schedule - this will be persisted with new format
    scheduleStore.persist(programId, PROGRAM_TYPE, STREAM_SCHEDULE_1, MAP_1, 0L, 0L, 0L, 0L, true);
    // Create one stream schedule - based on the old format
    txnl.execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            // Create a programId without version so that we can create a old format schedule
            ProgramId defaultProgramId = new ProgramId(programId.getNamespace(), programId.getApplication(), programId.getType(), programId.getProgram());
            String newRowKey = scheduleStore.getRowKey(defaultProgramId, PROGRAM_TYPE, STREAM_SCHEDULE_1.getName());
            Row row = table.get(Bytes.toBytes(scheduleStore.getRowKey(programId, PROGRAM_TYPE, STREAM_SCHEDULE_1.getName())));
            Assert.assertFalse(row.isEmpty());
            byte[] oldRowKey = Bytes.toBytes(scheduleStore.removeAppVersion(newRowKey));
            for (Map.Entry<byte[], byte[]> entry : row.getColumns().entrySet()) {
                table.put(oldRowKey, entry.getKey(), entry.getValue());
            }
        }
    });
    // Make sure there are only two stream size schedules
    txnl.execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            Scanner scanner = table.scan(startKey, stopKey);
            int numRows = 0;
            while (true) {
                Row row = scanner.next();
                if (row == null) {
                    break;
                }
                numRows++;
            }
            scanner.close();
            Assert.assertEquals(2, numRows);
        }
    });
    // This delete should have deleted both the old and new row format
    scheduleStore.delete(programId, PROGRAM_TYPE, STREAM_SCHEDULE_1.getName());
    txnl.execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            Scanner scanner = table.scan(startKey, stopKey);
            if (defaultVersion) {
                Assert.assertNull(scanner.next());
            } else {
                Assert.assertNotNull(scanner.next());
                Assert.assertNull(scanner.next());
            }
            scanner.close();
        }
    });
    // If the version is not default, we need to delete the row which didn't have a version
    if (!defaultVersion) {
        txnl.execute(new TransactionExecutor.Subroutine() {

            @Override
            public void apply() throws Exception {
                // Create a programId without version so that we can create row key to delete the old format schedule
                ProgramId defaultProgramId = new ProgramId(programId.getNamespace(), programId.getApplication(), programId.getType(), programId.getProgram());
                String newRowKey = scheduleStore.getRowKey(defaultProgramId, PROGRAM_TYPE, STREAM_SCHEDULE_1.getName());
                byte[] oldRowKey = Bytes.toBytes(scheduleStore.removeAppVersion(newRowKey));
                Row row = table.get(oldRowKey);
                Assert.assertFalse(row.isEmpty());
                table.delete(oldRowKey);
            }
        });
    }
}
Also used : Scanner(co.cask.cdap.api.dataset.table.Scanner) Table(co.cask.cdap.api.dataset.table.Table) TransactionExecutor(org.apache.tephra.TransactionExecutor) ProgramId(co.cask.cdap.proto.id.ProgramId) DatasetId(co.cask.cdap.proto.id.DatasetId) TransactionAware(org.apache.tephra.TransactionAware) Row(co.cask.cdap.api.dataset.table.Row)

Example 2 with TransactionAware

use of org.apache.tephra.TransactionAware in project cdap by caskdata.

the class DynamicPartitionerWithAvroTest method runDynamicPartitionerMapReduce.

private void runDynamicPartitionerMapReduce(final List<? extends GenericRecord> records, boolean allowConcurrentWriters, boolean expectedStatus) throws Exception {
    ApplicationWithPrograms app = deployApp(AppWithMapReduceUsingAvroDynamicPartitioner.class);
    final long now = System.currentTimeMillis();
    final Multimap<PartitionKey, GenericRecord> keyToRecordsMap = groupByPartitionKey(records, now);
    // write values to the input kvTable
    final KeyValueTable kvTable = datasetCache.getDataset(INPUT_DATASET);
    Transactions.createTransactionExecutor(txExecutorFactory, kvTable).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() {
            // the keys are not used; it matters that they're unique though
            for (int i = 0; i < records.size(); i++) {
                kvTable.write(Integer.toString(i), records.get(i).toString());
            }
        }
    });
    String allowConcurrencyKey = "dataset." + OUTPUT_DATASET + "." + PartitionedFileSetArguments.DYNAMIC_PARTITIONER_ALLOW_CONCURRENCY;
    // run the partition writer m/r with this output partition time
    ImmutableMap<String, String> arguments = ImmutableMap.of(OUTPUT_PARTITION_KEY, Long.toString(now), allowConcurrencyKey, Boolean.toString(allowConcurrentWriters));
    long startTime = System.currentTimeMillis();
    boolean status = runProgram(app, AppWithMapReduceUsingAvroDynamicPartitioner.DynamicPartitioningMapReduce.class, new BasicArguments(arguments));
    Assert.assertEquals(expectedStatus, status);
    if (!expectedStatus) {
        // if we expect the program to fail, no need to check the output data for expected results
        return;
    }
    // Verify notifications
    List<Notification> notifications = getDataNotifications(startTime);
    Assert.assertEquals(1, notifications.size());
    Assert.assertEquals(NamespaceId.DEFAULT.dataset(OUTPUT_DATASET), DatasetId.fromString(notifications.get(0).getProperties().get("datasetId")));
    // this should have created a partition in the pfs
    final PartitionedFileSet pfs = datasetCache.getDataset(OUTPUT_DATASET);
    final Location pfsBaseLocation = pfs.getEmbeddedFileSet().getBaseLocation();
    Transactions.createTransactionExecutor(txExecutorFactory, (TransactionAware) pfs).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws IOException {
            Map<PartitionKey, PartitionDetail> partitions = new HashMap<>();
            for (PartitionDetail partition : pfs.getPartitions(null)) {
                partitions.put(partition.getPartitionKey(), partition);
                // check that the mapreduce wrote the output partition metadata to all the output partitions
                Assert.assertEquals(AppWithMapReduceUsingAvroDynamicPartitioner.DynamicPartitioningMapReduce.METADATA, partition.getMetadata().asMap());
            }
            Assert.assertEquals(3, partitions.size());
            Assert.assertEquals(keyToRecordsMap.keySet(), partitions.keySet());
            // Check relative paths of the partitions. Also check that their location = pfs baseLocation + relativePath
            for (Map.Entry<PartitionKey, PartitionDetail> partitionKeyEntry : partitions.entrySet()) {
                PartitionDetail partitionDetail = partitionKeyEntry.getValue();
                String relativePath = partitionDetail.getRelativePath();
                int zip = (int) partitionKeyEntry.getKey().getField("zip");
                Assert.assertEquals(Long.toString(now) + Path.SEPARATOR + zip, relativePath);
                Assert.assertEquals(pfsBaseLocation.append(relativePath), partitionDetail.getLocation());
            }
            for (Map.Entry<PartitionKey, Collection<GenericRecord>> keyToRecordsEntry : keyToRecordsMap.asMap().entrySet()) {
                Set<GenericRecord> genericRecords = new HashSet<>(keyToRecordsEntry.getValue());
                Assert.assertEquals(genericRecords, readOutput(partitions.get(keyToRecordsEntry.getKey()).getLocation()));
            }
        }
    });
}
Also used : HashSet(java.util.HashSet) PartitionedFileSet(co.cask.cdap.api.dataset.lib.PartitionedFileSet) Set(java.util.Set) PartitionDetail(co.cask.cdap.api.dataset.lib.PartitionDetail) Notification(co.cask.cdap.proto.Notification) ApplicationWithPrograms(co.cask.cdap.internal.app.deploy.pipeline.ApplicationWithPrograms) BasicArguments(co.cask.cdap.internal.app.runtime.BasicArguments) GenericRecord(org.apache.avro.generic.GenericRecord) TransactionExecutor(org.apache.tephra.TransactionExecutor) PartitionedFileSet(co.cask.cdap.api.dataset.lib.PartitionedFileSet) IOException(java.io.IOException) KeyValueTable(co.cask.cdap.api.dataset.lib.KeyValueTable) TransactionAware(org.apache.tephra.TransactionAware) PartitionKey(co.cask.cdap.api.dataset.lib.PartitionKey) HashMap(java.util.HashMap) Map(java.util.Map) ImmutableMap(com.google.common.collect.ImmutableMap) Location(org.apache.twill.filesystem.Location)

Example 3 with TransactionAware

use of org.apache.tephra.TransactionAware in project cdap by caskdata.

the class AbstractTransactionContext method checkpoint.

@Override
public void checkpoint() throws TransactionFailureException {
    Preconditions.checkState(currentTx != null, "Cannot checkpoint tx that has not been started");
    persist();
    try {
        currentTx = txClient.checkpoint(currentTx);
        // update the current transaction with all TransactionAwares
        for (TransactionAware txAware : getTransactionAwares()) {
            txAware.updateTx(currentTx);
        }
    } catch (TransactionFailureException e) {
        abort(e);
    } catch (Throwable e) {
        abort(new TransactionFailureException(String.format("Exception from checkpoint for transaction %d.", currentTx.getTransactionId()), e));
    }
}
Also used : TransactionFailureException(org.apache.tephra.TransactionFailureException) TransactionAware(org.apache.tephra.TransactionAware)

Example 4 with TransactionAware

use of org.apache.tephra.TransactionAware in project cdap by caskdata.

the class QueueTest method testRollback.

@Test(timeout = TIMEOUT_MS)
public void testRollback() throws Exception {
    QueueName queueName = QueueName.fromFlowlet(NamespaceId.DEFAULT.getEntityName(), "app", "flow", "flowlet", "queuerollback");
    ConsumerConfig consumerConfig = new ConsumerConfig(0, 0, 1, DequeueStrategy.FIFO, null);
    configureGroups(queueName, ImmutableList.of(consumerConfig));
    try (QueueProducer producer = queueClientFactory.createProducer(queueName);
        QueueConsumer consumer = queueClientFactory.createConsumer(queueName, consumerConfig, 1)) {
        TransactionContext txContext = createTxContext(producer, consumer, new TransactionAware() {

            boolean canCommit = false;

            @Override
            public void startTx(Transaction tx) {
            }

            @Override
            public void updateTx(Transaction tx) {
            }

            @Override
            public Collection<byte[]> getTxChanges() {
                return ImmutableList.of();
            }

            @Override
            public boolean commitTx() throws Exception {
                // Flip-flop between commit success/failure.
                boolean res = canCommit;
                canCommit = !canCommit;
                return res;
            }

            @Override
            public void postTxCommit() {
            }

            @Override
            public boolean rollbackTx() throws Exception {
                return true;
            }

            @Override
            public String getTransactionAwareName() {
                return "test";
            }
        });
        // First, try to enqueue and commit would fail
        txContext.start();
        try {
            producer.enqueue(new QueueEntry(Bytes.toBytes(1)));
            txContext.finish();
            // If reaches here, it's wrong, as exception should be thrown.
            Assert.assertTrue(false);
        } catch (TransactionFailureException e) {
            txContext.abort();
        }
        // Try to enqueue again. Within the same transaction, dequeue should be empty.
        txContext.start();
        producer.enqueue(new QueueEntry(Bytes.toBytes(1)));
        Assert.assertTrue(consumer.dequeue().isEmpty());
        txContext.finish();
        // This time, enqueue has been committed, dequeue would see the item
        txContext.start();
        try {
            Assert.assertEquals(1, Bytes.toInt(consumer.dequeue().iterator().next()));
            txContext.finish();
            // If reaches here, it's wrong, as exception should be thrown.
            Assert.assertTrue(false);
        } catch (TransactionFailureException e) {
            txContext.abort();
        }
        // Dequeue again, since last tx was rollback, this dequeue should see the item again.
        txContext.start();
        Assert.assertEquals(1, Bytes.toInt(consumer.dequeue().iterator().next()));
        txContext.finish();
    }
}
Also used : QueueEntry(co.cask.cdap.data2.queue.QueueEntry) TransactionFailureException(org.apache.tephra.TransactionFailureException) TransactionFailureException(org.apache.tephra.TransactionFailureException) QueueConsumer(co.cask.cdap.data2.queue.QueueConsumer) Transaction(org.apache.tephra.Transaction) QueueProducer(co.cask.cdap.data2.queue.QueueProducer) TransactionContext(org.apache.tephra.TransactionContext) TransactionAware(org.apache.tephra.TransactionAware) Collection(java.util.Collection) ConsumerConfig(co.cask.cdap.data2.queue.ConsumerConfig) QueueName(co.cask.cdap.common.queue.QueueName) Test(org.junit.Test)

Example 5 with TransactionAware

use of org.apache.tephra.TransactionAware in project cdap by caskdata.

the class HBaseQueueTest method testReconfigure.

@Test(timeout = 30000L)
public void testReconfigure() throws Exception {
    final QueueName queueName = QueueName.fromFlowlet(NamespaceId.DEFAULT.getEntityName(), "app", "flow", "flowlet", "changeinstances");
    ConsumerGroupConfig groupConfig = new ConsumerGroupConfig(0L, 2, DequeueStrategy.HASH, "key");
    configureGroups(queueName, ImmutableList.of(groupConfig));
    // Enqueue 10 items
    createEnqueueRunnable(queueName, 10, 1, null).run();
    // Map from instance id to items dequeued
    final Multimap<Integer, Integer> dequeued = ArrayListMultimap.create();
    // Consume 2 items for each consumer instances
    for (int instanceId = 0; instanceId < groupConfig.getGroupSize(); instanceId++) {
        final ConsumerConfig consumerConfig = new ConsumerConfig(groupConfig, instanceId);
        try (QueueConsumer consumer = queueClientFactory.createConsumer(queueName, consumerConfig, 1)) {
            Transactions.createTransactionExecutor(executorFactory, (TransactionAware) consumer).execute(new TransactionExecutor.Subroutine() {

                @Override
                public void apply() throws Exception {
                    DequeueResult<byte[]> result = consumer.dequeue(2);
                    Assert.assertEquals(2, result.size());
                    for (byte[] data : result) {
                        dequeued.put(consumerConfig.getInstanceId(), Bytes.toInt(data));
                    }
                }
            });
        }
    }
    // Increase number of instances to 3
    changeInstances(queueName, 0L, 3);
    // Enqueue 10 more items
    createEnqueueRunnable(queueName, 10, 1, null).run();
    groupConfig = new ConsumerGroupConfig(0L, 3, DequeueStrategy.HASH, "key");
    // Dequeue everything
    while (dequeued.size() != 20) {
        for (int instanceId = 0; instanceId < groupConfig.getGroupSize(); instanceId++) {
            final ConsumerConfig consumerConfig = new ConsumerConfig(groupConfig, instanceId);
            try (QueueConsumer consumer = queueClientFactory.createConsumer(queueName, consumerConfig, 1)) {
                Transactions.createTransactionExecutor(executorFactory, (TransactionAware) consumer).execute(new TransactionExecutor.Subroutine() {

                    @Override
                    public void apply() throws Exception {
                        for (byte[] data : consumer.dequeue(20)) {
                            dequeued.put(consumerConfig.getInstanceId(), Bytes.toInt(data));
                        }
                    }
                });
            }
        }
    }
    // Instance 0 should see all evens before change instances
    Assert.assertEquals(ImmutableList.of(0, 2, 4, 6, 8, 0, 3, 6, 9), dequeued.get(0));
    // Instance 1 should see all odds before change instances
    Assert.assertEquals(ImmutableList.of(1, 3, 5, 7, 9, 1, 4, 7), dequeued.get(1));
    // Instance 2 should only see entries after change instances
    Assert.assertEquals(ImmutableList.of(2, 5, 8), dequeued.get(2));
    // All consumers should have empty dequeue now
    for (int instanceId = 0; instanceId < groupConfig.getGroupSize(); instanceId++) {
        final ConsumerConfig consumerConfig = new ConsumerConfig(groupConfig, instanceId);
        try (QueueConsumer consumer = queueClientFactory.createConsumer(queueName, consumerConfig, 1)) {
            Transactions.createTransactionExecutor(executorFactory, (TransactionAware) consumer).execute(new TransactionExecutor.Subroutine() {

                @Override
                public void apply() throws Exception {
                    DequeueResult<byte[]> result = consumer.dequeue(20);
                    Assert.assertTrue(result.isEmpty());
                }
            });
        }
    }
    // Enqueue 6 more items for the 3 instances
    createEnqueueRunnable(queueName, 6, 1, null).run();
    // Reduce to 1 consumer
    changeInstances(queueName, 0L, 1);
    // The consumer 0 should be able to consume all 10 new items
    dequeued.clear();
    final ConsumerConfig consumerConfig = new ConsumerConfig(0L, 0, 1, DequeueStrategy.HASH, "key");
    try (final QueueConsumer consumer = queueClientFactory.createConsumer(queueName, consumerConfig, 1)) {
        while (dequeued.size() != 6) {
            Transactions.createTransactionExecutor(executorFactory, (TransactionAware) consumer).execute(new TransactionExecutor.Subroutine() {

                @Override
                public void apply() throws Exception {
                    for (byte[] data : consumer.dequeue(1)) {
                        dequeued.put(consumerConfig.getInstanceId(), Bytes.toInt(data));
                    }
                }
            });
        }
    }
    Assert.assertEquals(ImmutableList.of(0, 1, 2, 3, 4, 5), dequeued.get(0));
}
Also used : TransactionExecutor(org.apache.tephra.TransactionExecutor) IOException(java.io.IOException) TableNotFoundException(org.apache.hadoop.hbase.TableNotFoundException) QueueConsumer(co.cask.cdap.data2.queue.QueueConsumer) DequeueResult(co.cask.cdap.data2.queue.DequeueResult) TransactionAware(org.apache.tephra.TransactionAware) ConsumerConfig(co.cask.cdap.data2.queue.ConsumerConfig) QueueName(co.cask.cdap.common.queue.QueueName) ConsumerGroupConfig(co.cask.cdap.data2.queue.ConsumerGroupConfig) Test(org.junit.Test) QueueTest(co.cask.cdap.data2.transaction.queue.QueueTest)

Aggregations

TransactionAware (org.apache.tephra.TransactionAware)97 Test (org.junit.Test)65 TransactionExecutor (org.apache.tephra.TransactionExecutor)48 Table (co.cask.cdap.api.dataset.table.Table)39 Transaction (org.apache.tephra.Transaction)34 IOException (java.io.IOException)28 PartitionedFileSet (co.cask.cdap.api.dataset.lib.PartitionedFileSet)24 DatasetAdmin (co.cask.cdap.api.dataset.DatasetAdmin)23 HBaseTable (co.cask.cdap.data2.dataset2.lib.table.hbase.HBaseTable)21 PartitionKey (co.cask.cdap.api.dataset.lib.PartitionKey)19 DataSetException (co.cask.cdap.api.dataset.DataSetException)16 List (java.util.List)13 Put (co.cask.cdap.api.dataset.table.Put)12 DatasetManagementException (co.cask.cdap.api.dataset.DatasetManagementException)11 TimePartitionedFileSet (co.cask.cdap.api.dataset.lib.TimePartitionedFileSet)11 Row (co.cask.cdap.api.dataset.table.Row)11 ArrayList (java.util.ArrayList)11 HashSet (java.util.HashSet)11 TransactionFailureException (org.apache.tephra.TransactionFailureException)11 PartitionNotFoundException (co.cask.cdap.api.dataset.PartitionNotFoundException)10