Search in sources :

Example 6 with PartitionConsumer

use of co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer in project cdap by caskdata.

the class PartitionConsumerTest method testPartitionConsumingWithPartitionAcceptor.

@Test
public void testPartitionConsumingWithPartitionAcceptor() throws Exception {
    final PartitionedFileSet dataset = dsFrameworkUtil.getInstance(pfsInstance);
    final TransactionAware txAwareDataset = (TransactionAware) dataset;
    // i will range from [0,10), s will always be 'partitionKeys1'
    final Set<PartitionKey> partitionKeys1 = new HashSet<>();
    for (int i = 0; i < 10; i++) {
        PartitionKey key = PartitionKey.builder().addIntField("i", i).addLongField("l", 17L).addStringField("s", "partitionKeys1").build();
        partitionKeys1.add(key);
    }
    // i will range from [0,15), s will always be 'partitionKeys2'
    final Set<PartitionKey> partitionKeys2 = new HashSet<>();
    for (int i = 0; i < 15; i++) {
        PartitionKey key = PartitionKey.builder().addIntField("i", i).addLongField("l", 17L).addStringField("s", "partitionKeys2").build();
        partitionKeys2.add(key);
    }
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            for (final PartitionKey partitionKey : partitionKeys1) {
                dataset.getPartitionOutput(partitionKey).addPartition();
            }
            for (final PartitionKey partitionKey : partitionKeys2) {
                dataset.getPartitionOutput(partitionKey).addPartition();
            }
        }
    });
    final PartitionConsumer partitionConsumer = new ConcurrentPartitionConsumer(dataset, new InMemoryStatePersistor());
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            List<Partition> consumedPartitions = new ArrayList<>();
            // specify a PartitionAcceptor that only limits to partitions where 's' field is equal to 'partitionKeys1'
            // so it will get all the partitions in partitionKeys1
            Iterables.addAll(consumedPartitions, partitionConsumer.consumePartitions(new CustomAcceptor("partitionKeys1")).getPartitions());
            // assert that we consumed all the partitions represented by partitionsKeys1
            Assert.assertEquals(partitionKeys1, toKeys(consumedPartitions));
            consumedPartitions.clear();
            // ask for partitions where 's' field is equal to 'partitionKeys2', but stop iterating upon 'i' field == 8
            Iterables.addAll(consumedPartitions, partitionConsumer.consumePartitions(new CustomAcceptor("partitionKeys2", 8)).getPartitions());
            // this will give us 8 of partitionKeys2
            Assert.assertEquals(8, consumedPartitions.size());
            // ask for the remainder of the partitions - i ranging from [8,15). Then, we will have all of 'partitionKeys2'
            Iterables.addAll(consumedPartitions, partitionConsumer.consumePartitions().getPartitions());
            Assert.assertEquals(partitionKeys2, toKeys(consumedPartitions));
        }
    });
}
Also used : ConcurrentPartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer) PartitionedFileSet(co.cask.cdap.api.dataset.lib.PartitionedFileSet) TransactionExecutor(org.apache.tephra.TransactionExecutor) TransactionAware(org.apache.tephra.TransactionAware) PartitionKey(co.cask.cdap.api.dataset.lib.PartitionKey) ArrayList(java.util.ArrayList) ImmutableList(com.google.common.collect.ImmutableList) List(java.util.List) ConcurrentPartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer) PartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer) HashSet(java.util.HashSet) Test(org.junit.Test)

Example 7 with PartitionConsumer

use of co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer in project cdap by caskdata.

the class PartitionConsumerTest method testSimpleConcurrency.

@Test
public void testSimpleConcurrency() throws Exception {
    final PartitionedFileSet dataset = dsFrameworkUtil.getInstance(pfsInstance);
    final TransactionAware txAwareDataset = (TransactionAware) dataset;
    final Set<PartitionKey> partitionKeys = new HashSet<>();
    for (int i = 0; i < 10; i++) {
        partitionKeys.add(generateUniqueKey());
    }
    // have ConcurrentPartitionConsumers that share the same state.
    InMemoryStatePersistor persistor = new InMemoryStatePersistor();
    ConsumerConfiguration configuration = ConsumerConfiguration.builder().setMaxRetries(3).build();
    final PartitionConsumer partitionConsumer1 = new ConcurrentPartitionConsumer(dataset, persistor, configuration);
    final PartitionConsumer partitionConsumer2 = new ConcurrentPartitionConsumer(dataset, persistor, configuration);
    final PartitionConsumer partitionConsumer3 = new ConcurrentPartitionConsumer(dataset, persistor, configuration);
    // add all ten keys to the partitioned fileset
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            for (final PartitionKey partitionKey : partitionKeys) {
                dataset.getPartitionOutput(partitionKey).addPartition();
            }
        }
    });
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            // with limit = 1, the returned iterator is only size 1, even though there are more unconsumed partitions
            List<PartitionDetail> consumedBy1 = partitionConsumer1.consumePartitions(1).getPartitions();
            Assert.assertEquals(1, consumedBy1.size());
            // partitionConsumer2 asks for 10 partitions, but 1 is currently in progress by partitionConsumer1, so it only
            // gets the remaining 9 partitions
            List<PartitionDetail> consumedBy2 = partitionConsumer2.consumePartitions(10).getPartitions();
            Assert.assertEquals(9, consumedBy2.size());
            // partitionConsumer3 tries to consume partitions, but all are marked in-progress by partitionConsumer 1 and 2
            Assert.assertEquals(0, partitionConsumer3.consumePartitions().getPartitions().size());
            // partitionConsumer1 aborts its partition, so it then becomes available for partitionConsumer3
            partitionConsumer1.onFinish(consumedBy1, false);
            consumedBy1.clear();
            // queries with limit=2, but only the 1 is available that partitionConsumer1 released
            List<PartitionDetail> consumedBy3 = partitionConsumer3.consumePartitions(2).getPartitions();
            Assert.assertEquals(1, consumedBy3.size());
            // partitionConsumers 2 and 3 marks that it successfully processed the partitions
            partitionConsumer3.onFinish(consumedBy3, true);
            // test onFinishWithKeys API
            List<PartitionKey> keysConsumedBy2 = Lists.transform(consumedBy2, new Function<PartitionDetail, PartitionKey>() {

                @Override
                public PartitionKey apply(PartitionDetail input) {
                    return input.getPartitionKey();
                }
            });
            partitionConsumer2.onFinishWithKeys(keysConsumedBy2, true);
            // at this point, all partitions are processed, so no additional partitions are available for consumption
            Assert.assertEquals(0, partitionConsumer3.consumePartitions().getPartitions().size());
            List<PartitionDetail> allProcessedPartitions = new ArrayList<>();
            allProcessedPartitions.addAll(consumedBy1);
            allProcessedPartitions.addAll(consumedBy2);
            allProcessedPartitions.addAll(consumedBy3);
            // ordering may be different, since all the partitions were added in the same transaction
            Assert.assertEquals(partitionKeys, toKeys(allProcessedPartitions));
        }
    });
}
Also used : ConcurrentPartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer) PartitionedFileSet(co.cask.cdap.api.dataset.lib.PartitionedFileSet) TransactionExecutor(org.apache.tephra.TransactionExecutor) PartitionDetail(co.cask.cdap.api.dataset.lib.PartitionDetail) Function(com.google.common.base.Function) TransactionAware(org.apache.tephra.TransactionAware) ConsumerConfiguration(co.cask.cdap.api.dataset.lib.partitioned.ConsumerConfiguration) PartitionKey(co.cask.cdap.api.dataset.lib.PartitionKey) ArrayList(java.util.ArrayList) ImmutableList(com.google.common.collect.ImmutableList) List(java.util.List) ConcurrentPartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer) PartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer) HashSet(java.util.HashSet) Test(org.junit.Test)

Example 8 with PartitionConsumer

use of co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer in project cdap by caskdata.

the class PartitionConsumerTest method testPartitionConsumer.

@Test
public void testPartitionConsumer() throws Exception {
    // exercises the edge case of partition consumption, when partitions are being consumed, while another in-progress
    // transaction has added a partition, but it has not yet committed, so the partition is not available for the
    // consumer
    PartitionedFileSet dataset1 = dsFrameworkUtil.getInstance(pfsInstance);
    PartitionedFileSet dataset2 = dsFrameworkUtil.getInstance(pfsInstance);
    TransactionManager txManager = dsFrameworkUtil.getTxManager();
    InMemoryTxSystemClient txClient = new InMemoryTxSystemClient(txManager);
    // producer simply adds initial partition
    TransactionContext txContext1 = new TransactionContext(txClient, (TransactionAware) dataset1);
    txContext1.start();
    PartitionKey partitionKey1 = generateUniqueKey();
    dataset1.getPartitionOutput(partitionKey1).addPartition();
    txContext1.finish();
    // consumer simply consumes initial partition
    TransactionContext txContext2 = new TransactionContext(txClient, (TransactionAware) dataset2);
    txContext2.start();
    PartitionConsumer partitionConsumer = new ConcurrentPartitionConsumer(dataset2, new InMemoryStatePersistor());
    List<? extends PartitionDetail> partitionIterator = partitionConsumer.consumePartitions().getPartitions();
    Assert.assertEquals(1, partitionIterator.size());
    Assert.assertEquals(partitionKey1, partitionIterator.get(0).getPartitionKey());
    txContext2.finish();
    // producer adds a second partition, but does not yet commit the transaction
    txContext1.start();
    PartitionKey partitionKey2 = generateUniqueKey();
    dataset1.getPartitionOutput(partitionKey2).addPartition();
    // consumer attempts to consume at a time after the partition was added, but before it committed. Because of this,
    // the partition is not visible and will not be consumed
    txContext2.start();
    Assert.assertTrue(partitionConsumer.consumePartitions().getPartitions().isEmpty());
    txContext2.finish();
    // producer commits the transaction in which the second partition was added
    txContext1.finish();
    // the next time the consumer runs, it processes the second partition
    txContext2.start();
    partitionIterator = partitionConsumer.consumePartitions().getPartitions();
    Assert.assertEquals(1, partitionIterator.size());
    Assert.assertEquals(partitionKey2, partitionIterator.get(0).getPartitionKey());
    txContext2.finish();
}
Also used : ConcurrentPartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer) TransactionManager(org.apache.tephra.TransactionManager) TransactionContext(org.apache.tephra.TransactionContext) PartitionKey(co.cask.cdap.api.dataset.lib.PartitionKey) PartitionedFileSet(co.cask.cdap.api.dataset.lib.PartitionedFileSet) ConcurrentPartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer) PartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer) InMemoryTxSystemClient(org.apache.tephra.inmemory.InMemoryTxSystemClient) Test(org.junit.Test)

Example 9 with PartitionConsumer

use of co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer in project cdap by caskdata.

the class PartitionConsumerTest method testOnFinishWithInvalidPartition.

@Test
public void testOnFinishWithInvalidPartition() throws Exception {
    // tests:
    // - attempts to abort a Partition that is not IN_PROGRESS
    // - attempts to commit a Partition that is already committed
    // both of these throw IllegalArgumentException
    final PartitionedFileSet dataset = dsFrameworkUtil.getInstance(pfsInstance);
    final TransactionAware txAwareDataset = (TransactionAware) dataset;
    ConsumerConfiguration configuration = ConsumerConfiguration.builder().setMaxRetries(3).build();
    final PartitionConsumer partitionConsumer = new ConcurrentPartitionConsumer(dataset, new InMemoryStatePersistor(), configuration);
    final PartitionKey partitionKey = generateUniqueKey();
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            dataset.getPartitionOutput(partitionKey).addPartition();
        }
    });
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            List<PartitionDetail> partitionDetails = partitionConsumer.consumePartitions(1).getPartitions();
            Assert.assertEquals(1, partitionDetails.size());
            // aborting the processing of the partition
            partitionConsumer.onFinish(partitionDetails, false);
            // abort were not found to have IN_PROGRESS state
            try {
                partitionConsumer.onFinish(partitionDetails, false);
                Assert.fail("Expected not to be able to abort a partition that is not IN_PROGRESS");
            } catch (IllegalStateException expected) {
            }
            // try to process the partition again, this time marking it as complete (by passing in true)
            partitionDetails = partitionConsumer.consumePartitions(1).getPartitions();
            Assert.assertEquals(1, partitionDetails.size());
            partitionConsumer.onFinish(partitionDetails, true);
            // is not found to have an IN_PROGRESS state
            try {
                partitionConsumer.onFinish(partitionDetails, true);
                Assert.fail("Expected not to be able to call onFinish on a partition is not IN_PROGRESS");
            } catch (IllegalArgumentException expected) {
            }
        }
    });
}
Also used : ConcurrentPartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer) PartitionedFileSet(co.cask.cdap.api.dataset.lib.PartitionedFileSet) TransactionExecutor(org.apache.tephra.TransactionExecutor) TransactionAware(org.apache.tephra.TransactionAware) ConsumerConfiguration(co.cask.cdap.api.dataset.lib.partitioned.ConsumerConfiguration) PartitionKey(co.cask.cdap.api.dataset.lib.PartitionKey) ArrayList(java.util.ArrayList) ImmutableList(com.google.common.collect.ImmutableList) List(java.util.List) ConcurrentPartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer) PartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer) Test(org.junit.Test)

Example 10 with PartitionConsumer

use of co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer in project cdap by caskdata.

the class PartitionConsumerTest method testPartitionConsumingWithFilterAndLimit.

@Test
public void testPartitionConsumingWithFilterAndLimit() throws Exception {
    final PartitionedFileSet dataset = dsFrameworkUtil.getInstance(pfsInstance);
    final TransactionAware txAwareDataset = (TransactionAware) dataset;
    final Set<PartitionKey> partitionKeys1 = new HashSet<>();
    for (int i = 0; i < 10; i++) {
        partitionKeys1.add(generateUniqueKey());
    }
    final Set<PartitionKey> partitionKeys2 = new HashSet<>();
    for (int i = 0; i < 15; i++) {
        partitionKeys2.add(generateUniqueKey());
    }
    final PartitionConsumer partitionConsumer = new ConcurrentPartitionConsumer(dataset, new InMemoryStatePersistor());
    // (consumption only happens at transaction borders)
    for (final PartitionKey partitionKey : partitionKeys1) {
        dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

            @Override
            public void apply() throws Exception {
                dataset.getPartitionOutput(partitionKey).addPartition();
            }
        });
    }
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            // Initial consumption results in the partitions corresponding to partitionKeys1 to be consumed because only
            // those partitions are added to the dataset at this point
            List<Partition> consumedPartitions = new ArrayList<>();
            // with limit = 1, the returned iterator is only size 1, even though there are more unconsumed partitions
            Iterables.addAll(consumedPartitions, partitionConsumer.consumePartitions(1).getPartitions());
            Assert.assertEquals(1, consumedPartitions.size());
            // ask for 5 more
            Iterables.addAll(consumedPartitions, partitionConsumer.consumePartitions(5).getPartitions());
            Assert.assertEquals(6, consumedPartitions.size());
            // ask for 5 more, but there are only 4 more unconsumed partitions (size of partitionKeys1 is 10).
            Iterables.addAll(consumedPartitions, partitionConsumer.consumePartitions(5).getPartitions());
            Assert.assertEquals(10, consumedPartitions.size());
            Assert.assertEquals(partitionKeys1, toKeys(consumedPartitions));
        }
    });
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            for (PartitionKey partitionKey : partitionKeys2) {
                dataset.getPartitionOutput(partitionKey).addPartition();
            }
        }
    });
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            // using the same PartitionConsumer (which remembers the PartitionConsumerState) to consume additional
            // partitions results in only the newly added partitions (corresponding to partitionKeys2) to be returned
            Assert.assertEquals(partitionKeys2, toKeys(partitionConsumer.consumePartitions().getPartitions()));
        }
    });
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            // consuming the partitions again, without adding any new partitions returns an empty iterator
            Assert.assertTrue(partitionConsumer.consumePartitions().getPartitions().isEmpty());
        }
    });
    dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() throws Exception {
            // creating a new PartitionConsumer resets the consumption state.
            // test combination of filter and limit
            // the partitionFilter will match partitionKeys [1, 7), of which there are 6
            final PartitionFilter partitionFilter = PartitionFilter.builder().addRangeCondition("i", 1, 7).build();
            final Predicate<PartitionDetail> predicate = new Predicate<PartitionDetail>() {

                @Override
                public boolean apply(PartitionDetail partitionDetail) {
                    return partitionFilter.match(partitionDetail.getPartitionKey());
                }
            };
            ConsumerConfiguration configuration = ConsumerConfiguration.builder().setPartitionPredicate(predicate).build();
            PartitionConsumer newPartitionConsumer = new ConcurrentPartitionConsumer(dataset, new InMemoryStatePersistor(), configuration);
            List<Partition> consumedPartitions = new ArrayList<>();
            // apply the filter (narrows it down to 6 elements) and apply a limit of 4 results in 4 consumed partitions
            Iterables.addAll(consumedPartitions, newPartitionConsumer.consumePartitions(4).getPartitions());
            Assert.assertEquals(4, consumedPartitions.size());
            // apply a limit of 3, using the same filter returns the remaining 2 elements that fit that filter
            Iterables.addAll(consumedPartitions, newPartitionConsumer.consumePartitions(3).getPartitions());
            Assert.assertEquals(6, consumedPartitions.size());
            // assert that the partitions returned have partition keys, where the i values range from [1, 7]
            Set<Integer> expectedIFields = new HashSet<>();
            for (int i = 1; i < 7; i++) {
                expectedIFields.add(i);
            }
            Set<Integer> actualIFields = new HashSet<>();
            for (Partition consumedPartition : consumedPartitions) {
                actualIFields.add((Integer) consumedPartition.getPartitionKey().getField("i"));
            }
            Assert.assertEquals(expectedIFields, actualIFields);
        }
    });
}
Also used : ConcurrentPartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer) ConsumablePartition(co.cask.cdap.api.dataset.lib.partitioned.ConsumablePartition) Partition(co.cask.cdap.api.dataset.lib.Partition) HashSet(java.util.HashSet) PartitionedFileSet(co.cask.cdap.api.dataset.lib.PartitionedFileSet) Set(java.util.Set) ConsumerWorkingSet(co.cask.cdap.api.dataset.lib.partitioned.ConsumerWorkingSet) PartitionedFileSet(co.cask.cdap.api.dataset.lib.PartitionedFileSet) TransactionExecutor(org.apache.tephra.TransactionExecutor) PartitionDetail(co.cask.cdap.api.dataset.lib.PartitionDetail) Predicate(co.cask.cdap.api.Predicate) PartitionFilter(co.cask.cdap.api.dataset.lib.PartitionFilter) TransactionAware(org.apache.tephra.TransactionAware) ConsumerConfiguration(co.cask.cdap.api.dataset.lib.partitioned.ConsumerConfiguration) PartitionKey(co.cask.cdap.api.dataset.lib.PartitionKey) ArrayList(java.util.ArrayList) ImmutableList(com.google.common.collect.ImmutableList) List(java.util.List) ConcurrentPartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer) PartitionConsumer(co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer) HashSet(java.util.HashSet) Test(org.junit.Test)

Aggregations

PartitionKey (co.cask.cdap.api.dataset.lib.PartitionKey)11 PartitionedFileSet (co.cask.cdap.api.dataset.lib.PartitionedFileSet)11 ConcurrentPartitionConsumer (co.cask.cdap.api.dataset.lib.partitioned.ConcurrentPartitionConsumer)11 PartitionConsumer (co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumer)11 Test (org.junit.Test)11 TransactionAware (org.apache.tephra.TransactionAware)10 TransactionExecutor (org.apache.tephra.TransactionExecutor)10 ImmutableList (com.google.common.collect.ImmutableList)9 ArrayList (java.util.ArrayList)9 List (java.util.List)9 ConsumerConfiguration (co.cask.cdap.api.dataset.lib.partitioned.ConsumerConfiguration)7 HashSet (java.util.HashSet)6 Partition (co.cask.cdap.api.dataset.lib.Partition)3 PartitionDetail (co.cask.cdap.api.dataset.lib.PartitionDetail)3 ConsumablePartition (co.cask.cdap.api.dataset.lib.partitioned.ConsumablePartition)3 PartitionConsumerResult (co.cask.cdap.api.dataset.lib.partitioned.PartitionConsumerResult)3 ConsumerWorkingSet (co.cask.cdap.api.dataset.lib.partitioned.ConsumerWorkingSet)2 Set (java.util.Set)2 Predicate (co.cask.cdap.api.Predicate)1 PartitionFilter (co.cask.cdap.api.dataset.lib.PartitionFilter)1