Examples with PartitionFilter io.cdap.cdap.api.dataset.lib.PartitionFilter used on opensource projects

Example 1 with PartitionFilter

use of io.cdap.cdap.api.dataset.lib.PartitionFilter in project cdap by caskdata.

the class MapReduceWithPartitionedTest method testPartitionedFileSetWithMR.

private void testPartitionedFileSetWithMR(boolean useCombineFileInputFormat) throws Exception {
    ApplicationWithPrograms app = deployApp(AppWithPartitionedFileSet.class, new AppWithPartitionedFileSet.AppConfig(useCombineFileInputFormat));
    // write a value to the input table
    final Table table = datasetCache.getDataset(AppWithPartitionedFileSet.INPUT);
    Transactions.createTransactionExecutor(txExecutorFactory, (TransactionAware) table).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() {
            table.put(Bytes.toBytes("x"), AppWithPartitionedFileSet.ONLY_COLUMN, Bytes.toBytes("1"));
        }
    });
    // a partition key for the map/reduce output
    final PartitionKey keyX = PartitionKey.builder().addStringField("type", "x").addLongField("time", 150000L).build();
    // run the partition writer m/r with this output partition time
    Map<String, String> runtimeArguments = Maps.newHashMap();
    Map<String, String> outputArgs = Maps.newHashMap();
    PartitionedFileSetArguments.setOutputPartitionKey(outputArgs, keyX);
    runtimeArguments.putAll(RuntimeArguments.addScope(Scope.DATASET, PARTITIONED, outputArgs));
    Assert.assertTrue(runProgram(app, AppWithPartitionedFileSet.PartitionWriter.class, new BasicArguments(runtimeArguments)));
    // this should have created a partition in the tpfs
    final PartitionedFileSet dataset = datasetCache.getDataset(PARTITIONED);
    Transactions.createTransactionExecutor(txExecutorFactory, (TransactionAware) dataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() {
            Partition partition = dataset.getPartition(keyX);
            Assert.assertNotNull(partition);
            String path = partition.getRelativePath();
            Assert.assertTrue(path.contains("x"));
            Assert.assertTrue(path.contains("150000"));
        }
    });
    // delete the data in the input table and write a new row
    Transactions.createTransactionExecutor(txExecutorFactory, (TransactionAware) table).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() {
            table.delete(Bytes.toBytes("x"));
            table.put(Bytes.toBytes("y"), AppWithPartitionedFileSet.ONLY_COLUMN, Bytes.toBytes("2"));
        }
    });
    // a new partition key for the next map/reduce
    final PartitionKey keyY = PartitionKey.builder().addStringField("type", "y").addLongField("time", 200000L).build();
    // now run the m/r again with a new partition time, say 5 minutes later
    PartitionedFileSetArguments.setOutputPartitionKey(outputArgs, keyY);
    runtimeArguments.putAll(RuntimeArguments.addScope(Scope.DATASET, PARTITIONED, outputArgs));
    Assert.assertTrue(runProgram(app, AppWithPartitionedFileSet.PartitionWriter.class, new BasicArguments(runtimeArguments)));
    // this should have created a partition in the tpfs
    Transactions.createTransactionExecutor(txExecutorFactory, (TransactionAware) dataset).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() {
            Partition partition = dataset.getPartition(keyY);
            Assert.assertNotNull(partition);
            String path = partition.getRelativePath();
            Assert.assertNotNull(path);
            Assert.assertTrue(path.contains("y"));
            Assert.assertTrue(path.contains("200000"));
        }
    });
    // a partition filter that matches the outputs of both map/reduces
    PartitionFilter filterXY = PartitionFilter.builder().addRangeCondition("type", "x", "z").build();
    // now run a map/reduce that reads all the partitions
    runtimeArguments = Maps.newHashMap();
    Map<String, String> inputArgs = Maps.newHashMap();
    PartitionedFileSetArguments.setInputPartitionFilter(inputArgs, filterXY);
    runtimeArguments.putAll(RuntimeArguments.addScope(Scope.DATASET, PARTITIONED, inputArgs));
    runtimeArguments.put(AppWithPartitionedFileSet.ROW_TO_WRITE, "a");
    Assert.assertTrue(runProgram(app, AppWithPartitionedFileSet.PartitionReader.class, new BasicArguments(runtimeArguments)));
    // this should have read both partitions - and written both x and y to row a
    final Table output = datasetCache.getDataset(AppWithPartitionedFileSet.OUTPUT);
    Transactions.createTransactionExecutor(txExecutorFactory, (TransactionAware) output).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() {
            Row row = output.get(Bytes.toBytes("a"));
            Assert.assertEquals("1", row.getString("x"));
            Assert.assertEquals("{type=x, time=150000}", row.getString("x_key"));
            Assert.assertEquals("2", row.getString("y"));
            Assert.assertEquals("{type=y, time=200000}", row.getString("y_key"));
        }
    });
    // a partition filter that matches the output key of the first map/reduce
    PartitionFilter filterX = PartitionFilter.builder().addValueCondition("type", "x").addRangeCondition("time", null, 160000L).build();
    // now run a map/reduce that reads a range of the partitions, namely the first one
    inputArgs.clear();
    PartitionedFileSetArguments.setInputPartitionFilter(inputArgs, filterX);
    runtimeArguments.putAll(RuntimeArguments.addScope(Scope.DATASET, PARTITIONED, inputArgs));
    runtimeArguments.put(AppWithPartitionedFileSet.ROW_TO_WRITE, "b");
    Assert.assertTrue(runProgram(app, AppWithPartitionedFileSet.PartitionReader.class, new BasicArguments(runtimeArguments)));
    // this should have read the first partition only - and written only x to row b
    Transactions.createTransactionExecutor(txExecutorFactory, (TransactionAware) output).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() {
            Row row = output.get(Bytes.toBytes("b"));
            Assert.assertEquals("1", row.getString("x"));
            Assert.assertEquals("{type=x, time=150000}", row.getString("x_key"));
            Assert.assertNull(row.get("y"));
            Assert.assertNull(row.get("y_key"));
        }
    });
    // a partition filter that matches no key
    PartitionFilter filterMT = PartitionFilter.builder().addValueCondition("type", "nosuchthing").build();
    // now run a map/reduce that reads an empty range of partitions (the filter matches nothing)
    inputArgs.clear();
    PartitionedFileSetArguments.setInputPartitionFilter(inputArgs, filterMT);
    runtimeArguments.putAll(RuntimeArguments.addScope(Scope.DATASET, PARTITIONED, inputArgs));
    runtimeArguments.put(AppWithPartitionedFileSet.ROW_TO_WRITE, "n");
    Assert.assertTrue(runProgram(app, AppWithPartitionedFileSet.PartitionReader.class, new BasicArguments(runtimeArguments)));
    // this should have read no partitions - and written nothing to row n
    Transactions.createTransactionExecutor(txExecutorFactory, (TransactionAware) output).execute(new TransactionExecutor.Subroutine() {

        @Override
        public void apply() {
            Row row = output.get(Bytes.toBytes("n"));
            Assert.assertTrue(row.isEmpty());
        }
    });
}

Example 2 with PartitionFilter

use of io.cdap.cdap.api.dataset.lib.PartitionFilter in project cdap by caskdata.

the class TimePartitionedFileSetTest method testPartitionsForTimeRange.

@Test
public void testPartitionsForTimeRange() throws Exception {
    for (Object[] test : rangeTests) {
        try {
            long start = test[0] instanceof Long ? (Long) test[0] : DATE_FORMAT.parse((String) test[0]).getTime();
            long stop = test[1] instanceof Long ? (Long) test[1] : DATE_FORMAT.parse((String) test[1]).getTime();
            List<PartitionFilter> filters = TimePartitionedFileSetDataset.partitionFiltersForTimeRange(start, stop);
            // Assert.assertEquals(test.length - 2, filters.size());
            Set<String> expectedSet = Sets.newHashSet();
            for (int i = 2; i < test.length; i++) {
                expectedSet.add((String) test[i]);
            }
            Set<String> actualSet = Sets.newHashSet();
            for (PartitionFilter filter : filters) {
                actualSet.add(filter == null ? null : filter.toString());
            }
            Assert.assertEquals(expectedSet, actualSet);
        } catch (Throwable t) {
            throw new Exception("Failed for range " + test[0] + "..." + test[1], t);
        }
    }
}

Example 3 with PartitionFilter

use of io.cdap.cdap.api.dataset.lib.PartitionFilter in project cdap by caskdata.

the class TimePartitionedFileSetTest method testTimePartitionedInputArguments.

@Test
public void testTimePartitionedInputArguments() throws Exception {
    final long time8 = DATE_FORMAT.parse("10/17/2014 8:42 am").getTime();
    final long time9 = DATE_FORMAT.parse("10/17/2014 9:42 am").getTime();
    final String path8 = "8:42";
    final String path9 = "9:42";
    final PartitionFilter filter9 = PartitionFilter.builder().addRangeCondition("hour", 9, null).build();
    // add a few partitions
    {
        final TimePartitionedFileSet dataset = dsFrameworkUtil.getInstance(TPFS_INSTANCE);
        final TransactionAware txAwareDataset = (TransactionAware) dataset;
        dsFrameworkUtil.newInMemoryTransactionExecutor(txAwareDataset).execute(new TransactionExecutor.Subroutine() {

            @Override
            public void apply() throws Exception {
                dataset.addPartition(time8, path8);
                dataset.addPartition(time9, path9);
            }
        });
    }
    // test specifying time range for input
    Map<String, String> arguments = Maps.newHashMap();
    TimePartitionedFileSetArguments.setInputStartTime(arguments, time8 - 30 * MINUTE);
    TimePartitionedFileSetArguments.setInputEndTime(arguments, time8 + 30 * MINUTE);
    testInputConfiguration(arguments, path8);
    // add a partition filter. it should not have an effect as long as there is a time range
    TimePartitionedFileSetArguments.setInputPartitionFilter(arguments, filter9);
    testInputConfiguration(arguments, path8);
    // test specifying input with a partition filter
    arguments.clear();
    TimePartitionedFileSetArguments.setInputPartitionFilter(arguments, filter9);
    testInputConfiguration(arguments, path9);
    // test specifying only a start time or only an end time for input, or none
    arguments.clear();
    TimePartitionedFileSetArguments.setInputStartTime(arguments, time8 + 30 * MINUTE);
    testInputConfigurationFailure(arguments, " with only a start time");
    arguments.clear();
    TimePartitionedFileSetArguments.setInputEndTime(arguments, time8 + 30 * MINUTE);
    testInputConfigurationFailure(arguments, " with only an end time");
}

Example 4 with PartitionFilter

use of io.cdap.cdap.api.dataset.lib.PartitionFilter 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);
        }
    });
}

Example 5 with PartitionFilter

use of io.cdap.cdap.api.dataset.lib.PartitionFilter in project cdap by caskdata.

the class PartitionedFileSetTest method testPartitionConsumingWithFilterAndLimit.

@Test
public void testPartitionConsumingWithFilterAndLimit() throws Exception {
    final PartitionedFileSet dataset = dsFrameworkUtil.getInstance(pfsInstance);
    final TransactionAware txAwareDataset = (TransactionAware) dataset;
    final Set<PartitionKey> partitionKeys1 = Sets.newHashSet();
    for (int i = 0; i < 10; i++) {
        partitionKeys1.add(generateUniqueKey());
    }
    final Set<PartitionKey> partitionKeys2 = Sets.newHashSet();
    for (int i = 0; i < 15; i++) {
        partitionKeys2.add(generateUniqueKey());
    }
    final SimplePartitionConsumer partitionConsumer = new SimplePartitionConsumer(dataset);
    // (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 = Lists.newArrayList();
            // with limit = 1, the returned iterator is only size 1, even though there are more unconsumed partitions
            Iterables.addAll(consumedPartitions, partitionConsumer.consumePartitions(1));
            Assert.assertEquals(1, consumedPartitions.size());
            // ask for 5 more
            Iterables.addAll(consumedPartitions, partitionConsumer.consumePartitions(5));
            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));
            Assert.assertEquals(10, consumedPartitions.size());
            Set<PartitionKey> retrievedKeys = Sets.newHashSet();
            for (Partition consumedPartition : consumedPartitions) {
                retrievedKeys.add(consumedPartition.getPartitionKey());
            }
            Assert.assertEquals(partitionKeys1, retrievedKeys);
        }
    });
    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
            List<Partition> consumedPartitions = Lists.newArrayList();
            Iterables.addAll(consumedPartitions, partitionConsumer.consumePartitions(1));
            // even though we set limit to 1 in the previous call to consumePartitions, we get all the elements of
            // partitionKeys2, because they were all added in the same transaction
            Set<PartitionKey> retrievedKeys = Sets.newHashSet();
            for (Partition consumedPartition : consumedPartitions) {
                retrievedKeys.add(consumedPartition.getPartitionKey());
            }
            Assert.assertEquals(partitionKeys2, retrievedKeys);
        }
    });
    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().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
            SimplePartitionConsumer newPartitionConsumer = new SimplePartitionConsumer(dataset);
            List<Partition> consumedPartitions = Lists.newArrayList();
            // 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());
                }
            };
            // 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, predicate));
            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, predicate));
            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);
        }
    });
}