Search in sources :

Example 6 with PCollectionTuple

use of org.apache.beam.sdk.values.PCollectionTuple in project beam by apache.

the class BatchLoads method expand.

@Override
public WriteResult expand(PCollection<KV<DestinationT, TableRow>> input) {
    Pipeline p = input.getPipeline();
    final String stepUuid = BigQueryHelpers.randomUUIDString();
    PCollectionView<String> tempFilePrefix = p.apply("Create", Create.of((Void) null)).apply("GetTempFilePrefix", ParDo.of(new DoFn<Void, String>() {

        @ProcessElement
        public void getTempFilePrefix(ProcessContext c) {
            c.output(resolveTempLocation(c.getPipelineOptions().getTempLocation(), "BigQueryWriteTemp", stepUuid));
        }
    })).apply("TempFilePrefixView", View.<String>asSingleton());
    // Create a singleton job ID token at execution time. This will be used as the base for all
    // load jobs issued from this instance of the transform.
    PCollectionView<String> jobIdTokenView = p.apply("TriggerIdCreation", Create.of("ignored")).apply("CreateJobId", MapElements.via(new SimpleFunction<String, String>() {

        @Override
        public String apply(String input) {
            return stepUuid;
        }
    })).apply(View.<String>asSingleton());
    PCollection<KV<DestinationT, TableRow>> inputInGlobalWindow = input.apply("rewindowIntoGlobal", Window.<KV<DestinationT, TableRow>>into(new GlobalWindows()).triggering(DefaultTrigger.of()).discardingFiredPanes());
    PCollectionView<Map<DestinationT, String>> schemasView = inputInGlobalWindow.apply(new CalculateSchemas<>(dynamicDestinations));
    TupleTag<WriteBundlesToFiles.Result<DestinationT>> writtenFilesTag = new TupleTag<WriteBundlesToFiles.Result<DestinationT>>("writtenFiles") {
    };
    TupleTag<KV<ShardedKey<DestinationT>, TableRow>> unwrittedRecordsTag = new TupleTag<KV<ShardedKey<DestinationT>, TableRow>>("unwrittenRecords") {
    };
    PCollectionTuple writeBundlesTuple = inputInGlobalWindow.apply("WriteBundlesToFiles", ParDo.of(new WriteBundlesToFiles<>(stepUuid, unwrittedRecordsTag, maxNumWritersPerBundle, maxFileSize)).withOutputTags(writtenFilesTag, TupleTagList.of(unwrittedRecordsTag)));
    PCollection<WriteBundlesToFiles.Result<DestinationT>> writtenFiles = writeBundlesTuple.get(writtenFilesTag).setCoder(WriteBundlesToFiles.ResultCoder.of(destinationCoder));
    // If the bundles contain too many output tables to be written inline to files (due to memory
    // limits), any unwritten records will be spilled to the unwrittenRecordsTag PCollection.
    // Group these records by key, and write the files after grouping. Since the record is grouped
    // by key, we can ensure that only one file is open at a time in each bundle.
    PCollection<WriteBundlesToFiles.Result<DestinationT>> writtenFilesGrouped = writeBundlesTuple.get(unwrittedRecordsTag).setCoder(KvCoder.of(ShardedKeyCoder.of(destinationCoder), TableRowJsonCoder.of())).apply(GroupByKey.<ShardedKey<DestinationT>, TableRow>create()).apply(ParDo.of(new WriteGroupedRecordsToFiles<DestinationT>(tempFilePrefix, maxFileSize)).withSideInputs(tempFilePrefix)).setCoder(WriteBundlesToFiles.ResultCoder.of(destinationCoder));
    // PCollection of filename, file byte size, and table destination.
    PCollection<WriteBundlesToFiles.Result<DestinationT>> results = PCollectionList.of(writtenFiles).and(writtenFilesGrouped).apply(Flatten.<Result<DestinationT>>pCollections());
    TupleTag<KV<ShardedKey<DestinationT>, List<String>>> multiPartitionsTag = new TupleTag<KV<ShardedKey<DestinationT>, List<String>>>("multiPartitionsTag") {
    };
    TupleTag<KV<ShardedKey<DestinationT>, List<String>>> singlePartitionTag = new TupleTag<KV<ShardedKey<DestinationT>, List<String>>>("singlePartitionTag") {
    };
    // Turn the list of files and record counts in a PCollectionView that can be used as a
    // side input.
    PCollectionView<Iterable<WriteBundlesToFiles.Result<DestinationT>>> resultsView = results.apply("ResultsView", View.<WriteBundlesToFiles.Result<DestinationT>>asIterable());
    // This transform will look at the set of files written for each table, and if any table has
    // too many files or bytes, will partition that table's files into multiple partitions for
    // loading.
    PCollection<Void> singleton = p.apply("singleton", Create.of((Void) null).withCoder(VoidCoder.of()));
    PCollectionTuple partitions = singleton.apply("WritePartition", ParDo.of(new WritePartition<>(singletonTable, tempFilePrefix, resultsView, multiPartitionsTag, singlePartitionTag)).withSideInputs(tempFilePrefix, resultsView).withOutputTags(multiPartitionsTag, TupleTagList.of(singlePartitionTag)));
    List<PCollectionView<?>> writeTablesSideInputs = Lists.newArrayList(jobIdTokenView, schemasView);
    writeTablesSideInputs.addAll(dynamicDestinations.getSideInputs());
    Coder<KV<ShardedKey<DestinationT>, List<String>>> partitionsCoder = KvCoder.of(ShardedKeyCoder.of(NullableCoder.of(destinationCoder)), ListCoder.of(StringUtf8Coder.of()));
    // If WriteBundlesToFiles produced more than MAX_NUM_FILES files or MAX_SIZE_BYTES bytes, then
    // the import needs to be split into multiple partitions, and those partitions will be
    // specified in multiPartitionsTag.
    PCollection<KV<TableDestination, String>> tempTables = partitions.get(multiPartitionsTag).setCoder(partitionsCoder).apply("MultiPartitionsReshuffle", Reshuffle.<ShardedKey<DestinationT>, List<String>>of()).apply("MultiPartitionsWriteTables", ParDo.of(new WriteTables<>(false, bigQueryServices, jobIdTokenView, schemasView, WriteDisposition.WRITE_EMPTY, CreateDisposition.CREATE_IF_NEEDED, dynamicDestinations)).withSideInputs(writeTablesSideInputs));
    // This view maps each final table destination to the set of temporary partitioned tables
    // the PCollection was loaded into.
    PCollectionView<Map<TableDestination, Iterable<String>>> tempTablesView = tempTables.apply("TempTablesView", View.<TableDestination, String>asMultimap());
    singleton.apply("WriteRename", ParDo.of(new WriteRename(bigQueryServices, jobIdTokenView, writeDisposition, createDisposition, tempTablesView)).withSideInputs(tempTablesView, jobIdTokenView));
    // Write single partition to final table
    partitions.get(singlePartitionTag).setCoder(partitionsCoder).apply("SinglePartitionsReshuffle", Reshuffle.<ShardedKey<DestinationT>, List<String>>of()).apply("SinglePartitionWriteTables", ParDo.of(new WriteTables<>(true, bigQueryServices, jobIdTokenView, schemasView, writeDisposition, createDisposition, dynamicDestinations)).withSideInputs(writeTablesSideInputs));
    PCollection<TableRow> empty = p.apply("CreateEmptyFailedInserts", Create.empty(TypeDescriptor.of(TableRow.class)));
    return WriteResult.in(input.getPipeline(), new TupleTag<TableRow>("failedInserts"), empty);
}
Also used : TupleTag(org.apache.beam.sdk.values.TupleTag) Result(org.apache.beam.sdk.io.gcp.bigquery.WriteBundlesToFiles.Result) PCollectionTuple(org.apache.beam.sdk.values.PCollectionTuple) TupleTagList(org.apache.beam.sdk.values.TupleTagList) PCollectionList(org.apache.beam.sdk.values.PCollectionList) List(java.util.List) GlobalWindows(org.apache.beam.sdk.transforms.windowing.GlobalWindows) KV(org.apache.beam.sdk.values.KV) Pipeline(org.apache.beam.sdk.Pipeline) PCollectionView(org.apache.beam.sdk.values.PCollectionView) TableRow(com.google.api.services.bigquery.model.TableRow) Map(java.util.Map)

Example 7 with PCollectionTuple

use of org.apache.beam.sdk.values.PCollectionTuple in project beam by apache.

the class ReplacementOutputsTest method taggedExtraReplacementThrows.

@Test
public void taggedExtraReplacementThrows() {
    PCollectionTuple original = PCollectionTuple.of(intsTag, ints).and(strsTag, strs);
    thrown.expect(IllegalArgumentException.class);
    thrown.expectMessage("Missing original output");
    thrown.expectMessage(moreIntsTag.toString());
    thrown.expectMessage(moreReplacementInts.toString());
    ReplacementOutputs.tagged(original.expand(), PCollectionTuple.of(strsTag, replacementStrs).and(moreIntsTag, moreReplacementInts).and(intsTag, replacementInts));
}
Also used : PCollectionTuple(org.apache.beam.sdk.values.PCollectionTuple) Test(org.junit.Test)

Example 8 with PCollectionTuple

use of org.apache.beam.sdk.values.PCollectionTuple in project beam by apache.

the class PTransformTranslationTest method multiMultiParDo.

private static AppliedPTransform<?, ?, ?> multiMultiParDo(Pipeline pipeline) {
    PCollectionView<String> view = pipeline.apply(Create.of("foo")).apply(View.<String>asSingleton());
    PCollection<Long> input = pipeline.apply(GenerateSequence.from(0));
    ParDo.MultiOutput<Long, KV<Long, String>> parDo = ParDo.of(new TestDoFn()).withSideInputs(view).withOutputTags(new TupleTag<KV<Long, String>>() {
    }, TupleTagList.of(new TupleTag<KV<String, Long>>() {
    }));
    PCollectionTuple output = input.apply(parDo);
    Map<TupleTag<?>, PValue> inputs = new HashMap<>();
    inputs.putAll(parDo.getAdditionalInputs());
    inputs.putAll(input.expand());
    return AppliedPTransform.<PCollection<Long>, PCollectionTuple, ParDo.MultiOutput<Long, KV<Long, String>>>of("MultiParDoInAndOut", inputs, output.expand(), parDo, pipeline);
}
Also used : HashMap(java.util.HashMap) TupleTag(org.apache.beam.sdk.values.TupleTag) KV(org.apache.beam.sdk.values.KV) PValue(org.apache.beam.sdk.values.PValue) PCollection(org.apache.beam.sdk.values.PCollection) ParDo(org.apache.beam.sdk.transforms.ParDo) PCollectionTuple(org.apache.beam.sdk.values.PCollectionTuple)

Example 9 with PCollectionTuple

use of org.apache.beam.sdk.values.PCollectionTuple in project beam by apache.

the class PipelineTest method testTupleInjectionTransform.

/**
   * Tests that Pipeline supports putting an element into a tuple as a transform.
   */
@Test
@Category(ValidatesRunner.class)
public void testTupleInjectionTransform() throws Exception {
    PCollection<Integer> input = pipeline.apply(Create.<Integer>of(1, 2, 3, 4));
    TupleTag<Integer> tag = new TupleTag<Integer>();
    PCollectionTuple output = input.apply("ProjectTag", new TupleInjectionTransform<Integer>(tag));
    PAssert.that(output.get(tag)).containsInAnyOrder(1, 2, 3, 4);
    pipeline.run();
}
Also used : TupleTag(org.apache.beam.sdk.values.TupleTag) PCollectionTuple(org.apache.beam.sdk.values.PCollectionTuple) Category(org.junit.experimental.categories.Category) Test(org.junit.Test)

Example 10 with PCollectionTuple

use of org.apache.beam.sdk.values.PCollectionTuple in project beam by apache.

the class PipelineTest method testTupleProjectionTransform.

/**
   * Tests that Pipeline supports pulling an element out of a tuple as a transform.
   */
@Test
@Category(ValidatesRunner.class)
public void testTupleProjectionTransform() throws Exception {
    PCollection<Integer> input = pipeline.apply(Create.<Integer>of(1, 2, 3, 4));
    TupleTag<Integer> tag = new TupleTag<Integer>();
    PCollectionTuple tuple = PCollectionTuple.of(tag, input);
    PCollection<Integer> output = tuple.apply("ProjectTag", new TupleProjectionTransform<Integer>(tag));
    PAssert.that(output).containsInAnyOrder(1, 2, 3, 4);
    pipeline.run();
}
Also used : TupleTag(org.apache.beam.sdk.values.TupleTag) PCollectionTuple(org.apache.beam.sdk.values.PCollectionTuple) Category(org.junit.experimental.categories.Category) Test(org.junit.Test)

Aggregations

PCollectionTuple (org.apache.beam.sdk.values.PCollectionTuple)31 TupleTag (org.apache.beam.sdk.values.TupleTag)27 Test (org.junit.Test)26 Category (org.junit.experimental.categories.Category)13 StringUtils.byteArrayToJsonString (org.apache.beam.sdk.util.StringUtils.byteArrayToJsonString)8 Matchers.containsString (org.hamcrest.Matchers.containsString)8 KV (org.apache.beam.sdk.values.KV)6 PCollection (org.apache.beam.sdk.values.PCollection)5 PCollectionView (org.apache.beam.sdk.values.PCollectionView)4 PValue (org.apache.beam.sdk.values.PValue)4 Pipeline (org.apache.beam.sdk.Pipeline)3 ValueState (org.apache.beam.sdk.state.ValueState)3 DoFn (org.apache.beam.sdk.transforms.DoFn)3 TupleTagList (org.apache.beam.sdk.values.TupleTagList)3 Instant (org.joda.time.Instant)3 TableRow (com.google.api.services.bigquery.model.TableRow)2 List (java.util.List)2 Map (java.util.Map)2 KeyedWorkItem (org.apache.beam.runners.core.KeyedWorkItem)2 StatefulParDo (org.apache.beam.runners.direct.ParDoMultiOverrideFactory.StatefulParDo)2