Examples with Sink org.apache.beam.sdk.io.FileIO.Sink used on opensource projects

Example 1 with Sink

use of org.apache.beam.sdk.io.FileIO.Sink in project DataflowTemplates by GoogleCloudPlatform.

the class GenericRecordsToGcsPartitioned method expand.

@Override
public PCollection<PartitionMetadata> expand(PCollection<GenericRecord> input) {
    Schema schema = SchemaUtils.parseAvroSchema(serializedAvroSchema);
    Sink<GenericRecord> sink;
    switch(outputFileFormat) {
        case PARQUET:
            sink = ParquetIO.sink(schema);
            break;
        case AVRO:
            sink = new AvroSinkWithJodaDatesConversion<>(schema);
            break;
        default:
            throw new UnsupportedOperationException("Output format is not implemented: " + outputFileFormat);
    }
    if (partitionColumnName == null || partitioningSchema == null) {
        LOG.info("PartitionColumnName or/and PartitioningSchema not provided. " + "Writing to GCS without partition");
        return input.apply("Write to Storage with No Partition", FileIO.<GenericRecord>write().withSuffix(outputFileFormat.getFileSuffix()).via(sink).to(gcsPath)).getPerDestinationOutputFilenames().apply("MapFileNames", MapElements.into(TypeDescriptors.strings()).via((SerializableFunction<KV<Void, String>, String>) KV::getValue)).apply(MapElements.via(new SimpleFunction<String, PartitionMetadata>() {

            @Override
            public PartitionMetadata apply(String path) {
                return PartitionMetadata.builder().setValues(ImmutableList.of("1")).setLocation(withoutFileName(path)).build();
            }
        }));
    }
    ZoneId zoneId = getZoneId(schema);
    return input.apply(FileIO.<List<KV<String, Integer>>, GenericRecord>writeDynamic().by((GenericRecord r) -> partitioningSchema.toPartition(Instant.ofEpochMilli(partitionColumnValueToMillis(r.get(partitionColumnName))).atZone(zoneId))).withDestinationCoder(ListCoder.of(KvCoder.of(StringUtf8Coder.of(), VarIntCoder.of()))).via(sink).to(gcsPath).withNumShards(// must be 1 as we can only have 1 file per Dataplex partition
    1).withNaming(p -> Write.defaultNaming(partitionToPath(p), outputFileFormat.getFileSuffix()))).getPerDestinationOutputFilenames().apply(MapElements.via(new SimpleFunction<KV<List<KV<String, Integer>>, String>, PartitionMetadata>() {

        @Override
        public PartitionMetadata apply(KV<List<KV<String, Integer>>, String> partitionAndPath) {
            if (partitionAndPath.getKey() == null) {
                throw new IllegalStateException("Partition is null for path " + partitionAndPath.getValue());
            }
            if (partitionAndPath.getValue() == null) {
                throw new IllegalStateException("Path is null for partition " + partitionAndPath.getKey());
            }
            return PartitionMetadata.builder().setValues(partitionAndPath.getKey().stream().map(e -> String.valueOf(e.getValue())).collect(toImmutableList())).setLocation(withoutFileName(partitionAndPath.getValue())).build();
        }
    }));
}