Examples with InputUDFToTableRow com.google.cloud.teleport.v2.transforms.UDFTextTransformer.InputUDFToTableRow used on opensource projects

Example 1 with InputUDFToTableRow

use of com.google.cloud.teleport.v2.transforms.UDFTextTransformer.InputUDFToTableRow in project DataflowTemplates by GoogleCloudPlatform.

the class DataStreamToBigQuery method run.

/**
 * Runs the pipeline with the supplied options.
 *
 * @param options The execution parameters to the pipeline.
 * @return The result of the pipeline execution.
 */
public static PipelineResult run(Options options) {
    /*
     * Stages:
     *   1) Ingest and Normalize Data to FailsafeElement with JSON Strings
     *   2) Write JSON Strings to TableRow Collection
     *       - Optionally apply a UDF
     *   3) BigQuery Output of TableRow Data
     *     a) Map New Columns & Write to Staging Tables
     *     b) Map New Columns & Merge Staging to Target Table
     *   4) Write Failures to GCS Dead Letter Queue
     */
    Pipeline pipeline = Pipeline.create(options);
    DeadLetterQueueManager dlqManager = buildDlqManager(options);
    String bigqueryProjectId = getBigQueryProjectId(options);
    String dlqDirectory = dlqManager.getRetryDlqDirectoryWithDateTime();
    String tempDlqDir = dlqManager.getRetryDlqDirectory() + "tmp/";
    InputUDFToTableRow<String> failsafeTableRowTransformer = new InputUDFToTableRow<String>(options.getJavascriptTextTransformGcsPath(), options.getJavascriptTextTransformFunctionName(), options.getPythonTextTransformGcsPath(), options.getPythonTextTransformFunctionName(), options.getRuntimeRetries(), FAILSAFE_ELEMENT_CODER);
    /*
     * Stage 1: Ingest and Normalize Data to FailsafeElement with JSON Strings
     *   a) Read DataStream data from GCS into JSON String FailsafeElements (datastreamJsonRecords)
     *   b) Reconsume Dead Letter Queue data from GCS into JSON String FailsafeElements
     *     (dlqJsonRecords)
     *   c) Flatten DataStream and DLQ Streams (jsonRecords)
     */
    PCollection<FailsafeElement<String, String>> datastreamJsonRecords = pipeline.apply(new DataStreamIO(options.getStreamName(), options.getInputFilePattern(), options.getInputFileFormat(), options.getGcsPubSubSubscription(), options.getRfcStartDateTime()).withFileReadConcurrency(options.getFileReadConcurrency()));
    // Elements sent to the Dead Letter Queue are to be reconsumed.
    // A DLQManager is to be created using PipelineOptions, and it is in charge
    // of building pieces of the DLQ.
    PCollection<FailsafeElement<String, String>> dlqJsonRecords = pipeline.apply("DLQ Consumer/reader", dlqManager.dlqReconsumer(options.getDlqRetryMinutes())).apply("DLQ Consumer/cleaner", ParDo.of(new DoFn<String, FailsafeElement<String, String>>() {

        @ProcessElement
        public void process(@Element String input, OutputReceiver<FailsafeElement<String, String>> receiver) {
            receiver.output(FailsafeElement.of(input, input));
        }
    })).setCoder(FAILSAFE_ELEMENT_CODER);
    PCollection<FailsafeElement<String, String>> jsonRecords = PCollectionList.of(datastreamJsonRecords).and(dlqJsonRecords).apply("Merge Datastream & DLQ", Flatten.pCollections());
    /*
     * Stage 2: Write JSON Strings to TableRow PCollectionTuple
     *   a) Optionally apply a Javascript or Python UDF
     *   b) Convert JSON String FailsafeElements to TableRow's (tableRowRecords)
     */
    PCollectionTuple tableRowRecords = jsonRecords.apply("UDF to TableRow/udf", failsafeTableRowTransformer);
    PCollection<TableRow> shuffledTableRows = tableRowRecords.get(failsafeTableRowTransformer.transformOut).apply("UDF to TableRow/ReShuffle", Reshuffle.<TableRow>viaRandomKey().withNumBuckets(100));
    /*
     * Stage 3: BigQuery Output of TableRow Data
     *   a) Map New Columns & Write to Staging Tables (writeResult)
     *   b) Map New Columns & Merge Staging to Target Table (null)
     *
     *   failsafe: writeResult.getFailedInsertsWithErr()
     */
    // TODO(beam 2.23): InsertRetryPolicy should be CDC compliant
    Set<String> fieldsToIgnore = getFieldsToIgnore(options.getIgnoreFields());
    WriteResult writeResult = shuffledTableRows.apply("Map to Staging Tables", new DataStreamMapper(options.as(GcpOptions.class), options.getOutputProjectId(), options.getOutputStagingDatasetTemplate(), options.getOutputStagingTableNameTemplate()).withDataStreamRootUrl(options.getDataStreamRootUrl()).withDefaultSchema(BigQueryDefaultSchemas.DATASTREAM_METADATA_SCHEMA).withDayPartitioning(true).withIgnoreFields(fieldsToIgnore)).apply("Write Successful Records", BigQueryIO.<KV<TableId, TableRow>>write().to(new BigQueryDynamicConverters().bigQueryDynamicDestination()).withFormatFunction(element -> removeTableRowFields(element.getValue(), fieldsToIgnore)).withFormatRecordOnFailureFunction(element -> element.getValue()).withoutValidation().ignoreInsertIds().withCreateDisposition(CreateDisposition.CREATE_NEVER).withWriteDisposition(WriteDisposition.WRITE_APPEND).withExtendedErrorInfo().withMethod(BigQueryIO.Write.Method.STREAMING_INSERTS).withFailedInsertRetryPolicy(InsertRetryPolicy.retryTransientErrors()));
    if (options.getApplyMerge()) {
        shuffledTableRows.apply("Map To Replica Tables", new DataStreamMapper(options.as(GcpOptions.class), options.getOutputProjectId(), options.getOutputDatasetTemplate(), options.getOutputTableNameTemplate()).withDataStreamRootUrl(options.getDataStreamRootUrl()).withDefaultSchema(BigQueryDefaultSchemas.DATASTREAM_METADATA_SCHEMA).withIgnoreFields(fieldsToIgnore)).apply("BigQuery Merge/Build MergeInfo", new MergeInfoMapper(bigqueryProjectId, options.getOutputStagingDatasetTemplate(), options.getOutputStagingTableNameTemplate(), options.getOutputDatasetTemplate(), options.getOutputTableNameTemplate())).apply("BigQuery Merge/Merge into Replica Tables", BigQueryMerger.of(MergeConfiguration.bigQueryConfiguration().withMergeWindowDuration(Duration.standardMinutes(options.getMergeFrequencyMinutes()))));
    }
    /*
     * Stage 4: Write Failures to GCS Dead Letter Queue
     */
    PCollection<String> udfDlqJson = PCollectionList.of(tableRowRecords.get(failsafeTableRowTransformer.udfDeadletterOut)).and(tableRowRecords.get(failsafeTableRowTransformer.transformDeadletterOut)).apply("UDF Failures/Flatten", Flatten.pCollections()).apply("UDF Failures/Sanitize", MapElements.via(new StringDeadLetterQueueSanitizer()));
    PCollection<String> bqWriteDlqJson = writeResult.getFailedInsertsWithErr().apply("BigQuery Failures", MapElements.via(new BigQueryDeadLetterQueueSanitizer()));
    PCollectionList.of(udfDlqJson).and(bqWriteDlqJson).apply("Write To DLQ/Flatten", Flatten.pCollections()).apply("Write To DLQ/Writer", DLQWriteTransform.WriteDLQ.newBuilder().withDlqDirectory(dlqDirectory).withTmpDirectory(tempDlqDir).build());
    // Execute the pipeline and return the result.
    return pipeline.run();
}

Example 2 with InputUDFToTableRow

use of com.google.cloud.teleport.v2.transforms.UDFTextTransformer.InputUDFToTableRow in project DataflowTemplates by GoogleCloudPlatform.

the class PubSubCdcToBigQuery method run.

/**
 * Runs the pipeline to completion with the specified options. This method does not wait until the
 * pipeline is finished before returning. Invoke {@code result.waitUntilFinish()} on the result
 * object to block until the pipeline is finished running if blocking programmatic execution is
 * required.
 *
 * @param options The execution options.
 * @return The pipeline result.
 */
public static PipelineResult run(Options options) {
    Pipeline pipeline = Pipeline.create(options);
    DeadLetterQueueManager dlqManager = buildDlqManager(options);
    String gcsOutputDateTimeDirectory = null;
    if (options.getDeadLetterQueueDirectory() != null) {
        gcsOutputDateTimeDirectory = dlqManager.getRetryDlqDirectory() + "YYYY/MM/DD/HH/mm/";
    }
    CoderRegistry coderRegistry = pipeline.getCoderRegistry();
    coderRegistry.registerCoderForType(CODER.getEncodedTypeDescriptor(), CODER);
    coderRegistry.registerCoderForType(FAILSAFE_ELEMENT_CODER.getEncodedTypeDescriptor(), FAILSAFE_ELEMENT_CODER);
    InputUDFToTableRow<String> failsafeTableRowTransformer = new InputUDFToTableRow<String>(options.getJavascriptTextTransformGcsPath(), options.getJavascriptTextTransformFunctionName(), options.getPythonTextTransformGcsPath(), options.getPythonTextTransformFunctionName(), options.getRuntimeRetries(), FAILSAFE_ELEMENT_CODER);
    BigQueryTableConfigManager bqConfigManager = new BigQueryTableConfigManager((String) options.as(GcpOptions.class).getProject(), (String) options.getOutputDatasetTemplate(), (String) options.getOutputTableNameTemplate(), (String) options.getOutputTableSpec());
    /*
     * Steps:
     *  1) Read messages in from Pub/Sub
     *  2) Transform the PubsubMessages into TableRows
     *     - Transform message payload via UDF
     *     - Convert UDF result to TableRow objects
     *  3) Write successful records out to BigQuery
     *     - Automap new objects to BigQuery if enabled
     *     - Write records to BigQuery tables
     *  4) Write failed records out to BigQuery
     */
    /*
     * Step #1: Read messages in from Pub/Sub
     */
    PCollection<PubsubMessage> messages = pipeline.apply("ReadPubSubSubscription", PubsubIO.readMessagesWithAttributes().fromSubscription(options.getInputSubscription()));
    PCollection<FailsafeElement<String, String>> jsonRecords;
    if (options.getDeadLetterQueueDirectory() != null) {
        PCollection<FailsafeElement<String, String>> failsafeMessages = messages.apply("ConvertPubSubToFailsafe", ParDo.of(new PubSubToFailSafeElement()));
        PCollection<FailsafeElement<String, String>> dlqJsonRecords = pipeline.apply(dlqManager.dlqReconsumer()).apply(ParDo.of(new DoFn<String, FailsafeElement<String, String>>() {

            @ProcessElement
            public void process(@Element String input, OutputReceiver<FailsafeElement<String, String>> receiver) {
                receiver.output(FailsafeElement.of(input, input));
            }
        })).setCoder(FAILSAFE_ELEMENT_CODER);
        jsonRecords = PCollectionList.of(failsafeMessages).and(dlqJsonRecords).apply(Flatten.pCollections());
    } else {
        jsonRecords = messages.apply("ConvertPubSubToFailsafe", ParDo.of(new PubSubToFailSafeElement()));
    }
    PCollectionTuple convertedTableRows = jsonRecords.apply(Reshuffle.<FailsafeElement<String, String>>viaRandomKey().withNumBuckets(options.getThreadCount())).apply("ApplyUdfAndConvertToTableRow", failsafeTableRowTransformer);
    /*
     * Step #3: Write the successful records out to BigQuery
     *   Either extract table destination only
     *   or extract table destination and auto-map new columns
     */
    PCollection<KV<TableId, TableRow>> tableEvents;
    if (options.getAutoMapTables()) {
        tableEvents = convertedTableRows.get(failsafeTableRowTransformer.transformOut).apply("Map Data to BigQuery Tables", new BigQueryMappers(bqConfigManager.getProjectId()).buildBigQueryTableMapper(bqConfigManager.getDatasetTemplate(), bqConfigManager.getTableTemplate()).withDefaultSchemaFromGCS(options.getSchemaFilePath()));
    } else {
        tableEvents = convertedTableRows.get(failsafeTableRowTransformer.transformOut).apply("ExtractBigQueryTableDestination", BigQueryDynamicConverters.extractTableRowDestination(bqConfigManager.getProjectId(), bqConfigManager.getDatasetTemplate(), bqConfigManager.getTableTemplate()));
    }
    /*
     * Step #3: Cont.
     *    - Write rows out to BigQuery
     */
    // TODO(https://github.com/apache/beam/pull/12004): Switch out alwaysRetry
    WriteResult writeResult = tableEvents.apply("WriteSuccessfulRecords", BigQueryIO.<KV<TableId, TableRow>>write().to(new BigQueryDynamicConverters().bigQueryDynamicDestination()).withFormatFunction(element -> element.getValue()).withoutValidation().withCreateDisposition(CreateDisposition.CREATE_NEVER).withWriteDisposition(WriteDisposition.WRITE_APPEND).withExtendedErrorInfo().withMethod(BigQueryIO.Write.Method.STREAMING_INSERTS).withFailedInsertRetryPolicy(InsertRetryPolicy.alwaysRetry()));
    // TODO: Cover tableRowRecords.get(TRANSFORM_DEADLETTER_OUT) error values
    if (options.getDeadLetterQueueDirectory() != null) {
        writeResult.getFailedInsertsWithErr().apply("DLQ: Write Insert Failures to GCS", MapElements.via(new BigQueryDeadLetterQueueSanitizer())).apply("Creating " + options.getWindowDuration() + " Window", Window.into(FixedWindows.of(DurationUtils.parseDuration(options.getWindowDuration())))).apply("DLQ: Write File(s)", TextIO.write().withWindowedWrites().withNumShards(20).to(new WindowedFilenamePolicy(gcsOutputDateTimeDirectory, "error", "-SSSSS-of-NNNNN", ".json")).withTempDirectory(FileBasedSink.convertToFileResourceIfPossible(options.getDeadLetterQueueDirectory())));
        PCollection<FailsafeElement<String, String>> transformDeadletter = PCollectionList.of(ImmutableList.of(convertedTableRows.get(failsafeTableRowTransformer.udfDeadletterOut), convertedTableRows.get(failsafeTableRowTransformer.transformDeadletterOut))).apply("Flatten", Flatten.pCollections()).apply("Creating " + options.getWindowDuration() + " Window", Window.into(FixedWindows.of(DurationUtils.parseDuration(options.getWindowDuration()))));
        PCollection<String> dlqWindowing = transformDeadletter.apply("Sanitize records", MapElements.via(new StringDeadLetterQueueSanitizer())).setCoder(StringUtf8Coder.of());
        dlqWindowing.apply("DLQ: Write File(s)", TextIO.write().withWindowedWrites().withNumShards(20).to(new WindowedFilenamePolicy(gcsOutputDateTimeDirectory, "error", "-SSSSS-of-NNNNN", ".json")).withTempDirectory(FileBasedSink.convertToFileResourceIfPossible(gcsOutputDateTimeDirectory + "tmp/")));
    } else {
        PCollection<FailsafeElement<String, String>> failedInserts = writeResult.getFailedInsertsWithErr().apply("WrapInsertionErrors", MapElements.into(FAILSAFE_ELEMENT_CODER.getEncodedTypeDescriptor()).via((BigQueryInsertError e) -> BigQueryConverters.wrapBigQueryInsertError(e))).setCoder(FAILSAFE_ELEMENT_CODER);
        /*
       * Step #4: Write records that failed table row transformation
       * or conversion out to BigQuery deadletter table.
       */
        PCollectionList.of(ImmutableList.of(convertedTableRows.get(failsafeTableRowTransformer.udfDeadletterOut), convertedTableRows.get(failsafeTableRowTransformer.transformDeadletterOut))).apply("Flatten", Flatten.pCollections()).apply("WriteFailedRecords", ErrorConverters.WriteStringMessageErrors.newBuilder().setErrorRecordsTable(BigQueryConverters.maybeUseDefaultDeadletterTable(options.getOutputDeadletterTable(), bqConfigManager.getOutputTableSpec(), DEFAULT_DEADLETTER_TABLE_SUFFIX)).setErrorRecordsTableSchema(ResourceUtils.getDeadletterTableSchemaJson()).build());
        // 5) Insert records that failed insert into deadletter table
        failedInserts.apply("WriteFailedRecords", ErrorConverters.WriteStringMessageErrors.newBuilder().setErrorRecordsTable(BigQueryConverters.maybeUseDefaultDeadletterTable(options.getOutputDeadletterTable(), bqConfigManager.getOutputTableSpec(), DEFAULT_DEADLETTER_TABLE_SUFFIX)).setErrorRecordsTableSchema(ResourceUtils.getDeadletterTableSchemaJson()).build());
    }
    return pipeline.run();
}

Example 3 with InputUDFToTableRow

use of com.google.cloud.teleport.v2.transforms.UDFTextTransformer.InputUDFToTableRow in project DataflowTemplates by GoogleCloudPlatform.

the class PubSubCdcToBigQueryTest method testPubSubCdcToBigQueryApplyJavaScriptUDF.

/**
 * Tests the {@link PubSubCdcToBigQuery} pipeline end-to-end.
 */
@Test
public void testPubSubCdcToBigQueryApplyJavaScriptUDF() throws Exception {
    // Test input
    final String payload = "{\"ticker\": \"GOOGL\", \"price\": 1006.94}";
    final PubsubMessage message = new PubsubMessage(payload.getBytes(), ImmutableMap.of("id", "123", "type", "custom_event"));
    final Instant timestamp = new DateTime(2022, 2, 22, 22, 22, 22, 222, DateTimeZone.UTC).toInstant();
    final FailsafeElementCoder<String, String> coder = FailsafeElementCoder.of(StringUtf8Coder.of(), StringUtf8Coder.of());
    CoderRegistry coderRegistry = pipeline.getCoderRegistry();
    coderRegistry.registerCoderForType(coder.getEncodedTypeDescriptor(), coder);
    // Parameters
    String transformPath = TRANSFORM_FILE_PATH;
    String transformFunction = "transform";
    PubSubCdcToBigQuery.Options options = PipelineOptionsFactory.create().as(PubSubCdcToBigQuery.Options.class);
    options.setJavascriptTextTransformGcsPath(transformPath);
    options.setJavascriptTextTransformFunctionName(transformFunction);
    InputUDFToTableRow<String> deadletterHandler = new InputUDFToTableRow<String>(options.getJavascriptTextTransformGcsPath(), options.getJavascriptTextTransformFunctionName(), options.getPythonTextTransformGcsPath(), options.getPythonTextTransformFunctionName(), options.getRuntimeRetries(), coder);
    // Build pipeline
    PCollectionTuple transformOut = pipeline.apply("CreateInput", Create.timestamped(TimestampedValue.of(message, timestamp)).withCoder(PubsubMessageWithAttributesCoder.of())).apply("ConvertPubSubToFailsafe", ParDo.of(new PubSubToFailSafeElement())).apply("ConvertMessageToTableRow", deadletterHandler);
    transformOut.get(deadletterHandler.udfDeadletterOut).setCoder(coder);
    transformOut.get(deadletterHandler.transformDeadletterOut).setCoder(coder);
    // Assert
    PAssert.that(transformOut.get(deadletterHandler.udfDeadletterOut)).empty();
    PAssert.that(transformOut.get(deadletterHandler.transformDeadletterOut)).empty();
    PAssert.that(transformOut.get(deadletterHandler.transformOut)).satisfies(collection -> {
        TableRow result = collection.iterator().next();
        assertThat(result.get("ticker"), is(equalTo("GOOGL")));
        assertThat(result.get("price"), is(equalTo(1006.94)));
        return null;
    });
    // Execute pipeline
    pipeline.run();
}