dpcollins-google commented on code in PR #23861: URL: https://github.com/apache/beam/pull/23861#discussion_r1009043366
########## sdks/java/core/src/main/java/org/apache/beam/sdk/transforms/WithWatermarkDelayFn.java: ########## @@ -0,0 +1,109 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +package org.apache.beam.sdk.transforms; + +import java.util.Arrays; +import java.util.Collections; +import javax.annotation.Nullable; +import org.apache.beam.sdk.transforms.DoFn.BoundedPerElement; +import org.apache.beam.sdk.transforms.splittabledofn.ManualWatermarkEstimator; +import org.apache.beam.sdk.transforms.splittabledofn.RestrictionTracker; +import org.apache.beam.sdk.transforms.splittabledofn.SplitResult; +import org.apache.beam.sdk.transforms.splittabledofn.WatermarkEstimators; +import org.apache.beam.sdk.transforms.windowing.BoundedWindow; +import org.joda.time.Duration; +import org.joda.time.Instant; + +/** + * Adds a watermark a fixed delay behind the extracted timestamp of the source element. Does not + * allow the watermark to pass the timestamp of the source element. + */ +@BoundedPerElement +class WithWatermarkDelayFn<T> extends DoFn<T, T> { + + private final SerializableFunction<T, Instant> fn; + private final Duration watermarkDelay; + + WithWatermarkDelayFn(SerializableFunction<T, Instant> fn, Duration watermarkDelay) { + this.fn = fn; + this.watermarkDelay = watermarkDelay; + } + + @ProcessElement + public void processElement( + @Element T input, + @Timestamp Instant inputTimestamp, + RestrictionTracker<Integer, Integer> restrictionTracker, + OutputReceiver<T> output, + ManualWatermarkEstimator<Instant> estimator) { + Instant ts = fn.apply(input); + output.outputWithTimestamp(input, ts); + Instant watermark = ts.minus(watermarkDelay); + // Don't advance the watermark past the input element timestamp. + watermark = Collections.min(Arrays.asList(watermark, inputTimestamp)); + if (watermark.isAfter(estimator.currentWatermark())) { + estimator.setWatermark(watermark); Review Comment: > There is no guarantee on the order of input elements and watermarks are monotonic Are you referring to the order that elements are propagated from source to SDF or the order that elements are generated by the source? If I directly have source -> WithTimestamps.withWatermarkDelay, will that do the right thing in 1) the beam model or 2) all currently existing runners? > What usually protects against this is the fact that the unprocessed elements upstream hold the watermark back I thought this still happens at SDFs? I remember Pub/Sub Lite's Source having problems where the impulse function for the SDF wasn't advancing the watermark, holding it back for the rest of the pipeline. Is this "a thing I'm doing wrong" or "a thing you currently believe is infeasible" to maintain this property? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
