tkaymak commented on code in PR #38255: URL: https://github.com/apache/beam/pull/38255#discussion_r3142402605
########## runners/spark/4/src/main/java/org/apache/beam/runners/spark/structuredstreaming/translation/batch/GroupByKeyTranslatorBatch.java: ########## @@ -0,0 +1,298 @@ +/* + * 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.runners.spark.structuredstreaming.translation.batch; + +import static org.apache.beam.repackaged.core.org.apache.commons.lang3.ArrayUtils.EMPTY_BYTE_ARRAY; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.eligibleForGlobalGroupBy; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.eligibleForGroupByWindow; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.explodeWindowedKey; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.valueKey; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.valueValue; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.windowedKV; +import static org.apache.beam.runners.spark.structuredstreaming.translation.helpers.CoderHelpers.toByteArray; +import static org.apache.beam.runners.spark.structuredstreaming.translation.helpers.EncoderHelpers.collectionEncoder; +import static org.apache.beam.runners.spark.structuredstreaming.translation.helpers.EncoderHelpers.encoderOf; +import static org.apache.beam.runners.spark.structuredstreaming.translation.helpers.EncoderHelpers.kvEncoder; +import static org.apache.beam.runners.spark.structuredstreaming.translation.helpers.EncoderHelpers.windowedValueEncoder; +import static org.apache.beam.runners.spark.structuredstreaming.translation.utils.ScalaInterop.concat; +import static org.apache.beam.runners.spark.structuredstreaming.translation.utils.ScalaInterop.fun1; +import static org.apache.beam.runners.spark.structuredstreaming.translation.utils.ScalaInterop.fun2; +import static org.apache.beam.runners.spark.structuredstreaming.translation.utils.ScalaInterop.javaIterator; +import static org.apache.beam.runners.spark.structuredstreaming.translation.utils.ScalaInterop.seqOf; +import static org.apache.beam.sdk.transforms.windowing.PaneInfo.NO_FIRING; +import static org.apache.beam.vendor.guava.v32_1_2_jre.com.google.common.base.Preconditions.checkState; +import static org.apache.spark.sql.functions.col; +import static org.apache.spark.sql.functions.collect_list; +import static org.apache.spark.sql.functions.explode; +import static org.apache.spark.sql.functions.max; +import static org.apache.spark.sql.functions.min; +import static org.apache.spark.sql.functions.struct; + +import java.io.Serializable; +import org.apache.beam.runners.core.InMemoryStateInternals; +import org.apache.beam.runners.core.ReduceFnRunner; +import org.apache.beam.runners.core.StateInternalsFactory; +import org.apache.beam.runners.core.SystemReduceFn; +import org.apache.beam.runners.spark.SparkCommonPipelineOptions; +import org.apache.beam.runners.spark.structuredstreaming.translation.TransformTranslator; +import org.apache.beam.runners.spark.structuredstreaming.translation.batch.functions.GroupAlsoByWindowViaOutputBufferFn; +import org.apache.beam.sdk.coders.KvCoder; +import org.apache.beam.sdk.transforms.GroupByKey; +import org.apache.beam.sdk.transforms.windowing.BoundedWindow; +import org.apache.beam.sdk.transforms.windowing.GlobalWindow; +import org.apache.beam.sdk.transforms.windowing.PaneInfo.PaneInfoCoder; +import org.apache.beam.sdk.transforms.windowing.TimestampCombiner; +import org.apache.beam.sdk.values.KV; +import org.apache.beam.sdk.values.PCollection; +import org.apache.beam.sdk.values.WindowedValue; +import org.apache.beam.sdk.values.WindowedValues; +import org.apache.beam.sdk.values.WindowingStrategy; +import org.apache.spark.sql.Column; +import org.apache.spark.sql.Dataset; +import org.apache.spark.sql.Encoder; +import org.apache.spark.sql.TypedColumn; +import org.apache.spark.sql.types.DataType; +import org.apache.spark.sql.types.DataTypes; +import org.checkerframework.checker.nullness.qual.NonNull; +import scala.Tuple2; +import scala.collection.Iterator; +import scala.collection.JavaConverters; +import scala.collection.immutable.List; + +/** + * Translator for {@link GroupByKey} using {@link Dataset#groupByKey} with the built-in aggregation + * function {@code collect_list} when applicable. + * + * <p>Note: Using {@code collect_list} isn't any worse than using {@link ReduceFnRunner}. In the + * latter case the entire group (iterator) has to be loaded into memory as well. Either way there's + * a risk of OOM errors. When enabling {@link + * SparkCommonPipelineOptions#getPreferGroupByKeyToHandleHugeValues()}, a more memory sensitive + * iterable is used that can be traversed just once. Attempting to traverse the iterable again will + * throw. + * + * <ul> + * <li>When using the default global window, window information is dropped and restored after the + * aggregation. + * <li>For non-merging windows, windows are exploded and moved into a composite key for better + * distribution. Though, to keep the amount of shuffled data low, this is only done if values + * are assigned to a single window or if there are only few keys and distributing data is + * important. After the aggregation, windowed values are restored from the composite key. + * <li>All other cases are implemented using the SDK {@link ReduceFnRunner}. + * </ul> + */ +class GroupByKeyTranslatorBatch<K, V> + extends TransformTranslator< + PCollection<KV<K, V>>, PCollection<KV<K, Iterable<V>>>, GroupByKey<K, V>> { + + /** Literal of binary encoded Pane info. */ + private static final Column PANE_NO_FIRING = lit(toByteArray(NO_FIRING, PaneInfoCoder.of())); + + /** Defaults for value in single global window. */ + private static final List<Column> GLOBAL_WINDOW_DETAILS = + windowDetails(lit(new byte[][] {EMPTY_BYTE_ARRAY})); + + GroupByKeyTranslatorBatch() { + super(0.2f); + } + + @Override + public void translate(GroupByKey<K, V> transform, Context cxt) { + WindowingStrategy<?, ?> windowing = cxt.getInput().getWindowingStrategy(); + TimestampCombiner tsCombiner = windowing.getTimestampCombiner(); + + Dataset<WindowedValue<KV<K, V>>> input = cxt.getDataset(cxt.getInput()); + + KvCoder<K, V> inputCoder = (KvCoder<K, V>) cxt.getInput().getCoder(); + KvCoder<K, Iterable<V>> outputCoder = (KvCoder<K, Iterable<V>>) cxt.getOutput().getCoder(); + + Encoder<V> valueEnc = cxt.valueEncoderOf(inputCoder); + Encoder<K> keyEnc = cxt.keyEncoderOf(inputCoder); + + // In batch we can ignore triggering and allowed lateness parameters + final Dataset<WindowedValue<KV<K, Iterable<V>>>> result; + + boolean useCollectList = + !cxt.getOptions() + .as(SparkCommonPipelineOptions.class) + .getPreferGroupByKeyToHandleHugeValues(); + if (useCollectList && eligibleForGlobalGroupBy(windowing, false)) { + // Collects all values per key in memory. This might be problematic if there's + // few keys only + // or some highly skewed distribution. + result = + input + .groupBy(col("value.key").as("key")) + .agg(collect_list(col("value.value")).as("values"), timestampAggregator(tsCombiner)) + .select( + inGlobalWindow( + keyValue(col("key").as(keyEnc), col("values").as(iterableEnc(valueEnc))), + windowTimestamp(tsCombiner))); + + } else if (eligibleForGlobalGroupBy(windowing, true)) { + // Produces an iterable that can be traversed exactly once. However, on the plus + // side, data is + // not collected in memory until serialized or done by the user. + result = + cxt.getDataset(cxt.getInput()) + .groupByKey(valueKey(), keyEnc) + .mapValues(valueValue(), cxt.valueEncoderOf(inputCoder)) + .mapGroups(fun2((k, it) -> KV.of(k, iterableOnce(it))), cxt.kvEncoderOf(outputCoder)) + .map(fun1(WindowedValues::valueInGlobalWindow), cxt.windowedEncoder(outputCoder)); + + } else if (useCollectList + && eligibleForGroupByWindow(windowing, false) + && (windowing.getWindowFn().assignsToOneWindow() || transform.fewKeys())) { + // Using the window as part of the key should help to better distribute the + // data. However, if + // values are assigned to multiple windows, more data would be shuffled around. + // If there's few + // keys only, this is still valuable. + // Collects all values per key & window in memory. + result = + input + .select(explode(col("windows")).as("window"), col("value"), col("timestamp")) + .groupBy(col("value.key").as("key"), col("window")) + .agg(collect_list(col("value.value")).as("values"), timestampAggregator(tsCombiner)) + .select( + inSingleWindow( + keyValue(col("key").as(keyEnc), col("values").as(iterableEnc(valueEnc))), + col("window").as(cxt.windowEncoder()), + windowTimestamp(tsCombiner))); + + } else if (eligibleForGroupByWindow(windowing, true) + && (windowing.getWindowFn().assignsToOneWindow() || transform.fewKeys())) { + // Using the window as part of the key should help to better distribute the + // data. However, if + // values are assigned to multiple windows, more data would be shuffled around. + // If there's few + // keys only, this is still valuable. + // Produces an iterable that can be traversed exactly once. However, on the plus + // side, data is + // not collected in memory until serialized or done by the user. + Encoder<Tuple2<BoundedWindow, K>> windowedKeyEnc = + cxt.tupleEncoder(cxt.windowEncoder(), keyEnc); + result = + cxt.getDataset(cxt.getInput()) + .flatMap(explodeWindowedKey(valueValue()), cxt.tupleEncoder(windowedKeyEnc, valueEnc)) + .groupByKey(fun1(t -> t._1()), windowedKeyEnc) + .mapValues(fun1(t -> t._2()), valueEnc) + .mapGroups( + fun2((wKey, it) -> windowedKV(wKey, iterableOnce(it))), + cxt.windowedEncoder(outputCoder)); + + } else { + // Collects all values per key in memory. This might be problematic if there's + // few keys only + // or some highly skewed distribution. + + // FIXME Revisit this case, implementation is far from ideal: + // - iterator traversed at least twice, forcing materialization in memory + + // group by key, then by windows + result = + input + .groupByKey(valueKey(), keyEnc) + .flatMapGroups( + new GroupAlsoByWindowViaOutputBufferFn<>( + windowing, + (SerStateInternalsFactory) key -> InMemoryStateInternals.forKey(key), + SystemReduceFn.buffering(inputCoder.getValueCoder()), + cxt.getOptionsSupplier()), + cxt.windowedEncoder(outputCoder)); + } Review Comment: Happy to do this in a follow up, I don't want to blow up the PR right now further, unless you think it makes sense to include it now. ########## runners/spark/4/src/main/java/org/apache/beam/runners/spark/structuredstreaming/translation/batch/GroupByKeyTranslatorBatch.java: ########## @@ -0,0 +1,298 @@ +/* + * 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.runners.spark.structuredstreaming.translation.batch; + +import static org.apache.beam.repackaged.core.org.apache.commons.lang3.ArrayUtils.EMPTY_BYTE_ARRAY; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.eligibleForGlobalGroupBy; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.eligibleForGroupByWindow; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.explodeWindowedKey; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.valueKey; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.valueValue; +import static org.apache.beam.runners.spark.structuredstreaming.translation.batch.GroupByKeyHelpers.windowedKV; +import static org.apache.beam.runners.spark.structuredstreaming.translation.helpers.CoderHelpers.toByteArray; +import static org.apache.beam.runners.spark.structuredstreaming.translation.helpers.EncoderHelpers.collectionEncoder; +import static org.apache.beam.runners.spark.structuredstreaming.translation.helpers.EncoderHelpers.encoderOf; +import static org.apache.beam.runners.spark.structuredstreaming.translation.helpers.EncoderHelpers.kvEncoder; +import static org.apache.beam.runners.spark.structuredstreaming.translation.helpers.EncoderHelpers.windowedValueEncoder; +import static org.apache.beam.runners.spark.structuredstreaming.translation.utils.ScalaInterop.concat; +import static org.apache.beam.runners.spark.structuredstreaming.translation.utils.ScalaInterop.fun1; +import static org.apache.beam.runners.spark.structuredstreaming.translation.utils.ScalaInterop.fun2; +import static org.apache.beam.runners.spark.structuredstreaming.translation.utils.ScalaInterop.javaIterator; +import static org.apache.beam.runners.spark.structuredstreaming.translation.utils.ScalaInterop.seqOf; +import static org.apache.beam.sdk.transforms.windowing.PaneInfo.NO_FIRING; +import static org.apache.beam.vendor.guava.v32_1_2_jre.com.google.common.base.Preconditions.checkState; +import static org.apache.spark.sql.functions.col; +import static org.apache.spark.sql.functions.collect_list; +import static org.apache.spark.sql.functions.explode; +import static org.apache.spark.sql.functions.max; +import static org.apache.spark.sql.functions.min; +import static org.apache.spark.sql.functions.struct; + +import java.io.Serializable; +import org.apache.beam.runners.core.InMemoryStateInternals; +import org.apache.beam.runners.core.ReduceFnRunner; +import org.apache.beam.runners.core.StateInternalsFactory; +import org.apache.beam.runners.core.SystemReduceFn; +import org.apache.beam.runners.spark.SparkCommonPipelineOptions; +import org.apache.beam.runners.spark.structuredstreaming.translation.TransformTranslator; +import org.apache.beam.runners.spark.structuredstreaming.translation.batch.functions.GroupAlsoByWindowViaOutputBufferFn; +import org.apache.beam.sdk.coders.KvCoder; +import org.apache.beam.sdk.transforms.GroupByKey; +import org.apache.beam.sdk.transforms.windowing.BoundedWindow; +import org.apache.beam.sdk.transforms.windowing.GlobalWindow; +import org.apache.beam.sdk.transforms.windowing.PaneInfo.PaneInfoCoder; +import org.apache.beam.sdk.transforms.windowing.TimestampCombiner; +import org.apache.beam.sdk.values.KV; +import org.apache.beam.sdk.values.PCollection; +import org.apache.beam.sdk.values.WindowedValue; +import org.apache.beam.sdk.values.WindowedValues; +import org.apache.beam.sdk.values.WindowingStrategy; +import org.apache.spark.sql.Column; +import org.apache.spark.sql.Dataset; +import org.apache.spark.sql.Encoder; +import org.apache.spark.sql.TypedColumn; +import org.apache.spark.sql.types.DataType; +import org.apache.spark.sql.types.DataTypes; +import org.checkerframework.checker.nullness.qual.NonNull; +import scala.Tuple2; +import scala.collection.Iterator; +import scala.collection.JavaConverters; +import scala.collection.immutable.List; + +/** + * Translator for {@link GroupByKey} using {@link Dataset#groupByKey} with the built-in aggregation + * function {@code collect_list} when applicable. + * + * <p>Note: Using {@code collect_list} isn't any worse than using {@link ReduceFnRunner}. In the + * latter case the entire group (iterator) has to be loaded into memory as well. Either way there's + * a risk of OOM errors. When enabling {@link + * SparkCommonPipelineOptions#getPreferGroupByKeyToHandleHugeValues()}, a more memory sensitive + * iterable is used that can be traversed just once. Attempting to traverse the iterable again will + * throw. + * + * <ul> + * <li>When using the default global window, window information is dropped and restored after the + * aggregation. + * <li>For non-merging windows, windows are exploded and moved into a composite key for better + * distribution. Though, to keep the amount of shuffled data low, this is only done if values + * are assigned to a single window or if there are only few keys and distributing data is + * important. After the aggregation, windowed values are restored from the composite key. + * <li>All other cases are implemented using the SDK {@link ReduceFnRunner}. + * </ul> + */ +class GroupByKeyTranslatorBatch<K, V> + extends TransformTranslator< + PCollection<KV<K, V>>, PCollection<KV<K, Iterable<V>>>, GroupByKey<K, V>> { + + /** Literal of binary encoded Pane info. */ + private static final Column PANE_NO_FIRING = lit(toByteArray(NO_FIRING, PaneInfoCoder.of())); + + /** Defaults for value in single global window. */ + private static final List<Column> GLOBAL_WINDOW_DETAILS = + windowDetails(lit(new byte[][] {EMPTY_BYTE_ARRAY})); + + GroupByKeyTranslatorBatch() { + super(0.2f); + } + + @Override + public void translate(GroupByKey<K, V> transform, Context cxt) { + WindowingStrategy<?, ?> windowing = cxt.getInput().getWindowingStrategy(); + TimestampCombiner tsCombiner = windowing.getTimestampCombiner(); + + Dataset<WindowedValue<KV<K, V>>> input = cxt.getDataset(cxt.getInput()); + + KvCoder<K, V> inputCoder = (KvCoder<K, V>) cxt.getInput().getCoder(); + KvCoder<K, Iterable<V>> outputCoder = (KvCoder<K, Iterable<V>>) cxt.getOutput().getCoder(); + + Encoder<V> valueEnc = cxt.valueEncoderOf(inputCoder); + Encoder<K> keyEnc = cxt.keyEncoderOf(inputCoder); + + // In batch we can ignore triggering and allowed lateness parameters + final Dataset<WindowedValue<KV<K, Iterable<V>>>> result; + + boolean useCollectList = + !cxt.getOptions() + .as(SparkCommonPipelineOptions.class) + .getPreferGroupByKeyToHandleHugeValues(); + if (useCollectList && eligibleForGlobalGroupBy(windowing, false)) { + // Collects all values per key in memory. This might be problematic if there's + // few keys only + // or some highly skewed distribution. + result = + input + .groupBy(col("value.key").as("key")) + .agg(collect_list(col("value.value")).as("values"), timestampAggregator(tsCombiner)) + .select( + inGlobalWindow( + keyValue(col("key").as(keyEnc), col("values").as(iterableEnc(valueEnc))), + windowTimestamp(tsCombiner))); + + } else if (eligibleForGlobalGroupBy(windowing, true)) { + // Produces an iterable that can be traversed exactly once. However, on the plus + // side, data is + // not collected in memory until serialized or done by the user. + result = + cxt.getDataset(cxt.getInput()) + .groupByKey(valueKey(), keyEnc) + .mapValues(valueValue(), cxt.valueEncoderOf(inputCoder)) + .mapGroups(fun2((k, it) -> KV.of(k, iterableOnce(it))), cxt.kvEncoderOf(outputCoder)) + .map(fun1(WindowedValues::valueInGlobalWindow), cxt.windowedEncoder(outputCoder)); + + } else if (useCollectList + && eligibleForGroupByWindow(windowing, false) + && (windowing.getWindowFn().assignsToOneWindow() || transform.fewKeys())) { + // Using the window as part of the key should help to better distribute the + // data. However, if + // values are assigned to multiple windows, more data would be shuffled around. + // If there's few + // keys only, this is still valuable. + // Collects all values per key & window in memory. + result = + input + .select(explode(col("windows")).as("window"), col("value"), col("timestamp")) + .groupBy(col("value.key").as("key"), col("window")) + .agg(collect_list(col("value.value")).as("values"), timestampAggregator(tsCombiner)) + .select( + inSingleWindow( + keyValue(col("key").as(keyEnc), col("values").as(iterableEnc(valueEnc))), + col("window").as(cxt.windowEncoder()), + windowTimestamp(tsCombiner))); + + } else if (eligibleForGroupByWindow(windowing, true) + && (windowing.getWindowFn().assignsToOneWindow() || transform.fewKeys())) { + // Using the window as part of the key should help to better distribute the + // data. However, if + // values are assigned to multiple windows, more data would be shuffled around. + // If there's few + // keys only, this is still valuable. + // Produces an iterable that can be traversed exactly once. However, on the plus + // side, data is + // not collected in memory until serialized or done by the user. + Encoder<Tuple2<BoundedWindow, K>> windowedKeyEnc = + cxt.tupleEncoder(cxt.windowEncoder(), keyEnc); + result = + cxt.getDataset(cxt.getInput()) + .flatMap(explodeWindowedKey(valueValue()), cxt.tupleEncoder(windowedKeyEnc, valueEnc)) + .groupByKey(fun1(t -> t._1()), windowedKeyEnc) + .mapValues(fun1(t -> t._2()), valueEnc) + .mapGroups( + fun2((wKey, it) -> windowedKV(wKey, iterableOnce(it))), + cxt.windowedEncoder(outputCoder)); + + } else { + // Collects all values per key in memory. This might be problematic if there's + // few keys only + // or some highly skewed distribution. + + // FIXME Revisit this case, implementation is far from ideal: + // - iterator traversed at least twice, forcing materialization in memory + + // group by key, then by windows + result = + input + .groupByKey(valueKey(), keyEnc) + .flatMapGroups( + new GroupAlsoByWindowViaOutputBufferFn<>( + windowing, + (SerStateInternalsFactory) key -> InMemoryStateInternals.forKey(key), + SystemReduceFn.buffering(inputCoder.getValueCoder()), + cxt.getOptionsSupplier()), + cxt.windowedEncoder(outputCoder)); + } Review Comment: Happy to do this in a follow up, I don't want to blow up the PR right now any further, unless you think it makes sense to include it now. -- 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]
