[
https://issues.apache.org/jira/browse/BEAM-5392?focusedWorklogId=195602&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-195602
]
ASF GitHub Bot logged work on BEAM-5392:
----------------------------------------
Author: ASF GitHub Bot
Created on: 07/Feb/19 11:07
Start Date: 07/Feb/19 11:07
Worklog Time Spent: 10m
Work Description: iemejia commented on pull request #7601: [BEAM-5392]
GroupByKey optimized for non-merging windows
URL: https://github.com/apache/beam/pull/7601#discussion_r254632472
##########
File path:
runners/spark/src/main/java/org/apache/beam/runners/spark/translation/GroupNonMergingWindowsFunctions.java
##########
@@ -0,0 +1,242 @@
+/*
+ * 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.translation;
+
+import java.io.Serializable;
+import java.util.Arrays;
+import java.util.Iterator;
+import java.util.Objects;
+import org.apache.beam.runners.spark.coders.CoderHelpers;
+import org.apache.beam.sdk.coders.ByteArrayCoder;
+import org.apache.beam.sdk.coders.Coder;
+import org.apache.beam.sdk.transforms.windowing.BoundedWindow;
+import org.apache.beam.sdk.util.WindowedValue;
+import org.apache.beam.sdk.util.WindowedValue.FullWindowedValueCoder;
+import org.apache.beam.sdk.values.KV;
+import org.apache.beam.sdk.values.WindowingStrategy;
+import
org.apache.beam.vendor.guava.v20_0.com.google.common.collect.AbstractIterator;
+import org.apache.beam.vendor.guava.v20_0.com.google.common.collect.Iterables;
+import org.apache.beam.vendor.guava.v20_0.com.google.common.collect.Iterators;
+import
org.apache.beam.vendor.guava.v20_0.com.google.common.collect.PeekingIterator;
+import
org.apache.beam.vendor.guava.v20_0.com.google.common.primitives.UnsignedBytes;
+import org.apache.spark.HashPartitioner;
+import org.apache.spark.api.java.JavaRDD;
+import org.joda.time.Instant;
+import scala.Tuple2;
+
+/** Functions for GroupByKey with Non-Merging windows translations to Spark. */
+public class GroupNonMergingWindowsFunctions {
+
+ static <K, V, W extends BoundedWindow>
+ JavaRDD<WindowedValue<KV<K, Iterable<V>>>> groupByKeyAndWindow(
+ JavaRDD<WindowedValue<KV<K, V>>> rdd,
+ Coder<K> keyCoder,
+ Coder<V> valueCoder,
+ WindowingStrategy<?, W> windowingStrategy) {
+ final Coder<W> windowCoder = windowingStrategy.getWindowFn().windowCoder();
+ final WindowedValue.FullWindowedValueCoder<byte[]> windowedValueCoder =
+ WindowedValue.getFullCoder(ByteArrayCoder.of(), windowCoder);
+ return rdd.flatMapToPair(
+ (WindowedValue<KV<K, V>> windowedValue) -> {
+ final byte[] keyBytes =
+ CoderHelpers.toByteArray(windowedValue.getValue().getKey(),
keyCoder);
+ final byte[] valueBytes =
+
CoderHelpers.toByteArray(windowedValue.getValue().getValue(), valueCoder);
+ return Iterators.transform(
+ windowedValue.explodeWindows().iterator(),
+ item -> {
+ Objects.requireNonNull(item, "Exploded window can not be
null.");
+ @SuppressWarnings("unchecked")
+ final W window = (W)
Iterables.getOnlyElement(item.getWindows());
+ final byte[] windowBytes =
CoderHelpers.toByteArray(window, windowCoder);
+ final byte[] windowValueBytes =
+ CoderHelpers.toByteArray(
+ WindowedValue.of(
+ valueBytes, item.getTimestamp(), window,
item.getPane()),
+ windowedValueCoder);
+ final WindowedKey windowedKey = new WindowedKey(keyBytes,
windowBytes);
+ return new Tuple2<>(windowedKey, windowValueBytes);
+ });
+ })
+ .repartitionAndSortWithinPartitions(new
HashPartitioner(rdd.getNumPartitions()))
+ .mapPartitions(
+ it ->
+ new GroupByKeyIterator<>(
+ it, keyCoder, valueCoder, windowingStrategy,
windowedValueCoder))
+ .filter(Objects::nonNull); // filter last null element from
GroupByKeyIterator
+ }
+
+ /**
+ * Transform stream of sorted key values into stream of value iterators for
each key. This
+ * iterator can be iterated only once!
+ *
+ * @param <K> type of key iterator emits
+ * @param <V> type of value iterator emits
+ */
+ static class GroupByKeyIterator<K, V, W extends BoundedWindow>
+ implements Iterator<WindowedValue<KV<K, Iterable<V>>>> {
+
+ private final PeekingIterator<Tuple2<WindowedKey, byte[]>> inner;
+ private final Coder<K> keyCoder;
+ private final Coder<V> valueCoder;
+ private final WindowingStrategy<?, W> windowingStrategy;
+ private final FullWindowedValueCoder<byte[]> windowedValueCoder;
+ private boolean hasNext = true;
+
+ private WindowedKey currentKey = null;
+
+ GroupByKeyIterator(
+ Iterator<Tuple2<WindowedKey, byte[]>> inner,
+ Coder<K> keyCoder,
+ Coder<V> valueCoder,
+ WindowingStrategy<?, W> windowingStrategy,
+ WindowedValue.FullWindowedValueCoder<byte[]> windowedValueCoder) {
+ this.inner = Iterators.peekingIterator(inner);
+ this.keyCoder = keyCoder;
+ this.valueCoder = valueCoder;
+ this.windowingStrategy = windowingStrategy;
+ this.windowedValueCoder = windowedValueCoder;
+ }
+
+ @Override
+ public boolean hasNext() {
+ return hasNext;
+ }
+
+ @Override
+ public WindowedValue<KV<K, Iterable<V>>> next() {
+ while (inner.hasNext()) {
+ final WindowedKey nextKey = inner.peek()._1;
+ if (nextKey.equals(currentKey)) {
+ // we still did not see all values for a given key
+ inner.next();
+ continue;
+ }
+ currentKey = nextKey;
+ final WindowedValue<KV<K, V>> decodedItem = decodeItem(inner.peek());
+ return decodedItem.withValue(
+ KV.of(decodedItem.getValue().getKey(), new ValueIterator(inner,
currentKey)));
+ }
+ hasNext = false;
+ return null;
+ }
+
+ class ValueIterator implements Iterable<V> {
+
+ boolean usedAsIterable = false;
+ private final PeekingIterator<Tuple2<WindowedKey, byte[]>> inner;
+ private WindowedKey currentKey;
+
+ ValueIterator(PeekingIterator<Tuple2<WindowedKey, byte[]>> inner,
WindowedKey currentKey) {
+ this.inner = inner;
+ this.currentKey = currentKey;
+ }
+
+ @Override
+ public Iterator<V> iterator() {
+ if (usedAsIterable) {
+ throw new IllegalStateException(
+ "ValueIterator can't be iterated more than once,"
+ + "otherwise there could be data lost");
+ }
+ usedAsIterable = true;
+ return new AbstractIterator<V>() {
+ @Override
+ protected V computeNext() {
+ if (inner.hasNext() && currentKey.equals(inner.peek()._1)) {
+ return decodeValue(inner.next()._2);
+ }
+ return endOfData();
+ }
+ };
+ }
+ }
+
+ private V decodeValue(byte[] windowedValueBytes) {
+ final WindowedValue<byte[]> windowedValue =
+ CoderHelpers.fromByteArray(windowedValueBytes, windowedValueCoder);
+ return CoderHelpers.fromByteArray(windowedValue.getValue(), valueCoder);
+ }
+
+ private WindowedValue<KV<K, V>> decodeItem(Tuple2<WindowedKey, byte[]>
item) {
+ final K key = CoderHelpers.fromByteArray(item._1.getKey(), keyCoder);
+ final WindowedValue<byte[]> windowedValue =
+ CoderHelpers.fromByteArray(item._2, windowedValueCoder);
+ final V value = CoderHelpers.fromByteArray(windowedValue.getValue(),
valueCoder);
+ @SuppressWarnings("unchecked")
+ final W window = (W)
Iterables.getOnlyElement(windowedValue.getWindows());
+ final Instant timestamp =
+ windowingStrategy
+ .getTimestampCombiner()
+ .assign(
+ window,
+ windowingStrategy
+ .getWindowFn()
+ .getOutputTime(windowedValue.getTimestamp(), window));
+ return WindowedValue.of(KV.of(key, value), timestamp, window,
windowedValue.getPane());
+ }
+ }
+
+ /** Composite key of key and window for groupByKey transformation. */
+ public static class WindowedKey implements Comparable<WindowedKey>,
Serializable {
+
+ private final byte[] key;
+ private final byte[] window;
+
+ WindowedKey(byte[] key, byte[] window) {
+ this.key = key;
+ this.window = window;
+ }
+
+ @Override
+ public boolean equals(Object o) {
+ if (this == o) {
+ return true;
+ }
+ if (o == null || getClass() != o.getClass()) {
+ return false;
+ }
+ WindowedKey that = (WindowedKey) o;
+ return Arrays.equals(key, that.key) && Arrays.equals(window,
that.window);
+ }
+
+ @Override
+ public int hashCode() {
+ int result = Arrays.hashCode(key);
+ result = 31 * result + Arrays.hashCode(window);
+ return result;
+ }
+
+ public byte[] getKey() {
+ return key;
+ }
+
+ public byte[] getWindow() {
Review comment:
remove public
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
Issue Time Tracking
-------------------
Worklog Id: (was: 195602)
Time Spent: 2h (was: 1h 50m)
> GroupByKey on Spark: All values for a single key need to fit in-memory at once
> ------------------------------------------------------------------------------
>
> Key: BEAM-5392
> URL: https://issues.apache.org/jira/browse/BEAM-5392
> Project: Beam
> Issue Type: Bug
> Components: runner-spark
> Affects Versions: 2.6.0
> Reporter: David Moravek
> Assignee: David Moravek
> Priority: Major
> Labels: performance, triaged
> Time Spent: 2h
> Remaining Estimate: 0h
>
> Currently, when using GroupByKey, all values for a single key need to fit
> in-memory at once.
>
> There are following issues, that need to be addressed:
> a) We can not use Spark's _groupByKey_, because it requires all values to fit
> in memory for a single key (it is implemented as "list combiner")
> b) _ReduceFnRunner_ iterates over values multiple times in order to group
> also by window
>
> Solution:
>
> In Dataflow Worker code, there are optimized versions of ReduceFnRunner, that
> can take advantage of having elements for a single key sorted by timestamp.
>
> We can use Spark's `{{repartitionAndSortWithinPartitions}}` in order to meet
> this constraint.
>
> For non-merging windows, we can put window itself into the key resulting in
> smaller groupings.
>
> This approach was already tested in ~100TB input scale on Spark 2.3.x.
> (custom Spark runner).
>
> I'll submit a patch once the Dataflow Worker code donation is complete.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
