You are viewing a plain text version of this content. The canonical link for it is here.
Posted to github@beam.apache.org by "twosom (via GitHub)" <gi...@apache.org> on 2024/12/08 12:28:28 UTC

[PR] Batch optimized SparkRunner groupByKey [beam]

twosom opened a new pull request, #33322:
URL: https://github.com/apache/beam/pull/33322

   **Please** add a meaningful description for your change here
   
   fixes #20943
   
   This PR improves the performance of GroupByKey transform in SparkRunner by replacing the current implementation that uses Spark's `groupByKey` with `combineByKey`. 
   
   The current implementation uses Spark's `groupByKey` which causes all the data to be shuffled across the network before grouping. This can lead to significant performance overhead and potential OOM issues with large datasets.
   
   By switching to Spark's `combineByKey`, we can:
   - Perform partial combining at the map side before shuffling
   - Reduce the amount of data transferred across the network
   - Improve memory utilization and overall performance
   
   
   ## Before optimized - `6.1GiB` for shuffle
   <img width="1257" alt="스크린샷 2024-12-08 오후 7 42 03" src="https://github.com/user-attachments/assets/871f066c-3802-425e-80e1-119edd6d3e19">
   
   ## After optimized - `1487.7MiB` for shuffle.
   <img width="1250" alt="스크린샷 2024-12-08 오후 9 22 45" src="https://github.com/user-attachments/assets/53b5197d-6df4-42bc-9c8a-7494ee8d6b47">
   
   
   
   ------------------------
   
   Thank you for your contribution! Follow this checklist to help us incorporate your contribution quickly and easily:
   
    - [ ] Mention the appropriate issue in your description (for example: `addresses #123`), if applicable. This will automatically add a link to the pull request in the issue. If you would like the issue to automatically close on merging the pull request, comment `fixes #<ISSUE NUMBER>` instead.
    - [ ] Update `CHANGES.md` with noteworthy changes.
    - [ ] If this contribution is large, please file an Apache [Individual Contributor License Agreement](https://www.apache.org/licenses/icla.pdf).
   
   See the [Contributor Guide](https://beam.apache.org/contribute) for more tips on [how to make review process smoother](https://github.com/apache/beam/blob/master/CONTRIBUTING.md#make-the-reviewers-job-easier).
   
   To check the build health, please visit [https://github.com/apache/beam/blob/master/.test-infra/BUILD_STATUS.md](https://github.com/apache/beam/blob/master/.test-infra/BUILD_STATUS.md)
   
   GitHub Actions Tests Status (on master branch)
   ------------------------------------------------------------------------------------------------
   [![Build python source distribution and wheels](https://github.com/apache/beam/workflows/Build%20python%20source%20distribution%20and%20wheels/badge.svg?branch=master&event=schedule)](https://github.com/apache/beam/actions?query=workflow%3A%22Build+python+source+distribution+and+wheels%22+branch%3Amaster+event%3Aschedule)
   [![Python tests](https://github.com/apache/beam/workflows/Python%20tests/badge.svg?branch=master&event=schedule)](https://github.com/apache/beam/actions?query=workflow%3A%22Python+Tests%22+branch%3Amaster+event%3Aschedule)
   [![Java tests](https://github.com/apache/beam/workflows/Java%20Tests/badge.svg?branch=master&event=schedule)](https://github.com/apache/beam/actions?query=workflow%3A%22Java+Tests%22+branch%3Amaster+event%3Aschedule)
   [![Go tests](https://github.com/apache/beam/workflows/Go%20tests/badge.svg?branch=master&event=schedule)](https://github.com/apache/beam/actions?query=workflow%3A%22Go+tests%22+branch%3Amaster+event%3Aschedule)
   
   See [CI.md](https://github.com/apache/beam/blob/master/CI.md) for more information about GitHub Actions CI or the [workflows README](https://github.com/apache/beam/blob/master/.github/workflows/README.md) to see a list of phrases to trigger workflows.
   


-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "twosom (via GitHub)" <gi...@apache.org>.
twosom commented on PR #33322:
URL: https://github.com/apache/beam/pull/33322#issuecomment-2526095933

   Run Java Spark PortableValidatesRunner Batch


-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "twosom (via GitHub)" <gi...@apache.org>.
twosom commented on PR #33322:
URL: https://github.com/apache/beam/pull/33322#issuecomment-2526038177

   Run Java Spark PortableValidatesRunner Batch


-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "damccorm (via GitHub)" <gi...@apache.org>.
damccorm merged PR #33322:
URL: https://github.com/apache/beam/pull/33322


-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "damccorm (via GitHub)" <gi...@apache.org>.
damccorm commented on code in PR #33322:
URL: https://github.com/apache/beam/pull/33322#discussion_r1884006358


##########
runners/spark/src/main/java/org/apache/beam/runners/spark/translation/GroupCombineFunctions.java:
##########
@@ -49,18 +53,36 @@ public static <K, V> JavaRDD<KV<K, Iterable<WindowedValue<V>>>> groupByKeyOnly(
       @Nullable Partitioner partitioner) {
     // we use coders to convert objects in the PCollection to byte arrays, so they
     // can be transferred over the network for the shuffle.
-    JavaPairRDD<ByteArray, byte[]> pairRDD =
-        rdd.map(new ReifyTimestampsAndWindowsFunction<>())
-            .mapToPair(TranslationUtils.toPairFunction())
-            .mapToPair(CoderHelpers.toByteFunction(keyCoder, wvCoder));
-
-    // If no partitioner is passed, the default group by key operation is called
-    JavaPairRDD<ByteArray, Iterable<byte[]>> groupedRDD =
-        (partitioner != null) ? pairRDD.groupByKey(partitioner) : pairRDD.groupByKey();
-
-    return groupedRDD
-        .mapToPair(CoderHelpers.fromByteFunctionIterable(keyCoder, wvCoder))
-        .map(new TranslationUtils.FromPairFunction<>());
+    final JavaPairRDD<ByteArray, byte[]> pairRDD =
+        rdd.mapPartitionsToPair(

Review Comment:
   > Additionally, Spark can better optimize execution with combined operations, reducing serialization overhead.
   
   Got it - this was the piece that I was missing - thanks!



-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "twosom (via GitHub)" <gi...@apache.org>.
twosom commented on code in PR #33322:
URL: https://github.com/apache/beam/pull/33322#discussion_r1882995425


##########
runners/spark/src/main/java/org/apache/beam/runners/spark/translation/GroupCombineFunctions.java:
##########
@@ -49,18 +53,36 @@ public static <K, V> JavaRDD<KV<K, Iterable<WindowedValue<V>>>> groupByKeyOnly(
       @Nullable Partitioner partitioner) {
     // we use coders to convert objects in the PCollection to byte arrays, so they
     // can be transferred over the network for the shuffle.
-    JavaPairRDD<ByteArray, byte[]> pairRDD =
-        rdd.map(new ReifyTimestampsAndWindowsFunction<>())
-            .mapToPair(TranslationUtils.toPairFunction())
-            .mapToPair(CoderHelpers.toByteFunction(keyCoder, wvCoder));
-
-    // If no partitioner is passed, the default group by key operation is called
-    JavaPairRDD<ByteArray, Iterable<byte[]>> groupedRDD =
-        (partitioner != null) ? pairRDD.groupByKey(partitioner) : pairRDD.groupByKey();
-
-    return groupedRDD
-        .mapToPair(CoderHelpers.fromByteFunctionIterable(keyCoder, wvCoder))
-        .map(new TranslationUtils.FromPairFunction<>());
+    final JavaPairRDD<ByteArray, byte[]> pairRDD =
+        rdd.mapPartitionsToPair(

Review Comment:
   Yes, there are several advantages to using `mapPartitionsToPair` and `mapPartitions`. 
   It processes an entire partition at once instead of individual elements, reducing function call overhead. 
   This approach reduces intermediate RDDs and stages, resulting in significant shuffle size reduction (6.1GiB → 1487.7MiB). 
   Additionally, Spark can better optimize execution with combined operations, reducing serialization overhead. While slightly more complex to read, the performance benefits outweigh this trade-off for a critical operation like `GroupByKey`.



-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "twosom (via GitHub)" <gi...@apache.org>.
twosom commented on PR #33322:
URL: https://github.com/apache/beam/pull/33322#issuecomment-2526097374

   assign set of reviewers


-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "github-actions[bot] (via GitHub)" <gi...@apache.org>.
github-actions[bot] commented on PR #33322:
URL: https://github.com/apache/beam/pull/33322#issuecomment-2526104136

   Assigning reviewers. If you would like to opt out of this review, comment `assign to next reviewer`:
   
   R: @damccorm for label build.
   
   Available commands:
   - `stop reviewer notifications` - opt out of the automated review tooling
   - `remind me after tests pass` - tag the comment author after tests pass
   - `waiting on author` - shift the attention set back to the author (any comment or push by the author will return the attention set to the reviewers)
   
   The PR bot will only process comments in the main thread (not review comments).


-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "twosom (via GitHub)" <gi...@apache.org>.
twosom commented on code in PR #33322:
URL: https://github.com/apache/beam/pull/33322#discussion_r1882991701


##########
runners/spark/src/test/java/org/apache/beam/runners/spark/translation/GroupNonMergingWindowsFunctionsTest.java:
##########
@@ -121,54 +112,6 @@ public void testGbkIteratorValuesCannotBeReiterated() throws Coder.NonDeterminis
     }
   }
 
-  @Test
-  @SuppressWarnings({"rawtypes", "unchecked"})
-  public void testGroupByKeyInGlobalWindowWithPartitioner() {
-    // mocking
-    Partitioner mockPartitioner = mock(Partitioner.class);
-    JavaRDD mockRdd = mock(JavaRDD.class);
-    Coder mockKeyCoder = mock(Coder.class);
-    Coder mockValueCoder = mock(Coder.class);
-    JavaPairRDD mockRawKeyValues = mock(JavaPairRDD.class);
-    JavaPairRDD mockGrouped = mock(JavaPairRDD.class);
-
-    when(mockRdd.mapToPair(any())).thenReturn(mockRawKeyValues);
-    when(mockRawKeyValues.groupByKey(any(Partitioner.class)))
-        .thenAnswer(
-            invocation -> {
-              Partitioner partitioner = invocation.getArgument(0);
-              assertEquals(partitioner, mockPartitioner);
-              return mockGrouped;
-            });
-    when(mockGrouped.map(any())).thenReturn(mock(JavaRDD.class));
-
-    GroupNonMergingWindowsFunctions.groupByKeyInGlobalWindow(
-        mockRdd, mockKeyCoder, mockValueCoder, mockPartitioner);
-
-    verify(mockRawKeyValues, never()).groupByKey();
-    verify(mockRawKeyValues, times(1)).groupByKey(any(Partitioner.class));
-  }
-
-  @Test
-  @SuppressWarnings({"rawtypes", "unchecked"})
-  public void testGroupByKeyInGlobalWindowWithoutPartitioner() {
-    // mocking
-    JavaRDD mockRdd = mock(JavaRDD.class);
-    Coder mockKeyCoder = mock(Coder.class);
-    Coder mockValueCoder = mock(Coder.class);
-    JavaPairRDD mockRawKeyValues = mock(JavaPairRDD.class);
-    JavaPairRDD mockGrouped = mock(JavaPairRDD.class);
-
-    when(mockRdd.mapToPair(any())).thenReturn(mockRawKeyValues);
-    when(mockRawKeyValues.groupByKey()).thenReturn(mockGrouped);
-
-    GroupNonMergingWindowsFunctions.groupByKeyInGlobalWindow(
-        mockRdd, mockKeyCoder, mockValueCoder, null);
-
-    verify(mockRawKeyValues, times(1)).groupByKey();
-    verify(mockRawKeyValues, never()).groupByKey(any(Partitioner.class));

Review Comment:
   Because the `GroupNonMergingWindowsFunctions.groupByKeyInGlobalWindow` method no longer uses RDD's `groupByKey` operation. 
   
   The implementation has been replaced with `combineByKey`, making these tests that verify `groupByKey` calls no longer relevant.



-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "damccorm (via GitHub)" <gi...@apache.org>.
damccorm commented on code in PR #33322:
URL: https://github.com/apache/beam/pull/33322#discussion_r1884006707


##########
runners/spark/src/test/java/org/apache/beam/runners/spark/translation/GroupNonMergingWindowsFunctionsTest.java:
##########
@@ -121,54 +112,6 @@ public void testGbkIteratorValuesCannotBeReiterated() throws Coder.NonDeterminis
     }
   }
 
-  @Test
-  @SuppressWarnings({"rawtypes", "unchecked"})
-  public void testGroupByKeyInGlobalWindowWithPartitioner() {
-    // mocking
-    Partitioner mockPartitioner = mock(Partitioner.class);
-    JavaRDD mockRdd = mock(JavaRDD.class);
-    Coder mockKeyCoder = mock(Coder.class);
-    Coder mockValueCoder = mock(Coder.class);
-    JavaPairRDD mockRawKeyValues = mock(JavaPairRDD.class);
-    JavaPairRDD mockGrouped = mock(JavaPairRDD.class);
-
-    when(mockRdd.mapToPair(any())).thenReturn(mockRawKeyValues);
-    when(mockRawKeyValues.groupByKey(any(Partitioner.class)))
-        .thenAnswer(
-            invocation -> {
-              Partitioner partitioner = invocation.getArgument(0);
-              assertEquals(partitioner, mockPartitioner);
-              return mockGrouped;
-            });
-    when(mockGrouped.map(any())).thenReturn(mock(JavaRDD.class));
-
-    GroupNonMergingWindowsFunctions.groupByKeyInGlobalWindow(
-        mockRdd, mockKeyCoder, mockValueCoder, mockPartitioner);
-
-    verify(mockRawKeyValues, never()).groupByKey();
-    verify(mockRawKeyValues, times(1)).groupByKey(any(Partitioner.class));
-  }
-
-  @Test
-  @SuppressWarnings({"rawtypes", "unchecked"})
-  public void testGroupByKeyInGlobalWindowWithoutPartitioner() {
-    // mocking
-    JavaRDD mockRdd = mock(JavaRDD.class);
-    Coder mockKeyCoder = mock(Coder.class);
-    Coder mockValueCoder = mock(Coder.class);
-    JavaPairRDD mockRawKeyValues = mock(JavaPairRDD.class);
-    JavaPairRDD mockGrouped = mock(JavaPairRDD.class);
-
-    when(mockRdd.mapToPair(any())).thenReturn(mockRawKeyValues);
-    when(mockRawKeyValues.groupByKey()).thenReturn(mockGrouped);
-
-    GroupNonMergingWindowsFunctions.groupByKeyInGlobalWindow(
-        mockRdd, mockKeyCoder, mockValueCoder, null);
-
-    verify(mockRawKeyValues, times(1)).groupByKey();
-    verify(mockRawKeyValues, never()).groupByKey(any(Partitioner.class));

Review Comment:
   Oh I see - thanks



-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "twosom (via GitHub)" <gi...@apache.org>.
twosom commented on PR #33322:
URL: https://github.com/apache/beam/pull/33322#issuecomment-2540169029

   @damccorm
   Thanks for your question! I've added the explanation as a comment on the PR.


-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "damccorm (via GitHub)" <gi...@apache.org>.
damccorm commented on code in PR #33322:
URL: https://github.com/apache/beam/pull/33322#discussion_r1882614060


##########
runners/spark/src/test/java/org/apache/beam/runners/spark/translation/GroupNonMergingWindowsFunctionsTest.java:
##########
@@ -121,54 +112,6 @@ public void testGbkIteratorValuesCannotBeReiterated() throws Coder.NonDeterminis
     }
   }
 
-  @Test
-  @SuppressWarnings({"rawtypes", "unchecked"})
-  public void testGroupByKeyInGlobalWindowWithPartitioner() {
-    // mocking
-    Partitioner mockPartitioner = mock(Partitioner.class);
-    JavaRDD mockRdd = mock(JavaRDD.class);
-    Coder mockKeyCoder = mock(Coder.class);
-    Coder mockValueCoder = mock(Coder.class);
-    JavaPairRDD mockRawKeyValues = mock(JavaPairRDD.class);
-    JavaPairRDD mockGrouped = mock(JavaPairRDD.class);
-
-    when(mockRdd.mapToPair(any())).thenReturn(mockRawKeyValues);
-    when(mockRawKeyValues.groupByKey(any(Partitioner.class)))
-        .thenAnswer(
-            invocation -> {
-              Partitioner partitioner = invocation.getArgument(0);
-              assertEquals(partitioner, mockPartitioner);
-              return mockGrouped;
-            });
-    when(mockGrouped.map(any())).thenReturn(mock(JavaRDD.class));
-
-    GroupNonMergingWindowsFunctions.groupByKeyInGlobalWindow(
-        mockRdd, mockKeyCoder, mockValueCoder, mockPartitioner);
-
-    verify(mockRawKeyValues, never()).groupByKey();
-    verify(mockRawKeyValues, times(1)).groupByKey(any(Partitioner.class));
-  }
-
-  @Test
-  @SuppressWarnings({"rawtypes", "unchecked"})
-  public void testGroupByKeyInGlobalWindowWithoutPartitioner() {
-    // mocking
-    JavaRDD mockRdd = mock(JavaRDD.class);
-    Coder mockKeyCoder = mock(Coder.class);
-    Coder mockValueCoder = mock(Coder.class);
-    JavaPairRDD mockRawKeyValues = mock(JavaPairRDD.class);
-    JavaPairRDD mockGrouped = mock(JavaPairRDD.class);
-
-    when(mockRdd.mapToPair(any())).thenReturn(mockRawKeyValues);
-    when(mockRawKeyValues.groupByKey()).thenReturn(mockGrouped);
-
-    GroupNonMergingWindowsFunctions.groupByKeyInGlobalWindow(
-        mockRdd, mockKeyCoder, mockValueCoder, null);
-
-    verify(mockRawKeyValues, times(1)).groupByKey();
-    verify(mockRawKeyValues, never()).groupByKey(any(Partitioner.class));

Review Comment:
   Why did we need to get rid of these 2 tests?



##########
runners/spark/src/main/java/org/apache/beam/runners/spark/translation/GroupCombineFunctions.java:
##########
@@ -49,18 +53,36 @@ public static <K, V> JavaRDD<KV<K, Iterable<WindowedValue<V>>>> groupByKeyOnly(
       @Nullable Partitioner partitioner) {
     // we use coders to convert objects in the PCollection to byte arrays, so they
     // can be transferred over the network for the shuffle.
-    JavaPairRDD<ByteArray, byte[]> pairRDD =
-        rdd.map(new ReifyTimestampsAndWindowsFunction<>())
-            .mapToPair(TranslationUtils.toPairFunction())
-            .mapToPair(CoderHelpers.toByteFunction(keyCoder, wvCoder));
-
-    // If no partitioner is passed, the default group by key operation is called
-    JavaPairRDD<ByteArray, Iterable<byte[]>> groupedRDD =
-        (partitioner != null) ? pairRDD.groupByKey(partitioner) : pairRDD.groupByKey();
-
-    return groupedRDD
-        .mapToPair(CoderHelpers.fromByteFunctionIterable(keyCoder, wvCoder))
-        .map(new TranslationUtils.FromPairFunction<>());
+    final JavaPairRDD<ByteArray, byte[]> pairRDD =
+        rdd.mapPartitionsToPair(

Review Comment:
   Is there an advantage to doing this in a single mapPartitionsToPair function (as opposed to the original implementation)? It is a little harder to read/reuses a little less code
   
   The same question applies to the mapping after the combineByKey



-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

Re: [PR] Batch optimized SparkRunner groupByKey [beam]

Posted by "github-actions[bot] (via GitHub)" <gi...@apache.org>.
github-actions[bot] commented on PR #33322:
URL: https://github.com/apache/beam/pull/33322#issuecomment-2526092428

   Checks are failing. Will not request review until checks are succeeding. If you'd like to override that behavior, comment `assign set of reviewers`


-- 
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: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org