Parallelize segment staging copy to deep store - #19082
Open
Vamsi-klu wants to merge 1 commit into
Open
Conversation
Batch Spark/Hadoop generation jobs spent wall-clock time on a single-threaded PinotFS move of hundreds of segment tars from staging to deep store. Parallel copy with bounded concurrency cuts job latency without changing push semantics.
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #19082 +/- ##
=========================================
Coverage 65.47% 65.47%
+ Complexity 1421 1415 -6
=========================================
Files 3426 3426
Lines 217315 217384 +69
Branches 34509 34525 +16
=========================================
+ Hits 142283 142333 +50
- Misses 63513 63527 +14
- Partials 11519 11524 +5
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
Contributor
Author
|
Ready for review — all required CI checks are green on this PR. Issue: #7510 Could the following folks take a look when convenient? Formal GitHub "Request review" is unavailable from a fork contributor account on Thank you! |
Jackie-Jiang
added
enhancement
Contributor
There was a problem hiding this comment.
Pull request overview
This PR removes the remaining single-threaded bottleneck in batch ingestion segment staging by parallelizing the staging → deep-store “move” step, and wiring Spark-3 and Hadoop runners to use a configurable bounded parallelism.
Changes:
- Extend
SegmentGenerationJobUtils.moveFiles(...)with a bounded thread-pool implementation and astagingCopyParallelismextra-config parser (default + max cap). - Update Spark-3 and Hadoop segment generation runners to pass the resolved staging copy parallelism to
moveFiles(...). - Add unit tests validating layout preservation across parallelism levels, overwrite behavior, and failure propagation.
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.
| File | Description |
|---|---|
pinot-plugins/pinot-batch-ingestion/pinot-batch-ingestion-spark-3/.../SparkSegmentGenerationJobRunner.java |
Uses stagingCopyParallelism from extra configs and invokes parallel moveFiles(...) for staging → output move. |
pinot-plugins/pinot-batch-ingestion/pinot-batch-ingestion-hadoop/.../HadoopSegmentGenerationJobRunner.java |
Uses stagingCopyParallelism from extra configs and invokes parallel moveFiles(...) for staging → output move. |
pinot-plugins/pinot-batch-ingestion/pinot-batch-ingestion-common/.../SegmentGenerationJobUtilsTest.java |
Adds unit coverage for parallel move equivalence, overwrite semantics, and failure aggregation. |
pinot-plugins/pinot-batch-ingestion/pinot-batch-ingestion-common/.../SegmentGenerationJobUtils.java |
Implements bounded parallel move, adds config parsing, and refactors file listing/move into helpers. |
Comment on lines
+267
to
271
| if (!overwrite && fs.exists(destFileUri)) { | ||
| LOGGER.warn("Can't overwrite existing output segment tar file: {}", destFileUri); | ||
| } else { | ||
| fs.move(sourceFileUri, destFileUri, true); | ||
| } |
Comment on lines
+249
to
+259
| private static List<URI> listSourceFiles(PinotFS fs, URI sourceDir) | ||
| throws IOException, URISyntaxException { | ||
| List<URI> sourceFileUris = new ArrayList<>(); | ||
| for (String sourcePath : fs.listFiles(sourceDir, true)) { | ||
| URI sourceFileUri = SegmentGenerationUtils.getFileURI(sourcePath, sourceDir); | ||
| String sourceFilename = SegmentGenerationUtils.getFileName(sourceFileUri); | ||
| URI destFileUri = | ||
| SegmentGenerationUtils.getRelativeOutputPath(sourceDir, sourceFileUri, destDir).resolve(sourceFilename); | ||
|
|
||
| if (!overwrite && fs.exists(destFileUri)) { | ||
| LOGGER.warn("Can't overwrite existing output segment tar file: {}", destFileUri); | ||
| } else { | ||
| fs.move(sourceFileUri, destFileUri, true); | ||
| if (fs.isDirectory(sourceFileUri)) { | ||
| continue; | ||
| } | ||
| sourceFileUris.add(sourceFileUri); | ||
| } | ||
| return sourceFileUris; |
Comment on lines
+161
to
+163
| int effectiveParallelism = Math.max(1, Math.min(parallelism, sourceFileUris.size())); | ||
| LOGGER.info("Moving {} files from [{}] to [{}] with parallelism {}", sourceFileUris.size(), sourceDir, destDir, | ||
| effectiveParallelism); |
Comment on lines
+189
to
+197
| } catch (InterruptedException e) { | ||
| interrupted = true; | ||
| future.cancel(true); | ||
| if (firstFailure == null) { | ||
| firstFailure = new IOException("Interrupted while moving files from " + sourceDir + " to " + destDir, e); | ||
| } else { | ||
| firstFailure.addSuppressed(e); | ||
| } | ||
| } catch (Exception e) { |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
Batch Spark/Hadoop segment generation jobs were spending a large share of wall-clock time on a single-threaded
PinotFScopy/move of segment tars from the staging directory to deep store (S3/HDFS/…). For jobs that produce hundreds of segments, that serial hop dominated the SLA even after segment build and controller push were already parallelized.This is the remaining bottleneck called out in #7510 (staging→deep-store copy), not tar push (which already supports
pushParallelism).Impact
moveis a remote copy+delete.How
SegmentGenerationJobUtils.moveFileswith a bounded thread-pool parallel implementation and an optional parallelism argument.Test plan
SegmentGenerationJobUtilsTestcovering parallel vs serial layout equivalence, overwrite true/false, and failure aggregation (LocalPinotFS)../mvnw -pl pinot-plugins/pinot-batch-ingestion/pinot-batch-ingestion-common -am -Dtest=SegmentGenerationJobUtilsTest -Dsurefire.failIfNoSpecifiedTests=false testRelated
closes: #7510
Reviewers
Suggested: xiangfu0, kkrugler (issue discussion / batch ingestion)
Was generative AI tooling used to co-author this PR?
Generated-by: Grok Build (xAI)