[GLUTEN-6960][VL] Limit Velox untracked global memory manager's usage #6988
Conversation
This file contains 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
|
Run Gluten Clickhouse CI |
|
Run Gluten Clickhouse CI |
|
Run Gluten Clickhouse CI |
|
Run Gluten Clickhouse CI |
|
Run Gluten Clickhouse CI |
|
CH failure doesn't seem to be related cc @zzcclp https://opencicd.kyligence.com/job/gluten/job/gluten-ci/11749 |
|
Run Gluten Clickhouse CI |
marin-ma
approved these changes
Aug 26, 2024
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Outdated suggestions cannot be applied.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.
Limit Velox's global memory manager's usage to 0.75 * Spark overhead memory by default. The overhead memory is calculated by same manner with vanilla Spark.
Velox's global memory manager is used for some global allocations that don't belong to a specific query or task. These allocations are not tracked by Spark task memory manager.
After the patch,
spark.memory.offHeap.sizeandspark.executor.memoryOverhead(withspark.executor.memoryOverheadFactor/spark.executor.minMemoryOverheadfor higher version Spark, probably) will co-work for Velox backend's memory management.spark.memory.offHeap.sizelimits major of the memory allocations happened during Velox query execution, e.g., hash tables, sort buffers, shuffle buffers, etc.spark.executor.memoryOverheadlimits the memory allocations that are hardly tracked by Spark task memory manager, or that don't belong to a specific query. E.g., allocations happen during spilling, or possibly the global cache size (unimplemented).Edit: As the change fails a couple of CI tests, we are setting the internal Velox global poll size limit to
Long.MaxValueby default in the PR, to bypass the check for sometime, until the relevant bugs from Velox get fixed.