Avoid concatentating multiple host buffers when reading Parquet #11911
+144
−152
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
Signed-off-by: Jason Lowe <[email protected]>
jlowe
added
the
cudf_dependency
abellina
reviewed
Jan 3, 2025
| @@ -2123,7 +2123,7 @@ class MultiFileCloudOrcPartitionReader( | |||
| } | |||
| } catch { | |||
| case e: Throwable => | |||
| hostBuffers.foreach(_.hmb.safeClose()) | |||
| hostBuffers.foreach(_.hmbs.foreach(_.safeClose())) | |||
There was a problem hiding this comment.
Not technically part of this PR, but couldn't safeClose throw and then we don't see e? We should probably also use the Seq variant of safeClose(e)
| @@ -2737,7 +2713,7 @@ class MultiFileCloudParquetPartitionReader( | |||
| } | |||
| } catch { | |||
| case e: Throwable => | |||
| hostBuffers.foreach(_.hmb.safeClose()) | |||
| hostBuffers.foreach(_.hmbs.safeClose()) | |||
There was a problem hiding this comment.
same here, could we use safeClose(e)?
sql-plugin/src/main/scala/com/nvidia/spark/rapids/GpuParquetScan.scala
Outdated
Show resolved
Hide resolved
revans2
previously approved these changes
Jan 6, 2025
|
I was able to run this in one of our performance cluster. I ran the MULTITHREADED reader, and I compared with and without the patch. The benchmark is ~2% faster, though it doesn't detect a significant improvement there, and one query was 40% faster. q9 is on average 11% better, but it's deemed not a significant change. |
abellina
previously approved these changes
Jan 7, 2025
|
I'll resolve conflicts. |
|
build |
revans2
previously approved these changes
Jan 8, 2025
|
build |
revans2
previously approved these changes
Jan 8, 2025
|
Seeing a segfault here all around avro. |
Fix issue in GpuAvroScan.readBatches
|
ok I reproed an issue locally, and fixed it here: 675fe5f Will re-run CI |
|
build |
revans2
approved these changes
Jan 14, 2025
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.
Depends upon rapidsai/cudf#17673.
Updates the multithreaded Parquet reader to leverage the new multiple host buffers reader interface. This removes the need to concatenate multiple host buffers into a single buffer before decoding the data via the GPU. This also makes it easier to accept "late arrivals" in the multithreaded combine reader after waking up with the GPU semaphore, since we only need to fabricate a new footer to accommodate the additional buffers in the read.