Support running Pandas UDFs on GPUs in Python processes. #640
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
Fixing an EOFExcetpion by creating a new file object on the same socket as output of the python worker process.
- Add a new object `PythonWorkerSemaphore`. - Add a new conf `spark.rapids.python.concurrentPythonWorkers` - Change class GpuSemaphore from `private` to `private[rapids]` - Let GpuSemaphore support not to initialize the GPU
Currently the limitation only works when pool memory is enabled. - Separate the configs for Python. - Add `OptionalConfEntry` for Python configs
including 5 types SQL_MAP_PANDAS_ITER_UDF SQL_GROUPED_AGG_PANDAS_UDF SQL_GROUPED_MAP_PANDAS_UDF SQL_SCALAR_PANDAS_ITER_UDF SQL_SCALAR_PANDAS_UDF by 4 physical plans GpuMapInPandasExec GpuAggregateInPandasExec GpuFlatMapGroupsInPandasExec GpuArrowEvalPythonExec along with a API update in python/worker.
These two types are SQL_COGROUPED_MAP_PANDAS_UDF and SQL_WINDOW_AGG_PANDAS_UDF, mapping to GpuFlatMapCoGroupsInPandasExec and GPuWindowInPandasExec respectively.
python Signed-off-by: Robert (Bobby) Evans <[email protected]>
Fixing an EOFExcetpion by creating a new file object on the same socket as output of the python worker process.
- Add a new object `PythonWorkerSemaphore`. - Add a new conf `spark.rapids.python.concurrentPythonWorkers` - Change class GpuSemaphore from `private` to `private[rapids]` - Let GpuSemaphore support not to initialize the GPU
Currently the limitation only works when pool memory is enabled. - Separate the configs for Python. - Add `OptionalConfEntry` for Python configs
including 5 types SQL_MAP_PANDAS_ITER_UDF SQL_GROUPED_AGG_PANDAS_UDF SQL_GROUPED_MAP_PANDAS_UDF SQL_SCALAR_PANDAS_ITER_UDF SQL_SCALAR_PANDAS_UDF by 4 physical plans GpuMapInPandasExec GpuAggregateInPandasExec GpuFlatMapGroupsInPandasExec GpuArrowEvalPythonExec along with a API update in python/worker.
These two types are SQL_COGROUPED_MAP_PANDAS_UDF and SQL_WINDOW_AGG_PANDAS_UDF, mapping to GpuFlatMapCoGroupsInPandasExec and GPuWindowInPandasExec respectively.
firestarman
requested review from
GaryShen2008,
jlowe,
NvTimLiu and
revans2
as code owners
Also always check python module configs
revans2
previously approved these changes
Sep 10, 2020
|
build |
|
build |
|
build |
revans2
approved these changes
Sep 10, 2020
|
This looks good to me if you don't have any other things you want to get in feel free to merge it. |
pxLi
reviewed
Sep 11, 2020
nartal1
pushed a commit
to nartal1/spark-rapids
that referenced
this pull request
Jun 9, 2021
Add support to run Pandas UDFs on GPUs, mainly consisting of two things: Overriding all the 6 related plans to build GPU context of device and memory for Python processes. Introducing 2 new python modules rapids.worker and rapids.daemon to execute the GPU memory initialization by leveraging RMM Python APIs. Signed-off-by: Firestarman <[email protected]> Co-authored-by: Liangcai Li <[email protected]> Co-authored-by: Robert (Bobby) Evans <[email protected]> Co-authored-by: shotai <[email protected]>
nartal1
pushed a commit
to nartal1/spark-rapids
that referenced
this pull request
Jun 9, 2021
Add support to run Pandas UDFs on GPUs, mainly consisting of two things: Overriding all the 6 related plans to build GPU context of device and memory for Python processes. Introducing 2 new python modules rapids.worker and rapids.daemon to execute the GPU memory initialization by leveraging RMM Python APIs. Signed-off-by: Firestarman <[email protected]> Co-authored-by: Liangcai Li <[email protected]> Co-authored-by: Robert (Bobby) Evans <[email protected]> Co-authored-by: shotai <[email protected]>
tgravescs
pushed a commit
to tgravescs/spark-rapids
that referenced
this pull request
Nov 30, 2023
Signed-off-by: Peixin Li <[email protected]> Signed-off-by: Peixin Li <[email protected]>
tgravescs
pushed a commit
to tgravescs/spark-rapids
that referenced
this pull request
Nov 30, 2023
* Update submodule cudf to f817d96d8bdc47da9fb2725d0e5a7b18586a29ee (NVIDIA#635) Signed-off-by: spark-rapids automation <[email protected]> Signed-off-by: spark-rapids automation <[email protected]> * Fixing empty columns when casting to integer or decimal crashing (NVIDIA#633) * fixing empty columns Signed-off-by: Mike Wilson <[email protected]> * cudf submodule commit to v22.10.00 (NVIDIA#640) Signed-off-by: Peixin Li <[email protected]> Signed-off-by: Peixin Li <[email protected]> * try use new token to fix automerge permission * verify automerge fix of Token permission (NVIDIA#643) Signed-off-by: Peixin Li <[email protected]> Signed-off-by: Peixin Li <[email protected]> * Revert not working automerge fix [skip ci] (NVIDIA#644) * Revert "verify automerge fix of Token permission (NVIDIA#643)" This reverts commit 8261117. * Revert "try use new token to fix automerge permission" This reverts commit 2a9acde. Signed-off-by: Peixin Li <[email protected]> Signed-off-by: Peixin Li <[email protected]> * Auto-merge use submodule in BASE ref Signed-off-by: Peixin Li <[email protected]> Signed-off-by: spark-rapids automation <[email protected]> Signed-off-by: Mike Wilson <[email protected]> Signed-off-by: Peixin Li <[email protected]> Co-authored-by: Jenkins Automation <[email protected]> Co-authored-by: Mike Wilson <[email protected]>
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.
This PR is to add support to run Pandas UDFs on GPUs, mainly consisting of two things:
rapids.workerandrapids.daemonto execute the GPU memory initialization by leveraging RMM Python APIs.