Cuda.deviceSynchronize as a last resort if we cannot spill enough #6849
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: Alessandro Bellina <[email protected]>
abellina
added
cudf_dependency
|
This is going to fail to build until we get cuDF changes in. |
tgravescs
reviewed
Oct 19, 2022
| logInfo(s"Device allocation of $allocSize bytes failed, device store has " + | ||
| s"$storeSize bytes. Total RMM allocated is ${Rmm.getTotalBytesAllocated} bytes.") | ||
| s"$storeSize bytes. $attemptMsg" + |
There was a problem hiding this comment.
Should be good here: 6d10d77
jlowe
reviewed
Oct 19, 2022
sql-plugin/src/main/scala/com/nvidia/spark/rapids/DeviceMemoryEventHandler.scala
Outdated
Show resolved
Hide resolved
jlowe
reviewed
Oct 19, 2022
sql-plugin/src/main/scala/com/nvidia/spark/rapids/DeviceMemoryEventHandler.scala
Show resolved
Hide resolved
jlowe
approved these changes
Oct 19, 2022
|
build |
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Outdated suggestions cannot be applied.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.
Signed-off-by: Alessandro Bellina [email protected]
Closes #6768
This depends on: rapidsai/cudf#11940
This PR adds
Cuda.deviceSynchronizecalls when the device store is empty. The idea is that these calls should allow GPU work to finish (like pending frees) and allow the async pool to resolve any temporary state (make available blocks that were recently freed for other streams to consume).Adds an internal config
spark.rapids.memory.gpu.oomMaxRetries. It is set to 2 retries by default.