JNI Set RMM_LOG_LEVEL and RMM_LOG_ACTIVE_LEVEL to allow setting log level at compile time - #19435
Conversation
…vel at compile time Signed-off-by: Alessandro Bellina <abellina@nvidia.com>
abellina
added
Spark
| endif() | ||
|
|
||
| # Set RMM logging level | ||
| set(RMM_LOGGING_LEVEL |
There was a problem hiding this comment.
There's a related conversation here about what should be precompiled in RMM here: rapidsai/rmm#1983 (comment)
It feels like these two topics might be solvable in similar ways.
There was a problem hiding this comment.
Assuming rmm is still being from source as part of the build, this will work, yes. It won't have any effect if you are using precompiled binaries, but since Spark doesn't do that this seems fine. The actual logs observed at runtime may still require setting the runtime log level if the logging functions are used.
|
/build |
|
/ok to test |
@abellina, there was an error processing your request: See the following link for more information: https://docs.gha-runners.nvidia.com/cpr/e/1/ |
|
/ok to test df05035 |
|
/merge |
7 participants
We have confusion from the customer side when they look at logs and see ERRORs from our pinned memory pool, since it is using a pooled_memory_resource that when exhausted we handle gracefully (here's a report NVIDIA/cudf-spark#13131):
The problem stems from us setting RMM_LOGGING_LEVEL in our cmake builds, but this is not getting propagated to RMM itself, somehow. I have been able to make this propagate, and silence the error @revans2 posted, but I am not 100% sure this is the right thing to do. Asking folks here for some help (@bdice) to check the CMakeLists change.
Description
Checklist