[GitHub] DickJC123 commented on issue #9560: Expand gpu-kernel-launch synchronous error checking.

[GitBox <gi...@apache.org>]

DickJC123 commented on issue #9560: Expand gpu-kernel-launch synchronous error checking.
URL: https://github.com/apache/incubator-mxnet/pull/9560#issuecomment-361466348
 
 
   The main python thread is protected by the try-catch created by the API_BEGIN() and API_END() macros in src/c_api_common.h.  Any exceptions thrown in separate worker threads are not caught by this try-catch, so those exceptions cause terminate() to be called which kills the CI.  One thing I considered was setting up MXNET_ENGINE_TYPE=NaiveEngine which performs all actions on the one python thread.  The problem with this approach is that the Engine is a singleton and the environment variable is only checked when the first test of the CI calls for an engine. Once again, one needs a separate process to get a fresh engine singleton for a particular test.
   
   So I've today finished a bit of polishing on the way to create the separate test process.  I've confirmed that the test runs properly under both Win and Linux Python3 CI environments, and is appropriately skipped for Python2 environments, where the needed fork-exec support is missing.
   
   I'm satisfied with the PR as it stands now, and hope it will be considered for merging.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services