Azure ML Components and a mysterious segmentation fault error
Here’s yet another evidence of poor error handling and user-friendly error messages from Azure. The other day I tried to execute a custom Azure ML pipeline component, which failed with the following error:
Execution failed. User process 'python' terminated by signal with name SIGSEGV. There is a segmentation fault possibly caused by an out of memory error. Please check log file 'user_logs/std_log.txt' for error details.
As expected, std_log.txt was empty, and there weren’t any other clues about what was going.
Turns out I was trying to run a component that used onnxruntime-gpu on an instance that didn’t have a GPU.