Business Context
During the transfer of a Credit Risk model from the training environment to the production CAS server, the file was corrupted (truncated). The pipeline must detect this invalid format to prevent the deployment of a broken model.
Data Preparation
Creation of a CAS table with random non-ONNX binary data to simulate a corrupted file.
Copied!
| 1 | DATA casuser.onnx_corrupted; |
| 2 | LENGTH model_blob varbinary; |
| 3 | /* Random garbage data that does not conform to ONNX spec */ |
| 4 | model_blob = 'FFFFFFFF000000...'x; |
| 5 | OUTPUT; |
| 6 | RUN; |
Étapes de réalisation
1
Attempt to validate the corrupted binary object.
Copied!
| 1 | PROC CAS; |
| 2 | /* We expect this to return a specific validation error in the result */ |
| 3 | astore.check RESULT=bad_res / |
| 4 | onnx={TABLE={caslib='casuser', name='onnx_corrupted'}}; |
| 5 | PRINT bad_res; |
| 6 | RUN; |
Expected Result
The action executes but returns a validation failure status in the results. The log or output table explicitly states that the binary data is not a valid ONNX format, confirming the system correctly catches bad inputs.