When running in mlserver multiprocessing the logging doesn't propagate upstream correctly, for example if an exception happens in the tempo code, only the stack trace of the mlserver code is shown. We should explore a way to provide a logging and connect the underlying handlers to simplify the reporting of issues, whilst still providing granularity/modularity when multiple model classes are running in a single mlserver instance. This issue may be better (or part of it) in mlserver codebase.
When running in mlserver multiprocessing the logging doesn't propagate upstream correctly, for example if an exception happens in the tempo code, only the stack trace of the mlserver code is shown. We should explore a way to provide a logging and connect the underlying handlers to simplify the reporting of issues, whilst still providing granularity/modularity when multiple model classes are running in a single mlserver instance. This issue may be better (or part of it) in mlserver codebase.