As engineers we would like to easily follow and reproduce the incoming requests (valid/invalid) and log them immediately after receiving them.

This will help following the tracebacks and understand which kinds of incoming requests are causing server errors, so we can then scale and upgrade our systems.

This ticket raised due to errors that occurred the previous days in the revertrisk-multiningual model server: https://phabricator.wikimedia.org/T409657

Goal

• Make the logging better
• Log immediately the input request before crashing - or - in the crash message.
• Upgrade error handling