[Question] How to implement readinessProbe for custom serving runtime.

[haiminh2001]

Hi, I am using Modelmesh with Tritonserver. I encountered a critical issue: Triton crashed loop back but mm kept receiving traffic and failed.

In the serving runtime yaml, I see no readinessProbe define for triton container, which might be the reason of the bug.

I have some questions:

• Not implementing readinessProbe is unintended or on purpose ?
• Even if the readinessProbe fails, can it prevent the server from receiving traffic since as far as I know, modelmesh use headless service, then it send requests directly to the IP of the failed pod?
• If my Triton server fails, what should I do?