Overview
Liveness probes are similar to startup probes, except that they run after the startup probe has completed successfully. Typically, Liveness probes are used to evaluate whether the application running in your container is in a healthy state. If a container’s liveness probe reaches the failure state, the container will be reallocated to a different node. Liveness probes are a good place to check for unrecoverable application and GPU runtime errors, including CUDA errors in NVIDIA workloads or ROCm/HIP errors in AMD workloads.Successful Liveness Probe
Failed Liveness Probe
Configuration
Check the checkbox to enable the liveness probe. Currently, the supported protocols areexec, gRPC, TCP, and
HTTP/1.X
