Add container_startup_health_check_timeout arg to Predictor.update_endpoint

[stevenpitts]

Describe the feature you'd like

Predictor's update_endpoint function should have a container_startup_health_check_timeout argument that gets passed to the new production variant.

Similar behavior can be seen in Session's create_endpoint_config,

How would this feature be used? Please describe.

Currently, I cannot call update_endpoint with a health check timeout. With this new argument, I could pass container_startup_health_check_timeout=60 and it would make the new production variant with ContainerStartupHealthCheckTimeoutInSeconds=60, so that a bad model deployment would fail after only one minute.

Whether container_startup_health_check_timeout triggers the creation of a new production deployment or it is only used when one of the other production variant fields is passed, I leave up to the implementer.

Describe alternatives you've considered

Instead of using update_endpoint, I could create the endpoint config manually and switch the endpoint over manually.

Additional context

There are a variety of other places where container_startup_health_check_timeout would be nice to have, but adding it to all those places would greatly increase the scope of work here.