I logged a feature-request issue for this but wanted to try it in a discussion-format to see what others are doing.
To summarize, when rolling out normal Kubernetes deployments, it’s common practice to use a preStop hook and SIGTERM handling in the services. This also makes HPA work well.
With Istio and the local caching of endpoints, that same solution doesn’t apply. Blue/green does help solve the problem for deployments but HPA isn’t addressed by that since there’s no user orchestration to inject.
How are others using HPA on your own deployments with Istio in the environment? Is it standard practice to just use some Istio retry logic to mitigate these issues? Should we expect something more elegant down the road?