There are extremely rare scenarios when shutting down some activity workers does make sense. These are ones where deploying workers is part of the workflow. For example to perform some CPU intensive computation.
In all other scenarios workers are expected to be running permanently as they are shared across many workflows.
The reason Im asking is Im seeing high qps to temporal servers from various worker services, not sure if this is a burden on the server so we should conserver resources.
This is not expected as workers use long poll to receive tasks. So no high request load is expected while system is idle. Could you provide more stats on the load?