My main goal is to avoid a big delay in the health status updating to "Critical" when I can predict my service will go down. I would combine this with the http health check already in place.
Considered solutions:
I have tried a TTL check, but this introduces the burden of converting the service to constantly send it's current status.
Using a TTL check with a really high ttl + sending "healthy" once on restart came to mind, but if this initial request fails the service stays unhealthy way too long.
Reducing the interval of my http health-check would mitigate the problem a bit, but also creates a bigger overhead.