So sorry—I forgot to disable tcp_checks. Just redeployed with those disabled too
Ok, I think I know what’s happening.
Even outside of the configured health checks, we also have “passive” health checks. If our proxy can’t connect to your app when you make a request (such as when it’s restarting), this marks it as unhealthy for a few seconds. 10s to be precise, that matches what you’re seeing.
This is meant to detect struggling instances and route around them.
I think I can tweak this.
Awesome! I was going to say, to have it be consistently 10.something seconds seemed a bit suspicious…
I don’t know how tall an order this would be, but it would be especially cool if the tweak could be user-applied, in case of needing again in a different app
I just changed something in our proxy. Can you let me know if it’s working better?
I only deployed the change on the server where your instance currently is, so don’t deploy or it might fall on a different server 
Hmm, unfortunately it’s the same 