If the pod can't connect to the DB couldn't that be a possible indicator that the coder pod is broken (perhaps broken networking)?

If the pod can't connect to the DB couldn't that be a possible indicator that the coder pod is broken (perhaps broken networking)?

Wouldn't coderd eventually crash anyway if the DB became unavailable?

Update: looks like it doesn't, but the DB being unavailable isn't necessarily something that restarting the pod will fix. How about we instead use /healthz for liveness and /api/v2/buildinfo for readiness? That way, coderd will simply no longer have traffic routed to it if it can't connect to the database.

I don't think using buildinfo over healthz really has any benefits. They're both initialized at essentially the same time during startup, and the only difference seems to be is that buildinfo is behind the API ratelimiter.

👍🏽

There is an expectation that pinging a health-check endpoint is essentially free, both in terms of latency and downstream resource costs.

The API ratelimiter is enough of a reason to prefer healthz IMO.