Hacker News Reader

Feels really strange to adjust the liveness checks in a way that potentially causes healthy instances to get thrown out (while still having a delay due to the polling interval) instead of just deterministically making it dead when you start shutting it down.

View on HN

Isn't that why it's step 2 of 3? It sounds like with the SIGTERM issue fixed it could be dialed back to its previous value. But it also sounds like they triaged reasonably, reducing the incidence to buy more time to debug the deeper cause.

A pod can be unready but still live, which is what should happen if its a temporary failure.

Reply on news.ycombinator.com