Bennett Lynch

@wilkinsona For those not able to upgrade to 3.x yet, can you please advise how to workaround this issue in 2.7? I'm encountering this issue with a setup like so:...

That seems to have done the trick. Thank you, @wilkinsona.

Hi @ramaraochavali. I am fairly certain from looking at the metrics. Example of requests received by pod with no warmup:  Example of requests received by pod with 5m Istio...

> Did you enable "PILOT_SEND_UNHEALTHY_ENDPOINTS" by any chance? No it's not enabled or explicitly configured.

Hi @ramaraochavali. Just curious if you were able to confirm that the current behavior is indeed based on container start time, and whether this is intentional or not?

Interesting. My metrics seem to tell a different story but I'll try experimenting with more prolonged durations to confirm the behavior.

I encountered what appears to be the same issue and wanted to share some extra data/anecdotes. The contention around `okhttp3.internal.concurrent.TaskRunner` and `okhttp3.internal.connection.RealConnection` caused my server worker threads (all using OkHttp...

Is it possible to remove the lock usage in favor of a single `AtomicInteger` which tracks outstanding calls requested? `startAnotherThread` logic could then use CAS semantics to attempt to decrement...

@bringyou Nice, good option to have in our back pocket. :) @swankjesse Understood on the need for a holistic approach. Few questions if you can spare the time: 1. If...

Thanks, @swankjesse. Do the different threads hold the lock during the entire connection attempt, or just until the thread starts processing? I see here that it will just attempt to...