Q: Can a stream cancel operation hang due to an unresponsive DNS query?

[lmcdasi]

I'm encountering an issue with an on-premises service where, intermittently, a DNS error occurs. This error repeats several times, and after a few retries, the service hangs.
It appears that a DNS query never completes, causing the InetAddress request to block. Since this happens during stream creation, the thread responsible for creating the session hits its deadline.
When the deadline is reached, the code attempts to cancel the stream. My question is: can a stream cancel operation be blocked if the underlying DNS resolution is still hanging or unresponsive?
Any insights into how gRPC handles stream cancellation in the presence of unresolved DNS queries would be appreciated.

[ejona86]

RPCs/Streams are completely separate from DNS and connection establishment. RPC threads are not used at all for those operations. If you cancel a stream, or its deadline expires, that happens near-instantly. It is not dependent on I/O. (At time there are some thread-hops, so there is a small processing delay, but no I/O.)

[lmcdasi]

Is it possible to inject a different Listener2 and have a custom onError ? Or any other suggestion(s) ? Thank you.

[ejona86]

And have it do what? Suggestion for what? If DNS doesn't complete, then gRPC can't make a connection.

[lmcdasi]

I understand that. But rather than reaching a state where the DNS hangs a VM I prefer to trigger a shutdown of that VM and avoid having a manual intervention.
The Listner onError would help me achieve this.

[ejona86]

Hangs a VM? The RPC should fail. If it isn't failing, you may just have to wait a bit longer for the DNS timeout. It sounds like you may want to know when the channel is unhealthy. You can use managedChannel.getState(false) to notice when it is TRANSIENT_FAILURE.