call to send_datagram_wait can block indefinitely

[uniquefine]

I came across an edgecase where send_datagram_wait blocks indefinitely during a benchmark.

It happens when the path MTU changes and oversized packets are dropped.

Dropping packets in the outgoing queue should unblock callers of send_datagram_wait that are waiting for Event::DatagramsUnblocked.

If no datagram can be sent e.g. because all packets in the outgoing queue were too large for max_payload then the callers will wait forever.

[uniquefine]

Would it make sense to just emit a DatagramUnblocked event here?

[Ralith]

Good catch, thanks! I think your proposed solution makes sense.

[uniquefine]

Okay, I'll create PR 👍

[uniquefine]

Fix is here #2462

Can you give me some pointers for how to create a test for this scenario?

[Ralith]

In quinn-proto/tests/mod.rs, there's a number of tests that manipulate Pair::mtu. Something similar could work here; queue a bunch of large datagrams, drop the MTU, and ensure that some are still queued when the MTU change is detected.