Re-running workflow after the flaky pubsub test.. Strange that it never fails locally. I suspect it happens because there are >1 threads being used but not enough time for all of them.

LGTM, but shouldn't the exit_daemon be called after all futures are either complete or dropped?

shouldn't the exit_daemon be called after all futures are either complete or dropped?

With the conformance tests for block/get called "should get a block added as CIDv0 with a CIDv1" it creates a block_get request which is we currently don't support finding, so it stays blocking forever, after the shutdown. The shutdown signal happens and the future resolves, but warp (or hyper) will wait for all spawned tasks to complete before exiting, which will never happen as the conformance test keeps the connection alive.

With the shutdown signal now forcing the subscription futures to complete the request will be dropped and warp will exit normally. It'll still take some time for the conformance tests to get an http client timeout (separate timeout from the 2s test timeout).