There is a couple issues in this test:

• The initial poll_node is not disconnected and not part of the quorum, which means this is an avalanche peer that will never respond to polls. This does not cause any failure but slows the test a bit and create false positive errors in the log.
• The stalling vote does not consume all the polls, and there can be a lot of them because stalling requires a lot of voting rounds. This can eventually cause the node to stop polling because too many polls are already in flight.