Dodgy trace in Diffusion testnet
bolt12 opened this issue · comments
I found a test failure in one of my PRs (https://ci.iog.io/build/1226040/nixlog/2) where looking at the logs I see that there's an async Hot->Cold demotion, followed by the node forgetting that peer, followed by that peer making it back into the known set and trying to be promoted again. During these steps, the cleanup connection handler didn't had chance to run and actually clean the connection leading to a TrConnectionExists
trace. Which makes me think that there should be some kind of synchronization between the cleanup handler and forgetting about the peer.