Graphsync Transfer Failures

[whyrusleeping]

I'm noticing a lot of graphsync transfer failures for a number of different reasons (known and unknown).

• The first issue is that yamux will periodically time out and close the whole connection, this is actually a bug and should be resolved in feat: improve ping accuracy libp2p/go-yamux#35
• data transfers appear to fail rather often when restarted with 'message too large'
• data transfer doesnt appear to automatically restart anything (we probably should, some limited number of times)
• counting of data sent and received appears to be incorrect around restarts
• data transfers appear to fail due to inactivity somewhat often, for inexplicable reasons
• data transfers get removed due to some timeout, but it appears to be different between client and server. I think I had a transfer get removed by the miner side while the client was still transferring data

[jennijuju]

defined tasks are now created in https://github.com/filecoin-project/lotus/milestone/8