Matt Joiner

Do you know for certain that the burst size is indeed sufficiently high? I think the problem was the same in #759.

@AskAlexSharov I'm not sure, I think we covered it thoroughly in the other issue. Either there's an issue propagating the correct value somewhere in anacrolix/torrent, or the wrong version or...

I don't currently have time to look into this in depth right now. Note that goroutines do not map to threads 1:1. Check the threads profile to see how many...

Yeah things that come to mind that could block, are related to C? Perhaps disabling uTP, piece completion using sqlite, any memory-mapping. But I'm pretty sure that none of those...

Thanks for resolving the issue more. I believe the open syscall is not reliably asynchronous on many platforms, so a dedicated thread may be necessary in the Go runtime. On...

The x.pe addresses should be attempted on all diallers, and should result in a connection attempt in any that can interpret the address. That should be both uTP and TCP...

This looks related to https://github.com/anacrolix/torrent/issues/712, which comes out of an issue with Windows reporting UDP warnings as errors. I believe it's a longstanding issue and I think the Go core...

How easily can you reproduce this? Is there a particular torrent or configuration that causes it? The invariants around tracking piece availability are quite complex, it would be great if...

Thanks. The piece relative availability is an optimization around picking pieces on rarity. There are several factors involved, so a refactor and unit tests might shake these out if we...

I think for that to occur, this line would be executed: https://github.com/anacrolix/torrent/blob/f1e56b4017cc6108b44cf2a7fdc15c9f4b271061/peerconn.go#L1433. Could you confirm you have a non-zero value when you get that error message? It's possible we should...