[CLOSED] Retry logic for missing requested pieces

[feross]

Issue by fisch0920
Friday May 09, 2014 at 22:54 GMT
Originally opened as https://github.com/feross/bittorrent-client/issues/5

---

Messing around with small torrents, about 90% of the time the torrents download successfully and the client cleanly exits, but the rest of the time a piece or a few pieces are missing that were requested but never resolved (maybe because of the external server; maybe because of something we're doing that's only affecting a small percent of BT clients). When in this stalled state, the client keeps connecting to new peers, but no progress is ever made on finishing the torrent.

Here's an example log showing the bug in action. Pieces 3 and 5 are both requested but never responded to. Methinks the client/swarm needs better retry logic for missing pieces.

[feross]

Comment by feross
Monday May 12, 2014 at 08:01 GMT

---

Yeah, I think we just need to add a handler on the 'timeout' wire event and/or come up with a more aggressive strategy for requesting pieces.

[feross]

Comment by feross
Sunday May 18, 2014 at 07:30 GMT

---

Pretty sure this was fixed in #10. When a timeout happens, bittorrent-protocol should call the request callback with an error and then self.storage.cancelBlock gets called.

Let me know if I'm mistaken and I'll re-open.

[feross]

Comment by fisch0920
Sunday May 18, 2014 at 09:21 GMT

---

Yep; this should be fixed now.