rqbit/TODO.md

• when we have the whole torrent, there's no point talking to peers that also have the whole torrent and keep reconnecting to them.
• per-file stats
• [x (partial)] per-peer stats
• use some concurrent hashmap e.g. flurry or dashmap
• tracing instead of logging. Debugging peers: RUST_LOG=[{peer=.*}]=debug test-log for tests
• reopen read only is bugged
• initializing/checking
  • blocks the whole process. Need to break it up. On slower devices (rpi) just hangs for a good while
  • checking torrents should be visible right away
• server persistence
  • it would be nice to restart the server and keep the state
• torrent actions
  • pause/unpause
  • remove including from disk
• DHT
  • bootstrapping is lame
  • many nodes in "Unknown" status, do smth about it
  • for torrents with a few seeds might be cool to re-query DHT once in a while.
  • don't leak memory when deleting torrents (i.e. remove torrent information (seen peers etc) once the torrent is deleted)
  • Routing table - is it balanced properly?
  • Don't query Bad nodes
  • [-] Buckets that have not been changed in 15 minutes should be "refreshed." (per RFC)
    • Did it, but it's flawed: starts repeating the same queries again as neighboring refreshes don't know about the other ones, and DHT returns the same nodes again and again.
  • it's sending many requests now way too fast, locks up Mac OS UI annoyingly
  • store peers sent to us with "announce_peer"
  • announced peers should be persisted (partial)
  • clean up announced peer cache
    • only send a token to torrents really close to us
  • After the search is exhausted, the client then inserts the peer contact information for itself onto the responding nodes with IDs closest to the infohash of the torrent.
  • Ensure that if we query the "returned" nodes, they are even closer to our request than the responding node id was.

incoming peers:

• error managing peer: expected extended handshake, but got Bitfield(<94 bytes>)
• do not announce when merely listing the torrent

someday:

• cancellation from the client-side for the lib (i.e. stop the torrent manager)
• favicons for Web UI

desktop:

• on first run show options
• allow to change options later (even with a session restart)
• look at logs - allow writing them to files? Set RUST_LOG through API

persistence:

• store total uploaded bytes, so that on restart it comes back up

efficiency:

• once the torrent is completed, we don't need to remember old peers

refactor:

• session persistence: should add torrents even if we haven't resolved it yet

• where are peers stored

• http api pause/unpause etc

• when a live torrent fails writing to disk, it should transition to error state

• something is wrong when unpausing - can't finish. Recalculate needed/have from chunk tracker.

• silence this: WARN torrent{id=0}:external_peer_adder: librqbit::spawn_utils: finished with error: no longer live

• start from error state should be possible from UI

• checking is very slow on raspberry checked. nothing much can be done here. Even if raspberry's own libssl.so is used it's still super slow (sha1)

• .rqbit-session.json file has 0 bytes when disk full. I guess fs::rename does this when disk is full? at least on linux. Couldn't repro on MacOS

• reopen:

  • in general, the only time the file should be write-only, is when it's live and not yet fully downloaded
  • initializing: open read-only if file has all pieces
  • on piece validated open read-only all files that were copleted
  • would be nice to have some abstraction that walks files and their pieces
  • nit: optimize open write/read/write right away on first start
  • peers: if finished they are all paused forever, but if we change the list of files, we need to restart them

• opened_files: track HAVE progress

  • actually track
  • show in API and UI
  • refresh when downloading (it doesn't somehow)
  • on restart, this is not computed, compute