# TODO

* readme
  * document privilege dropping and cpu pinning

* config
  * add flag to print parsed config when starting
  * fail on unrecognized keys

* quit whole program if any thread panics

* implement socket_recv_size and ipv6_only in glommio implementations

* newer glommio versions might use SIGUSR1 internally, see glommio fe33e30

* CI
  * file transfer CI for all implementations
  * test access lists?
  * cargo-deny

* aquatic_udp
  * look at proper cpu pinning (check that one thread gets bound per core)
    * then consider so_attach_reuseport_cbpf
  * what poll event capacity is actually needed?
  * stagger connection cleaning intervals?
  * load test
      * move additional request sending to for each received response, maybe
        with probability 0.2

* aquatic_http:
  * clean out connections regularly
    * Rc<RefCell<ValidUntil>> which get set on successful request parsing and
      successful response sending. Clone kept in connection slab which gets cleaned
      periodically (= cancel tasks). Means that task handle will need to be stored in slab.
      Config vars kill_idle_connections: bool, max_idle_connection_time. Remove keepalive.
    * handle panicked/cancelled tasks?
  * optimize?
    * get_peer_addr only once (takes 1.2% of runtime)
    * queue response: allocating takes 2.8% of runtime
  * use futures-rustls for load test
  * consider better error type for request parsing, so that better error
    messages can be sent back (e.g., "full scrapes are not supported")

* aquatic_ws
  * mio
    * shard torrent state. this could decrease dropped messages too, since
      request handlers won't send large batches of them
    * connection cleaning interval
    * use access list cache
    * use write event interest for handshakes too
    * deregistering before closing is required by mio, but it hurts performance
      * blocked on https://github.com/snapview/tungstenite-rs/issues/51
    * connection closing: send tls close message etc?
  * glommio
    * proper cpu set pinning
    * RES memory still high after traffic stops, even if torrent maps and connection slabs go down to 0 len and capacity
      * replacing indexmap_amortized / simd_json with equivalents doesn't help
    * SinkExt::send maybe doesn't wake up properly?
      * related to https://github.com/sdroege/async-tungstenite/blob/master/src/compat.rs#L18 ?
  * general
    * large amount of temporary allocations in serialize_20_bytes, pretty many in deserialize_20_bytes

# Less important

* extract response peers: extract "one extra" to compensate for removal,
  of sender if present in selection? maybe make criterion benchmark,
  optimize

## aquatic_http_load_test
* how handle large number of peers for "popular" torrents in keepalive mode?
  maybe it is ok

## aquatic_http
* test torrent transfer with real clients
  * scrape: does it work (serialization etc), and with multiple hashes?
  * 'left' optional in magnet requests? Probably not. Transmission sends huge
    positive number.
* compact=0 should result in error response

## aquatic_ws_load_test
* very small amount of connections means very small number of peers per
  torrents, so tracker handling of large number is not really assessed

# Not important

## aquatic_ws
* write new version of extract_response_peers which checks for equality with
  peer sending request? It could return an arrayvec or smallvec by the way
  (but then the size needs to be adjusted together with the corresponding
  config var, or the config var needs to be removed)

## aquatic_cli_helpers
* Include config field comments in exported toml (likely quite a bit of work)

# Don't do

* general: PGO didn't seem to help way back

## aquatic_http
* request from path:
  * deserialize 20 bytes: possibly rewrite (just check length of underlying
    bytes == 20 and then copy them), also maybe remove String from map for
    these cases too. doesn't really improve performance
  * crazy http parsing: check for newline with memchr, take slice until
    there. then iter over space newlines/just take relevant data. Not faster
    than httparse and a lot worse

## aquatic_udp_protocol
* Use `bytes` crate: seems to worsen performance somewhat
* Zerocopy (https://docs.rs/zerocopy/0.3.0/zerocopy/index.html) for requests
  and responses. Doesn't improve performance