improve http performance and ws readability

CHANGELOG.md

@@ -15,10 +15,6 @@
 #### Added
 
 * Add support for reporting peer client information
-* Speed up parsing and serialization of requests and responses with
-  [zerocopy](https://crates.io/crates/zerocopy)
-* Store torrents with up to two peers without an extra heap allocation for the
-  peers.
 
 #### Changed
 
@@ -28,10 +24,16 @@
 * Remove support for unbounded worker channels
 * Add backpressure in socket workers. They will postpone reading from the
   socket if sending a request to a swarm worker failed
+* Avoid a heap allocation for torrents with two or less peers. This can save
+  a lot of memory if many torrents are tracked
+* Improve announce performance by avoiding having to filter response peers
+* In announce response statistics, don't include announcing peer
 * Distribute announce responses from swarm workers over socket workers to
   decrease performance loss due to underutilized threads
 * Harden ConnectionValidator to make IP spoofing even more costly
 * Remove config key `network.poll_event_capacity` (always use 1)
+* Speed up parsing and serialization of requests and responses by using
+  [zerocopy](https://crates.io/crates/zerocopy)
 
 ### aquatic_http
 
@@ -41,6 +43,13 @@
 * Support running without TLS
 * Support running behind reverse proxy
 
+#### Changed
+
+* Index peers by packet source IP and provided port instead of by source ip
+  and peer id. This is likely slightly faster.
+* Improve announce performance by avoiding having to filter response peers
+* In announce response statistics, don't include announcing peer
+
 #### Fixed
 
 * Fix bug where clean up after closing connections wasn't always done

TODO.md

@@ -2,22 +2,21 @@
 
 ## High priority
 
-* if peer_clients is on, add task to generate prometheus exports on regular
-  interval to clean up data
+* general
+  * add task to generate prometheus exports on regular interval to clean up
+    data. this is important if peer_clients is activated
+
+* http
+  * consider storing small number of peers without extra heap allocation
+  * add CI transfer test for http without TLS
 
 * aquatic_bench
   * Opentracker "slow to get up to speed", is it due to getting faster once
     inserts are rarely needed since most ip-port combinations have been sent?
     In that case, a shorter duration (e.g., 30 seconds) would be a good idea.
-  * Maybe investigate aquatic memory use.
-    * Would it use significantly less memory to store peers in an ArrayVec if
-      there are only, say, 2 of them?
 
-* CI transfer test
-  * add HTTP without TLS
-
-* http
-  * panic sentinel not working
+* general
+  * panic sentinel not working? at least seemingly not in http?
 
 ## Medium priority
 
@@ -42,10 +41,6 @@
 
 * aquatic_ws
   * Add cleaning task for ConnectionHandle.announced_info_hashes?
-  * 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 ?
 
 * Performance hyperoptimization (receive interrupts on correct core)
   * If there is no network card RSS support, do eBPF XDP CpuMap redirect based on packet info, to
@@ -63,7 +58,6 @@
     * thiserror?
     * CI
     * uring load test?
-  * what poll event capacity is actually needed?
   * load test
       * move additional request sending to for each received response, maybe
         with probability 0.2