mirror of
https://relay.ngit.dev/npub15qydau2hjma6ngxkl2cyar74wzyjshvl65za5k5rl69264ar2exs5cyejr/ngit-grasp.git
synced 2026-10-05 15:08:24 +00:00
Accepted sockets retained Nagle while the relay flushes individual EVENT and EOSE frames. A delayed-ACK peer could hold small response batches for about 40 ms, including when the immediate peer is a loopback proxy. Enable TCP_NODELAY before handing each accepted socket to Hyper. Reject only that connection if setting the option fails. Protocol framing, proxy configuration and client deadlines remain unchanged. This addresses a measured transport delay, not every reported production timeout. Retain opt-in delayed-ACK and concurrent LMDB read benchmarks for the HTTP/WebSocket path. The controlled delayed-ACK reproduction originally fell from 1.050 s to 15.17 ms for 25 two-event reads. Benchmarks have bounded completion deadlines and validate complete EVENT/EOSE responses; latency assertions remain opt-in because busy CI workers can distort timings. Validation: the original before/after benchmark passed; the rebuilt PR is validated again after restoring its released SDK dependency.