Hello! The following patch series introduces conditional rearm of write timeouts (and also does some related cleanup).
Notably, this might be beneficial on Linux under memory pressure, when requested send buffers cannot be allocated and writev() returns EAGAIN without any progress, yet write events are reported by kernel, resulting in a busy loop. And, more importantly, making existing write timeout handling ineffective. Suggested change is to only rearm write timeouts if some progress was made, in most cases detected by comparing c->sent with the original value as saved at the start of the event handler. Comments are welcome. -- Maxim Dounin
