0.9.2
Patch release preventing large buffered HTTP/1 responses from truncating under downstream backpressure
Highlights
- Complete buffered responses under backpressure — sync workers now honor partial
sendmsg()results and resume at the exact unsent header or body offset. - Production-driven regression proof — deterministic partial-write tests are paired with a real slow-reader socket transfer whose response exceeds the constrained kernel send buffer.
Upgrade notes
- This is a backward-compatible patch release with no configuration migration.
- Applications serving large buffered HTTP/1 responses through a slow client or proxy should upgrade from 0.9.1.
- The normal full-write path keeps one
sendmsg()call and the existing response buffers. The completion fallback runs only when the socket accepts a prefix.
Fixed
- Prevent the sync worker from silently dropping the unsent suffix of a large buffered response when
socket.sendmsg()returns a legal partial-write byte count. (#312)
Install
Core HTTP/1.1 server:
uv add "bengal-pounce==0.9.2"
All optional protocol dependencies:
uv add "bengal-pounce[full]==0.9.2"