Add task-based async with persistent event loop#11
Closed
Conversation
Implement socketpair-based FD reactor model as alternative to NIF-based WSGI/ASGI marshalling. The new backend_mode option allows choosing between: - nif: Traditional NIF-based marshalling (default, unchanged) - fd_reactor: Socketpair + reactor model with streaming body support New components: - priv/hornbeam_http/: HTTP parser package extracted from gunicorn with PROXY protocol v1/v2 support - priv/hornbeam_reactor_http.py: HTTP protocol for FD reactor - src/hornbeam_proxy_protocol.erl: PROXY v2 binary encoder - src/hornbeam_socketpair.erl: Unix socketpair creation - src/hornbeam_reactor_pool.erl: Reactor context pool management - src/hornbeam_proxy_bridge.erl: Request/response relay via socketpair Modified hornbeam_handler.erl to route requests based on backend_mode. HTTP/2 requests are translated to HTTP/1.1 for the reactor path. Includes test suite and benchmarking framework for comparing modes.
The py:bind/1 function doesn't exist in current erlang_python. Use py:contexts_started() and py:context() instead.
High-performance HTTP parser using picohttpparser library: - Pure C Python extension (no Cython required) - SIMD optimizations (SSE4.2/AVX2 on x86, NEON on ARM) - 12.7x faster than pure Python parser - ~1.7M requests/sec parsing throughput Includes build_environ() and build_asgi_scope() helpers for WSGI/ASGI integration. Build: cd priv/hornbeam_http_fast && python setup.py build_ext --inplace
Add pico_parser_fast.c with zero-copy optimizations: - Returns HttpRequest object with lazy attribute access - Avoids dict/string allocations until accessed - Achieves 8M+ req/s parsing throughput (50x faster than Python) Update hornbeam_reactor_http.py to use fast parser: - Prefer pico_parser_fast, fallback to pico_parser, then Python - Refactor HTTPProtocol to use parsed_request dict format - Implement fast body reading for Content-Length and chunked - Use dict-based build_environ and build_asgi_scope - Full WSGI cycle: 269k req/s, ASGI cycle: similar performance Update setup.py to build pico_parser_fast extension.
- Reuse shared event loop instead of creating new one per request - Add ASGIState class with __slots__ for efficient state management - Move receive/send coroutines to module level to reduce allocations - ASGI performance: 18k -> 27k req/s (50% improvement) - Remaining overhead (~33µs) is inherent to run_until_complete()
When running inside Erlang VM, use erlang.new_event_loop() instead of asyncio.new_event_loop(). The Erlang-backed event loop provides: - Sub-millisecond latency via enif_select - Zero polling (event-driven) - GIL release during I/O waits Falls back to standard asyncio when running outside Erlang VM.
Supports async_pending for non-blocking task submission.
Tasks signal completion via erlang.send(reactor_pid, {write_ready, fd}).
- Update hornbeam.erl to use py:contexts_started() - Update hornbeam_handler.erl to use py:call (automatic routing) - Update hornbeam_lifespan.erl to use py:call with timeout option - Update hornbeam_socketpair.erl to use py_nif:fd_close - Add local variable caching in HTTP protocol hot paths - Use memoryview for zero-copy buffer writes - Pre-allocate ASGI disconnect message
- Add StreamingASGISender with flow control (max pending chunks) - Add StreamingASGIReceive for streaming request body from Erlang - Add run_asgi_streaming() entry point for message-based streaming - Add StreamingResponse and run_wsgi_streaming() for WSGI - Add handle_asgi_streaming/handle_wsgi_streaming to handler - Add stream_body_loop for forwarding chunks to client - Add handle_streaming to proxy bridge with stream_response - Remove PROXY v2 overhead, use plain HTTP/1.1 - Optimize header map construction with maps:from_list - Optimize write_all with non-blocking retry loop - Fix fd_reactor_SUITE test setup and assertions - Fix fd_reactor_benchmark application startup and formatting
benoitc
force-pushed
the
feature/task-based-asgi
branch
from
d379f3f to
4621bfe
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary