feat: default to AMPHP bidi streaming, optimized and multi-worker - #3
Merged
Conversation
Retire Swoole from the examples, README, and live demo (it stays as an alternative transport) and make AmpStreamingServer the default everywhere, then cut the bidi transport's per-request cost and scale it across cores. Migration to AMP as the default: - examples/endpoint.php + tracing.php + bin/restate-serve now serve over AmpStreamingServer (bidi HTTP/2 h2c); amp-endpoint.php folded into endpoint.php - docker-compose / Makefile / docker images: examples-endpoint runs the amp image; bidi verified live against Restate 1.5.2 (only V7 cancellation needs >=1.7) - README: AMP featured as the default, Swoole/PSR-15/Lambda listed as alternatives; dropped the shepherd.dev (Psalm) badge Transport optimization (greeter, 16-core host, Restate 1.5.2): - TCP_NODELAY on the server socket + per-slice frame coalescing in AmpStreamTransport: 495 -> ~660 req/s (+35%), p50 89 -> 63 ms (amphp leaves Nagle on, which collides with delayed-ACK for a ~40 ms per-invocation stall) - inline fast-path: a non-parking handler runs in the request fiber and returns one ReadableBuffer (no async task, no outbound Queue, zero extra event-loop hops); only a parked handler falls back to the streaming queue via the new SwitchableOutputSink. +15-22% single-worker, tighter tail - multi-worker: listen($host, $port, $workers) pre-forks N processes that share the port via SO_REUSEPORT, lifting the single-event-loop ceiling. 16 workers reach 1470 req/s @-c200 and 1709 @-c400 (beats single-worker Swoole's ~1228), and the single-process -c400 collapse (123 req/s) is gone Benchmarks: e2e harness parameterized by TRANSPORT (swoole|amp) with a compare target; new bench-endpoint-amp + a transport comparison in docs/BENCHMARKS.md. Tests: cover the inline fast-path (SwitchableOutputSink + StreamingInline: completed / parked-resolve / EOF-while-parked suspension). 509 unit tests green.
|
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
Codecov flagged 3 uncovered patch lines: the EOF-before-journal return in InvocationDriver::tryStartInline and the null vm/fiber guard in RequestProcessor::continueStreamingFromPark. Add two tests exercising both — an inline attempt that ends before the journal is ready (empty completion) and a mis-constructed parked result that must close the channel rather than dereference null.
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
2 participants
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.
No description provided.