yangjj-iso opened a new pull request, #7665: URL: https://github.com/apache/shenyu/pull/7665
Refs #7649 and #7650. Add an opt-in `minimal-http` reactor profile and an independently packageable `shenyu-bootstrap-http` distribution using the existing Global, Divide, URI, HTTP Client and Response starters, with WebSocket configuration synchronization. The default full distribution and integration ownership remain unchanged. Include foreground Windows/Unix launchers, Docker recipes, dependency/packaged-JAR inventories, inherited LICENSE/NOTICE coverage and capability-extension instructions. Add packaged routing smoke tests and a disposable real H2 Admin test covering authenticated REST rule updates, deletion/recreation, selector deletion and successful/failed HTTP proxying. A dedicated JDK 17 workflow builds and checks the opt-in assembly without feeding its inactive module into ordinary CI module selection. ### Local validation - 35-module minimal `clean install` and 163-module full/Admin release packaging passed, including test compilation, Checkstyle and RAT. Java test execution was skipped for these packaging builds. - Independent `-f shenyu-bootstrap-http/pom.xml package` and resolved runtime inventory passed. - Five Python harness tests and five CI routing cases passed. - Both packaged assemblies passed the WebSocket protocol routing smoke; incomplete initial synchronization kept readiness unhealthy while liveness remained healthy. - Six real Admin/gateway checks passed using disposable H2, a generated JWT key and an authenticated WebSocket connection. - Three final same-host runs per assembly, with fixed-arrival-rate k6 traffic, direct-upstream controls, matching JVM/configuration/CPU affinity and JFR capture. All 200/500/1000 RPS points passed with zero response errors and dropped iterations. The local Windows/Temurin 21 comparison measured 108 vs 384 packaged JARs, 43.74 vs 128.45 MiB compressed archives, median sync-plus-proxy readiness of 3.906 vs 5.531 seconds, and median idle working set of 250.3 vs 363.8 MiB. These are local observations, not a general QPS improvement claim; 1000 RPS was the highest tested rate and did not reach saturation. Recorded submillisecond latency medians have a measurement floor on this host. Commands, per-run results, complete JAR inventory CSVs and raw k6/JFR evidence hashes are in `docs/performance/minimal-http.md` and `docs/performance/results/minimal-http-2026-10-15/`. Large raw samples/profiles are retained locally and bundled separately, rather than committed to the repository. ### Remaining acceptance evidence — keep draft Docker Desktop did not provide a working daemon after a launch attempt, so image byte sizes remain unmeasured. True OS-cold-cache readiness, a saturation-bracketing sweep under container quotas and the independent JDK 17 CI run remain to be completed. First-observed/warm-filesystem timings are explicitly distinguished from controlled cold-cache timings. This PR does not close #7649 while these acceptance items remain outstanding. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
