| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
Automate ERD refresh for release branches | 2 个月前 | |
release/v0.7.67 (#693) * release: v0.7.67 Security release: upgrade to fleetbase/core-api v1.6.67, which generates API keys from the CSPRNG (keys created in the same second were identical across organizations). composer update without -W, so only fleetbase/core-api moves in the lock; packages/core-api gitlink at v1.6.67. * perf: turn Xdebug off in the images; benchmark with concurrent users The v0.7.67 PR benchmark stalled: every request hit k6's 30s timeout (75% errors, 9.6 req/s), so the noisy tenant was never answered at all, let alone throttled, and the isolation checks failed as a side effect. - docker/Dockerfile: Xdebug is installed in the base stage for coverage and ran in its default develop mode in every image, production included. On the dev stack a GET /v1/orders took about twice as long with it (10.2s vs 4.8s median under emulation), and while loaded it keeps the OPcache JIT off. Default XDEBUG_MODE=off; coverage runs set XDEBUG_MODE=coverage explicitly. - k6 throughput now ramps concurrent virtual users (THROUGHPUT_MODE=vus, THROUGHPUT_PEAK_VUS=16) instead of a fixed 50 req/s arrival rate. An open arrival rate above what the runner serves grows the queue without bound until everything times out, and the backlog swamped the noisy-neighbour phase that followed. THROUGHPUT_MODE=rate keeps the old model for hosts of known capacity. - NOISY_RATE defaults to 4 req/s (240/min, double the 120/min limit). - RELEASE.md notes the Xdebug change. | 28 天前 |