carlos / platform Public

startup-largebox-test-281

Human

Test only, #281 ask 2, at 6662ee8c (review rounds 1 and 2 applied). End-to-end startup on a 300-instance box with the sweep, layout recovery and hold restoration running for real (real instancectl.Server kept cold through its Systemctl seam, real journals and hold markers, edge in production TLS mode, agent as a child process). The gate probe answers in ~40-120 ms while the sweep waits. Requests, sidecar requests, worker wakes and scheduled deliveries wait at the latch and complete after it opens. Held hosts are fenced before Start and never started. Mid-move hosts get 503, never the stale socket. Overrun gives the 30 s wake-timeout 503. Shutdown mid-startup joins in under 1 ms with no goroutine leak. TestRunAgentStartupWiring checks that runAgent, activatorStart.Run, startActivator and recoverLayoutAtStart take each step on their own goroutine and in order, and that nothing before the edge waits on startup. Port retries only on the agent's own bind failure, never on a missed deadline. ~31 s wall; -count=5 green under nice. Not merged.

Commits 4
Files changed 1 +1794 −0
Discussions 0 0 open
Tasks 0/0 done

Discussions

  • No discussions on this branch yet.

Tasks

  • No tasks yet.

Activity