1 Commits
Author SHA1 Message Date
Krystie 86cc21c7a4 [glm-grade=B+] ops: self-healing watchdog for DNS2 (container, port 3001, Caddyfile, caddy)
30s systemd timer heals four real failure classes: container down (docker
start -> start.sh fallback), rogue host process on :3001, Caddyfile wiped
by foreign generators (known-good snapshot + size/site-block/marker
gates), caddy down/not serving. Telegram alerts with 15-min per-class
cooldown; stamp only burned on successful send. Adversarial review B-
(both blockers fixed: snapshot poisoning via multi-gate integrity +
refresh lockout, deployment). Live kill-tested twice: full recovery in
one cycle, alerts delivered, cooldown verified.
2026-08-14 23:59:50 -07:00