amadan
Wake budget: a configurable health window, and keep-warm Feature Request open
The platform gives a waking instance 30 seconds to become healthy. On 2026-09-12, two of the three restarts of ips.bdn.oncarlos.com failed that check and succeeded on the next attempt (wake failed: ... instance not healthy within 30s, then instance woke in 15.014s). One release was out of service for about a minute, the other for nearly two.
Most of that is ours to fix: a boot that misses 30 seconds is too slow. But two platform knobs would make the failure mode kinder:
- A per-app health window, so an app with a known long first boot (a migration, a large template parse) isn't killed and restarted in the middle of it.
- Keep-warm: an option on
instances create/set-*to never hibernate an instance, or to keep N warm. For an app whose first reader otherwise pays wake plus boot plus a cold screen, it's the difference between a slow first click and a normal one.
Related: #6 (timing at the platform boundary is invisible to an app).
--- Filed from the ips repository (Turnkey's insolvency practice system, built on rastrillo v0.26.0 and deployed on CARLOS), where this was worked around. Paths without a repository prefix are in ips.