Skupper Model Provider — Operational Lifecycle

Skupper Model Provider — Operational Lifecycle A lifecycle diagram generated by Archify. 01 / Operational phases 02 / Recovery 03 / Terminal Clean · no infrastructure · Operational phases · entry 01 Clean no infrastructure entry Setup · 6-phase idempotent · Operational phases · one-time 02 Setup 6-phase idempotent one-time Ready (Down) · VAN stopped · Operational phases · idle 03 Ready (Down) VAN stopped idle VAN + Models Up · up.sh + hosted-model-ctl · Operational phases · starting 04 VAN + Models Up up.sh + hosted-model-ctl starting Running · serving inference · Operational phases · live 05 Running serving inference live Degraded · partial availability · Recovery · partial Degraded partial availability partial Crashed · router/controller down · Recovery · auto-fix Crashed router/controller down auto-fix Teardown · remove all infra · Terminal · terminal Teardown remove all infra terminal up.sh down.sh systemd 10s teardown.sh rebuild Legend start active state waiting terminal success failure / exit

Daily Operations

  • • up.sh starts controllers then routers across all sites
  • • hosted-model-ctl starts GPU containers on providers
  • • down.sh reverses: models, routers, controllers

Partial Availability

  • • Unreachable hosts are skipped, not fatal
  • • Models only started on reachable providers
  • • Combined status report: up/skipped/down per site

Auto-Recovery

  • • systemd start-watch.sh detects crash
  • • Restart=on-failure within 10 seconds
  • • No manual intervention for transient failures