Manual validation of distributed runners on Toys cluster
Hands-on validation of distributed runners on the Toys cluster. This is the "bang on things" pass to build confidence before redeploying to Garden.
Prerequisites
The following issues were discovered during the initial Garden dogfooding (MIR-890) and need to be resolved first:
- MIR-920 — Fix runner API cert IP SANs (
app run/execbroken on runner sandboxes) - MIR-902 —
miren runner installcommand for clean provisioning - MIR-919 — Human-readable runner names
- MIR-917 —
miren runner removeto clean up stale nodes - MIR-918 — Runner version entity updates on restart
Testing checklist
- Fresh runner provisioning via
miren runner install+miren runner join - App deploy schedules sandboxes across coordinator and runner
-
miren app runandmiren execwork on runner-hosted sandboxes -
miren sandbox listshows runner names - Runner restart preserves version info
- Stale runner cleanup with
miren runner remove - Deploy rolls sandboxes correctly across nodes
- Network overlay (flannel) routes traffic between nodes
Environment
- Cluster: Toys (
gcloud compute ssh miren-toys --project=miren-development) - Local CLI connected to Toys for client-side testing