Verification
Experimental: verification increases confidence; it does not certify production readiness.
scripts/verify-fleet.sh runs layered checks from the repository root.
Start with the read-only hub audit:
./scripts/verify-fleet.sh --layer 1 --suite hubLayers
Section titled “Layers”| Layer | Purpose | Safety |
|---|---|---|
1 |
Read-only audit and health checks | Safe after deployment |
2 |
Synthetic create, verify, and cleanup | Changes test resources |
3 |
Rebuild workflow followed by checks | Destructive; approved maintenance window only |
Registered suite names include hub, core, edge, backup, tenancy (console auth / RBAC), llm, apps, regression, and all. Legacy names (nested, catalog) may appear in --help; use apps for chart-library checks.
App and chart smoke tests
Section titled “App and chart smoke tests”| Script | Purpose |
|---|---|
verify-fleet-app.sh |
whoami install + pod probe (FLEET_APP_CLUSTER) |
verify-fleet-chart.sh |
Hub library lists seeded charts |
verify-fleet-llama-app-lifecycle.sh |
llama-cpp install, PATCH, LLM backend sync (FLEET_LLM_CLUSTER on GPU Ready cluster) |
verify-fleet-llama-profile.sh |
Profile CRUD, quick/extended verify, hub-proxied chat (check id llm.profile-verify) |
verify-fleet-llm.sh |
Hub LiteLLM enabled + smoke |
verify-litellm-routing.sh |
Multi-spoke load balance + session sticky (x-litellm-model-api-base) |
verify-llm-inference-e2e.sh |
End-to-end LLM API through console auth |
See FleetApp, fleet-llm, and Fleet Llama profiles.
Cluster reinstall presets
Section titled “Cluster reinstall presets”Cluster reinstall is an L2 operation. Built-in preset names: lab-sn01, lab-3n01, lab-ha01 (homelab profile in the check registry). Example:
./scripts/verify-fleet.sh --layer 2 --cluster-preset lab-sn01 --cluster-preset-onlyA reinstall can delete cluster state; confirm the target and recovery plan first.
For HA presets, hub finalize Jobs allow up to 90 minutes. Default flow polling is 90 × 20s (30 minutes). Override fleet_flow_poll_retries (for example 270 for 90 minutes) for HA flows.
Reports land under reports/verify-<layer>-<timestamp>/. Protect reports if they contain operational identifiers.
Use layers 2 and 3 only with a reviewed target and recovery plan. See scripts/fleet-verify/README.md for the check registry.