Monitoring UI (detail)

Re-verified 2026-09-16 against DashboardGrid, ClusterBento, and the monitor/message-flow API clients.

The dashboard is the landing page; the Browser Terminal is one tab among many.

Current composition:

1. Cluster plane — one full-width Cluster card. Its bento cells cover lifecycle/readiness, live activity, resource pulse, topology, recent runtime, and the read-only Capacity Gate decision preview. Job details live on the dedicated BLAST Jobs/Results routes rather than in a standalone dashboard card.
2. Message Flow — an optional compact strip between Cluster and Resource planes. It renders only while the Service Bus integration is effective-enabled and opens the full producer → request queue → worker → completion flow.
3. Resource plane — ACR, Storage, and preview-gated Browser Terminal cards. Storage reports the private-network posture and database readiness; ACR compares repository tags with IMAGE_TAGS; Terminal probes the loopback sidecar path.
4. Sidecar runtime — one card for all six containers in ca-elb-dashboard, including CPU, memory, restart count, health, and recent HTTP activity. It is hidden on narrow mobile layouts.

The global refresh control offers Live, Slow, and Manual modes. Live uses SSE for sidecar and job invalidation with polling fallbacks; the optional Message Flow strip independently polls every 4-5 seconds while active or idle. Do not describe every dashboard surface as a fixed 30-second poll.

All numbers must come from real Azure / Kubernetes APIs. Never fabricate or cache stale data without showing a last-refreshed timestamp or a clear degraded state.