Skip to content

Dashboard

Responsibility

User/admin UI backend: proxies orchestration, scheduler, and storage operations.

Inbound Authentication

  • user session/JWT context
  • admin-only guards for sensitive operations

Outbound Authentication

  • forwards user/admin headers to orchestrator/scheduler/storage
  • uses dedicated scheduler/orchestrator reload tokens for admin reload buttons

Key Config

  • identity URLs and portal scope
  • scheduler/orchestrator reload token env vars
  • CAPABILITY_REGISTRY_URL — AMQP trigger test, S3 admin, Postgres snapshots, and /api/diag resolve provider URLs from the registry under service_name (http.service, rabbitmq.service, s3.service, postgres.service, …), not short aliases or EGRESS_* env. Platform URLs (orchestrator, scheduler, gateway, storage, identity, tempo, loki, capability_registry) stay on their own env.
  • VARIABLES_SERVICE_URL — base URL for the variables capability provider (used by GET /api/runtime/variables-list/{root_run_id} to show run-scoped variable rows on the Run detail panel; defaults to http://localhost:8094)

Health

  • GET /healthz
  • GET /readyz

Cluster Health panel

Admin UI Services → Cluster Health aggregates live probes for discovered services.

Discovery

Sources (merged):

  1. Capability registry — every registered provider with a service_url appears automatically
  2. Builtin platform/infra — dashboard, capability-registry, rabbitmq-trigger-subscriber, loki, tempo (when URLs are configured)
  3. Extras — CLUSTER_HEALTH_EXTRA_SERVICES_JSON for pods that do not register

Registry metadata (metadata.cluster_health)

On the service registration payload:

Field Default Meaning
check_version true Compare /healthz version to the K8s deploy pin
required false When true, a down service marks cluster health unhealthy
health_path /healthz Probe path
ready_path /readyz Probe path

Example for an external provider that should be visible but not fail on MeshFlows build pins:

{
  "service_name": "acme.provider",
  "service_url": "http://acme:8080",
  "metadata": {
    "cluster_health": {
      "check_version": false,
      "required": false
    }
  }
}

Extras without registry

Dashboard env:

CLUSTER_HEALTH_EXTRA_SERVICES_JSON='[{"name":"acme.provider","url":"http://acme:8080","check_version":false}]'

Optional: CLUSTER_HEALTH_REQUIRED_SERVICES=orchestrator,gateway,... (comma-separated names that must be present and healthy).

Version match notes

  • Expected pin comes from the Kubernetes image tag when the service is listed in CLUSTER_HEALTH_DEPLOYMENT_MAP (dashboard code).
  • Without a map entry there is no K8s image pin; version match falls back or is meaningless — for external providers prefer check_version: false.
  • Only required: true (metadata or CLUSTER_HEALTH_REQUIRED_SERVICES) makes the cluster unhealthy when that provider is down.
  • Short rule: register for visibility; use check_version: false for providers that are not on the MeshFlows build/product pin.

Related: Capability provider deploy integration, Capability provider contract.

Operations panel

Dedicated sidebar section (same level as Workflows / Triggers). Not part of Services, and not the designer tree.

  • Permission: meshflows:operations:view for the screen (also implied by meshflows:platform:admin / meshflows:*). Operators get this by default.
  • All instances and the Operations tree are limited to workflows the user may run (meshflows:workflow:run:* or meshflows:workflow:run:<name>). Platform admin sees all.
  • Users with only operations:view (no run scopes) see an empty ledger. Users with only these ops scopes do not see Workflows/Triggers, Runtime Assets, Services, or Secrets.
  • Flow folders in the Operations tree come from the MeshFlows deployment (group on workflow YAML)
  • UI loads GET /api/runs (dashboard proxy, operations-gated) from orchestrator Postgres run ledger
  • Default time window is the last 1 day (since_minutes=1440). Presets: 30 minutes, 1 hour, 3 hours, 1 day. Custom from–to uses local browser time and sends since/until as UTC ISO
  • Selecting a flow in the Operations tree filters instances; All instances shows every flow; Open goes to the existing run detail
  • Expand a row for ledger steps plus subsequent flows via GET /api/runs/{run_id}/related
  • Extra columns come from seeded flows/dashboards/*.yaml (context.<key> on the list row). Chain milestones (type: milestone) roll up on the runtime-root parent as context.milestone (latest milestone_at in the tree)
  • Depends on orchestrator ORCH_MESSAGE_DATABASE_DSN; returns 503 / empty state when unset

Logic Apps Import (analysis + scaffolding preview)

Admin UI Platform → Logic Apps Import analyzes Azure Logic Apps Standard workflow.json or Consumption ARM templates (Microsoft.Logic/workflows → properties.definition) and returns a gap/mapping report plus MeshFlows YAML scaffolding stubs (download / preview; create-in-cluster not enabled yet).

  • POST /api/tools/logicapps/analyze — admin-gated; body is the document or { "workflow": {…}, "source_name": "…" }; response includes input_kind, optional arm metadata, and scaffolding preview files
  • POST /api/tools/logicapps/scaffolding/download — same body; returns ZIP (workflows/*.yaml, triggers/*.yaml, placeholders, import-report.json)
  • Peek-lock / Invictus ack → ordinary AMQP trigger; former complete/ack → type: log
  • ARM [parameters] / [concat] remain unresolved placeholders in scaffolding

See Logic Apps import.

Workflow Validation Integration

Dashboard is the UI backend for workflow authoring and should surface orchestrator validation diagnostics unchanged.

Current behavior:

  • dashboard proxies reload and workflow-management operations
  • POST /api/runtime/validate-workflow forwards YAML to orchestrator POST /admin/validate (pass-through JSON)

Response body matches orchestrator POST /admin/validate (typically valid, errors with Pydantic locations, etc.).