Dashboard

The coordinator ships with an embedded monitoring dashboard — a single-page application served from the same HTTP server as /metrics and /statusz. No CDN, no external dependencies, no build step: the HTML, JavaScript (Alpine.js + Chart.js), and CSS are compiled into the Go binary via //go:embed. It works on-prem, in air-gapped networks, and behind VPNs.

Enabling the dashboard

Pass --metrics-addr to the coordinator (or the collapsed urutau run CLI). The dashboard is served on the same address:

# Single-process mode
urutau run -f pipeline.yaml --metrics-addr :9090

# Distributed mode (coordinator)
urutau-coordinator run -f pipeline.yaml --metrics-addr :9090

On Kubernetes, set spec.coordinator.metricsAddr in the CDCPipeline CR:

CDCPipeline CR with metrics

apiVersion: urutau.maltzsama.github.io/v1alpha1
kind: CDCPipeline
spec:
  coordinator:
    metricsAddr: ":9090"

Open http://localhost:9090 in a browser. The dashboard is the root page; /metrics (Prometheus) and /statusz (JSON) remain available at their usual paths.

Views

Overview

The landing page shows the pipeline at a glance:

• Pipeline name and run ID
• Status indicator: healthy / degraded / failed
• Source kind and sink type
• Workers: online count, snapshot progress
• Uptime since start
• Maintenance enabled/disabled

Streams (Tables)

One card per source → target stream. Each card shows:

Field	Meaning
Source / Target	schema.table → namespace.table
Write mode	upsert, append, or append-idempotent
Position	Current CDC position (with copy button)
Lag	Time since last commit
Rows	Total written + rate (rows/s)
Commits	Total + failures
Commit latency	Average ms per commit
Equality deletes	Count (upsert mode)
Snapshot progress	0..1 during initial backfill

Clicking a stream card opens a drawer with throughput and commit latency charts (1-hour client-side buffer, window selector for 5m/15m/1h/6h).

Workers

One card per connected worker:

Field	Meaning
Worker name	<pipeline>-<target>-<index>
Status	starting, snapshotting, streaming
Epoch	Reset count (bumped on supervisor reset)
Assigned tables	Tables this worker owns
Last ack	Seconds since last acknowledgement
Inflight bytes	Unacked batch bytes
Committed positions	Per-table last committed position

Events

A scrollable, color-coded log of operational events:

• Commit failure — catalog rejected a commit
• Worker reset — supervisor restarted a stale worker
• Schema change — DDL detected in source
• Maintenance — compaction/expiry/orphan pass completed
• Pipeline — start, stop, snapshot transitions

Filter by event type and worker. Events are kept in a 1000-entry ring buffer in memory.

Logs

The coordinator's structured log tail, with level filter (debug / info / warn / error). Useful for debugging without SSH-ing into the pod.

Actions

The dashboard exposes two actions, both requiring a confirmation dialog:

Action	What it does
Cancel pipeline	Sends Shutdown to all workers; terminates the pipeline gracefully
Restart worker	Bumps the worker's epoch, cancels its session, and triggers a supervisor reset

Actions are logged as events and visible in the Events view.

Real-time updates

The dashboard uses Server-Sent Events (SSE) for live updates. On connect, the server pushes a full snapshot (pipeline, tables, workers, events, logs). After that, only deltas are pushed — no polling, no WebSocket complexity. The browser's EventSource handles reconnection automatically.

The SSE endpoint is GET /api/v1/stream. A periodic comment (: ping) keeps idle proxies from timing out the connection.

API endpoints

The dashboard's JSON API is also available for programmatic access. See Dashboard API for the full reference.

Endpoint	Method	Description
/api/v1/pipeline	GET	Pipeline summary
/api/v1/tables	GET	All streams
/api/v1/tables/{name}	GET	One stream (by source or target name)
/api/v1/workers	GET	All workers
/api/v1/workers/{name}	GET	One worker
/api/v1/events	GET	Event log (filter: type, worker, limit)
/api/v1/logs	GET	Coordinator log tail (filter: level, limit)
/api/v1/stream	GET	SSE stream
/api/v1/actions/cancel	POST	Cancel pipeline
/api/v1/actions/restart/{worker}	POST	Restart worker
/healthz	GET	Liveness probe
/readyz	GET	Readiness probe

Security

The dashboard assumes a trusted network (same trust model as the gRPC control plane). There is no authentication in v0.2.0 — the endpoint is intended for internal monitoring, not public exposure.

For untrusted networks, put the dashboard behind a reverse proxy with authentication, or restrict access via network policies.

Architecture

The dashboard is a separate package (internal/dashboard) that reads coordinator state through the State interface — the same dependency direction the rest of the orchestration keeps. The coordinator implements State; the dashboard never imports the coordinator.

Dashboard architecture

coordinator.go
  └── dashboard.Handler
        ├── /api/v1/*  (JSON API)
        ├── /api/v1/stream  (SSE)
        ├── /healthz, /readyz  (probes)
        └── /*  (embedded SPA)

The SPA assets (index.html, app.js, style.css, plus vendored Alpine.js/Chart.js/Pico.css) are embedded at compile time via //go:embed static/*. No runtime filesystem access.