# Dash — uptime observatory A self-hosted uptime monitoring dashboard in a single binary. A Go check engine with SQLite persistence streams live status over SSE into a hand-built React frontend: three bespoke SVG circular visualizations with spring physics, full RTL (Farsi) support, and a shareable public status mode. ![Dash](docs/dashboard.png) ## What it does - **Five check types** — HTTP (status + latency), TCP port, TLS certificate expiry (days remaining), ICMP ping, and Docker (container running + healthcheck status via the daemon socket). Each target runs on its own interval. - **Flap detection** — N consecutive failures before a target is marked down, M consecutive successes before it recovers. No alert noise from a single dropped packet. - **Incident grouping** — one incident spans the entire down period, from the first failed check of the streak until recovery. - **Live stream** — `/api/stream` pushes every probe and every confirmed state transition over Server-Sent Events. The UI never polls. - **Every result persisted** — timestamp, latency, status, error string in SQLite (pure Go driver, no CGO), pruned after `history_days`. ## The frontend Three circular instruments, all hand-built SVG + Framer Motion (no charting libraries): 1. **Status rings** — one per target. The outer arc fills to the uptime percentage over the selected window (24H/7D) and springs to every new value; the inner brass gauge encodes latency on a log scale (1ms–2.5s); each completed probe lands as a visible pulse; confirmed state changes ripple outward. Click a ring and it travels into a detail view (shared layout transition) with a latency sparkline, a per-target day dial, and recent incidents. 2. **Orbital overview** — every target orbits the system core at Kepler-flavored speeds (inner orbits run faster). Healthy bodies drift calmly with motion trails, degraded ones breathe, a down body breaks its orbit outward and flares red over the scar of its old path. The core carries fleet-wide uptime and a brass scanner hand. 3. **Day dial** — a 24-hour clock face. Each of the last 24 hours is one segment at its true wall-clock position, colored by status. Hours without data stay hairline-thin. A brass hand marks now. One glance, one full day. Plus: a boot sequence where the instruments draw themselves in, staggered; `prefers-reduced-motion` collapses all motion to fully-formed static rendering; **فارسی / RTL** is a first-class layout mode (mirrored layout, Vazirmatn type, Persian digits and the Persian solar calendar via `fa-IR`); **dark and light themes** (persisted, both tuned variants of the same brass / ink design language); and a **Status** mode — the same instruments stripped to a clean, shareable public status page. Continuous motion (the orbital field, the scanner hand) runs outside React entirely — one animation-frame loop mutates SVG attributes directly, so the page stays at full frame rate regardless of target count. No SVG filters, no blend modes on hot layers. ## Build Requires Go ≥ 1.22 and Node ≥ 20. The frontend must be built first because the binary embeds it via `go:embed`. ```sh # 1. build the frontend into web/dist cd web && npm install && npm run build && cd .. # 2. compile the single binary (web/dist is embedded) go build -o dash . ``` The result is one self-contained file, `dash`. Copy it and a `targets.yaml` to any Linux/macOS/Windows box — nothing else to install. ## Run ```sh ./dash # uses ./targets.yaml, ./dash.db, :8080 ./dash -config /etc/dash/targets.yaml -db /var/lib/dash.db -listen :9090 ``` Open http://localhost:8080. > **ICMP checks** need raw-socket privileges. Run as root, or grant the > binary `CAP_NET_RAW` (`setcap cap_net_raw+ep ./dash`), or allow > unprivileged ping (`sysctl net.ipv4.ping_group_range="0 2147483647"`). > Other check types work without any privileges. ## Configuration — targets.yaml ```yaml settings: listen: ":8080" # bind address db: "dash.db" # sqlite file history_days: 30 # raw check retention targets: - id: example-web # unique slug (required) name: "Example Website" type: http # http | tcp | tls | icmp url: https://example.com # http → url, tcp/tls → address, icmp → host interval: 15s # time between checks (default 30s) timeout: 5s # per-check timeout (default 5s) fail_threshold: 3 # consecutive fails → down (default 3) recover_threshold: 2 # consecutive oks → up (default 2) expect_status: 200 # http only: require exact status degraded_latency_ms: 1500 # mark degraded above this latency - id: github-tls name: "GitHub TLS" type: tls address: github.com:443 interval: 5m warn_days: 21 # tls only: degraded when cert expires sooner - id: my-container name: "My Container" type: docker container: my-container # docker container name interval: 30s # socket: /var/run/docker.sock (default) ``` > **Docker checks** read the daemon socket. Run the binary as a user in the > `docker` group (or `SupplementaryGroups=docker` in the systemd unit). A > container is *up* when running, *degraded* while its healthcheck reports > `starting`, and failing when stopped or `unhealthy`. ## API | Endpoint | Description | | --- | --- | | `GET /api/targets` | Current state of all targets: flap-filtered status, last check, 24h/7d uptime, latency percentiles, open incident | | `GET /api/targets/{id}/history?range=24h` | Time series: downsampled points plus 24 aggregate buckets (`range`: `30m`, `24h`, `7d`, …) | | `GET /api/incidents?range=7d` | Grouped incidents, ongoing first | | `GET /api/stream` | SSE: `snapshot` on connect, then `check` per probe and `state` per confirmed transition | ## Development ```sh go run . & # backend on :8080 cd web && npm run dev # vite on :5173, proxies /api → :8080 ``` ## Layout ``` main.go, webfs.go entrypoint + embedded frontend internal/config targets.yaml parsing, defaults, validation internal/checker the four probe implementations internal/engine scheduler, flap detection, incidents, SSE hub internal/store sqlite persistence (modernc.org/sqlite, no CGO) internal/api REST + SSE handlers web/ React + TypeScript + Vite + Tailwind v4 + Motion ```