baudflow

Overview

Baudflow is a self-hosted network monitoring platform built with Phoenix LiveView, Oban, and PostgreSQL. It runs automated Ookla speed tests and continuous TCP-connect pings on a schedule, streams every run in real time, and turns the results into health verdicts, SLA compliance, and alerts.

A speed test on a schedule tells you a number. Baudflow tells you whether your connection is healthy, when it broke, how badly, and whether your provider is delivering what you pay for — and it keeps the full raw record so the answer is auditable, not anecdotal.

Design principles

  • Raw data is sacred; derived views are cheap. Every test’s complete result is stored. Health verdicts, SLA percentages, and heatmaps are derived on the fly, so changing your thresholds or mode re-derives every past view instantly. Nothing is backfilled, nothing is lost.
  • Behaviours, not branches. Test backends and notification channels are behaviour implementations. Adding a channel or a backend is a self-contained module, not a new fork in a worker.
  • One concern per stage. Scheduling, running, evaluating, and notifying each live in their own context, wired together by Oban jobs rather than reaching into each other’s internals.

What a bare speed test leaves out

A scheduled speed test gives you one number per run. Baudflow closes the gaps that number alone leaves open:

  • Adaptive cadence. An hourly test gives a single point per hour — a blur during an outage and wasted bandwidth the rest of the time. A breached schedule runs faster until it recovers, then slows back down, giving you a detailed outage timeline without paying for it in bandwidth.
  • First-class ping. Latency, jitter, and loss degrade before throughput moves. A continuous TCP-connect ping runner with its own live console catches what a speed test misses; most outages show up here first.
  • SLA + raw retention. “It feels slow” is not a support ticket. Baudflow stores every result in full and rolls it into SLA percentages, breach timelines, and heatmaps — the kind of record a provider actually has to answer for.
  • A real alert pipeline. Threshold checks bolted onto a cron fire on every re-check. Baudflow’s event → policy → template → channel pipeline gates a breach to fire exactly once, and also notifies on recovery and on outright failure.

Feature catalog

Real-time visibility

Baudflow streams progress over PubSub as each test runs — no spinner, polling, or waiting for a result to land.

  • Live CRT oscilloscope — Ookla NDJSON streamed line-by-line, with phase tracking and phosphor afterglow as the waveform builds.
  • Live ping diagnostics — a dedicated /ping page streams each TCP-connect sample: cumulative average, jitter, low/high, and loss, updating live.
  • Manual runs + history — trigger a test instantly or let the scheduler run. Every run is logged with success, failure, or timeout status.
  • Per-result drill-down — ping, jitter, packet loss, server, ISP. Every measurement is fully inspectable, with the raw JSON preserved.

Health intelligence

Health is derived just-in-time from thresholds, never stored, so changing your mode re-derives every view instantly.

  • Adaptive cadence — a breached schedule switches to a faster escalated cron until it recovers, capturing outage detail without wasted bandwidth.
  • Auto / absolute / off thresholds — Auto judges against your own rolling baseline (zero-config); Absolute uses fixed Mbps/ms; Off disables verdicts.
  • Breach streaks & transitions — healthy → breach → recovered transitions, tracked atomically per schedule and surfaced to alerts and escalation.
  • SLA compliance — the percentage of tests that met your promised speeds: “delivered 91.3% of the month” instead of a bare Mbps number.
  • Health heatmap — a GitHub-style calendar wall revealing evening congestion, weekend slowdowns, and recurring outages at a glance.

Alerting

Event → policy → template → channel. Each layer is one concern; the policy decides whether to notify before any rendering work is done.

  • Four-layer pipeline — a pure policy module, a per-channel template renderer, and a channel behaviour. Adding a channel is one module.
  • Breach-streak gating — a breach fires exactly once, when the streak first reaches your threshold, not on every re-check. Less noise.
  • Recovery & failure alerts — get notified when service recovers and when a run outright fails, not only on degradation.
  • ntfy — push to a self-hosted or hosted ntfy.sh topic, wired through application env.
  • Webhooks + templates — POST JSON to Home Assistant, Grafana, n8n, or a custom endpoint, with an editable EEx body. A blank URL disables it.

Integrations

Standard interfaces, plain text, no lock-in. Baudflow slots into the observability setup you already run.

  • Prometheus /metrics — eight baudflow_* gauges in hand-rolled text format, NaN for stale values. Scrape it like any exporter.
  • /health endpoint — a lightweight liveness check for Uptime Kuma or a Kubernetes probe.
  • Embeddable heatmap — a chrome-less /heatmap/embed route for iframing into Home Assistant or Grafana dashboards.
  • Raw JSON retention — the full result is stored on every measurement; derived views are cheap, raw data is sacred. Pruned by retention.
  • Multi-arch Docker — a non-root, Ookla-bundled image for amd64 and arm64. One container, one Postgres.

Next steps