Direct data access

The read-only rule constrains the browser, not you. The Postgres on the box is a normal Postgres, and querying it directly is a supported feature — your data, no export ceremony.

On the Pi#

psql -U guppi -h /var/run/postgresql guppi

The installer trusts the local guppi role on the Unix socket, so no password is needed. (There's no guppi OS user — the hub runs as whoever installed it.)

The tables you care about:

Table / view	What's in it
telemetry_points	start here — one row per sample: recorded_at (timestamp), value, path (e.g. psu1.1.voltage), unit, rig_id (plus device_id, text_value for string signals, and a few more)
telemetry	the compact frame storage behind that view — one row per frame with samples packed into arrays, addressed by channel_id. Query telemetry_points unless you know you want frames
telemetry_channels	one row per signal: channel_id ↔ rig_id + device_id + path + unit
test_executions	every test run: status, timestamps, full result document (result_json)
artifacts	captured waveforms: metadata + storage_path; bytes live under /var/lib/guppi/artifacts
rigs	paired rigs

telemetry / telemetry_points are the real samples — every admitted sample, exactly as recorded. You may also notice telemetry_rollup_* tables: those are pre-computed min/max/mean summaries the charts use to stay fast at wide zooms. For analysis, query the raw tables — never the rollups — so nothing you read was averaged or decimated.

Example — one signal, last hour, as CSV:

psql -U guppi -h /var/run/postgresql guppi -c "\copy (
  SELECT recorded_at, value FROM telemetry_points
  WHERE path = 'psu1.1.voltage' AND recorded_at > now() - interval '1 hour'
  ORDER BY recorded_at
) TO '/tmp/volt.csv' CSV HEADER"

From another machine#

Two options:

• HTTP (no setup): PostgREST already serves read-only JSON on port 3010 — curl 'http://bench.local:3010/telemetry_points?path=eq.psu1.1.voltage&limit=100'. Handy for scripts and notebooks; capped at 10 000 rows per request.
• SQL (opt-in): Postgres listens only on the local socket by default. To open it to your LAN, edit postgresql.conf (listen_addresses) and pg_hba.conf yourself — standard Postgres administration, at your own discretion on your own network.

Notebooks / Grafana#

Anything that speaks Postgres or HTTP works. pandas via PostgREST:

import pandas as pd
df = pd.read_json(
    "http://bench.local:3010/telemetry_points"
    "?path=eq.psu1.1.voltage&order=recorded_at.desc&limit=10000"
)

Waveform artifacts are canonical little-endian float32 — read a capture with numpy.fromfile(path, dtype="<f4") using the storage_path from the artifacts row.

One rule#

Treat direct access as read-only. The hub owns writes; inserting or mutating rows underneath it can confuse ingest and the UI. (Deleting old telemetry rows is fine — there's no automatic sweeper, so this is how you reclaim space; the Storage tab does the same thing through the UI.)