Skip to content

TODO.md

Latest commit

 

History

History
44 lines (33 loc) · 3.24 KB

TODO.md

File metadata and controls

44 lines (33 loc) · 3.24 KB
Raw

To do

Things to check or finish that need something this repo can't do on its own.

Check when PM2.5 is high

Does the sensor's pm02Compensated equal the cloud's pm02_corrected? Both read 0 on the day it was tried, so it is not known. The bot records pm02Compensated (preferring it to the plain pm02) as its PM2.5. If the two differ, a chart would show a step wherever the bot's own readings meet the cloud's.

When the air has real particles in it (smoke, incense, a dusty day), run these within a minute of each other, with TOKEN set to the AIRGRADIENT_API_TOKEN and the sensor's address in place of the name if .local does not resolve:

curl -s "https://api.airgradient.com/public/api/v1/locations/153523/measures/current?token=$TOKEN" | jq '{timestamp, pm02, pm02_corrected}'
curl -s http://airgradient_d83bda1c1e08.local/measures/current | jq '{pm02, pm02Compensated}'

Equal (or within a few percent): nothing to do. Different: compare over a longer stretch with the measures/past call (5-minute rows, stamped with the start of the interval) before deciding which to trust.

Try on the real machines

  • The sensor's own server: AIRGRADIENT_LOCAL_URL on a bot on the sensor's network. Check the log says "live readings from the sensor itself", that "air now" is fresh, and that a chart of today has no step where the bot's readings begin.
  • Without a cloud account (no AIRGRADIENT_API_TOKEN): nothing should be asked of the cloud, and the first charts only reach back to when the bot began recording.
  • ecobot.service as a real unit (MemoryDenyWriteExecute, AF_NETLINK for finding the sensor by name, StateDirectory, CacheDirectory): start it, then ask for a chart, an analysis question and something that reads the caches.
  • The Docker image with zeroconf in it: the image check on a pull request builds it.
  • ALERT_COOLDOWN_MINUTES in the production env file: the default is now 60; a line setting it to 30 keeps 30.

Try Home Assistant as the weather station's source, then redesign the storage

Set HOMEASSISTANT_URL and HOMEASSISTANT_TOKEN (and leave ECOWITT_SOURCE at auto, or try homeassistant with no Ecowitt keys) and run it for a week or two. It has only been run against a fake Home Assistant. Check in the log and the charts:

  • the start-up line says where the readings come from, and the live poll and minute pulls don't warn (or warn once and recover);
  • a chart of a few hours is drawn a reading a minute, a day or a week looks right, and a chart past Home Assistant's 10 days still has its bands (from the cache);
  • a rain alert comes within seconds of the gauge's report; "weather now" is fresh;
  • with both set, nothing is asked of the cloud except what Home Assistant lacks.

Then the storage redesign: docs/storage-plan.md has the measurements, the design, the migration and the steps (the requirements that were here are its section 2, the decisions its section 6.1). Steps 0 and 1 are done (the measurements, and the new storage checked on the real cache); next is step 2, the station on it. Its step 4 takes the AirGradient's readings from the monitor itself first, Home Assistant as the fallback and the cloud last (AIRGRADIENT_SOURCE; section 3.12 has how each source's PM2.5 is checked).