Whether each part of nixkeeper is on time: its sync and hourly updates, the five digests, start-runs, the tests on main of every repository, and the tokens that expire. One page, status.nixkeeper.com, and an issue here for anything that needs attention.
Every hour, check.yml looks at each row of
nixkeeper_status/checks.py:
- runs: when a workflow last ran, and whether its last runs passed;
- data: when a digest's data (or the page's) was made, from its
meta.json, as a run can pass and still publish nothing new; - start-runs: when it last started a run (it starts them from outside GitHub, whose own schedules come hours late when it's busy);
- tests on main in every repository;
- tokens: the security tracker's (it renews itself), start-runs' and digest-versions' GitLab token (renewed by hand), before they expire;
- package lists: mistakes the sync found in them, such as a list naming a package nixpkgs no longer has: any is red;
- what nixkeeper's last sync read from each source (Repology's projects,
Hydra's evaluations, the security tracker's CVEs...) and how old it is,
from the page's
index.json; nixkeeper.com points here for them; - and backlogs worth seeing (logs or diffs still to read), never late.
Each part says when it's late, from a week of real runs: green is on time, yellow is late, red needs attention; grey couldn't be read (GitHub's trouble, not the part's). A row is as bad as its worst part.
A row that turns red opens an issue here, labelled status, with what's
wrong and links to its runs. It stays open while the row is late, and
closes itself with a comment once it's on time again. Watching this
repository is all it takes to be told: GitHub sends issues by mail and to
its mobile app. When GitHub can't be read for most rows, no issue opens or
closes.
With a Telegram bot set up, each issue opened or closed is also a message,
and so is a check that fails: create a bot with
@BotFather, send it any message, find your chat's
id in https://api.telegram.org/bot<token>/getUpdates (message.chat.id),
and add both as the repository's Actions secrets TELEGRAM_BOT_TOKEN and
TELEGRAM_CHAT_ID. Starting the check by hand with "Send a test message"
checks them. Without them, nothing is sent.
The check is started by GitHub's schedule and by start-runs, so it keeps running when either stops. If both stop, the page says its results are old.
index.html on GitHub Pages (from main, served as it is:
.nojekyll; at status.nixkeeper.com by CNAME), reading
data/status.json from the data branch, which each check replaces (main
plus one commit: the issues are the record of what went wrong).
?data=URL reads another status.json.
assets/ has nixkeeper's mark (the tab's icon, coloured by the status) and
its lockup, copied from nixkeeper's page/ (generated there by
scripts/brand/, MIT like this repository).
python3 -m nixkeeper_status check --out data checks every row and writes
data/status.json (Python 3.11 or later, standard library only; with
GITHUB_TOKEN set, GitHub allows more than 60 requests an hour).
--issues opens and closes the issues (with GITHUB_REPOSITORY),
--dry-run says which it would. nix run . -- check does the same;
nix flake check runs the tests (which never ask GitHub), lint and the
formatting check; nix fmt formats.
To check something new, add a row to checks.py.
This repository's code and documentation are written with the assistance of an AI (Claude, by Anthropic), with a person in the loop: nixkeeper's maintainer decides what to build, and reviews and tests every change before it's merged. The check itself is automation: a scheduled workflow that reads nixkeeper's public runs and data, and opens and closes issues in this repository only. It never comments on, closes or merges anything in nixpkgs.
MIT