docs · administration

Data retention and server metrics

What is pruned

Once a day, the server cleans up what would otherwise only grow:

  • Mentions & replies: each person keeps their newest 500 entries, none older than 180 days.
  • Leftovers of deleted messages, rooms and accounts: read positions, hidden rooms, unfinished jobs and the like.
  • Expired invite links, a day after they expire.
  • Data exports, after 7 days.
  • Usage statistics: the counts taken every 5 minutes are kept a year, and each person's activity 90 days. Tick no_stats on the site record to stop recording them at all.

Messages and files are never pruned. Every room keeps its full history.

Metrics

GET /api/chat/metrics reports the server's state for monitoring. Only superusers can read it.

  • Queues: for each background job (Telegram, linked servers, link previews, thumbnails, file moves, cleanup), how many jobs wait, how many failed at least once, and the oldest one's age.
  • Workers: when each one last ran, and its last error.
  • Realtime: connections, events waiting, slow clients dropped.
  • The server: uptime, versions, database size, memory, and how many requests are waiting their turn.

It answers in JSON, or in Prometheus's text format with ?format=prometheus (or when the request asks for text, as scrapers do). The /stats page shows the same to a superuser.

Scraping it with Prometheus

A scraper needs a superuser token. Make a long-lived one with PocketBase's impersonate call: signed in as a superuser, POST /api/collections/_superusers/impersonate/{id} with a long duration in seconds, where {id} is the superuser's id. Then:

scrape_configs:
  - job_name: chirkle
    scheme: https
    metrics_path: /api/chat/metrics
    params:
      format: [prometheus]
    authorization:
      credentials: <the superuser token>
    static_configs:
      - targets: [chat.example.com]

Health

GET /api/health answers 200 while the server is up. The Docker image and Dokku use it.

Edit this page on GitHub