Dezhurka: Telegram On-Call Tool as Opsgenie Shuts Down in 2027

How it works

Dezhurka does not replace monitoring. It receives alerts from Prometheus Alertmanager, Grafana Alerting or Zabbix via webhooks, each service having a secret integration key. It checks the schedule, messages the current on-call engineer in Telegram and by email, and duplicates the incident to a team chat. If nobody presses the "Take it" button within 5 minutes, it calls the next person, then the team lead; after two rounds, the timeline records that escalation is exhausted.

Repeated alerts with the same key count as one incident. For Alertmanager the key is the label fingerprint; for Zabbix it is {EVENT.ID}, so problem and recovery are matched. Once someone takes the incident, all Telegram messages are edited to show who is handling it.

Built for closed environments

The bot uses long polling via getUpdates instead of a Telegram webhook, so only outbound access to api.telegram.org is needed, with no public HTTPS endpoint. Only Telegram accounts linked through an HMAC-signed personal link can press buttons. The panel uses Basic authentication and checks Origin on POST forms.

The stack is Python 3.12, FastAPI and SQLAlchemy with SQLite or Postgres, roughly 2,000 lines of code and 24 tests, shipped as one Docker container. Everything runs in a single process with one asyncio.Lock, a deliberate choice to avoid double wake-ups or lost button presses. People, schedules and policies live in a YAML file, with secrets pulled from environment variables. Shift swaps are entered in the panel and override the rotation.

The source is on GitHub, with a one-command demo installer for clean Ubuntu that sets up Dezhurka with Prometheus, Alertmanager, Grafana and optionally Zabbix. The code is open for inspection; terms are in the repository's LICENSE.