Incident Management Platform · View 08 of 34 · 3 · Structure
Decisions
- The paging path is four layers of small components written for one job each: an edge that sheds load per integration, services that decide and dispatch, the state they share, and the software that speaks to carriers. Nothing below them is called at run time.
- Delivery software runs inside the platform rather than behind a third-party aggregator. Asterisk and Jasmin speak SIP and SMPP to two carriers directly, which is what makes a keypress or an SMS reply an acknowledgement the platform receives itself (ADR-20).
- Record and insight stores are ordinary, well-understood databases. PostgreSQL holds seven years of events, ClickHouse holds thirteen months of attempts, Ceph holds the tamper evidence.
Build and adopt
- Built: ingest gateway, incident engine, escalation timers, dispatcher, ack receiver, schedule resolver, incident service, review module, console. Everything else is adopted open source (ADR-05). The built parts are the ones that encode the requirement; the adopted parts are the ones that encode decades of other people's operations.
Omitted
- Prometheus, Loki and the OpenTelemetry collectors that watch every layer (view 28), and the responder mobile app, which is a client of the ack receiver.
