Real-time updates

Operators see acknowledgements, escalations, alert repeats and notes from others within 2 seconds, without refreshing. Rationale: ADR 0003.

Hub contract

Path /hubs/incidents. Server-to-client only; all writes go through REST so validation and rate limiting live in one place.

Method	Argument	Sent when
IncidentChanged	Incident	any field changes: create, transition, escalation, assignee
TimelineAppended	TimelineEntry	a timeline entry is written: transitions, notes, alerts

Production runs Azure SignalR Service in default mode; locally the hub is in-process. The API code is the same: AddAzureSignalR() is applied only when the connection setting exists.

Connection and broadcast

sequenceDiagram
    autonumber
    participant B as Browser (console)
    participant API as Incidents API
    participant S as Azure SignalR Service
    participant B2 as Other operator

    B->>API: POST /hubs/incidents/negotiate
    API-->>B: { url: SignalR endpoint, accessToken }
    B->>S: WebSocket connect with token
    B2->>API: POST /api/incidents/{id}/acknowledge
    API->>API: commit transition
    API->>S: IncidentChanged(incident), TimelineAppended(entry)
    S-->>B: IncidentChanged
    S-->>B: TimelineAppended
    B->>B: replace incident in list cache, append to timeline

The API holds no client connections; the service does. API replicas can scale to zero or restart without dropping operators.

Reconnect and consistency

stateDiagram-v2
    [*] --> Connecting
    Connecting --> Connected: negotiate + WebSocket
    Connected --> Reconnecting: connection lost
    Reconnecting --> Connected: within retry schedule (0, 2, 10, 30 s)
    Connected --> Resync: reconnected
    Resync --> Connected: refetch open incidents
    Reconnecting --> Polling: retries exhausted
    Polling --> Connecting: every 30 s, refetch + reconnect attempt

Messages missed while disconnected are not replayed. The client treats REST as the source of truth and refetches the open list after every reconnect, so the hub only needs to be fast, not reliable. Each IncidentChanged carries the full incident, which makes applying it idempotent and order-tolerant by comparing timestamps.

Limits

Item	Free_F1	Standard_S1 (per unit)
Concurrent connections	20	1 000
Messages per day	20 000	1 000 000
SLA	none	99.9%

The demo runs on Free_F1. A team of 40 operators with the console open all day fits one Standard_S1 unit; moving is a SKU change in Bicep, not a code change.