WebSocket page

performance/websocket/ — activity and timing for Django Channels consumers instrumented with WebSocketLoggingMixin. Every event (connect, disconnect, receive, send) is recorded in WebSocketEventLog; unlike API requests, WebSocket events are not sampled — all of them are stored when ENABLED is on, so counts here are exact, not weighted.

KPI tiles

Tile	Source	Meaning
Active connections	connects − disconnects over the range (floored at 0)	Approximate live connections. A long range inflates this if sessions predate the window; read it alongside the lifecycle chart.
Avg processing time	Avg(processing_time_ms)	Mean time the wrapped handler took across all events.
Inbound messages	Count(event_type = receive)	Client → server frames.
Outbound messages	Count(event_type = send)	Server → client frames.

Connection lifecycle chart

Connects vs disconnects per hour (TruncHour). A persistent gap between the two lines means connections are accumulating (or leaking); they should track each other over a healthy period.

Messages per consumer

Inbound vs outbound counts grouped by consumer_name. Identifies which consumer carries the traffic and whether it is read-heavy or write-heavy.

Connection duration histogram

Disconnect events carry connection_duration_ms (time since the matching connect). Up to 5,000 recent durations are bucketed into 10 bins from 0 to the longest observed duration, labelled in seconds. It shows whether sessions are mostly short-lived or long-lived.

Consumer breakdown table

One row per consumer (consumer_name), combining several aggregates:

Column	Source	Meaning
Active	connects − disconnects for that consumer	Approximate live connections on it.
Avg processing	Avg(processing_time_ms)	Mean handler time.
p95 processing	95th percentile of that consumer's processing times	Tail latency for its events.
Avg message size	Avg(message_size_bytes)	Typical frame size.
Fanout	outbound / max(inbound, 1)	Broadcast ratio — see below.

Fanout is the number of outbound (server → client) frames per inbound frame. A value near 1 is request/response; a high value flags a broadcast-heavy consumer (one inbound message triggering many sends), which is where WebSocket load tends to concentrate.

Event latency table

Processing-time percentiles grouped by consumer + event type (_ws_event_latency), sorted by p95 descending, over up to 50,000 recent events. This is the breakdown that exposes slow connect events specifically — the real user-facing WebSocket pain (a slow handshake) that a single consumer-wide average would hide. Each row shows count and p50/p95/p99 for that (consumer, event) pair.

Refresh endpoint

performance/api/websocket/ returns the page context as JSON, staff-only.