[Apollon77]

Addresses #1213.

Nothing sent traffic on an idle WebSocket connection. Browsers never send WebSocket pings themselves, and the server did not send any either. A relay that closes idle connections, such as Home Assistant Cloud remote access, therefore dropped the dashboard every 1–2.5 minutes. The Home Assistant frontend stays connected over the same path because its server pings every 55 s. Supervisor and Core ingress pass ping/pong frames through.

• The server pings each client of its WebSocket API every 30 s. A client that leaves 3 pings in a row without a pong or a message is terminated. Before this, dead connections were only cleaned up when a send stalled. No ping is sent while a connection is in backpressure mode, because a ping waits behind the backlog there; the existing send watchdog covers that case.
• The dashboard makes one automatic reconnect attempt when a connection drops, then falls back to the "Connection lost" page. Event listeners are registered once; before, every reconnect added another set.
• ws-client connect() now fails when the server closes the connection before it is ready. Before, the call never returned, which could leave the dashboard on "Connecting…".

Not done from the issue: showing the result of a running commissioning after a reconnect. Once the connection stays up, that case is rare.

Remote-path behaviour is not yet confirmed; the reporter will recheck after a release.

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

A drained burst can leave a connection in queued mode indefinitely, suppressing keepalive pings and allowing proxy idle timeouts to recur.

1 open issue

• Restore direct mode after queued burst drains · New

What changed in this PR

This PR addresses #1213 by adding WebSocket keepalives and improving dashboard recovery from dropped connections.

Changes:

• Send periodic server pings and terminate unresponsive clients, while deferring to backpressure handling during queued sends.
• Attempt one automatic dashboard reconnect and register event listeners only once.
• Reject client connection attempts when the server closes before sending server information, with regression coverage.

File	Description
packages/​ws-controller/​test/​WsBackpressureLatencyReproTest.ts	Adapt the socket fake to the keepalive interface.
packages/​ws-controller/​test/​WebSocketConnectionTest.ts	Test ping intervals, responses, termination, backpressure, and disposal.
packages/​ws-controller/​src/​server/​WebSocketConnection.ts	Add keepalive timers and unanswered-ping tracking.
packages/​ws-client/​test/​WsClientTest.ts	Test connection rejection on early server closure.
packages/​ws-client/​src/​connection.ts	Reject pending connection setup when the socket closes.
packages/​dashboard/​src/​pages/​matter-dashboard-app.ts	Add automatic reconnect and avoid duplicate listeners.
CHANGELOG.md	Document keepalive and connection-recovery changes.

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

Keepalive behavior through the affected remote-proxy path remains unconfirmed and needs human validation.

No open issues 🎉

1 resolved since last review

• Restore direct mode after queued burst drains

_{Review effort: Balanced}