Detect dead connections via a read timeout and reconnect

[ptr727]

Why

The Host Command connections (event stream and command client) read with no timeout and rely on the peer closing the socket to surface a lost connection. On a hard power-off there is no TCP RST, and the event stream's ECHO keepalive writes only fail once the OS abandons the connection (about 15 minutes). Until then:

• the event stream read blocks forever, so Connected / Disconnected / Reconnected never fire and consumers keep treating the controller as online;
• the command client reads time out (it already passes read_timeout) but the timeout does not close the socket, and _get_connection only reconnects when the connection is closed, so every later request reuses the dead socket and keeps timing out.

Net effect: a powered-off or crashed controller is not detected for about 15 minutes, and per-object data does not recover until then.

How it is implemented

• Event stream (_command_client/events.py): read with a timeout of 2 * KEEPALIVE_INTERVAL. The ECHO keepalive round-trips every 60s on a healthy link, so 120s of silence means the link is dead. ClientTimeoutError subclasses ClientConnectionError, so it already routes into the existing reconnect path.
• Both clients: on ClientConnectionError, close the connection so the next _get_connection opens a fresh socket instead of reusing the stale one. Without this, a read timeout leaves the socket open and the event stream re-emits Reconnected on a dead socket, flapping every timeout and never recovering.

Two small follow-on commits keep CI green: a Generator (not Iterator) return type on a @contextmanager example to clear the only pyright reportDeprecated error, and Python 3.14 added to the lint matrix.

How it was tested

Validated against a real InFusion controller by correlating Home Assistant entity state with ground-truth ICMP reachability across several power-off / power-on cycles:

• Connection loss detected about 108-120s after the link drops (previously not detected within 8+ minutes).
• Connection state no longer flaps while the controller is unreachable (single latched transition).
• Event stream and per-object command data both recover within a few seconds of the controller returning (previously stuck until the ~15 min OS timeout).
• Confirmed ECHO round-trips every 60s (ECHO -> R:ECHO), so a healthy but quiet link never reaches the read timeout.

ruff check, ruff format --check, and pyright 1.1.410 pass across the 3.10 - 3.14 matrix.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds explicit read timeouts and socket teardown on command/event streams to detect dead TCP links promptly and force clean reconnection instead of waiting on OS-level TCP timeouts.

Changes:

• Introduces a read timeout for event stream reads tied to the keepalive interval.
• Closes connections explicitly on ClientConnectionError (including read timeouts) to avoid reusing stale sockets.
• Updates an example type annotation and expands CI lint matrix to include Python 3.14.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
src/aiovantage/_command_client/events.py	Adds READ_TIMEOUT and closes the connection on timeout/connection loss to trigger reconnect behavior.
src/aiovantage/_command_client/client.py	Ensures command requests close the socket on connection/timeout errors to avoid reusing an unknown-state connection.
examples/loads/control_load.py	Adjusts contextmanager return type annotation to Generator.
.github/workflows/lint.yml	Adds Python 3.14 to the lint test matrix.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.