Transient connection errors (TimeoutError/ClientConnectorError) are only retried in fetch_events

[iMicknl]

Summary

Only fetch_events is decorated with @retry_on_connection_failure. Every other HTTP-making method on OverkizClient — including the command path (_execute_action_group_direct) and all setup/state/refresh calls — lacks it, so a transient TimeoutError / aiohttp.ClientConnectorError propagates out raw on the very first occurrence instead of being retried.

This surfaced in Home Assistant: a ConnectionTimeoutError (subclass of the builtin TimeoutError) raised from a cover close command escaped as an unhandled traceback, because execute_action_group → _execute_action_group_direct carries only @retry_on_too_many_executions + @retry_on_auth_error.

Decorator definition

retry_on_connection_failure = backoff.on_exception(
    backoff.expo,
    (TimeoutError, ClientConnectorError),
    max_tries=5,
    max_time=120,
    ...
)

Audit of HTTP methods missing @retry_on_connection_failure

fetch_events is the only method that has it. Everything below makes an HTTP request (_get/_post/_put/_delete) and does not:

• Command / execution: _execute_action_group_direct, execute_persisted_action_group, schedule_persisted_action_group, cancel_execution
• Core polling/setup: get_setup, get_devices, get_gateways, get_state, refresh_states, refresh_device_states, get_current_execution, get_current_executions, get_execution_history, register_event_listener, unregister_event_listener, get_api_version
• Everything else: get_diagnostic_data, get_device_definition, get_action_groups, get_places, all get_reference_*, all firmware methods, all local-token methods, all developer-mode methods, search_reference_devices, etc.

Note: ServerDisconnectedError is retried (once) only as a side effect of retry_on_auth_error's relogin path; plain TimeoutError / ClientConnectorError are retried nowhere except fetch_events.

Suggested fix

Apply @retry_on_connection_failure consistently to the request-issuing methods — or, more robustly, centralize it in the _get/_post/_put/_delete helpers so every call gets uniform transient-connection retry, removing the need to remember it per-method. The command path in particular should retry transient timeouts before giving up.

Context

Found while reviewing home-assistant/core#173155. On the HA side we added a catch that converts these to HomeAssistantError so the user sees a clean error instead of a traceback, but the underlying retry gap belongs here in the library.