v0.16.2

Fixed

• MQTT reconnection retry loop: Fixed a critical bug where
  force_reconnect() would fail to retry after a transient auth service or
  network error. Previously, when reconnection failed during the initial
  attempt, the method would return instead of rescheduling retries with
  exponential backoff. This caused the integration to become permanently stuck
  disconnected, requiring manual Home Assistant restart to recover. The fix
  converts force_reconnect() to loop internally with exponential backoff
  (2s, 5s, 15s, 30s, 60s cap), continuously retrying until successful
  reconnection or task cancellation. This ensures automatic recovery from
  transient auth/network failures. Fixes
  issue #100.

v0.16.1

Fixed

• MQTT reconnection rate-limiting bug: Fixed a critical bug in the
  reconnection rate-limit logic that prevented the integration from recovering
  after MQTT connection failures. When a reconnection attempt was rate-limited
  (< 30 seconds since the last attempt), the timeout counter was incorrectly
  reset to 0, preventing the reconnection threshold from ever being reached
  again. This caused the integration to remain disconnected indefinitely even
  when the MQTT service became available. The fix removes the counter reset
  from the rate-limit case, allowing the counter to accumulate normally so
  reconnection proceeds once the rate-limit interval expires. Fixes
  issue #100.

v0.16.0

Fixed

• MQTT reconnect alignment with nwp500-python v9.0.0: The coordinator no
  longer tears down the MQTT client after three disconnected update cycles while
  the library's own hardened internal reconnect loop is already running. Forced
  reconnect remains as a last-resort escape hatch only for repeated
  coordinator-level request timeouts on a connection that still appears up, and
  the integration now listens for the library's reconnection_failed event to
  trigger Home Assistant reauth when the internal loop stops permanently.
• AWS CRT clean-session warning workaround removed: nwp500-python v9.2.0
  fixes issue #97 upstream:
  _await_ack() and the subscribe/unsubscribe acknowledgement waits now
  attach a done callback whenever a shielded AWS CRT future is abandoned, so
  the eventual result/exception is always retrieved and logged at debug level
  by the library instead of leaking to asyncio's global exception handler.
  The coordinator's temporary global asyncio exception-handler
  install/restore machinery (_nwp500_exception_handler,
  _install_exception_handler, _restore_exception_handler, and the shared
  refcount globals) is no longer needed and has been removed.
• Recirculation Active binary sensor: Removed the redundant "Recirculation
  Active" binary sensor (recirculation_use), which read the same
  DeviceStatus.recirc_operation_busy field as the existing "Recirculation
  Operation Busy" sensor and duplicated it with a nonexistent recirc_use
  fallback. Use the existing "Recirculation Operation Busy" sensor instead.

Changed

• Internal API cleanup / Python 3.14 modernization: Removed redundant
  from __future__ import annotations imports from all
  custom_components/nwp500/ modules now that this integration is
  Python 3.14-only, updated stale nwp500-python version-pinned comments
  to reflect current library behavior, and replaced the auth-client
  shutdown dunder call with the library's public close() API. Initial
  auth setup still relies on NavienAuthClient.__aenter__() because
  nwp500-python does not yet expose a matching public connect/open
  lifecycle method for the coordinator's longer-lived auth session.
• Library Dependency: nwp500-python: Upgraded to
  9.2.0. No
  breaking changes affecting this integration. Notable changes:
  • Fixes issue #97:
    _await_ack() in mqtt/connection.py and the subscribe/unsubscribe
    acknowledgement waits in mqtt/subscriptions.py now attach a done
    callback whenever a shielded AWS CRT future is abandoned due to a
    timeout or cancellation, so its eventual result/exception is always
    retrieved and logged at debug level instead of leaking as an unhandled
    "Future exception was never retrieved" asyncio warning. This obsoletes
    this integration's own global asyncio exception-handler workaround,
    which has been removed (see "Fixed" above).
  • Internal refactors with no behavior change: CLI formatting stacks
    merged behind a single Rich renderer, mqtt/client.py slimmed via
    mixins, event-name constants de-duplicated between events.py and
    mqtt_events.py, and awscrt types (e.g. mqtt.QoS) wrapped behind a
    library-owned nwp500.QoS enum on public MQTT signatures. This
    integration does not use awscrt.mqtt.QoS directly, so no code changes
    were required.
• Library Dependency: nwp500-python: Upgraded to 9.0.0 (BREAKING). This
  is a major version bump on the library side that trims its public API
  surface and removes dead code; see the
  full release notes
  for the complete list. Notable changes affecting this integration:
  • Removed the never-raised TokenExpiredError exception. The coordinator's
    authentication error handling no longer imports or catches it (the
    remaining TokenRefreshError/AuthenticationError handling is
    unaffected, since the library never actually raised this class).
  • Numerous bug fixes relevant to this integration's stability: queued
    control commands are now preserved in order and expire after a
    configurable age instead of replaying stale commands after long
    outages; MQTT message dispatch now runs on the event loop instead of
    the AWS CRT network thread, fixing a potential
    RuntimeError: dictionary changed size during iteration during
    reconnects; the reconnection loop now survives all library errors
    instead of dying silently on auth/token errors during an outage;
    error_detected events are now emitted when the error code changes
    between two non-zero values; and sub-zero Fahrenheit temperature
    conversions are now rounded correctly.
  • No changes were required to this integration's own use of
    build_reservation_entry/build_tou_period, since it already imports
    them from nwp500.encoding rather than the removed top-level
    re-exports.
• Water heater mode enums: Replaced duplicated magic-number protocol values
  with nwp500.enums.CurrentOperationMode / DhwOperationSetting members and
  consolidated DHW mode-to-state translation so the water heater entity and its
  extra state attributes share one base mapping while preserving the vacation
  restore behavior.

v0.15.5

Changed

• Library Dependency: nwp500-python: Upgraded to 8.1.3. Fixes a
  thread-safety bug in on_connection_resumed where Task.cancel() was called
  directly from an AWS IoT SDK background thread. When the event loop was busy,
  the cancellation could be silently dropped, leaving a stale
  _reconnect_with_backoff task that would complete its sleep and tear down an
  otherwise healthy connection — restarting the
  disconnect → reconnect → AWS_ERROR_MQTT_UNEXPECTED_HANGUP cycle. See
  nwp500-python PR #89.

v0.15.4

Fixed

• MQTT reconnection recovery loop: The force_reconnect method was updating _last_reconnect_time before attempting setup, which prevented the integration from retrying failed reconnections. If setup failed, the rate-limiting check in the coordinator would see the recent timestamp and block further attempts for 30 seconds, keeping the device stuck in a disconnected state. Now _last_reconnect_time is updated only after successful setup, allowing retries on failed attempts while still preventing excessive reconnect attempts on success.

v0.15.3

Fixed

• MQTT client ID conflict: All HA instances authenticated with the same Navien account previously shared the same MQTT client ID (navien-ha-{user_seq}). AWS IoT Core only allows one active connection per client ID, so running more than one instance (e.g. production + local dev) caused a continuous ping-pong of disconnections. The client ID now incorporates the HA installation's unique instance UUID: navien-ha-{user_seq}-{ha_uuid[:8]}, making it stable across restarts and unique per installation.
• MQTT subscriptions lost after auto-reconnect: When the AWS IoT SDK reconnected internally before the integration's own reconnect logic fired, session_present=False was reported but no resubscribe_all() was called — leaving the client connected but with zero active topic subscriptions. Device status updates stopped flowing silently until a forced full reconnect occurred. Fix pending release in nwp500-python (see fix/mqtt-resubscribe-on-session-lost).
• Persistent MQTT session: Changed clean_session from True (library default) to False. Combined with the now-unique stable client ID, the AWS IoT broker resumes the existing session on reconnect (session_present=True), so subscriptions are preserved server-side and AWS_ERROR_MQTT_CANCELLED_FOR_CLEAN_SESSION errors are eliminated.

Changed

• Library Dependency: nwp500-python: Upgraded to 8.1.2

v0.15.2

Release v0.15.2

v0.15.1

Changed

• Library Dependency: nwp500-python: Upgraded from v8.0.0 to v8.1.0.
  • v8.1.0 (2026-05-16): Multiple bug fixes — see release notes

v0.15.0

Release v0.15.0

v0.14.4

Fixed

• README: Corrected stale version number (was showing 0.3.0, now reflects current version)

Changed

• CI: Removed ignore: brands from HACS validation workflow — brand assets are now provided via custom_components/nwp500/brand/ directory

---

Full changelog: https://github.com/eman/ha_nwp500/blob/main/CHANGELOG.md