v2.21.1 crashes HA Core on load (.fault file written, all entities transition to unavailable)

[jflammia]

Summary

After HACS auto-installed ha_carrier v2.21.1 on 2026-05-08 at 07:13 EDT, HA Core crashed at 07:15:22 EDT with /config/home-assistant.log.fault written (empty body, suggesting logging itself broke during the crash). HA auto-restarted but multiple integrations failed to re-initialize correctly, leading to a long degraded state until I disabled ha_carrier by renaming its directory.

Environment

• HA Core: 2026.5.0 (stable, not beta)
• HA OS (Home Assistant Yellow appliance), aarch64
• ha_carrier previous version: 2.21.0 (working)
• ha_carrier new version: 2.21.1 (released 2026-05-08T02:02 UTC, release notes)
• carrier-api requirement bumped to 2.11.3 in this release
• HACS-managed install, auto-update enabled

Timeline

Time (EDT)	Event
07:13	HACS auto-pulled v2.21.1, files in /config/custom_components/ha_carrier/ written
07:15:22	HA Core crash; /config/home-assistant.log.fault created (0 bytes, mtime = crash time); all stable entities (sun.sun, zone.home, person.justin) transitioned to unavailable simultaneously
07:15+	HA auto-restart; recovered into degraded state — many integrations stuck in not_loaded (homekit_controller, apple_tv, eufy_security, wled, roborock, etc.); Hue Bridge + UniFi Network "loaded" but entities unavailable
Subsequent restart attempts	Each restart wrote a new .fault file — cycle continued until ha_carrier was removed from load path

Recovery

1. mv /config/custom_components/ha_carrier /config/custom_components/ha_carrier.DISABLED-2026-05-08
2. Restart HA Core
3. → All integrations except ha_carrier loaded cleanly; entities recovered

Diagnostic data

• /config/home-assistant.log — does not exist during the broken state. Only rotated home-assistant.log.1 (Nov 2025) and home-assistant.log.fault (0-byte) present. Suggests logger itself crashed on ha_carrier import/setup.
• /config/custom_components/ha_carrier/__pycache__/ was generated at 07:15 — Python compiled the new code but execution failed.
• HA Core showed state=RUNNING per its API even while in this broken state — the crash + restart left HA "running" but with cascading integration failures.

Suspected cause

carrier-api 2.11.3 bump appears to be the only change in this release per the changelog. The crash may originate inside the carrier-api dependency or its interaction with the v2.21.0 → 2.21.1 wiring.

The integration loads OK on its own (config_entry shows state=loaded after restart) but its presence appears to disrupt other integrations during HA Core startup — possibly a long blocking call or unhandled exception during async setup that holds up the setup_after_dependencies queue.

Workaround

Disable the integration (rename directory) until v2.21.2+ ships with a fix.

Request

Could you investigate v2.21.1 → carrier-api 2.11.3 path? Even a hint about which config invariant changed would help users who still want to use the integration. Happy to provide more diagnostic output if helpful (HA logs are tricky here because the active log file is missing — but the .fault file timestamp + integration entry IDs are available).

Thanks for maintaining this integration!