[stakach]

Adds end to end InfluxDB ingest specs (Redis → StatusEvents → InfluxDB, read back with Flux) covering every payload format, and fixes what they found:

• A dead MQTT broker blocked InfluxDB. Each message was retried ~30s while disconnected, filling the publisher queue and blocking broadcast for every sink. Publishing is now skipped while the client is closed; reconnecting already republishes all state.
• Resyncs were written to InfluxDB, recording a duplicate point for every status on each broker reconnect. Resyncs now go only to sinks that retain state.
• Buffered points could wait indefinitely under a steady trickle, and were dropped on stop. They now commit by age and on stop.
• Default filters never matched: they ran on raw JSON, where strings are quoted. They now apply to parsed field and tag values.

Review

• Full suite: 78 examples, 0 failures.
• New specs fail on master for the dead-broker, resync, trickle, stop and filter cases.
• Not covered: a half-open broker connection still applies backpressure until keepalive notices (~60–90s), and a slow InfluxDB can likewise slow MQTT.

🤖 Generated with Claude Code