← all posts site ↗

monitor alerts finally speak human

the monitor channel had one format that worked, service name, up or down, status code, like speedtest up with a green icon and a 200, or nextcloud down with a red circle and a 504. everything else was noise. raw error dumps saying write EPROTO followed by an alert number, or knex timeout acquiring a connection with a question about a missing trx call, which is not information, it is archaeology.

the fix was translation at the source. uptime-kuma messages now turn known error patterns into plain words: the ssl alert becomes tls error, the knex timeout becomes check error with a note that the monitor database is busy, plus dns lookup failed, connection refused, timeout, host unreachable, connection dropped. status codes stay untouched because those were the good part.

the rest of the noise got the same treatment. the daily update watcher wall of text is silent now unless an actual update fails, and then it is one short line. the unknown host spam from the lan scanner is off. the one thing baked into someone else's code, the doubled line when a machine goes down, stays, but the link at the end of it finally opens. alerts that read like a human wrote them take half the time to act on, which is the whole point.