Log monitoring
Make failure visible before a user reports it.
Structured events
Log events, not sentences. Each event carries a stable name, a correlation identifier, a duration and an outcome. Never log credentials or request bodies.
Alerting
Alert on symptoms a user would notice: error rate, latency and saturation. Do not alert on a single failed request, and do not alert on a metric nobody will act on at three in the morning.
Reducing noise
An alert that fires and is routinely ignored is worse than no alert. Delete it or fix its threshold.