Cross-region write latency caused cascading alerts that mimicked a database outage

Short summary

A team received critical alerts suggesting their database was down, but RDS metrics showed CPU under 20% and no pressure anywhere. The real cause was cross-region latency: their EU app made ~25 primary writes per request across the Atlantic to a US-East database, with each round trip adding hundreds of milliseconds. Prisma's default 5-second transaction timeout was exceeded because 25 cross-region queries stacked up to over 9 seconds, causing cascading failures that looked like a database outage.

• •Database alerts were caused by cross-region write latency, not actual database failure
• •A single request made ~25 primary writes crossing the Atlantic, exceeding Prisma's 5s transaction timeout
• •RDS metrics (CPU <20%, normal latency) quickly disproved the 'database is down' narrative

Generated with AI, which can make mistakes.