Atlas, given current load growth we'll hit single-region limits by end of Q3. What's the migration path to multi-region?
ATLAS18:13
Three viable paths. Ranked by risk/reward:1. Active-passive failover (lowest risk) — replicate primary region to us-east-2, DNS-based failover. Gets us resilience but not horizontal capacity.2. Active-active with regional sharding — users pinned to nearest region. Needs cross-region sync for shared state (billing, auth).3. Full cell-based architecture — each cell is autonomous, users assigned to cells. Highest reinvention, best ceiling.Start with (1) as an insurance policy. Plan (2) for H1 next year.
You18:32
What's the failover RTO we can realistically commit to?
ATLAS18:33
With DNS-based failover and 60s TTL: 5-8 minutes realistic RTO, 15 minutes worst case. If we need sub-minute, need BGP anycast which is 6-week lift minimum.
⏎ to send · ⇧⏎ for new line · this is a mocked interface