Cross-region data replication plan
A step-by-step blueprint for deploying Meridian workloads across multiple geographic regions with sub-second write propagation and automatic failover.
Topology selection
Choose active-active for latency-sensitive reads or active-passive for write-heavy workloads. Meridian supports both via a single config toggle in the cluster manifest.
Write-ahead log shipping
Every mutation is durably committed to a local WAL before fan-out. The replication stream uses compressed binary frames over mTLS, keeping cross-region bandwidth under 12 MB/s per 10k ops.
Conflict resolution
Last-writer-wins with hybrid logical clocks is the default. For CRDT semantics, enable the causal consistency module — it ships vector clocks alongside each payload and merges deterministically.
Health probing
Regional health endpoints are polled every 500 ms. If a region exceeds a 2-second p99 latency threshold for three consecutive probes, the control plane drains connections and promotes a standby.
Failback procedure
Once the primary region recovers, Meridian replays the delta log from the standby. After catch-up, traffic shifts back gradually over a 60-second weighted round-robin window to avoid thundering herd.
Estimated latency matrix
| Region pair | P50 write | P99 write | P50 read |
|---|---|---|---|
| us-east → eu-west | 68 ms | 112 ms | 4 ms |
| us-east → ap-southeast | 142 ms | 198 ms | 6 ms |
| eu-west → ap-southeast | 118 ms | 164 ms | 5 ms |