Cross-Region Replicas

Cross-region replicas provide a read-only warm copy in another geography. They are almost always async; accept replication lag as your cross-region RPO unless you invest in specialized sync architectures.

Recipe

us-east primary, eu-west async standby for DR read and promotion candidate.

# eu-west standby primary_conninfo targets us-east primary
# Use TLS and dedicated replication network path

-- On primary: monitor cross-region standby
SELECT application_name, client_addr,
       pg_wal_lsn_diff(sent_lsn, replay_lsn) AS bytes_behind,
       replay_lag
FROM pg_stat_replication
WHERE client_addr << '10.1.%';  -- eu-west CIDR example

When to reach for this: Regulatory residency, regional read scaling, or DR promotion when primary region is lost.

Working Example

Deploy cross-region physical replica with slot and lag SLO:

# eu-west: base backup from us-east primary
pg_basebackup -h primary.us-east.db.internal -U replicator \
  -D /var/lib/postgresql/18/main -Fp -Xs -P -R \
  -C -S eu_west_dr_slot

-- primary_conninfo on eu-west includes sslmode=verify-full
-- Lag alert query (run on primary)
SELECT application_name,
       pg_wal_lsn_diff(pg_current_wal_lsn(), replay_lsn) AS dr_lag_bytes
FROM pg_stat_replication
WHERE application_name = 'eu_west_dr';

# Optional: eu-west serves read-only reporting via HAProxy replica frontend
# Writes still route to us-east only until DR declaration

What this demonstrates:

• Cross-region link adds WAN RTT to replay lag.
• Physical slot on primary retains WAL for DR replica during outages.
• DR replica is read-only until promotion runbook executes.

Deep Dive

Lag Budget

Set RPO tier 1 cross-region to measured p99 lag plus archive fallback.

Promotion Runbook Outline

1. Declare disaster (see DR Runbook & Comms)
2. Stop writes to primary region (if partially alive)
3. Verify DR replica lag acceptable or accept data loss window
4. pg_ctl promote on DR or Patroni failover in DR scope
5. Repoint DNS / HAProxy / PgBouncer to DR region
6. Rebuild replication back to primary region when recovered

Cascading Cross-Region

Hub in us-east feeds multiple regions (see Cascading Replicas) to reduce primary fan-out.

Gotchas

• Treating DR replica as HA - WAN partition causes large lag; not sub-minute failover. Fix: Same-region Patroni for HA; cross-region for DR.
• Sync replication cross-region - Commit latency unacceptable for most apps. Fix: Async with documented RPO.
• No slot on DR replica - Region brief network blip loses WAL forever. Fix: Permanent eu_west_dr_slot.
• Reads on stale DR replica - Users see old data. Fix: Route only analytics to DR; document lag in UI if needed.
• Promotion without etcd in DR - Patroni state confused. Fix: Separate DCS scope per region or manual promote runbook.

Alternatives

FAQs

Related

• DR Basics - RPO/RTO tiers
• Streaming Replication Basics - standby setup
• Cascading Replicas - hub-spoke
• Cloud PITR - managed cross-region backup

Stack versions: This page was written for PostgreSQL 18.4 (stable 18, maintenance 17), pgvector 0.8+, PgBouncer 1.x, Patroni 3.x, and PostGIS 3.5+.