Failover routing
Active-passive
resource "aws_route53_record" "primary" {
zone_id = var.zone_id
name = "app.example.com"
type = "A"
set_identifier = "primary"
failover_routing_policy { type = "PRIMARY" }
health_check_id = aws_route53_health_check.app_primary.id # or alias evaluate_target_health
alias {
name = aws_lb.primary.dns_name
zone_id = aws_lb.primary.zone_id
evaluate_target_health = true
}
}
resource "aws_route53_record" "secondary" {
zone_id = var.zone_id
name = "app.example.com"
type = "A"
set_identifier = "secondary"
failover_routing_policy { type = "SECONDARY" }
alias {
name = aws_cloudfront_distribution.maintenance.domain_name
zone_id = aws_cloudfront_distribution.maintenance.hosted_zone_id
evaluate_target_health = false
}
}
- The primary must have a health check (or evaluate-target-health); otherwise Route 53 can't know when to fail over.
- While the primary is healthy, all queries get the primary. When it's unhealthy, queries get the secondary.
- The secondary can also have a health check (if both are unhealthy, Route 53 returns the primary).
- Common secondaries: a DR region's ALB (pilot light / warm standby), or a static maintenance page on S3 + CloudFront ("we'll be right back").
For a static failover page, prepare in advance: the page, its CloudFront distribution and certificate for the same hostname (us-east-1 ACM), and a tested runbook — not something to build during an outage.
Failover timing
Total time ≈ health check detection (interval × failure threshold, e.g. 10 s × 3 = 30 s) + Route 53 propagation (seconds) + client cache TTL (60 s if configured well) + client retry behaviour. Plan for ~1–2 minutes with good settings, longer for clients that ignore TTL.
"Failover isn't switching even though the primary is down"
Check: the health check is actually attached to the primary record (or evaluate-target-health is on); the health check targets the right endpoint/port/path and reports unhealthy; Route 53 health checkers can reach the endpoint (security groups allow their IP ranges — otherwise the check is always failing or always passing depending on setup); for alias targets, that the ALB really has no healthy targets (an ALB with one healthy target is still "healthy"); and client caches (TTL).