# ALB active-active routing evidence — 31 July 2026

## Scope

This record verifies the Ireland application tier only. It does not claim
regional disaster recovery or redundancy of the VPC, Virtual Private Gateway
or Site-to-Site VPN connections.

## Live baseline

- ALB:
  `eg334s-personal-demo-site-alb-349585711.eu-west-1.elb.amazonaws.com`
- Target group: `eg334s-personal-targets`
- `eu-west-1a`: original hub `i-074bf0cb910a99e02`, private IP
  `10.1.1.61`
- `eu-west-1b`: ASG-managed secondary `i-0001c5bff73898175`, private IP
  `10.1.2.110`
- Secondary ASG: `min=1`, `desired=1`, `max=1`
- Both targets reported `healthy` after the test and at final verification.

## Steady-state load distribution

Thirty sequential HTTP requests were sent through the ALB after both targets
became healthy:

- original target responses: **15**
- secondary target responses: **15**
- failed requests: **0**

Result: **PASS** for active-active steady-state traffic distribution.

## Abrupt primary-stop test

The original target was stopped while the ALB was polled continuously. The
test captured 150 samples:

- HTTP 200 responses: **143**
- responses verified from the surviving secondary: **141**
- transient failures: **7**
  - HTTP 502: **3**
  - connection timeout / curl status 000: **4**

The transient window lasted approximately 41 seconds while target health
converged. The target group's verified settings were a 15-second interval,
5-second timeout and unhealthy threshold of two.

Result: **NOT MET** for a strict zero-interruption claim. The surviving target
served the large majority of requests, but the abrupt EC2 stop was visible to
clients before the ALB removed the failed target.

## Accurate claim and operating guidance

The service is **active-active across two Availability Zones**, with two
normally running healthy targets. It is not described as guaranteed
zero-interruption failover.

For planned maintenance, deregister and drain the selected target before
stopping it. For unplanned failure, more aggressive target-health settings may
reduce detection time, but client retry, connection draining and application
state must also be designed and tested before any zero-downtime claim.

The older 30 July 2026 cold-recovery drill remains separate historical
evidence: it restored application service in 218 seconds against the accepted
600-second objective.