A postmortem analysis of a 2019 Atlassian outage caused by AWS Elastic Load Balancer (ELB) not being 'warm' enough during a blue-green deployment. When Atlassian flipped DNS from the old ELB to a newly created one, the new ELB only had a fraction of the internal load balancer servers needed to handle production traffic. It took ~5 minutes for the ELB to auto-scale and recover. Key takeaways: load balancers are not infinitely scalable magic — they take 1–7 minutes to spin up new internal servers, they handle gradual surges better than sudden spikes, and pre-warming ELBs before expected traffic is critical. The fix was switching from full infrastructure replacement (blue-green) to rolling deployments that reuse the existing load balancer.
•13m watch time
2 Impressions