Success Story
Justo cuts daily development cost by nearly 50% with end-to-end FinOps on AWS
Justo is an e-commerce and software platform for restaurants and delivery (identified through apps such as orion-delivery-app, pos.bejusto.com and delivery.bejusto.com). Its infrastructure runs on AWS with databases on MongoDB Atlas. Craftech executed an architecture-review, modernization and strong operational cost-optimization (FinOps) project.
-49.83%
daily development cost (from ~US$53 to ~US$26)
9.8 TB
old ALB logs in S3 purged via Lifecycle
Challenge
Justo had several pockets of inefficiency and unnecessary spend. In storage (S3) it had 45 TB total, of which 9.8 TB were just old ALB logs accumulated since 2021 with no retention policies, averaging USD 2,258.74 monthly. In compute (ECS), production services used on-demand Fargate without leveraging Spot, with suboptimal autoscaling that kept resources idle. Internal traffic (apps to MongoDB databases) was routed through public IPs and the NAT Gateway, inflating data transfer costs. There were obsolete and 'junk' cloud resources (abandoned dev clusters, EC2, ELBs, Target Groups, EFS and unused databases) still billing, and EKS clusters on version 1.25 at risk of paying AWS's costly Extended Support.
Solution
Craftech, through Squad 2, ran an end-to-end FinOps process tackling usage, pricing and architecture. In ECS and Fargate it unified all services to the Fargate Spot capacity provider, with a base of 2 On-Demand nodes for availability and all horizontal scaling on Spot, and redefined CPU- and memory-based scaling rules. It executed an aggressive deletion of unused infrastructure (including a dev EKS cluster 'el-tofu', orphaned EC2, Load Balancers, inactive Target Groups, EFS and obsolete S3 buckets). It configured S3 Lifecycle rules to automatically purge ALB logs older than 3 months. Using VPC Flow Logs it identified the services driving NAT Gateway overspend (traffic to MongoDB's port 27017) and planned VPC Peering/PrivateLink, unifying Load Balancers and migrating Ingress to internal LBs. As preventive modernization, it upgraded the internal-tooling EKS clusters from 1.25 to 1.29 to avoid Extended Support, and applied compression and caching on CloudFront.
Results
Cleaning up inactive development environments (removing Concourse and scaling down the justo-dev namespace, among others) reduced the daily development cost by 49.83% in the first days, from ~US$53 to ~US$26 per day. The transition to Fargate Spot in production and the automatic removal of terabytes of old S3 logs secured a structural reduction in the monthly AWS bill. And the timely EKS upgrades avoided future financial penalties from AWS Extended Support.
Stack & technologies
