The biggest auction of the night
was the one most likely to fall over.

We split the application into containerized services on Amazon ECS Fargate, separating bidding, video, chat, payments and inventory.

• Each service scales against its own workload now, so a rise in viewers doesn't make every part of the application grow in step.
• AWS Auto Scaling expands and contracts with real demand, which means a big event doesn't need capacity provisioned hours ahead or operators kept on standby.
• An Application Load Balancer routes requests to healthy targets and drains traffic during deployments, so a slow service can no longer take the entry point down with it.
• AWS Lambda handles event-driven execution and asynchronous workflows outside the main request paths.

All the features drew on the same shared resources, so the busiest one decided how well the rest performed.

The legacy database was a single point of contention, sized for a normal day. Amazon RDS replaced it. Read replicas separate catalog and analytical reads from the transactional writes that carry bids, and automated backups support recovery, replacing the manual playbook that used to follow an incident.

The platform couldn't stop for its own modernization, because there's always another auction, so the team introduced services incrementally behind the load balancer and shifted traffic across as the migration progressed. The cutover happened with zero downtime and no scheduled outage windows.

The product team put its effort where bidders were dropping out.

• Google and Apple sign-in joined phone authentication, and persistent sessions mean a bidder doesn't have to keep logging back into the mobile app.
• Checkout expanded to wallets, Apple Pay, Google Pay, GiroPay, EPS and stored cards alongside the existing payment route, and the wallet can pre-authorize funds before an event so a winning bid settles against a cleared balance instead of failing at the last step.
• Amazon SNS sends a notification when a followed seller goes live, when an auction is about to close, or when a saved item is about to appear.
• mParticle pulls viewing, bidding and checkout events together into one shared customer data layer, so product and marketing teams can plan re-engagement based on what actually happened in the funnel.

The team made these changes because, for an impulse purchase, every extra step between a buyer wanting an item and owning it is a chance to lose them.

We rebuilt the release path around automated checks, so every change is verified by the pipeline before it ships.

• AWS CodePipeline automates build, test and deployment. Tests act as release gates, and rolling back a release is one defined operation.
• AWS CloudFormation keeps the infrastructure in version-controlled templates, so development, staging and production are provisioned the same way instead of drifting apart over time.
• Amazon CloudWatch dashboards and structured logs track bid latency, payment success, video quality and concurrent users, with alerts tuned to live-event demand, so operations gets signals that general server metrics never would.
• AWS WAF protects the public edge against bot traffic and scraping, IAM roles limit service access, and AWS KMS and AWS Certificate Manager handle encryption and certificates.