0ms
downtime on release
Boss of the World
Game backend
ChallengePlayer counts spike without warning during live events โ the old single-server setup needed a human to scale it in time.
ApproachMoved to EKS with horizontal pod autoscaling wired to real player-count metrics, plus blue/green rollouts for releases.
ResultReleases ship with 0ms downtime, and peak-hour spikes scale themselves โ no on-call page required.
EKS
GitHub Actions
15+
independent services
Buyss Pass
Marketplace platform
ChallengeA monolithic marketplace meant one risky deploy could take down checkout, listings, and search all at once.
ApproachSplit the platform into 15+ independently deployed microservices, each with its own pipeline and scaling policy.
ResultTeams ship to one service without touching the rest, and a bad deploy stays contained to its own container.
Docker
CI/CD
10x
concurrency spikes absorbed
Dominos
Multiplayer backend
ChallengeSessions can 10x in seconds when a match goes viral, and cold-started pods were too slow to keep up with the rush.
ApproachPre-warmed node pools and tuned load-balancer health checks so Kubernetes absorbs the spike before players feel any lag.
Result10x concurrency spikes absorbed with no dropped connections and no manual scaling.
Kubernetes
Load balancing
EKS+ฮป
hybrid scaling model
Wagmi
Hybrid compute
ChallengeSteady baseline traffic and unpredictable bursts don't fit one compute model โ over-provisioning either wastes budget or under-serves peaks.
ApproachRouted workloads by shape: EKS handles the steady state, Lambda absorbs the bursts, sized and deployed with Terraform.
ResultRight compute for every workload shape, without paying full-cluster prices for burst capacity.
EKS + Lambda
Terraform