Read summarized version with
About project
Working time:
Since 2025
Industry:
Biopharma
The service:
Managed Services
Overview
The customer is a US biopharma company whose research teams use a data platform Dedicatted built to ingest, process and share experimental data. The platform runs as containerized services and workers on Amazon ECS with AWS Fargate.
The Challenge
After launch the platform needed operations, but the research IT team had no container specialist and did not want to adopt Kubernetes to get one.
1. Night-Time Failures
Batch workers failed overnight and were noticed by scientists the next morning.
2. Stalled Deployments
A deployment that stalled halfway left services running mixed versions.
3. Ageing Images
Base images and Fargate platform versions fell behind, and vulnerabilities piled up in scans.
4. Unreviewed Cost
Nobody looked at why compute cost rose in heavy research months.
Technology & Approach
We took over day-to-day operations of the ECS services, with alarms on what matters, safe deployments and a monthly report.
Implementation
Onboarding
Inventoried clusters, services, task definitions, capacity providers, load balancers and pipelines, set alarms on failed deployments, task restarts, queue depth, latency and errors, and wrote deployment and rollback runbooks.
Monthly Operations
Our engineers respond to incidents around the clock, triage failed deployments and roll back when needed, update base images and Fargate platform versions every month, scan images in Amazon ECR, and tune capacity and autoscaling.
Deployment Safety
Blue/green deployments with AWS CodeDeploy and automatic rollback on alarms, configured for each service.
Reporting
A monthly report covers incidents, deployments, vulnerabilities and cost by service, with rightsizing proposals.
Customization & Client Adaptation
Research workloads behave differently from a web product, so operations were tuned to them.
Research-Aware Alerting
Worker alarms watch queue depth and job age, so a stuck pipeline is caught before scientists notice.
Fargate Spot for Batch
Interruptible batch workers run on Fargate Spot, with on-demand capacity kept for user-facing services.
Vulnerabilities with Owners
Image scan findings are fixed in the monthly update or tracked with an owner.
No Kubernetes Required
The team kept the simplicity of Amazon ECS and gained operations without changing platforms.
If you find this case interesting, we recommend taking a closer look at
The Results
Overnight job failures are now handled before the research day starts, every deployment rolls back automatically when alarms fire, and moving batch workers to Fargate Spot reduced compute cost by about a third.
Long-Term Value
- Images and platform versions kept current every month.
- Runbooks that make incidents routine.
- Cost by service reviewed every month.
Business Impact
- Scientists no longer lose mornings to failed jobs.
- Releases ship without half-finished deployments.
- Security scan findings no longer pile up.
