Cohestra vs AWS Managed Service for Apache Flink

Cohestra is a drop-in replacement for AWS MSF (Managed Service for Apache Flink) and similar managed Flink services. You bring your own Kubernetes cluster; Cohestra provides the management layer.

Feature Comparison

Feature	AWS MSF	Cohestra
Infrastructure	AWS-managed, no cluster access	Any Kubernetes (EKS, GKE, AKS, on-prem)
Flink Version	Managed, often months behind	Any version — you control the image
Deployment	AWS Console / CloudFormation	REST API / SDK / GitOps / CI pipeline
Rollback	Manual redeploy from console	One-command with savepoint preservation
Autoscaling	Basic KPU-based	Custom SDK — Kafka lag, CPU, any metric
State Management	Opaque S3 buckets	You own checkpoint/savepoint storage
Operation History	CloudWatch logs	Durable Temporal workflows with full audit
Cluster Freeze	Not available	Namespace-level mutation freeze
Health Gates	Basic	Checkpoint, restart, backpressure, lag, sink
Multi-Cloud	AWS only	Any cloud or on-prem
Cost	Per-KPU pricing (~$0.11/KPU-hour)	Kubernetes node cost only
License	Proprietary	Apache 2.0
Vendor Lock-in	High	None

Why Replace MSF?

1. Cost at Scale

MSF charges per KPU (Kinesis Processing Unit). A modest job with 8 KPUs costs ~$630/month. At 50 jobs, that's $31,500/month in MSF fees alone — before data transfer and storage.

With Cohestra on EKS, the same workloads run on your existing nodes. Typical savings: 60-80% at scale.

2. Flink Version Control

MSF locks you to specific Flink versions. When Flink 2.x shipped with major performance improvements, MSF users waited months. With Cohestra, update your Docker image and deploy.

3. Custom Autoscaling

MSF's autoscaling is a black box. Cohestra's Autoscaler SDK lets you build autoscalers that react to your actual metrics — Kafka consumer lag from CloudWatch or Confluent, TaskManager CPU, custom business metrics.

4. Operational Transparency

Every Cohestra operation is a durable Temporal workflow. You can query any deployment's full history, see exactly what savepoint was used for a rollback, and audit who approved a production deploy. MSF gives you CloudWatch logs.

5. No Vendor Lock-in

Switching away from MSF means rewriting CloudFormation templates, migration tooling, and operational runbooks. Cohestra runs on standard Kubernetes — move between EKS, GKE, and on-prem with the same Helm chart and API.

Migration Guide

From MSF to Cohestra on EKS

1. Set up Cohestra — Install the Helm chart on your Kubernetes cluster
2. Containerize your Flink job — Build a Docker image with your JAR
3. Register with Cohestra — PUT /api/v1/deployments/{env}/{ns}/{name}
4. Deploy — POST .../deploy with your image digest
5. Set up autoscaling — Replace KPU scaling with custom autoscaler
6. Decommission MSF — Delete MSF application after verification

Mapping MSF Concepts to Cohestra

MSF Concept	Cohestra Equivalent
Application	Deployment (registered via PUT)
Snapshot	Savepoint (via POST .../savepoint)
Application update	Deploy (via POST .../deploy)
Scaling	Scale (via POST .../scale)
CloudWatch metrics	Health summary (via GET .../actor)
KPU autoscaling	Custom autoscaler SDK