EC2 Managed Node Group Support

[sashaneb]

I wanted to share some findings from our deployment that might help others. We noticed the module currently uses Fargate exclusively, and after analyzing our costs, we think adding EC2 managed node group support could significantly reduce hosting costs for steady-state deployments.

What We Found

For our minimal Langfuse deployment, Fargate compute costs run around ~$520/month:

| Component             | vCPU | Memory | Fargate Cost/mo |
|-----------------------|------|--------|-----------------|
| ClickHouse × 3        | 6    | 24 GB  | ~$270           |
| Zookeeper × 3         | 3    | 6 GB   | ~$110           |
| Langfuse web + worker | 4    | 8 GB   | ~$140           |

The same workload on EC2 (2× t3.large) would cost around ~$122/month - roughly 75% savings.

Why EC2 Might Be a Better Default

Fargate is excellent for bursty, unpredictable, or scale-to-zero workloads. However, Langfuse runs ClickHouse, Zookeeper, and web/worker pods continuously, which means:

• Bin-packing benefits: EC2 nodes can run multiple pods efficiently sharing resources, while Fargate isolates each pod in its own micro-VM
• Autoscaling opportunities: EC2 node groups work with Cluster Autoscaler or Karpenter, enabling scale-down during quiet periods
• Spot instance support: For non-production or fault-tolerant setups, Spot instances can reduce costs by another 60-70%

Would you be open to adding EC2 managed node group support as an option? Something like:

variable "compute_type" {
    description = "EKS compute type: 'fargate' or 'ec2'"
    type        = string
    default     = "ec2"
}

[Steffen911]

@sashaneb Thank you for the suggestion and the supporting data. We'd be happy to accept community contributions that add support for different compute types. We'd prefer to keep ECS Fargate as the default though as this keeps operational load to a minimum.