flyte-binary Helm chart. The result is a single-cluster, production-capable Flyte installation backed by S3 and RDS.
Prerequisites
EKS cluster
EKS cluster
An EKS cluster with at least 3 nodes (e.g.,
m5.xlarge). Flyte’s control-plane components are lightweight but tasks need node capacity.S3 bucket
S3 bucket
Create one or two S3 buckets — one for Flyte metadata and one for user task data (they can be the same bucket):
RDS PostgreSQL
RDS PostgreSQL
Create a PostgreSQL 13+ RDS instance in the same VPC as your EKS cluster. Note the endpoint, username, password, and database name.The Flyte database should be named
flyteadmin by default.IAM roles (IRSA)
IAM roles (IRSA)
Flyte uses IAM Roles for Service Accounts (IRSA) so that pods can access S3 without static credentials.You need two IAM roles:
- Backend role (
FLYTE_BACKEND_IAM_ARN): used by theflyte-binarypod itself to access S3 for metadata - User role (
FLYTE_USER_IAM_ARN): assumed by task pods in each project namespace
s3:GetObject, s3:PutObject, s3:DeleteObject, and s3:ListBucket on your buckets.Enable OIDC on your cluster and create the roles with eksctl:Installation
EKS values reference
This is the completeeks-starter.yaml with all required fields:
Verify the installation
Configure flytectl
IAM policy reference
The backend IAM role (attached to theflyte-binary service account) needs at minimum:
What’s next
- Enable authentication with your identity provider
- Deploy Spark or Ray plugins
- Configure secrets management