Kubernetes Cluster Prerequisites

Estimated reading: 3 minutes

This section outlines the Kubernetes infrastructure requirements necessary for deploying and operating Robility Flow in a containerized environment. Robility Flow is designed to run on industry-standard Kubernetes platforms, enabling scalability, high availability, workload isolation, and simplified application lifecycle management. Prior to deployment, the Kubernetes cluster must meet the following version, sizing, storage, and operating system requirements.

Supported Kubernetes Distributions

Robility Flow has been validated and certified on the following Kubernetes distributions. The minimum supported Kubernetes version is 1.27.

1. Amazon Elastic Kubernetes Service (EKS) – 1.27 or higher
2. Azure Kubernetes Service (AKS) – 1.27 or higher
3. Google Kubernetes Engine (GKE) – Standard or Autopilot, 1.27 or higher

Pilot / Proof-of-Concept (PoC) Sizing

For pilot, evaluation, and proof-of-concept deployments, a small Kubernetes cluster is sufficient to host Robility Flow services and perform functional validation. A single-node cluster may be used for demonstration purposes; however, a two-node cluster is recommended to provide better resource availability and operational flexibility.

1. Worker nodes: 2 nodes, each with 8 vCPU and 32 GB RAM
2. Storage: 1 storage class backed by SSD-class persistent volumes (minimum 250 GB provisioned)
3. Operating system: Linux (Ubuntu 22.04 LTS, RHEL 9, or Amazon Linux 2023 recommended)

Kubernetes resource requests by service.

Service	Replicas (Min → Max)	CPU Request	Memory Request	Notes
RobilityFlow Designer	1 → 3	2 vCPU	4 GB	Stateless service and eligible for Horizontal Pod Autoscaling (HPA).
RobilityFlow Runtime	2 → 10	4 vCPU	8 GB	Supports Horizontal Pod Autoscaling based on CPU utilization and queue depth.
PostgreSQL	1 Primary + 1 Replica	4 vCPU	8 GB	Deployed as a StatefulSet with Persistent Volume Claims (PVC) of 100 GB or higher.
Redis	1 (or 3-Node Cluster)	1 vCPU	2 GB	Requires a 10 GB PVC when AOF (Append Only File) persistence is enabled.
Ingress Controller	2	0.5 vCPU	256 MB	Configured with anti-affinity rules to ensure replicas are distributed across different nodes for high availability.

Production Sizing

Production deployments should target a minimum of 3 worker nodes spread across availability zones. Horizontal Pod Autoscaler (HPA) is pre-configured for the Runtime and Designer deployments.

Service	Replicas	CPU Limit	Memory Limit	Notes
RobilityFlow Designer	3	2 vCPU	4 GB	CDN-cached static assets recommended.
RobilityFlow Runtime	4 – 20 (HPA)	4 vCPU	8 GB	Scale on CPU utilization ≥ 70% or based on queue depth.
PostgreSQL (Primary)	1	8 vCPU	32 GB	PVC 500 GB SSD; enable WAL archiving for backup and recovery.
PostgreSQL (Replica)	1 – 2	4 vCPU	16 GB	Read replica used for reporting and analytical queries.
Redis	3-Node Cluster	2 vCPU	4 GB	Cluster mode with AOF persistence and RDB snapshots enabled.
Ingress Controller	3	1 vCPU	512 MB	Replicas should be distributed across availability zones for high availability.

Cluster autoscaler or Karpenter (AWS) is recommended for production so that Runtime pods can scale out automatically during peak flow execution periods.

Persistent Storage Classes

At least one Kubernetes StorageClass backed by a block-storage provider must be available and set as default. Required characteristics:

1. Access mode: ReadWriteOnce (RWO) for PostgreSQL and Redis PersistentVolumeClaims
2. Reclaim policy: Retain (recommended for stateful workloads)
3. Volume type: SSD-backed (gp3 on AWS, Premium_LRS on Azure, pd-ssd on GCP)
4. Dynamic provisioning: Enabled via a CSI driver (EBS CSI, Azure Disk CSI, GCP PD CSI)

Container Image Registry

RobilityFlow platform images are distributed through the Robility Hub registry. The cluster must be able to pull images from this registry, or images must be mirrored to a customer-managed OCI registry.

1. Robility Hub registry: https://robilityai.azurecr.io
2. Image pull secrets must be configured in the robilityflow Kubernetes namespace
3. If using a private mirror, ensure images are synced before installation and on each platform update

Kubernetes Cluster Prerequisites

Supported Kubernetes Distributions

Pilot / Proof-of-Concept (PoC) Sizing

Production Sizing

Persistent Storage Classes

Container Image Registry

Kubernetes Cluster Prerequisites

CONTENTS