Platform Setup Guide

This guide describes how to configure Michelangelo server components in Kubernetes cluster. It focuses on the configuration surfaces (ConfigMaps, fields, and key parameters).

Overview

Michelangelo consists of four core server components:

1. API Server – Central gRPC API
2. Controller Manager – Kubernetes controllers
3. Worker – Workflow execution (Temporal workers + compute integration)
4. UI + Envoy – Frontend and proxy

Each component exposes server-side configuration through ConfigMaps and overlays.

This document explains:

• Where each component's configuration lives
• What fields can be customized
• What each field means
• How to apply changes using Kustomize overlays

Michelangelo Service architecture diagram

The following diagram shows the relationship between each of the services in Michelangelo eco-system.

Server Configuration

API Server Configuration

Key Fields

apiserver:
  yarpc:
    host: 0.0.0.0
    port: 15566
  k8s:
    qps: 300
    burst: 600
  metadataStorage:
    enableMetadataStorage: false
  crdSync:
    enableCRDUpdate: true
    enableIncompatibleUpdate: false

Field Explanations

Field	Description
yarpc.host/port	gRPC bind address + port
k8s.qps/burst	Throttling limits for Kubernetes API calls
enableMetadataStorage	Enables metadata persistence
enableCRDUpdate	Controls whether CRDs can be sync'd
enableIncompatibleUpdate	Allows breaking CRD changes (use only during major migrations)

Controller Manager Configuration

Key Fields

controllermgr:
  metricsBindAddress: 8091
  healthProbeBindAddress: 8081
  leaderElection: false
  leaderElectionID: michelangelo.your-organization.com
  port: 9443

controllers:
  rayCluster:
    k8sQps: 300
    k8sBurst: 600

minio:
  awsRegion: ap-southeast-1
  awsEndpointUrl: s3.ap-southeast-1.amazonaws.com
  useIam: true

workflowClient:
  service: temporal-frontend
  host: temporal.your-domain.com:7233
  transport: grpc
  domain: uniflow

Field Explanations

Field	Description
metricsBindAddress	Controller metrics port
healthProbeBindAddress	Health check port
leaderElection	Enable for production HA
minio.*	S3 / MinIO backend configuration
workflowClient.*	Temporal client configuration
controllers.*	Each controller components' configuration

Worker Configuration

Key Fields

worker:
  address: michelangelo-apiserver.your-domain.com:443
  maApiServiceName: ma-apiserver
  useTLS: true

logging:
  level: info
  development: true
  encoding: console

workflow-engine:
  host: temporal.your-domain.com:7233
  transport: grpc
  provider: temporal
  workers:
    - domain: default
      taskList: production-uniflow
  client:
    domain: uniflow

Field Explanations

Field	Description
worker.address	API server endpoint used by workers
workflow-engine.host	Temporal endpoint
workers[].taskList	Worker task list to poll
client.domain	Temporal workflow domain

UI & Envoy Configuration

Envoy Proxy

ConfigMap:

static_resources:
  listeners:
    - address:
        socket_address: { address: 0.0.0.0, port_value: 8081 }

  clusters:
    - name: michelangelo-apiserver
      load_assignment:
        endpoints:
          - lb_endpoints:
              - endpoint:
                  address:
                    socket_address:
                      address: michelangelo-apiserver
                      port_value: 15566

Public UI Config

ConfigMap:

config.json: |
  {
    "apiBaseUrl": "https://michelangelo-envoy.<your-domain>"
  }

Environment Overrides / Domain Settings

You must customize domain-specific values in overlays:

Location	Fields to Update
Worker ConfigMap	API server domain, compute domain, Temporal host
UI Public Config	"apiBaseUrl"
Envoy Config	CORS allowed origins, API cluster hostname
Ingress	Hostnames for API server & UI
Controller Manager	S3 region, endpoint, Temporal host

Object Store Configuration

Object storage (MinIO / S3) is used by Michelangelo for artifacts and metadata.

Controller Manager Object Store Settings

These live in the controller manager ConfigMap:

minio:
  awsRegion: sample   # AWS region 
  awsEndpointUrl: sample.amazonaws.com
  useIam: true                # Use IAM roles for authentication

Fields

• awsRegion – The AWS region of your S3 bucket.
• awsEndpointUrl – S3 endpoint (s3.amazonaws.com or regional endpoint).
• useIam – Set to true in production (do not hardcode keys in config).

Storage Setup Checklist (from original guide)

• Configure AWS credentials/IAM roles for pods that need S3 access.
• Verify region and endpoint in the ConfigMap match your S3 setup.
• Test connectivity from worker/controller pods to the bucket.

Workflow Engine Configuration (Temporal/Cadence)

Michelangelo uses a workflow engine (Temporal or Cadence) for orchestrating workflows. Most of your current guide examples use Temporal, and Cadence is used in sandbox/dev.

Controller Manager Workflow Client

From controllermgr-configmap.yaml:

workflowClient:
  service: temporal-frontend    # Temporal service name
  host: temporal.your-domain.com:7233  # Temporal endpoint
  transport: grpc               # Transport protocol
  domain: uniflow               # Temporal domain

Fields

• service – Workflow engine frontend service name (temporal-frontend / cadence-frontend).
• host – Full endpoint (host:port).
• transport – Typically grpc.
• domain – Temporal domain (or Cadence domain) to target.

Worker Workflow Engine Settings

From worker-configmap.yaml:

workflow-engine:
  host: temporal(/cadence).your-domain.com:7233
  transport: grpc
  provider: temporal/cadence
  workers:
    - domain: default
      taskList: production-uniflow
  client:
    domain: uniflow

Fields

• provider – temporal (or cadence); can be extended to cadence if needed.
• host – Temporal/Cadence endpoint.
• workers[].domain – Domain where worker polls for tasks.
• workers[].taskList – Task list (queue) used for workflow tasks.
• client.domain – Client domain for starting workflows.

Temporal Setup (from original external dependencies)

• Ensure Temporal is accessible at the configured endpoint.
• Create required domains (uniflow, default, production-uniflow).
• Configure task lists such as production-uniflow.