📦 Install with Helm

The fastest way to run Mars in a cluster is the official Helm chart — it packages the server, its configuration and (optionally) Redis for you. Two flavors are available:

Mode	Default	What you get
standalone	✅	One replica, SQLite database, in-memory queue. Zero external dependencies — perfect for evaluation and small teams.
cluster	—	Multiple replicas, MySQL database, Redis-backed message queue. For high availability.

Prerequisites

• kubectl configured against your cluster
• helm v3+
• For cluster mode: an external MySQL reachable from inside the cluster

1. Add the chart repository

helm repo add mars-charts https://duc-cnzj.github.io/mars-charts/
helm repo update
helm search repo mars-charts        # should list mars-charts/mars

2. Prepare your values

Start from the chart's defaults and override what you need:

helm show values mars-charts/mars > mars-values.yaml

The chart turns a standalone_conf / cluster_conf block into the app's config.yaml automatically. At minimum, set the mode and your own admin password:

image:
  repository: registry.cn-hangzhou.aliyuncs.com/duc-cnzj/mars
  tag: "latest"              # pin a release instead of latest

mode: "standalone"           # or "cluster"

standalone_conf: |
  external_ip: "1.2.3.4"             # your cluster's externally reachable IP
  admin_password: "change-me"        # 🔒 change the default!
  kubeconfig: ""                     # empty = use the in-cluster config
  git_server_plugin:
    name: gitlab
    args:
      token: "your-gitlab-token"     # GitLab → Settings → Access Tokens
      baseurl: "https://gitlab.com/api/v4"

3. Install

helm upgrade --install mars mars-charts/mars \
  -n mars --create-namespace \
  -f mars-values.yaml

Watch it come up:

kubectl get pods -n mars -w

4. Open the web UI

Access	values	Result
Port-forward (default ClusterIP)	service.type: ClusterIP	kubectl port-forward -n mars svc/mars 4000:6000 → open http://127.0.0.1:4000
NodePort	service.type: NodePort	http://<node-ip>:<node-port>
LoadBalancer	service.type: LoadBalancer	http://<lb-ip>
Ingress	ingress.enabled: true + hosts	http://your-domain

Log in with admin and the admin_password you set.

Cluster mode (high availability)

Two replicas, the bundled Redis for the message queue, and an external MySQL for storage:

mode: "cluster"
replicaCount: 2

# bundled Redis (bitnami)
redis:
  enabled: true
  auth:
    password: "mars-redis-password"
  master:
    persistence:
      enabled: false
  replica:
    replicaCount: 0

cluster_conf: |
  db_driver: 'mysql'
  db_host: your-mysql-host
  db_port: 3306
  db_username: root
  db_password: ""
  db_database: mars

  ws_sender_plugin:
    name: ws_sender_redis
    args:
      addr: "{{.Release.Name}}-redis-master.{{.Release.Namespace}}.svc.cluster.local:6379"
      password: "mars-redis-password"
      db: 1

TIP

cluster_conf is rendered as a Go template — you can reference release metadata with {{.Release.Name}} and {{.Release.Namespace}}, exactly like the ws_sender_redis address above.

Storage

By default the chart mounts an emptyDir — the SQLite database and uploads are lost when the pod restarts. Keep them with a PVC:

persistent:
  enabled: true
  storage: 10Gi
  # storageClassName: your-class

If uploads go to S3 instead, persistence can stay off.

Expose gRPC (optional)

For the SDK / CLI over gRPC:

grpc:
  enabled: true
  service:
    enabled: true
    type: ClusterIP
  ingress:
    enabled: true
    annotations:
      nginx.ingress.kubernetes.io/backend-protocol: "GRPC"
    hosts:
      - host: grpc.mars.example.com

Canary releases (advanced)

The chart can run a canary Deployment/Service/Ingress alongside the stable one — a new image tag receives a slice of traffic based on nginx-ingress rules (header, cookie or weight). Requires an nginx-ingress controller and ingress.enabled: true:

canary:
  enabled: true
  image:
    tag: "next"              # new version
  replicas: 1
  ingress:
    byHeader: "x-canary"     # route when this header == "always"
    weight: "10"             # or route 10% of traffic

Upgrade

helm repo update
helm upgrade --install mars mars-charts/mars -n mars -f mars-values.yaml

The pod is recreated when the config checksum changes.

Security notes

• The chart binds its ServiceAccount to cluster-admin — Mars needs cluster-wide permissions to deploy apps and watch resources for you. Don't run it in a cluster you don't fully trust.
• Change the default admin_password (default 123456).
• Provide a private_key (RSA) — Mars uses it to sign access tokens. The example values include a working one; generate your own.
• Prometheus scraping is enabled by default on port 9091 (prometheus.io/scrape: "true").