Skip to main content

Helm Production Setup and HA

The chart generates every credential on first install and keeps them in one Secret (Credentials), so a production values file carries no passwords at all. Set them explicitly only when the values must come from your own secret store — or when you render manifests outside the cluster (GitOps), where generation cannot work:

# Only if you are supplying it yourself. Command substitution does not run inside a values file,
# so pass it at install time: --set-string global.jwt.secret="$JWT_SECRET"
JWT_SECRET=$(openssl rand -base64 32)

A reference values file for production-grade deployments:

global:
mainAddress: "elchi.company.com"
tlsEnabled: true
jwt:
# secret: "" # omit → generated once and kept in elchi-stack-secrets
accessTokenDuration: "1h"
refreshTokenDuration: "24h"
elchiBackend:
controlPlaneDefaultReplicas: 3
controllerDefaultReplicas: 3
versions:
- tag: v1.8.4-v0.14.0-envoy1.40.2

# The defaults run one replica of each, sized for a first try
envoy:
replicas: 2
elchi-collector:
replicaCount: 2

# Resource limits
elchi:
replicas: 3
resources:
requests:
memory: "512Mi"
cpu: "500m"
limits:
memory: "1Gi"
cpu: "1000m"

High availability​

  • Point global.mainAddress at a load balancer or VIP in front of every node, by DNS name. See Choosing mainAddress.
  • Run at least 3 replicas for every critical component.
  • Configure pod anti-affinity rules to spread replicas across nodes.
  • Use an external MongoDB replica set for durable persistence.
  • Hand metrics off to an external VictoriaMetrics cluster.
  • Set both resource requests and limits on every workload.