High Availability

Sharding is used to distribute workload by installing multiple Navecd Controllers and Navecd uses Leases to make sure that there is only one running Controller per Shard.

While that works for most cases, sometimes it is desired to have backup instances running on different nodes to reduce downtime in case of node failures. You can use Pod Anti-Affinity and increase the replica count of your Navecd Controllers to counter that situation:

navecd/primary_system.cue

package navecd

import (
  "github.com/kharf/navecd/schema/component"
)

primaryProjectControllerDeployment: component.#Manifest & {
  ...
  content: {
    apiVersion: "apps/v1"
    kind:       "Deployment"
    metadata: {
      name:      "project-controller-primary"
      namespace: ns.content.metadata.name
      labels:    _primaryLabels
    }
    spec: {
      selector: matchLabels: _primaryLabels
      replicas: 2
  ...
    }
  }
}

navecd/ha.cue

package navecd

import (
  "github.com/kharf/navecd/schema/component"
)

primaryProjectControllerDeployment: component.#Manifest & {
  content: spec: template: spec: affinity: podAntiAffinity: {
    requiredDuringSchedulingIgnoredDuringExecution: [{
      labelSelector: matchExpressions: [{
        key:      _shardKey
        operator: "In"
        values: ["primary"]
      }]
      topologyKey: "kubernetes.io/hostname"
    }]
  }
}