Redis: Single Replica to Redis in High Availability (HA) Mode Migration - Cognigy Documentation

Prepare Redis and Redis persistent HA Configuration

Before upgrading to Cognigy.AI 4.65, you must perform the following steps and modify the values.yaml of your Cognigy.AI Helm Release accordingly.

Self-Managed Redis Installations

If you do not use Redis and Redis persistent services delivered with AI Helm Chart and prefer self-managed external Redis services instead, ensure that you have the following variables in your values.yaml:

statefulRedis:
  enabled: false

statefulRedisPersistent:
  enabled: false

1. To continue using self-managed Redis installations, disable new Redis and Redis persistent HA deployments with the following setting in values.yaml:

redisHa:
     enabled: false

redisPersistentHa:
     enabled: false

2. Remove the statefulRedis and statefulRedisPersistent sections from values.yaml and skip the remaining steps in this migration guide.

Cloud Infrastructure Configuration

1. Redis and Redis persistent in HA mode are provisioned with 3 replicas to increase service availability. Ensure that your Kubernetes cluster has enough free capacity for additional Redis and Redis persistent pods in HA setups. In total, both configurations require an additional provision of 3 CPU cores and 3 GB RAM in the cluster.
2. Check reclaimPolicy for the existing redis-persistent StorageClass with the following command:

kubectl get storageclass redis-persistent -o yaml

If reclaimPolicy: Delete, you can skip the Persistent Volume Clean-up section of this guide. If reclaimPolicy: Retain is set, you must manually clean up the Persistent Volume and the underlying disk as described in the Persistent Volume Clean-up section. 3. Redis HA default settings imply that you run your cluster with 3 Availability Zones (AZ) (Cognigy.AI recommended setup), and the Helm Release spawns HA replicas across 3 AZs. If your cluster is provisioned without Availability Zones, override zonal podAntiAffinity in values.yaml:

redisHa:
  replica:
    affinity:
      podAntiAffinity:
        requiredDuringSchedulingIgnoredDuringExecution: []

redisPersistentHa:
  replica:
    affinity:
      podAntiAffinity:
        requiredDuringSchedulingIgnoredDuringExecution: []

4. If your cloud provider is neither AWS nor Azure, create redis-persistent-ha StorageClass manually. Set all the parameters for the new redis-persistent-ha StorageClass equal to the existing redis-persistent StorageClass:

• Get existing redis-persistent StorageClass and store it in the redis-persistent-ha.yaml file:

kubectl get storageclass redis-persistent -o yaml > redis-persistent-ha.yaml

• Open the redis-persistent-ha.yaml file in a text editor. Change the name: field to redis-persistent-ha. Remove the uid:, resourceVersion: and creationTimestamp: fields.
  • Save the file and create the new redis-persistent-ha StorageClass by applying it to the cluster: kubectl apply -f redis-persistent-ha.yaml.
  • Check that the new StorageClass is created in the cluster: kubectl get storageclass redis-persistent-ha -o yaml.

5. If your cloud provider is either AWS or Azure, the redis-persistent-ha StorageClass will be created automatically. Before upgrading the Helm Release, ensure that:
   • On AWS: the gp3 storage and the ebs.csi.aws.com provisioner are enabled in your cluster.
   • On Azure: the Premium_LRS storage account type and the disk.csi.azure.com provisioner are enabled in your cluster.
   • Alternatively, you can override redisPersistentHa settings under storageClass: section to match the parameters of the existing redis-persistent StorageClass.

Migrate Custom Redis and Redis Persistent Configuration

• If you do not have any custom configuration under the statefulRedis and statefulRedisPersistent sections in your Cognigy.AI Helm Release values.yaml, skip this section.
• If you have a custom configuration under the statefulRedis and/or statefulRedisPersistent sections, you need to copy it under redisHa and redisPersistentHa respectively as follows:
  • If statefulRedis.auth.password is defined in cleartext, copy its value under redisHa.auth.password.
  • If statefulRedisPersistent.auth.password is defined in cleartext, copy its value under redisPersistentHa.auth.password.
  • If a custom statefulRedis.auth.existingSecret is defined, copy its value under redisHa.auth.existingSecret. Ensure that the corresponding custom secret exists in the cluster.
  • If a custom statefulRedisPersistent.auth.existingSecret is defined, copy its value under redisPersistentHa.auth.existingSecret. Ensure that the corresponding custom secret exists in the cluster.
  • If custom resources for statefulRedis are defined in your values.yaml, copy the resources section (including both requests and limits) to redisHa.replica.resources. Set the maxmemory setting under the redisHa.replica.configuration parameter to 85% of resources.limits.memory.
  • If custom resources for statefulRedisPersistent are defined in your values.yaml, copy the resources section (including both requests and limits) to redisPersistentHa.replica.resources.

Upgrade Cognigy.AI Helm Release to 4.65

1. Double check that all parameters in your values.yaml are adjusted as described above.
2. Perform the upgrade of the Cognigy.AI Helm Release to 4.65 as usual. During the upgrade:
   • New redis-ha-node and redis-persistent-ha-node StatefulSets, along with their corresponding pods, will be created in the cluster.
   • Old redis and redis-persistent Deployments and corresponding pods will be removed from the cluster.
   • Cognigy.AI services will reconnect to Redis and Redis persistent HA setups.
   • Verify that all the pods are running as expected by executing: kubectl get pods -n=cognigy-ai.

Persistent Volume Clean-up

After upgrading Cognigy.AI to 4.65 and verifying that the release works properly, you can clean up the remaining persistent volume (PV) for the old redis-persistent Deployment:

1. If reclaimPolicy: Delete was set for the old redis-persistent StorageClass, skip this section. The underlying PV and PVC will be deleted automatically.
2. If reclaimPolicy: Retain was set for the old redis-persistent StorageClass, manually remove the PV associated with the old redis-persistent Deployment and the underlying disk in your cloud infrastructure:
   • Get PVs in your cluster: kubectl get pv.
   • Write down the name of the PV in the Released state and CLAIM: cognigy-ai/redis-persistent (referenced below as PV_NAME).
   • Delete the pv with kubectl delete pv PV_NAME.
   • Delete the storage disk corresponding to PV_NAME in your cloud environment.