Azure

​

Components and Sizing Recommendations

​

Component: AI Gateway

• Deployment: Deploy as a Docker container in your Kubernetes cluster using Helm Charts.
• Instance Type: Azure Standard B2ms instance, with at least 4GiB of memory and two vCPUs.
• High Availability: Deploy across multiple zones for high reliability.

​

Component: Logs Store (optional)

• Options: Azure Cosmos DB, Azure Blob Storage.
• Sizing: Each log document is ~10kb in size (uncompressed).

​

Component: Cache (Prompts, Configs & Virtual Keys)

• Options: Azure Cache for Redis or self-hosted Redis.
• Deployment: Deploy in the same VNet as the Portkey Gateway.

​

Deployment Steps

​

Prerequisites

Ensure the following tools are installed:

• Docker
• Kubectl
• Helm (v3 or above)

​

Step 1: Clone the Portkey Repo Containing Helm Chart

git clone https://github.com/Portkey-AI/helm-chart

​

Step 2: Update values.yaml for Helm

Modify the values.yaml file in the Helm chart directory to include the Docker registry credentials and necessary environment variables. You can find the sample file at ./helm-chart/helm/enterprise/values.yaml.

Image Credentials Configuration

imageCredentials:
  name: portkey-enterprise-registry-credentials
  create: true
  registry: kubernetes.io/dockerconfigjson
  username:
  password:

The Portkey team will share the credentials for your image.

Environment Variables Configuration

Can be fetched from a secrets store

environment:
  data:
    SERVICE_NAME: "gateway_enterprise"
    PORTKEY_CLIENT_AUTH: ""
    PORTKEY_ORGANISATION_ID: ""
    ORGANISATIONS_TO_SYNC: ""
    LOG_STORE: ""

    # If you're using Cosmos DB as Logs Store
    COSMOS_DB_CONNECTION_URL: ""
    COSMOS_DATABASE: ""
    COSMOS_COLLECTION_NAME: ""

    # If you're using Azure Blob Storage as Logs Store
    AZURE_STORAGE_ACCOUNT_NAME: ""
    AZURE_STORAGE_ACCOUNT_KEY: ""
    AZURE_CONTAINER_NAME: ""

    # Analytics Store credentials shared by Portkey
    ANALYTICS_STORE: ""
    ANALYTICS_STORE_USER: ""
    ANALYTICS_STORE_PASSWORD: ""
    ANALYTICS_STORE_HOST: ""
    ANALYTICS_STORE_TABLE: ""
    ANALYTICS_STORE_FEEDBACK_TABLE: ""

    # Your cache details
    CACHE_STORE: ""
    REDIS_URL: ""
    REDIS_TLS_ENABLED: ""

    # For semantic cache, when using Pinecone
    PINECONE_NAMESPACE: ""
    PINECONE_INDEX_HOST: ""
    PINECONE_API_KEY: ""
    SEMCACHE_OPENAI_EMBEDDINGS_API_KEY: ""
    SEMCACHE_OPENAI_EMBEDDINGS_MODEL: "text-embedding-3-small"

Notes on the Log Store

LOG_STORE can be:

• s3, for Azure Blob Storage
• mongo, for Azure Cosmos DB

If the LOG_STORE is mongo, the following environment variables are needed:

MONGO_DB_CONNECTION_URL:
MONGO_DATABASE:
MONGO_COLLECTION_NAME:

If the LOG_STORE is s3, the following values are mandatory:

LOG_STORE_REGION:
LOG_STORE_ACCESS_KEY:
LOG_STORE_SECRET_KEY:
LOG_STORE_GENERATIONS_BUCKET:

You need to generate the Access Key and Secret Key from the respective providers.

Notes on Cache

If CACHE_STORE is set as redis, a Redis instance will also get deployed in the cluster. If you are using custom Redis, then leave it blank. The following values are mandatory:

REDIS_URL:
REDIS_TLS_ENABLED:

REDIS_URL defaults to redis://redis:6379 and REDIS_TLS_ENABLED defaults to false.

Notes on Analytics Store

This is hosted in Portkey’s control plane and these credentials will be shared by the Portkey team.

The following are mandatory and are shared by the Portkey Team.

PORTKEY_CLIENT_AUTH:
ORGANISATIONS_TO_SYNC:

​

Step 3: Deploy Using Helm Charts

Navigate to the directory containing your Helm chart and run the following command to deploy the application:

helm install portkey-gateway ./helm/enterprise --namespace portkeyai --create-namespace

This command installs the Helm chart into the _portkeyai_ namespace.

​

Step 4: Verify the Deployment

Check the status of your deployment to ensure everything is running correctly:

kubectl get pods -n portkeyai

​

Step 5: Port Forwarding (Optional)

To access the service over the internet, use port forwarding:

kubectl port-forward  -n portkeyai 443:8787

Replace _<pod-name>_ with the name of your pod.

​

Uninstalling the Deployment

If you need to remove the deployment, run:

helm uninstall portkey-app --namespace portkeyai

This command will uninstall the Helm release and clean up the resources.

​

Network Configuration

​

Step 1: Allow Access to the Service

To make the service accessible from outside the cluster, define a Service of type LoadBalancer in your values.yaml or Helm templates. Specify the desired port for external access.

service:
  type: LoadBalancer
  port:
  targetPort: 8787

Replace _<desiredport>_ with the port number for external access with the port the application listens on internally.

​

Step 2: Ensure Outbound Network Access

By default, Kubernetes allows full outbound access, but if your cluster has NetworkPolicies that restrict egress, configure them to allow outbound traffic.

Example NetworkPolicy for Outbound Access:

apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: allow-all-egress
  namespace: portkeyai
spec:
  podSelector: {}
  policyTypes:
  - Egress
  egress:
  - to:
    - ipBlock:
        cidr: 0.0.0.0/0

This allows the gateway to access LLMs hosted within your VNet and outside as well. This also enables connection for the sync service to the Portkey Control Plane.

​

Step 3: Configure Inbound Access for Portkey Control Plane

Ensure the Portkey control plane can access the service either over the internet or through VNet peering.

Over the Internet:

• Ensure the LoadBalancer security group allows inbound traffic on the specified port.
• Document the public IP/hostname and port for the control plane connection.

Through VNet Peering:

• Set up VNet peering between your Azure account and the control plane’s Azure account. Requires manual setup by Portkey Team.

This guide provides the necessary steps and configurations to deploy Portkey on Azure effectively, ensuring high availability, scalability, and integration with your existing infrastructure.