Skip to content

Latest commit

 

History

History
361 lines (293 loc) · 13.7 KB

README.md

File metadata and controls

361 lines (293 loc) · 13.7 KB
Raw

Haystack RAG Application Helm Chart

This Helm chart deploys the Haystack reference RAG application.

Architecture

The application consists of several components:

API Gateway

  • NGINX-based reverse proxy
  • Serves static files (Reach build, CSS, images)
  • Routes traffic to indexing and query services

Search layer

  • OpenSearch: Document storage and search engine

    • Single-node deployment
    • Secure authentication

Backend services

  • Indexing Service: Document processing

    • File upload handling
    • Document indexing
    • Persistent storage for uploads

  • Query Service: Search operations

    • RAG pipeline
    • OpenAI integration
    • Configurable embedding options
    • Read-only access to persistent storage

Frontend

  • Simple React-based UI
  • File upload interface
  • Search functionality

Prerequisites

  • Kubernetes 1.31+
  • Helm 3.16+
  • Storage class with ReadWriteOnce support
  • OpenAI API key

Installation

Quick start

  1. Edit values file for your environment:
global:
  environment: production
  image:
    registryPath: your-registry
    pullPolicy: Always
  secrets:
    useExternalSecrets: true
    name: "hra-secrets"
..
  1. Add external secrets (or edit values.yaml):
kubectl create secret generic hra-secrets \
  --from-literal=opensearch-user=admin \
  --from-literal=opensearch-password=<password> \
  --from-literal=openai-api-key=<key>

  1. Check the persistent storage configuration in values.yaml

  2. Check gateway/ingress configuration in values.yaml

  3. Install the chart:

helm install hra . -f values.yaml

Configuration reference

Global settings

global:
  environment: development
  image:
    registryPath: your-registry
    pullPolicy: Always
  secrets:
  ..
  openai:
    apiKey: your-api-key
..

Component settings

See values.yaml and the table below for detailed configuration options for:

  • OpenSearch deployment
  • Backend services (indexing and query)
  • Frontend service
  • API Gateway
  • GKE Gateway configuration (for GKE environments)
  • Ingress configuration

Please note that the config section in values.yaml is used to configure the backend services.

search:
  opensearch:
    enabled: true
    replicas: 1
    # ... other OpenSearch settings

backend:
  indexing:
    enabled: true
    replicas: 2
    # ... other indexing service settings
  query:
    enabled: true
    replicas: 3
    # ... other query service settings
  config:
    llm:
      generator: openai
      useOpenAIEmbedder: false
    tokenizers:
      parallelism: false
    logging:
      level: INFO
      haystackLevel: INFO
    indexing:
      onStartup: true

frontend:
  enabled: true
  replicas: 2
  # ... other frontend settings

# Example GKE Gateway configuration
gkeGateway:
  enabled: true
  routes:
    - path: /
      service: gateway-api-gw
      port: 8080

Storage configuration

persistence:
  fileStorage:
    enabled: true
    size: 10Gi
    storageClass: standard-rwo-regional  # GKE storage class
    accessMode: ReadWriteOnce

Validation

The chart includes schema validation. Validate your values:

helm lint --strict -f values.yaml .

Maintenance

Scaling

Components can be scaled independently:

helm upgrade hra . --set backend.query.replicas=3

Updates

helm upgrade hra . -f values.yaml

Troubleshooting

  1. OpenSearch issues

    • Check memory settings and limits
    • Verify storage class availability
    • Validate security credentials

  2. Service connectivity

    • Check service names and ports
    • Verify API Gateway nginx configuration (locations, timeouts)
    • Check gateway/ingress settings (host, path, service, port, timeouts)

  3. Resource issues

    • Monitor resource usage
    • Ensure ephemeral storage is sufficient
    • Check pod logs
    • Verify PVC status

  4. Health checks

    • Check readiness probes
    • Check liveness probes
    • Check OpenSearch cluster health endpoint

  5. Configuration and secrets

    • Check values.yaml
    • Check external secrets (e.g., OpenAI API key)

Security Considerations

  1. API Keys & secrets

    • Use external secrets management when possible (Vault, AWS Secrets Manager, etc.)
    • Rotate credentials regularly
    • Never commit sensitive values to version control

  2. Improvements to network security

    • Enable TLS for ingress/gateway
    • Configure network policies
    • Use secure OpenSearch configuration
    • Consider using GKE Gateway's built-in security features

Chart Values

Value Type Description Default
global.environment string Environment name (development/production) development
global.image.registryPath string Container registry path gcr.io/your-project-id
global.image.pullPolicy string Image pull policy Always
global.secrets.useExternalSecrets boolean Use external secrets true
global.secrets.name string Name of the secrets "hra-secrets"
global.secrets.opensearch.adminUser string (If not external secrets) OpenSearch username "admin"
global.secrets.opensearch.adminPassword string (If not external secrets) OpenSearch password "your-password-here"
global.secrets.openai.apiKey string (If not external secrets) OpenAI API key "sk-proj-999"
search.opensearch.enabled boolean Enable OpenSearch deployment true
search.opensearch.replicas integer Number of OpenSearch replicas 1
search.opensearch.image.imageName string OpenSearch image name opensearch
search.opensearch.image.tag string OpenSearch image tag "2.18.0"
search.opensearch.image.pullPolicy string Image pull policy IfNotPresent
search.opensearch.securityContext.runAsUser integer User ID to run container 1000
search.opensearch.podSecurityContext.fsGroup integer File system group ID 1000
search.opensearch.service.ports[0].port integer Service port for REST API 9200
search.opensearch.service.ports[0].targetPort integer Container port for REST API 9200
search.opensearch.service.ports[0].name string Port name for REST API rest-api
search.opensearch.resources.requests.memory string Memory request "2Gi"
search.opensearch.resources.requests.cpu string CPU request "500m"
search.opensearch.resources.requests.ephemeral-storage string Ephemeral storage request "2Gi"
search.opensearch.resources.limits.memory string Memory limit "4Gi"
search.opensearch.resources.limits.cpu string CPU limit "1000m"
search.opensearch.resources.limits.ephemeral-storage string Ephemeral storage limit "4Gi"
search.opensearch.volumeClaimTemplates.size string Storage size 30Gi
search.opensearch.volumeClaimTemplates.storageClass string Storage class standard-rwo-regional
search.opensearch.config.javaOpts string Java options "-Xms512m -Xmx512m"
search.opensearch.readinessProbe.tcpSocket.port string Readiness probe port rest-api
search.opensearch.readinessProbe.initialDelaySeconds integer Initial delay for readiness 90
search.opensearch.readinessProbe.periodSeconds integer Period for readiness check 10
search.opensearch.readinessProbe.timeoutSeconds integer Timeout for readiness 10
search.opensearch.readinessProbe.failureThreshold integer Failure threshold for readiness 10
search.opensearch.livenessProbe.tcpSocket.port string Liveness probe port rest-api
search.opensearch.livenessProbe.initialDelaySeconds integer Initial delay for liveness 180
search.opensearch.livenessProbe.periodSeconds integer Period for liveness check 10
search.opensearch.livenessProbe.timeoutSeconds integer Timeout for liveness 5
search.opensearch.livenessProbe.failureThreshold integer Failure threshold for liveness 10
backend.indexing.enabled boolean Enable indexing service true
backend.indexing.replicas integer Number of indexing replicas 1
backend.indexing.image.imageName string Indexing service image name hra-indexing
backend.indexing.image.tag string Indexing service image tag latest
backend.indexing.service.ports[0].port integer Service port 8001
backend.indexing.service.ports[0].targetPort integer Container port 8001
backend.indexing.service.ports[0].name string Port name indexing-api
backend.indexing.service.readinessProbe.httpGet.path string Readiness probe path /health
backend.indexing.service.readinessProbe.httpGet.port string Readiness probe port indexing-api
backend.indexing.service.readinessProbe.initialDelaySeconds integer Initial delay for readiness 30
backend.indexing.service.readinessProbe.periodSeconds integer Period for readiness check 10
backend.indexing.service.readinessProbe.timeoutSeconds integer Timeout for readiness 5
backend.indexing.service.readinessProbe.failureThreshold integer Failure threshold for readiness 18
backend.indexing.resources.requests.memory string Memory request "1Gi"
backend.indexing.resources.requests.cpu string CPU request "250m"
backend.indexing.resources.requests.ephemeral-storage string Ephemeral storage request "10Gi"
backend.indexing.resources.limits.memory string Memory limit "2Gi"
backend.indexing.resources.limits.cpu string CPU limit "500m"
backend.indexing.resources.limits.ephemeral-storage string Ephemeral storage limit "20Gi"
backend.query.enabled boolean Enable query service true
backend.query.replicas integer Number of query replicas 1
backend.query.image.imageName string Query service image name hra-query
backend.query.image.tag string Query service image tag latest
backend.query.service.ports[0].port integer Service port 8002
backend.query.service.ports[0].targetPort integer Container port 8002
backend.query.service.ports[0].name string Port name query-api
backend.query.service.type string Service type ClusterIP
backend.storage.volumeName string Storage volume name file-storage
backend.storage.mountPath string Storage mount path /app/files
backend.config.llm.generator string LLM generator type openai
backend.config.llm.useOpenAIEmbedder boolean Use OpenAI embedder false
backend.config.tokenizers.parallelism boolean Enable tokenizer parallelism false
backend.config.logging.level string Logging level INFO
backend.config.logging.haystackLevel string Haystack logging level INFO
backend.config.indexing.onStartup boolean Index on startup true
frontend.enabled boolean Enable frontend deployment true
frontend.replicas integer Number of frontend replicas 1
frontend.image.imageName string Frontend image name hra-frontend
frontend.image.tag string Frontend image tag latest
frontend.service.ports[].port integer Service port 3000
frontend.service.ports[0].targetPort integer Container port 3000
frontend.service.ports[0].name string Port name react-app
frontend.service.type string Service type ClusterIP
frontend.service.readinessProbe.httpGet.path string Readiness probe path /
frontend.service.readinessProbe.httpGet.port string Readiness probe port react-app
frontend.resources.requests.memory string Memory request "256Mi"
frontend.resources.requests.cpu string CPU request "100m"
frontend.resources.limits.memory string Memory limit "512Mi"
frontend.resources.limits.cpu string CPU limit "200m"
apiGateway.enabled boolean Enable API Gateway true
apiGateway.replicas integer Number of Gateway replicas 1
apiGateway.image.registryPath string Registry path docker.io/library
apiGateway.image.imageName string Gateway image name nginx
apiGateway.image.tag string Gateway image tag alpine
apiGateway.service.ports[0].port integer Service port 8080
apiGateway.service.ports[0].targetPort integer Container port 8080
apiGateway.service.type string Service type ClusterIP
apiGateway.configMap.name string ConfigMap name api-gateway-config
apiGateway.resources.requests.cpu string CPU request "100m"
apiGateway.resources.requests.memory string Memory request "256Mi"
apiGateway.resources.limits.cpu string CPU limit "200m"
apiGateway.resources.limits.memory string Memory limit "512Mi"
gkeGateway.enabled boolean Enable GKE Gateway false
gkeGateway.routes[0].path string Route path /
gkeGateway.routes[0].service string Service name gateway-api-gw
gkeGateway.routes[0].port integer Service port 8080
persistence.fileStorage.enabled boolean Enable persistent storage true
persistence.fileStorage.size string Storage size 30Gi
persistence.fileStorage.storageClass string Storage class name standard-rwo-regional
persistence.fileStorage.accessMode string Storage access mode ReadWriteOnce

Monitoring & Observability

Metrics

..

Logging

..

Contributing

  1. Fork the repository
  2. Create a feature branch
  3. Submit a pull request