Skip to main content

​
Helm Values

This update introduces the following configurations: global.otel configuration involves services responsible for generating telemetry data, specifically including dify-api and dify-worker. enterpriseCollector is a new service designed to interact with your custom OpenTelemetry (OTEL) system. In most cases, keeping the default configuration values is sufficient. 
global:
  otel:
    # Requires enterpriseCollector.enabled to be true
    enabled: true
    # Sampling rate (0.0 to 1.0, empty means default)
    samplingRate: "0.2"
    # Batch export schedule delay in milliseconds
    batchExportScheduleDelay: "5000"
    # Maximum queue size for batching
    maxQueueSize: "2048"
    # Maximum export batch size
    maxExportBatchSize: "512"
    # Metric export interval in milliseconds
    metricExportInterval: "60000"
    # Batch export timeout in milliseconds
    batchExportTimeout: "30000"
    # Metric export timeout in milliseconds
    metricExportTimeout: "30000"
    sender:
      otlp:
        # Custom headers for OTLP requests (key=value,key2=value2)
        headers: ""
      # Service name reported to OTEL collector
      serviceName: "dify-api-otel"
      # Include input/output content in traces (privacy toggle)
      includeContent: false
enterpriseCollector:
  enabled: true
  image:
    repository: langgenius/dify-ee-collector
    tag: "3.9.6"
  replicas: 1
  resources: {}
  nodeSelector: {}
  affinity: {}
  tolerations: []
  extraEnv:
  # Apply your own Environment Variables if necessary. Below is demo.
  # extraEnv:
  #   - name: ENV_FROM_COMMUNITY1
  #     value: env123
  #   - name: ENV_FROM_COMMUNITY2
  #     value: env123
  serviceAccountName: ""
  # readinessProbe:
  #   tcpSocket:
  #     port: 8080
  #   initialDelaySeconds: 30
  #   periodSeconds: 10
  #   successThreshold: 1
  #   timeoutSeconds: 3
  # livenessProbe:
  #   tcpSocket:
  #     port: 8080
  #   initialDelaySeconds: 20
  #   periodSeconds: 10
  #   successThreshold: 1
  #   timeoutSeconds: 3
  extraConfig: |
    # yaml config for otel collector
    # you can use this to fully customize the otel collector
    # see https://opentelemetry.io/docs/collector/configuration/
  # The size of the sending queues are related to the memory limit and the batch size
  # the default value assumes 512 MiB memory limit and 512 batch size
  # adjust correspondingly when you configure different memory limits and batch sizes
  sendingQueues:
    trace:
      size: 64
    metric:
      size: 256

​
ServiceMonitor Integration

Although the architecture of enterpriseCollector is designed to transmit telemetry data to user systems, pulling data remains a common method for collecting metrics. Therefore, it also exposes a metrics endpoint on port 8889, which can be seamlessly integrated with Prometheus ServiceMonitor. Below is an example. 
apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: dify-enterprise-collector
  namespace: monitoring
spec:
  endpoints:
  - interval: 5s
    path: /metrics
    port: http-metrics
    scheme: http
  namespaceSelector:
    matchNames:
    - {dify's namespace}
  selector:
    matchLabels:
      app: {dify's release name}-enterprise-collector