Skip to main content

​
Why you need to configure a metrics data source

The “Load Monitoring” charts in Plugins Management rely on standard Kubernetes container metrics to draw CPU / memory / network resource curves. Because different Kubernetes distributions and monitoring stacks expose metrics differently, you must choose an appropriate metrics data source during deployment. If your cluster/monitoring system cannot provide the required metrics, set the source to disabled to avoid empty charts or repeated errors in the UI.

​
Prerequisites (required metrics)

Make sure your cluster or metrics system can provide the following metrics. If any of them are missing, charts may be incomplete or not displayed:
  • container_cpu_usage_seconds_total
  • container_memory_working_set_bytes
  • container_network_receive_bytes_total
  • container_network_transmit_bytes_total

​
Display Helm chart values

helm show values dify/dify

​
How to configure (Helm values)

Below are the plugin-manager metrics data source configurations (typically set in values.yaml or your deployment configuration). Use an external Prometheus as the metrics source (recommended for production due to stronger persistence and query capabilities): 
plugin_manager:
  metric:
    source: "prometheus"

externalPrometheus:
  enabled: true
  endpoint: "http://prometheus-kube-prometheus-prometheus:9090"
  timeout: 10s
  username: ""
  password: ""
  insecure: true
  • externalPrometheus.endpoint: Prometheus endpoint (plugin-manager must be able to reach it inside the cluster network)
  • externalPrometheus.timeout: Query timeout
  • externalPrometheus.username/password: Set these if Prometheus requires authentication
  • externalPrometheus.insecure: Allow insecure TLS (use only if you clearly understand the risks)

​
B. cAdvisor (actively collected by plugin-manager)

When configured to cAdvisor mode, plugin-manager collects cAdvisor metrics via the node proxy: 
plugin_manager:
  metric:
    source: "cadvisor"
Notes:
  • Permissions required: You must grant ClusterRole permissions to read nodes and nodes/proxy, otherwise metrics cannot be collected through the node proxy.
  • When to use: Collected metrics are typically stored temporarily and are not persisted. This mode is better for lightweight/temporary troubleshooting and should not be your only data source for long-term audit or capacity planning.

​
C. disabled

When metrics are disabled, the UI will not show CPU/memory/network monitoring curves: 
plugin_manager:
  metric:
    source: "disabled"

​
Troubleshooting

  • Charts are empty or missing lines: First confirm the four prerequisite metrics exist. If your monitoring stack does not provide them, set plugin_manager.metric.source to disabled.
  • Prometheus mode cannot fetch data: Verify externalPrometheus.endpoint is reachable within the cluster (DNS/Service/NetworkPolicy), and check whether authentication or TLS settings are required.
  • Permission errors in cAdvisor mode: Confirm the cluster-level permissions for nodes / nodes/proxy are granted.

​
References