prometheus

Author: lework Version: 0.0.2 Type: tool

Description

Creating a Dify Plugin

dify plugin init

Packaging the Plugin

dify plugin package ./prometheus

Prometheus Query Plugin

A Prometheus query plugin for Dify that allows users to query Prometheus metric data using PromQL.

Features

• Query Prometheus metric data using PromQL
• Support for time range queries (relative and absolute time)
• Support for basic authentication and token authentication
• Automatic formatting of results for easy understanding
• Support for Markdown table output, clearly displaying the latest value of each metric with labels as columns
• Kubernetes Pod resource metrics query, getting CPU, memory usage and restart count for Pods

Configuration

Before using this plugin, you need to configure the following information:

• API URL: The URL of the Prometheus server, e.g., http://localhost:9090
• Username/Password: (Optional) Username and password for basic authentication
• Token: (Optional) Bearer token for authentication

Tools

1. Prometheus Query

Parameters

• PromQL Query Statement: Required, the PromQL query statement to execute
• Start Time: Optional, the start time of the query, supports the following formats:
  • RFC3339/ISO8601 format: 2023-01-01T00:00:00Z
  • Relative time: 1h, 2d, 3w, 4m, 5y, etc.
  • Default value: 1h (1 hour ago)
• End Time: Optional, the end time of the query, supports the same formats as the start time
  • Default value: now (current time)
• Step: Optional, the resolution step of the query
  • Format: 15s, 1m, 1h, etc.
  • Default value: 15s (15 seconds)

Examples

Query CPU Usage

query: 100 - (avg by (instance) (irate(node_cpu_seconds_total{mode="idle"}[5m])) * 100)
start_time: 1h
end_time: now
step: 1m

Query Memory Usage

query: 100 * (1 - (node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes))
start_time: 12h
end_time: now
step: 5m

2. Kubernetes Pod Resource Metrics Query

Get resource usage information for Kubernetes Pods, including CPU, memory usage and restart count, displayed in Markdown table format.

Parameters

• Namespace: Optional, Kubernetes namespace name, if not specified, queries all namespaces
• Label Selector: Optional, label selector to filter Pods, e.g., app=myapp,component=database
• Pod Name Pattern: Optional, regular expression to filter Pod names, e.g., frontend-.*

Examples

Query Pod Resource Usage in a Specific Namespace

namespace: default

Query Pod Resource Usage for a Specific Application

selector: app=myapp

Query Pod Resource Usage with a Specific Prefix

pod_name_pattern: frontend-.*

Return Result

Returns a Markdown table containing the following information:

Pod Name	Age	Namespace	Node	Ready	Phase	CPU Usage %	Memory Usage MiB	CPU Request	CPU Limit	Memory Request MiB	Memory Limit MiB	Restarts in 24h
pod-1	1.0 day	default	node1	✓	Running	0.819%	500 MiB	0.100	1	500 MiB	2 GiB	0
pod-2	1.7 days	default	node2	✓	Running	0.380%	256 MiB	0.100	1	256 MiB	2 GiB	0

Return Results

Markdown Table Format (New Feature)

For matrix type query results, the plugin will generate a Markdown table, using labels as table columns, making it easier to clearly display the data:

### node_cpu_seconds_total

| timestamp           | instance       | job        | mode   | value |
| :------------------ | :------------- | :--------- | :----- | ----: |
| 2023-03-11T18:15:00 | 10.131.242.148 | prometheus | idle   | 0.278 |
| 2023-03-11T18:15:00 | 10.131.242.149 | prometheus | system | 0.156 |

_Showing the most recent data points_

JSON Format

When a Markdown table cannot be generated, the plugin will return the complete result in JSON format:

{
  "success": true,
  "result_type": "matrix",
  "data": [
    {
      "metric": "metric_name",
      "labels": {
        "instance": "localhost:9090",
        "job": "prometheus"
      },
      "values": [
        {
          "timestamp": "2023-01-01T00:00:00.000000",
          "value": 12.34
        },
        ...
      ]
    },
    ...
  ]
}