[OSCP]KusciaTask 系统资源指标采集、暴露及统一导出

[renji555]

Fixedhttps://github.com//issues/400

[UniqueMarvin]

这里也建议直接调用 Containerd 的 pkg 去实现，不要使用 crictl stats 实现，这个可以搜一下大模型，有很多成熟的实现

[UniqueMarvin]

类似这样：

package main

import (
	"context"
	"fmt"
	"log"
	"time"

	"github.com/containerd/containerd"
	"github.com/containerd/containerd/cio"
	"github.com/containerd/containerd/namespaces"
	"github.com/containerd/typeurl"
	v1 "github.com/containerd/cgroups/stats/v1" // Import cgroups stats
)

func main() {
	// Create a new client connected to the default containerd socket
	client, err := containerd.New("/run/containerd/containerd.sock")
	if err != nil {
		log.Fatal(err)
	}
	defer client.Close()

	// Define a context with the appropriate namespace
	ctx := namespaces.WithNamespace(context.Background(), "default")

	// Specify the container ID you want to monitor
	containerID := "your-container-id"

	// Fetch the container by its ID
	container, err := client.LoadContainer(ctx, containerID)
	if err != nil {
		log.Fatalf("failed to load container: %v", err)
	}

	// Get the task associated with the container
	task, err := container.Task(ctx, cio.Load)
	if err != nil {
		log.Fatalf("failed to load task for container: %v", err)
	}

	// Get the metrics for the task
	metrics, err := task.Metrics(ctx)
	if err != nil {
		log.Fatalf("failed to get metrics: %v", err)
	}

	// Parse the metrics into a cgroups metrics structure
	data, err := typeurl.UnmarshalAny(metrics.Data)
	if err != nil {
		log.Fatalf("failed to unmarshal metrics data: %v", err)
	}

	// Assert the data to the expected type
	stats := data.(*v1.Metrics)

	// Print the CPU and memory usage
	fmt.Printf("CPU Usage: %v\n", stats.CPU.Usage.Total)
	fmt.Printf("Memory Usage: %v\n", stats.Memory.Usage.Usage)
}

[renji555]

那task的相关任务指标是否可以通过crictl stats获得呢

[renji555]

这种方法只能获得CPU和内存使用，剩下的Iodes和disk该如何解决呢

[yushiqie]

containerd 有metrics plugin，不建议裸写。
https://github.com/containerd/containerd/blob/main/docs/ops.md