allard
392 lines
11 KiB
Plaintext
392 lines
11 KiB
Plaintext
package main
|
|
|
|
import (
|
|
"context"
|
|
"encoding/json"
|
|
"fmt"
|
|
"io/ioutil"
|
|
"log"
|
|
"net/http"
|
|
"os"
|
|
"strconv"
|
|
"strings"
|
|
"time"
|
|
|
|
"golang.org/x/crypto/ssh"
|
|
v1 "k8s.io/api/core/v1"
|
|
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
|
|
"k8s.io/client-go/kubernetes"
|
|
"k8s.io/client-go/rest"
|
|
metrics "k8s.io/metrics/pkg/client/clientset/versioned"
|
|
"gopkg.in/yaml.v3"
|
|
)
|
|
|
|
// Node and NodePool structs
|
|
type Node struct {
|
|
Name string `yaml:"name" json:"name"`
|
|
IP string `yaml:"ip" json:"ip"`
|
|
Status string `yaml:"status" json:"status"`
|
|
CPU int `yaml:"cpu" json:"cpu"`
|
|
Memory int `yaml:"memory" json:"memory"`
|
|
Role string `yaml:"role" json:"role"`
|
|
Cluster string `yaml:"cluster" json:"cluster"`
|
|
LastActive string `yaml:"last_active" json:"last_active"`
|
|
Pods int `yaml:"pods,omitempty" json:"pods,omitempty"`
|
|
Temperature float64 `yaml:"temperature,omitempty" json:"temperature,omitempty"`
|
|
}
|
|
|
|
type NodePool struct {
|
|
Nodes []Node `yaml:"nodes" json:"nodes"`
|
|
}
|
|
|
|
// mustIntEnv
|
|
func mustIntEnv(name string, def int) int {
|
|
if val := os.Getenv(name); val != "" {
|
|
if i, err := strconv.Atoi(val); err == nil {
|
|
return i
|
|
}
|
|
}
|
|
return def
|
|
}
|
|
|
|
// loadPool
|
|
func loadPool(file string) (*NodePool, error) {
|
|
data, err := ioutil.ReadFile(file)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
var pool NodePool
|
|
err = yaml.Unmarshal(data, &pool)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
return &pool, nil
|
|
}
|
|
|
|
// savePool
|
|
func savePool(file string, pool *NodePool) error {
|
|
data, err := yaml.Marshal(pool)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
return ioutil.WriteFile(file, data, 0644)
|
|
}
|
|
|
|
// initNodePool
|
|
func initNodePool(clientset *kubernetes.Clientset, poolFile string) (*NodePool, error) {
|
|
ctx := context.Background()
|
|
|
|
pool, err := loadPool(poolFile)
|
|
if err != nil {
|
|
log.Println("Failed to load existing node pool, starting fresh:", err)
|
|
pool = &NodePool{}
|
|
}
|
|
|
|
nodes, err := clientset.CoreV1().Nodes().List(ctx, metav1.ListOptions{})
|
|
if err != nil {
|
|
return nil, fmt.Errorf("failed to list nodes: %w", err)
|
|
}
|
|
|
|
poolMap := map[string]*Node{}
|
|
for i := range pool.Nodes {
|
|
poolMap[pool.Nodes[i].Name] = &pool.Nodes[i]
|
|
}
|
|
|
|
updatedNodes := []Node{}
|
|
for _, n := range nodes.Items {
|
|
ip := nodeIP(&n)
|
|
status := "online"
|
|
role := "worker"
|
|
if isControlPlane(&n) {
|
|
role = "microk8s-controlplane"
|
|
}
|
|
|
|
node := Node{
|
|
Name: n.Name,
|
|
IP: ip,
|
|
Status: status,
|
|
Role: role,
|
|
Cluster: os.Getenv("CLUSTER_NAME"),
|
|
CPU: 0,
|
|
Memory: 0,
|
|
LastActive: time.Now().Format(time.RFC3339),
|
|
}
|
|
|
|
if oldNode, ok := poolMap[n.Name]; ok {
|
|
node.Status = oldNode.Status
|
|
node.LastActive = oldNode.LastActive
|
|
node.CPU = oldNode.CPU
|
|
node.Memory = oldNode.Memory
|
|
node.Cluster = oldNode.Cluster
|
|
node.Role = oldNode.Role
|
|
}
|
|
|
|
updatedNodes = append(updatedNodes, node)
|
|
}
|
|
|
|
pool.Nodes = updatedNodes
|
|
savePool(poolFile, pool)
|
|
log.Printf("Initialized node pool with %d nodes", len(pool.Nodes))
|
|
return pool, nil
|
|
}
|
|
|
|
// runSSH
|
|
func runSSH(host, user, pass, cmd string) (string, error) {
|
|
config := &ssh.ClientConfig{
|
|
User: user,
|
|
Auth: []ssh.AuthMethod{ssh.Password(pass)},
|
|
HostKeyCallback: ssh.InsecureIgnoreHostKey(),
|
|
Timeout: 10 * time.Second,
|
|
}
|
|
client, err := ssh.Dial("tcp", fmt.Sprintf("%s:22", host), config)
|
|
if err != nil {
|
|
return "", err
|
|
}
|
|
defer client.Close()
|
|
session, err := client.NewSession()
|
|
if err != nil {
|
|
return "", err
|
|
}
|
|
defer session.Close()
|
|
out, err := session.CombinedOutput(cmd)
|
|
return string(out), err
|
|
}
|
|
|
|
// isControlPlane
|
|
func isControlPlane(n *v1.Node) bool {
|
|
if _, ok := n.Labels["node.kubernetes.io/microk8s-controlplane"]; ok {
|
|
return true
|
|
}
|
|
if _, ok := n.Labels["node-role.kubernetes.io/control-plane"]; ok {
|
|
return true
|
|
}
|
|
return false
|
|
}
|
|
|
|
// nodeIP
|
|
func nodeIP(n *v1.Node) string {
|
|
for _, addr := range n.Status.Addresses {
|
|
if addr.Type == v1.NodeInternalIP {
|
|
return addr.Address
|
|
}
|
|
}
|
|
return ""
|
|
}
|
|
|
|
// updatePerNodeUtilization
|
|
func updatePerNodeUtilization(cs *kubernetes.Clientset, ms *metrics.Clientset, pool *NodePool, sshUser, sshPass string) error {
|
|
ctx := context.Background()
|
|
|
|
nodes, err := cs.CoreV1().Nodes().List(ctx, metav1.ListOptions{})
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
metricsList, err := ms.MetricsV1beta1().NodeMetricses().List(ctx, metav1.ListOptions{})
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
capCPU := map[string]int64{}
|
|
capMem := map[string]int64{}
|
|
nodeMap := map[string]*v1.Node{}
|
|
for i, n := range nodes.Items {
|
|
capCPU[n.Name] = n.Status.Capacity.Cpu().MilliValue()
|
|
capMem[n.Name] = n.Status.Capacity.Memory().Value()
|
|
nodeMap[n.Name] = &nodes.Items[i]
|
|
}
|
|
|
|
usageCPU := map[string]int64{}
|
|
usageMem := map[string]int64{}
|
|
for _, m := range metricsList.Items {
|
|
usageCPU[m.Name] = m.Usage.Cpu().MilliValue()
|
|
usageMem[m.Name] = m.Usage.Memory().Value()
|
|
}
|
|
|
|
pods, err := cs.CoreV1().Pods("").List(ctx, metav1.ListOptions{})
|
|
if err != nil {
|
|
return err
|
|
}
|
|
podCount := map[string]int{}
|
|
for _, p := range pods.Items {
|
|
if p.Spec.NodeName != "" {
|
|
podCount[p.Spec.NodeName]++
|
|
}
|
|
}
|
|
|
|
for i := range pool.Nodes {
|
|
n := &pool.Nodes[i]
|
|
name := n.Name
|
|
|
|
cpuCap, ok1 := capCPU[name]
|
|
memCap, ok2 := capMem[name]
|
|
cpuUse, ok3 := usageCPU[name]
|
|
memUse, ok4 := usageMem[name]
|
|
|
|
if ok1 && ok2 && ok3 && ok4 && cpuCap > 0 && memCap > 0 {
|
|
n.CPU = int((cpuUse * 100) / cpuCap)
|
|
n.Memory = int((memUse * 100) / memCap)
|
|
}
|
|
|
|
n.Pods = podCount[name]
|
|
|
|
if kn, ok := nodeMap[name]; ok {
|
|
n.Temperature = getNodeTemp(kn, sshUser, sshPass)
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// getNodeTemp
|
|
func getNodeTemp(node *v1.Node, sshUser, sshPass string) float64 {
|
|
ip := nodeIP(node)
|
|
out, err := runSSH(ip, sshUser, sshPass, "cat /proc/device-tree/model")
|
|
if err != nil {
|
|
return 0
|
|
}
|
|
hw := strings.ToUpper(strings.TrimSpace(out))
|
|
|
|
var cmd string
|
|
switch {
|
|
case strings.Contains(hw, "RASPBERRY"):
|
|
cmd = "vcgencmd measure_temp | egrep -o '[0-9]+\\.[0-9]+'"
|
|
case strings.Contains(hw, "ODROID"):
|
|
cmd = "awk '{printf \"%3.1f\", $1/1000}' /sys/class/thermal/thermal_zone0/temp"
|
|
default:
|
|
return 0
|
|
}
|
|
|
|
tempStr, err := runSSH(ip, sshUser, sshPass, cmd)
|
|
if err != nil {
|
|
return 0
|
|
}
|
|
|
|
t, err := strconv.ParseFloat(strings.TrimSpace(tempStr), 64)
|
|
if err != nil {
|
|
return 0
|
|
}
|
|
return t
|
|
}
|
|
|
|
// ... other functions: ensureControlPlanes, clusterUtilization, deactivateOneWorkerSafe, activateOneWorker etc.
|
|
// (exactly as in your last working version)
|
|
|
|
// startWebGUI
|
|
func startWebGUI(poolFile string) {
|
|
http.HandleFunc("/status", func(w http.ResponseWriter, r *http.Request) {
|
|
log.Println("Serving /status")
|
|
|
|
pool, err := loadPool(poolFile)
|
|
if err != nil {
|
|
log.Println("Cannot load node pool:", err)
|
|
http.Error(w, "cannot load node pool", http.StatusInternalServerError)
|
|
return
|
|
}
|
|
|
|
w.Header().Set("Content-Type", "application/json")
|
|
json.NewEncoder(w).Encode(pool)
|
|
})
|
|
|
|
fs := http.FileServer(http.Dir("/app/web"))
|
|
http.Handle("/", fs)
|
|
|
|
go func() {
|
|
log.Println("Web GUI running at :8080")
|
|
if err := http.ListenAndServe(":8080", nil); err != nil {
|
|
log.Fatal(err)
|
|
}
|
|
}()
|
|
}
|
|
|
|
// main
|
|
func main() {
|
|
poolFile := os.Getenv("NODE_LIST_FILE")
|
|
sshUser := os.Getenv("NODE_SSH_USER")
|
|
sshPass := os.Getenv("NODE_SSH_PASS")
|
|
minCPU := mustIntEnv("MIN_CPU", 20)
|
|
maxCPU := mustIntEnv("MAX_CPU", 80)
|
|
minMem := mustIntEnv("MIN_MEM", 30)
|
|
maxMem := mustIntEnv("MAX_MEM", 80)
|
|
desiredCP := mustIntEnv("DESIRED_CONTROL_PLANES", 3)
|
|
waitSec := mustIntEnv("DEACTIVATE_WAIT_SEC", 120)
|
|
clusterName := os.Getenv("CLUSTER_NAME")
|
|
if clusterName == "" {
|
|
log.Fatal("CLUSTER_NAME must be set")
|
|
}
|
|
|
|
log.Println("=== startup configuration ===")
|
|
log.Printf("NODE_LIST_FILE=%q", poolFile)
|
|
log.Printf("NODE_SSH_USER=%q", sshUser)
|
|
log.Printf("MIN_CPU=%d", minCPU)
|
|
log.Printf("MAX_CPU=%d", maxCPU)
|
|
log.Printf("MIN_MEM=%d", minMem)
|
|
log.Printf("MAX_MEM=%d", maxMem)
|
|
log.Printf("DESIRED_CONTROL_PLANES=%d", desiredCP)
|
|
log.Printf("DEACTIVATE_WAIT_SEC=%d", waitSec)
|
|
log.Printf("CLUSTER_NAME=%q", clusterName)
|
|
log.Println("=============================")
|
|
|
|
startWebGUI(poolFile)
|
|
|
|
config, err := rest.InClusterConfig()
|
|
if err != nil {
|
|
log.Fatal(err)
|
|
}
|
|
clientset, err := kubernetes.NewForConfig(config)
|
|
if err != nil {
|
|
log.Fatal(err)
|
|
}
|
|
|
|
_, err = initNodePool(clientset, poolFile)
|
|
if err != nil {
|
|
log.Fatal("Failed to initialize node pool:", err)
|
|
}
|
|
|
|
for {
|
|
pool, err := loadPool(poolFile)
|
|
if err != nil {
|
|
log.Println("cannot load node pool:", err)
|
|
time.Sleep(30 * time.Second)
|
|
continue
|
|
}
|
|
|
|
metricsClient, err := metrics.NewForConfig(config)
|
|
if err != nil {
|
|
log.Fatal(err)
|
|
}
|
|
|
|
ensureControlPlanes(clientset, pool, poolFile, sshUser, sshPass, clusterName, desiredCP)
|
|
|
|
clusterCPU, clusterMem, err := clusterUtilization(clientset, metricsClient)
|
|
if err != nil {
|
|
log.Println("failed to compute utilization:", err)
|
|
time.Sleep(30 * time.Second)
|
|
continue
|
|
}
|
|
|
|
log.Printf("Cluster CPU: %d%%, Mem: %d%%", clusterCPU, clusterMem)
|
|
|
|
if clusterCPU < minCPU && clusterMem < minMem {
|
|
log.Println("Below minimum thresholds, attempting to deactivate one worker node...")
|
|
deactivateOneWorkerSafe(clientset, pool, poolFile, sshUser, sshPass, clusterName, waitSec)
|
|
}
|
|
|
|
if clusterCPU > maxCPU || clusterMem > maxMem {
|
|
log.Println("Above maximum thresholds, attempting to activate a free node from the node-pool...")
|
|
activateOneWorker(clientset, pool, poolFile, sshUser, sshPass, clusterName)
|
|
}
|
|
|
|
err = updatePerNodeUtilization(clientset, metricsClient, pool, sshUser, sshPass)
|
|
if err != nil {
|
|
log.Println("failed to update per-node utilization:", err)
|
|
}
|
|
|
|
if err := savePool(poolFile, pool); err != nil {
|
|
log.Println("failed to save node pool:", err)
|
|
}
|
|
|
|
time.Sleep(30 * time.Second)
|
|
}
|
|
}
|