mirror of
https://github.com/Dokploy/dokploy
synced 2025-06-26 18:27:59 +00:00
Feat/monitoring (#1267) Cloud Version
* feat: add start monitoring remote servers * reafctor: update * refactor: update * refactor: update * refactor: update * refactor: update * refactor: update * refactor: update * refactor: * refactor: add metrics * feat: add disk monitoring * refactor: translate to english * refacotor: add stats * refactor: remove color * feat: add log server metrics * refactor: remove unused deps * refactor: add origin * refactor: add logs * refactor: update * feat: add series monitoring * refactor: add system monitoring * feat: add benchmark to optimize data * refactor: update fn * refactor: remove comments * refactor: update * refactor: exclude items * feat: add refresh rate * feat: add monitoring remote servers * refactor: update * refactor: remove unsued volumes * refactor: update monitoring * refactor: add more presets * feat: add container metrics * feat: add docker monitoring * refactor: update conversion * refactor: remove unused code * refactor: update * refactor: add docker compose logs * refactor: add docker cli * refactor: add install curl * refactor: add get update * refactor: add monitoring remote servers * refactor: add containers config * feat: add container specification * refactor: update path * refactor: add server filter * refactor: simplify logic * fix: verify if file exist before get stats * refactor: update * refactor: remove unused deps * test: add test for containers * refactor: update * refactor add memory collector * refactor: update * refactor: update * refactor: update * refactor: remove * refactor: add memory * refactor: add server memory usage * refactor: change memory * refactor: update * refactor: update * refactor: add container metrics * refactor: comment code * refactor: mount proc bind * refactor: change interval with node cron * refactor: remove opening file * refactor: use streams * refactor: remove unused ws * refactor: disable live when is all * refactor: add sqlite * refactor: update * feat: add golang benchmark * refactor: update go * refactor: update dockerfile * refactor: update db * refactor: add env * refactor: separate logic * refactor: split logic * refactor: update logs * refactor: update dockerfile * refactor: hide .env * refactor: update * chore: hide ,.ebnv * refactor: add end angle * refactor: update * refactor: update * refactor: update * refactor: update * refactor: update * refactor: update monitoring * refactor: add mount db * refactor: add metrics and url callback * refactor: add middleware * refactor: add threshold property * feat: add memory and cpu threshold notification * feat: send notifications to the server * feat: add metrics for dokploy server * refactor: add dokploy server to monitoring * refactor: update methods * refactor: add admin to useeffect * refactor: stop monitoring containers if elements are 0 * refactor: cancel request if appName is empty * refactor: reuse methods * chore; add feat monitoring * refactor: set base url * refactor: adjust monitoring * refactor: delete migrations * feat: add columns * fix: add missing flag * refactor: add free metrics * refactor: add paid monitoring * refactor: update methods * feat: improve ui * feat: add container stats * refactor: add all container metrics * refactor: add color primary * refactor: change default rate limiting refresher * refactor: update retention days * refactor: use json instead of individual properties * refactor: lint * refactor: pass json env * refactor: update * refactor: delete * refactor: update * refactor: fix types * refactor: add retention days * chore: add license * refactor: create db * refactor: update path * refactor: update setup * refactor: update * refactor: create files * refactor: update * refactor: delete * refactor: update * refactor: update token metrics * fix: typechecks * refactor: setup web server * refactor: update error handling and add monitoring * refactor: add local storage save * refactor: add spacing * refactor: update * refactor: upgrade drizzle * refactor: delete * refactor: uppgrade drizzle kit * refactor: update search with jsonB * chore: upgrade drizzle * chore: update packages * refactor: add missing type * refactor: add serverType * refactor: update url * refactor: update * refactor: update * refactor: hide monitoring on self hosted * refactor: update server * refactor: update * refactor: update * refactor: pin node version
This commit is contained in:
Mauricio Siu
GitHub
261
apps/monitoring/monitoring/monitor.go
Normal file
261
apps/monitoring/monitoring/monitor.go
Normal file
@@ -0,0 +1,261 @@
|
||||
package monitoring
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"io"
|
||||
"net/http"
|
||||
"os"
|
||||
"os/exec"
|
||||
"runtime"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"github.com/shirou/gopsutil/v3/cpu"
|
||||
"github.com/shirou/gopsutil/v3/disk"
|
||||
"github.com/shirou/gopsutil/v3/host"
|
||||
"github.com/shirou/gopsutil/v3/mem"
|
||||
"github.com/shirou/gopsutil/v3/net"
|
||||
|
||||
"github.com/mauriciogm/dokploy/apps/monitoring/config"
|
||||
"github.com/mauriciogm/dokploy/apps/monitoring/database"
|
||||
)
|
||||
|
||||
type SystemMetrics struct {
|
||||
CPU string `json:"cpu"`
|
||||
CPUModel string `json:"cpuModel"`
|
||||
CPUCores int32 `json:"cpuCores"`
|
||||
CPUPhysicalCores int32 `json:"cpuPhysicalCores"`
|
||||
CPUSpeed float64 `json:"cpuSpeed"`
|
||||
OS string `json:"os"`
|
||||
Distro string `json:"distro"`
|
||||
Kernel string `json:"kernel"`
|
||||
Arch string `json:"arch"`
|
||||
MemUsed string `json:"memUsed"`
|
||||
MemUsedGB string `json:"memUsedGB"`
|
||||
MemTotal string `json:"memTotal"`
|
||||
Uptime uint64 `json:"uptime"`
|
||||
DiskUsed string `json:"diskUsed"`
|
||||
TotalDisk string `json:"totalDisk"`
|
||||
NetworkIn string `json:"networkIn"`
|
||||
NetworkOut string `json:"networkOut"`
|
||||
Timestamp string `json:"timestamp"`
|
||||
}
|
||||
|
||||
type AlertPayload struct {
|
||||
ServerType string `json:"ServerType"`
|
||||
Type string `json:"Type"`
|
||||
Value float64 `json:"Value"`
|
||||
Threshold float64 `json:"Threshold"`
|
||||
Message string `json:"Message"`
|
||||
Timestamp string `json:"Timestamp"`
|
||||
Token string `json:"Token"`
|
||||
}
|
||||
|
||||
func getRealOS() string {
|
||||
if content, err := os.ReadFile("/etc/os-release"); err == nil {
|
||||
lines := strings.Split(string(content), "\n")
|
||||
var id, name, version string
|
||||
for _, line := range lines {
|
||||
if strings.HasPrefix(line, "PRETTY_NAME=") {
|
||||
return strings.Trim(strings.TrimPrefix(line, "PRETTY_NAME="), "\"")
|
||||
} else if strings.HasPrefix(line, "NAME=") {
|
||||
name = strings.Trim(strings.TrimPrefix(line, "NAME="), "\"")
|
||||
} else if strings.HasPrefix(line, "VERSION=") {
|
||||
version = strings.Trim(strings.TrimPrefix(line, "VERSION="), "\"")
|
||||
} else if strings.HasPrefix(line, "ID=") {
|
||||
id = strings.Trim(strings.TrimPrefix(line, "ID="), "\"")
|
||||
}
|
||||
}
|
||||
if name != "" && version != "" {
|
||||
return fmt.Sprintf("%s %s", name, version)
|
||||
}
|
||||
if name != "" {
|
||||
return name
|
||||
}
|
||||
if id != "" {
|
||||
return id
|
||||
}
|
||||
}
|
||||
|
||||
if content, err := os.ReadFile("/etc/system-release"); err == nil {
|
||||
text := strings.ToLower(string(content))
|
||||
switch {
|
||||
case strings.Contains(text, "red hat"):
|
||||
return "rhel"
|
||||
case strings.Contains(text, "centos"):
|
||||
return "centos"
|
||||
case strings.Contains(text, "fedora"):
|
||||
return "fedora"
|
||||
}
|
||||
}
|
||||
|
||||
cmd := exec.Command("uname", "-a")
|
||||
if output, err := cmd.Output(); err == nil {
|
||||
osInfo := strings.ToLower(string(output))
|
||||
switch {
|
||||
case strings.Contains(osInfo, "debian"):
|
||||
return "debian"
|
||||
case strings.Contains(osInfo, "ubuntu"):
|
||||
return "ubuntu"
|
||||
case strings.Contains(osInfo, "centos"):
|
||||
return "centos"
|
||||
case strings.Contains(osInfo, "fedora"):
|
||||
return "fedora"
|
||||
case strings.Contains(osInfo, "red hat"):
|
||||
return "rhel"
|
||||
case strings.Contains(osInfo, "arch"):
|
||||
return "arch"
|
||||
case strings.Contains(osInfo, "darwin"):
|
||||
return "darwin"
|
||||
}
|
||||
}
|
||||
|
||||
return runtime.GOOS
|
||||
}
|
||||
|
||||
func GetServerMetrics() database.ServerMetric {
|
||||
v, _ := mem.VirtualMemory()
|
||||
c, _ := cpu.Percent(0, false)
|
||||
cpuInfo, _ := cpu.Info()
|
||||
diskInfo, _ := disk.Usage("/")
|
||||
netInfo, _ := net.IOCounters(false)
|
||||
hostInfo, _ := host.Info()
|
||||
distro := getRealOS()
|
||||
|
||||
cpuModel := ""
|
||||
if len(cpuInfo) > 0 {
|
||||
cpuModel = fmt.Sprintf("%s %s", cpuInfo[0].VendorID, cpuInfo[0].ModelName)
|
||||
}
|
||||
|
||||
memTotalGB := float64(v.Total) / 1024 / 1024 / 1024
|
||||
memUsedGB := float64(v.Used) / 1024 / 1024 / 1024
|
||||
memUsedPercent := (memUsedGB / memTotalGB) * 100
|
||||
|
||||
var networkIn, networkOut float64
|
||||
if len(netInfo) > 0 {
|
||||
networkIn = float64(netInfo[0].BytesRecv) / 1024 / 1024
|
||||
networkOut = float64(netInfo[0].BytesSent) / 1024 / 1024
|
||||
}
|
||||
return database.ServerMetric{
|
||||
Timestamp: time.Now().UTC().Format(time.RFC3339Nano),
|
||||
CPU: c[0],
|
||||
CPUModel: cpuModel,
|
||||
CPUCores: int32(runtime.NumCPU()),
|
||||
CPUPhysicalCores: int32(len(cpuInfo)),
|
||||
CPUSpeed: float64(cpuInfo[0].Mhz),
|
||||
OS: getRealOS(),
|
||||
Distro: distro,
|
||||
Kernel: hostInfo.KernelVersion,
|
||||
Arch: hostInfo.KernelArch,
|
||||
MemUsed: memUsedPercent,
|
||||
MemUsedGB: memUsedGB,
|
||||
MemTotal: memTotalGB,
|
||||
Uptime: hostInfo.Uptime,
|
||||
DiskUsed: float64(diskInfo.UsedPercent),
|
||||
TotalDisk: float64(diskInfo.Total) / 1024 / 1024 / 1024,
|
||||
NetworkIn: networkIn,
|
||||
NetworkOut: networkOut,
|
||||
}
|
||||
}
|
||||
|
||||
func ConvertToSystemMetrics(metric database.ServerMetric) SystemMetrics {
|
||||
return SystemMetrics{
|
||||
CPU: fmt.Sprintf("%.2f", metric.CPU),
|
||||
CPUModel: metric.CPUModel,
|
||||
CPUCores: metric.CPUCores,
|
||||
CPUPhysicalCores: metric.CPUPhysicalCores,
|
||||
CPUSpeed: metric.CPUSpeed,
|
||||
OS: metric.OS,
|
||||
Distro: metric.Distro,
|
||||
Kernel: metric.Kernel,
|
||||
Arch: metric.Arch,
|
||||
MemUsed: fmt.Sprintf("%.2f", metric.MemUsed),
|
||||
MemUsedGB: fmt.Sprintf("%.2f", metric.MemUsedGB),
|
||||
MemTotal: fmt.Sprintf("%.2f", metric.MemTotal),
|
||||
Uptime: metric.Uptime,
|
||||
DiskUsed: fmt.Sprintf("%.2f", metric.DiskUsed),
|
||||
TotalDisk: fmt.Sprintf("%.2f", metric.TotalDisk),
|
||||
NetworkIn: fmt.Sprintf("%.2f", metric.NetworkIn),
|
||||
NetworkOut: fmt.Sprintf("%.2f", metric.NetworkOut),
|
||||
Timestamp: metric.Timestamp,
|
||||
}
|
||||
}
|
||||
|
||||
func CheckThresholds(metrics database.ServerMetric) error {
|
||||
cfg := config.GetMetricsConfig()
|
||||
cpuThreshold := float64(cfg.Server.Thresholds.CPU)
|
||||
memThreshold := float64(cfg.Server.Thresholds.Memory)
|
||||
callbackURL := cfg.Server.UrlCallback
|
||||
metricsToken := cfg.Server.Token
|
||||
|
||||
// log.Printf("CPU threshold: %.2f%%", cpuThreshold)
|
||||
// log.Printf("Current CPU usage: %.2f%%", metrics.CPU)
|
||||
// log.Printf("Memory threshold: %.2f%%", memThreshold)
|
||||
// log.Printf("Callback URL: %s", callbackURL)
|
||||
// log.Printf("Metrics token: %s", metricsToken)
|
||||
|
||||
if cpuThreshold == 0 && memThreshold == 0 {
|
||||
return nil
|
||||
}
|
||||
|
||||
if cpuThreshold > 0 && metrics.CPU > cpuThreshold {
|
||||
alert := AlertPayload{
|
||||
ServerType: cfg.Server.ServerType,
|
||||
Type: "CPU",
|
||||
Value: metrics.CPU,
|
||||
Threshold: cpuThreshold,
|
||||
Message: fmt.Sprintf("CPU usage (%.2f%%) exceeded threshold (%.2f%%)", metrics.CPU, cpuThreshold),
|
||||
Timestamp: metrics.Timestamp,
|
||||
Token: metricsToken,
|
||||
}
|
||||
if err := sendAlert(callbackURL, alert); err != nil {
|
||||
return fmt.Errorf("failed to send CPU alert: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
if memThreshold > 0 && metrics.MemUsed > memThreshold {
|
||||
alert := AlertPayload{
|
||||
ServerType: cfg.Server.ServerType,
|
||||
Type: "Memory",
|
||||
Value: metrics.MemUsed,
|
||||
Threshold: memThreshold,
|
||||
Message: fmt.Sprintf("Memory usage (%.2f%%) exceeded threshold (%.2f%%)", metrics.MemUsed, memThreshold),
|
||||
Timestamp: metrics.Timestamp,
|
||||
Token: metricsToken,
|
||||
}
|
||||
if err := sendAlert(callbackURL, alert); err != nil {
|
||||
return fmt.Errorf("failed to send memory alert: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
return nil
|
||||
}
|
||||
|
||||
func sendAlert(callbackURL string, payload AlertPayload) error {
|
||||
if callbackURL == "" {
|
||||
return fmt.Errorf("callback URL is not set")
|
||||
}
|
||||
wrappedPayload := map[string]interface{}{
|
||||
"json": payload,
|
||||
}
|
||||
|
||||
jsonData, err := json.Marshal(wrappedPayload)
|
||||
if err != nil {
|
||||
return fmt.Errorf("failed to marshal alert payload: %v", err)
|
||||
}
|
||||
|
||||
resp, err := http.Post(callbackURL, "application/json", bytes.NewBuffer(jsonData))
|
||||
if err != nil {
|
||||
return fmt.Errorf("failed to send POST request: %v", err)
|
||||
}
|
||||
defer resp.Body.Close()
|
||||
|
||||
if resp.StatusCode != http.StatusOK {
|
||||
bodyBytes, _ := io.ReadAll(resp.Body)
|
||||
return fmt.Errorf("received non-OK response status: %s, body: %s", resp.Status, string(bodyBytes))
|
||||
}
|
||||
|
||||
return nil
|
||||
}
|
||||
Reference in New Issue
Block a user