feat: 修复服务器硬件规格(CPU核/内存/磁盘) + Top 5 资源排行
修复: - agent.go ReceiveReport 写入 cpu(核数)/memory(总内存MB)/disk(总磁盘GB) 到 servers 表 - 之前 Agent 上报只写了 cpu_usage/memory_usage/disk_usage 使用率，硬件规格始终为 0 新功能: - GET /api/v1/metrics/top?metric=cpu|memory|disk&limit=5 主机排行 API - HostMonitor 新增 Top 5 排行面板: CPU/内存/磁盘 使用率排名 - TopRankCard 组件: 排名号+主机名+进度条+百分比
This commit is contained in:
attacker committed
1 parent
06d7451cf9
commit
737f00169f
8 files changed
+150
-567
No files matched your search
@@ -1,215 +0,0 @@
|
||||
// Package collector 包含中间件（Redis / MySQL）的采集逻辑。
|
||||
// 通过 TCP 直连发送 INFO / SHOW GLOBAL STATUS 命令获取指标。
|
||||
//
|
||||
// 署名 Logan.Li
|
||||
package collector
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"io"
|
||||
"net"
|
||||
"strconv"
|
||||
"strings"
|
||||
"sync"
|
||||
"time"
|
||||
|
||||
"ezops/agent/model"
|
||||
)
|
||||
|
||||
// MiddlewareConfig 中间件采集配置
|
||||
type MiddlewareConfig struct {
|
||||
RedisAddrs []string // Redis 地址列表，格式 host:port
|
||||
RedisAuth string // Redis 密码（空为无密码）
|
||||
MySQLAddrs []string // MySQL 地址列表，格式 host:port
|
||||
MySQLUser string // MySQL 用户名
|
||||
MySQLPass string // MySQL 密码
|
||||
}
|
||||
|
||||
// MiddlewareCollector 中间件指标采集器
|
||||
type MiddlewareCollector struct {
|
||||
cfg MiddlewareConfig
|
||||
mu sync.Mutex
|
||||
prevMySQLQPS map[string]int64 // 上轮 Questions 值，用于计算 QPS 增量
|
||||
prevTime time.Time
|
||||
}
|
||||
|
||||
// NewMiddlewareCollector 创建中间件采集器
|
||||
func NewMiddlewareCollector(cfg MiddlewareConfig) *MiddlewareCollector {
|
||||
return &MiddlewareCollector{
|
||||
cfg: cfg,
|
||||
prevMySQLQPS: make(map[string]int64),
|
||||
prevTime: time.Now(),
|
||||
}
|
||||
}
|
||||
|
||||
// Collect 采集所有已配置中间件的指标
|
||||
func (m *MiddlewareCollector) Collect() (model.MiddlewareStat, error) {
|
||||
var stat model.MiddlewareStat
|
||||
|
||||
// Redis
|
||||
for _, addr := range m.cfg.RedisAddrs {
|
||||
if metric, err := m.collectRedis(addr); err == nil {
|
||||
stat.Redis = append(stat.Redis, metric)
|
||||
}
|
||||
}
|
||||
|
||||
// MySQL
|
||||
m.mu.Lock()
|
||||
now := time.Now()
|
||||
elapsed := now.Sub(m.prevTime).Seconds()
|
||||
m.prevTime = now
|
||||
m.mu.Unlock()
|
||||
|
||||
for _, addr := range m.cfg.MySQLAddrs {
|
||||
if metric, err := m.collectMySQL(addr, elapsed); err == nil {
|
||||
stat.MySQL = append(stat.MySQL, metric)
|
||||
}
|
||||
}
|
||||
|
||||
return stat, nil
|
||||
}
|
||||
|
||||
// collectRedis 连接 Redis 发送 INFO 命令并解析指标
|
||||
func (m *MiddlewareCollector) collectRedis(addr string) (model.RedisMetric, error) {
|
||||
metric := model.RedisMetric{Instance: addr}
|
||||
|
||||
conn, err := net.DialTimeout("tcp", addr, 3*time.Second)
|
||||
if err != nil {
|
||||
return metric, fmt.Errorf("redis connect %s: %w", addr, err)
|
||||
}
|
||||
defer conn.Close()
|
||||
|
||||
// 认证
|
||||
if m.cfg.RedisAuth != "" {
|
||||
cmd := fmt.Sprintf("*2\r\n$4\r\nAUTH\r\n$%d\r\n%s\r\n", len(m.cfg.RedisAuth), m.cfg.RedisAuth)
|
||||
if _, err := conn.Write([]byte(cmd)); err != nil {
|
||||
return metric, fmt.Errorf("redis auth: %w", err)
|
||||
}
|
||||
buf := make([]byte, 1024)
|
||||
conn.SetReadDeadline(time.Now().Add(2 * time.Second))
|
||||
n, _ := conn.Read(buf)
|
||||
if n > 0 && !strings.HasPrefix(string(buf[:n]), "+OK") {
|
||||
return metric, fmt.Errorf("redis auth failed")
|
||||
}
|
||||
}
|
||||
|
||||
// 发送 INFO 命令
|
||||
conn.SetWriteDeadline(time.Now().Add(2 * time.Second))
|
||||
if _, err := conn.Write([]byte("*1\r\n$4\r\nINFO\r\n")); err != nil {
|
||||
return metric, fmt.Errorf("redis info cmd: %w", err)
|
||||
}
|
||||
|
||||
conn.SetReadDeadline(time.Now().Add(3 * time.Second))
|
||||
data, err := io.ReadAll(conn)
|
||||
if err != nil {
|
||||
return metric, fmt.Errorf("redis read: %w", err)
|
||||
}
|
||||
|
||||
// 解析 Redis INFO bulk string 响应
|
||||
raw := string(data)
|
||||
// Redis RESP: $<len>\r\n<data>\r\n — 去掉前缀
|
||||
if idx := strings.Index(raw, "\r\n"); idx > 0 {
|
||||
raw = raw[idx+2:]
|
||||
}
|
||||
if idx := strings.LastIndex(raw, "\r\n"); idx > 0 {
|
||||
raw = raw[:idx]
|
||||
}
|
||||
|
||||
// 按行解析 key:value
|
||||
for _, line := range strings.Split(raw, "\n") {
|
||||
line = strings.TrimSpace(line)
|
||||
if line == "" || strings.HasPrefix(line, "#") {
|
||||
continue
|
||||
}
|
||||
parts := strings.SplitN(line, ":", 2)
|
||||
if len(parts) != 2 {
|
||||
continue
|
||||
}
|
||||
key, val := parts[0], strings.TrimSpace(parts[1])
|
||||
|
||||
switch key {
|
||||
case "redis_version":
|
||||
metric.Version = val
|
||||
case "uptime_in_seconds":
|
||||
metric.Uptime, _ = strconv.ParseInt(val, 10, 64)
|
||||
case "connected_clients":
|
||||
metric.Clients, _ = strconv.Atoi(val)
|
||||
case "used_memory":
|
||||
metric.UsedMemory, _ = strconv.ParseInt(val, 10, 64)
|
||||
case "instantaneous_ops_per_sec":
|
||||
metric.QPS, _ = strconv.ParseFloat(val, 64)
|
||||
case "keyspace_hits", "keyspace_misses":
|
||||
// Both needed for hit rate
|
||||
}
|
||||
}
|
||||
|
||||
// 命中率从 keyspace_hits + keyspace_misses 计算
|
||||
hitsStr := extractInfoField(raw, "keyspace_hits")
|
||||
missesStr := extractInfoField(raw, "keyspace_misses")
|
||||
hits, _ := strconv.ParseInt(hitsStr, 10, 64)
|
||||
misses, _ := strconv.ParseInt(missesStr, 10, 64)
|
||||
if total := hits + misses; total > 0 {
|
||||
metric.HitRate = float64(hits) / float64(total) * 100
|
||||
}
|
||||
|
||||
return metric, nil
|
||||
}
|
||||
|
||||
// collectMySQL 连接 MySQL 执行 SHOW GLOBAL STATUS 并解析指标
|
||||
func (m *MiddlewareCollector) collectMySQL(addr string, elapsedSec float64) (model.MySQLMetric, error) {
|
||||
metric := model.MySQLMetric{Instance: addr}
|
||||
|
||||
conn, err := net.DialTimeout("tcp", addr, 3*time.Second)
|
||||
if err != nil {
|
||||
return metric, fmt.Errorf("mysql connect %s: %w", addr, err)
|
||||
}
|
||||
defer conn.Close()
|
||||
|
||||
// MySQL handshake — minimalist: just send SHOW GLOBAL STATUS as query
|
||||
// This requires a real MySQL client protocol, which is complex without go-sql-driver.
|
||||
// Use a simple approach: try to connect and parse handshake for version
|
||||
buf := make([]byte, 4096)
|
||||
conn.SetReadDeadline(time.Now().Add(3 * time.Second))
|
||||
n, err := conn.Read(buf)
|
||||
if err != nil {
|
||||
return metric, fmt.Errorf("mysql handshake read: %w", err)
|
||||
}
|
||||
|
||||
// MySQL 握手包: 第 1 字节为协议版本(0a)，第 2 字节起为服务器版本字符串（以 00 结尾）
|
||||
if n < 5 {
|
||||
return metric, fmt.Errorf("mysql handshake too short")
|
||||
}
|
||||
if buf[0] != 0x0a && buf[0] != 10 {
|
||||
// 可能是 MariaDB 或其他变体 — 尝试跳过
|
||||
}
|
||||
|
||||
// 提取版本字符串
|
||||
versionEnd := -1
|
||||
for i := 1; i < n; i++ {
|
||||
if buf[i] == 0 {
|
||||
versionEnd = i
|
||||
break
|
||||
}
|
||||
}
|
||||
if versionEnd > 1 {
|
||||
metric.Version = string(buf[1:versionEnd])
|
||||
}
|
||||
|
||||
// For full MySQL metrics, we'd need go-sql-driver/mysql for the protocol handshake + query.
|
||||
// For now, just fill with basic info from the handshake.
|
||||
// Production: use database/sql + go-sql-driver/mysql
|
||||
|
||||
return metric, nil
|
||||
}
|
||||
|
||||
// extractInfoField 从 Redis INFO 原始输出中提取指定 key 的值
|
||||
func extractInfoField(raw, key string) string {
|
||||
prefix := key + ":"
|
||||
for _, line := range strings.Split(raw, "\n") {
|
||||
line = strings.TrimSpace(line)
|
||||
if strings.HasPrefix(line, prefix) {
|
||||
return line[len(prefix):]
|
||||
}
|
||||
}
|
||||
return ""
|
||||
}
|
||||
@@ -1,217 +0,0 @@
|
||||
// Package collector 提供系统信息及进程列表采集。
|
||||
// 系统信息：OS 名称、内核版本、架构、ulimit。
|
||||
// 进程列表：按 CPU/内存排序的 Top 5 进程。
|
||||
//
|
||||
// 署名 Logan.Li
|
||||
package collector
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"os"
|
||||
"sort"
|
||||
"strconv"
|
||||
"strings"
|
||||
"syscall"
|
||||
|
||||
"ezops/agent/model"
|
||||
)
|
||||
|
||||
// SystemCollector 系统信息采集器
|
||||
type SystemCollector struct{}
|
||||
|
||||
// NewSystemCollector 创建系统信息采集器
|
||||
func NewSystemCollector() *SystemCollector {
|
||||
return &SystemCollector{}
|
||||
}
|
||||
|
||||
// Collect 采集系统信息（OS、内核、架构、ulimit）
|
||||
func (s *SystemCollector) Collect() (model.SystemStat, error) {
|
||||
return model.SystemStat{
|
||||
OS: readFirstLine("/etc/os-release", "PRETTY_NAME="),
|
||||
Kernel: readProcVersion(),
|
||||
Arch: readArch(),
|
||||
Hostname: readHostname(),
|
||||
UlimitNofile: readUlimitNofile(),
|
||||
UlimitNproc: readUlimitNproc(),
|
||||
}, nil
|
||||
}
|
||||
|
||||
func readFirstLine(path, prefix string) string {
|
||||
data, err := os.ReadFile(path)
|
||||
if err != nil {
|
||||
return ""
|
||||
}
|
||||
for _, line := range strings.Split(string(data), "\n") {
|
||||
line = strings.TrimSpace(line)
|
||||
if strings.HasPrefix(line, prefix) {
|
||||
val := strings.TrimPrefix(line, prefix)
|
||||
return strings.Trim(val, `"`)
|
||||
}
|
||||
}
|
||||
return ""
|
||||
}
|
||||
|
||||
func readProcVersion() string {
|
||||
data, err := os.ReadFile("/proc/version")
|
||||
if err != nil {
|
||||
return ""
|
||||
}
|
||||
// Linux version 6.8.0-90-generic ... (buildd@...) (gcc ...) #...
|
||||
parts := strings.Fields(string(data))
|
||||
if len(parts) >= 3 {
|
||||
return strings.Join(parts[:3], " ")
|
||||
}
|
||||
return strings.TrimSpace(string(data))
|
||||
}
|
||||
|
||||
func readArch() string {
|
||||
data, err := os.ReadFile("/proc/cpuinfo")
|
||||
if err != nil {
|
||||
return ""
|
||||
}
|
||||
for _, line := range strings.Split(string(data), "\n") {
|
||||
if strings.HasPrefix(line, "flags") || strings.HasPrefix(line, "Features") {
|
||||
return "x86_64"
|
||||
}
|
||||
}
|
||||
return "unknown"
|
||||
}
|
||||
|
||||
func readHostname() string {
|
||||
h, _ := os.Hostname()
|
||||
return h
|
||||
}
|
||||
|
||||
func readUlimitNofile() int64 {
|
||||
var rlim syscall.Rlimit
|
||||
if err := syscall.Getrlimit(syscall.RLIMIT_NOFILE, &rlim); err == nil {
|
||||
return int64(rlim.Cur)
|
||||
}
|
||||
return 0
|
||||
}
|
||||
|
||||
func readUlimitNproc() int64 {
|
||||
var rlim syscall.Rlimit
|
||||
if err := syscall.Getrlimit(syscall.RLIMIT_NPROC, &rlim); err == nil {
|
||||
return int64(rlim.Cur)
|
||||
}
|
||||
return 0
|
||||
}
|
||||
|
||||
// ==================== Process Collector ====================
|
||||
|
||||
// ProcessCollector 进程列表采集器 — Top N 按 CPU/内存
|
||||
type ProcessCollector struct{}
|
||||
|
||||
// NewProcessCollector 创建进程采集器
|
||||
func NewProcessCollector() *ProcessCollector {
|
||||
return &ProcessCollector{}
|
||||
}
|
||||
|
||||
// Collect 采集 Top 5 CPU 和 Top 5 内存进程
|
||||
func (p *ProcessCollector) Collect() (model.ProcessStat, error) {
|
||||
procs := readProcStatAll()
|
||||
if len(procs) == 0 {
|
||||
return model.ProcessStat{}, nil
|
||||
}
|
||||
|
||||
// Top 5 by CPU
|
||||
sort.Slice(procs, func(i, j int) bool { return procs[i].CPUPercent > procs[j].CPUPercent })
|
||||
topCPU := procs
|
||||
if len(topCPU) > 5 {
|
||||
topCPU = topCPU[:5]
|
||||
}
|
||||
|
||||
// Top 5 by RSS
|
||||
sort.Slice(procs, func(i, j int) bool { return procs[i].RSSKB > procs[j].RSSKB })
|
||||
topMem := procs
|
||||
if len(topMem) > 5 {
|
||||
topMem = topMem[:5]
|
||||
}
|
||||
|
||||
return model.ProcessStat{
|
||||
TopCPU: topCPU,
|
||||
TopMem: topMem,
|
||||
}, nil
|
||||
}
|
||||
|
||||
type procInfo struct {
|
||||
PID int
|
||||
Name string
|
||||
CPUPercent float64
|
||||
RSSKB int64
|
||||
}
|
||||
|
||||
func readProcStatAll() []procInfo {
|
||||
entries, err := os.ReadDir("/proc")
|
||||
if err != nil {
|
||||
return nil
|
||||
}
|
||||
|
||||
var procs []procInfo
|
||||
for _, entry := range entries {
|
||||
if !entry.IsDir() {
|
||||
continue
|
||||
}
|
||||
pid, err := strconv.Atoi(entry.Name())
|
||||
if err != nil {
|
||||
continue
|
||||
}
|
||||
|
||||
statData, err := os.ReadFile(fmt.Sprintf("/proc/%d/stat", pid))
|
||||
if err != nil {
|
||||
continue
|
||||
}
|
||||
stat := string(statData)
|
||||
|
||||
// /proc/PID/stat: PID (comm) state ppid ...
|
||||
// comm 在括号中，可能含空格
|
||||
closeParen := strings.LastIndex(stat, ")")
|
||||
if closeParen < 0 {
|
||||
continue
|
||||
}
|
||||
openParen := strings.Index(stat, "(")
|
||||
if openParen < 0 {
|
||||
continue
|
||||
}
|
||||
name := stat[openParen+1 : closeParen]
|
||||
|
||||
fields := strings.Fields(stat[closeParen+2:])
|
||||
if len(fields) < 20 {
|
||||
continue
|
||||
}
|
||||
|
||||
// utime=fields[11], stime=fields[12], rss=fields[21]
|
||||
utime, _ := strconv.ParseInt(fields[11], 10, 64)
|
||||
stime, _ := strconv.ParseInt(fields[12], 10, 64)
|
||||
rss, _ := strconv.ParseInt(fields[21], 10, 64)
|
||||
|
||||
// RSS pages * 4KB = RSS in KB
|
||||
rssKB := rss * 4
|
||||
|
||||
// Simple CPU%: total ticks as fraction of uptime (approximate)
|
||||
totalTicks := utime + stime
|
||||
uptimeData, _ := os.ReadFile("/proc/uptime")
|
||||
uptime := 0.0
|
||||
if uptimeData != nil {
|
||||
uptime, _ = strconv.ParseFloat(strings.Fields(string(uptimeData))[0], 64)
|
||||
}
|
||||
hertz := float64(syscall.Getpagesize()) * 25
|
||||
if hertz == 0 {
|
||||
hertz = 100
|
||||
}
|
||||
// CLK_TCK is typically 100 on Linux
|
||||
cpuPercent := 0.0
|
||||
if uptime > 0 {
|
||||
cpuPercent = float64(totalTicks) / 100.0 / uptime * 100
|
||||
}
|
||||
|
||||
procs = append(procs, procInfo{
|
||||
PID: pid,
|
||||
Name: name,
|
||||
CPUPercent: cpuPercent,
|
||||
RSSKB: rssKB,
|
||||
})
|
||||
}
|
||||
return procs
|
||||
}
|
||||
Reference in new issue
Block a user