513 lines
15 KiB
Go
513 lines
15 KiB
Go
package handlers
|
||
|
||
import (
|
||
"encoding/json"
|
||
"fmt"
|
||
"math"
|
||
"net/http"
|
||
"sort"
|
||
"strconv"
|
||
"strings"
|
||
"time"
|
||
|
||
"ezops/internal/models"
|
||
ezredis "ezops/pkg/redis"
|
||
|
||
"github.com/gin-gonic/gin"
|
||
"gorm.io/gorm"
|
||
"gorm.io/gorm/clause"
|
||
)
|
||
|
||
// AgentHandler 处理 Agent 上报数据 // Logan.Li
|
||
type AgentHandler struct {
|
||
db *gorm.DB
|
||
cache ezredis.Cache // 可选 Redis 缓存（用于热数据存储）
|
||
hubMetrics *MetricsHub // WebSocket 广播 metrics 更新
|
||
}
|
||
|
||
func NewAgentHandler(db *gorm.DB, cache ezredis.Cache, hub *MetricsHub) *AgentHandler {
|
||
return &AgentHandler{db: db, cache: cache, hubMetrics: hub}
|
||
}
|
||
|
||
// ReceiveReport 接收 Agent 上报的指标数据
|
||
// POST /api/v1/agent/report
|
||
func (h *AgentHandler) ReceiveReport(c *gin.Context) {
|
||
var req models.AgentReportRequest
|
||
if err := c.ShouldBindJSON(&req); err != nil {
|
||
c.JSON(http.StatusBadRequest, models.Response{Code: 400, Message: "请求格式错误: " + err.Error()})
|
||
return
|
||
}
|
||
|
||
// 解析上报时间
|
||
reportTime := time.Now()
|
||
if req.Timestamp != "" {
|
||
if t, err := time.Parse(time.RFC3339, req.Timestamp); err == nil {
|
||
reportTime = t
|
||
}
|
||
}
|
||
|
||
// 序列化磁盘和端口信息
|
||
diskJSON, _ := json.Marshal(req.Disks)
|
||
portJSON, _ := json.Marshal(req.Ports)
|
||
netJSON, _ := json.Marshal(req.Network)
|
||
dioJSON, _ := json.Marshal(req.DiskIO)
|
||
|
||
// 存储到数据库
|
||
report := models.AgentReport{
|
||
Hostname: req.Hostname,
|
||
CPUUsage: req.CPU.UsagePercent,
|
||
CPUCores: req.CPU.CoreCount,
|
||
Load1: req.CPU.Load1,
|
||
Load5: req.CPU.Load5,
|
||
Load15: req.CPU.Load15,
|
||
MemTotal: req.Memory.TotalGB,
|
||
MemUsed: req.Memory.UsedGB,
|
||
MemUsage: req.Memory.UsagePercent,
|
||
DiskInfo: string(diskJSON),
|
||
PortInfo: string(portJSON),
|
||
NetInfo: string(netJSON),
|
||
DiskIOInfo: string(dioJSON),
|
||
ReportTime: reportTime,
|
||
}
|
||
|
||
if err := h.db.Create(&report).Error; err != nil {
|
||
c.JSON(http.StatusInternalServerError, models.Response{Code: 500, Message: "保存失败: " + err.Error()})
|
||
return
|
||
}
|
||
|
||
// 同步更新 servers 表的实时指标（三级匹配：hostname → client IP → hostname LIKE）
|
||
diskAvg := avgDiskUsage(req.Disks)
|
||
|
||
// 计算硬件规格（四舍五入到整数，避免 31.34→31GB 的显示偏差）
|
||
totalDiskGB := 0.0
|
||
for _, d := range req.Disks {
|
||
totalDiskGB += d.TotalGB
|
||
}
|
||
memGB := math.Ceil(req.Memory.TotalGB) // 31.34 → 32, 62.79 → 63（物理内存向上取整）
|
||
memMB := int64(memGB * 1024) // GB → MB
|
||
diskGB := int64(totalDiskGB + 0.5) // 39.99 → 40
|
||
|
||
clientIP := c.ClientIP()
|
||
|
||
// 三级匹配：优先用 Agent 上报的 IP，其次用请求源 IP，最后用报文的 hostname
|
||
matchIP := req.IP
|
||
if matchIP == "" {
|
||
matchIP = clientIP
|
||
}
|
||
|
||
// 直接 SQL 更新 asset_servers 监控字段
|
||
// 若 AgentVersion 非空则同步写入，否则保留数据库中已有版本号（防止旧版Agent空字段覆盖已设置的版本）
|
||
updateSQL := `UPDATE asset_servers SET
|
||
cpu_usage = ?, memory_usage = ?, disk_usage = ?,
|
||
cpu = ?, memory = ?, disk = ?,
|
||
status = ?, agent_status = ?` + func() string {
|
||
if req.AgentVersion != "" {
|
||
return `, agent_version = ?`
|
||
}
|
||
return ``
|
||
}() + `,
|
||
hostname = CASE WHEN hostname != ? AND ip = ? THEN ? ELSE hostname END,
|
||
updated_at = NOW()`
|
||
args := []interface{}{
|
||
req.CPU.UsagePercent, req.Memory.UsagePercent, diskAvg,
|
||
req.CPU.CoreCount, memMB, diskGB,
|
||
"running", "online",
|
||
}
|
||
if req.AgentVersion != "" {
|
||
args = append(args, req.AgentVersion)
|
||
}
|
||
args = append(args,
|
||
// CASE WHEN hostname != ? AND ip = ? THEN ? ELSE hostname END
|
||
req.Hostname, matchIP, req.Hostname,
|
||
)
|
||
if req.System != nil {
|
||
if req.System.OSVersion != "" {
|
||
updateSQL += ", os = ?"
|
||
args = append(args, req.System.OSVersion)
|
||
}
|
||
if req.System.Kernel != "" {
|
||
updateSQL += ", kernel = ?"
|
||
args = append(args, req.System.Kernel)
|
||
}
|
||
if req.System.Uptime > 0 {
|
||
updateSQL += ", uptime = ?"
|
||
args = append(args, req.System.Uptime)
|
||
}
|
||
}
|
||
updateSQL += " WHERE hostname = ? OR ip = ?"
|
||
args = append(args, req.Hostname, matchIP)
|
||
|
||
result := h.db.Exec(updateSQL, args...)
|
||
|
||
// 如果 hostname 和 IP 都没匹配到，尝试用 hostname 反向匹配 IP 字段
|
||
if result.RowsAffected == 0 && req.Hostname != matchIP {
|
||
updateSQL2 := updateSQL + " OR ip = ?"
|
||
args2 := append(args, req.Hostname)
|
||
result = h.db.Exec(updateSQL2, args2...)
|
||
}
|
||
|
||
// 如果还是没匹配到 —— 自动注册新服务器（UPSERT 防重复）
|
||
if result.RowsAffected == 0 {
|
||
osName := ""
|
||
if req.System != nil {
|
||
osName = req.System.OSVersion
|
||
}
|
||
newServer := models.Server{
|
||
Hostname: req.Hostname,
|
||
IP: matchIP,
|
||
Port: 22,
|
||
Username: "root",
|
||
AuthType: "password",
|
||
OS: osName,
|
||
CPU: req.CPU.CoreCount,
|
||
Memory: memMB,
|
||
Disk: diskGB,
|
||
Status: "running",
|
||
AgentStatus: "online",
|
||
AgentVersion: req.AgentVersion,
|
||
CPUUsage: req.CPU.UsagePercent,
|
||
MemoryUsage: req.Memory.UsagePercent,
|
||
DiskUsage: diskAvg,
|
||
}
|
||
// ON DUPLICATE KEY UPDATE — 若 hostname 已存在则更新字段（防重复注册）
|
||
if err := h.db.Clauses(clause.OnConflict{
|
||
Columns: []clause.Column{{Name: "hostname"}},
|
||
DoUpdates: clause.AssignmentColumns([]string{
|
||
"ip", "os", "cpu", "memory", "disk",
|
||
"status", "agent_status",
|
||
"cpu_usage", "memory_usage", "disk_usage",
|
||
"updated_at",
|
||
}),
|
||
}).Create(&newServer).Error; err != nil {
|
||
fmt.Printf("[agent] 自动注册主机 %q (IP: %s) 失败: %v\n", req.Hostname, matchIP, err)
|
||
} else {
|
||
fmt.Printf("[agent] 自动注册新主机: %q (IP: %s, CPU=%d核, 内存=%dMB, 磁盘=%dGB, OS=%q)\n",
|
||
req.Hostname, matchIP, req.CPU.CoreCount, memMB, diskGB, osName)
|
||
}
|
||
}
|
||
|
||
// WebSocket 广播：向所有前端客户端推送该主机的最新指标 // Logan.Li
|
||
if h.hubMetrics != nil && result.RowsAffected > 0 {
|
||
h.hubMetrics.Broadcast(MetricsUpdate{
|
||
Type: "metrics_update",
|
||
Hostname: req.Hostname,
|
||
CPU: req.CPU.UsagePercent,
|
||
Memory: req.Memory.UsagePercent,
|
||
Disk: diskAvg,
|
||
Status: "running",
|
||
TCPEstablished: req.Network.TCPEstablished,
|
||
TCPTimeWait: req.Network.TCPTimeWait,
|
||
NetRxBPS: req.Network.NetRxBPS,
|
||
NetTxBPS: req.Network.NetTxBPS,
|
||
DiskReadBPS: req.DiskIO.ReadBPS,
|
||
DiskWriteBPS: req.DiskIO.WriteBPS,
|
||
})
|
||
}
|
||
|
||
// 写入 Redis 有序集合（热数据，保留7天）
|
||
if h.cache != nil {
|
||
timestamp := float64(reportTime.Unix())
|
||
ttl := 7 * 24 * time.Hour // 7天
|
||
|
||
// 使用 reportTime.UnixNano() 确保 member 唯一（避免相同值碰撞）
|
||
h.writeHotMetric(c, "metrics:cpu:"+req.Hostname, req.CPU.UsagePercent, timestamp, ttl, reportTime)
|
||
h.writeHotMetric(c, "metrics:mem:"+req.Hostname, req.Memory.UsagePercent, timestamp, ttl, reportTime)
|
||
h.writeHotMetric(c, "metrics:load1:"+req.Hostname, req.CPU.Load1, timestamp, ttl, reportTime)
|
||
}
|
||
|
||
// 保存成功
|
||
c.JSON(http.StatusOK, models.Response{Code: 200, Message: "ok"})
|
||
}
|
||
|
||
// writeHotMetric 将单个指标写入 Redis 有序集合并维护 TTL 和过期数据清理
|
||
func (h *AgentHandler) writeHotMetric(c *gin.Context, key string, value float64, score float64, ttl time.Duration, reportTime time.Time) {
|
||
ctx := c.Request.Context()
|
||
|
||
// member 格式: "value:nano_timestamp" 保证唯一性
|
||
member := fmt.Sprintf("%.4f:%d", value, reportTime.UnixNano())
|
||
|
||
if err := h.cache.ZAdd(ctx, key, ezredis.SortedSetMember{
|
||
Score: score,
|
||
Member: member,
|
||
}); err != nil {
|
||
// 静默失败，不影响主流程
|
||
return
|
||
}
|
||
|
||
// 设置 key 过期时间
|
||
_ = h.cache.Expire(ctx, key, ttl)
|
||
|
||
// 清理 7 天前的旧数据
|
||
cutoff := float64(time.Now().Add(-ttl).Unix())
|
||
_ = h.cache.ZRemRangeByScore(ctx, key, "-inf", fmt.Sprintf("%.0f", cutoff))
|
||
}
|
||
|
||
// GetLatestReports 获取所有服务器的 Agent 最新状态（含未安装 Agent 的主机）
|
||
// GET /api/v1/agent/latest
|
||
func (h *AgentHandler) GetLatestReports(c *gin.Context) {
|
||
var results []models.AgentLatestResponse
|
||
threeMinAgo := time.Now().Add(-3 * time.Minute)
|
||
// 以 servers 表为主，LEFT JOIN 最近 3 分钟内的 Agent 上报
|
||
// 未装 Agent 的主机也会显示（agent_status='not_installed'），打破"不显示→无法安装"的循环
|
||
h.db.Raw(`
|
||
SELECT
|
||
s.id as server_id,
|
||
s.hostname,
|
||
COALESCE(s.ip, '') as ip,
|
||
COALESCE(a.cpu_usage, 0) as cpu_usage,
|
||
COALESCE(a.cpu_cores, 0) as cpu_cores,
|
||
COALESCE(a.mem_usage, 0) as mem_usage,
|
||
COALESCE(a.mem_total, 0) as mem_total,
|
||
COALESCE(a.mem_used, 0) as mem_used,
|
||
COALESCE(s.disk_usage, 0) as disk_usage,
|
||
a.disk_info,
|
||
a.port_info,
|
||
a.net_info,
|
||
a.disk_io_info,
|
||
a.report_time,
|
||
COALESCE(s.agent_version, '') as agent_version,
|
||
CASE
|
||
WHEN a.hostname IS NOT NULL THEN COALESCE(s.agent_status, 'online')
|
||
ELSE 'not_installed'
|
||
END as agent_status,
|
||
COALESCE(s.os, '') as os,
|
||
COALESCE(s.kernel, '') as kernel,
|
||
COALESCE(s.uptime, 0) as uptime,
|
||
COALESCE(s.disk, 0) as disk_total,
|
||
COALESCE(r.name, '') as region,
|
||
CASE WHEN s.password != '' OR s.ssh_key != '' THEN true ELSE false END as has_auth
|
||
FROM asset_servers s
|
||
LEFT JOIN (
|
||
SELECT * FROM monitor_agent_reports
|
||
WHERE id IN (
|
||
SELECT MAX(id) FROM monitor_agent_reports
|
||
WHERE report_time >= ?
|
||
GROUP BY hostname
|
||
)
|
||
) a ON a.hostname = s.hostname
|
||
LEFT JOIN asset_regions r ON r.id = s.region_id
|
||
ORDER BY a.report_time DESC, s.id ASC
|
||
`, threeMinAgo).Scan(&results)
|
||
|
||
if results == nil {
|
||
results = []models.AgentLatestResponse{}
|
||
}
|
||
|
||
c.JSON(http.StatusOK, models.Response{
|
||
Code: 200,
|
||
Message: "success",
|
||
Data: results,
|
||
})
|
||
}
|
||
|
||
// GetHostHistory 获取指定主机的历史指标
|
||
// GET /api/v1/agent/history/:hostname?hours=1
|
||
func (h *AgentHandler) GetHostHistory(c *gin.Context) {
|
||
hostname := c.Param("hostname")
|
||
hours := 1
|
||
if v := c.Query("hours"); v != "" {
|
||
fmt.Sscanf(v, "%d", &hours)
|
||
}
|
||
since := time.Now().Add(-time.Duration(hours) * time.Hour)
|
||
|
||
var reports []models.AgentReport
|
||
h.db.Where("hostname = ? AND report_time >= ?", hostname, since).
|
||
Order("report_time ASC").
|
||
Find(&reports)
|
||
|
||
c.JSON(http.StatusOK, models.Response{
|
||
Code: 200,
|
||
Message: "success",
|
||
Data: reports,
|
||
})
|
||
}
|
||
|
||
// AggregateMetrics 聚合 Redis 热数据到 MySQL klines
|
||
// POST /api/v1/agent/aggregate
|
||
func (h *AgentHandler) AggregateMetrics(c *gin.Context) {
|
||
var req models.AggregateReq
|
||
if err := c.ShouldBindJSON(&req); err != nil {
|
||
c.JSON(http.StatusBadRequest, models.Response{Code: 400, Message: "请求格式错误: " + err.Error()})
|
||
return
|
||
}
|
||
|
||
if h.cache == nil {
|
||
c.JSON(http.StatusServiceUnavailable, models.Response{Code: 503, Message: "Redis 缓存不可用"})
|
||
return
|
||
}
|
||
|
||
// 支持的聚合间隔及对应的回溯时长
|
||
intervals := []struct {
|
||
name string
|
||
duration time.Duration
|
||
}{
|
||
{"1m", 1 * time.Minute},
|
||
{"5m", 5 * time.Minute},
|
||
{"3m", 3 * time.Minute},
|
||
{"1h", 1 * time.Hour},
|
||
{"6h", 6 * time.Hour},
|
||
{"1d", 24 * time.Hour},
|
||
}
|
||
|
||
redisKey := fmt.Sprintf("metrics:%s:%s", req.Metric, req.Hostname)
|
||
ctx := c.Request.Context()
|
||
var totalUpserted int64
|
||
|
||
for _, iv := range intervals {
|
||
// 回溯窗口：取 interval × 200 倍，确保覆盖足够多的数据点
|
||
lookback := iv.duration * 200
|
||
now := time.Now()
|
||
since := now.Add(-lookback)
|
||
minScore := fmt.Sprintf("%d", since.Unix())
|
||
maxScore := fmt.Sprintf("%d", now.Unix()+1)
|
||
|
||
members, err := h.cache.ZRangeByScore(ctx, redisKey, minScore, maxScore)
|
||
if err != nil || len(members) == 0 {
|
||
continue
|
||
}
|
||
|
||
// 按间隔分桶
|
||
buckets := make(map[int64][]float64) // bucketTime (unix) → values
|
||
for _, m := range members {
|
||
ts := int64(m.Score)
|
||
bucketStart := (ts / int64(iv.duration.Seconds())) * int64(iv.duration.Seconds())
|
||
val := parseMemberValue(m.Member)
|
||
buckets[bucketStart] = append(buckets[bucketStart], val)
|
||
}
|
||
|
||
// 计算 max/min/avg/p95
|
||
for bucketTs, values := range buckets {
|
||
if len(values) == 0 {
|
||
continue
|
||
}
|
||
maxV := values[0]
|
||
minV := values[0]
|
||
sum := 0.0
|
||
|
||
for _, v := range values {
|
||
sum += v
|
||
if v > maxV {
|
||
maxV = v
|
||
}
|
||
if v < minV {
|
||
minV = v
|
||
}
|
||
}
|
||
avg := sum / float64(len(values))
|
||
|
||
// 计算 P95（排序后取 95% 位置的值）
|
||
p95 := maxV
|
||
if len(values) >= 20 {
|
||
sorted := make([]float64, len(values))
|
||
copy(sorted, values)
|
||
sort.Float64s(sorted)
|
||
idx := int(float64(len(sorted)) * 0.95)
|
||
if idx >= len(sorted) {
|
||
idx = len(sorted) - 1
|
||
}
|
||
p95 = sorted[idx]
|
||
}
|
||
|
||
agg := models.MetricsAggregate{
|
||
Hostname: req.Hostname,
|
||
Metric: req.Metric,
|
||
Interval: iv.name,
|
||
Avg: math.Round(avg*100) / 100,
|
||
MaxV: math.Round(maxV*100) / 100,
|
||
MinV: math.Round(minV*100) / 100,
|
||
P95: math.Round(p95*100) / 100,
|
||
Count: len(values),
|
||
BucketTime: time.Unix(bucketTs, 0),
|
||
}
|
||
|
||
// Upsert: hostname + metric + interval + bucket_time 唯一
|
||
result := h.db.Clauses(clause.OnConflict{
|
||
Columns: []clause.Column{
|
||
{Name: "hostname"},
|
||
{Name: "metric"},
|
||
{Name: "interval"},
|
||
{Name: "bucket_time"},
|
||
},
|
||
DoUpdates: clause.AssignmentColumns([]string{
|
||
"avg", "max_v", "min_v", "p95", "count",
|
||
}),
|
||
}).Create(&agg)
|
||
|
||
if result.Error == nil {
|
||
totalUpserted++
|
||
}
|
||
}
|
||
}
|
||
|
||
c.JSON(http.StatusOK, models.Response{
|
||
Code: 200,
|
||
Message: fmt.Sprintf("聚合完成，写入/更新 %d 条记录", totalUpserted),
|
||
Data: gin.H{"upserted": totalUpserted},
|
||
})
|
||
}
|
||
|
||
// GetAggregate 查询聚合数据（原 GetKline）
|
||
// GET /api/v1/agent/aggregate?hostname=&metric=&interval=&from=&to=
|
||
func (h *AgentHandler) GetAggregate(c *gin.Context) {
|
||
var req models.AggregateQueryReq
|
||
if err := c.ShouldBindQuery(&req); err != nil {
|
||
c.JSON(http.StatusBadRequest, models.Response{Code: 400, Message: "参数错误: " + err.Error()})
|
||
return
|
||
}
|
||
|
||
from, err := time.Parse(time.RFC3339, req.From)
|
||
if err != nil {
|
||
c.JSON(http.StatusBadRequest, models.Response{Code: 400, Message: "from 格式错误，需 RFC3339: " + err.Error()})
|
||
return
|
||
}
|
||
to, err := time.Parse(time.RFC3339, req.To)
|
||
if err != nil {
|
||
c.JSON(http.StatusBadRequest, models.Response{Code: 400, Message: "to 格式错误，需 RFC3339: " + err.Error()})
|
||
return
|
||
}
|
||
|
||
var items []models.MetricsAggregate
|
||
result := h.db.Where("hostname = ? AND metric = ? AND `interval` = ? AND bucket_time >= ? AND bucket_time <= ?",
|
||
req.Hostname, req.Metric, req.Interval, from, to).
|
||
Order("bucket_time ASC").
|
||
Find(&items)
|
||
|
||
if result.Error != nil {
|
||
c.JSON(http.StatusInternalServerError, models.Response{Code: 500, Message: "查询失败: " + result.Error.Error()})
|
||
return
|
||
}
|
||
|
||
if items == nil {
|
||
items = []models.MetricsAggregate{}
|
||
}
|
||
|
||
c.JSON(http.StatusOK, models.Response{
|
||
Code: 200,
|
||
Message: "success",
|
||
Data: items,
|
||
})
|
||
}
|
||
|
||
// parseMemberValue 从 member 字符串 "45.2300:1234567890123456789" 中提取浮点值
|
||
func parseMemberValue(member string) float64 {
|
||
idx := strings.Index(member, ":")
|
||
if idx < 0 {
|
||
v, _ := strconv.ParseFloat(member, 64)
|
||
return v
|
||
}
|
||
v, _ := strconv.ParseFloat(member[:idx], 64)
|
||
return v
|
||
}
|
||
|
||
// avgDiskUsage 计算磁盘平均使用率
|
||
func avgDiskUsage(disks []models.DiskMetric) float64 {
|
||
if len(disks) == 0 {
|
||
return 0
|
||
}
|
||
var sum float64
|
||
for _, d := range disks {
|
||
sum += d.UsagePercent
|
||
}
|
||
return math.Round(sum/float64(len(disks))*100) / 100
|
||
}
|