feat(aiops): 新增「智能运维」——AI 日志分析(P0,不扫日志内容)
=== 为什么这样设计(成本优先)===
全集群 370GB/天,把日志原文丢给模型是不可能的(一次几百万 token)。
所以整条链路只做一件事:**把 GB 级日志压成 KB 级信号**。
P0 的信号来源全部是索引/快照,**零日志内容扫描**:
体积 + 趋势 → index/volume_range(实测 24h 仅 54ms)
条数/流数 → index/stats,且只在 ≤2 小时窗口用
(全天 index/stats 实测 20s,绝不能进 AI 路径)
应用排行 → 平台小时级预聚合快照(毫秒)
基线对比 → 上一个等长窗口的 volume_range
告警 → 平台投递记录
实测一次分析:3~6 秒、输入 600~1400 tokens(约 ¥0.005/次)。
=== 后端 ===
models/ai.go AIConfig(APIKey 加密存储)+ AIAnalysisRun(审计)
services/ai_config.go 读写配置、掩码返回、每日 token 用量统计
services/ai_provider.go OpenAI 兼容客户端 + 连通性测试
(DeepSeek/通义/内网 vLLM/Ollama 通用)
services/ai_redact.go 脱敏:凭据/邮箱/手机号/身份证 → 占位符,
**保留内网 IP 与端口**(排查必需)
services/ai_signal.go 信号采集 + 渲染成紧凑文本(省 token)
services/ai_analyze.go 结构化输出解析、配额闸门、审计落库
controllers/ai.go /ai/config /ai/config/test /ai/analyze /ai/runs
成本闸门(三层):
1) 单次:时间范围 ≤24h、maxTokens ≤1500
2) 每用户每日 token 上限(默认 50 万,可配,超限直接拒绝并说明)
3) 审计只留最近 500 条,避免表膨胀
=== 前端 ===
新页面 /aiops(侧栏「智能运维」,管理员),页内单层 tab:
日志分析 / 模型配置 / 分析历史
- 分析:数据源 + 时间预设(15m/1h/6h/24h) + 可选提问 → 结论/证据/建议/不确定 四区
顶部显示本次消耗(in/out tokens、耗时),可查看「发给模型的完整内容」
- 配置:启用开关、Base URL、模型、API Key(掩码)、温度、输出上限、超时、
每日配额、额外脱敏正则;按钮顺序改为「测试连接 → 保存配置」
- 历史:服务端分页(20/页),点行看详情与 payload
=== 两个真实踩坑(已修)===
1) 密码含 @ 导致脱敏漏掉(mysql://user:P@ssw0rd@host)
→ URL basic-auth 规则改为贪婪匹配到最后一个 @,由单测锁定
2) 结论里的时间是 UTC(容器无 tzdata,time.Local=UTC),
模型写"凌晨 6 点下跌"而运维当时是下午 2 点
→ 显式用东八区展示(LOKISTACK_DISPLAY_TZ=UTC 可覆盖)
=== 实测(69147008fcf4)===
配置保存 → key 掩码 sk-***e286;连通性测试返回「正常」
分析 15 分钟窗口:4.1s,结论识别出"窗口末段体积断崖约 -60%"
信号时间 14:23~14:38(本地时间正确)
历史分页 total=3,tokens/耗时字段完整
单测:脱敏 2 个用例 + 模型输出解析 4 个用例 + 6 条 AI 路由断言
This commit is contained in:
1 parent
3861b8b128
commit
7be41f0a6b
19 files changed
+2236
-1
No files matched your search
@@ -0,0 +1,243 @@
|
||||
package services
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"lokistack/models"
|
||||
"lokistack/repository"
|
||||
)
|
||||
|
||||
// AIAnalysisResult 结构化结论。
|
||||
//
|
||||
// 强制模型输出这个结构而不是自由文本,原因:
|
||||
// 1. 前端能分区渲染(结论/证据/建议/不确定),可读性远好于一段散文;
|
||||
// 2. 每条结论都要挂 evidence,逼模型把话落在数据上,降低编造;
|
||||
// 3. unknowns 单独一栏,明确"不知道什么"比编一个答案有价值得多。
|
||||
type AIAnalysisResult struct {
|
||||
Summary string `json:"summary"`
|
||||
Findings []AIFinding `json:"findings"`
|
||||
Actions []string `json:"actions"`
|
||||
Unknowns []string `json:"unknowns"`
|
||||
RawText string `json:"-"`
|
||||
ParseErr string `json:"parseError,omitempty"`
|
||||
}
|
||||
|
||||
type AIFinding struct {
|
||||
Title string `json:"title"`
|
||||
Evidence string `json:"evidence"`
|
||||
Severity string `json:"severity"` // high | medium | low
|
||||
}
|
||||
|
||||
// AIAnalysisOutcome 一次分析的完整返回(含成本信息)。
|
||||
type AIAnalysisOutcome struct {
|
||||
Result AIAnalysisResult `json:"result"`
|
||||
Signals *AISignalBundle `json:"signals"`
|
||||
Prompt string `json:"prompt"`
|
||||
Run models.AIAnalysisRun `json:"run"`
|
||||
Warnings []string `json:"warnings"`
|
||||
}
|
||||
|
||||
const aiSystemPrompt = `你是资深 SRE,负责分析 Loki 日志平台的运行状况。
|
||||
|
||||
严格规则:
|
||||
1. 只能基于我给你的"信号数据"下结论。**不许编造**日志内容、错误信息、时间点或服务名。
|
||||
2. 每一条结论都必须在 evidence 里引用信号数据中的具体数字或条目。
|
||||
3. 如果信号数据不足以判断根因,把它写进 unknowns,不要猜。
|
||||
4. 你看不到日志原文(只有统计与排行),因此不要描述"日志里写了什么"。
|
||||
5. 输出必须是**单个 JSON 对象**,不要 markdown 代码块,不要任何解释性前后缀。
|
||||
|
||||
JSON 结构:
|
||||
{
|
||||
"summary": "两三句话的整体判断",
|
||||
"findings": [{"title": "简短结论", "evidence": "引用具体数据", "severity": "high|medium|low"}],
|
||||
"actions": ["可执行的下一步动作,按优先级排序"],
|
||||
"unknowns": ["无法从当前数据判断的点"]
|
||||
}`
|
||||
|
||||
// RunAIAnalysis 执行一次分析:采集信号 → 脱敏 → 调模型 → 解析 → 落审计。
|
||||
func RunAIAnalysis(ctx context.Context, cfg models.AIConfig, src *models.LogSource, userID uint, username string, admin bool, from, to time.Time, question string) (*AIAnalysisOutcome, error) {
|
||||
if !cfg.Enabled {
|
||||
return nil, fmt.Errorf("AI 分析未启用:请先在「智能运维 → 模型配置」中开启并保存")
|
||||
}
|
||||
|
||||
// ── 每日配额闸门(成本控制第一层)──
|
||||
if cfg.DailyTokenCap > 0 {
|
||||
used := AITokenUsedToday(userID)
|
||||
if used >= cfg.DailyTokenCap {
|
||||
run := models.AIAnalysisRun{
|
||||
UserID: userID, Username: username, SourceID: src.ID, SourceName: src.Name,
|
||||
Status: "quota-exceeded", Error: fmt.Sprintf("今日 AI token 用量已达上限(%d/%d)", used, cfg.DailyTokenCap),
|
||||
CreatedAt: time.Now(),
|
||||
}
|
||||
persistAIRun(&run)
|
||||
return nil, fmt.Errorf("今日 AI 用量已达上限(%d/%d tokens),可在模型配置中调整上限或明天再试", used, cfg.DailyTokenCap)
|
||||
}
|
||||
}
|
||||
|
||||
started := time.Now()
|
||||
signals, err := CollectAISignals(ctx, src, userID, admin, from, to)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
// 扫描预算闸门:P0 全是索引/快照,实际扫描为 0,这里留出扩展位
|
||||
if cfg.MaxScanBytes > 0 && signals.TotalBytes > cfg.MaxScanBytes*8 {
|
||||
signals.Warnings = append(signals.Warnings,
|
||||
fmt.Sprintf("该范围日志体积 %.1f GB 偏大,已只使用索引统计(未读取日志内容)",
|
||||
float64(signals.TotalBytes)/1024/1024/1024))
|
||||
}
|
||||
|
||||
// ── 脱敏(第二层:隐私)──
|
||||
rawPrompt := SignalsAsText(signals)
|
||||
redacted, invalidRules := RedactText(rawPrompt, cfg.RedactExtra)
|
||||
if question = strings.TrimSpace(question); question != "" {
|
||||
redacted += "\n\n运维补充的问题:" + RedactTextOnly(question, cfg.RedactExtra)
|
||||
}
|
||||
warnings := append([]string{}, signals.Warnings...)
|
||||
for _, r := range invalidRules {
|
||||
warnings = append(warnings, "额外脱敏规则无效已跳过:"+r)
|
||||
}
|
||||
|
||||
// ── 调模型 ──
|
||||
reply, promptTokens, completionTokens, callErr := aiChat(ctx, cfg, []aiMessage{
|
||||
{Role: "system", Content: aiSystemPrompt},
|
||||
{Role: "user", Content: redacted},
|
||||
})
|
||||
|
||||
run := models.AIAnalysisRun{
|
||||
UserID: userID, Username: username, SourceID: src.ID, SourceName: src.Name,
|
||||
RangeMinutes: signals.RangeMinutes, Model: cfg.Model, Question: question,
|
||||
PromptPayload: truncateStr(redacted, 20000),
|
||||
PromptTokens: promptTokens, CompletionTokens: completionTokens,
|
||||
ScanBytes: 0, // P0 不扫内容
|
||||
DurationMs: time.Since(started).Milliseconds(),
|
||||
CreatedAt: time.Now(),
|
||||
}
|
||||
if callErr != nil {
|
||||
run.Status = "failed"
|
||||
run.Error = callErr.Error()
|
||||
run.DurationMs = time.Since(started).Milliseconds()
|
||||
persistAIRun(&run)
|
||||
return nil, callErr
|
||||
}
|
||||
|
||||
result := parseAIResult(reply)
|
||||
conclusion, _ := json.Marshal(result)
|
||||
run.Status = "ok"
|
||||
run.Conclusion = string(conclusion)
|
||||
run.DurationMs = time.Since(started).Milliseconds()
|
||||
persistAIRun(&run)
|
||||
|
||||
return &AIAnalysisOutcome{
|
||||
Result: result, Signals: signals, Prompt: redacted, Run: run, Warnings: warnings,
|
||||
}, nil
|
||||
}
|
||||
|
||||
// RedactTextOnly 只脱敏不做规则校验提示(用于短文本如用户提问)。
|
||||
func RedactTextOnly(text string, extra []string) string {
|
||||
out, _ := RedactText(text, extra)
|
||||
return out
|
||||
}
|
||||
|
||||
// parseAIResult 宽松解析模型输出。
|
||||
//
|
||||
// 模型常见的"不听话"有三种:包了 ```json 代码块、前后加了解释、
|
||||
// 直接返回纯文本。前两种剥壳后重试,第三种降级为 summary=原文,
|
||||
// 保证用户至少能看到内容,而不是一个"解析失败"。
|
||||
func parseAIResult(reply string) AIAnalysisResult {
|
||||
text := strings.TrimSpace(reply)
|
||||
text = strings.TrimPrefix(text, "```json")
|
||||
text = strings.TrimPrefix(text, "```")
|
||||
text = strings.TrimSuffix(text, "```")
|
||||
text = strings.TrimSpace(text)
|
||||
if idx := strings.Index(text, "{"); idx > 0 {
|
||||
text = text[idx:]
|
||||
}
|
||||
if end := strings.LastIndex(text, "}"); end >= 0 && end < len(text)-1 {
|
||||
text = text[:end+1]
|
||||
}
|
||||
|
||||
var parsed AIAnalysisResult
|
||||
if err := json.Unmarshal([]byte(text), &parsed); err == nil && (parsed.Summary != "" || len(parsed.Findings) > 0) {
|
||||
return parsed
|
||||
}
|
||||
return AIAnalysisResult{
|
||||
Summary: strings.TrimSpace(reply),
|
||||
Findings: []AIFinding{},
|
||||
Actions: []string{},
|
||||
Unknowns: []string{},
|
||||
RawText: reply,
|
||||
ParseErr: "模型未按 JSON 格式返回,已按纯文本展示",
|
||||
}
|
||||
}
|
||||
|
||||
// persistAIRun 落审计(同时用于每日 token 统计)。
|
||||
func persistAIRun(run *models.AIAnalysisRun) {
|
||||
db := repository.GetDB()
|
||||
if db == nil {
|
||||
return
|
||||
}
|
||||
newID := db.NextID("aiAnalysisRuns")
|
||||
db.Lock()
|
||||
defer db.Unlock()
|
||||
run.ID = newID
|
||||
db.AIAnalysisRuns = append(db.AIAnalysisRuns, *run)
|
||||
// 只保留最近 500 条,避免审计表无限膨胀(明细可通过日志追溯)
|
||||
if len(db.AIAnalysisRuns) > 500 {
|
||||
db.AIAnalysisRuns = db.AIAnalysisRuns[len(db.AIAnalysisRuns)-500:]
|
||||
}
|
||||
db.RequestSave()
|
||||
}
|
||||
|
||||
// ListAIRuns 分页查询审计记录(按时间倒序)。
|
||||
func ListAIRuns(page, pageSize int) ([]models.AIAnalysisRun, int) {
|
||||
if page <= 0 {
|
||||
page = 1
|
||||
}
|
||||
if pageSize <= 0 || pageSize > 100 {
|
||||
pageSize = 20
|
||||
}
|
||||
db := repository.GetDB()
|
||||
if db == nil {
|
||||
return nil, 0
|
||||
}
|
||||
db.RLock()
|
||||
all := make([]models.AIAnalysisRun, len(db.AIAnalysisRuns))
|
||||
copy(all, db.AIAnalysisRuns)
|
||||
db.RUnlock()
|
||||
|
||||
total := len(all)
|
||||
// 倒序(新的在前)
|
||||
for i, j := 0, len(all)-1; i < j; i, j = i+1, j-1 {
|
||||
all[i], all[j] = all[j], all[i]
|
||||
}
|
||||
start := (page - 1) * pageSize
|
||||
if start >= total {
|
||||
return []models.AIAnalysisRun{}, total
|
||||
}
|
||||
end := start + pageSize
|
||||
if end > total {
|
||||
end = total
|
||||
}
|
||||
return all[start:end], total
|
||||
}
|
||||
|
||||
// GetAIRun 取单条审计记录(用于查看完整 payload)。
|
||||
func GetAIRun(id uint) *models.AIAnalysisRun {
|
||||
db := repository.GetDB()
|
||||
if db == nil {
|
||||
return nil
|
||||
}
|
||||
db.RLock()
|
||||
defer db.RUnlock()
|
||||
for i := range db.AIAnalysisRuns {
|
||||
if db.AIAnalysisRuns[i].ID == id {
|
||||
run := db.AIAnalysisRuns[i]
|
||||
return &run
|
||||
}
|
||||
}
|
||||
return nil
|
||||
}
|
||||
@@ -0,0 +1,185 @@
|
||||
package services
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"lokistack/models"
|
||||
"lokistack/pkg/crypto"
|
||||
"lokistack/repository"
|
||||
)
|
||||
|
||||
const aiConfigSettingKey = "ai_config"
|
||||
|
||||
// 成本相关的默认值(宁可保守:额度小、扫描少,用户觉得不够再调)
|
||||
const (
|
||||
defaultAIMaxTokens = 1500
|
||||
defaultAITimeoutSec = 120
|
||||
defaultAITemperature = 0.1 // 分析场景要稳定复现,不要发挥
|
||||
defaultAIDailyTokenCap = 500000
|
||||
defaultAIMaxScanBytes = 2 << 30 // 2GiB
|
||||
)
|
||||
|
||||
// GetAIConfig 读配置(返回真实 APIKey,仅供内部调用)。
|
||||
func GetAIConfig() models.AIConfig {
|
||||
cfg := models.AIConfig{
|
||||
Provider: "openai-compatible",
|
||||
Temperature: defaultAITemperature,
|
||||
MaxTokens: defaultAIMaxTokens,
|
||||
TimeoutSec: defaultAITimeoutSec,
|
||||
DailyTokenCap: defaultAIDailyTokenCap,
|
||||
MaxScanBytes: defaultAIMaxScanBytes,
|
||||
}
|
||||
db := repository.GetDB()
|
||||
if db == nil {
|
||||
return cfg
|
||||
}
|
||||
db.RLock()
|
||||
raw := ""
|
||||
for _, s := range db.Settings {
|
||||
if s.Key == aiConfigSettingKey {
|
||||
raw = s.Value
|
||||
break
|
||||
}
|
||||
}
|
||||
db.RUnlock()
|
||||
if raw == "" {
|
||||
return cfg
|
||||
}
|
||||
var stored models.AIConfig
|
||||
if err := json.Unmarshal([]byte(raw), &stored); err != nil {
|
||||
return cfg
|
||||
}
|
||||
if stored.APIKey != "" {
|
||||
stored.APIKey = crypto.DecryptField(stored.APIKey)
|
||||
}
|
||||
if stored.Provider == "" {
|
||||
stored.Provider = "openai-compatible"
|
||||
}
|
||||
if stored.MaxTokens <= 0 {
|
||||
stored.MaxTokens = defaultAIMaxTokens
|
||||
}
|
||||
if stored.TimeoutSec <= 0 {
|
||||
stored.TimeoutSec = defaultAITimeoutSec
|
||||
}
|
||||
if stored.MaxScanBytes <= 0 {
|
||||
stored.MaxScanBytes = defaultAIMaxScanBytes
|
||||
}
|
||||
return stored
|
||||
}
|
||||
|
||||
// MaskedAIConfig 返回给前端的版本:APIKey 只留头尾。
|
||||
func MaskedAIConfig() models.AIConfig {
|
||||
cfg := GetAIConfig()
|
||||
cfg.APIKey = maskKey(cfg.APIKey)
|
||||
return cfg
|
||||
}
|
||||
|
||||
func maskKey(key string) string {
|
||||
if key == "" {
|
||||
return ""
|
||||
}
|
||||
if len(key) <= 8 {
|
||||
return "***"
|
||||
}
|
||||
return key[:3] + "***" + key[len(key)-4:]
|
||||
}
|
||||
|
||||
// SaveAIConfig 保存配置。
|
||||
//
|
||||
// apiKey 传空 = 沿用旧值(前端拿到的是掩码,不能把掩码存回去);
|
||||
// 传 "***" 开头的掩码也视为沿用,避免用户只是改模型名却把 key 清空。
|
||||
func SaveAIConfig(in models.AIConfig, apiKey string) error {
|
||||
old := GetAIConfig()
|
||||
key := strings.TrimSpace(apiKey)
|
||||
switch {
|
||||
case key == "":
|
||||
key = old.APIKey
|
||||
case strings.Contains(key, "***"):
|
||||
key = old.APIKey
|
||||
}
|
||||
if in.Enabled && (strings.TrimSpace(in.BaseURL) == "" || strings.TrimSpace(in.Model) == "") {
|
||||
return fmt.Errorf("启用 AI 分析需要填写 Base URL 与模型名")
|
||||
}
|
||||
if in.Enabled && key == "" {
|
||||
return fmt.Errorf("启用 AI 分析需要填写 API Key(内网无鉴权的服务可填任意占位值)")
|
||||
}
|
||||
if in.BaseURL != "" && !strings.HasPrefix(in.BaseURL, "http") {
|
||||
return fmt.Errorf("Base URL 必须以 http:// 或 https:// 开头(例如 https://api.deepseek.com/v1)")
|
||||
}
|
||||
|
||||
enc, err := crypto.EncryptField(key)
|
||||
if err != nil {
|
||||
return fmt.Errorf("加密 API Key 失败: %w", err)
|
||||
}
|
||||
saved := models.AIConfig{
|
||||
Enabled: in.Enabled,
|
||||
Provider: "openai-compatible",
|
||||
BaseURL: strings.TrimSuffix(strings.TrimSpace(in.BaseURL), "/"),
|
||||
APIKey: enc,
|
||||
Model: strings.TrimSpace(in.Model),
|
||||
Temperature: in.Temperature,
|
||||
MaxTokens: in.MaxTokens,
|
||||
TimeoutSec: in.TimeoutSec,
|
||||
DailyTokenCap: in.DailyTokenCap,
|
||||
MaxScanBytes: in.MaxScanBytes,
|
||||
RedactExtra: in.RedactExtra,
|
||||
}
|
||||
if saved.MaxTokens <= 0 {
|
||||
saved.MaxTokens = defaultAIMaxTokens
|
||||
}
|
||||
if saved.TimeoutSec <= 0 {
|
||||
saved.TimeoutSec = defaultAITimeoutSec
|
||||
}
|
||||
if saved.MaxScanBytes <= 0 {
|
||||
saved.MaxScanBytes = defaultAIMaxScanBytes
|
||||
}
|
||||
data, err := json.Marshal(saved)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
db := repository.GetDB()
|
||||
if db == nil {
|
||||
return fmt.Errorf("数据库未初始化")
|
||||
}
|
||||
newID := db.NextID("settings")
|
||||
db.Lock()
|
||||
defer db.Unlock()
|
||||
for i := range db.Settings {
|
||||
if db.Settings[i].Key == aiConfigSettingKey {
|
||||
db.Settings[i].Value = string(data)
|
||||
db.RequestSave()
|
||||
return nil
|
||||
}
|
||||
}
|
||||
db.Settings = append(db.Settings, models.SystemSetting{ID: newID, Key: aiConfigSettingKey, Value: string(data)})
|
||||
db.RequestSave()
|
||||
return nil
|
||||
}
|
||||
|
||||
// truncateToDay 取当天 00:00(本地时区),用于"今日已用 token"统计。
|
||||
func truncateToDay() time.Time {
|
||||
now := time.Now()
|
||||
return time.Date(now.Year(), now.Month(), now.Day(), 0, 0, 0, 0, now.Location())
|
||||
}
|
||||
|
||||
// AITokenUsedToday 某用户今日已消耗的 token 总量(每日配额闸门用)。
|
||||
func AITokenUsedToday(userID uint) int {
|
||||
db := repository.GetDB()
|
||||
if db == nil {
|
||||
return 0
|
||||
}
|
||||
today := truncateToDay()
|
||||
db.RLock()
|
||||
defer db.RUnlock()
|
||||
total := 0
|
||||
for _, r := range db.AIAnalysisRuns {
|
||||
if r.UserID == userID && !r.CreatedAt.Before(today) {
|
||||
total += r.PromptTokens + r.CompletionTokens
|
||||
}
|
||||
}
|
||||
return total
|
||||
}
|
||||
@@ -0,0 +1,128 @@
|
||||
package services
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"context"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"io"
|
||||
"net/http"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"lokistack/models"
|
||||
)
|
||||
|
||||
// OpenAI 兼容 Chat Completions 客户端。
|
||||
//
|
||||
// 只依赖 /chat/completions 这一个端点,所以 DeepSeek、通义、
|
||||
// 内网 vLLM、Ollama(/v1)、各类网关都能直接用,不需要按厂商写适配。
|
||||
|
||||
type aiMessage struct {
|
||||
Role string `json:"role"`
|
||||
Content string `json:"content"`
|
||||
}
|
||||
|
||||
type aiChatRequest struct {
|
||||
Model string `json:"model"`
|
||||
Messages []aiMessage `json:"messages"`
|
||||
Temperature float64 `json:"temperature"`
|
||||
MaxTokens int `json:"max_tokens,omitempty"`
|
||||
Stream bool `json:"stream"`
|
||||
}
|
||||
|
||||
type aiChatResponse struct {
|
||||
Choices []struct {
|
||||
Message struct {
|
||||
Content string `json:"content"`
|
||||
} `json:"message"`
|
||||
FinishReason string `json:"finish_reason"`
|
||||
} `json:"choices"`
|
||||
Usage struct {
|
||||
PromptTokens int `json:"prompt_tokens"`
|
||||
CompletionTokens int `json:"completion_tokens"`
|
||||
TotalTokens int `json:"total_tokens"`
|
||||
} `json:"usage"`
|
||||
Error *struct {
|
||||
Message string `json:"message"`
|
||||
Type string `json:"type"`
|
||||
} `json:"error"`
|
||||
}
|
||||
|
||||
// aiChat 调一次对话补全。
|
||||
//
|
||||
// 错误信息必须具体:把上游返回的 message 原样带出来,不要包装成
|
||||
// "调用失败" —— 常见原因(key 错、模型名不存在、baseURL 路径少了 /v1、
|
||||
// 内网模型没起)都在上游 message 里,藏起来只会让用户去猜。
|
||||
func aiChat(ctx context.Context, cfg models.AIConfig, messages []aiMessage) (string, int, int, error) {
|
||||
if strings.TrimSpace(cfg.BaseURL) == "" || strings.TrimSpace(cfg.Model) == "" {
|
||||
return "", 0, 0, fmt.Errorf("未配置模型:请先在「智能运维 → 模型配置」填写 Base URL 与模型名")
|
||||
}
|
||||
url := strings.TrimSuffix(cfg.BaseURL, "/") + "/chat/completions"
|
||||
body, err := json.Marshal(aiChatRequest{
|
||||
Model: cfg.Model,
|
||||
Messages: messages,
|
||||
Temperature: cfg.Temperature,
|
||||
MaxTokens: cfg.MaxTokens,
|
||||
Stream: false,
|
||||
})
|
||||
if err != nil {
|
||||
return "", 0, 0, err
|
||||
}
|
||||
|
||||
timeout := time.Duration(cfg.TimeoutSec) * time.Second
|
||||
if timeout <= 0 {
|
||||
timeout = 120 * time.Second
|
||||
}
|
||||
reqCtx, cancel := context.WithTimeout(ctx, timeout)
|
||||
defer cancel()
|
||||
|
||||
req, err := http.NewRequestWithContext(reqCtx, http.MethodPost, url, bytes.NewReader(body))
|
||||
if err != nil {
|
||||
return "", 0, 0, fmt.Errorf("构造请求失败: %w", err)
|
||||
}
|
||||
req.Header.Set("Content-Type", "application/json")
|
||||
if cfg.APIKey != "" {
|
||||
req.Header.Set("Authorization", "Bearer "+cfg.APIKey)
|
||||
}
|
||||
|
||||
resp, err := http.DefaultClient.Do(req)
|
||||
if err != nil {
|
||||
if reqCtx.Err() == context.DeadlineExceeded {
|
||||
return "", 0, 0, fmt.Errorf("模型响应超时(%d 秒):可调大「超时」或换更快的模型", cfg.TimeoutSec)
|
||||
}
|
||||
return "", 0, 0, fmt.Errorf("连接模型服务失败: %w", err)
|
||||
}
|
||||
defer resp.Body.Close()
|
||||
|
||||
raw, _ := io.ReadAll(io.LimitReader(resp.Body, 1<<20))
|
||||
if resp.StatusCode >= 400 {
|
||||
var parsed aiChatResponse
|
||||
if json.Unmarshal(raw, &parsed) == nil && parsed.Error != nil && parsed.Error.Message != "" {
|
||||
return "", 0, 0, fmt.Errorf("模型返回 %d:%s", resp.StatusCode, parsed.Error.Message)
|
||||
}
|
||||
return "", 0, 0, fmt.Errorf("模型返回 %d:%s", resp.StatusCode, truncateStr(string(raw), 300))
|
||||
}
|
||||
|
||||
var parsed aiChatResponse
|
||||
if err := json.Unmarshal(raw, &parsed); err != nil {
|
||||
return "", 0, 0, fmt.Errorf("解析模型响应失败(可能不是 OpenAI 兼容接口): %s", truncateStr(string(raw), 200))
|
||||
}
|
||||
if len(parsed.Choices) == 0 {
|
||||
return "", 0, 0, fmt.Errorf("模型没有返回任何内容(finish_reason 为空)")
|
||||
}
|
||||
return parsed.Choices[0].Message.Content, parsed.Usage.PromptTokens, parsed.Usage.CompletionTokens, nil
|
||||
}
|
||||
|
||||
// TestAIConnection 连通性测试:发一句最小请求,验证 baseURL/key/model 三件套。
|
||||
func TestAIConnection(cfg models.AIConfig) (string, error) {
|
||||
cfg.MaxTokens = 32
|
||||
cfg.TimeoutSec = 30
|
||||
reply, _, _, err := aiChat(context.Background(), cfg, []aiMessage{
|
||||
{Role: "user", Content: "只回复两个字:正常"},
|
||||
})
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
return strings.TrimSpace(reply), nil
|
||||
}
|
||||
@@ -0,0 +1,64 @@
|
||||
package services
|
||||
|
||||
import (
|
||||
"regexp"
|
||||
"strings"
|
||||
)
|
||||
|
||||
// 脱敏:日志里经常夹带凭据,送给外部模型前必须处理。
|
||||
//
|
||||
// 原则:
|
||||
// - 宁可多脱一点,也不能漏掉凭据;
|
||||
// - 但**不要脱掉排查必需的信息**:内网 IP、端口、主机名、命名空间、
|
||||
// Pod 名、错误码都要保留,否则模型给不出有用结论。
|
||||
// - 每条规则都替换成带类型的占位符(<IP> / <TOKEN>),
|
||||
// 这样模型仍能看出"这里原本是个 token",语义不丢。
|
||||
|
||||
type redactRule struct {
|
||||
pattern *regexp.Regexp
|
||||
replacement string
|
||||
}
|
||||
|
||||
var builtinRedactRules = []redactRule{
|
||||
// Bearer / JWT —— 一定是凭据
|
||||
{regexp.MustCompile(`(?i)bearer\s+[A-Za-z0-9\-._~+/]{12,}=*`), "Bearer <TOKEN>"},
|
||||
{regexp.MustCompile(`\beyJ[A-Za-z0-9_-]{8,}\.[A-Za-z0-9_-]{8,}\.[A-Za-z0-9_-]{4,}`), "<JWT>"},
|
||||
// key=value 形式的敏感字段
|
||||
{regexp.MustCompile(`(?i)\b(password|passwd|pwd|secret|token|api[_-]?key|access[_-]?key|secret[_-]?key)\b\s*[=:]\s*("?[^\s"',;]{4,}"?)`), "$1=<REDACTED>"},
|
||||
// URL 里的 basic auth:scheme://user:pass@host
|
||||
// 密码里可能含 @(如 P@ssw0rd),所以取到**最后一个 @** 之前的所有内容:
|
||||
// 用贪婪 [^\s/]+ 再回溯到最后一个 @,能正确切出 password 与 host。
|
||||
{regexp.MustCompile(`://([^:/\s]+):([^\s/]+)@`), "://$1:<REDACTED>@"},
|
||||
// AK/SK 形式(云厂商常见 16~40 位字母数字混合,带已知前缀)
|
||||
{regexp.MustCompile(`\b(AKIA|LTAI|AKID)[A-Za-z0-9]{12,}\b`), "<CLOUD_AK>"},
|
||||
// 手机号 / 身份证
|
||||
{regexp.MustCompile(`\b1[3-9]\d{9}\b`), "<PHONE>"},
|
||||
{regexp.MustCompile(`\b\d{17}[\dXx]\b`), "<ID_CARD>"},
|
||||
// 邮箱
|
||||
{regexp.MustCompile(`\b[A-Za-z0-9._%+\-]+@[A-Za-z0-9.\-]+\.[A-Za-z]{2,}\b`), "<EMAIL>"},
|
||||
}
|
||||
|
||||
// RedactText 对文本做脱敏,extra 是用户在设置里加的额外正则。
|
||||
//
|
||||
// 额外正则写错(如括号不闭合)不能让整个分析失败:静默跳过,
|
||||
// 但上层会把这些失效规则一并提示给用户。
|
||||
func RedactText(text string, extra []string) (string, []string) {
|
||||
out := text
|
||||
for _, rule := range builtinRedactRules {
|
||||
out = rule.pattern.ReplaceAllString(out, rule.replacement)
|
||||
}
|
||||
var invalid []string
|
||||
for _, raw := range extra {
|
||||
raw = strings.TrimSpace(raw)
|
||||
if raw == "" {
|
||||
continue
|
||||
}
|
||||
re, err := regexp.Compile(raw)
|
||||
if err != nil {
|
||||
invalid = append(invalid, raw)
|
||||
continue
|
||||
}
|
||||
out = re.ReplaceAllString(out, "<REDACTED>")
|
||||
}
|
||||
return out, invalid
|
||||
}
|
||||
@@ -0,0 +1,87 @@
|
||||
package services
|
||||
|
||||
import (
|
||||
"strings"
|
||||
"testing"
|
||||
)
|
||||
|
||||
// 脱敏的两条底线:
|
||||
// 1) 凭据必须挡住(否则把 key 发给外部模型);
|
||||
// 2) 排查必需的信息必须留着(IP/端口/命名空间/错误码),
|
||||
// 否则模型给不出有用结论 —— 这比"脱得太干净"更常见。
|
||||
func TestRedactText(t *testing.T) {
|
||||
in := `2026-10-30 12:00:01 ERROR redis client: dial tcp 172.19.16.17:6379: connect: connection refused
|
||||
password=Sup3rS3cret! user-admin
|
||||
Authorization: Bearer eyJhbGciOiJIUzI1NiJ9.eyJzdWIiOiIxIn0.abcdefghijklmn
|
||||
url: mysql://appuser:P@ssw0rd@10.2.3.4:3306/cex
|
||||
contact: ops@example.com phone: 13800138000
|
||||
accessKey=AKIAIOSFODNN7EXAMPLE`
|
||||
|
||||
out, invalid := RedactText(in, nil)
|
||||
if len(invalid) != 0 {
|
||||
t.Fatalf("内置规则不应产生无效正则: %v", invalid)
|
||||
}
|
||||
|
||||
mustRedact := []string{"Sup3rS3cret", "eyJhbGciOiJIUzI1NiJ9", "P@ssw0rd", "ops@example.com", "13800138000", "AKIAIOSFODNN7EXAMPLE"}
|
||||
for _, secret := range mustRedact {
|
||||
if strings.Contains(out, secret) {
|
||||
t.Errorf("敏感内容未脱敏: %q\n%s", secret, out)
|
||||
}
|
||||
}
|
||||
|
||||
mustKeep := []string{"172.19.16.17:6379", "connection refused", "10.2.3.4:3306", "redis"}
|
||||
for _, keep := range mustKeep {
|
||||
if !strings.Contains(out, keep) {
|
||||
t.Errorf("排查必需的信息被误删: %q\n%s", keep, out)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 用户自定义正则写错不能拖垮整次分析:跳过并在返回值里报告。
|
||||
func TestRedactTextInvalidExtraRule(t *testing.T) {
|
||||
out, invalid := RedactText("hello internal-id-12345", []string{`(unclosed`, `internal-id-\d+`})
|
||||
if len(invalid) != 1 || invalid[0] != `(unclosed` {
|
||||
t.Errorf("应报告 1 条无效规则,实际 %v", invalid)
|
||||
}
|
||||
if strings.Contains(out, "internal-id-12345") {
|
||||
t.Errorf("有效自定义规则应生效: %s", out)
|
||||
}
|
||||
}
|
||||
|
||||
// 模型经常不按 JSON 返回,解析必须宽容,不能只报"解析失败"。
|
||||
func TestParseAIResultTolerance(t *testing.T) {
|
||||
cases := []struct {
|
||||
name string
|
||||
in string
|
||||
want string // 期望的 summary 片段
|
||||
}{
|
||||
{
|
||||
name: "标准 JSON",
|
||||
in: `{"summary":"日志量正常","findings":[],"actions":[],"unknowns":[]}`,
|
||||
want: "日志量正常",
|
||||
},
|
||||
{
|
||||
name: "包了 ```json 代码块",
|
||||
in: "```json\n{\"summary\":\"体积上涨 2 倍\",\"findings\":[{\"title\":\"t\",\"evidence\":\"e\",\"severity\":\"high\"}],\"actions\":[],\"unknowns\":[]}\n```",
|
||||
want: "体积上涨 2 倍",
|
||||
},
|
||||
{
|
||||
name: "前后加了解释",
|
||||
in: `分析如下:{"summary":"存在告警","findings":[],"actions":[],"unknowns":[]} 希望有帮助`,
|
||||
want: "存在告警",
|
||||
},
|
||||
{
|
||||
name: "纯文本(降级)",
|
||||
in: "看起来一切正常,没有异常日志。",
|
||||
want: "看起来一切正常",
|
||||
},
|
||||
}
|
||||
for _, c := range cases {
|
||||
t.Run(c.name, func(t *testing.T) {
|
||||
got := parseAIResult(c.in)
|
||||
if !strings.Contains(got.Summary, c.want) {
|
||||
t.Errorf("summary 不符: got %q want contains %q", got.Summary, c.want)
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,279 @@
|
||||
package services
|
||||
|
||||
import (
|
||||
"context"
|
||||
"fmt"
|
||||
"os"
|
||||
"sort"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"lokistack/models"
|
||||
"lokistack/repository"
|
||||
)
|
||||
|
||||
// AI 分析的"信号包":把 GB 级日志压成 KB 级结构化数据。
|
||||
//
|
||||
// 成本优先原则(P0 的硬约束):
|
||||
// - 体积/趋势 → index/volume_range(实测 24h 仅 54ms,不读日志内容)
|
||||
// - 条数/流数 → index/stats,只允许在 ≤2 小时窗口上用
|
||||
// (全天 index/stats 实测 20 秒,绝不能进 AI 路径)
|
||||
// - 应用排行 → 平台按小时预聚合的快照(毫秒级)
|
||||
// - 告警 → 平台库
|
||||
//
|
||||
// 也就是说:**P0 完全不扫日志内容**,一次分析对 Loki 的压力接近 0。
|
||||
type AISignalBundle struct {
|
||||
SourceID uint `json:"sourceId"`
|
||||
SourceName string `json:"sourceName"`
|
||||
Selector string `json:"selector"`
|
||||
From string `json:"from"`
|
||||
To string `json:"to"`
|
||||
RangeMinutes int `json:"rangeMinutes"`
|
||||
|
||||
TotalBytes int64 `json:"totalBytes"`
|
||||
TotalEntries int64 `json:"totalEntries"`
|
||||
// EntriesExact 条数是否为精确值(false = 范围过大跳过了 index/stats)
|
||||
EntriesExact bool `json:"entriesExact"`
|
||||
|
||||
Streams int64 `json:"streams"`
|
||||
Chunks int64 `json:"chunks"`
|
||||
|
||||
Trend []AITrendPoint `json:"trend"`
|
||||
Apps []AISignalApp `json:"apps"`
|
||||
|
||||
Baseline *AIBaseline `json:"baseline,omitempty"`
|
||||
Alerts []AIAlert `json:"alerts"`
|
||||
|
||||
// SnapshotHours 平台按小时预聚合已留存的小时数(影响排行的可信度)
|
||||
SnapshotHours int `json:"snapshotHours"`
|
||||
// Warnings 采集过程中的降级说明(如"范围过大,条数为估算")
|
||||
Warnings []string `json:"warnings"`
|
||||
}
|
||||
|
||||
// aiDisplayZone 结论中展示时间用的时区。
|
||||
//
|
||||
// 为什么不直接用 time.Local:静态编译的容器镜像通常没有 /usr/share/zoneinfo,
|
||||
// TZ 也没设,time.Local 就是 UTC —— 结果是模型在结论里写"凌晨 6 点日志下跌",
|
||||
// 而运维当时是下午 2 点,直接被当成数据错误。
|
||||
// 默认东八区(本地运维环境),需要别的时区可用 LOKISTACK_DISPLAY_TZ=UTC 覆盖。
|
||||
func aiDisplayZone() *time.Location {
|
||||
if strings.EqualFold(strings.TrimSpace(os.Getenv("LOKISTACK_DISPLAY_TZ")), "UTC") {
|
||||
return time.UTC
|
||||
}
|
||||
return time.FixedZone("CST", 8*3600)
|
||||
}
|
||||
|
||||
type AITrendPoint struct {
|
||||
Time string `json:"time"`
|
||||
Bytes int64 `json:"bytes"`
|
||||
}
|
||||
|
||||
type AISignalApp struct {
|
||||
App string `json:"app"`
|
||||
Namespace string `json:"namespace"`
|
||||
Entries int64 `json:"entries"`
|
||||
Bytes int64 `json:"bytes"`
|
||||
Share string `json:"share"`
|
||||
}
|
||||
|
||||
type AIBaseline struct {
|
||||
// PreviousBytes 上一周期同时长的体积(等长窗口对比)
|
||||
PreviousBytes int64 `json:"previousBytes"`
|
||||
CurrentBytes int64 `json:"currentBytes"`
|
||||
DeltaPercent float64 `json:"deltaPercent"`
|
||||
ComparisonNote string `json:"comparisonNote"`
|
||||
}
|
||||
|
||||
type AIAlert struct {
|
||||
RuleName string `json:"ruleName"`
|
||||
Status string `json:"status"`
|
||||
Count int `json:"count"`
|
||||
CreatedAt string `json:"createdAt"`
|
||||
}
|
||||
|
||||
// CollectAISignals 采集一次分析所需的全部信号。
|
||||
func CollectAISignals(ctx context.Context, src *models.LogSource, userID uint, admin bool, from, to time.Time) (*AISignalBundle, error) {
|
||||
if src == nil {
|
||||
return nil, fmt.Errorf("数据源不存在")
|
||||
}
|
||||
if !to.After(from) {
|
||||
return nil, fmt.Errorf("时间范围无效")
|
||||
}
|
||||
// 复用查询授权:AI 能看到的数据范围必须与当前用户一致,不多不少
|
||||
selector, err := AuthorizedLogQuery(src, userID, admin, `{namespace=~".+"}`)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
bundle := &AISignalBundle{
|
||||
SourceID: src.ID,
|
||||
SourceName: src.Name,
|
||||
Selector: selector,
|
||||
// 时间统一按本地时区展示:Loki 返回的是 UTC,
|
||||
// 直接给模型 UTC 会让结论里出现"凌晨 6 点"这种和用户作息对不上的时间。
|
||||
From: from.In(aiDisplayZone()).Format("2006-01-02 15:04"),
|
||||
To: to.In(aiDisplayZone()).Format("2006-01-02 15:04"),
|
||||
RangeMinutes: int(to.Sub(from).Minutes()),
|
||||
Alerts: []AIAlert{},
|
||||
Trend: []AITrendPoint{},
|
||||
Apps: []AISignalApp{},
|
||||
}
|
||||
|
||||
// ── 体积 + 趋势:volume_range(只读索引元数据)──
|
||||
buckets, err := LokiHistogram(ctx, src, selector, from.UnixNano(), to.UnixNano())
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
for _, b := range buckets {
|
||||
var bytes int64
|
||||
switch v := b["count"].(type) {
|
||||
case int64:
|
||||
bytes = v
|
||||
case float64:
|
||||
bytes = int64(v)
|
||||
}
|
||||
ts, _ := b["timestamp"].(string)
|
||||
if parsed, perr := time.Parse(time.RFC3339, ts); perr == nil {
|
||||
ts = parsed.In(aiDisplayZone()).Format("15:04")
|
||||
}
|
||||
bundle.TotalBytes += bytes
|
||||
bundle.Trend = append(bundle.Trend, AITrendPoint{Time: ts, Bytes: bytes})
|
||||
}
|
||||
// 趋势只保留尾部若干桶,避免 prompt 过长
|
||||
if len(bundle.Trend) > 24 {
|
||||
bundle.Trend = bundle.Trend[len(bundle.Trend)-24:]
|
||||
}
|
||||
|
||||
// ── 条数:仅短窗口走 index/stats(长窗口太慢,直接跳过并标注)──
|
||||
if bundle.RangeMinutes <= 120 {
|
||||
if st, statErr := LokiStatsCached(ctx, src, selector, from.UnixNano(), to.UnixNano()); statErr == nil {
|
||||
bundle.TotalEntries = st.Entries
|
||||
bundle.Streams = st.Streams
|
||||
bundle.Chunks = st.Chunks
|
||||
bundle.EntriesExact = true
|
||||
} else {
|
||||
bundle.Warnings = append(bundle.Warnings, "条数统计失败:"+truncateStr(statErr.Error(), 120))
|
||||
}
|
||||
} else {
|
||||
bundle.Warnings = append(bundle.Warnings, "时间范围超过 2 小时,条数为估算(跳过 index/stats:全天统计实测需 20 秒)")
|
||||
}
|
||||
|
||||
// ── 应用排行:平台按小时预聚合快照 ──
|
||||
if rows := GetAppRanking(src.ID, from, to, 10, "entries"); len(rows) > 0 {
|
||||
var total int64
|
||||
for _, r := range rows {
|
||||
total += r.Entries
|
||||
}
|
||||
for _, r := range rows {
|
||||
share := ""
|
||||
if total > 0 {
|
||||
share = fmt.Sprintf("%.1f%%", float64(r.Entries)/float64(total)*100)
|
||||
}
|
||||
bundle.Apps = append(bundle.Apps, AISignalApp{
|
||||
App: r.App, Namespace: r.Namespace, Entries: r.Entries, Bytes: r.Bytes, Share: share,
|
||||
})
|
||||
}
|
||||
}
|
||||
hours, _ := AppVolumeCoverageInfo(src.ID)
|
||||
bundle.SnapshotHours = hours
|
||||
if len(bundle.Apps) == 0 {
|
||||
bundle.Warnings = append(bundle.Warnings, "应用排行暂无数据:平台按小时预聚合还在积累(当前留存 0 小时)")
|
||||
}
|
||||
|
||||
// ── 基线对比:与上一个等长窗口比 ──
|
||||
prevFrom := from.Add(-to.Sub(from))
|
||||
prevTo := from
|
||||
if prevBuckets, perr := LokiHistogram(ctx, src, selector, prevFrom.UnixNano(), prevTo.UnixNano()); perr == nil {
|
||||
var prevBytes int64
|
||||
for _, b := range prevBuckets {
|
||||
switch v := b["count"].(type) {
|
||||
case int64:
|
||||
prevBytes += v
|
||||
case float64:
|
||||
prevBytes += int64(v)
|
||||
}
|
||||
}
|
||||
if prevBytes > 0 || bundle.TotalBytes > 0 {
|
||||
delta := 0.0
|
||||
if prevBytes > 0 {
|
||||
delta = (float64(bundle.TotalBytes) - float64(prevBytes)) / float64(prevBytes) * 100
|
||||
}
|
||||
bundle.Baseline = &AIBaseline{
|
||||
PreviousBytes: prevBytes,
|
||||
CurrentBytes: bundle.TotalBytes,
|
||||
DeltaPercent: delta,
|
||||
ComparisonNote: fmt.Sprintf("与上一个 %d 分钟等长窗口对比", bundle.RangeMinutes),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// ── 告警:窗口内的投递记录 ──
|
||||
db := repository.GetDB()
|
||||
if db != nil {
|
||||
db.RLock()
|
||||
for _, d := range db.AlertDeliveries {
|
||||
if d.CreatedAt.Before(from) || d.CreatedAt.After(to) {
|
||||
continue
|
||||
}
|
||||
bundle.Alerts = append(bundle.Alerts, AIAlert{
|
||||
RuleName: d.RuleName, Status: d.Status, Count: d.Count,
|
||||
CreatedAt: d.CreatedAt.In(aiDisplayZone()).Format("2006-01-02 15:04"),
|
||||
})
|
||||
}
|
||||
db.RUnlock()
|
||||
}
|
||||
sort.Slice(bundle.Alerts, func(i, j int) bool { return bundle.Alerts[i].CreatedAt > bundle.Alerts[j].CreatedAt })
|
||||
if len(bundle.Alerts) > 10 {
|
||||
bundle.Alerts = bundle.Alerts[:10]
|
||||
}
|
||||
return bundle, nil
|
||||
}
|
||||
|
||||
// SignalsAsText 把信号包渲染成给模型看的紧凑文本(不是 JSON,省 token)。
|
||||
func SignalsAsText(b *AISignalBundle) string {
|
||||
var sb strings.Builder
|
||||
fmt.Fprintf(&sb, "数据源:%s\n", b.SourceName)
|
||||
fmt.Fprintf(&sb, "时间范围:%s ~ %s(%d 分钟)\n", b.From, b.To, b.RangeMinutes)
|
||||
fmt.Fprintf(&sb, "日志体积:%.2f GB\n", float64(b.TotalBytes)/1024/1024/1024)
|
||||
if b.EntriesExact {
|
||||
fmt.Fprintf(&sb, "日志条数:%d\n", b.TotalEntries)
|
||||
} else {
|
||||
sb.WriteString("日志条数:(范围过大未精确统计)\n")
|
||||
}
|
||||
if b.Streams > 0 || b.Chunks > 0 {
|
||||
fmt.Fprintf(&sb, "流数/块数:%d / %d\n", b.Streams, b.Chunks)
|
||||
}
|
||||
if b.Baseline != nil {
|
||||
fmt.Fprintf(&sb, "基线对比:上一等长窗口 %.2f GB → 当前 %.2f GB,变化 %+.1f%%(%s)\n",
|
||||
float64(b.Baseline.PreviousBytes)/1024/1024/1024,
|
||||
float64(b.Baseline.CurrentBytes)/1024/1024/1024,
|
||||
b.Baseline.DeltaPercent, b.Baseline.ComparisonNote)
|
||||
}
|
||||
if len(b.Apps) > 0 {
|
||||
sb.WriteString("\n应用日志量排行(来自平台小时级预聚合):\n")
|
||||
for i, a := range b.Apps {
|
||||
fmt.Fprintf(&sb, " %d. %s/%s 条数=%d 占比=%s\n", i+1, a.Namespace, a.App, a.Entries, a.Share)
|
||||
}
|
||||
}
|
||||
if len(b.Trend) > 0 {
|
||||
sb.WriteString("\n体积趋势(最近分桶):\n")
|
||||
for _, t := range b.Trend {
|
||||
fmt.Fprintf(&sb, " %s %.1f MB\n", t.Time, float64(t.Bytes)/1024/1024)
|
||||
}
|
||||
}
|
||||
if len(b.Alerts) > 0 {
|
||||
sb.WriteString("\n窗口内告警:\n")
|
||||
for _, a := range b.Alerts {
|
||||
fmt.Fprintf(&sb, " %s %s 命中=%d 状态=%s\n", a.CreatedAt, a.RuleName, a.Count, a.Status)
|
||||
}
|
||||
} else {
|
||||
sb.WriteString("\n窗口内告警:无\n")
|
||||
}
|
||||
if len(b.Warnings) > 0 {
|
||||
sb.WriteString("\n数据限制说明:\n")
|
||||
for _, w := range b.Warnings {
|
||||
fmt.Fprintf(&sb, " - %s\n", w)
|
||||
}
|
||||
}
|
||||
return sb.String()
|
||||
}
|
||||
Reference in new issue
Block a user