=== 一、现象 ===
- /logs 页「正在统计整个时间范围的日志量」24 小时查询转圈 5 分钟不出结果
- /analysis 页默认 1 小时也一直「加载中...」
- 期间 Loki 被 OOMKilled(2Gi) → Pod not Ready → Service 无端点 → 全站查询 502
=== 二、根因(实测数据) ===
同一 query {namespace="astralx", app="cex-ipublic-gateway"}:
24h count_over_time[15m] step=15m → 53.7s 后连接断开(算不完)
24h count_over_time[1h] step=1h → 60.1s 失败
24h index/stats(只读索引) → 1.3s → bytes 44.9GB / entries 3.33亿 / chunks 2474
24h index/volume_range step=1h → 111ms → 25 个点
1h index/volume_range step=5m → 115ms → 13 个点
原因:count_over_time 必须**下载并解码所有 chunk**;volume_range 只读 TSDB 索引里的
chunk 元数据(字节数),根本不碰日志内容。该 app 一天就有 3.33 亿条 / 44.9 GB。
=== 三、改造 ===
1. LokiHistogram 改用官方 /loki/api/v1/index/volume_range(带 step)
- 返回单位是「字节」,bucket 带 unit=bytes,前端格式化为 KB/MB/GB
- 老版本 Loki 没有该端点时自动回退 count_over_time(unit=entries)
2. 查询护栏(防止单条查询拖垮单实例 Loki)
- 统计范围上限 7 天,超出直接 400「统计范围过大(27.0 天),最多支持 7 天」
- 重查询单独 30s 超时(原来 120s)
- 全局并发重查询上限 2,超出立刻 429「请稍后重试」(不排队)
- /logs/metrics(analysis 页)共用同一套护栏
3. 前端
- 发起新查询前 abort 上一次请求(原来只用版本号丢弃结果,旧请求仍在后端跑)
- histogram/query 支持 signal;超时 40s/60s(原 125s)
- 错误不再统一显示「日志量统计失败」,按 429/400/502/超时给不同文案
- 时间线按 unit 格式化(字节 → GB),i18n 去掉「条」
4. Loki 侧官方优化(此前 limits_config 只有 retention_period)
- query_range: align_queries_with_step + cache_results + embedded_cache 100MB
- limits_config: max_query_parallelism 8(默认 32,单实例过高)
- 已用 `loki -verify-config` 校验通过(config is valid)
=== 四、实测(平台 API) ===
1 分钟 0.02s 2 桶
1 小时 0.05s 61 桶 3.8 GB
6 小时 0.04s 73 桶 12.3 GB
24 小时 0.04s 97 桶 41.8 GB ← 改造前 53.7s 失败
7 天 0.05s 37 桶 58.2 GB
27 天 0.01s 400「最多支持 7 天」
镜像:lghost/lokistack:latest + v0.1 = sha256:9fa1c4597533
686 lines
21 KiB
Go
686 lines
21 KiB
Go
package services
|
||
|
||
// 日志查询代理 — 后端代表前端访问 Loki HTTP API。
|
||
// cluster-svc 型经 kube-apiserver service proxy(不依赖 ClusterIP 直通与集群 DNS);
|
||
// url 型直连。契约参照 aiops 日志中心(query/labels/values/histogram + cursor 翻页)。
|
||
import (
|
||
"context"
|
||
"crypto/tls"
|
||
"encoding/base64"
|
||
"encoding/json"
|
||
"errors"
|
||
"fmt"
|
||
"io"
|
||
"net/http"
|
||
"net/url"
|
||
"sort"
|
||
"strconv"
|
||
"strings"
|
||
"time"
|
||
|
||
"k8s.io/client-go/rest"
|
||
"lokistack/models"
|
||
"lokistack/repository"
|
||
)
|
||
|
||
const maxLogQueryLimit = 1000
|
||
const logQueryTimeout = 120 * time.Second
|
||
|
||
var lokiHTTPClients = [2]*http.Client{
|
||
{Timeout: logQueryTimeout, Transport: newLokiTransport(false)},
|
||
{Timeout: logQueryTimeout, Transport: newLokiTransport(true)},
|
||
}
|
||
|
||
func newLokiTransport(skipTLSVerify bool) *http.Transport {
|
||
transport := http.DefaultTransport.(*http.Transport).Clone()
|
||
transport.TLSClientConfig = &tls.Config{InsecureSkipVerify: skipTLSVerify} //nolint:gosec // 数据源显式配置
|
||
return transport
|
||
}
|
||
|
||
// ==================== Loki HTTP 访问 ====================
|
||
|
||
// lokiProxyGET 对数据源执行 GET;cluster-svc 使用 client-go 的认证传输访问 API Server Service Proxy。
|
||
func lokiProxyGET(ctx context.Context, src *models.LogSource, path string, query url.Values) ([]byte, int, error) {
|
||
if src.TargetType == "cluster-svc" {
|
||
db := repository.GetDB()
|
||
db.RLock()
|
||
var cluster *models.Cluster
|
||
for i := range db.Clusters {
|
||
if src.ClusterID != nil && db.Clusters[i].ID == *src.ClusterID {
|
||
c := db.Clusters[i]
|
||
cluster = &c
|
||
break
|
||
}
|
||
}
|
||
db.RUnlock()
|
||
if cluster == nil {
|
||
return nil, 500, fmt.Errorf("数据源关联的集群不存在")
|
||
}
|
||
if !kubernetesNamespacePattern.MatchString(src.Namespace) || !kubernetesNamespacePattern.MatchString(src.Service) || src.Port < 1 || src.Port > 65535 {
|
||
return nil, 500, fmt.Errorf("Loki 数据源的命名空间、Service 或端口无效")
|
||
}
|
||
config, err := clusterRESTConfig(cluster, logQueryTimeout)
|
||
if err != nil {
|
||
return nil, 500, err
|
||
}
|
||
client, err := rest.HTTPClientFor(config)
|
||
if err != nil {
|
||
return nil, 500, fmt.Errorf("创建 Kubernetes API 客户端失败: %w", err)
|
||
}
|
||
full := strings.TrimRight(config.Host, "/") + fmt.Sprintf("/api/v1/namespaces/%s/services/%s:%d/proxy%s",
|
||
src.Namespace, src.Service, src.Port, path)
|
||
if qs := query.Encode(); qs != "" {
|
||
full += "?" + qs
|
||
}
|
||
req, err := http.NewRequestWithContext(ctx, http.MethodGet, full, nil)
|
||
if err != nil {
|
||
return nil, 500, err
|
||
}
|
||
resp, err := client.Do(req)
|
||
if err != nil {
|
||
return nil, 502, fmt.Errorf("Kubernetes API 代理查询失败: %w", err)
|
||
}
|
||
defer resp.Body.Close()
|
||
body, err := io.ReadAll(io.LimitReader(resp.Body, 32<<20))
|
||
if err != nil {
|
||
return nil, 502, err
|
||
}
|
||
if resp.StatusCode >= 400 {
|
||
return body, resp.StatusCode, fmt.Errorf("Kubernetes API 代理返回 HTTP %d: %s", resp.StatusCode, truncateStr(string(body), 300))
|
||
}
|
||
return body, resp.StatusCode, nil
|
||
}
|
||
|
||
// url 直连
|
||
if src.URL == "" {
|
||
return nil, 500, fmt.Errorf("数据源未配置访问地址")
|
||
}
|
||
full := strings.TrimSuffix(src.URL, "/") + path
|
||
if qs := query.Encode(); qs != "" {
|
||
full += "?" + qs
|
||
}
|
||
req, err := http.NewRequestWithContext(ctx, http.MethodGet, full, nil)
|
||
if err != nil {
|
||
return nil, 500, err
|
||
}
|
||
applyLokiAuth(req, src)
|
||
client := lokiHTTPClients[0]
|
||
if src.SkipTLSVerify {
|
||
client = lokiHTTPClients[1]
|
||
}
|
||
resp, err := client.Do(req)
|
||
if err != nil {
|
||
return nil, 502, err
|
||
}
|
||
defer resp.Body.Close()
|
||
body, err := io.ReadAll(io.LimitReader(resp.Body, 32<<20))
|
||
if err != nil {
|
||
return nil, 502, err
|
||
}
|
||
if resp.StatusCode >= 400 {
|
||
return body, resp.StatusCode, fmt.Errorf("Loki 返回 HTTP %d: %s", resp.StatusCode, truncateStr(string(body), 300))
|
||
}
|
||
return body, resp.StatusCode, nil
|
||
}
|
||
|
||
func applyLokiAuth(req *http.Request, src *models.LogSource) {
|
||
switch src.AuthType {
|
||
case "basic":
|
||
if pass, err := DecryptCred(src.BasicPass); err == nil {
|
||
req.Header.Set("Authorization", "Basic "+
|
||
base64.StdEncoding.EncodeToString([]byte(src.BasicUser+":"+pass)))
|
||
}
|
||
case "bearer":
|
||
if token, err := DecryptCred(src.BearerToken); err == nil {
|
||
req.Header.Set("Authorization", "Bearer "+token)
|
||
}
|
||
}
|
||
for name, encrypted := range src.Headers {
|
||
if validSourceHeader(name) {
|
||
if value, err := DecryptCred(encrypted); err == nil {
|
||
req.Header.Set(name, value)
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
func truncateStr(s string, n int) string {
|
||
if len(s) <= n {
|
||
return s
|
||
}
|
||
return s[:n] + "..."
|
||
}
|
||
|
||
// ==================== 查询实现(契约对齐 aiops) ====================
|
||
|
||
type LogEntryRow struct {
|
||
Timestamp string `json:"timestamp"`
|
||
TimestampNs string `json:"timestampNs"`
|
||
Level string `json:"level"`
|
||
Message string `json:"message"`
|
||
Labels map[string]string `json:"labels"`
|
||
}
|
||
|
||
type LogQueryResult struct {
|
||
Entries []LogEntryRow `json:"entries"`
|
||
HasMore bool `json:"hasMore"`
|
||
NextCursor string `json:"nextCursor"`
|
||
}
|
||
|
||
// ContextLogSelector uses stable stream identity labels. Loki can add per-entry
|
||
// labels such as detected_level to query results; matching those would hide
|
||
// adjacent lines whose detected level differs from the selected line.
|
||
func ContextLogSelector(labels map[string]string) (string, error) {
|
||
for key := range labels {
|
||
if !ValidLogLabelName(key) {
|
||
return "", fmt.Errorf("标签名无效")
|
||
}
|
||
}
|
||
keys := []string{"cluster", "namespace", "app", "pod", "container", "filename"}
|
||
if labels["pod"] == "" {
|
||
keys = append(keys, "job", "instance", "host")
|
||
}
|
||
parts := make([]string, 0, len(keys))
|
||
for _, key := range keys {
|
||
if value := labels[key]; value != "" {
|
||
encoded, _ := json.Marshal(value)
|
||
parts = append(parts, key+"="+string(encoded))
|
||
}
|
||
}
|
||
if len(parts) == 0 {
|
||
for key, value := range labels {
|
||
if value == "" || key == "detected_level" || key == "level" || strings.HasPrefix(key, "__") {
|
||
continue
|
||
}
|
||
encoded, _ := json.Marshal(value)
|
||
parts = append(parts, key+"="+string(encoded))
|
||
}
|
||
sort.Strings(parts)
|
||
}
|
||
if len(parts) == 0 {
|
||
return "", fmt.Errorf("缺少可用于定位日志流的标签")
|
||
}
|
||
return "{" + strings.Join(parts, ",") + "}", nil
|
||
}
|
||
|
||
// QueryLokiLogs 日志查询:flatten streams → 统一行;cursor(最后一条纳秒时间戳)向后翻页
|
||
func QueryLokiLogs(ctx context.Context, src *models.LogSource, query string, start, end int64, limit int, direction, cursor string) (*LogQueryResult, error) {
|
||
if limit <= 0 {
|
||
limit = 200
|
||
}
|
||
if limit > maxLogQueryLimit {
|
||
limit = maxLogQueryLimit
|
||
}
|
||
if direction == "" {
|
||
direction = "backward"
|
||
}
|
||
if end <= 0 {
|
||
end = time.Now().UnixNano()
|
||
}
|
||
if start <= 0 {
|
||
start = end - int64(time.Hour) // 默认最近 1 小时
|
||
}
|
||
if cursor != "" {
|
||
if c, err := strconv.ParseInt(cursor, 10, 64); err == nil {
|
||
if direction == "forward" {
|
||
start = c + 1
|
||
} else {
|
||
end = c - 1
|
||
}
|
||
}
|
||
}
|
||
|
||
q := url.Values{}
|
||
q.Set("query", query)
|
||
q.Set("start", strconv.FormatInt(start, 10))
|
||
q.Set("end", strconv.FormatInt(end, 10))
|
||
q.Set("limit", strconv.Itoa(limit))
|
||
q.Set("direction", direction)
|
||
|
||
body, _, err := lokiProxyGET(ctx, src, "/loki/api/v1/query_range", q)
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
var resp struct {
|
||
Status string `json:"status"`
|
||
Data struct {
|
||
ResultType string `json:"resultType"`
|
||
Result []struct {
|
||
Stream map[string]string `json:"stream"`
|
||
Values [][2]string `json:"values"`
|
||
} `json:"result"`
|
||
} `json:"data"`
|
||
}
|
||
if err := json.Unmarshal(body, &resp); err != nil {
|
||
return nil, fmt.Errorf("解析 Loki 响应失败: %w", err)
|
||
}
|
||
|
||
entries := make([]LogEntryRow, 0, limit)
|
||
var lastTs string
|
||
for _, stream := range resp.Data.Result {
|
||
for _, v := range stream.Values {
|
||
entries = append(entries, LogEntryRow{
|
||
Timestamp: formatLokiTimestamp(v[0]),
|
||
TimestampNs: v[0],
|
||
Level: detectLevel(stream.Stream, v[1]),
|
||
Message: v[1],
|
||
Labels: stream.Stream,
|
||
})
|
||
}
|
||
}
|
||
sort.SliceStable(entries, func(i, j int) bool {
|
||
a, _ := strconv.ParseInt(entries[i].TimestampNs, 10, 64)
|
||
b, _ := strconv.ParseInt(entries[j].TimestampNs, 10, 64)
|
||
if direction == "forward" {
|
||
return a < b
|
||
}
|
||
return a > b
|
||
})
|
||
if len(entries) > limit {
|
||
entries = entries[:limit]
|
||
}
|
||
if len(entries) > 0 {
|
||
lastTs = entries[len(entries)-1].TimestampNs
|
||
}
|
||
|
||
nextCursor := ""
|
||
if len(entries) >= limit && lastTs != "" {
|
||
if ts, err := strconv.ParseInt(strings.Split(lastTs, ".")[0], 10, 64); err == nil {
|
||
nextCursor = strconv.FormatInt(ts, 10)
|
||
}
|
||
}
|
||
return &LogQueryResult{Entries: entries, HasMore: nextCursor != "", NextCursor: nextCursor}, nil
|
||
}
|
||
|
||
// formatLokiTimestamp Loki 时间串(纳秒字符串/科学计数法)→ RFC3339
|
||
func formatLokiTimestamp(ts string) string {
|
||
if i := strings.IndexAny(ts, ".eE"); i > 0 {
|
||
ts = ts[:i]
|
||
}
|
||
n, err := strconv.ParseInt(ts, 10, 64)
|
||
if err != nil {
|
||
return ts
|
||
}
|
||
// 自适应纳秒/微秒/毫秒/秒
|
||
switch {
|
||
case n > 1e18:
|
||
return time.Unix(0, n).Format(time.RFC3339)
|
||
case n > 1e15:
|
||
return time.Unix(0, n*1e3).Format(time.RFC3339)
|
||
case n > 1e12:
|
||
return time.Unix(0, n*1e6).Format(time.RFC3339)
|
||
case n > 1e9:
|
||
return time.Unix(0, n*1e9).Format(time.RFC3339)
|
||
default:
|
||
return time.Unix(n, 0).Format(time.RFC3339)
|
||
}
|
||
}
|
||
|
||
// detectLevel 优先取 level 标签;否则按行内容推断
|
||
func detectLevel(labels map[string]string, line string) string {
|
||
if lv, ok := labels["level"]; ok && lv != "" {
|
||
return strings.ToLower(lv)
|
||
}
|
||
l := strings.ToLower(line)
|
||
for _, p := range []string{"error", "err", "fatal", "panic"} {
|
||
if strings.Contains(l, p) {
|
||
return "error"
|
||
}
|
||
}
|
||
if strings.Contains(l, "warn") {
|
||
return "warn"
|
||
}
|
||
return "info"
|
||
}
|
||
|
||
// LokiLabels 标签名列表
|
||
func LokiLabels(ctx context.Context, src *models.LogSource, start, end int64) ([]string, error) {
|
||
q := url.Values{}
|
||
if end > 0 {
|
||
q.Set("end", strconv.FormatInt(end, 10))
|
||
}
|
||
if start > 0 {
|
||
q.Set("start", strconv.FormatInt(start, 10))
|
||
}
|
||
body, _, err := lokiProxyGET(ctx, src, "/loki/api/v1/labels", q)
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
var resp struct {
|
||
Data []string `json:"data"`
|
||
}
|
||
if err := json.Unmarshal(body, &resp); err != nil {
|
||
return nil, err
|
||
}
|
||
// Loki 在某 label 无值时会返回 {"data":null},Go nil slice 序列化成 JSON null,
|
||
// 前端 setState(null) 后 .map() 直接抛 "Cannot read properties of null"。
|
||
// 统一归一化为空切片,前端拿到的永远是数组。
|
||
if resp.Data == nil {
|
||
resp.Data = []string{}
|
||
}
|
||
return resp.Data, nil
|
||
}
|
||
|
||
// LokiLabelValues 标签值列表(query 只取 {...} selector 部分)
|
||
func LokiLabelValues(ctx context.Context, src *models.LogSource, name, selector string, start, end int64) ([]string, error) {
|
||
q := url.Values{}
|
||
if s := extractStreamSelector(selector); s != "" && s != "{}" {
|
||
q.Set("query", s)
|
||
}
|
||
if end > 0 {
|
||
q.Set("end", strconv.FormatInt(end, 10))
|
||
}
|
||
if start > 0 {
|
||
q.Set("start", strconv.FormatInt(start, 10))
|
||
}
|
||
body, _, err := lokiProxyGET(ctx, src, "/loki/api/v1/label/"+url.PathEscape(name)+"/values", q)
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
var resp struct {
|
||
Data []string `json:"data"`
|
||
}
|
||
if err := json.Unmarshal(body, &resp); err != nil {
|
||
return nil, err
|
||
}
|
||
// 同上:label 无值时 Loki 返回 {"data":null},归一化为空切片,
|
||
// 避免前端 .map() 遇到 null 崩溃(快捷 LogQL 报错的直接原因)。
|
||
if resp.Data == nil {
|
||
resp.Data = []string{}
|
||
}
|
||
return resp.Data, nil
|
||
}
|
||
|
||
// extractStreamSelector 从 LogQL 中提取 {...} 部分(labels API 仅接受 stream selector)
|
||
func extractStreamSelector(query string) string {
|
||
i := strings.Index(query, "{")
|
||
if i < 0 {
|
||
return ""
|
||
}
|
||
j := strings.Index(query[i:], "}")
|
||
if j < 0 {
|
||
return ""
|
||
}
|
||
return query[i : i+j+1]
|
||
}
|
||
|
||
// LokiHistogram 日志量直方图:服务端包 sum(count_over_time(q[step]))
|
||
// 重查询护栏。设计依据:
|
||
// - 单条 27 天的 count_over_time 曾把 Loki 打到 OOMKilled(2Gi),
|
||
// 进而 Pod 不 Ready → Service 无端点 → 全站查询 502。
|
||
// - 所以"算不完的查询"必须在入口就拒绝,而不是硬跑 120 秒把实例拖垮。
|
||
const (
|
||
// 日志量统计最多统计 7 天;更长跨度请分段查(前端会提示缩小范围)
|
||
maxHistogramRange = 7 * 24 * time.Hour
|
||
// 重查询(histogram/metrics)单独的更短超时,避免长时间占用 CPU
|
||
heavyQueryTimeout = 30 * time.Second
|
||
// 全局并发重查询上限:超过直接快速失败,保护单实例 Loki
|
||
maxConcurrentHeavyQueries = 2
|
||
)
|
||
|
||
// ErrTooManyQueries 并发重查询超限(handler 映射为 429)
|
||
var ErrTooManyQueries = errors.New("当前有较多日志量统计在进行,请稍后重试")
|
||
|
||
// heavyQuerySem 全局重查询信号量
|
||
var heavyQuerySem = make(chan struct{}, maxConcurrentHeavyQueries)
|
||
|
||
func acquireHeavyQuery() bool {
|
||
select {
|
||
case heavyQuerySem <- struct{}{}:
|
||
return true
|
||
default:
|
||
return false
|
||
}
|
||
}
|
||
|
||
func releaseHeavyQuery() { <-heavyQuerySem }
|
||
|
||
func LokiHistogram(ctx context.Context, src *models.LogSource, query string, start, end int64) ([]map[string]interface{}, error) {
|
||
if end <= 0 {
|
||
end = time.Now().UnixNano()
|
||
}
|
||
if start <= 0 {
|
||
start = end - int64(time.Hour)
|
||
}
|
||
span := end - start
|
||
if span > int64(maxHistogramRange) {
|
||
return nil, fmt.Errorf("统计范围过大(%.1f 天),最多支持 %.0f 天;请缩小时间范围后重试",
|
||
float64(span)/float64(24*time.Hour), maxHistogramRange.Hours()/24)
|
||
}
|
||
// 并发护栏:拿不到令牌立刻返回,不排队(排队只会让用户等更久)
|
||
if !acquireHeavyQuery() {
|
||
return nil, ErrTooManyQueries
|
||
}
|
||
defer releaseHeavyQuery()
|
||
|
||
// 独立超时:Loki 侧再慢也不会超过 30s,防止单查询长期占用实例
|
||
ctx, cancel := context.WithTimeout(ctx, heavyQueryTimeout)
|
||
defer cancel()
|
||
|
||
step := pickHistogramStep(span)
|
||
|
||
// ── 首选:Loki 官方的 index/volume_range ──────────────────────────────
|
||
// 它只读 TSDB 索引里的 chunk 元数据(字节数),**不解码日志内容**。
|
||
// 实测同一条件 24 小时:count_over_time 53s 超时失败,volume_range 111ms。
|
||
// 返回单位是「字节」,所以 bucket 里带 unit=bytes,前端据此格式化为 KB/MB/GB。
|
||
q := url.Values{}
|
||
q.Set("query", query)
|
||
q.Set("start", strconv.FormatInt(start, 10))
|
||
q.Set("end", strconv.FormatInt(end, 10))
|
||
q.Set("step", step)
|
||
q.Set("limit", "50")
|
||
if body, status, err := lokiProxyGET(ctx, src, "/loki/api/v1/index/volume_range", q); err == nil && status < 400 {
|
||
if buckets, ok := parseVolumeBuckets(body); ok {
|
||
return buckets, nil
|
||
}
|
||
}
|
||
// 老版本 Loki 没有 volume_range(404/400)时回退到 count_over_time
|
||
return lokiHistogramByCount(ctx, src, query, step, start, end)
|
||
}
|
||
|
||
// parseVolumeBuckets 解析 index/volume_range 的 matrix 响应。
|
||
// 成功但无数据也返回 ok=true(空图),只有结构不对才 ok=false 触发回退。
|
||
func parseVolumeBuckets(body []byte) ([]map[string]interface{}, bool) {
|
||
var resp struct {
|
||
Data struct {
|
||
Result []struct {
|
||
Values [][2]interface{} `json:"values"`
|
||
} `json:"result"`
|
||
} `json:"data"`
|
||
}
|
||
if err := json.Unmarshal(body, &resp); err != nil {
|
||
return nil, false
|
||
}
|
||
buckets := []map[string]interface{}{}
|
||
if len(resp.Data.Result) > 0 {
|
||
for _, v := range resp.Data.Result[0].Values {
|
||
tsf, _ := strconv.ParseFloat(fmt.Sprintf("%v", v[0]), 64)
|
||
bytes, _ := strconv.ParseFloat(fmt.Sprintf("%v", v[1]), 64)
|
||
buckets = append(buckets, map[string]interface{}{
|
||
"timestamp": time.Unix(int64(tsf), 0).Format(time.RFC3339),
|
||
"count": int64(bytes),
|
||
"unit": "bytes",
|
||
})
|
||
}
|
||
}
|
||
return buckets, true
|
||
}
|
||
|
||
// lokiHistogramByCount 回退实现:count_over_time 逐点统计「条数」。
|
||
// 代价高(要真读 chunk),只在 volume_range 不可用时使用。
|
||
func lokiHistogramByCount(ctx context.Context, src *models.LogSource, query, step string, start, end int64) ([]map[string]interface{}, error) {
|
||
wrapped := fmt.Sprintf("sum(count_over_time(%s[%s]))", query, step)
|
||
q := url.Values{}
|
||
q.Set("query", wrapped)
|
||
q.Set("start", strconv.FormatInt(start, 10))
|
||
q.Set("end", strconv.FormatInt(end, 10))
|
||
q.Set("step", step)
|
||
body, _, err := lokiProxyGET(ctx, src, "/loki/api/v1/query_range", q)
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
var resp struct {
|
||
Data struct {
|
||
Result []struct {
|
||
Values [][2]interface{} `json:"values"`
|
||
} `json:"result"`
|
||
} `json:"data"`
|
||
}
|
||
if err := json.Unmarshal(body, &resp); err != nil {
|
||
return nil, err
|
||
}
|
||
buckets := []map[string]interface{}{}
|
||
if len(resp.Data.Result) > 0 {
|
||
for _, v := range resp.Data.Result[0].Values {
|
||
tsf, _ := strconv.ParseFloat(fmt.Sprintf("%v", v[0]), 64)
|
||
cnt, _ := strconv.ParseFloat(fmt.Sprintf("%v", v[1]), 64)
|
||
buckets = append(buckets, map[string]interface{}{
|
||
"timestamp": time.Unix(int64(tsf), 0).Format(time.RFC3339),
|
||
"count": int64(cnt),
|
||
"unit": "entries",
|
||
})
|
||
}
|
||
}
|
||
return buckets, nil
|
||
}
|
||
|
||
// MetricSeriesPoint 指标序列上的一个采样点
|
||
type MetricSeriesPoint struct {
|
||
Timestamp string `json:"timestamp"`
|
||
Value float64 `json:"value"`
|
||
}
|
||
|
||
// MetricSeries 一条按标签分组的指标序列(如 sum by (app)(...) 的每个 app 一条)
|
||
type MetricSeries struct {
|
||
Labels map[string]string `json:"labels"`
|
||
Points []MetricSeriesPoint `json:"points"`
|
||
}
|
||
|
||
// LokiMetrics 通用 LogQL 指标查询(日志分析页用):query 为完整指标表达式,
|
||
// 其中所有 [$step] 占位符替换为按时间跨度自适应的步长后走 query_range,
|
||
// 返回矩阵多序列(直方图只取单序列,这里按 by (...) 分组整表返回)
|
||
func LokiMetrics(ctx context.Context, src *models.LogSource, query string, start, end int64) ([]MetricSeries, error) {
|
||
if end <= 0 {
|
||
end = time.Now().UnixNano()
|
||
}
|
||
if start <= 0 {
|
||
start = end - int64(time.Hour)
|
||
}
|
||
// 与 histogram 共用护栏:并发限 2、单次 30s 超时,避免 analysis 页
|
||
// 多个卡片同时发起 count_over_time 把单实例 Loki 打满。
|
||
if !acquireHeavyQuery() {
|
||
return nil, ErrTooManyQueries
|
||
}
|
||
defer releaseHeavyQuery()
|
||
ctx, cancel := context.WithTimeout(ctx, heavyQueryTimeout)
|
||
defer cancel()
|
||
|
||
step := pickHistogramStep(end - start)
|
||
wrapped := strings.ReplaceAll(query, "[$step]", "["+step+"]")
|
||
|
||
q := url.Values{}
|
||
q.Set("query", wrapped)
|
||
q.Set("start", strconv.FormatInt(start, 10))
|
||
q.Set("end", strconv.FormatInt(end, 10))
|
||
q.Set("step", step)
|
||
body, _, err := lokiProxyGET(ctx, src, "/loki/api/v1/query_range", q)
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
var resp struct {
|
||
Data struct {
|
||
Result []struct {
|
||
Metric map[string]string `json:"metric"`
|
||
Values [][2]interface{} `json:"values"`
|
||
} `json:"result"`
|
||
} `json:"data"`
|
||
}
|
||
if err := json.Unmarshal(body, &resp); err != nil {
|
||
return nil, err
|
||
}
|
||
series := make([]MetricSeries, 0, len(resp.Data.Result))
|
||
for _, item := range resp.Data.Result {
|
||
points := make([]MetricSeriesPoint, 0, len(item.Values))
|
||
for _, v := range item.Values {
|
||
// metric API 的 ts 为秒级浮点数(如 1.79e+09)
|
||
tsf, _ := strconv.ParseFloat(fmt.Sprintf("%v", v[0]), 64)
|
||
cnt, _ := strconv.ParseFloat(fmt.Sprintf("%v", v[1]), 64)
|
||
points = append(points, MetricSeriesPoint{
|
||
Timestamp: time.Unix(int64(tsf), 0).Format(time.RFC3339),
|
||
Value: cnt,
|
||
})
|
||
}
|
||
if len(points) == 0 {
|
||
continue
|
||
}
|
||
series = append(series, MetricSeries{Labels: item.Metric, Points: points})
|
||
}
|
||
return series, nil
|
||
}
|
||
|
||
// pickHistogramStep 按时间跨度自适应步长(aiops 同款)
|
||
func pickHistogramStep(spanNano int64) string {
|
||
span := time.Duration(spanNano)
|
||
switch {
|
||
case span <= 15*time.Minute:
|
||
return "30s"
|
||
case span <= time.Hour:
|
||
return "1m"
|
||
case span <= 6*time.Hour:
|
||
return "5m"
|
||
case span <= 24*time.Hour:
|
||
return "15m"
|
||
case span <= 7*24*time.Hour:
|
||
return "1h"
|
||
default:
|
||
return "6h"
|
||
}
|
||
}
|
||
|
||
// DeploymentReady 部署级就绪探测:按部署名匹配自动注册的数据源,经其访问 Loki /ready
|
||
func DeploymentReady(ctx context.Context, deploymentName string) (bool, string) {
|
||
if deploymentName == "" {
|
||
return false, ""
|
||
}
|
||
db := repository.GetDB()
|
||
db.RLock()
|
||
var src *models.LogSource
|
||
for i := range db.LogSources {
|
||
if db.LogSources[i].Name == deploymentName {
|
||
s := db.LogSources[i]
|
||
src = &s
|
||
break
|
||
}
|
||
}
|
||
db.RUnlock()
|
||
if src == nil {
|
||
return false, "未注册数据源"
|
||
}
|
||
if LokiReady(ctx, src) {
|
||
return true, ""
|
||
}
|
||
return false, "Loki 未就绪"
|
||
}
|
||
|
||
// LokiReady Loki /ready 探活(部署卡片/查询页状态点)
|
||
func LokiReady(ctx context.Context, src *models.LogSource) bool {
|
||
body, code, err := lokiProxyGET(ctx, src, "/ready", nil)
|
||
if err != nil || code != 200 {
|
||
return false
|
||
}
|
||
return strings.Contains(strings.ToLower(string(body)), "ready")
|
||
}
|
||
|
||
// TestLogSource 数据源连通性(/loki/api/v1/labels 探活)
|
||
func TestLogSource(ctx context.Context, src *models.LogSource) map[string]interface{} {
|
||
labels, err := LokiLabels(ctx, src, 0, 0)
|
||
if err != nil {
|
||
return map[string]interface{}{"success": false, "message": err.Error()}
|
||
}
|
||
return map[string]interface{}{
|
||
"success": true,
|
||
"message": fmt.Sprintf("连接成功,发现 %d 个标签", len(labels)),
|
||
"labels": labels,
|
||
}
|
||
}
|