opsx
3191af8b6a
=== 一、问题 ===
查看日志上下文时，前后各取几条是**写死**在控制器里的：
span := 5 * time.Minute
QueryLokiLogs(..., ts-span, ts-1, 5, "backward") // 5 条
QueryLokiLogs(..., ts+1, ts+span, 5, "forward") // 5 条
三处写死：条数 5、时间窗 5 分钟、且用户无法调整。
写死的三个实际弊端：
1. 排障时想看更多（比如前后 20 条找调用链）做不到；
2. 只想看紧邻 2 条时，仍然要扫 5 分钟窗口、返回 5 条，浪费；
3. 时间窗与条数不匹配——某服务日志稀疏时，5 分钟窗口里可能一条都没有，
于是"前后各 5 条"实际返回 0 条。
=== 二、改动 ===
1. backend/dto: LogContextRequest 新增 before/after（可选）
2. backend/controllers: normalizeContextLines()
- 不传或 ≤0 → 默认 2 条
- 上限 50 条（防止一次弹窗拉几百条）
- 时间窗改为**按条数自适应**：每条按 30s 估算，最少 1 分钟、最多 2 小时
（不再固定 5 分钟，解决"稀疏服务取不到上下文"的问题）
3. frontend
- 上下文弹窗标题栏加「前后各 [2/5/10/20/50]」选择器，切换后原地重新拉取
- 选择结果写入 localStorage，下次沿用
- 每页条数：数字输入框 → 预设下拉 [100/500/1000/2000/5000]
（原来只能手输且硬顶 1000，现在上限提到 Loki 默认的 5000）
- i18n：去掉"前后各最多 5 条 · 前后 5 分钟"的写死文案
=== 三、实测（平台 API）===
before=2 after=2 → 2 / 2 条
before=5 after=5 → 5 / 5 条
before=10 after=10 → 10 / 10 条
before=50 after=50 → 50 / 50 条
before=999 after=999 → 50 / 50 条（超上限自动收敛）
不传 → 2 / 2 条（默认值生效）
镜像：lghost/lokistack:latest + v0.1 = sha256:d19239fa65f3
410 lines
12 KiB
Go
410 lines
12 KiB
Go
package controllers
|
||
|
||
import (
|
||
"context"
|
||
"errors"
|
||
"net/http"
|
||
"strconv"
|
||
"strings"
|
||
"time"
|
||
|
||
"lokistack/dto"
|
||
"lokistack/models"
|
||
"lokistack/services"
|
||
"lokistack/utils"
|
||
|
||
"github.com/gin-gonic/gin"
|
||
)
|
||
|
||
// 上下文前后条数的默认值与上限（原来是硬编码 5）
|
||
const (
|
||
DefaultContextLines = 2
|
||
MaxContextLines = 50
|
||
)
|
||
|
||
// normalizeContextLines 把前端传入的条数收敛到 [1, MaxContextLines]，
|
||
// 不传(0)或负数一律用默认值。
|
||
func normalizeContextLines(n int) int {
|
||
if n <= 0 {
|
||
return DefaultContextLines
|
||
}
|
||
if n > MaxContextLines {
|
||
return MaxContextLines
|
||
}
|
||
return n
|
||
}
|
||
|
||
func (ctrl *LogsController) Context(c *gin.Context) {
|
||
var req dto.LogContextRequest
|
||
if !utils.BindAndValidate(c, &req) {
|
||
return
|
||
}
|
||
src := services.GetLogSource(req.SourceID)
|
||
if src == nil {
|
||
utils.BadRequest(c, "数据源不存在")
|
||
return
|
||
}
|
||
ts, err := strconv.ParseInt(req.TimestampNs, 10, 64)
|
||
if err != nil || ts < 1 {
|
||
utils.BadRequest(c, "时间戳无效")
|
||
return
|
||
}
|
||
selector, err := services.ContextLogSelector(req.Labels)
|
||
if err != nil {
|
||
utils.BadRequest(c, err.Error())
|
||
return
|
||
}
|
||
query, err := services.AuthorizedLogQuery(src, c.GetUint("userId"), c.GetString("role") == "admin", selector)
|
||
if err != nil {
|
||
utils.Forbidden(c, err.Error())
|
||
return
|
||
}
|
||
// 前后条数由前端传入（默认 2 条，上限 50 —— 再多一次弹窗就是几百条日志）
|
||
beforeN := normalizeContextLines(req.Before)
|
||
afterN := normalizeContextLines(req.After)
|
||
|
||
// 时间窗随条数自适应：原来固定 5 分钟，条数调大时 5 分钟可能装不下，
|
||
// 条数调小时又白扫一大片。按「每条 30s」估算，最少 1 分钟，最多 2 小时。
|
||
span := int64(time.Duration(max(beforeN, afterN)*30) * time.Second)
|
||
if span < int64(time.Minute) {
|
||
span = int64(time.Minute)
|
||
}
|
||
if span > int64(2*time.Hour) {
|
||
span = int64(2 * time.Hour)
|
||
}
|
||
|
||
before, err := services.QueryLokiLogs(c.Request.Context(), src, query, ts-span, ts-1, beforeN, "backward", "")
|
||
if err != nil {
|
||
utils.Error(c, 502, err.Error())
|
||
return
|
||
}
|
||
after, err := services.QueryLokiLogs(c.Request.Context(), src, query, ts+1, ts+span, afterN, "forward", "")
|
||
if err != nil {
|
||
utils.Error(c, 502, err.Error())
|
||
return
|
||
}
|
||
utils.Success(c, gin.H{"before": before.Entries, "after": after.Entries})
|
||
}
|
||
|
||
// LogsController 日志查询代理 + 数据源管理
|
||
type LogsController struct{}
|
||
|
||
func NewLogsController() *LogsController {
|
||
return &LogsController{}
|
||
}
|
||
|
||
// ==================== 数据源 ====================
|
||
|
||
func (ctrl *LogsController) ListSources(c *gin.Context) {
|
||
if c.GetString("role") == "admin" {
|
||
services.EnsureLogSourcesForRunningDeployments()
|
||
}
|
||
list := services.ListVisibleLogSources(c.GetUint("userId"), c.GetString("role") == "admin")
|
||
utils.Success(c, utils.PageData{List: list, Total: int64(len(list)), Page: 1, PageSize: 100})
|
||
}
|
||
|
||
func (ctrl *LogsController) CreateSource(c *gin.Context) {
|
||
var req dto.LogSourceRequest
|
||
if !utils.BindAndValidate(c, &req) {
|
||
return
|
||
}
|
||
if req.Name == "" || req.TargetType == "" {
|
||
utils.BadRequest(c, "名称与类型必填")
|
||
return
|
||
}
|
||
if err := services.ValidateSourceAccess(req.Access); err != nil {
|
||
utils.BadRequest(c, err.Error())
|
||
return
|
||
}
|
||
src := &models.LogSource{
|
||
Name: req.Name, TargetType: req.TargetType,
|
||
Namespace: req.Namespace, Service: req.Service, Port: req.Port,
|
||
URL: req.URL, AuthType: req.AuthType, BasicUser: req.BasicUser,
|
||
BasicPass: req.BasicPass, BearerToken: req.BearerToken, Headers: req.Headers, SkipTLSVerify: req.SkipTLSVerify, IsDefault: req.IsDefault, Access: req.Access,
|
||
}
|
||
if req.TargetType == "cluster-svc" {
|
||
if req.ClusterID > 0 {
|
||
id := req.ClusterID
|
||
src.ClusterID = &id
|
||
} else {
|
||
utils.BadRequest(c, "cluster-svc 类型必须选择集群")
|
||
return
|
||
}
|
||
}
|
||
created, err := services.CreateLogSource(src)
|
||
if err != nil {
|
||
utils.Error(c, 400, err.Error())
|
||
return
|
||
}
|
||
utils.Success(c, created.Sanitized())
|
||
}
|
||
|
||
func (ctrl *LogsController) UpdateSource(c *gin.Context) {
|
||
id, err := strconv.ParseUint(c.Param("id"), 10, 64)
|
||
if err != nil {
|
||
utils.BadRequest(c, "ID 无效")
|
||
return
|
||
}
|
||
var updates map[string]interface{}
|
||
if !utils.BindAndValidate(c, &updates) {
|
||
return
|
||
}
|
||
if src := services.GetLogSource(uint(id)); src != nil && src.Managed && updates["isDefault"] == nil {
|
||
utils.BadRequest(c, "平台部署的数据源由系统管理,不可编辑")
|
||
return
|
||
}
|
||
if err := services.UpdateLogSource(uint(id), updates); err != nil {
|
||
utils.Error(c, 400, err.Error())
|
||
return
|
||
}
|
||
utils.Success(c, gin.H{"message": "updated"})
|
||
}
|
||
|
||
func (ctrl *LogsController) DeleteSource(c *gin.Context) {
|
||
id, err := strconv.ParseUint(c.Param("id"), 10, 64)
|
||
if err != nil {
|
||
utils.BadRequest(c, "ID 无效")
|
||
return
|
||
}
|
||
if src := services.GetLogSource(uint(id)); src != nil && src.Managed {
|
||
utils.BadRequest(c, "平台部署的数据源随部署管理,请先删除对应部署")
|
||
return
|
||
}
|
||
if err := services.DeleteLogSource(uint(id)); err != nil {
|
||
utils.Error(c, 400, err.Error())
|
||
return
|
||
}
|
||
utils.Success(c, gin.H{"message": "deleted"})
|
||
}
|
||
|
||
// TestSource 数据源连通性测试(按 id 或表单参数)
|
||
func (ctrl *LogsController) TestSource(c *gin.Context) {
|
||
var req dto.LogSourceRequest
|
||
if !utils.BindAndValidate(c, &req) {
|
||
return
|
||
}
|
||
var src *models.LogSource
|
||
if req.ID > 0 {
|
||
src = services.GetLogSource(req.ID)
|
||
if src == nil {
|
||
utils.NotFound(c, "数据源不存在")
|
||
return
|
||
}
|
||
} else {
|
||
src = &models.LogSource{
|
||
TargetType: req.TargetType, Namespace: req.Namespace, Service: req.Service,
|
||
Port: req.Port, URL: req.URL, AuthType: req.AuthType,
|
||
BasicUser: req.BasicUser, BasicPass: req.BasicPass, BearerToken: req.BearerToken, Headers: req.Headers, SkipTLSVerify: req.SkipTLSVerify,
|
||
}
|
||
if req.TargetType == "cluster-svc" && req.ClusterID > 0 {
|
||
id := req.ClusterID
|
||
src.ClusterID = &id
|
||
}
|
||
}
|
||
utils.Success(c, services.TestLogSource(c.Request.Context(), src))
|
||
}
|
||
|
||
// ==================== 查询代理 ====================
|
||
|
||
func (ctrl *LogsController) Query(c *gin.Context) {
|
||
var req dto.LogQueryRequest
|
||
if !utils.BindAndValidate(c, &req) {
|
||
return
|
||
}
|
||
src := services.GetLogSource(req.SourceID)
|
||
if src == nil {
|
||
utils.BadRequest(c, "数据源不存在")
|
||
return
|
||
}
|
||
query, err := services.AuthorizedLogQuery(src, c.GetUint("userId"), c.GetString("role") == "admin", req.Query)
|
||
if err != nil {
|
||
utils.Forbidden(c, err.Error())
|
||
return
|
||
}
|
||
result, err := services.QueryLokiLogs(c.Request.Context(), src, query,
|
||
req.Start, req.End, req.Limit, req.Direction, req.Cursor)
|
||
if err != nil {
|
||
utils.Error(c, 502, err.Error())
|
||
return
|
||
}
|
||
utils.Success(c, result)
|
||
}
|
||
|
||
func (ctrl *LogsController) Labels(c *gin.Context) {
|
||
id, _ := strconv.ParseUint(c.Query("sourceId"), 10, 64)
|
||
src := services.GetLogSource(uint(id))
|
||
if src == nil {
|
||
utils.BadRequest(c, "数据源不存在")
|
||
return
|
||
}
|
||
if _, err := services.AuthorizedLogQuery(src, c.GetUint("userId"), c.GetString("role") == "admin", "{}"); err != nil {
|
||
utils.Forbidden(c, err.Error())
|
||
return
|
||
}
|
||
start, _ := strconv.ParseInt(c.Query("start"), 10, 64)
|
||
end, _ := strconv.ParseInt(c.Query("end"), 10, 64)
|
||
labels, err := services.LokiLabels(c.Request.Context(), src, start, end)
|
||
if err != nil {
|
||
utils.Error(c, 502, err.Error())
|
||
return
|
||
}
|
||
utils.Success(c, labels)
|
||
}
|
||
|
||
func (ctrl *LogsController) LabelValues(c *gin.Context) {
|
||
id, _ := strconv.ParseUint(c.Query("sourceId"), 10, 64)
|
||
src := services.GetLogSource(uint(id))
|
||
if src == nil {
|
||
utils.BadRequest(c, "数据源不存在")
|
||
return
|
||
}
|
||
start, _ := strconv.ParseInt(c.Query("start"), 10, 64)
|
||
end, _ := strconv.ParseInt(c.Query("end"), 10, 64)
|
||
requestedQuery := c.Query("query")
|
||
if strings.TrimSpace(requestedQuery) == "" {
|
||
requestedQuery = "{}"
|
||
}
|
||
query, err := services.AuthorizedLogQuery(src, c.GetUint("userId"), c.GetString("role") == "admin", requestedQuery)
|
||
if err != nil {
|
||
utils.Forbidden(c, err.Error())
|
||
return
|
||
}
|
||
values, err := services.LokiLabelValues(c.Request.Context(), src,
|
||
c.Param("name"), query, start, end)
|
||
if err != nil {
|
||
utils.Error(c, 502, err.Error())
|
||
return
|
||
}
|
||
utils.Success(c, values)
|
||
}
|
||
|
||
// MyScope 返回当前登录用户在某数据源上的授权范围。
|
||
//
|
||
// 快捷 LogQL 用它做两件事：
|
||
// 1. 把授权策略里配置的 namespace 作为默认选中值（管理员配了 default/astralx 后开箱即用）
|
||
// 2. 把该值合并进 namespace 候选列表 —— 否则像 default 这种「策略里有、Loki 里没日志」
|
||
// 的命名空间不会出现在下拉里，用户会以为策略没生效
|
||
func (ctrl *LogsController) MyScope(c *gin.Context) {
|
||
id, _ := strconv.ParseUint(c.Query("sourceId"), 10, 64)
|
||
src := services.GetLogSource(uint(id))
|
||
if src == nil {
|
||
utils.BadRequest(c, "数据源不存在")
|
||
return
|
||
}
|
||
if c.GetString("role") == "admin" {
|
||
// 管理员的查询范围不受策略限制（unrestricted=true），
|
||
// 但仍返回策略里的 labels 作为快捷 LogQL 的默认选中值——
|
||
// 否则管理员在小黑框里看到的下拉永远没有策略配的那个 namespace，
|
||
// 会误以为「默认策略没生效」。
|
||
labels := services.UserAccessLabels(src, c.GetUint("userId"))
|
||
if labels == nil {
|
||
labels = map[string]string{}
|
||
}
|
||
utils.Success(c, gin.H{"labels": labels, "unrestricted": true})
|
||
return
|
||
}
|
||
labels := services.UserAccessLabels(src, c.GetUint("userId"))
|
||
if labels == nil {
|
||
utils.Forbidden(c, "无权查询此数据源")
|
||
return
|
||
}
|
||
utils.Success(c, gin.H{"labels": labels, "unrestricted": len(labels) == 0})
|
||
}
|
||
|
||
// Health Loki /ready 探活(查询页状态点)
|
||
func (ctrl *LogsController) Health(c *gin.Context) {
|
||
id, _ := strconv.ParseUint(c.Query("sourceId"), 10, 64)
|
||
src := services.GetLogSource(uint(id))
|
||
if src == nil {
|
||
utils.BadRequest(c, "数据源不存在")
|
||
return
|
||
}
|
||
if _, err := services.AuthorizedLogQuery(src, c.GetUint("userId"), c.GetString("role") == "admin", "{}"); err != nil {
|
||
utils.Forbidden(c, err.Error())
|
||
return
|
||
}
|
||
ctx, cancel := context.WithTimeout(c.Request.Context(), 15*time.Second)
|
||
defer cancel()
|
||
utils.Success(c, gin.H{"ready": services.LokiReady(ctx, src)})
|
||
}
|
||
|
||
// Stats 返回该查询在时间范围内的「总条数/字节/块数」，用于前端展示总页数。
|
||
// 走 index/stats（只读索引），成本远低于真数日志。
|
||
func (ctrl *LogsController) Stats(c *gin.Context) {
|
||
var req dto.LogQueryRequest
|
||
if !utils.BindAndValidate(c, &req) {
|
||
return
|
||
}
|
||
src := services.GetLogSource(req.SourceID)
|
||
if src == nil {
|
||
utils.BadRequest(c, "数据源不存在")
|
||
return
|
||
}
|
||
query, err := services.AuthorizedLogQuery(src, c.GetUint("userId"), c.GetString("role") == "admin", req.Query)
|
||
if err != nil {
|
||
utils.Forbidden(c, err.Error())
|
||
return
|
||
}
|
||
st, err := services.LokiStatsFor(c.Request.Context(), src, query, req.Start, req.End)
|
||
if err != nil {
|
||
utils.Error(c, 502, err.Error())
|
||
return
|
||
}
|
||
utils.Success(c, st)
|
||
}
|
||
|
||
func (ctrl *LogsController) Histogram(c *gin.Context) {
|
||
var req dto.LogQueryRequest
|
||
if !utils.BindAndValidate(c, &req) {
|
||
return
|
||
}
|
||
src := services.GetLogSource(req.SourceID)
|
||
if src == nil {
|
||
utils.BadRequest(c, "数据源不存在")
|
||
return
|
||
}
|
||
query, err := services.AuthorizedLogQuery(src, c.GetUint("userId"), c.GetString("role") == "admin", req.Query)
|
||
if err != nil {
|
||
utils.Forbidden(c, err.Error())
|
||
return
|
||
}
|
||
buckets, err := services.LokiHistogram(c.Request.Context(), src, query, req.Start, req.End)
|
||
if err != nil {
|
||
switch {
|
||
case errors.Is(err, services.ErrTooManyQueries):
|
||
// 429：并发重查询超限，让前端提示"稍后重试"，而不是当成服务故障
|
||
utils.Error(c, http.StatusTooManyRequests, err.Error())
|
||
case strings.Contains(err.Error(), "统计范围过大"):
|
||
utils.Error(c, http.StatusBadRequest, err.Error())
|
||
default:
|
||
utils.Error(c, 502, err.Error())
|
||
}
|
||
return
|
||
}
|
||
utils.Success(c, buckets)
|
||
}
|
||
|
||
// Metrics 日志分析指标查询(仅 admin,路由层 AdminAuth 拦截)
|
||
func (ctrl *LogsController) Metrics(c *gin.Context) {
|
||
var req dto.LogMetricQueryRequest
|
||
if !utils.BindAndValidate(c, &req) {
|
||
return
|
||
}
|
||
src := services.GetLogSource(req.SourceID)
|
||
if src == nil {
|
||
utils.BadRequest(c, "数据源不存在")
|
||
return
|
||
}
|
||
query, err := services.AuthorizedLogQuery(src, c.GetUint("userId"), c.GetString("role") == "admin", req.Query)
|
||
if err != nil {
|
||
utils.Forbidden(c, err.Error())
|
||
return
|
||
}
|
||
series, err := services.LokiMetrics(c.Request.Context(), src, query, req.Start, req.End)
|
||
if err != nil {
|
||
utils.Error(c, 502, err.Error())
|
||
return
|
||
}
|
||
utils.Success(c, series)
|
||
}
|