Files
LokiStack/backend/services/log_source.go
T
opsx ad0d35f72a
CI / Backend (go vet + test) (push) Canceled after 0s
CI / Frontend (lint + build + test) (push) Canceled after 0s
feat(alerts,logs): A 方案—告警窗口 15m→5m 消重复扫描;数据源可编辑/删除且删后不被自动重建
=== 1. 告警窗口 15m → 5m(阈值同比例 1/3)===
背景:检查间隔已改 5 分钟,但规则窗口仍是 15 分钟 ——
同一个 15 分钟窗口每 5 分钟被完整重扫一遍,等于 3 倍重复扫描。
在 370GB/天、全命名空间 selector 下,这是 Loki CPU 的最大稳态来源。

    FATAL            10 → 4        窗口 5m
    Go panic         10 → 4        窗口 5m
    JVM OOM           5 → 2        窗口 5m
    StackOverflow     5 → 2        窗口 5m
    应用错误激增  100000 → 35000    窗口 5m(改名 error > 3.5万)
    ingress 5xx      2000 → 700    窗口 5m
    ingress 4xx     20000 → 7000   窗口 5m
    DB 连接失败      1000 → 350    窗口 5m

阈值按 1/3 缩放是必须的:窗口缩短 3 倍后,同样的流量在 5 分钟内只有
1/3 的条数,阈值不动就永远打不到了。
迁移 v10→v11 只改"窗口仍是 15 且名字匹配内置规则"的项,用户手调的不动。

=== 2. 数据源支持编辑/删除(含托管源)===
原来前后端都把托管源锁死:
  - 前端 `!managed &&` 直接不渲染编辑/删除按钮
  - 后端 DeleteSource 返回「平台部署的数据源随部署管理,请先删除对应部署」
用户的诉求是"数据源必须能编辑和删除",且确实有场景(不想让某个源出现在
查询下拉里、想改认证头)。

=== 3. 修「删了又回来」的真 bug ===
EnsureLogSourcesForRunningDeployments() 在 **每次打开数据源列表和启动时**
都会给 running 部署补齐数据源 —— 所以删掉托管源后,刷新页面它又回来了,
用户看到的现象就是"删不掉"。

新增 LokiDeployment.AutoSourceDisabled:
  - 删除托管源时置位(DisableAutoSourceFor)
  - EnsureLogSourceForDeployment 见到该开关直接 return
  - 部署被编辑/重新部署时重置(EnableAutoSourceFor),即"你要用它了就恢复"
前端配套文案:
  - 编辑托管源弹窗提示"重新部署会覆盖集群/Service/端口,认证与请求头保留"
  - 删除确认明确"删除后不会再自动重建,重新部署该 Loki 才恢复"

=== 实测(20300eef92d2)===
  v11 迁移后 8 条规则全部 窗口=5m,阈值 4/4/2/2/35000/700/7000/350
  go test ./... 全绿
2026-10-30 15:28:30 +00:00

629 lines
17 KiB
Go
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
// 日志数据源管理 — CRUD、默认源、部署自动注册(Managed 标记)。
package services
import (
"context"
"encoding/json"
"fmt"
"log"
"regexp"
"sort"
"strings"
"time"
"lokistack/models"
"lokistack/pkg/crypto"
"lokistack/repository"
)
// ==================== 数据源 CRUD ====================
func GetLogSource(id uint) *models.LogSource {
db := repository.GetDB()
db.RLock()
defer db.RUnlock()
for i := range db.LogSources {
if db.LogSources[i].ID == id {
s := db.LogSources[i]
return &s
}
}
return nil
}
func ListLogSources() []map[string]interface{} {
db := repository.GetDB()
db.RLock()
defer db.RUnlock()
out := make([]map[string]interface{}, 0, len(db.LogSources))
for i := range db.LogSources {
out = append(out, db.LogSources[i].Sanitized())
}
return out
}
func ListVisibleLogSources(userID uint, admin bool) []map[string]interface{} {
db := repository.GetDB()
db.RLock()
defer db.RUnlock()
out := make([]map[string]interface{}, 0)
for i := range db.LogSources {
s := &db.LogSources[i]
if admin || policyGroupAccess(db.LogPolicyGroups, s.ID, userID) != nil || sourceAccess(s, userID) != nil {
item := s.Sanitized()
if !admin {
item = map[string]interface{}{"id": s.ID, "name": s.Name, "isDefault": s.IsDefault}
}
out = append(out, item)
}
}
return out
}
// UserAccessLabels 返回某用户在指定数据源上的授权 labels。
//
// 返回值语义:
// - nil → 该用户对此数据源完全无权限
// - 空 map → 有权限但不限定范围(可查全部)
// - 非空 map → 只能查匹配这些 label 的日志(如 {"namespace":"astralx"})
func UserAccessLabels(src *models.LogSource, userID uint) map[string]string {
db := repository.GetDB()
var access *models.LogSourceAccess
if db != nil {
db.RLock()
access = policyGroupAccess(db.LogPolicyGroups, src.ID, userID)
db.RUnlock()
}
if access == nil {
access = sourceAccess(src, userID)
}
if access == nil {
return nil
}
if access.Labels == nil {
return map[string]string{}
}
return access.Labels
}
func sourceAccess(src *models.LogSource, userID uint) *models.LogSourceAccess {
for i := range src.Access {
if src.Access[i].UserID == userID {
return &src.Access[i]
}
}
return nil
}
var labelNamePattern = regexp.MustCompile(`^[a-zA-Z_][a-zA-Z0-9_]*$`)
func ValidLogLabelName(name string) bool { return labelNamePattern.MatchString(name) }
// AuthorizedLogQuery constrains the Loki stream selector server-side. Non-admins must be explicitly granted access.
func AuthorizedLogQuery(src *models.LogSource, userID uint, admin bool, query string) (string, error) {
if admin {
return query, nil
}
db := repository.GetDB()
var access *models.LogSourceAccess
if db != nil {
db.RLock()
access = policyGroupAccess(db.LogPolicyGroups, src.ID, userID)
db.RUnlock()
}
if access == nil {
access = sourceAccess(src, userID)
}
if access == nil {
return "", fmt.Errorf("无权查询此数据源")
}
q := strings.TrimSpace(query)
if !strings.HasPrefix(q, "{") {
return "", fmt.Errorf("普通用户查询必须以日志流选择器开头")
}
close := strings.IndexByte(q, '}')
if close < 0 || strings.Contains(q[close+1:], "{") {
return "", fmt.Errorf("查询只能包含一个日志流选择器")
}
if len(access.Labels) == 0 {
return q, nil
}
keys := make([]string, 0, len(access.Labels))
for k := range access.Labels {
if !labelNamePattern.MatchString(k) {
return "", fmt.Errorf("访问策略包含无效标签名")
}
keys = append(keys, k)
}
sort.Strings(keys)
matchers := make([]string, 0, len(keys))
for _, k := range keys {
v, _ := json.Marshal(access.Labels[k])
matchers = append(matchers, k+"="+string(v))
}
inner := strings.TrimSpace(q[1:close])
if inner != "" {
inner += ","
}
return "{" + inner + strings.Join(matchers, ",") + "}" + q[close+1:], nil
}
func CreateLogSource(s *models.LogSource) (*models.LogSource, error) {
if err := encryptSourceCredentials(s); err != nil {
return nil, err
}
db := repository.GetDB()
newID := db.NextID("logSources")
db.Lock()
defer db.Unlock()
s.ID = newID
if s.IsDefault {
for i := range db.LogSources {
db.LogSources[i].IsDefault = false
}
}
// 首个数据源自动设默认
if len(db.LogSources) == 0 {
s.IsDefault = true
}
db.LogSources = append(db.LogSources, *s)
db.RequestSave()
return s, nil
}
func encryptSourceCredentials(s *models.LogSource) error {
if s.BasicPass != "" {
v, err := crypto.EncryptField(s.BasicPass)
if err != nil {
return err
}
s.BasicPass = v
}
if s.BearerToken != "" {
v, err := crypto.EncryptField(s.BearerToken)
if err != nil {
return err
}
s.BearerToken = v
}
for name, value := range s.Headers {
if !validSourceHeader(name) {
return fmt.Errorf("自定义 Header 名称无效: %s", name)
}
v, err := crypto.EncryptField(value)
if err != nil {
return err
}
s.Headers[name] = v
}
return nil
}
func validSourceHeader(name string) bool {
if name == "" || strings.EqualFold(name, "Host") || strings.EqualFold(name, "Authorization") {
return false
}
for _, c := range name {
if !(c >= 'a' && c <= 'z' || c >= 'A' && c <= 'Z' || c >= '0' && c <= '9' || c == '-') {
return false
}
}
return true
}
func UpdateLogSource(id uint, patch map[string]interface{}) error {
db := repository.GetDB()
db.Lock()
defer db.Unlock()
for i := range db.LogSources {
if db.LogSources[i].ID != id {
continue
}
if v, ok := patch["name"].(string); ok && v != "" {
db.LogSources[i].Name = v
}
if v, ok := patch["namespace"].(string); ok {
db.LogSources[i].Namespace = v
}
if v, ok := patch["service"].(string); ok {
db.LogSources[i].Service = v
}
if v, ok := patch["port"].(float64); ok {
db.LogSources[i].Port = int(v)
}
if v, ok := patch["url"].(string); ok {
db.LogSources[i].URL = v
}
if v, ok := patch["authType"].(string); ok {
db.LogSources[i].AuthType = v
}
if v, ok := patch["basicUser"].(string); ok {
db.LogSources[i].BasicUser = v
}
if v, ok := patch["bearerToken"].(string); ok {
if v != "" {
enc, err := crypto.EncryptField(v)
if err != nil {
return err
}
db.LogSources[i].BearerToken = enc
}
}
if v, ok := patch["skipTLSVerify"].(bool); ok {
db.LogSources[i].SkipTLSVerify = v
}
if raw, ok := patch["headers"]; ok {
data, err := json.Marshal(raw)
if err != nil {
return err
}
var headers map[string]string
if err := json.Unmarshal(data, &headers); err != nil {
return err
}
for name, value := range headers {
if !validSourceHeader(name) {
return fmt.Errorf("自定义 Header 名称无效: %s", name)
}
if value == "" {
if old, ok := db.LogSources[i].Headers[name]; ok {
headers[name] = old
}
continue
}
enc, err := crypto.EncryptField(value)
if err != nil {
return err
}
headers[name] = enc
}
db.LogSources[i].Headers = headers
}
if v, ok := patch["basicPass"].(string); ok && v != "" {
if enc, err := crypto.EncryptField(v); err == nil {
db.LogSources[i].BasicPass = enc
}
}
if v, ok := patch["isDefault"].(bool); ok && v {
for j := range db.LogSources {
db.LogSources[j].IsDefault = db.LogSources[j].ID == id
}
}
if raw, ok := patch["access"]; ok {
data, err := json.Marshal(raw)
if err != nil {
return err
}
var access []models.LogSourceAccess
if err := json.Unmarshal(data, &access); err != nil {
return err
}
if err := ValidateSourceAccess(access); err != nil {
return err
}
db.LogSources[i].Access = access
}
db.LogSources[i].UpdatedAt = time.Now()
db.RequestSave()
return nil
}
return fmt.Errorf("数据源不存在")
}
func ValidateSourceAccess(access []models.LogSourceAccess) error {
seen := map[uint]bool{}
for _, rule := range access {
if rule.UserID == 0 || seen[rule.UserID] {
return fmt.Errorf("访问策略的用户 ID 无效或重复")
}
seen[rule.UserID] = true
for k := range rule.Labels {
if !labelNamePattern.MatchString(k) {
return fmt.Errorf("访问策略包含无效标签名: %s", k)
}
}
}
return nil
}
func DeleteLogSource(id uint) error {
db := repository.GetDB()
db.Lock()
defer db.Unlock()
for i := range db.LogSources {
if db.LogSources[i].ID == id {
db.LogSources = append(db.LogSources[:i], db.LogSources[i+1:]...)
groups := db.LogPolicyGroups[:0]
for _, group := range db.LogPolicyGroups {
if group.SourceID != id {
groups = append(groups, group)
}
}
db.LogPolicyGroups = groups
db.RequestSave()
return nil
}
}
return fmt.Errorf("数据源不存在")
}
// reachableEndpoint 把部署记录里的访问地址转换成「平台侧能真正连上」的地址。
//
// 主机模式创建部署时存的是容器的监听地址(0.0.0.0:3100 / [::]:3100),
// 那是容器内部视角,平台进程在别的机器/集群里,用它必然 connection refused。
// 换成该主机的真实 IP 再返回;主机信息缺失时原样返回(至少不会更差)。
func reachableEndpoint(d *models.LokiDeployment) string {
endpoint := strings.TrimSpace(d.Endpoint)
if endpoint == "" {
return endpoint
}
if !strings.Contains(endpoint, "0.0.0.0") && !strings.Contains(endpoint, "[::]") {
return endpoint
}
hostIP := ""
if d.Host != nil {
hostIP = strings.TrimSpace(d.Host.HostIP)
}
if hostIP == "" {
// 没带 Host 关联时回库查一次,避免调用方漏预加载
db := repository.GetDB()
if db != nil && d.HostID != nil {
db.RLock()
for i := range db.Hosts {
if db.Hosts[i].ID == *d.HostID {
hostIP = strings.TrimSpace(db.Hosts[i].HostIP)
break
}
}
db.RUnlock()
}
}
if hostIP == "" {
return endpoint
}
endpoint = strings.Replace(endpoint, "0.0.0.0", hostIP, 1)
endpoint = strings.Replace(endpoint, "[::]", hostIP, 1)
return endpoint
}
// CleanupManagedLogSource 删除某部署名对应的**平台托管**数据源。
//
// 为什么必须有:部署删除后它自动注册的数据源若不清掉,「日志查询」下拉里会
// 留一个指向已删实例的源,选中必然连接失败(线上残留过 host-verify-loki)。
//
// 连带处理:
// - 该源上的查询授权策略组:DeleteLogSource 内部会一并删除
// - 绑定了该源的告警规则:置为禁用(保留配置便于审计,但不再触发,
// 否则引擎每次评估都会因为连不上而写失败记录)
// - 手填的数据源(Managed=false)不动:那是用户自己建的,可能指向别的实例
func CleanupManagedLogSource(name string) {
db := repository.GetDB()
if db == nil || name == "" {
return
}
db.RLock()
var sourceID uint
for _, src := range db.LogSources {
if src.Managed && src.Name == name {
sourceID = src.ID
break
}
}
db.RUnlock()
if sourceID == 0 {
// 部署本来就没注册数据源(如部署失败),无需处理
return
}
if err := DeleteLogSource(sourceID); err != nil {
log.Printf("[deploy-delete] 清理数据源 %q(id=%d) 失败: %v", name, sourceID, err)
}
db.Lock()
defer db.Unlock()
disabled := 0
for i := range db.AlertRules {
if db.AlertRules[i].SourceID == sourceID && db.AlertRules[i].Enabled {
db.AlertRules[i].Enabled = false
disabled++
}
}
if disabled > 0 {
db.RequestSave()
log.Printf("[deploy-delete] %d 条告警规则因数据源被删而禁用", disabled)
}
}
// CleanupOrphanManagedLogSources 启动时清理孤儿托管数据源:
// 部署记录已删除、或所属集群已删除的 managed 源。
//
// 为什么需要:CleanupManagedLogSource 只在「删除部署」的代码路径里执行;
// 如果部署是在旧版本(没有清理逻辑)删掉的,或删除过程中进程退出,
// 托管源就会永久残留,在数据源列表里指向一个不存在的实例。
func CleanupOrphanManagedLogSources() {
db := repository.GetDB()
if db == nil {
return
}
db.Lock()
names := make(map[string]bool, len(db.Deployments))
for _, d := range db.Deployments {
names[d.Name] = true
}
clusters := make(map[uint]bool, len(db.Clusters))
for _, c := range db.Clusters {
clusters[c.ID] = true
}
kept := make([]models.LogSource, 0, len(db.LogSources))
removed, removedDefault := 0, false
for _, s := range db.LogSources {
orphan := false
if s.Managed {
if !names[s.Name] {
orphan = true
}
if s.ClusterID != nil && !clusters[*s.ClusterID] {
orphan = true
}
}
if orphan {
removed++
if s.IsDefault {
removedDefault = true
}
continue
}
kept = append(kept, s)
}
if removed > 0 {
if removedDefault && len(kept) > 0 {
kept[0].IsDefault = true
}
db.LogSources = kept
db.RequestSave()
log.Printf("[startup] 清理 %d 个孤儿托管数据源", removed)
}
db.Unlock()
}
// AutoDetectLogSource 探测数据源的可用连接方式,返回最终选择(cluster-svc / url)。
//
// 交付环境组合很多:平台可能在主机或集群内,Loki 也可能在主机或集群内。
// cluster-svc 走 kube-apiserver 的 service proxy,不依赖集群 DNS;url 直连
// 则需要平台能访问到该地址。这里按「代理优先、URL 兜底」实测,避免按部署
// 模式硬编码导致某种组合连不上。
func AutoDetectLogSource(ctx context.Context, src *models.LogSource) string {
if src.ClusterID != nil {
probe := *src
probe.TargetType = "cluster-svc"
if _, status, err := lokiProxyGET(ctx, &probe, "/ready", nil); err == nil && status == 200 {
src.TargetType = "cluster-svc"
return "cluster-svc"
}
}
if src.URL != "" {
probe := *src
probe.TargetType = "url"
if _, status, err := lokiProxyGET(ctx, &probe, "/ready", nil); err == nil && status == 200 {
src.TargetType = "url"
return "url"
}
}
return ""
}
// EnsureLogSourceForDeployment 部署成功后自动注册查询数据源(host→url;k8s→cluster-svc)。
// 已存在同名数据源时:平台托管的自动刷新连接信息,用户手填的不覆盖。
func EnsureLogSourceForDeployment(d *models.LokiDeployment) {
if d == nil || d.Endpoint == "" {
return
}
// 用户主动删过这个自动数据源 → 不再补回来(重新部署/编辑部署会重置该开关)
if d.AutoSourceDisabled {
return
}
svc := "loki"
port := 3100
target := &models.LogSource{
Name: d.Name, TargetType: "cluster-svc",
ClusterID: d.ClusterID, Namespace: d.Namespace, Service: svc, Port: port,
AuthType: "none", Managed: true,
}
if d.Mode == "host-binary" || d.Mode == "host-docker" {
target.TargetType = "url"
// 主机模式部署的 endpoint 记的是**容器监听地址**(0.0.0.0:3100),
// 直接拿来当数据源 URL 会得到 dial tcp 0.0.0.0:3100: connection refused ——
// 平台进程不在那台主机上,0.0.0.0 只在本机有意义。
// 这里换成主机真实 IP,平台才能连上。
target.URL = reachableEndpoint(d)
target.ClusterID = nil
} else {
// k8s 部署:集群内 Service 地址也记下来,作为 URL 直连候选
target.URL = d.Endpoint
}
// 实测可用性后决定连接方式:主机/集群内平台、主机/集群内 Loki 都兼容
detectCtx, cancel := context.WithTimeout(context.Background(), 6*time.Second)
if method := AutoDetectLogSource(detectCtx, target); method != "" {
log.Printf("[log-source] %q 自动选择连接方式: %s", target.Name, method)
}
cancel()
db := repository.GetDB()
db.Lock()
for i := range db.LogSources {
if db.LogSources[i].Name != target.Name {
continue
}
if !db.LogSources[i].Managed {
db.Unlock()
return
}
db.LogSources[i].TargetType = target.TargetType
db.LogSources[i].ClusterID = target.ClusterID
db.LogSources[i].Namespace = target.Namespace
db.LogSources[i].Service = target.Service
db.LogSources[i].Port = target.Port
db.LogSources[i].URL = target.URL
db.LogSources[i].UpdatedAt = time.Now()
db.Unlock()
db.RequestSave()
return
}
db.Unlock()
_, _ = CreateLogSource(target)
}
// DisableAutoSourceFor 标记某个部署不再自动注册数据源。
//
// 触发场景:用户在「日志查询 → 数据源」里删掉了平台自动创建的数据源。
// 不置位的话,EnsureLogSourcesForRunningDeployments 会立刻把它补回来。
func DisableAutoSourceFor(name string) {
db := repository.GetDB()
if db == nil {
return
}
db.Lock()
defer db.Unlock()
for i := range db.Deployments {
if db.Deployments[i].Name == name && !db.Deployments[i].AutoSourceDisabled {
db.Deployments[i].AutoSourceDisabled = true
log.Printf("[log-source] %q 的自动数据源已被用户删除,不再自动重建(重新部署可恢复)", name)
db.RequestSave()
}
}
}
// EnableAutoSourceFor 重新允许自动注册(部署被编辑/重新部署时调用)。
func EnableAutoSourceFor(name string) {
db := repository.GetDB()
if db == nil {
return
}
db.Lock()
defer db.Unlock()
for i := range db.Deployments {
if db.Deployments[i].Name == name && db.Deployments[i].AutoSourceDisabled {
db.Deployments[i].AutoSourceDisabled = false
db.RequestSave()
}
}
}
// EnsureLogSourcesForRunningDeployments 补齐运行中部署的数据源,避免查询页默认没有数据源。
func EnsureLogSourcesForRunningDeployments() {
db := repository.GetDB()
db.RLock()
running := make([]models.LokiDeployment, 0, len(db.Deployments))
for i := range db.Deployments {
d := db.Deployments[i]
if d.Status == "running" && d.Endpoint != "" {
running = append(running, d)
}
}
db.RUnlock()
for i := range running {
EnsureLogSourceForDeployment(&running[i])
}
}