=== 一、两个线上 bug(172.19.16.8 真机取证)===
1) host 标签是字面量 ${HOSTNAME}
Promtail 不做环境变量展开(除非 -config.expand-env + systemd Environment),
实测 21,825 条日志的 host 全是 "${HOSTNAME}",无法按主机过滤。
→ 安装时 `hostname -s` 探测真实主机名并渲染进配置,结果存 Collector.ResolvedHostname,
让「预览配置」与线上一致。
2) 系统日志时间戳解析失效
TencentOS 4 的 /var/log/messages 是 ISO8601(2026-07-17T15:38:24.923784+08:00),
旧正则按传统 syslog(Oct 28 18:11:06)写死 → timestamp 阶段匹配不上,用的是采集时间,
回填历史日志时间全错。
→ 安装时探测目标机格式:iso8601 → RFC3339Nano;rfc3164 → Jan _2 15:04:05 + Local;
系统日志路径同时探测(Ubuntu 没有 messages,写死会一条都采不到)。
实测:写一条 10 分钟前的 ISO8601 探针行,Loki 里 loki_ts 与该行自身时间偏差 **0 秒**;
另写一条 2 小时前的被 Loki 以 entry too far behind 拒绝 —— 恰好反证时间戳是按事件时间
解析后推送的(旧实现会用采集时间,反而被接收)。
=== 二、B 档:结构化多 Job 编辑器 ===
后端
- models/scrape_job.go:ScrapeJob + ValidateScrapeJobs
校验 job 名格式、绝对路径、**敏感路径拦截**(/etc/shadow、/etc/sudoers、/root/.ssh、
id_rsa、*.pem…)、正则可编译、job/host/__path__ 保留标签名、数量上限。
- services/promtail_host_config.go:多 job 渲染;多路径展开成多个 static_configs;
pipeline 顺序硬约束 cri → multiline → json → timestamp → drop → limit;
encoding 写在 scrape_config 级(放进 static_configs 里 promtail 直接起不来)。
- services/host_profile.go:探测 distro / init / hostname / 时区 / 架构 / journald
与真实存在的系统日志路径。
- CollectorService:renderConfig 实时渲染、PreviewConfig(只渲染不落库)、
Update 支持 scrapeJobs 并校验、安装结果(hostname/syslog 格式/配置快照)落库。
- 新增 POST /api/v1/collectors/:id/preview → {config, warnings, hostname, syslogFormat}。
- installHost:配置先写 .new + `promtail -check-syntax` 通过才替换,重启后轮询
systemctl is-active,失败把 journalctl 尾部带回前端。
前端
- 采集项结构化编辑器:名称/路径/标签常驻,多行合并 / JSON / 自定义时间戳 / 命中丢弃 /
限流折叠在「高级」里;「预览 YAML」弹窗在保存前看到将要下发的配置。
- 本地即时校验 + 旧数据(只有 logPath)自动迁移成 app + system 两个采集项。
=== 三、验证(host-verify-promtail @172.19.16.8)===
- 预览:host=pre-cex-ops-jenkins(无 ${HOSTNAME}),2 job / 3 条系统日志路径 /
ISO8601 正则 / 多行 + drop + 限流 全部按填写渲染。
- 部署:promtail active、NRestarts=0、-check-syntax 通过,主机侧配置与预览逐行一致。
- Loki:新数据 host 全为 pre-cex-ops-jenkins;job=system 415 条/5min、job=app 40 条/5min。
- 负向 7/7 返回指名到字段的 400:/etc/shadow、/root/.ssh/id_rsa、相对路径、坏正则、
job 名带空格、保留标签名、空路径。
镜像:lghost/lokistack:latest + v0.1 = sha256:26172f51187d
534 lines
18 KiB
Go
534 lines
18 KiB
Go
package services
|
||
|
||
import (
|
||
"context"
|
||
"encoding/json"
|
||
"fmt"
|
||
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
|
||
"strings"
|
||
"time"
|
||
|
||
"lokistack/models"
|
||
)
|
||
|
||
// CollectorDeployer 采集器部署器,按 type 分发到三种安装/卸载流程
|
||
type CollectorDeployer struct {
|
||
collector *models.Collector
|
||
loki *models.LokiDeployment
|
||
logFunc func(level, message, detail string)
|
||
}
|
||
|
||
func NewCollectorDeployer(c *models.Collector, l *models.LokiDeployment, logFunc func(level, message, detail string)) *CollectorDeployer {
|
||
return &CollectorDeployer{collector: c, loki: l, logFunc: logFunc}
|
||
}
|
||
|
||
func (d *CollectorDeployer) log(level, message, detail string) {
|
||
if d.logFunc != nil {
|
||
d.logFunc(level, message, detail)
|
||
}
|
||
}
|
||
|
||
// Install 按采集器类型分发安装
|
||
func (d *CollectorDeployer) Install(ctx context.Context) error {
|
||
endpoint := d.loki.Endpoint
|
||
if endpoint == "" {
|
||
return fmt.Errorf("关联的 Loki 尚未部署,缺少推送地址(endpoint),请先成功部署该 Loki")
|
||
}
|
||
d.log("info", fmt.Sprintf("开始部署采集器 %s(类型 %s)", d.collector.Name, d.collector.Type), "")
|
||
|
||
var err error
|
||
switch d.collector.Type {
|
||
case "host-promtail":
|
||
err = d.installHost(ctx, endpoint)
|
||
case "k8s-daemonset":
|
||
err = d.installDaemonset(ctx, endpoint)
|
||
case "k8s-fluentbit-sidecar":
|
||
err = d.installSidecar(ctx, endpoint)
|
||
default:
|
||
return fmt.Errorf("不支持的采集器类型: %s", d.collector.Type)
|
||
}
|
||
if err != nil {
|
||
d.log("error", fmt.Sprintf("采集器部署失败: %v", err), "")
|
||
return err
|
||
}
|
||
d.log("success", "采集器部署完成", "")
|
||
return nil
|
||
}
|
||
|
||
// Uninstall 按采集器类型分发卸载
|
||
func (d *CollectorDeployer) Uninstall(ctx context.Context) error {
|
||
d.log("info", fmt.Sprintf("开始卸载采集器 %s(类型 %s)", d.collector.Name, d.collector.Type), "")
|
||
switch d.collector.Type {
|
||
case "host-promtail":
|
||
return d.uninstallHost(ctx)
|
||
case "k8s-daemonset":
|
||
return d.uninstallDaemonset(ctx)
|
||
case "k8s-fluentbit-sidecar":
|
||
return d.uninstallSidecar(ctx)
|
||
}
|
||
return nil
|
||
}
|
||
|
||
// ===== host-promtail:SSH 安装 Promtail 二进制 + systemd =====
|
||
|
||
func (d *CollectorDeployer) installHost(ctx context.Context, endpoint string) error {
|
||
host := d.collector.Host
|
||
if host == nil {
|
||
return fmt.Errorf("主机采集器缺少目标主机")
|
||
}
|
||
client, err := NewSSHClientWithAuth(HostSSHAuth(host))
|
||
if err != nil {
|
||
return err
|
||
}
|
||
defer client.Close()
|
||
|
||
// 检测架构
|
||
archOut, _, _ := client.Execute(ctx, "uname -m")
|
||
arch := strings.TrimSpace(archOut)
|
||
switch arch {
|
||
case "x86_64":
|
||
arch = "amd64"
|
||
case "aarch64", "arm64":
|
||
arch = "arm64"
|
||
default:
|
||
return fmt.Errorf("不支持的主机架构: %s", arch)
|
||
}
|
||
|
||
// 确定 Promtail 版本(优先使用关联的 Loki 版本)
|
||
version := "3.0.0"
|
||
if d.loki != nil && d.loki.Version != "" {
|
||
version = d.loki.Version
|
||
}
|
||
|
||
// 创建必要目录
|
||
d.log("info", "准备目录 /opt/promtail /etc/promtail /var/lib/promtail", "")
|
||
if _, _, err := client.Execute(ctx, "mkdir -p /opt/promtail /etc/promtail /var/lib/promtail"); err != nil {
|
||
return fmt.Errorf("创建 Promtail 目录失败: %w", err)
|
||
}
|
||
|
||
// 下载 Promtail
|
||
d.log("info", fmt.Sprintf("下载 Promtail %s(%s)", version, arch), "")
|
||
url := fmt.Sprintf("https://github.com/grafana/loki/releases/download/v%s/promtail-linux-%s.zip", version, arch)
|
||
dl := fmt.Sprintf("cd /tmp && curl -fsSL -o p.zip '%s' && unzip -o p.zip -d /opt/promtail/ && chmod +x /opt/promtail/promtail-linux-%s", url, arch)
|
||
if _, _, err := client.Execute(ctx, dl); err != nil {
|
||
return fmt.Errorf("下载 Promtail 失败: %w", err)
|
||
}
|
||
client.Execute(ctx, "ln -sf /opt/promtail/promtail-linux-* /usr/local/bin/promtail 2>/dev/null || true")
|
||
|
||
// 生成并上传配置
|
||
logPath := d.collector.LogPath
|
||
if logPath == "" {
|
||
logPath = "/var/log/*.log"
|
||
}
|
||
// 目标机真实身份与 syslog 格式只能在安装时探测(见 Bug3 / Bug4):
|
||
// - host 标签写真实主机名,promtail 不做 ${HOSTNAME} 展开;
|
||
// - 系统日志按实际格式(ISO8601 / 传统 syslog)选正则。
|
||
hostname := detectHostname(ctx, client, host)
|
||
syslogFormat := detectSyslogFormat(ctx, client)
|
||
d.log("info", fmt.Sprintf("目标机 %s,syslog 格式 %s", hostname, syslogFormat), "")
|
||
|
||
// 系统日志路径同样必须探测:Ubuntu 只有 syslog/auth.log,没有 messages
|
||
sysLogPaths := detectSystemLogPaths(ctx, client)
|
||
d.log("info", fmt.Sprintf("系统日志路径:%s", strings.Join(sysLogPaths, ",")), "")
|
||
|
||
conf, err := PromtailConfigForHost(HostPromtailOptions{
|
||
Endpoint: endpoint,
|
||
Cluster: "host",
|
||
LabelsJSON: d.collector.Labels,
|
||
Hostname: hostname,
|
||
SyslogFormat: syslogFormat,
|
||
SystemLogPaths: sysLogPaths,
|
||
LogPath: logPath,
|
||
Jobs: d.collector.ScrapeJobs,
|
||
})
|
||
if err != nil {
|
||
return fmt.Errorf("生成 Promtail 配置失败(现有配置未改动): %w", err)
|
||
}
|
||
d.collector.ResolvedHostname = hostname
|
||
d.collector.SyslogFormat = syslogFormat
|
||
d.collector.SystemLogPaths = sysLogPaths
|
||
|
||
// 先写临时文件并做语法校验,通过后才替换正式配置 ——
|
||
// 配置坏了会把正在跑的采集干掉,不能直接覆盖。
|
||
d.log("info", "上传 Promtail 配置并校验", "")
|
||
if _, _, err := client.Execute(ctx, fmt.Sprintf("cat > /etc/promtail/promtail.yaml.new << 'PROMTAIL_EOF'\n%s\nPROMTAIL_EOF", conf)); err != nil {
|
||
return fmt.Errorf("上传配置失败: %w", err)
|
||
}
|
||
if out, _, err := client.Execute(ctx, "/usr/local/bin/promtail -check-syntax -config.file=/etc/promtail/promtail.yaml.new 2>&1"); err != nil {
|
||
// 老版本 promtail 可能没有 -check-syntax:只有明确报配置错误才阻断
|
||
msg := strings.TrimSpace(out)
|
||
if strings.Contains(msg, "flag provided but not defined") {
|
||
d.log("info", "当前 promtail 不支持 -check-syntax,跳过语法校验", "")
|
||
} else {
|
||
return fmt.Errorf("Promtail 配置校验未通过(现有配置未改动):%s", truncateForMessage(msg, 600))
|
||
}
|
||
}
|
||
if _, _, err := client.Execute(ctx, "mv -f /etc/promtail/promtail.yaml.new /etc/promtail/promtail.yaml"); err != nil {
|
||
return fmt.Errorf("替换 Promtail 配置失败: %w", err)
|
||
}
|
||
|
||
// systemd 服务(生产级)
|
||
service := `[Unit]
|
||
Description=Promtail Log Collector
|
||
Wants=network-online.target
|
||
After=network-online.target
|
||
|
||
[Service]
|
||
Type=simple
|
||
User=root
|
||
Group=root
|
||
ExecStart=/usr/local/bin/promtail -config.file=/etc/promtail/promtail.yaml
|
||
ExecReload=/bin/kill -HUP $MAINPID
|
||
Restart=always
|
||
RestartSec=10
|
||
LimitNOFILE=65536
|
||
StandardOutput=append:/var/log/promtail.log
|
||
StandardError=append:/var/log/promtail.log
|
||
|
||
[Install]
|
||
WantedBy=multi-user.target
|
||
`
|
||
d.log("info", "安装 systemd 服务", "")
|
||
if _, _, err := client.Execute(ctx, fmt.Sprintf("cat > /etc/systemd/system/promtail.service << 'EOF'\n%s\nEOF", service)); err != nil {
|
||
return fmt.Errorf("创建 systemd 服务失败: %w", err)
|
||
}
|
||
if _, _, err := client.Execute(ctx, "systemctl daemon-reload && systemctl enable promtail && systemctl restart promtail"); err != nil {
|
||
return fmt.Errorf("启动 Promtail 失败: %w", err)
|
||
}
|
||
|
||
// 必须校验服务真的起来了:
|
||
// 原来只判断 systemctl 命令有没有执行错误,结果 Promtail 因配置/参数错误
|
||
// 崩溃重启 40+ 次,平台仍然显示"采集器已部署",用户以为采集是好的。
|
||
if err := waitServiceActive(ctx, client, "promtail", 20*time.Second); err != nil {
|
||
out, _, _ := client.Execute(ctx, "journalctl -u promtail -n 15 --no-pager 2>/dev/null | tail -15")
|
||
tail := strings.TrimSpace(out)
|
||
if tail == "" {
|
||
out, _, _ = client.Execute(ctx, "tail -10 /var/log/promtail.log 2>/dev/null")
|
||
tail = strings.TrimSpace(out)
|
||
}
|
||
return fmt.Errorf("Promtail 启动后未处于运行状态:%v\n最近日志:\n%s", err, truncateForMessage(tail, 900))
|
||
}
|
||
|
||
// 保存配置快照
|
||
d.collector.Config = conf
|
||
d.log("success", "Promtail 部署完成", "")
|
||
return nil
|
||
}
|
||
|
||
// waitServiceActive 轮询 systemctl is-active,直到 active 或超时。
|
||
// Promtail 起不来时 systemctl restart 依然返回 0(只是排队拉起),
|
||
// 所以必须自己查状态,否则会把"崩溃重启中"误报成部署成功。
|
||
func waitServiceActive(ctx context.Context, client *SSHClient, service string, timeout time.Duration) error {
|
||
deadline := time.Now().Add(timeout)
|
||
var last string
|
||
for time.Now().Before(deadline) {
|
||
out, _, err := client.Execute(ctx, fmt.Sprintf("systemctl is-active %s 2>&1", service))
|
||
last = strings.TrimSpace(out)
|
||
if err == nil && last == "active" {
|
||
return nil
|
||
}
|
||
time.Sleep(2 * time.Second)
|
||
}
|
||
if last == "" {
|
||
last = "unknown"
|
||
}
|
||
return fmt.Errorf("%s 状态=%s", service, last)
|
||
}
|
||
|
||
func truncateForMessage(s string, max int) string {
|
||
if len(s) <= max {
|
||
return s
|
||
}
|
||
return s[:max] + "…"
|
||
}
|
||
|
||
// detectHostname 取目标机真实主机名(Bug3 修复点)。
|
||
// promtail 不做环境变量展开,把 ${HOSTNAME} 写进配置只会在 Loki 里留下一堆
|
||
// 字面量标签,无法按主机过滤。优先 `hostname -s`,失败退回平台里登记的主机名/IP。
|
||
func detectHostname(ctx context.Context, client *SSHClient, host *models.Host) string {
|
||
if out, _, err := client.Execute(ctx, "hostname -s 2>/dev/null || hostname 2>/dev/null"); err == nil {
|
||
if v := strings.TrimSpace(strings.SplitN(out, "\n", 2)[0]); v != "" && !strings.Contains(v, "${") {
|
||
return v
|
||
}
|
||
}
|
||
if host != nil {
|
||
if v := strings.TrimSpace(host.Name); v != "" {
|
||
return v
|
||
}
|
||
if v := strings.TrimSpace(host.HostIP); v != "" {
|
||
return v
|
||
}
|
||
}
|
||
return "unknown-host"
|
||
}
|
||
|
||
// detectSyslogFormat 探测目标机 /var/log 的时间戳格式(Bug4 修复点)。
|
||
//
|
||
// TencentOS / RHEL 系(rsyslog 新版): 2026-07-17T15:38:24.923784+08:00 → iso8601
|
||
// Debian / 老 rsyslog: Oct 28 18:11:06 → syslog
|
||
//
|
||
// 探测不到(文件不存在/为空)时按传统 syslog 处理(保守默认)。
|
||
func detectSyslogFormat(ctx context.Context, client *SSHClient) string {
|
||
const probe = `f=""; for p in /var/log/messages /var/log/syslog /var/log/secure; do [ -s "$p" ] && f="$p" && break; done; [ -n "$f" ] && head -n 1 "$f"`
|
||
out, _, err := client.Execute(ctx, probe)
|
||
if err != nil || strings.TrimSpace(out) == "" {
|
||
return SyslogFormatClassic
|
||
}
|
||
line := strings.TrimSpace(strings.SplitN(out, "\n", 2)[0])
|
||
if len(line) >= 19 && line[4] == '-' && line[7] == '-' && line[10] == 'T' {
|
||
return SyslogFormatISO8601
|
||
}
|
||
return SyslogFormatClassic
|
||
}
|
||
|
||
func (d *CollectorDeployer) uninstallHost(ctx context.Context) error {
|
||
host := d.collector.Host
|
||
if host == nil {
|
||
return nil
|
||
}
|
||
client, err := NewSSHClientWithAuth(HostSSHAuth(host))
|
||
if err != nil {
|
||
return err
|
||
}
|
||
defer client.Close()
|
||
client.Execute(ctx, "systemctl stop promtail 2>/dev/null || true")
|
||
client.Execute(ctx, "systemctl disable promtail 2>/dev/null || true")
|
||
client.Execute(ctx, "rm -rf /etc/systemd/system/promtail.service /usr/local/bin/promtail /etc/promtail /opt/promtail /var/lib/promtail")
|
||
client.Execute(ctx, "systemctl daemon-reload")
|
||
return nil
|
||
}
|
||
|
||
// ===== k8s-daemonset:以原生清单安装 Promtail DaemonSet =====
|
||
|
||
func (d *CollectorDeployer) installDaemonset(ctx context.Context, endpoint string) error {
|
||
cluster := d.collector.Cluster
|
||
if cluster == nil {
|
||
return fmt.Errorf("集群采集器缺少目标集群")
|
||
}
|
||
|
||
// 默认部署到关联 Loki 的命名空间(统一 log 命名空间策略)
|
||
ns := d.collector.Namespace
|
||
if ns == "" && d.loki != nil {
|
||
ns = d.loki.Namespace
|
||
}
|
||
if ns == "" {
|
||
ns = "log"
|
||
}
|
||
|
||
clusterName := "unknown"
|
||
if cluster.Name != "" {
|
||
clusterName = cluster.Name
|
||
}
|
||
|
||
manifest := PromtailDaemonsetYAML(ns, endpoint, clusterName, d.collector.Labels,
|
||
d.collector.TargetNamespaces, d.collector.AppLabelSource, d.collector.AppInclude)
|
||
|
||
// 关键顺序:先本地校验清单,再删旧资源。
|
||
// 否则模板有误时会出现「旧 DaemonSet 已删、新的建不出来」→ 日志采集直接中断。
|
||
if err := validateManifest(manifest); err != nil {
|
||
d.log("error", fmt.Sprintf("生成的 Promtail 清单无效,已中止且未改动现有资源: %v", err), "")
|
||
return fmt.Errorf("Promtail 清单校验失败(现有采集器未受影响): %w", err)
|
||
}
|
||
|
||
// 删除历史残留并通过 Kubernetes API 提交资源。
|
||
for _, kind := range []string{"daemonset", "configmap", "serviceaccount"} {
|
||
_ = deleteK8sResource(ctx, cluster, kind, ns, "promtail")
|
||
}
|
||
if err := d.applyYAML(ctx, nil, manifest, "promtail"); err != nil {
|
||
return err
|
||
}
|
||
d.log("info", fmt.Sprintf("等待 DaemonSet promtail 就绪(命名空间 %s)...", ns), "")
|
||
if err := waitDaemonSet(ctx, cluster, ns, "promtail"); err != nil {
|
||
return fmt.Errorf("Promtail DaemonSet 未就绪: %w", err)
|
||
}
|
||
svcURL := endpoint
|
||
|
||
d.collector.Config = svcURL
|
||
d.log("success", fmt.Sprintf("Promtail DaemonSet 已部署到 %s(推送至 %s)", ns, svcURL), "")
|
||
return nil
|
||
}
|
||
|
||
// applyYAML 通过 API Server 应用采集器清单。
|
||
func (d *CollectorDeployer) applyYAML(ctx context.Context, _ []string, manifest, label string) error {
|
||
if d.collector.Cluster == nil {
|
||
return fmt.Errorf("采集器缺少集群")
|
||
}
|
||
if err := applyManifest(ctx, d.collector.Cluster, manifest); err != nil {
|
||
d.log("error", fmt.Sprintf("应用 %s 失败: %v", label, err), "")
|
||
return err
|
||
}
|
||
return nil
|
||
}
|
||
|
||
func (d *CollectorDeployer) uninstallDaemonset(ctx context.Context) error {
|
||
cluster := d.collector.Cluster
|
||
if cluster == nil {
|
||
return nil
|
||
}
|
||
|
||
ns := d.collector.Namespace
|
||
if ns == "" && d.loki != nil {
|
||
ns = d.loki.Namespace
|
||
}
|
||
if ns == "" {
|
||
ns = "log"
|
||
}
|
||
// clusterrole / clusterrolebinding 用 loki-promtail;
|
||
// 数组末两项是历史遗留命名,一并清理避免升级后留下孤儿资源。
|
||
for _, item := range [][2]string{
|
||
{"daemonset", "promtail"}, {"configmap", "promtail"}, {"serviceaccount", "promtail"},
|
||
{"clusterrole", "loki-promtail"}, {"clusterrolebinding", "loki-promtail"},
|
||
{"clusterrole", "opsxloki-promtail"}, {"clusterrolebinding", "opsxloki-promtail"},
|
||
} {
|
||
if err := deleteK8sResource(ctx, cluster, item[0], ns, item[1]); err != nil {
|
||
return err
|
||
}
|
||
}
|
||
return nil
|
||
}
|
||
|
||
// ===== k8s-fluentbit-sidecar:kubectl patch 注入 Fluent Bit sidecar =====
|
||
|
||
func (d *CollectorDeployer) installSidecar(ctx context.Context, endpoint string) error {
|
||
cluster := d.collector.Cluster
|
||
if cluster == nil {
|
||
return fmt.Errorf("Sidecar 采集器缺少目标集群")
|
||
}
|
||
target := d.collector.TargetDeployment
|
||
if target == "" {
|
||
return fmt.Errorf("Sidecar 采集器缺少目标 Deployment")
|
||
}
|
||
client, err := clusterClient(cluster, 30*time.Second)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
|
||
ns := d.collector.Namespace
|
||
if ns == "" {
|
||
ns = "default"
|
||
}
|
||
logPath := d.collector.LogPath
|
||
if logPath == "" {
|
||
logPath = "/var/log/app/*.log"
|
||
}
|
||
app := target
|
||
|
||
// 1. 检查目标 deployment 是否存在
|
||
d.log("info", fmt.Sprintf("检查目标 Deployment %s/%s", ns, target), "")
|
||
deployment, err := client.AppsV1().Deployments(ns).Get(ctx, target, metav1.GetOptions{})
|
||
if err != nil {
|
||
return fmt.Errorf("读取目标 Deployment %s/%s: %w", ns, target, err)
|
||
}
|
||
deploymentJSON, err := json.Marshal(deployment)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
mount, err := resolveSidecarLogMount(deploymentJSON, target, logPath, d.collector.ID)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
hasSidecar := false
|
||
for _, container := range deployment.Spec.Template.Spec.Containers {
|
||
if container.Name == "fluent-bit" {
|
||
hasSidecar = true
|
||
break
|
||
}
|
||
}
|
||
|
||
// 3. 部署 ConfigMap
|
||
lokiHost, lokiPort := parseLokiURL(endpoint)
|
||
conf := FluentbitConfig(lokiHost, lokiPort, logPath, app, ns)
|
||
configMapName := fmt.Sprintf("fluent-bit-config-%d", d.collector.ID)
|
||
cm := FluentbitConfigMapYAML(conf, ns, configMapName)
|
||
d.log("info", "部署 Fluent Bit ConfigMap", "")
|
||
if err := applyManifest(ctx, cluster, cm); err != nil {
|
||
return fmt.Errorf("ConfigMap 部署失败: %w", err)
|
||
}
|
||
d.log("info", fmt.Sprintf("为 %s 配置 Fluent Bit Sidecar", target), "")
|
||
if err := patchDeployment(ctx, cluster, ns, target, SidecarPatchYAML(configMapName, mount)); err != nil {
|
||
return fmt.Errorf("Sidecar 注入失败: %w", err)
|
||
}
|
||
if hasSidecar {
|
||
if err := restartDeployment(ctx, cluster, ns, target); err != nil {
|
||
return fmt.Errorf("Sidecar 配置已更新,但重启 Deployment 失败: %w", err)
|
||
}
|
||
}
|
||
d.log("info", "等待 Pod 重建", "")
|
||
if err := waitDeployment(ctx, cluster, ns, target); err != nil {
|
||
return fmt.Errorf("Sidecar Deployment 未就绪: %w", err)
|
||
}
|
||
|
||
// DNS 自愈:CoreDNS 异常集群上 svc 域名不可解析 → 用 Loki ClusterIP 重建 ConfigMap 并滚动重启
|
||
if strings.Contains(endpoint, ".svc") {
|
||
host := strings.Split(strings.Split(endpoint, "//")[1], ":")[0]
|
||
if podDNSBroken(ctx, cluster, ns, target, "fluent-bit", host) {
|
||
d.log("warn", "fluent-bit 无法解析 Loki 服务名,回退 ClusterIP…", "")
|
||
if ip, ierr := serviceClusterIP(ctx, cluster, lokiNamespaceOf(d.loki, endpoint), "loki"); ierr == nil && ip != "" {
|
||
conf2 := FluentbitConfig(ip, lokiPortOf(endpoint), logPath, target, ns)
|
||
cm2 := FluentbitConfigMapYAML(conf2, ns, configMapName)
|
||
if err := applyManifest(ctx, cluster, cm2); err == nil {
|
||
_ = restartDeployment(ctx, cluster, ns, target)
|
||
d.log("info", "fluent-bit 推送地址已回退为 ClusterIP: "+ip, "")
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
return nil
|
||
}
|
||
|
||
func (d *CollectorDeployer) uninstallSidecar(ctx context.Context) error {
|
||
cluster := d.collector.Cluster
|
||
if cluster == nil || d.collector.TargetDeployment == "" {
|
||
return nil
|
||
}
|
||
client, err := clusterClient(cluster, 30*time.Second)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
|
||
ns := d.collector.Namespace
|
||
if ns == "" {
|
||
ns = "default"
|
||
}
|
||
deployment, err := client.AppsV1().Deployments(ns).Get(ctx, d.collector.TargetDeployment, metav1.GetOptions{})
|
||
if err != nil {
|
||
return err
|
||
}
|
||
deploymentJSON, err := json.Marshal(deployment)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
removePatch, err := sidecarRemovePatchYAML(deploymentJSON, d.collector.ID)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if err := patchDeployment(ctx, cluster, ns, d.collector.TargetDeployment, removePatch); err != nil {
|
||
return fmt.Errorf("移除 Sidecar 失败: %w", err)
|
||
}
|
||
if err := deleteK8sResource(ctx, cluster, "configmap", ns, fmt.Sprintf("fluent-bit-config-%d", d.collector.ID)); err != nil {
|
||
return err
|
||
}
|
||
|
||
return nil
|
||
}
|
||
|
||
// lokiPortOf 从 endpoint 提取端口(默认 3100)
|
||
func lokiPortOf(endpoint string) string {
|
||
rest := strings.SplitN(endpoint, "//", 2)
|
||
if len(rest) != 2 {
|
||
return "3100"
|
||
}
|
||
hostPart := rest[1]
|
||
if i := strings.Index(hostPart, ":"); i >= 0 {
|
||
port := hostPart[i+1:]
|
||
if j := strings.IndexAny(port, "/"); j >= 0 {
|
||
port = port[:j]
|
||
}
|
||
if port != "" {
|
||
return port
|
||
}
|
||
}
|
||
return "3100"
|
||
}
|