feat: opsx-loki 全面升级 — 前端重构/用户与LDAP/部署向导/日志查询
基于 WEB-UI-Development-Guide 标准的整体升级,并更名 LokiDeploy → opsx-loki。 前端(Gin+React+shadcn/ui): - 按 02/07 章重构:页面域化拆分(无超 400 行文件)、PageContainer/ListQuery、 反馈分级(成功静默/失败 toast/危险确认)、401 单飞刷新+队列重放 - 暗色模式(next-themes)+ 语义 Token(--primary 品牌色/success/warning/info) - 部署向导(状态推导四步:环境→存储→Loki→采集,内嵌表单闭环) - 日志查询页:LogQL 标签构建器/拖拽直方图/日志上下文/Live Tail/查询历史 - 按用户+标签策略的查询授权(服务端强制合并策略) 后端(Go+Gin+JSON DB): - 简单用户管理(仅 admin,无 RBAC)+ TokenVersion 会话吊销 + 登录 IP 限流 - LDAP 接入:配置(AES 加密存储/hasPassword 脱敏)/测试连接/批量导入/登录 - 存储配置测试连接(SigV4 零依赖)+ 桶不存在自动创建 - Loki 三模式部署修复:SSH 密钥认证、3.x 配置模板、ruler 对象存储嵌套块、 凭证解密、孤儿 deploying 状态恢复、Grafana 部署(可选默认关)+面板导入 - DaemonSet 采集纯 YAML(去 helm 依赖)+ DNS 自愈回退 ClusterIP + 集群名标签 - 日志查询代理:cluster-svc 经 kube-apiserver proxy/url 直连 + cursor 翻页 - 统一 log 命名空间;部署就绪探活;Grafana 访问地址自动记录 修复:run.sh 增量编译检测、Sanitized 脱敏补全、LokiDeployment Update 白名单、 健康检查 CreatedAt 零值等 10+ 存量 bug。 数据兼容:DB 沿用旧 lokideploy.json(新名优先);localStorage 一次性迁移; HKDF 加密前缀保留(已加密凭证不受影响)。
This commit is contained in:
logan committed
1 parent
327e282356
commit
f9c6c1d826
204 files changed
+14107
-9103
No files matched your search
@@ -7,7 +7,7 @@ import (
|
||||
"os/exec"
|
||||
"strings"
|
||||
|
||||
"lokideploy/models"
|
||||
"opsxloki/models"
|
||||
)
|
||||
|
||||
// CollectorDeployer 采集器部署器，按 type 分发到三种安装/卸载流程
|
||||
@@ -97,11 +97,7 @@ func (d *CollectorDeployer) installHost(ctx context.Context, endpoint string) er
|
||||
if host == nil {
|
||||
return fmt.Errorf("主机采集器缺少目标主机")
|
||||
}
|
||||
password, err := DecryptCred(host.Password)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
client, err := NewSSHClient(host.HostIP, host.Port, host.User, password)
|
||||
client, err := NewSSHClientWithAuth(HostSSHAuth(host))
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
@@ -145,7 +141,7 @@ func (d *CollectorDeployer) installHost(ctx context.Context, endpoint string) er
|
||||
if logPath == "" {
|
||||
logPath = "/var/log/*.log"
|
||||
}
|
||||
conf := PromtailConfig(endpoint, logPath)
|
||||
conf := PromtailConfig(endpoint, logPath, "host", d.collector.Labels)
|
||||
d.log("info", "上传 Promtail 配置", "")
|
||||
if _, _, err := client.Execute(ctx, fmt.Sprintf("cat > /etc/promtail/promtail.yaml << 'EOF'\n%s\nEOF", conf)); err != nil {
|
||||
return fmt.Errorf("上传配置失败: %w", err)
|
||||
@@ -191,8 +187,7 @@ func (d *CollectorDeployer) uninstallHost(ctx context.Context) error {
|
||||
if host == nil {
|
||||
return nil
|
||||
}
|
||||
password, _ := DecryptCred(host.Password)
|
||||
client, err := NewSSHClient(host.HostIP, host.Port, host.User, password)
|
||||
client, err := NewSSHClientWithAuth(HostSSHAuth(host))
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
@@ -209,7 +204,7 @@ func (d *CollectorDeployer) uninstallHost(ctx context.Context) error {
|
||||
func (d *CollectorDeployer) installDaemonset(ctx context.Context, endpoint string) error {
|
||||
cluster := d.collector.Cluster
|
||||
if cluster == nil {
|
||||
return fmt.Errorf("DaemonSet 采集器缺少目标集群")
|
||||
return fmt.Errorf("集群采集器缺少目标集群")
|
||||
}
|
||||
env, cleanup, err := kubeconfigEnv(cluster)
|
||||
if err != nil {
|
||||
@@ -217,37 +212,106 @@ func (d *CollectorDeployer) installDaemonset(ctx context.Context, endpoint strin
|
||||
}
|
||||
defer cleanup()
|
||||
|
||||
// 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 = "logging"
|
||||
ns = "log"
|
||||
}
|
||||
|
||||
values := PromtailHelmValues(endpoint)
|
||||
valuesFile, valuesCleanup, err := writeTempFile("promtail-values-*.yaml", values, 0600)
|
||||
clusterName := "unknown"
|
||||
if cluster.Name != "" {
|
||||
clusterName = cluster.Name
|
||||
}
|
||||
|
||||
// 幂等:先清掉同名残留(如历史失败安装留下的无注解资源)
|
||||
del := exec.CommandContext(ctx, "kubectl", "delete",
|
||||
"daemonset/promtail", "configmap/promtail", "serviceaccount/promtail",
|
||||
"-n", ns, "--ignore-not-found=true", "--wait=false")
|
||||
del.Env = env
|
||||
del.CombinedOutput()
|
||||
|
||||
// 纯 YAML 安装(SA/RBAC/ConfigMap/DaemonSet),不依赖后端安装 helm
|
||||
manifest := PromtailDaemonsetYAML(ns, endpoint, clusterName, d.collector.Labels)
|
||||
if err := d.applyYAML(ctx, env, manifest, "promtail"); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
d.log("info", fmt.Sprintf("等待 DaemonSet promtail 就绪(命名空间 %s)...", ns), "")
|
||||
waitCmd := exec.CommandContext(ctx, "kubectl", "rollout", "status", "daemonset/promtail",
|
||||
"-n", ns, "--timeout=5m")
|
||||
waitCmd.Env = env
|
||||
waitOut, _ := waitCmd.CombinedOutput()
|
||||
d.log("info", string(waitOut), "")
|
||||
|
||||
// DNS 自愈:CoreDNS 异常的集群上服务名不可解析,回退 Loki ClusterIP 重建配置
|
||||
svcURL := endpoint
|
||||
if strings.Contains(endpoint, ".svc") {
|
||||
probe := exec.CommandContext(ctx, "kubectl", "exec", "-n", ns,
|
||||
"daemonset/promtail", "--", "wget", "-q", "-O", "/dev/null", "-T", "5", endpoint+"/ready")
|
||||
probe.Env = env
|
||||
if pout, perr := probe.CombinedOutput(); perr != nil {
|
||||
d.log("warn", fmt.Sprintf("promtail 无法解析 Loki 服务名(%s),回退 ClusterIP...", strings.TrimSpace(string(pout))), "")
|
||||
ip := d.lokiClusterIP(ctx, env, endpoint)
|
||||
if ip != "" {
|
||||
// 只替换 host 部分,保留 scheme/port/path: http://loki.ns.svc:3100/... → http://IP:3100/...
|
||||
afterScheme := strings.SplitN(endpoint, "//", 2)
|
||||
if len(afterScheme) == 2 {
|
||||
rest := afterScheme[1]
|
||||
host := rest
|
||||
if i := strings.IndexAny(rest, ":/"); i >= 0 {
|
||||
host = rest[:i]
|
||||
}
|
||||
fallback := afterScheme[0] + "//" + strings.Replace(rest, host, ip, 1)
|
||||
if err := d.applyYAML(ctx, env, PromtailDaemonsetYAML(ns, fallback, clusterName, d.collector.Labels), "promtail"); err == nil {
|
||||
rs := exec.CommandContext(ctx, "kubectl", "rollout", "restart", "daemonset/promtail", "-n", ns)
|
||||
rs.Env = env
|
||||
rs.CombinedOutput()
|
||||
d.log("info", fmt.Sprintf("promtail 推送地址已回退为 ClusterIP: %s", fallback), "")
|
||||
svcURL = fallback
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
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 kubectl apply 一个 manifest(临时文件)
|
||||
func (d *CollectorDeployer) applyYAML(ctx context.Context, env []string, manifest, label string) error {
|
||||
f, cleanup, err := writeTempFile("collector-*.yaml", manifest, 0600)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
defer valuesCleanup()
|
||||
|
||||
d.log("info", fmt.Sprintf("helm install promtail（命名空间 %s）", ns), "")
|
||||
// 添加 grafana repo（幂等）
|
||||
repo := exec.CommandContext(ctx, "helm", "repo", "add", "grafana", "https://grafana.github.io/helm-charts")
|
||||
repo.Env = env
|
||||
repo.CombinedOutput()
|
||||
upd := exec.CommandContext(ctx, "helm", "repo", "update")
|
||||
upd.Env = env
|
||||
upd.CombinedOutput()
|
||||
|
||||
cmd := exec.CommandContext(ctx, "helm", "upgrade", "--install", "promtail", "grafana/promtail",
|
||||
"-f", valuesFile, "--namespace", ns, "--create-namespace", "--wait", "--timeout", "5m")
|
||||
defer cleanup()
|
||||
cmd := exec.CommandContext(ctx, "kubectl", "apply", "-f", f)
|
||||
cmd.Env = env
|
||||
out, err := cmd.CombinedOutput()
|
||||
if err != nil {
|
||||
return fmt.Errorf("helm 安装 Promtail 失败: %w\n%s", err, string(out))
|
||||
d.log("error", fmt.Sprintf("kubectl apply %s failed: %v", label, err), string(out))
|
||||
return fmt.Errorf("kubectl apply %s failed: %w\n%s", label, err, string(out))
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// lokiClusterIP 从 endpoint 的命名空间取 loki Service 的 ClusterIP(不经 DNS)
|
||||
func (d *CollectorDeployer) lokiClusterIP(ctx context.Context, env []string, endpoint string) string {
|
||||
ns := "log"
|
||||
if d.loki != nil && d.loki.Namespace != "" {
|
||||
ns = d.loki.Namespace
|
||||
}
|
||||
cmd := exec.CommandContext(ctx, "kubectl", "get", "svc", "loki", "-n", ns,
|
||||
"-o", "jsonpath={.spec.clusterIP}")
|
||||
cmd.Env = env
|
||||
out, _ := cmd.CombinedOutput()
|
||||
return strings.TrimSpace(string(out))
|
||||
}
|
||||
|
||||
func (d *CollectorDeployer) uninstallDaemonset(ctx context.Context) error {
|
||||
cluster := d.collector.Cluster
|
||||
if cluster == nil {
|
||||
@@ -259,10 +323,16 @@ func (d *CollectorDeployer) uninstallDaemonset(ctx context.Context) error {
|
||||
}
|
||||
defer cleanup()
|
||||
ns := d.collector.Namespace
|
||||
if ns == "" {
|
||||
ns = "logging"
|
||||
if ns == "" && d.loki != nil {
|
||||
ns = d.loki.Namespace
|
||||
}
|
||||
cmd := exec.CommandContext(ctx, "helm", "uninstall", "promtail", "--namespace", ns)
|
||||
if ns == "" {
|
||||
ns = "log"
|
||||
}
|
||||
cmd := exec.CommandContext(ctx, "kubectl", "delete",
|
||||
"daemonset/promtail", "configmap/promtail", "serviceaccount/promtail",
|
||||
"clusterrole/opsxloki-promtail", "clusterrolebinding/opsxloki-promtail",
|
||||
"-n", ns, "--ignore-not-found=true")
|
||||
cmd.Env = env
|
||||
cmd.CombinedOutput()
|
||||
return nil
|
||||
|
||||
Reference in new issue
Block a user