LokiStack/backend/services/promtail_host_test.go

package services

import (
	"strings"
	"testing"

	"gopkg.in/yaml.v3"

	"lokistack/models"
)

func renderHost(t *testing.T, o HostPromtailOptions) string {
	t.Helper()
	cfg, err := PromtailConfigForHost(o)
	if err != nil {
		t.Fatalf("渲染主机配置失败: %v", err)
	}
	var out map[string]interface{}
	if err := yaml.Unmarshal([]byte(cfg), &out); err != nil {
		t.Fatalf("生成的 promtail 配置不是合法 YAML: %v\n%s", err, cfg)
	}
	if _, ok := out["scrape_configs"]; !ok {
		t.Fatalf("缺少 scrape_configs\n%s", cfg)
	}
	return cfg
}

// 默认（未配置采集项）仍要生成 app + system 两个 job，且系统日志不能套 cri
func TestHostPromtailConfigDefaultJobs(t *testing.T) {
	cfg := renderHost(t, HostPromtailOptions{
		Endpoint:   "http://loki.log.svc:3100",
		Cluster:    "pre",
		Hostname:   "pre-cex-ops-jenkins",
		LogPath:    "/var/log/myapp/*.log",
		LabelsJSON: `{"env":"pre"}`,
	})
	if !strings.Contains(cfg, "job_name: app") || !strings.Contains(cfg, "job_name: system") {
		t.Fatalf("默认应有 app + system 两个 job：\n%s", cfg)
	}
	sysIdx := strings.Index(cfg, "job_name: system")
	if sysIdx < 0 {
		t.Fatal("找不到 system job")
	}
	if strings.Contains(cfg[sysIdx:], "cri: {}") {
		t.Fatal("system job 里错误地使用了 cri 阶段")
	}
	if !strings.Contains(cfg, "env: pre") {
		t.Fatal("自定义标签没写进 external_labels")
	}
}

// Bug3 回归：host 标签必须是真实主机名，不能是 ${HOSTNAME} 字面量
func TestHostPromtailHostnameIsConcrete(t *testing.T) {
	cfg := renderHost(t, HostPromtailOptions{
		Endpoint: "http://loki:3100",
		Hostname: "pre-cex-ops-jenkins",
		LogPath:  "/data/app/*.log",
	})
	if strings.Contains(cfg, "${HOSTNAME}") {
		t.Fatalf("host 标签仍是 ${HOSTNAME} 字面量（promtail 不做环境变量展开）：\n%s", cfg)
	}
	// 断言"每个 static_config 都带真实主机名"而不是写死数量：
	// 默认系统日志有多个文件（messages/secure/cron…），每个文件一条 static_config，
	// 数量会随目标机实际存在的日志文件变化。
	if n := strings.Count(cfg, "host: pre-cex-ops-jenkins"); n < 2 {
		t.Fatalf("app / system 两个 job 都应带真实主机名（实际 %d 处）：\n%s", n, cfg)
	}
	// 即使调用方漏传主机名，也不能退回占位符
	cfg2 := renderHost(t, HostPromtailOptions{Endpoint: "http://loki:3100", LogPath: "/data/app/*.log"})
	if strings.Contains(cfg2, "${") {
		t.Fatalf("空主机名时退化成了占位符：\n%s", cfg2)
	}
}

// Bug4 回归：syslog 格式决定正则与 timestamp 布局
func TestHostPromtailSyslogFormatSwitch(t *testing.T) {
	iso := renderHost(t, HostPromtailOptions{
		Endpoint: "http://loki:3100", Hostname: "h1", SyslogFormat: SyslogFormatISO8601,
		Jobs: []models.ScrapeJob{{Name: "system", System: true}},
	})
	if !strings.Contains(iso, `\d{4}-\d{2}-\d{2}T`) {
		t.Fatalf("iso8601 模式下应使用 ISO8601 正则：\n%s", iso)
	}
	if !strings.Contains(iso, "format: RFC3339Nano") {
		t.Fatalf("iso8601 模式下 timestamp 应用 RFC3339Nano：\n%s", iso)
	}
	if strings.Contains(iso, "Jan _2 15:04:05") {
		t.Fatalf("iso8601 模式下不该出现传统 syslog 布局：\n%s", iso)
	}

	classic := renderHost(t, HostPromtailOptions{
		Endpoint: "http://loki:3100", Hostname: "h1", SyslogFormat: SyslogFormatClassic,
		Jobs: []models.ScrapeJob{{Name: "system", System: true}},
	})
	if !strings.Contains(classic, "Jan _2 15:04:05") {
		t.Fatalf("传统 syslog 模式应使用经典布局：\n%s", classic)
	}
	if strings.Contains(classic, `\d{4}-\d{2}-\d{2}T`) {
		t.Fatalf("传统 syslog 模式不该出现 ISO8601 正则：\n%s", classic)
	}
}

// 容器路径仍保留 cri 解析；普通文件不套 cri（否则非 CRI 行会被丢）
func TestHostPromtailCRIStageOnlyForCRIPaths(t *testing.T) {
	cri := renderHost(t, HostPromtailOptions{
		Endpoint: "http://loki:3100", Hostname: "h1",
		Jobs: []models.ScrapeJob{{Name: "pods", Paths: []string{"/var/log/pods/*/*/*.log"}}},
	})
	if !strings.Contains(cri, "cri: {}") {
		t.Fatalf("CRI 路径应保留 cri 阶段：\n%s", cri)
	}
	plain := renderHost(t, HostPromtailOptions{
		Endpoint: "http://loki:3100", Hostname: "h1",
		Jobs: []models.ScrapeJob{{Name: "app", Paths: []string{"/data/app/logs/*.log"}}},
	})
	if strings.Contains(plain, "cri: {}") {
		t.Fatalf("普通应用日志不应使用 cri 阶段：\n%s", plain)
	}
}

// B 档：多 job / 多路径 / 多行 / 丢弃 / 限流都要真的渲染出来
func TestHostPromtailStructuredJobs(t *testing.T) {
	cfg := renderHost(t, HostPromtailOptions{
		Endpoint: "http://loki:3100", Hostname: "h1", SyslogFormat: SyslogFormatISO8601,
		Jobs: []models.ScrapeJob{
			{
				Name:   "app-api",
				Paths:  []string{"/data/app/logs/*.log", "/data/app/logs/error/*.log"},
				Labels: map[string]string{"app": "cex-api"},
				Multiline: &models.JobMultiline{
					FirstLine: `^\d{4}-\d{2}-\d{2}`,
					MaxLines:  500,
				},
				JSON:        true,
				Timestamp:   &models.JobTimestamp{Source: "time", Format: "RFC3339Nano"},
				DropIfMatch: []string{`"status":200`},
				RateLimit:   &models.JobRateLimit{Rate: 500, Burst: 900},
			},
			{Name: "system", System: true},
		},
	})
	for _, want := range []string{
		"job_name: app-api",
		"/data/app/logs/error/*.log",
		"app: cex-api",
		"max_lines: 500",
		"json: {}",
		"drop_counter_reason: drop_if_match",
		"rate: 500",
		"burst: 900",
		"job_name: system",
	} {
		if !strings.Contains(cfg, want) {
			t.Fatalf("配置缺少 %q：\n%s", want, cfg)
		}
	}
	// 两个路径应渲染成两个 static_configs
	if strings.Count(cfg, "targets:") != 3 { // app-api 两个路径 + system 默认路径 6 条? 见下
		// system 默认路径有 6 条，这里只断言 app-api 的两条
		if strings.Count(cfg, "/data/app/logs/") < 2 {
			t.Fatalf("多路径未展开成多个 static_configs：\n%s", cfg)
		}
	}
}

// 校验必须拦死敏感路径和坏正则（这两种都会把主机凭据搬进日志平台 / 让 promtail 起不来）
func TestValidateScrapeJobsRejectsUnsafe(t *testing.T) {
	bad := [][]models.ScrapeJob{
		{{Name: "x", Paths: []string{"/etc/shadow"}}},
		{{Name: "x", Paths: []string{"/root/.ssh/id_rsa"}}},
		{{Name: "x", Paths: []string{"relative/path.log"}}},
		{{Name: "x", Paths: []string{"/var/log/a.log"}, Multiline: &models.JobMultiline{FirstLine: "(["}}},
		{{Name: "x", Paths: []string{"/var/log/a.log"}, DropIfMatch: []string{"(["}}},
		{{Name: "x", Paths: []string{"/var/log/a.log"}, Labels: map[string]string{"job": "y"}}},
		{{Name: "", Paths: []string{"/var/log/a.log"}}},
		{{Name: "x", Paths: nil}},
	}
	for i, jobs := range bad {
		if err := models.ValidateScrapeJobs(jobs); err == nil {
			t.Fatalf("第 %d 组非法采集项应被拒绝: %+v", i+1, jobs)
		}
	}
	good := []models.ScrapeJob{{Name: "app", Paths: []string{"/var/log/a.log"}, Labels: map[string]string{"app_name": "x"}}}
	if err := models.ValidateScrapeJobs(good); err != nil {
		t.Fatalf("合法采集项被误拒: %v", err)
	}
}