yovision/Bell/server/app/alert/lifecycle/service_test.go

package lifecycle

import (
	"context"
	"fmt"
	"os"
	"sync"
	"testing"
	"time"

	"git.ilapage.cn/ila/yovision/Bell/server/app/alert/query"
	"git.ilapage.cn/ila/yovision/Bell/server/app/audit"
	"git.ilapage.cn/ila/yovision/Bell/server/app/auth"
	"git.ilapage.cn/ila/yovision/Bell/server/app/event"
	"git.ilapage.cn/ila/yovision/Bell/server/app/rule"
	"git.ilapage.cn/ila/yovision/Bell/server/internal/platform"
	"git.ilapage.cn/ila/yovision/Bell/server/migrations"
)

func TestConcurrentAckCloseAndRestart(t *testing.T) {
	url := os.Getenv("BELL_TEST_DATABASE_URL")
	if url == "" {
		t.Skip("BELL_TEST_DATABASE_URL is not set")
	}
	ctx := context.Background()
	db, err := platform.OpenDatabase(ctx, url)
	if err != nil {
		t.Fatal(err)
	}
	if err := migrations.Apply(ctx, db); err != nil {
		t.Fatal(err)
	}
	prefix := fmt.Sprintf("lifecycle-%d", time.Now().UnixNano())
	actors := make([]auth.User, 2)
	for i := range actors {
		var id string
		username := fmt.Sprintf("%s-%d", prefix, i)
		if err := db.QueryRow(ctx, `INSERT INTO bell_users(username,display_name,password_hash) VALUES($1,$2,'test-only-not-a-login-hash') RETURNING id::text`, username, username).Scan(&id); err != nil {
			t.Fatal(err)
		}
		if _, err := db.Exec(ctx, `INSERT INTO bell_user_roles(user_id,role_id) SELECT $1,id FROM bell_roles WHERE code='operator'`, id); err != nil {
			t.Fatal(err)
		}
		actors[i] = auth.User{ID: id, Username: username, DisplayName: username, Roles: []string{"operator"}}
	}
	eventType := prefix + "-event"
	ruleService := rule.Service{DB: db}
	createdRule, err := ruleService.Create(ctx, rule.CreateInput{Code: prefix, Name: "生命周期测试规则", EventType: &eventType, MinimumSeverity: "low"})
	if err != nil {
		t.Fatal(err)
	}
	eventService := event.Service{DB: db, AfterPersist: ruleService.Match}
	accepted, err := eventService.Ingest(ctx, event.Command{ProducerID: prefix, SourceEventID: "1", EventType: eventType, OccurredAt: time.Now().UTC(), Location: "test place", Severity: "high", Attributes: map[string]any{}})
	if err != nil {
		t.Fatal(err)
	}
	queryService := query.Service{DB: db}
	alerts, err := queryService.ForEvent(ctx, accepted.Event.ID)
	if err != nil {
		t.Fatal(err)
	}
	alertID := ""
	for _, item := range alerts {
		if item.PrimaryRuleID == createdRule.ID {
			alertID = item.ID
		}
	}
	if alertID == "" {
		t.Fatal("alert not created")
	}
	service := Service{DB: db, Audit: audit.Store{DB: db}, Query: queryService}
	const attempts = 20
	var wg sync.WaitGroup
	for i := 0; i < attempts; i++ {
		wg.Add(1)
		actor := actors[i%2]
		go func() { defer wg.Done(); _, _ = service.Ack(ctx, alertID, actor) }()
	}
	wg.Wait()
	detail, err := queryService.Get(ctx, alertID)
	if err != nil {
		t.Fatal(err)
	}
	if detail.Alert.Status != "acknowledged" || detail.Alert.AcknowledgedBy == nil {
		t.Fatalf("unexpected ack projection: %#v", detail.Alert)
	}
	var facts int
	if err := db.QueryRow(ctx, `SELECT count(*) FROM bell_alert_lifecycle_facts WHERE alert_id=$1 AND transition='acknowledged'`, alertID).Scan(&facts); err != nil {
		t.Fatal(err)
	}
	if facts != 1 {
		t.Fatalf("ack facts=%d", facts)
	}
	winner := actors[0]
	loser := actors[1]
	if winner.ID != *detail.Alert.AcknowledgedBy {
		winner, loser = loser, winner
	}
	if _, err := service.Close(ctx, alertID, CloseInput{Outcome: "site_normal"}, loser); err == nil {
		t.Fatal("non-owner closed alert")
	}
	closed, err := service.Close(ctx, alertID, CloseInput{Outcome: "site_normal", Note: "现场检查正常"}, winner)
	if err != nil {
		t.Fatal(err)
	}
	if closed.Detail.Alert.Status != "closed" {
		t.Fatal("alert not closed")
	}
	repeat, err := service.Close(ctx, alertID, CloseInput{Outcome: "site_normal", Note: "现场检查正常"}, winner)
	if err != nil || !repeat.Idempotent {
		t.Fatalf("repeat close: %#v %v", repeat, err)
	}
	if err := db.QueryRow(ctx, `SELECT count(*) FROM bell_alert_lifecycle_facts WHERE alert_id=$1`, alertID).Scan(&facts); err != nil {
		t.Fatal(err)
	}
	if facts != 2 {
		t.Fatalf("lifecycle facts=%d", facts)
	}
	if _, err := db.Exec(ctx, `DELETE FROM bell_alert_lifecycle_facts WHERE alert_id=$1`, alertID); err == nil {
		t.Fatal("lifecycle deletion succeeded")
	}
	db.Close()
	reopened, err := platform.OpenDatabase(ctx, url)
	if err != nil {
		t.Fatal(err)
	}
	defer reopened.Close()
	afterRestart, err := (query.Service{DB: reopened}).Get(ctx, alertID)
	if err != nil {
		t.Fatal(err)
	}
	if afterRestart.Alert.Status != "closed" || len(afterRestart.Timeline) != 2 {
		t.Fatalf("restart projection: %#v", afterRestart)
	}
}