fix: use READ_COMMITTED_SNAPSHOT on MSSQL (#39512)

MSSQL's default READ COMMITTED makes reads wait on writers, so the
runner pickup deadlocks with concurrent claims, flaking
`TestCreateTaskForRunnerConcurrentClaim`.

- Enable `READ_COMMITTED_SNAPSHOT` on MSSQL so it reads like PostgreSQL
and MySQL
- Read the pickup cursor before claiming, a lost claim could skip
waiting jobs
- Add tests that fail without consistent READ COMMITTED

Performance: Writes on MSSQL now also store the previous row version in
tempdb, the same versioning cost PostgreSQL and MySQL always pay, and
Azure SQL enables it by default. Reads no longer block on writers, and a
32-runner pickup stress test ran 2.5x faster with it.

---------

Signed-off-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: Giteabot <teabot@gitea.io>

models/actions/task.go

@@ -298,6 +298,12 @@ func CreateTaskForRunner(ctx context.Context, runner *ActionRunner) (*ActionTask
 		if err := e.Where(cond).Asc("updated", "id").Limit(pickTaskBatchSize).Find(&jobs); err != nil {
 			return nil, false, err
 		}
+		// A short page means no waiting jobs remain beyond it.
+		isLastPage := len(jobs) < pickTaskBatchSize
+		if !isLastPage {
+			last := jobs[len(jobs)-1] // read before a lost claim bumps Updated
+			cursorUpdated, cursorID = last.Updated, last.ID
+		}
 
 		for _, v := range jobs {
 			if !runner.CanMatchLabels(v.RunsOn) {
@@ -313,12 +319,9 @@ func CreateTaskForRunner(ctx context.Context, runner *ActionRunner) (*ActionTask
 			// Another runner claimed this job concurrently; try the next one.
 		}
 
-		// A short page means no waiting jobs remain beyond it.
-		if len(jobs) < pickTaskBatchSize {
+		if isLastPage {
 			return nil, false, nil
 		}
-		last := jobs[len(jobs)-1]
-		cursorUpdated, cursorID = last.Updated, last.ID
 	}
 }
 

models/db/engine_init.go

@@ -7,6 +7,7 @@ import (
 	"context"
 	"database/sql"
 	"fmt"
+	"time"
 
 	"gitea.dev/modules/log"
 	"gitea.dev/modules/setting"
@@ -109,6 +110,10 @@ func InitEngineWithMigration(ctx context.Context, migrateFunc func(context.Conte
 
 	preprocessDatabaseCollation(xormEngine)
 
+	if setting.Database.Type.IsMSSQL() {
+		enableMSSQLReadCommittedSnapshot(ctx, xormEngine)
+	}
+
 	// We have to run migrateFunc here in case the user is re-running installation on a previously created DB.
 	// If we do not then table schemas will be changed and there will be conflicts when the migrations run properly.
 	//
@@ -131,3 +136,12 @@ func InitEngineWithMigration(ctx context.Context, migrateFunc func(context.Conte
 
 	return nil
 }
+
+// enableMSSQLReadCommittedSnapshot stops MSSQL reads waiting on writers, like PostgreSQL and MySQL
+func enableMSSQLReadCommittedSnapshot(ctx context.Context, engine EngineMigration) {
+	ctx, cancel := context.WithTimeout(ctx, 5*time.Second) // ALTER waits for all other connections to close
+	defer cancel()
+	if _, err := engine.Context(ctx).Exec("IF (SELECT is_read_committed_snapshot_on FROM sys.databases WHERE database_id = DB_ID()) = 0 ALTER DATABASE CURRENT SET READ_COMMITTED_SNAPSHOT ON"); err != nil {
+		log.Error("Unable to set READ_COMMITTED_SNAPSHOT=ON: %v", err)
+	}
+}