sql-server - NHibernate下SQL Server死锁

Question

我有一个多线程应用程序，它适用于大型数据库(文件大小 >1 Gb，数据库有 38 个表，每个表有超过 500 K 的实体)。它使用 城堡 3.1.0.0 , NHibernate 3.3.1.4000 , FluentNibernate 1.3.0.733 , SQL Server 2012 .

NHibernate 配置如下:

config.SetProperty(Environment.CommandTimeout, "300");
config.SetProperty(Environment.BatchSize, "0"); 
config.SetProperty(Environment.GenerateStatistics, "true");
config.SetProperty(Environment.ReleaseConnections, "auto");
config.SetProperty(Environment.UseQueryCache, "true");
config.SetProperty(Environment.SqlExceptionConverter, typeof(MsSqlExceptionConverter).AssemblyQualifiedName);
//...
.MaxFetchDepth(1)

我用 每个线程一个 session (CaSTLe.Windsor) 和 做空交易 .每个数据库更新、保存、删除过程都被代码锁定:

public abstract class BaseEntityRepository<T, TId> : IBaseEntityRepository<T, TId> where T : BaseEntity<TId> {
    protected static readonly object Locker = new object();

    public bool Save(T item) {
        bool result = false;

        if ((item != null) && (item.IsTransient())) {
            lock (Locker) {
                using (ITransaction tr = Session.BeginTransaction()) {
                    try {                       
                        Session.Save(item);
                        if ((tr.IsActive) && (!tr.WasCommitted) && (!tr.WasRolledBack))
                            tr.Commit();  
                        result = true;
                    } catch {
                        if ((tr.IsActive) && (!tr.WasCommitted) && (!tr.WasRolledBack))
                            tr.Rollback();
                        Session.Clear();
                        throw;
                    }
                }
            }
        }
        return result;
    }

    //same for delete and update 

    public T Get(TId itemId) {
        T result = default(T);

        try {
            result = Session.Get<T>(itemId);
        } catch {
            throw;
        }
        return result;
    }

    public IList<T> Find(Expression<Func<T, bool>> predicate) {
        IList<T> result = new List<T>();
        try {
            result = Session.Query<T>().Where(predicate).ToList();
        } catch {
            throw;
        }
        return result;
    }
}

在我将 3 个数据库(每个 300-400 Mb)合并到一个大数据库(如上所述)之前，一切正常。我通过 Microsoft SQL Server Management Studio 导出/导入数据向导合并了数据库。合并后，我设置了主键，并为 设置了身份规范。身份证 列。在这之后我得到了很多 SQL Server 错误(此错误未重现，它们不时出现):

Transaction (Process ID X) was deadlocked on lock resources with another process and has been chosen as the deadlock victim. Rerun the transaction.

所以我读了 this article , this .我完成了所有步骤，发现两个线程在同一个关联对象 ID 上执行相同的页锁。我什至不明白这是怎么回事，因为我以编程方式锁定了每个实体更新，所以它甚至不可能是这样的问题。这就是我将数据库更改为快照隔离的原因:

ALTER DATABASE MY_DB SET ALLOW_SNAPSHOT_ISOLATION ON

ALTER DATABASE MY_DB SET READ_COMMITTED_SNAPSHOT ON

我也改变了 NHibernate 隔离级别 来自 已提交阅读 至 快照 来解决这个问题。在此之后，我收到了以下错误:

Could not execute command: UPDATE [Foo] SET Field1 = @p0, Field2= @p1 WHERE Id = @p2
System.Data.SqlClient.SqlException (0x80131904): Snapshot isolation transaction aborted due to update conflict. You cannot use snapshot isolation to access table 'dbo.Bar' directly or indirectly in database 'My_DB' to update, delete, or insert the row that has been modified or deleted by another transaction. Retry the transaction or change the isolation level for the update/delete statement.

in System.Data.SqlClient.SqlConnection.OnError(SqlException exception, Boolean breakConnection)
in System.Data.SqlClient.TdsParser.ThrowExceptionAndWarning()
in System.Data.SqlClient.TdsParser.Run(RunBehavior runBehavior, SqlCommand cmdHandler, SqlDataReader dataStream, BulkCopySimpleResultSet bulkCopyHandler, TdsParserStateObject stateObj)
in System.Data.SqlClient.SqlCommand.FinishExecuteReader(SqlDataReader ds, RunBehavior runBehavior, String resetOptionsString)
in System.Data.SqlClient.SqlCommand.RunExecuteReaderTds(CommandBehavior cmdBehavior, RunBehavior runBehavior, Boolean returnStream, Boolean async)
in System.Data.SqlClient.SqlCommand.RunExecuteReader(CommandBehavior cmdBehavior, RunBehavior runBehavior, Boolean returnStream, String method, DbAsyncResult result)
in System.Data.SqlClient.SqlCommand.InternalExecuteNonQuery(DbAsyncResult result, String methodName, Boolean sendToPipe)
in System.Data.SqlClient.SqlCommand.ExecuteNonQuery()
in NHibernate.AdoNet.AbstractBatcher.ExecuteNonQuery(IDbCommand cmd)

Could not synchronize database state with session NHibernate.StaleObjectStateException: Row was updated or deleted by another transaction (or unsaved-value mapping was incorrect): [Foo#544353]

in NHibernate.Persister.Entity.AbstractEntityPersister.Update(Object id, Object[] fields, Object[] oldFields, Object rowId, Boolean[] includeProperty, Int32 j, Object oldVersion, Object obj, SqlCommandInfo sql, ISessionImplementor session)
in NHibernate.Persister.Entity.AbstractEntityPersister.UpdateOrInsert(Object id, Object[] fields, Object[] oldFields, Object rowId, Boolean[] includeProperty, Int32 j, Object oldVersion, Object obj, SqlCommandInfo sql, ISessionImplementor session)
in NHibernate.Persister.Entity.AbstractEntityPersister.Update(Object id, Object[] fields, Int32[] dirtyFields, Boolean hasDirtyCollection, Object[] oldFields, Object oldVersion, Object obj, Object rowId, ISessionImplementor session)
in NHibernate.Action.EntityUpdateAction.Execute()
in NHibernate.Engine.ActionQueue.Execute(IExecutable executable)
in NHibernate.Engine.ActionQueue.ExecuteActions(IList list)
in NHibernate.Engine.ActionQueue.ExecuteActions()
in NHibernate.Event.Default.AbstractFlushingEventListener.PerformExecutions(IEventSource session)

NHibernate.TransactionException: Transaction not connected, or was disconnected

in NHibernate.Transaction.AdoTransaction.CheckNotZombied()
in NHibernate.Transaction.AdoTransaction.Rollback()
in My.DBLayer.Data.Repositories.BaseEntityRepository`2.Update(T item)

我没有 版本 DB 实体的属性，并且我的代码中没有 OptimisticLock.Version() 行，所以我隐式使用 PessimisticLock。我可以加 版本 和 乐观锁 ，但不要认为这会解决问题。

我试着做简单的测试:

Thread t1 = new Thread(m1);
Thread t2 = new Thread(m2);
t1.Start();
t2.Start();

private static void m1() {
    FooRepository rep1 = BootstrapContainer.Instance.Resolve<FooRepository>();
    Foo foo1 = rep1.Get(1);
    foo1.Field1 = "bbbb";
    Thread.Sleep(60*1000);
    rep1.Update(foo1);
}

private static void m2() {
    FooRepository rep2 = BootstrapContainer.Instance.Resolve<FooRepository>();
    Thread.Sleep(5*1000);
    Foo foo2 = rep2.Get(1);
    foo2.Field2 = "aaaaa";
    Thread.Sleep(5*1000);
    rep2.Update(foo2);
}

一切正常，没有任何错误。

为什么我有所有这些错误(我没有更改代码，只是将数据库合并为一个，并且在合并之前一切正常)？如果我使用锁定来防止同时更新不同线程中的实体，为什么会出现此错误。

Answer 1

您可以从查询并行化中获得死锁……也就是说，当将工作拆分为多个并行操作时，单个查询可能会对自身造成死锁。我曾多次遇到这种情况。如果您使用的是 hql，您可以将 OPTION(MAXDOP 1) 添加到您的查询/语句中以查看是否可以解决问题。

或者，您可以将整个服务器设置为 MAXDOP 1...这意味着您永远不会获得并行性( https://msdn.microsoft.com/en-us/library/ms189094.aspx )。如果没有对工作负载进行深入分析，我不会推荐这样做，但它可以帮助您了解并行性是否是死锁的根源。