最近在分析SQL Server的死锁时,发现一个比较有意思的现象,发现死锁当中一个会话的隔离级别为序列化(Serializable),这个是让人比较奇怪的地方,我们知道SQL Server数据库的默认隔离级别为已提交读(READ COMMITTED),除非人为设置事务隔离级别(TRANSACTION ISOLATION LEVEL),否则事务隔离级别会使用数据库的默认隔离级别。在分析了死锁相关的存储过程后,没有发现有人为修改事务隔离级别的地方。在分析过后,我们判断应该是在应用程序代码里面有设置隔离级别,下面我们通过一个小实验来构造这样的一个案例。
测试环境数据库为AdventureWorks2014,如下所示,我简单写了一点C#代码,截取黏贴部分C#代码在此,在这段代码中,我们使用TransactionScope,我们先更新Sales.SalesOrderDetail,然后查询 [Sales].[SalesOrderHeader]的相关数据来绑定Grid控件
try
{
using (TransactionScope scope = new TransactionScope())
{
using (SqlConnection conn = new SqlConnection(connString))
{
string cmdText = "UPDATE Sales.SalesOrderDetail SET OrderQty=2 WHERE SalesOrderID=43659 AND SalesOrderDetailID=1;";
SqlCommand cmd = new SqlCommand(cmdText, conn);
conn.Open();
cmd.ExecuteNonQuery();
}
using (SqlConnection conn = new SqlConnection(connString))
{
DataSet sqldataset = new DataSet();
string cmdText = "SELECT * FROM [Sales].[SalesOrderHeader] WHERE SalesOrderID=43659;";
SqlCommand cmd = new SqlCommand(cmdText, conn);
SqlDataAdapter sqladapter = new SqlDataAdapter(cmdText, conn);
sqladapter.Fill(sqldataset, "spt_values");
gvData.DataSource = sqldataset;
gvData.DataBind();
}
scope.Complete();
}
}
catch (TransactionAbortedException exc)
{
log.Error("错误", exc);
}
然后另外一个会话,就直接用SSMS开启一个事务(懒得构造C#代码案例,主要是太浪费时间了),主要执行下面逻辑:
BEGIN TRAN
UPDATE [Sales].[SalesOrderHeader] SET SubTotal = SubTotal + 10
WHERE SalesOrderID=43659;
WAITFOR DELAY '00:00:10';
SELECT TOP 10 * FROM Sales.SalesOrderDetail
--ROLLBACK TRAN;
执行上面SQL语句,然后运行最上面C#代码,立马就能构造出一个死锁案例,如下截图所示,测试环境为SQL Server 2014,我就使用扩展事件system_health捕获的死锁(当然,你可以使用任何方式,例如Profile或Trace捕获死锁相关信息),使用SQL将死锁的XML信息查出
如下所示,你会看到使用TransactionScope的会话的隔离级别为isolationlevel="serializable (4)", 具体可以参考下面死锁的XML文件。
<deadlock>
<victim-list>
<victimProcess id="process17676e108" />
</victim-list>
<process-list>
<process id="process17676e108" taskpriority="0" logused="384" waitresource="KEY: 7:72057594048479232 (0ca7b7436f59)" waittime="379" ownerId="46635671" transactionname="user_transaction" lasttranstarted="2019-04-02T23:26:21.150" XDES="0x17f0511f0" lockMode="S" schedulerid="1" kpid="13440" status="suspended" spid="61" sbid="0" ecid="0" priority="0" trancount="1" lastbatchstarted="2019-04-02T23:26:21.147" lastbatchcompleted="2019-04-02T23:26:09.343" lastattention="1900-01-01T00:00:00.343" clientapp="Microsoft SQL Server Management Studio - Query" hostname="MyNB00021" hostpid="9728" loginname="test" isolationlevel="read committed (2)" xactid="46635671" currentdb="7" lockTimeout="4294967295" clientoption1="671090784" clientoption2="390200">
<executionStack>
<frame procname="adhoc" line="8" stmtstart="282" stmtend="368" sqlhandle="0x020000002a285923f5e38f7347b53337195c56a4a1bc33080000000000000000000000000000000000000000">
unknown </frame>
</executionStack>
<inputbuf>
BEGIN TRAN
UPDATE [Sales].[SalesOrderHeader] SET SubTotal = SubTotal + 10
WHERE SalesOrderID=43659;
WAITFOR DELAY '00:00:10';
SELECT TOP 10 * FROM Sales.SalesOrderDetail </inputbuf>
</process>
<process id="process175603c28" taskpriority="0" logused="436" waitresource="KEY: 7:72057594048544768 (6a8a6db47ef5)" waittime="4420" ownerId="46635065" transactionname="user_transaction" lasttranstarted="2019-04-02T23:25:36.807" XDES="0x1762fa9f0" lockMode="S" schedulerid="1" kpid="51760" status="suspended" spid="63" sbid="0" ecid="0" priority="0" trancount="2" lastbatchstarted="2019-04-02T23:26:26.450" lastbatchcompleted="2019-04-02T23:25:36.807" lastattention="1900-01-01T00:00:00.807" clientapp=".Net SqlClient Data Provider" hostname="MyNB00021" hostpid="1700" loginname="kkk" isolationlevel="serializable (4)" xactid="46635065" currentdb="7" lockTimeout="4294967295" clientoption1="673316896" clientoption2="128056">
<executionStack>
<frame procname="AdventureWorks2014.Sales.iduSalesOrderDetail" line="18" stmtstart="982" stmtend="2448" sqlhandle="0x0300070076146e6c18e00a016ba3000000000000000000000000000000000000000000000000000000000000">
INSERT INTO [Production].[TransactionHistory]
([ProductID]
,[ReferenceOrderID]
,[ReferenceOrderLineID]
,[TransactionType]
,[TransactionDate]
,[Quantity]
,[ActualCost])
SELECT
inserted.[ProductID]
,inserted.[SalesOrderID]
,inserted.[SalesOrderDetailID]
,'S'
,GETDATE()
,inserted.[OrderQty]
,inserted.[UnitPrice]
FROM inserted
INNER JOIN [Sales].[SalesOrderHeader]
ON inserted.[SalesOrderID] = [Sales].[SalesOrderHeader].[SalesOrderID </frame>
<frame procname="adhoc" line="1" stmtstart="52" stmtend="262" sqlhandle="0x02000000abf4ee0ff24fea415c6f35709c721203030a173b0000000000000000000000000000000000000000">
unknown </frame>
<frame procname="adhoc" line="1" stmtend="186" sqlhandle="0x02000000b0cd40243d43ed1a51b1baa9cbf70d1628eae7880000000000000000000000000000000000000000">
unknown </frame>
</executionStack>
<inputbuf>
UPDATE Sales.SalesOrderDetail SET OrderQty=2 WHERE SalesOrderID=43659 AND SalesOrderDetailID=1; </inputbuf>
</process>
</process-list>
<resource-list>
<keylock hobtid="72057594048479232" dbid="7" objectname="AdventureWorks2014.Sales.SalesOrderDetail" indexname="PK_SalesOrderDetail_SalesOrderID_SalesOrderDetailID_old" id="lock154ffb300" mode="X" associatedObjectId="72057594048479232">
<owner-list>
<owner id="process175603c28" mode="X" />
</owner-list>
<waiter-list>
<waiter id="process17676e108" mode="S" requestType="wait" />
</waiter-list>
</keylock>
<keylock hobtid="72057594048544768" dbid="7" objectname="AdventureWorks2014.Sales.SalesOrderHeader" indexname="PK_SalesOrderHeader_SalesOrderID" id="lock155a8fa00" mode="X" associatedObjectId="72057594048544768">
<owner-list>
<owner id="process17676e108" mode="X" />
</owner-list>
<waiter-list>
<waiter id="process175603c28" mode="S" requestType="wait" />
</waiter-list>
</keylock>
</resource-list>
</deadlock>
我们也可以使用下面SQL语句来捕获会话的的隔离级别(根据实际情况调整), 在运行上面C#代码期间捕获会话信息,如下截图所示:
DECLARE @end_time DATETIME;
SET @end_time = DATEADD(SECOND, 10, GETDATE());
WHILE GETDATE() < @end_time
BEGIN
INSERT INTO mintor_isolcation_level
SELECT session_id ,
start_time ,
status ,
total_elapsed_time ,
CASE transaction_isolation_level
WHEN 1 THEN 'ReadUncomitted'
WHEN 2 THEN 'ReadCommitted'
WHEN 3 THEN 'Repeatable'
WHEN 4 THEN 'Serializable'
WHEN 5 THEN 'Snapshot'
ELSE 'Unspecified'
END AS transaction_isolation_level ,
sh.text ,
ph.query_plan
FROM sys.dm_exec_requests
CROSS APPLY sys.dm_exec_sql_text(sql_handle) sh
CROSS APPLY sys.dm_exec_query_plan(plan_handle) ph
END
因为上面的脚本执行时间太短,所以有可能捕获到的是相关SQL运行期间的触发器脚本。如果要清晰的捕获相关SQL,可以构造一个执行时间较长的SQL
是否有点意外,其实官方文档已有详细介绍(详见参考资料),摘抄部分信息如下,TransactionScope如果不指定隔离级别,默认情况下,事务隔离级别为Serializable
设置 TransactionScope 隔离级别
除超时值之外,TransactionScope 的有些重载构造函数还接受 TransactionOptions 类型的结构,用于指定隔离级别。 默认情况下,事务在隔离级别设置为 Serializable 的情况下执行。 通常对频繁执行读取的系统选择 Serializable 之外的隔离级别。 这需要全面地了解事务处理理论、事务本身的语义、所涉及的并发问题以及系统一致性的结果。
总结
这里只是一个案例,仅仅说明应用程序的驱动程序或API函数,有可能会需要(或默认)设定事务的隔离级别,这个一定要当心,避免由于人为失误导(不了解技术细节)致不小心提高事务隔离级别,造成不必要的死锁出现。另外,这里总结这篇文章,也仅仅是对这种案例感到有意思。
参考资料:
https://docs.microsoft.com/en-us/dotnet/framework/data/transactions/implementing-an-implicit-transaction-using-transaction-scope