ODP .NET表现得很奇怪

Question

应用程序本身在我之前的问题中描述.在DAL方面我使用

Oracle.ManagedDataAccess, Version=4.121.1.0, Culture=neutral, PublicKeyToken=89b483f429c47342

这是连接字符串:

User id=usr;Password=pwd;Data Source=database1;Validate connection=True;Connection timeout=8;Pooling=false

问题是有时ODP会在异常后引发:

Oracle.ManagedDataAccess.Client.OracleException (0xFFFFFC18): Connection request timed out
   in OracleInternal.ConnectionPool.PoolManager`3.CreateNewPR(Int32 reqCount, Boolean bForPoolPopulation, ConnectionString csWithDiffOrNewPwd, String instanceName)
   in OracleInternal.ConnectionPool.PoolManager`3.Get(ConnectionString csWithDiffOrNewPwd, Boolean bGetForApp, String affinityInstanceName, Boolean bForceMatch)
   in OracleInternal.ConnectionPool.OraclePoolManager.Get(ConnectionString csWithNewPassword, Boolean bGetForApp, String affinityInstanceName, Boolean bForceMatch)
   in OracleInternal.ConnectionPool.OracleConnectionDispenser`3.Get(ConnectionString cs, PM conPM, ConnectionString pmCS, SecureString securedPassword, SecureString securedProxyPassword)
   in Oracle.ManagedDataAccess.Client.OracleConnection.Open()
   in MySyncApp.DBRepository.GetChangedDataDB(DateTime startPeriod) in D:\MySyncApp\MySyncApp\DB.cs:line 23
   in MySyncApp.Program.<>c__DisplayClass30.<>c__DisplayClass32.<Synchronize>b__2f(ID id) in D:\MySyncApp\MySyncApp\Program.cs:line 441

但是在这个异常之后,当我查看Oracle中的会话时,我发现实际上连接正在存在并且只是标记为INACTIVE!因此,此类连接将继续挂在服务器端,逐渐耗尽可用会话的数量.

我的代码没有什么特别的,只是

public List<DataObj> GetChangedDataDB(DateTime startPeriod)
{
    List<DataObj> list = new List<DataObj>();
    using (OracleConnection conn = new OracleConnection(this._connstr))
    {
        conn.Open();

        using (OracleCommand comm = new OracleCommand("select data from table(usr.syncpackage.GetChanged(:pStart))", conn))
        {
            comm.CommandTimeout = 10;
            comm.Parameters.Add(":pStart", startPeriod);

            using (OracleDataReader reader = comm.ExecuteReader())
            {
                // ..omitted
            }
        }

    }
    return list;
}

此代码Parallel.ForEach循环运行,可同时从大量数据库中提取数据.甚至可能是与同一数据库的三个并行连接(从模式的不同部分提取数据,例如,从三个不同的企业部门提取).

甲骨文是

Oracle Database 11g企业版11.2.0.3.0版 - 64位生产

同步过程本身以10秒的间隔在计时器上触发.如果已经有运行任务,则下一个任务正在停止:

    public static void Synchronize(object obj)
    {            
        // ... omitted
        log.Info("ITERATION_COMMON_START");

        if (Program.State == "Running")
        {
            log.Info("ITERATION_COMMON_END_BY_EXISTING");
            return;
        }

        lock (Program.StateLock)
        {
            Program.State = "Running";
        }                   

        Parallel.ForEach(Global.config.dbs, new ParallelOptions { MaxDegreeOfParallelism = -1 }, (l) =>
        {
            Console.WriteLine("Started synchronization for {0}", l.key);
            DBRepository db = new DBRepository(l.connectionString);

            Parallel.ForEach(l.departments, new ParallelOptions { MaxDegreeOfParallelism = -1 }, (department) =>
            {                       
                DateTime ChangesFromTS = GetPreviousIterationTS;
                List<DataObj> cdata = db.GetChangedDataDB(ChangesFromTS);
                // ... doing the work here
            }

        }

        // Finishing work

        GC.Collect();            

        lock (Program.StateLock)
        {
            Program.State = "";
        }

    }

这是定期调用同步任务的计时器:

Program.getModifiedDataTimer = new Timer(Program.Synchronize, null, (int)Global.config.syncModifiedInterval * 1000, (int)Global.config.syncModifiedInterval * 1000);

Global.config.syncModifiedInterval 是几秒钟

当我打开池时,ODP的行为方式与此相同.Max pool size在具有相同异常的连接字符串中,它创建了比指令允许的更多连接.

请告诉我你对这些东西的想法和经验.

---

UPDATE

以下是引发异常时的Oracle跟踪:

(PUB) (ERR) OracleConnection.Open() (txnid=n/a) Oracle.ManagedDataAccess.Client.OracleException (0xFFFFFC18): Connection request timed out
   in OracleInternal.ConnectionPool.PoolManager`3.CreateNewPR(Int32 reqCount, Boolean bForPoolPopulation, ConnectionString csWithDiffOrNewPwd, String instanceName)
   in OracleInternal.ConnectionPool.PoolManager`3.Get(ConnectionString csWithDiffOrNewPwd, Boolean bGetForApp, String affinityInstanceName, Boolean bForceMatch)
   in OracleInternal.ConnectionPool.OraclePoolManager.Get(ConnectionString csWithNewPassword, Boolean bGetForApp, String affinityInstanceName, Boolean bForceMatch)
   in OracleInternal.ConnectionPool.OracleConnectionDispenser`3.Get(ConnectionString cs, PM conPM, ConnectionString pmCS, SecureString securedPassword, SecureString securedProxyPassword)
   in Oracle.ManagedDataAccess.Client.OracleConnection.Open()

---

更新#2

似乎这种连接因滞后连接而显示,就像发送oracle连接的请求被发送但其响应被忽略.或者,发送到服务器或从服务器发送的数据在到达目的地时正在被破坏.

即使我关闭了应用程序,连接仍会挂在服务器的会话列表中.当我杀死一个会话时,它会挂在列表中并带有"KILLED"标签.

---

更新#3

这是演示相同问题的演示应用程序.正如我之前所说,它出现在连接不良时,您可以使用WANem模拟器模拟此类连接.这是我用于数据库连接的相同组件.希望得到你的帮助.

Answer 1

您的连接超时非常低，一般为 8 秒，您发出的命令为 10 秒，尝试将其增加到一分钟，看看会发生什么。我不知道你的算法有多繁重，但只要某处的一个线程在不到 8 秒的时间内因查询而失败，你就会得到该异常。

另外，我找不到说明 ODP.net 不是线程安全的文档，但我也找不到任何说明它是线程安全的文档，所以如果没有其他帮助，请尝试此更改：

    Parallel.ForEach(Global.config.dbs, new ParallelOptions { MaxDegreeOfParallelism = -1 }, (l) =>
    {
        Console.WriteLine("Started synchronization for {0}", l.key);

        Parallel.ForEach(l.departments, new ParallelOptions { MaxDegreeOfParallelism = -1 }, (department) =>
        {                       
            // Now local to the executing thread.
            DBRepository db = new DBRepository(l.connectionString);

            DateTime ChangesFromTS = GetPreviousIterationTS;
            List<DataObj> cdata = db.GetChangedDataDB(ChangesFromTS);
            // ... doing the work here
        }

    }