NHibernate：在同一会话中保存不同类型的对象会破坏批处理

Question

newbie here, sorry if this is an obvious question.

It seems saving different types of objects in the same session breaks batching, cause significant performance drop.

ID generator is set to Increment (as Diego Mijelshon advised, I tried hilo("100"), but unfortunately same issue, Test1() is still about 5 times slower than Test2()):

public class CustomIdConvention : IIdConvention
{
    public void Apply(IIdentityInstance instance)
    {
        instance.GeneratedBy.Increment();
    }
}

AdoNetBatchSize is set to 1000:

MsSqlConfiguration.MsSql2008
.ConnectionString(connectionString)
.AdoNetBatchSize(1000)
.Cache(x => x
    .UseQueryCache()
    .ProviderClass<HashtableCacheProvider>())
.ShowSql();

These are the models:

public class TestClass1
{
    public virtual int Id { get; private set; }
}

public class TestClass2
{
    public virtual int Id { get; private set; }
}

These are the test methods. Test1() takes 62 seconds, Test2() takes only 11 seconds. (as Phill advised, I tried stateless sessions, but unfortunately same issue):

    [TestMethod]
    public void Test1()
    {
        int count = 50 * 1000;
        using (var session = SessionFactory.OpenSession())
        {
            using (var transaction = session.BeginTransaction())
            {
                for (int i = 0; i < count; i++)
                {
                    var x = new TestClass1();
                    var y = new TestClass2();
                    session.Save(x);
                    session.Save(y);
                }
                transaction.Commit();
            }
        }
    }

    [TestMethod]
    public void Test2()
    {
        int count = 50 * 1000;
        using (var session = SessionFactory.OpenSession())
        {
            using (var transaction = session.BeginTransaction())
            {
                for (int i = 0; i < count; i++)
                {
                    var x = new TestClass1();
                    session.Save(x);
                }
                transaction.Commit();
            }
        }
        using (var session = SessionFactory.OpenSession())
        {
            using (var transaction = session.BeginTransaction())
            {
                for (int i = 0; i < count; i++)
                {
                    var y = new TestClass2();
                    session.Save(y);
                }
                transaction.Commit();
            }
        }
    }

Any ideas?

Thanks!

Update:

The test project can be downloaded from here . You need to change the connectionString in the Main method. I changed all sessions to stateless sessions.

My restuls: Test1 = 59.11, Test2 = 7.60, Test3 = 7.72. Test1 is 7.7 times slower than Test2 & Test3!

Answer 1

Do not use increment. It's the worst possible generator.

Try changing it to HiLo.

Update:

It looks like the problem occurs when alternating saves of different entities, regardless of whether the session/transaction are separated or not.

This produces similar results to the second test method:

[TestMethod]
public void Test3()
{
    int count = 50 * 1000;
    using (var session = SessionFactory.OpenSession())
    {
        using (var transaction = session.BeginTransaction())
        {
            for (int i = 0; i < count; i++)
            {
                var x = new TestClass1();
                session.Save(x);
            }
            for (int i = 0; i < count; i++)
            {
                var y = new TestClass2();
                session.Save(y);
            }
            transaction.Commit();
        }
    }
}

My guess, without looking at NH's sources, is that it preserves the order because of possible relationships between the entities, even when there are none.

Answer 2

When you run test2 and test3, the insert's are batched together.

When you run test1, where you alternate the inserts, the inserts are issued as separate statements and are not batched together.

I found this out by profiling all three tests.

So as per Diego's answer, it must preserve the order that you're inserting, and batch them together.

I wrote a 4th test, I set the batch size to 10, then alternated when i changed from TestClass1 to TestClass2 so that I was doing 5 of TestClass1 and then 5 of TestClass2, to hit the batch size.

This pushed out batch's of 5 in the order they were processed.

public void Test4()
{
    int count = 10;
    using (var session = SessionFactory.OpenSession())
    using (var transaction = session.BeginTransaction())
    {
        for (int i = 0; i < count; i++)
        {
            if (i%2 == 0)
            {
                for (int j = 0; j < 5; j++)
                {
                    var x = new TestClass1();
                    session.Save(x);
                }
            }
            else
            {
                for (int j = 0; j < 5; j++)
                {
                    var y = new TestClass2();
                    session.Save(y);
                }
            }
        }
        transaction.Commit();
    }
}

Then I changed it to insert 3 at a time instead of 5. The batch's were in multiples of 3, so what must be happening is the batch size allows a batch of 1 type to go to specified amount, but groups only the same type together. While alternating causes separate insert statements.