The batch-size capability is extremely helpful in avoiding the N+1 select problem. However, NHibernate follows Hibernates semantics such that if the number of candidate entities is less than the batch-size, then the sqrt(batch-size) is attempted. If the number of candidates is still less than this, then a batch-size of 1 is used. Would it be possible to provide an alternate usage (through a config setting) such that the min(candiates, batch-size) is used.

e.g. if batch-size is 10 and 8 candidate entities are found, then a single query with 8 enttity primary keys is used). If 12 are found, then 2 queries are performed with 8 and 2 respectively. This would really give more control to help control the N+1 select problem.

Thanks,
craig

Is there a proper place to list feature requests?

craig

Our JIRA is probably the place, but adding features is very low priority for me right now (if the feature comes in the form of a patch, then the priority could be a bit higher).

Ok Sergey,

I know you are super-busy so I will work on creating a patch for you. I hope to be able to use a simple strategy pattern to accomodate alternate batching schemes.

Thanks again,
craig

It's not that I'm busy, I just don't want to destabilize 1.0.x with new features, and I haven't completely decided yet how the future versions of NH will be done. It's likely that I will just port Hibernate 3, trying to stay as close as possible to the Java version. If this turns out to be the case, then all the feature requests will have to go through Hibernate team first.

in H3, the algorithm has changed to use n-1 = n /2 instead of sqrt(n)
The issue should be minimized.

That makes sense. If I provide a patch, would you still consider including it?

craig

Yes. If the implementation, or at least the algorithm, is similar to H3, I will include it. By the way, I'm going to release 1.0.2 soon since there are only a few open issues left, so hurry up :-)

Hello Sergey,

Sorry for the late reply. I have been ill with the Flu for a couple of days. I plan on creating the patch tomorrow. Hopefully it is not too late.

thanks,
craig

No-no, it's not. Plenty of time left :)

Hello Sergey,

I have been looking through the Hibernate 3.1 code to try and match its sementics, but the NHibernate port differs enough, that I am not comfortable (or confident) with trying to match its semantics of the batching loaders and persisters. Since I am not familiar with the NHibernate porting decisions, I am not sure if the differences are irrelevant. In addition, the Hibernate 3.1 does not provide the ability to provide alternate batching strategies and that is the main thing I was going to add. Do you think it is valuable to provide the ability to plug in different batching strategies, or do you want NHibernate to use the new batching strategy in NHibernate 3.1 without providing the ability to change it.

thanks,
craig

It's enough to just change the current strategy to the one in Hibernate 3.1.