Hi,

I am fairly new to Hibernate Validator and have been trying it out over the last couple of days. One thing I noticed is that the first run of validations is very slow. For example, the first time I ran a set of validations it took just shy of 7 seconds. I then reran the same set of validations on the same exact data set and the second run took only 15 ms. Why the disparity? I'm guessing there's some caching going on so the second run is really fast. Are there any settings I can use that would help the first run perform better?

Hi,

a difference in validation time is expected, but your case sounds like a lot. How many validations are you executing? How many different beans and how many constraints per bean. I would love to get more information.

If you are validating many different beans, the validation time for the first validation is expected to be slower. The metadata for this particular bean time needs to be created and as you say it will be cached.

Most important performance recommendation is to not throw away the ValidatorFactory. The cache works per factory. Validator instances are lightweight and can be easily re-created.

Depending on your usecase you might want to consider some pre-warming of the caches, but there are no other special options to perform faster. Depending on your usecase and if you have very deep object graphs you could try the failFast which is specific to Hibernate Validator and will stop the validation on the first validation error.

--Hardy

Thank you for the info.

I'm executing roughly 20 validations but the object graph is deep and a few of those validations are on List objects so those validations are iterating through those lists.

Since this is part of a web application, I'm thinking I could build the factory and fire off the validations on app startup just to get that initial "slow" validation run out of the way. Is this a good idea?

Sounds reasonable. Can you give a order of magnitude of the elements in the list or maybe the number of validations? Also how many different bean types are involved?

The list lengths should be fairly small averaging about five to ten objects in the list. There is only one validation being performed on each of the lists right now with the potential for more.

As far as bean types go, our application has a fairly complex XML structure that we unmarshall into JAXB objects. The object graph can get quite deep with roughly 15 nodes being the deepest. Because we have so many different elements within our XML structure there are quite a few bean types as well.

Hopefully that's the kind of info you're looking for.