Hi all,

I recently join a new application team developpers and noticed some strange behaviors about the persistence layer of their application.
Is there any reason or any particular hibernate configuration that explains the more a simple (w\o any eargly fetched data) select query is repeatedly executed the more time is execution takes.
Here is a example of HQL query :


and the issuing SQL query :



In a test when I place this query into a loop from 1 to 6500 iterations and add timers its execution time is multiplied by 4 between the 1000 first iterations and the 1000 last ones (the average execution time increase from 7ms to 28ms).
The test is executed with a single session and a single transaction.
The query parameters are chosen to return any object so it should not be a matter of objects instanciation.
I already tried the same test with a direct JDBC connection resulting with a negligible execution time so this is not a database (Oracle) issue.
We also use a C3P0 connections pool that I tried to bypass but the result is the same.

This is probably due to a bad configuration but I can't figure it out.

Many thanks in advance for your highlights.

If your query is selecting a different object for each iteration the problem is probably that you are filling up the first-level cache in Hibernate. While this is not a problem by itself, it may result in a performance penalty due to the automatic dirty-checking of entities. The default setting is that Hibernate do this before executing a query (FlushMode.AUTO), so for each iteration more and more entities are dirty-checked. There are basically two ways to improve this.

* Use a different flush mode, for example, FlushMode.COMMIT. The flush mode is set on the session.
* Evict entities from the session at regular intervals.

Hi Nordborg,

This is the case, each query select a different object and the 1st level cache is growing up quickly.
As you mentioned, after switching the AUTO default session's flush mode to COMMIT this overhead vanished.

I've been working with Hibernate for some time now but I've never been told about this problem yet.
Many thanks to figure it out.

No I've to check that within the same transaction there is nowhere some code that expect the changes performed to managed objects being automatically synchronized with the database. In such a case I think I will have to flush them manually to the database.