FlushMode.AUTO practically unusable - should not flush before EVERY query ------------------------------------------------------------------------- Key: HHH-2399 URL: http://opensource.atlassian.com/projects/hibernate/browse/HHH-2399 Project: Hibernate3 Type: Improvement Versions: 3.1.3, 3.2.1, 3.2.2, 3.2.0.ga Environment: PostgreSQL 8.1.5, JDBC 2 and 3, Linux kernel 2.4.x, 2.6.x, Java 1.5.0_09 Reporter: Piotr Kołaczkowski Attachments: hbtest.zip The method flushIfRequired seems to flush before every query execution, even though there wasn't anything changed to the objects in the session, and the objects in the session are of different class than the object(s) queried. When having more than a few such objects in a session, it may lead to large, unneeded overheads (I noticed 10-200 times slowdown on simple queries). Our system uses many fast queries, that uasually return only one result or few results. The average query durations on different levels, measured by System.nanoTime() (see the attachment): - database level: 0.05 ms - pure JDBC code, prepared statements: 0.2 ms - pure JDBC code, no prepared statements: 0.7 ms - not cached hibernate query, no other objects associated with the session: 1.3 ms - cached hibernate query, object in the session cache, no other objects associated with the session: 1.1 ms - cached hibernate query, 10000 very simple objects of some other class in the session: >50 ms 10000 may seem large, but note these were very simple objects, having only 2 properties. On our production system we have classes with many properties, and even 20 objects in a session can cause a sub-millisecond query to run longer than 5 ms. We switched to FlushMode.COMMIT, and now the performance is much better. Can you do something about this? As far as I've read in JIRA, a similar issue exists in Hibernate 2.x, and is still OPEN / UNRESOLVED. BTW. Why is the 1st level cache so slow in my tests? Why reading just a 2 field object from the database cause so much overhead over pure JDBC? What am I doing wrong? There is so much marketing about using cglib reflection optimizer in the performance FAQ but even if I retrieved these objects using JDBC and reflection, I would achieve much better performance. After all 1 ms is very much time, if getting results from DB and sending them to JVM lasts 0.2 ms. -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://opensource.atlassian.com/projects/hibernate/secure/Administrators.... - For more information on JIRA, see: http://www.atlassian.com/software/jira