Incrementally decreasing performance in EJB3

[R van Vliet]

Hi guys,

I've run into an odd issue. I have a method that persists a rather complicated entity object (multiple OneToOne, OneToMany and ManyToMany relationships to other entities, all of which fetchtype.LAZY, no cascading).

Now, if I call that method 1000 times in a row it gets increasingly slow. Is there any known issue with EJB3 entity persistence that I should be aware of? I've read some things about FlushMode.AUTO being the issue but that seemed to be JBoss specific.

Any input would be much appreciated.

[Jimmy Clark]

Now, if I call that method 1000 times in a row it gets increasingly slow.

Post the code of the method you are calling.

Please explain how you call the method 1000 times. How long does the method take to execute on the fifth time? How long does it take on the 1000th time?
[ July 16, 2008: Message edited by: James Clark ]

[Rabin Rath]

It is not a good idea to fetch all the 1000 records and store it in to a HttpSession object. if you are doing like this the definitely you will get the performance issue.
You can do

1. restrict the number and size of the retrieving data.
2. suppose you restrict it to 100 but you need 1000 then you have to hit the data base 10 times.
3. use refersh() after each flush() method.
4. Use some valuable pool size also.

[R van Vliet]

Hi guys,

Thanks for the replies, I was ill the past week hence the slow reaction.

The code of the method is something like this (cant paste the actual code, NDA etc.) :

   
MyEntity myEntity = new MyEntity();
 
myEntity.setA(a);
myEntity.setB(b);
...
myEntity.setZ(z);
 
em.persist(myEntity);

So, extremely basic, there are 4 @OneToOne, 1 @OneToMany and 2 @ManyToMany annotions present in the entity, all correctly mapped, no cascading, all lazy.

Performance of the method goes up slowly :

    
#1:call took 13ms
#2:call took 12ms
#3:call took 12ms
#4:call took 14ms
#5:call took 13ms
...
#101:call took 34ms
#102:call took 31ms
...
#501:call took 109ms
#502:call took 112ms
...
#999:call took 239ms

The call is done in a for loop within a unit test. I really think it has something to do with flushmodes. If I use COMMIT instead of AUTO the problem is gone, but I'd still like to know why this happens.