Hi! I've a class Product with one-to-one mapping with class Inventory with eager-loading. The inital stockqty of Inventory is 10. One program will add class Inventory.stockqty by 10 and the other program will subtract Inventory.stockqty by 3.

The first and second programs Product object, embedded inside a SSLB, could be loaded right before the operation below:

Assuming the first program run slower than the second program and the second program. The result is Hibernate performed is 20 instead of 17! I tried logging the operations and it seems that the first program's (ran the operation after the second program) Inventory's stockqty value is 10 instead of 7. It seems that the first program's stockqty had been cached? Any suggestions?

Any help is greatly appreciated.

len.

Hibernate caches things. This is good in some circumstances, and perhaps bad in circumstances like yours. I think there are two ways to solve this:

1. You need to either circumvent the cache for this transaction, and use pessimistic locking (assuming the transaction itself (the read/update) doesnt take long... this locks the DB resources)

-or-

2. You can version your objects in the DB (e.g. with a timestamp) so when the update occurs at least an exception occurs so that you know someone else edited the data in the meantime... This is good for GUI's, but perhaps bad for daemons unless you can automatically retry the logic.

Chris

How do I circumvent the cache? It seems that I can't find any APIs to disable the cache upon performing query.

-len

Get the object (using cache), then call Session.refresh(object); There are other ways too, though this seems easy and appropriate.

No refresh method found here. I'm using 1.2.5 with eager-loading. Is eager-loading the problem? My database server is using repeatble read.

-len

I'ved modified my stockqty addition and subtraction, moved them to a dao method with LockMode.UPRADE to perform the before the operation. It still uses the old value.

-len

SessionFactory


But in your place I would disable second-level cache or would use optimististic version locking.

--
Leonid

I tried to turn my database server to serialization isolation and configured my hib-1.2.5 isolation level to 8. Seems to work but I'm not sure if it will have performance impact.

-len

It should have perfomance impact.
In chapter 5 of Hibernate In Action, authours, for the best perfomance, recommend to use version optimistic locking (<version> is mapped) with read commited or repeatable read isolation levels:
int TRANSACTION_READ_COMMITTED = 2; // most suitable
int TRANSACTION_REPEATABLE_READ = 4;
In that chapter they describe this choice.

--
Leonid

How do I disable 2nd level cache?
How do I use version optimistic locking?

I need to build a financial application with heavy updates. The inventory table is very critical in my application especially with multi-user and its running inside an ejb container with SLSB. I need to get an accurate inventory before performing any additions or substractions. I tried turning my db-server to read-committed but I'ved tried running multi-apps and my inventory didnt have accurate values.

-len

2nd level cache is not enabled by default, do not turn on it :)
http://www.hibernate.org/hib_docs/reference/en/html/performance.html#performance-cache
If you need to clear enabled 2nd level cache, use SessionFactory.evict.


http://www.hibernate.org/hib_docs/reference/en/html/mapping.html#mapping-declaration-version
http://www.hibernate.org/hib_docs/reference/en/html/transactions.html#transactions-optimistic

There are a lot of information and suggesions about 2nd level chache, versioning and suitable transaction isolation levels in chapter 5 of "Hibernate In Action".

--
Leonid

1.2.5 API doesnt have evict methods.

If I use session.connection and execute update stockqty=stockqty (+/-) to prevent cached data before writing to db?

If there is a way to write some entities into database bypassing hibernate (bypassing 2nd level cache) in your application, it is strongly recommend to turn off 2nd level cache for those entities. Use 2nd level cache for immutable referenced data if you need to improve your performance. Map version property, and ensure that application that write into database bypassing hibernate increments version and checks it before update.

--
Leonid

My 2nd level cache is turned off by default.