openjpa-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Michael Dick (JIRA)" <j...@apache.org>
Subject [jira] Created: (OPENJPA-1150) Add WriteBehind mode to cache updates before flushing to backing store
Date Tue, 30 Jun 2009 20:10:48 GMT
Add WriteBehind mode to cache updates before flushing to backing store
----------------------------------------------------------------------

                 Key: OPENJPA-1150
                 URL: https://issues.apache.org/jira/browse/OPENJPA-1150
             Project: OpenJPA
          Issue Type: New Feature
          Components: kernel
    Affects Versions: 2.0.0
            Reporter: Michael Dick
            Assignee: Michael Dick
             Fix For: 2.0.0


The current L2 / DataCache implementation is exhibits write-through behavior. When an entity
is updated its contents are updated in the datacache and then flushed to the backing store
when the transaction completes. 

This feature will enable write-behind behavior - instead of updating the backing store when
the transaction completes, the updates will be cached in memory until until triggered to truly
flush. This approach has the effect of reducing the load on the backing store and (hopefully)
improving overall throughput. 

Naturally there are some caveats to this approach. 

1. A DataCache must be used in conjunction with the WriteBehind cache. The DataCache will
be kept in sync with the WriteBehind cache reducing the possibility of stale reads. 
2. When operating in write-behind mode you need to take care that OpenJPA is the only application
accessing these rows. Updates from other applications may lead to errors when the write-behind
cache is flushed. 
2a. Some care must be taken when running on multiple JVMs. The DataCache & WriteBehindCache
need to be kept in sync across JVMs. The DataCache has some capability in this area and the
WriteBehindCache will further extend that. 
2b. The application must contain some logic to recover from a failed update. 
3. Database generated columns (ie IDENTITY, SEQUENCE) will not work correctly (at least not
with the initial version of write-behind)
3a. Table generated columns will work, but the SQL used to obtain the value from the table
will bypass the WriteBehindCache (ie you may see some database traffic in this case). 

This support may not be suitable for every application. Write heavy applications will see
great benefit as will applications which place high load on the database. The feature should
not be enabled by default and should be considered strictly optional. 

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message