| [CDO] How do Passive updates work ? [message #1695074] |
Mon, 11 May 2015 20:09  |
 Stephane fournier
Messages: 340
Registered: July 2009 |
Senior Member |
|
|
Hi there,
I would like to get some more detailed explanations about Passive Updates.
Let me introduce a brief use case:
- 2 client JVM (A, B running on different computers) connected to a same CDO server (running on another computer) accessing a resource containing data based on the famous Company ecore model.
- Let's say that JVM A, on computer "One", commits the fact a new Employee is created and added in a containment relationship of a company. After the transaction commit operation call, JVM A uses RMI to remotely calls JVM B (in RMI threads) passing the name of the employee (for instance).
- On JVM B side, RMI implementation tries to lookup the Employee from its name.
Depending on how the Passive updates are quick or not, JVM B find or not the previously added Employee. Indeed, RMI call handling is done in another thread, and on JVM B side, it might happen the update of the CDO transaction happens after the RMI call.
If I add a sleep of 200ms after the commit operation and before the RMI call, I never faced the issue. But usage of sleep() is not a proper solution always difficult to find the right value.
It seems to me, Passive updates, to call back remote clients, are NOT done during the commit itself, but in reaction of the end of the commit on server side.
Is it right ?
To recap : when client A calls transaction.commit(); and commit() steps out, we cannot make sure other remote clients are up-to-date immediately (even with passive update set to true) ?
If'm right, how can I make sure when I commit something in CDO every remote clients are up-to-date ?
I perfectly know I can register listener (CDO ones or EMF adapters) on the model stored in the server to be notified, but for other reasons I need to use RMI and pass attribute values to identify objects to lookup on other side.
Does anyone have an idea how to solve that integration (between RMI and CDO) issue ?
Thanks in adavance,
Stéphane.
|
|
|
|
| Re: [CDO] How do Passive updates work ? [message #1695098 is a reply to message #1695079] |
Tue, 12 May 2015 05:24  |
|
Hey Guys,
Alexandre, you gave the perfect answer. Instead of CDOTransaction.getLastCommitTime() you can also use
CDOTransaction.commit().getTimeStamp(), which is a little safer when multiple threads commit the transaction.
Cheers
/Eike
----
http://www.esc-net.de
http://thegordian.blogspot.com
http://twitter.com/eikestepper
Am 11.05.2015 um 22:52 schrieb Alexandre Borgoltz:
> I'd say it's up to the client to ensure they are "up-to-date" when they work with the model.
>
> One way you could do is the following:
>
>
> JVM A : perform your commit
> JVM A : retreive the corresponding repository's timestamp by CDOTransaction.getLastCommitTime()
> JVM A : include that timestamp as an additional parameter to your RMI call together with the employee's name
> JVM B : in the method called through RMI, call CDOView.waitForUpdate(timestamp) passing the timestamp received by RMI
> JVM B : use your model, it is up-to-date (at least enough so that the employee is there!)
>
>
> I hope it helps!
>
Cheers
/Eike
----
http://www.esc-net.de
http://thegordian.blogspot.com
http://twitter.com/eikestepper
|
|
|
|
|
|
Powered by
FUDForum. Page generated in 0.03504 seconds
] 
Search
Help
Register
Login
Home
