Hi Daniel,
On Tuesday 20 June 2017 06:39 PM, Daniel Fuchs wrote:
On 20/06/2017 12:42, Ujwal Vangapally wrote:
Thanks for the Review Daniel, Harsha.
Yes with this fix JConsole running on JDK 8 won't be able to connect
to a process running on higher version of Java containing this change.
I think even Harsha is agreeing this, his point is that the use case
where a client running on JDK 8 trying to connect to a process
running on latest releases is rare. As mostly for Local Management
both client and server will be running on same machine using same JDK
for starting both client and server .
please correct me if this is not the case.
Well - I believe it could be quite common to use VisualVM running
on JDK 8 to connect to JDK 9 processes. JMX is all about
interoperability so I'd be very reluctant to break this 'by
default'.
Agreed.
can we have this fix with interoperability issue between versions ?
Maybe you could make the fix conditional to a system property being
defined? Also are you sure what you are proposing is the correct fix
for this issue?
It's more of a workaround than a fix as we don't see any other way to
fix this issue.
Note that - if I'm not mistaken - you can work around the original issue
by specifying -Djava.rmi.server.hostname=<address> on the command line.
http://docs.oracle.com/javase/8/docs/technotes/guides/rmi/javarmiproperties.html
This will instruct RMI to put the specified <address> inside its stubs.
The inconvenience is that this is global (for all RMI stubs, just not
only those used by JMX) - but would that be an issue for the use case
mentioned in the bug?
We had the same consideration that setting the RMI property will change
the Interface address for all the stubs which may not be an acceptable
fix in this case.
One possible fix (enhancement) would be to define a JMX system property
that will control the Interface address that goes into RMI stubs. If
that seems like too much work, we can close this issue as 'won't fix' as
we don't have any acceptable fix.
best regards,
-- daniel
-Harsha
Thanks,
Ujwal.
On 6/20/2017 2:40 PM, Daniel Fuchs wrote:
Hi Harsha,
Maybe I'm missing something.
How is the local agent started?
If it's started when you connect jconsole to a process
by specifying the process ID - then I suspect this will
prevent e.g. jconsole or jvisualvm running on JDK 8 to
connect to your process.
Can you verify that it's not the case?
best regards,
-- daniel
On 20/06/2017 09:11, Harsha Wardhana B wrote:
Hi Daniel,
The fix is applicable only to local JMX agent and the most common
use case would be client and server running from same JVM.
It is highly unlikely that local JMX agent will be started to cater
for out-of-jvm clients.
I don't see how introducing this fix can cause new interoperability
problems. Can you please elaborate?
-Harsha
On Monday 19 June 2017 10:23 PM, Daniel Fuchs wrote:
Hi,
If I'm not mistaken then this will make it impossible
for earlier release to interoperate with newer releases
as the LocalRMIClientSocketFactory class will not be
present the client tries to deserialize the stub.
best regards,
-- daniel
On 19/06/2017 11:52, Ujwal Vangapally wrote:
Hi,
Kindly review the fix for bug below
https://bugs.openjdk.java.net/browse/JDK-8173180
webrev:
http://cr.openjdk.java.net/~uvangapally/webrev/2017/8173180/webrev.00/
Thanks,
Ujwal
