Hi Javier.

I share your concerns so let me clarify my posting. I wrote application correctly written are guaranteed to run. By correctly written, I mean applications that are not using deprecated, undocumented or exotic APIs.

You are right about support, it is up to the software vendor to provide or not to provide support for a new OS release. Sun is guaranteeing to its supported customers that adhere the compatibility program an eligible application binary will work. Sun doesn't guarantee support for non Sun products.

I wasn't suggesting to upgrade the OP production platform to a newer version of Solaris. I was suggesting to build a test platform to investigate the issue. Depending on the application, that might simple or not, especially if a different license is required to install and use the application.

Finally, a core file coming from a non instrumented binary is unlikely to help finding a memory leak.

Hi guys:

Going back to the original quote from Jlliagre "Applications correctly written for Solaris 8 are guaranteed to run on newer releases", which triggered my original response, here are my questions:

1. How do any of us know if the application is written "correctly"?

2. Even if you look at the "guarantee" from Sun, you can see that:

a> "If an application meets the program guidelines and does not recompile", meaning that it may not run and it
could fail right of the bat.

b> "If an application experiences compatibility problems when running on the Solaris 10 Operating System, Sun
will analyze the problem and provide the appropriate remedy as set forth in the Solaris Application Guarantee
Program", which once again it shows that altough Sun will try to help, upgrading to Solaris 10 may not be the
automatic solution to such problem.

Lastly, the original poster indicated that they don't have solaris 10 running on any boxes and there are a lot of
employers that will fire the employee if they install unapproved version in the work area. Trust me on this, if you deal with the government (mainly) here, there is so much red tape and procedures to follow that it takes forever to
get an approval.

Now, going back to the original problem, if the guy runs a savecore -L and generates a live crash dump, he could
try to send the resulting files (unix.x and vmcore.x) to the kernel team, which could tell him in detail why (meaning what library/function/thread/whatever) is not releasing memory) and then he could show those results to
the person or company that wrote the application to begin with.

I do agree with Jlliagre that if it was possible, running the app on 10 would make it easier to debug via newer tools like dtrace. But again, upgrading may not be a possibility here.

Have a good sunday,

Javier

And by the way, what application are we talking about? It would be nice to know.

LOL

Javier

Nobody knows. That shouldn't prevent to try running it.
This debate is pointless. The OP poster is unlikely to be engaged in this compatibility program anyway. My point is Sun engineering is committing to provide compatibility from older to newer Solaris releases. This is an important differentiator that is worth knowing compared with Linux, AIX, HP-UX. This is something real, I have no statistics but most application I tried worked as is or with very little change. This is something that might be useful in certain situations including the one this thread is about.
That might be the case or not, but nothing you and I can help to fix. Most companies, government owned or not, I use to work with are certainly willing to build whatever platform in order to help investigating an issue. Again, I do not suggest to migrate the production platform to Solaris 10 (although it would be a smart move), I suggest to build a test platform. This platform might be disconnected from the "work area" if that is a concern to the OP.
I'm afraid you are over-optimistic about what a kernel dump file contains and can be used to. A kernel memory leak will be detected, but as far as I understand, we are talking about an application leak. AFAIK, only application stacks, not their virtual memory, are stored in these dumps. The kernel team members are undoubtedly smart guys but I wouldn't expect them to exercise ESP.