No reachable instances in Native memory screen

[guillaume.philippot]

Hello,

I am investigating a memory leak on a dump of 3.7GB and the figures I see in the native memory screen under the Committed memory tab seem very strange.

Code: Select all

Private - 3702244KB
   Managed heaps - 2299593KB
      Normal heap - 2098178KB
         Generation #0 - 132488KB
            Gaps - 127635KB
            Unreachable instances - 4854KB
         Generation #1 - 6527KB
            Gaps - 1KB
            Unreachable instances - 6526KB
         Generation #2 - 1959163KB
            Gaps - 1005116KB
            Unreachable instances - 954046KB
      Large heap - 187027KB
         Overhead/unused - 100185KB
         Unreachable instances - 86842KB
      Overhead - 14388KB
         Other - 14388KB
   Code - 5984KB
   JIT - 7104KB
   Runtime heap - 34236KB
   Other data - 1355327KB
Shared - 0KB
Potentially shared - 489724KB

As you see, none of the normal heaps contain Reachable instances. It makes no sense to me as the total figures in the overview screen (see attachment) are completely different.

Could it be due to

• a corrupted dump ?
• a wrong interpretation of the figures ?
• a bug in the software ?

Your help and insight is greatly welcome.

Regards,
Guillaume

[Andreas Suurkuusk]

Thanks for the error report. This is a bug in the profiler that we have also noticed recently. Due to an optimization, the reachable instances are analysed in another stage of the heap snapshot collection. Now the reachable information is updated after the native memory information is reported, which causes all instances to be presented as unreachable in the native memory view.

We will fix this in the next release of .NET Memory Profiler. Hopefully, we will be able to present reachable and unreachable instances separately again, without incurring a too high performance cost. Otherwise we will just present instances, without specifying reachability in the native memory view. Information about unreachable instances can still be found in the Types view and Type details view.

[guillaume.philippot]

Thank you for your answser.

This is a bug in the profiler [...] which causes all instances to be presented as unreachable in the native memory view.

And what about the gaps ? Are those figures correct ? Do we really have >1GB of gaps in our heaps or should we just not rely on the native memory view until the bug is fixed ?

Regards,
Guillaume

[Andreas Suurkuusk]

We're not aware of an error related to the presentation of gaps in the Native memory view. However, the numbers presented in your session does not look correct. We will look into this and see if we find any problems related to the identification of gaps.

It would be good if we could import your memory dump so that we can try reproduce the problem. Is there any possibility that you can send us the memory dump? If there is, please contact us at support@scitech.se.