Author: wesleysmith
Subject: procdump -r not mapping all memory regions in dmp
Posted: 25 March 2013 at 11:08pm
I'm observing the same behavior as the original poster. As a test, I dumped the same w3wp.exe process both using reflection and without.
Without reflection:
With reflection:
The file sizes on the disk show evidence of the problem:
Regular.dmp is 704 Megs, while Reflected.dmp is 2 Megs and Reflected-reflected.dmp is 531 Megs. So, just from the disk sizes, it appears that several hundred Megs of data is potentially missing.
After loading both dumps into WinDBG (using the .ini file for the reflected dump), I find that commands like !verifyheap (from SOS) work fine on the regular dump file, but won't work on the reflected version:
Using .dumpdebug shows a large difference in the number of Memory64ListStreams, just like the original post.
So, are the reflected dumps just completely unreliable, or is there something that's triggering this problem for just some situations? Any way to work around this issue (while still using reflected dumps)?
Subject: procdump -r not mapping all memory regions in dmp
Posted: 25 March 2013 at 11:08pm
I'm observing the same behavior as the original poster. As a test, I dumped the same w3wp.exe process both using reflection and without.
Without reflection:
|
With reflection:
|
The file sizes on the disk show evidence of the problem:
Regular.dmp is 704 Megs, while Reflected.dmp is 2 Megs and Reflected-reflected.dmp is 531 Megs. So, just from the disk sizes, it appears that several hundred Megs of data is potentially missing.
After loading both dumps into WinDBG (using the .ini file for the reflected dump), I find that commands like !verifyheap (from SOS) work fine on the regular dump file, but won't work on the reflected version:
|
Using .dumpdebug shows a large difference in the number of Memory64ListStreams, just like the original post.
So, are the reflected dumps just completely unreliable, or is there something that's triggering this problem for just some situations? Any way to work around this issue (while still using reflected dumps)?