Visual Studio 2010 or WinDbg take longer than usual to load debug symbols.

Do you dislike a situation where applications on your machine have been working fine for weeks and months at a time and all of a sudden they begin to run very slowly? Well, I do too, especially when it comes to applications I use all day & every day.

At any given point in time I usually have Visual Studio and/or WinDbg running and today I noticed that my Visual Studio 2010 started to consume more CPU cycles than it does usually, and it was loading debug symbols for a release mode postmortem dump very, very slowly. First thing I did was to open the same dump in WinDbg, and guess what? WinDbg also was acting exactly the same. It was consuming approximately 30 to 40 percent of CPU cycles while in .reload /f /v and it was taking its sweet time.

On my machine, I use _NT_SYMBOL_PATH environment variable that controls my path to local symbol cache and remote symbol servers, and it is applicable to all debuggers that I use. I must point out that while I did come across several blogs that recommend against using this variable, it isn’t the root cause of the problem I had experienced today. After all, my debuggers were working fine last week (Friday) and not today (Monday).

So, what changed?

While I didn’t change anything, my machine is a member of a corporate domain so it is quite possible that “something” was changed on my machine for me.

Troubleshooting approach: simultaneous sniffer and process monitor capture on my machine while reloading symbols in VS 2010 or WinDbg.

Observations:

I noticed 3 to 7 second delays originating from my machine between each request to a remote symbol server (in this case: Microsoft public symbol server). Meaning: WinDbg or VS 2010 were blocked / busy doing something else before they even get to the point where they ask remote symbol server for a file, if it is missing in my cache.

sniffer

I also noticed excessive registry activity surrounding the registry keys shown below.

Continue reading

Share

WinDbg exhibits a memory leak when you debug postmortem managed dumps

Howdy there!
My name is Olegas and I’ll be blogging here from time to time. Prashant has a very nice collection already and I’ll be adding my 2 cents every once in a while.

Recently I’ve come across an interesting behavior in WinDbg and I decided to look into it a bit further.

The scenario:
• You are debugging a managed dump using WinDbg 6.11.0001.404 on 32bit platform.
• You are trying to dump hundreds of managed objects to inspect their properties. You use a script similar, but not limited to
.foreach (MyObj {!dumpheap -MT 00687320 -short}) {!do MyObj}

The observed behavior:
• After dumping few dozen objects, WinDbg begins to report
<Note: this object has an invalid CLASS field>
Invalid object
• Your 32bt machine begins to respond very slowly and you notice excessive paging.
• Perfmon shows a behavior within WinDbg consistent with a memory leak
Perfmon_Graph

Continue reading

Share