CPU 100% Server Help!!!

[ruen]

Hi, all,

We have a server (regular exe) running sometimes 100% CPU usage, which seems
to hang somewhere afterwards.
The problem is that attach to debugger doesn't work after it happens, and if
we run debugger, it's not happening. I saw an article saying about
WinDBG(attaching to thread), but it doesn't give us viewing "Thread" option
for us. Can WinDBG be used for only IIS?

We have multiple threads/multiple machine(DCOM communicating with multiple
server), but everything was fine for months until this happens. Only one exe
has problem.

Please, give us some advice. We really need help on this one.
Thanks in advance.

-- ruen

 

[Richard Liddiment]

What's the name of the EXE running at 100%?

 

[ruen]

It's our program, named server.exe. I could attach to WinDBG, but not sure
that would work with the 100% CPU usage case since visual studio doesn't
work. It's not just looping since if it were, we should be able to attach to
debugger. We tested it.

-- ruen

 

[Richard Liddiment]

I should have said in the first place - we have a clipper (dbase compiler)
application that can cause ntvdm.exe to run at 100% - so that's not going to
be your issue. Also I've encountered lsass.exe running at 100% that turned
out to be a conflict with an old NT4 primary domain controller. But if it's
your own app showing 100% you'll have to figure it out yourself!

What's it written in?
Is it exactly the same server.exe that ran OK for months or has it been
recompiled?
What's been installed on the windows server? Service packs, other
applications etc.
What does it do?

 

[ruen]

Thanks,

It is written in C++, and we could attach to windbg~ =) Which was cool since
it shows all thread callstack. We found some clue, but not yet well. It is
not same server.exe, of course, not~ Since last patch(we're game company),
it started to happening. So, basically, the new server.exe has a bug, which
we couldn't figure out which yet since it doesn't attach to debugger. (We
did a lot of examine and anlalyze code and log stuff to figure out where and
who crashes it, but not much luck) The point was that what kind of thing can
cause this kind of issue. If it were infinite loop, we still should be able
to attach to debugger, but it wasn't. And, it should happen with debugger.
But, it's not happening with debugger if we start from there. Some trash
memory or timing issue can do that?

Thanks,

-- ruen