Re: MDAC memory leak

Tech-Archive recommends: Repair Windows Errors & Optimize Windows Performance

See below...
On Mon, 12 May 2008 14:51:02 -0700 (PDT), prashant <lookforprashant@xxxxxxxxx> wrote:

On May 9, 9:31 pm, Joseph M. Newcomer <newco...@xxxxxxxxxxxx> wrote:
When allocations are done, particularly from libraries, there always arises the issue of
when the storage should be freed.  Most libraries place the decision of when to free
storage on the caller.  So if you comment out a call to a library, and storage leaks go
down, you need to investigate under what circumstances the storage is freed.  It could
well be that you are failing to notify the library that you are finished with the last set
of data before initiating a call to obtain new data.  Since I don't know what libraries
you are using, I can't offer advice, but in libraries I've written, and third-party
libraries I've used, there is typically a requirement to either delete the objects
returned or to call the library to delete the objects returned.

strcore leaks are hard to diagnose, because they mean you have an unfreed CString.  This
is actually fairly rare in most cases, and what you most often see is a CString as a
member of some class leaking because the instance of the object is still hanging around.
There's a capability of breaking on a particular memory allocation, so if the effect is
reproducible in that sequences of dumps always give you a leak for allocation {1024} then
you set the memory allocator breakpoint value to 1024, and you will see where it is
called.  You will probably find a CString inside a CSomething that is a member of an array
or list of a CWhatever, and the real problem is not that you are leaking a string, but
that you neglected to free the CWhatever object when you were done with it.  I tend to
leave strcore.cpp leaks as the very, very, very last leaks I will ever look for.  When
I've fixed the other leaks (and it is usually a systematic pattern; on the first test, I
leak 500 objects, on the second test I leak 3, because I fixed the bug).  The new smart
pointers also reduce memory leakage by making it impossible to leave dangling objects
around.

Oh, yes, the other cause is that you DID delete the CWhatever, but the CWhatever had a
CSomething* member, and you didn't delete the CSomething* in the ~CWhatever destructor.
That's usually where I go awry, and it is usually a stupid oversight on my part (I MEANT
to put the delete in, but got distracted in some way and forgot about it...)

So first look for major leaks, and worry about strcore leaks as the very, very last thing,
and you may not have to deal with them at all.

It helps if you take a look at the line that says

#define new DEBUG_NEW

or whatever it is (including its #if... conditionals) and make sure all your modules have
it.  This sometimes helps, because it forces the debug allocator call to be issued, which
includes the file/line information in the object.
                                        joe



On Fri, 9 May 2008 14:36:00 -0700, Prashant <Prash...@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
Joseph,

Thanks for you reply. Actually when we commented all the calls related to
the database the memory usage as well as the amount of leak came down
subsequently.

Also when we used some of the memory leak tools suggested on microsoft site
those are showing leak in strcore.cpp.

Thanks,
Prashant

"Joseph M. Newcomer" wrote:

See below...
On Wed, 7 May 2008 02:11:00 -0700, App shows memory leak on some machines.
<Appshowsmemoryleakonsomemachi...@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:

Actually I've originally used AfxBeginThread() function to launch thread and
just a return statement while exiting thread. And since I was getting handle
leak along with memory leak, I tried using CloseHandle() in the previously
mentioned way. After this change I found no difference in app's behavior.
I again revert back to original code where I've used a return statement in
the top level thread function. Still I can find same behavior (handle +
memory leak).

I'll like to add some more points here-

-I am launching only one thread from main thread.
Lets say ThreadFirst is launched by main thread. Now, all other threads are
getting launched by ThreadFirst.
It's required since main thread needs to do other task.

-Again there is a critical section (semaphore) in thread function. Many
threads are keep on hanging at the start of critical section waiting for
others threads to leave.
****
I suspect you are confused.  A CRITICAL_SECTION is a very efficient low-level
synchronization primitive.  A mutex is a considerably less efficient synchronization
primitive.  A semaphore is something completely different, and fulfills the purposes of a
semaphore (a counted exclusion object) and has little to do with mutual exclusion of
multiple threads.  So when you say "critical section (semaphore)" I become suspicious,
because semaphores would not be appropriate here.

And yes, if you have a synchronization object, and you have high contention, you are
likely to have a lot of threads hanging.

Some rules:
       Don't use synchronization.  Design your code so there is no need for
               concurrent access
       Lock data, not code
       Lock the smallest possible amount of data for the shortest possible time

In the absence of any code example showing the locking, it is hard to guess what you might
have done, but the most likely cause is that you have violated one or more of the above
rules.

Locking is done when threads rub together.  That means there is friction.  And just like
in mechanical systems, friction generates heat and wastes energy.  I tend to view
synchronization as something that should be avoided by making sure threads are not
actually rubbing.  See my essay "The Best Synchronization is No Synchronization" on my MVP
Tips site.  Every once in a while, you need concurrent access, but the more I work with
threads, the less I do it.
                               joe
****

Can it create any problem??

Thanks,
Digvijay

"Nick Schultz" wrote:

Use Process Explorer fromwww.sysinternals.com.  It is free and is MUCH
better than Task Manager, you can double click on a process and get very
detailed information, including a list of threads and memory usage.  It also
shows processes in a tree format rather than just a flat list...overall its
a better tool.

Go to the Options menu and select "Replace Task Manager" and Process
Explorer will pop up whenever you press CTRL+ALT+DEL.

Nick

"App shows memory leak on some machines."
<Appshowsmemoryleakonsomemachi...@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in
messagenews:91870733-F651-4E90-A5E7-88CB6170E89D@xxxxxxxxxxxxxxxx

Thank you Joseph for your reply.

I've used AfxBeginThread() function to launch thread and subsequently
AfxEndThread() to end it.
I tried using CloseHandle() and checked its return value, also checked
return value for GetLastError(), to make sure that handle is getting
closed.
Return values were as expected. But, it made no difference in behaviour of
exe.

This part of code looks like
//--------------------------------
CWinThread* ThPtr = AfxBeginThread(..);
bool b = ::CloseHandle(ThPtr->m_hThread);
int n = GetLastError();

thread function:
ThreadFunc()
{
.
.
.
AfxEndThread();
returb bRet;
}
//--------------------------------

I've used socket and DB calls inside thread function. For DB
communication,
I've used 'msado15.dll'.

I'll like to know, in which cases application will behave differently (in
terms of memory leaks) on different machine?

Thanks,
Digvijay

"Joseph M. Newcomer" wrote:

It is FAR more likely that what happened is that you are failing to close
the thread
handle of the thread, resulting in a leakage of thread stacks.  Are you
certain every
thread handle is being closed?

The task manager may suggest that memory usage is increasing (it is one
of the few valid
reports about memory it is actually capable of), but you need to rule out
lots of other
explanations before pointing to MDAC (not that it might not be the cause,
but the most
common error I've found in multithreaded leaks is stack leakage).  I
would suggest
downloading the free App Verifier from the Microsoft site and turning on
all memory
management options.
joe

On Fri, 2 May 2008 07:48:02 -0700, Prashant
<Prash...@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:

The application is a multithreaded exe involving socket communication.
The
memory usage in task manager increases when exe is running. It never
comes
down. In other words, for each thread it increases and never  comes down
even
after that thread dies.

Thanks,
Digvijay

"Joseph M. Newcomer" wrote:

How do you detect that there is a memory leak?  You assert this is
happening without
explaining why you think it is so.
joe
On Thu, 1 May 2008 05:11:01 -0700, App shows memory leak on some
machines. <App shows
memory leak on some machin...@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:

Hi,

My application (developed in VC++) is running fine on most of the
machines(without any memory leak). But, it is showing memory leaks on
production machine. Even on development side it is showing memory
leaks on
few machine. Why it is so?

...

read more »- Hide quoted text -

- Show quoted text -

Just for the information; In MDAC 2.82.3939 which comes with the
Microsoft Windows SP 2 all the memory leaks are fixed? The tools we
downloaded from microsoft site are still showing us leak in msado. So
we wanted to confirm this.
****
Leak detection does not say who "leaked" the storage; it says who "allocated" the storage.
If a library allocates storage but does not free it because the caller of the library
failed to call some method that released it, it is not the fault of the library; it is the
fault of the caller of the library.

Also, if storage is allocated in DllMain, it may not be freed until the DLL_PROCESS_DETACH
call, and consequently *appears* as if it were a storage leak, but in fact it is not. The
presence of a storage allocation attribution does not necessarily put the burden of
release on the code that did the allocation.

I've actually released libraries in which the documentation clearly states, usually in
about three places:

"Returns a pointer to an object of type <SOMETYPEHERE>. It is the responsibility of the
caller to free this storage by calling <FREESOMTYPE(SOMETYPE * p)" and gotten nasty notes
about how I'm leaking storage all over the place...
joe

Thanks,
Prashant
Joseph M. Newcomer [MVP]
email: newcomer@xxxxxxxxxxxx
Web: http://www.flounder.com
MVP Tips: http://www.flounder.com/mvp_tips.htm
.

Relevant Pages

  • Re: MDAC memory leak
    ... when the storage should be freed. ...  Most libraries place the decision of when to free ... strcore leaks are hard to diagnose, because they mean you have an unfreed CString. ... There's a capability of breaking on a particular memory allocation, ...
    (microsoft.public.vc.mfc)
  • Re: MDAC memory leak
    ... 2003 SP2 there are no leaks? ... microsoft site still showing leak in msado. ...  Most libraries place the decision of when to free ... There's a capability of breaking on a particular memory allocation, ...
    (microsoft.public.vc.mfc)
  • Re: tip50 Accepted
    ... This is also the version to be bundled with Tcl core 8.6 beta. ... Is it really true that memory leaks are being plugged in code which is ... It is really unbelievable that a developer would be hunting down ...
    (comp.lang.tcl)
  • Re: tip50 Accepted
    ... This is also the version to be bundled with Tcl core 8.6 beta. ... Is it really true that memory leaks are being plugged in code which is ...
    (comp.lang.tcl)
  • Re: FastMM 4.40 released (replacement memory manager for Delphi IDE and applications, free and open
    ... >> A fast replacement memory manager for Borland Delphi Win32 applications ... >> - Optionally reports memory leaks on program shutdown. ... >> the need for stack frames. ... If exception handling is not in place FastMM ...
    (borland.public.delphi.thirdpartytools.general)