MySQL Bugs: #7576: MyODBC leaking handles

Bug #7576	MyODBC leaking handles
Submitted:	29 Dec 2004 15:22	Modified:	25 Feb 2005 16:04
Reporter:	Robert Lindgren	Email Updates:
Status:	Closed	Impact on me:	None
Category:	Connector / ODBC	Severity:	S1 (Critical)
Version:	3.51.10	OS:	Windows (windows)
Assigned to:	Vasily Kishkin	CPU Architecture:	Any

Description:
Running Envox (www.envox.com) accessing a mysql database using odbc, causes handles to be leaked, in masses. 

This has been reported before, see http://lists.mysql.com/mysql/99413, and is a serious issue since it's not possible for people (read me) to port my current Oracle environment to MySQL. 

How to repeat:
I run MyODBC 3.51.10 against a remote 4.0.22 database, the url above refers to a old mail with a program which will show the leak, of if someone has envox installed I can provide a small script showing the problem.

Can you provide a test case without to involve envox ?
This will help us a lot for to try to repeat the behavior 
reported.

Thanks in advance.

As this has been reported more than 2 years ago with a test case, written in C I guessed you 
had the case provided by Envox at that time still available but that isn't the case I guess? I'll 
try to find something out. It's strange that you cannot see the problem from the odbc trace 
file.

Hi Robert,

The person who handled that case isn't not works longer with us,
and I would prefer to run a test case against Purify than just
a trace log file analyze.

Thanks.

Ok, I understand, the problem I have is that I don't have the small c program which where refered to in the earlier mailinglist discussion. So I thought that you might be having it since it's your list. But it looks like the attached files are not viewable from the web any longer.

What I do have is an Envox system proving the problem, but as it's not that many people having Envox installed for some reason :) I do understand that you would like me to provide some c-code proving the problem. I'll try to find something during next week.

Here is a note from one Envox developer made in the earlier talks between Envox and MySQL  
back in 2001/2002. What I don't really get from this note is, if he is talking about MyODBC 
version or if it's the server version. Anyway it leaks as much when connecting to a 4.1.8 
server. 
  
 I've been discussing with the MySQL developers and they tell me that the 
handles-leak problem will not be solved in 3.51.03, but in 4.1. This 
will most likely take some time before it is released and by knowing 
this, Envox and MySQL can't be used together. (Will eventually crash). 
Just wanted to inform you about this and make the suggestion to use MSDE 
instead which can be found on your MSoffice CD or MSSQL. 
 
Background: 
  Memory leak exist only when SQLExecuteDirect() (i.e. SQLExecute() and 
SQLPrepare()) is executed from separated thread (which is a requirement 
for our software). If it is executed in the same thread, memory leak 
would not appear.  
-----------

Hi Robert,

The below text you posted:

Background: 
Memory leak exist only when SQLExecuteDirect() (i.e. SQLExecute() and 
SQLPrepare()) is executed from separated thread (which is a requirement 
for our software). If it is executed in the same thread, memory leak 
would not appear.

Gives me a guess what is happens: the my_thread_end() isn't called when
additional threads are used in the same connection, then handle leaks
occurs for every additional thread. How I am not sure and I don't have
their application to test I would like to know if you are able for to test
an MyODBC dll it will be provided with the patch ?  

Thanks in advance.

Yeah I would be more than happy to test a patch. As soon as possible is the best time for me to test it :)

Ok. Thanks I will wait just for the next release Connector/ODBC for Windows
(it will be into 1 week). Then I will build the dll for your test.

When I run the Envox environment in Purify I see that for each dbopen-select-close there is 
one handle, a semaphore, left open, and it's myodbc3.dll that is indicated to be the dll using 
the semaphore. The semaphore is never closed. More and more are allocated for each query.

Uploaded some examples, showing handle leak problem that we are seeing. You'll need to  
change dsn and query I guess. But since it's a simple example that shouldn't be that hard.  
  
I've watched thread leakage with process explorer.  
http://www.sysinternals.com/ntw2k/freeware/procexp.shtml 
 
The threaded examples leak one handle per query.

Hi Robert,

I installed the 3.51.10-2 Connector/ODBC driver and tested the
sample: db_test_thread.c and Purify showed me just potential
leaks. Please update you driver with that and let me know
what you find.

That driver you can get from our download site.
Thanks in advance.

Well have you tried the db_test_thread10k.c? Run that and check memory usage of the  
process, please understand that the problem doesn't show up in purify at least not that  
obvious (but I've not used purify before so it might be good, but it sure doesn't do a good job  
in this case), please run the program I was recommended, processexplorer. I don't know why  
purify doesn't show the leak, well it does, a new handle is used in every run in the  
db_test_thread, but in the 10k version it's more obvious that the program leaks like h*ll.   
  
So please test the db_test_thread10k and tell how much memory that has used up compared  
to when it started.  
  
Since this problem has been around since 2001 I regard this as quite serious problem. And 
it's vital for us to get this leak out of the way, otherwise we'll have to stay with Oracle, which I 
guess we don't want, and in this case you neither.

Why doesn't the status change when I do a edit submission?

Showing memory usage when running the 10k threaded version.

Attachment: my_leak1.JPG (image/jpeg, text), 194.38 KiB.

Hi,

I think this bug #7576 is in relationship with bug description in issue #2776 (Memory leak in Connector/ODBC ? ! ).

Best Regards

I can't se how http://bugs.mysql.com/bug.php?id=2776 relates to this problem? And I'm not able to find a bug called (Memory leak in Connector/ODBC ? ! ). So please guide me to the right place so that I can read the description of that bug.

Same bug is present in 3.51.11 also.

Looks to be fixed prerelease MyODBC 3.51.11-2, whhiiiwhoooo :) I'm happy again :)

Will validate more deeply on monday morning.

It sure is fixed in 3.51.11-2. No more handle leaks. Thanks a bunch. You can now close this bug.

Thank you for the update.