Re: Problem in SQLFreeHandle (Statement) - Mailing list pgsql-odbc

I recompiled the driver to enable extra logging.

Here is the grep on ENTER_CONN_CS, the second address is the address of the mutex (pthread recursive).

This was compiled on OSX with Clang, and then also with GCC with the same result. Behavior doesn't reproduce on Windows.

[140735257568016][[SQLFreeHandle]]

[140735257568016]*** ENTER_CONN_CS [140210608634616] *** 

[140735257568016]*** ENTER_CONN_CS [140210608634616] *** 

[140735257568016]*** LEAVE_CONN_CS [140210608634616] *** 

[140735257568016]*** LEAVE_CONN_CS [140210608634616] *** 

[140735257568016]*** LEAVE_CONN_CS [140210608634616] *** 

[140735257568016][SQLGetConnectAttrW]

[140735257568016]*** ENTER_CONN_CS [140210608634616] *** 

[140735257568016]*** LEAVE_CONN_CS [140210608634616] *** 

[140735257568016][SQLGetConnectAttrW]

[140735257568016]*** ENTER_CONN_CS [140210608634616] *** 

[140735257568016]*** LEAVE_CONN_CS [140210608634616] *** 

[140735257568016][SQLGetConnectAttrW]

[140735257568016]*** ENTER_CONN_CS [140210608634616] *** 

[140735257568016]*** LEAVE_CONN_CS [140210608634616] *** 

[4466282496][[SQLAllocHandle]]

[4466282496]*** ENTER_CONN_CS [140210608634616] ***    <- new thread grabbed mutex, count=1

[4466282496]**** PGAPI_AllocStmt: hdbc = 0x7f8553857000, stmt = 0x7f8552d17100

[4466282496]*** LEAVE_CONN_CS [140210608634616] ***    <- count=0

[4466282496]**** PGAPI_Prepare: STMT_ALLOCATED, copy

[4466282496]*** ENTER_CONN_CS [140210608634616] ***    <- count=1

[4466282496]*** ENTER_CONN_CS [140210608634616] ***    <- count=2

[140735257568016]*** LEAVE_CONN_CS [140210608634616] *** <- this thread doesn't own, count=2

[4466282496]*** LEAVE_CONN_CS [140210608634616] ***      <- count=1 

[4466282496]*** LEAVE_CONN_CS [140210608634616] ***      <- count=0

[4466282496]*** LEAVE_CONN_CS [140210608634616] ***      <- count=0

[140735257568016][[SQLFreeHandle]]

[140735257568016]*** ENTER_CONN_CS [140210608634616] ***  <-thread blocks on mutex here, should be unlocked?

On Thu, Jun 12, 2014 at 6:30 AM, Heikki Linnakangas <hlinnakangas@vmware.com> wrote:

On 06/12/2014 05:30 AM, Jade Koskela wrote:

Hello,
I am using psql-odbc 9.3 on OSX.

 

Which version, exactly? There were some changes to locking in SQLFreeHandle in 09.03.0200. Not sure if those changes might've fixed this, or caused it.

Can easily you test on another platform, or with different version of unixODBC?

 

I have one thread running a query, and
another thread cancels the query using SQLCancel. The thread running the
query successfully cancels and exits the driver code. When we try to free
the statement handle, it blocks on a mutex doing ENTER_CONN_CS, or
something similar in SQLFreeHandle.

Does anyone have any input on this problem? I don't see who is holding the
mutex, the thread running the query is done.

 

I tried to reproduce this with the attached little test program. It works fine. Can you post the test program you're using, or modify the attached so that it exhibits the behavior you're seeing, please?

- Heikki