Re: [libpqxx-general] Memory leaks

[Rupert Kittinger]

Jeroen T. Vermeulen schrieb:
> On Fri, April 21, 2006 02:14, Dietmar Hummel wrote:
> 
>> I have a really big problem with libpqxx 2.5.1 on a gentoo x86 system. My
>> little program consumes so much memory, that it get killed by the OS
>> after some minutes of running.
> 
>> I use code like the following:
> 
> Okay, this is a pretty clear example (in other words, thanks for a good
> bug report :)
> 
> Two quick points for the record: (i) this is a relatively old version of
> libpqxx, so there's a small chance that the problem will just go away with
> an upgrade; and (ii) it may be useful at some point to know what compiler
> version and libpq version you are using.
> 
> 
>> template <typename TThis> void RoomBase<TThis>::ReadValue::operator()
>> ( argument_type &T )
>> {
>>         ...
>>         // Querystring wieder zusammenstellen
>>         std::string strQuery = this is the SQL-statement UPDATE xxx ...
>>
>>         // and the problem is the following line
>>         T.exec( strQuery );
>>
>>         ...
>> }
> 
>> BTW: EVERY! T.exec()-call consumes very much memory.
> 
> So the problem happens in this exec()...  This is of course *not* supposed
> to leak memory.  That is surprise #1.
> 
> Surprise #2 is that you're getting big leaks from an UPDATE statement. 
> That should not happen at all, because the result from an UPDATE doesn't
> contain much data!  If I read the Valgrind output right, you're losing
>> 500 blocks of >2 KB each.  Is this from a single call to this function?
> 
> 
>> Is there anything wrong? Do i miss a call to something like T.clear()?
> 
> There is definitely something wrong.  The exec() call is not supposed to
> use so much memory AFAICS (although it's all inside libpq so I don't
> really know what exactly goes on in there), and then the result value that
> is returned should get destroyed immediately, freeing up any memory it has
> allocated.
> 
> Is there any way you can see whether the pqxx::result destructor is being
> called for every result object that has been created?  In other words, is
> PQclear() being called right after the T.exec()?
> 
> The call to PQclear() is in a function pqxx::internal::freemem_result() in
> src/util.cxx.  You could set a breakpoint there in the debugger.
> 
> 
> Jeroen
> 
>

There is a memory leak in 2.5.1 that was fixed in 2.5.2. See the mailing 
list archives for details. I recommend upgrading to release 2.5.5 (which 
works great for me :-)

Rupert

-- 
Rupert Kittinger <[EMAIL PROTECTED]>
Krenngasse 32
A-8010 Graz
Austria

_______________________________________________
Libpqxx-general mailing list
Libpqxx-general@gborg.postgresql.org
http://gborg.postgresql.org/mailman/listinfo/libpqxx-general