Hi Carl,
I think memory pooling cannot be the problem since the requested memory
blocks should not differ much in size and therefore should be reused for a
new memory allocation request. Nevertheless, I modified the code a bit. Now,
it pauses for 30 seconds after 1000 iterations and prints the memory usage
to stdout. Then, it continues.
unsigned
mymem()
{
unsigned total, res;
std::ifstream is("/proc/self/statm");
is >> total >> res;
return res / 256; // convert to MB
}
void
test_qpid_memleak()
{
while (1) {
for (int i = 0; i < 1000; ++i) {
qpid::client::Connection connection__;
connection__.open("localhost", 5672);
connection__.close();
}
std::cout << "mem (1) = " << mymem() << std::endl;
std::cout << "sleeping ..." << std::endl;
sleep(30);
std::cout << "mem (2) = " << mymem() << std::endl;
}
}
int
main(int argc, char** argv)
{
test_qpid_memleak();
return 0;
}
The output is:
mem (1) = 25
sleeping ...
mem (2) = 25
mem (1) = 50
sleeping ...
mem (2) = 50
mem (1) = 99
sleeping ...
mem (2) = 99
mem (1) = 156
sleeping ...
mem (2) = 156
mem (1) = 248
sleeping ...
mem (2) = 248
mem (1) = 348
sleeping ...
mem (2) = 348
mem (1) = 456
sleeping ...
mem (2) = 456
As you can see memory usage is not decreasing. After 7000 iterations the
binary consumes 450MB of resident memory.
-Daniel
On Mon, Nov 23, 2009 at 2:46 PM, Carl Trieloff <[email protected]>wrote:
>
> Is it not just that the memory is not needed back yet from Linux? I.e. gcc
> does not really try
> to reuse memory until it needs to. If you run for say 30 seconds, then stop
> the client for a few,
> and then run it again. Or in the pause in-between start firefox... and see
> if tc_mem_consolidate
> is run and it drops back down
>
> Carl.
>
>
>
> Daniel Etzold wrote:
>
>> Hi Steve,
>> I've created a jira report.
>> https://issues.apache.org/jira/browse/QPID-2214
>>
>> Thanks,
>> Daniel
>>
>>
>> On Fri, Nov 20, 2009 at 4:23 PM, Steve Huston <[email protected]>
>> wrote:
>>
>>
>>
>>> Hi Daniel,
>>>
>>>
>>>
>>>> when executing the code below (connecting and disconnecting to a
>>>>
>>>>
>>> local
>>>
>>>
>>>> broker without sending any messages) the memory usage
>>>> increases constantly
>>>> and rapidly. After 10.000 iterations several hundred
>>>> megabytes of resident memory are used.
>>>>
>>>>
>>> Yikes.
>>>
>>>
>>>
>>>> When commenting out the lines "connection.open()" and
>>>> "close()" the memory usage does not increase.
>>>>
>>>> So, is there a memory leak in Connection open/close?
>>>>
>>>>
>>> It appears you may have found one, yes.
>>>
>>>
>>>
>>>> int
>>>> main(int argc, char** argv)
>>>> {
>>>> while (1) {
>>>> qpid::client::Connection connection;
>>>> connection.open("localhost", 5672);
>>>> connection.close();
>>>> }
>>>> }
>>>>
>>>> I'm running qpid 0.5 on a Debian Linux with gcc 4.3.3.
>>>>
>>>>
>>> Could you maybe run your test under valgrind and get a better idea
>>> where the leak is? Also, when you get this information, could you
>>> please create a jira report for it?
>>> (http://issues.apache.org/jira/browse/qpid) Please attach your test
>>> progam and valgrind report.
>>>
>>> Thanks,
>>> -Steve
>>>
>>> --
>>> Steve Huston, Riverace Corporation
>>> Total Lifecycle Support for Your Networked Applications
>>> http://www.riverace.com
>>>
>>>
>>> ---------------------------------------------------------------------
>>> Apache Qpid - AMQP Messaging Implementation
>>> Project: http://qpid.apache.org
>>> Use/Interact: mailto:[email protected]
>>>
>>>
>>>
>>>
>>
>>
>>
>
>
