Memory leak determined by valgrind

[pengweichu]

Hi, I'm using latest code of blockingconcurrentqueue in my project, I got below output by valgrind:

<error>
  <unique>0x3a</unique>
  <tid>1</tid>
  <kind>Leak_DefinitelyLost</kind>
  <xwhat>
    <text>24 bytes in 1 blocks are definitely lost in loss record 58 of 325</text>
    <leakedbytes>24</leakedbytes>
    <leakedblocks>1</leakedblocks>
  </xwhat>
  <stack>
    <frame>
      <ip>0x4C2A436</ip>
      <obj>/usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so</obj>
      <fn>operator new(unsigned long, std::nothrow_t const&amp;)</fn>
      <dir>/builddir/build/BUILD/valgrind-3.13.0/coregrind/m_replacemalloc</dir>
      <file>vg_replace_malloc.c</file>
      <line>377</line>
    </frame>
    <frame>
      <ip>0xAF57725</ip>
      <obj>/usr/lib64/libstdc++.so.6.0.22</obj>
      <fn>__cxa_thread_atexit</fn>
    </frame>
    <frame>
      <ip>0x621B1D</ip>
      <obj>/home/dev/project/bin/proxy</obj>
      <fn>moodycamel::details::ThreadExitNotifier::instance()</fn>
      <dir>/home/dev/project/common</dir>
      <file>concurrentqueue.h</file>
      <line>545</line>
    </frame>
    <frame>
      <ip>0x6219B1</ip>
      <obj>/home/dev/project/bin/proxy</obj>
      <fn>moodycamel::details::ThreadExitNotifier::subscribe(moodycamel::details::ThreadExitListener*)</fn>
      <dir>/home/dev/project/common</dir>
      <file>concurrentqueue.h</file>
      <line>510</line>
    </frame>
    <frame>
      <ip>0x62423C</ip>
      <obj>/home/dev/project/bin/proxy</obj>
      <fn>moodycamel::ConcurrentQueue&lt;resip::Data, proxy::LogQueue::Traits&gt;::get_or_add_implicit_producer()</fn>
      <dir>/home/dev/project/common</dir>
      <file>concurrentqueue.h</file>
      <line>3413</line>
    </frame>
    <frame>
      <ip>0x6236F9</ip>
      <obj>/home/dev/project/bin/proxy</obj>
      <fn>bool moodycamel::ConcurrentQueue&lt;resip::Data, proxy::LogQueue::Traits&gt;::inner_enqueue&lt;(moodycamel::ConcurrentQueue&lt;resip::Data, proxy::LogQueue::Traits&gt;::AllocationMode)0, resip::Data const&amp;&gt;(resip::Data const&amp;)</fn>
      <dir>/home/dev/project/common</dir>
      <file>concurrentqueue.h</file>
      <line>1291</line>
    </frame>
    <frame>
      <ip>0x623280</ip>
      <obj>/home/dev/project/bin/proxy</obj>
      <fn>moodycamel::ConcurrentQueue&lt;resip::Data, proxy::LogQueue::Traits&gt;::enqueue(resip::Data const&amp;)</fn>
      <dir>/home/dev/project/common</dir>
      <file>concurrentqueue.h</file>
      <line>914</line>
    </frame>
    <frame>
      <ip>0x622BDC</ip>
      <obj>/home/dev/project/bin/proxy</obj>
      <fn>moodycamel::BlockingConcurrentQueue&lt;resip::Data, proxy::LogQueue::Traits&gt;::enqueue(resip::Data const&amp;)</fn>
      <dir>/home/dev/project/common</dir>
      <file>blockingconcurrentqueue.h</file>
      <line>521</line>
    </frame>
    <frame>
      <ip>0x62061F</ip>
      <obj>/home/dev/project/bin/proxy</obj>
      <fn>proxy::LogQueue::add(resip::Data const&amp;)</fn>
      <dir>/home/dev/project/common</dir>
      <file>LogQueue.cxx</file>
      <line>46</line>
    </frame>
    <frame>
      <ip>0x62A60B</ip>
      <obj>/home/dev/project/bin/proxy</obj>
      <fn>proxy::AppExternalLogger::operator()(resip::Log::Level, resip::Subsystem const&amp;, resip::Data const&amp;, char const*, int, resip::Data const&amp;, resip::Data const&amp;)</fn>
      <dir>/home/dev/project/common</dir>
      <file>proxyLogger.cxx</file>
      <line>91</line>
    </frame>
    <frame>
      <ip>0x8F4E02</ip>
      <obj>/home/dev/project/bin/proxy</obj>
      <fn>resip::Log::Guard::~Guard()</fn>
    </frame>
    <frame>
      <ip>0x56EA75</ip>
      <obj>/home/dev/project/bin/proxy</obj>
      <fn>proxy::RedisConnection::queueSetClientName()::{lambda(cpp_redis::reply const&amp;)#1}::operator()(cpp_redis::reply const&amp;) const</fn>
      <dir>/home/dev/project/proxy</dir>
      <file>RedisConnection.cxx</file>
      <line>526</line>
    </frame>
  </stack>
</error>

[cameron314]

Not much I can do about that, it looks like your libstdc++ is leaking from __cxa_thread_atexit.

[pengweichu]

Not much I can do about that, it looks like your libstdc++ is leaking from __cxa_thread_atexit.

If remove macro "MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED" will avoid this issue ?
Does there has any negative results If remove macro "MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED" ?

Thanks

[cameron314]

You can remove MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED, and it will indeed work around this, but memory usage (and performance) will be impacted because every implicit producer will exist forever instead of being recycled when the thread it's associated to is destroyed.

[powerdevil]

Hi, I came across with issue too, but i'm using Clang Address sanitizer. It reports the leak is from line146: " elt *new_elt = new (std::nothrow) elt;", having no idea why.