memory pilling up in publish

[alanxz]

memory pilling up in publish

[alanxz]

I cannot reproduce the behavior you're describing with the following program:

#include <SimpleAmqpClient/SimpleAmqpClient.h>

#include <iostream>

using namespace AmqpClient;

int main()
{
  Channel::ptr_t channel = AmqpClient::Channel::Create();

  for (int i = 0; i < 1000000; ++i)
  {
    try
    {
      BasicMessage::ptr_t message = AmqpClient::BasicMessage::Create(std::string(1024*128, 'a'));
      channel->BasicPublish("", "noroute", message, true, true);
      return 0;
    }
    catch (MessageReturnedException &e)
    {
      std::cout << "Message delivery error: " << e.what() << std::endl;
    }
  }

  return 0;
}

That said rabbitmq-c, the library that SimpleAmqpClient wraps, has a limitation in the way it handles memory such that memory can only be released if all of the amqp_frame_t objects allocated since the last call to amqp_maybe_release_buffers() are no longer needed.

In your situation its likely SimpleAmqpClient has received other frames, probably from a consumer, from the broker while your code tries repeatedly to send the message, and thus cannot release memory.

There is a known issue that in certain pathological cases where you have two or more active consumers its possible that the library won't release any memory allocated until the AmqpClient::Channel object is destroyed. In the average case I don't this isn't an issue. A fix for this issue requires support from rabbitmq-c for finer grained memory control, which doesn't exist yet.

The workaround for this issue is to use multiple AmqpClient::Channel objects for each different consumer.

[chenyujian]

i review my code ,you are right, while consuming msg,i make a rpc request in one channel,so the memory is pilling up.