zeromq / cppzmq Goto Github PK

View Code? Open in Web Editor NEW

1.9K 83.0 757.0 696 KB

Header-only C++ binding for libzmq

Home Page: http://www.zeromq.org

License: MIT License

C++ 95.98% CMake 3.70% Shell 0.33%

c-plus-plus libzmq header-only binding zeromq 0mq

cppzmq's Issues

Missing FindZeroMQ.cmake

Attempting to build on Ubuntu 16.04 after installing ZeroMQ 4.2.1

CMake Error at CMakeLists.txt:4 (find_package):
By not providing "FindZeroMQ.cmake" in CMAKE_MODULE_PATH this project has
asked CMake to find a package configuration file provided by "ZeroMQ", but
CMake did not find one.

Could not find a package configuration file provided by "ZeroMQ" with any
of the following names:
ZeroMQConfig.cmake
zeromq-config.cmake
Add the installation prefix of "ZeroMQ" to CMAKE_PREFIX_PATH or set
"ZeroMQ_DIR" to a directory containing one of the above files. If "ZeroMQ"
provides a separate development package or SDK, be sure it has been
installed.

-- Configuring incomplete, errors occurred!

Am i invoking cmake incorrectly?

Memory leak in monitor_t?

The while loop (in monitor_t::monitor() method) can be aborted at line 700 and 717. If that happens the zmq_msg_close (&eventMsg); will never be executed. Couldn't that create a memory leak?

Packaging -> Need a tagged version

As per discussion on #107, I would like to package cppzmq for conda, it would be very nice to have a tagged version on GitHub, to have a fixed version of the code to refer to.

One thing that would help is that the tag always follow the same pattern and can be completely deduced from the version number. (typically M.m.p).

cmake install TARGETS given no ARCHIVE DESTINATIOn for static library target "cppzmq"

From the master (0feae8c) , and using cmake v2.8.12.2 on CentOS 6.9, I clone and immediately did:

cmake .

and get the following error message:

--cppzmq v4.2.3
CMake Error at CMakeLists.txt:30 (install):
  install TARGETS given no ARCHIVE DESTINATION for static library target "cppzmq".


-- Configuring incomplete, errors occurred!

No idea how to fix that. Stackoverflow wasn't helpful there either.

Hello, it is possible to "reuse" message_t into a recv loop? For instance, is the followed code correct?..or message_t varible has to create and destroy on each loop iteration? so I have to declare message_t msg inside while loop.

I'm wondering if reusing messate_t could be any internal memory leak?
...
void ReadThread(..)
{
message_t msg; //reused varible `` //... some code
while(true) `` {
`` poll(&items[0], items.size(), 1000);`` if (items [0].revents & ZMQ_POLLIN)`` { `` `` service->recv(&msg);`` `` // do something with msg..`` } `` }
}

Issue with no NULL-terminated strings.

Sender written in Python, receiver in C++.
recv() does not place an ending NULL after the received number of bytes but, as it does not return the number of bytes, it's not possible to do outside, either.
Result: Impossible to determine the end of the received string. It's just the example on strings on ZMQ tutorial for C.
The only solution seems to avoid the C++ port and use the underlying C functions, which return the length of the message. This method works perfectly with the Python sender.
Is this on purpose or a bug?
In this case, it's not possible to modify the Python sender to include the string length in the message itself.

The send and recv returns true even no connected.

Hi, when I use the cppzmq to connect a wrong IP(actually there no ZMQ server bind), the connect function does not trow exception. After then I try to send and received the data, but the functions always returns true! How could I detect the data is sent out and is recv ?

// connect 
try
{
    int linger = 0;
    mZmqSocket->setsockopt(ZMQ_LINGER, &linger, sizeof(linger));
    stringstream ss;
    ss << "tcp://" << mRemoteIP << ":" << mRemotePort;

    mZmqSocket->connect(ss.str().c_str());
}
catch (const zmq::error_t& e)
{
    cout << "Unable to connect ZMQ socket: " << e.what() << endl;
    return rt;
}


try
{
    while (!mZmqSocket->send(req))
    {
        resip::sleepMs(100);
    }
}

catch (const zmq::error_t& e)
{
    cout << "Failed to send request.";
    return rt;
}



// recv response
zmq::message_t response;
try
{
    if (mZmqSocket->recv(&response, ZMQ_DONTWAIT))
    {
        cout << "Received data from server: " << (char *)response.data() << endl;
    }
}
catch (const zmq::error_t& e)
{
    cout << "Got error with recv: " << e.what();
    return rt;
}

The send and recv functions always return true even no a ZMQ server side running!

Error with zmq::message_t::gets

Getting the following error trying to use the zmq.hpp on a new install. Don't know what change caused the error, I have an older version of the file that works fine.

/usr/local/include/zmq.hpp: In member function `const char* zmq::message_t::gets (const char*) const':
/usr/local/include/zmq.hpp:377:53: error: 'property' was not declared in this scope
  const char* value = zmq_msg_gets (&msg, property);

Support libzmq's Autotools Install: pkg-config

Following #127 and zeromq/libzmq#2621 (comment), the libzmq install via autotools is still a thing and used in packaging systems, too.

Albeit the autotools install lacks the install of CMake config modules, it would be great if we could add support for libzmq autotools-installs anyway.

One way could be to support the always-installed pkg-config module named libzmq in cppzmq's CMakeLists.txt.

For autotools based installs we could as a fallback to the current logic piggyback on pkg-config. This could look like this:

find_package(PkgConfig)
pkg_check_modules(PC_LIBZMQ QUIET libzmq)

set(ZeroMQ_VERSION ${PC_LIBZMQ_VERSION})
find_library(ZeroMQ_LIBRARY NAMES libzmq.so
             PATHS ${PC_LIBZMQ_LIBDIR} ${PC_LIBZMQ_LIBRARY_DIRS})
find_library(ZeroMQ_STATIC_LIBRARY NAMES libzmq.a
             PATHS ${PC_LIBZMQ_LIBDIR} ${PC_LIBZMQ_LIBRARY_DIRS})

connected returns true even as soon as the socket is initialized

The connected() function compares the internal ptr to null to see if it is connected, but ptr is initialized in init(...) which is called during construction.

Poll monitoring sockets

I would like to be able to use zmq::poll() for normal ZMQ sockets and monitoring ZMQ sockets, to handle them all within a single thread. However this isn't currently possible with the zmq::monitor_t class due to the blocking nature of the zmq::monitor_t::monitor() method and the lack of an accessor for the socket that is monitoring socketPtr.

Could we augment the zmq::monitor_t class with the functionality to allow polling? Perhaps something like the following would work (apologies in advance for syntax)...

class monitor_t {
    // ...
    inline operator void* () ZMQ_NOTHROW
    {
        return monitorPtr;
    }

    inline operator void const* () const ZMQ_NOTHROW
    {
        return monitorPtr;
    }

    void monitor_start(socket_t &socket, const char *addr_, int events = ZMQ_EVENT_ALL)
    {
        int rc = zmq_socket_monitor(socket.ptr, addr_, events);
        if (rc != 0)
            throw error_t ();

        socketPtr = socket.ptr;
        monitorPtr = zmq_socket (socket.ctxptr, ZMQ_PAIR);
        assert (monitorPtr);

        rc = zmq_connect (monitorPtr, addr_);
        assert (rc == 0);

        on_monitor_started();
    }

    bool monitor_poll()
    {
        zmq_msg_t eventMsg;
        zmq_msg_init (&eventMsg);
        int rc = zmq_msg_recv (&eventMsg, s, 0);
        if (rc == -1 && zmq_errno() == ETERM)
            return false;
        assert (rc != -1);

        // ...
    }

    void monitor(socket_t &socket, const char *addr_, int events = ZMQ_EVENT_ALL)
    {
        monitor_start(socket, addr_, events);

        bool keep_looping = monitor_poll();
        while (keep_looping) {
            keep_looping = monitor_poll();
        }
    }
    // ...

private:
    void* socketPtr;
    void* monitorPtr;
};

This would allow a usage similar to the following...

void Worker(zmq::socket_t* raw_zmq_data_sock)
{
    std::unique_ptr<zmq::socket_t> zmq_data_sock(raw_zmq_data_sock);
    std::unique_ptr<zmq::monitor_t> zmq_monitor_sock(new zmq::monitor_t);

    zmq_monitor_sock->start_monitor(*zmq_data_sock, "inproc://abc");

    std::vector<zmq::pollitem_t> poll_items = {
        { (void*)(*zmq_data_sock), 0, ZMQ_POLLIN | ZMQ_POLLERR, 0, },
        { (void*)(*zmq_monitor_sock), 0, ZMQ_POLLIN | ZMQ_POLLERR, 0, },
    };

    while (true) {
        poll_items[0].revents = 0;
        poll_items[1].revents = 0;
        items = zmq::poll(poll_items);

        // ...
        if (poll_items[1].revents) {
            zmq_monitor_sock->monitor_poll();
        }
    }

    // ...
}

I can put a pull request together if it sounds like a reasonable idea.

Creating a zmq::pollitem_t requires casting to void *

Looks like zmq::pollitem_t is a just typedef for zmq_pollitem_t. Is there any plans to make zmq::pollitem_t a C++ class on its own? The main reason is that creating a zmq::pollitem_t requires casting a zmq::socket_t to void *, which attracts bugs like:

zmq::pollitem_t { &socket, 0, ZMQ_POLLIN, 0 }

Doing a Google search shows that this bug is quite common and causes a lot of confusion. It would be good if the interface to zmq::pollitem_t can just take a socket without casting.

Abort signal thrown from zmq_send

The following is the string I'm trying to send as a message using zmq_send:
" Node[0]TC[0]: Thread tid 16814, cpu 14 started"

Intermittently, my program crashes only while sending this message, not for other messages. This string is generated using snprintf, so it is definitely NULL terminated.

Following is the stack trace of the crash:

#0 0x00007fa29b56d187 in raise () from /lib64/libc.so.6
#1 0x00007fa29b56e538 in abort () from /lib64/libc.so.6
#2 0x00007fa29ca9ba3e in zmq::zmq_abort(char const*) ()
from libzmq.so.4.2.0
#3 0x00007fa29caa4389 in zmq::msg_t::size() ()
from libzmq.so.4.2.0
#4 0x00007fa29cad1032 in zmq::v2_encoder_t::message_ready() ()
fromlibzmq.so.4.2.0
#5 0x00007fa29cac4801 in zmq::stream_engine_t::out_event() ()
from libzmq.so.4.2.0
#6 0x00007fa29ca9c8a4 in zmq::io_thread_t::in_event() ()
from libzmq.so.4.2.0
#7 0x00007fa29ca9b71e in zmq::epoll_t::loop() ()
from libzmq.so.4.2.0
#8 0x00007fa29cace915 in thread_routine ()
from libzmq.so.4.2.0
#9 0x00007fa2a18ec0a4 in start_thread () from /lib64/libpthread.so.0
#10 0x00007fa29b61d04d in clone () from /lib64/libc.so.6

I have reproduced the same with two version of zmq - 4.2.0 and 4.1.6.

zmq::send always releases its message

As a messages_t is an object, it'll be release anyway. So unless the message content is being moved, I can't see the benefit of zmq_send releasing the message.

In the case of creating a device that repeats a message to more than one destination socket, zmq_send_const must be called directly to stop the message from being released.

Can anyone explain why zmq_send behaves this way and why zmq should have a send_const member?

zmq_ctx_shutdown() is never called

When using the straight C API, if one has a listening thread that is blocking on a recv() function and the application shuts down, they call zmq_ctx_shutdown(context) and it signals the recv() function to exit and shutdown proceeds normally.

However, in the C++ wrapper you have this for the context_t destructor:

inline ~context_t ()
{
    close();
}

inline void close()
{
    if (ptr == NULL)
       return;
    int rc = zmq_ctx_destroy (ptr);
    ZMQ_ASSERT (rc == 0);
    ptr = NULL;
}

Note that this does not include a call to zmq_ctx_shutdown().

Would it not be better to first call zmq_ctx_shutdown() before zmq_ctx_destroy()? Without that, the caller has to manually call zmq_ctx_shutdown() before the destructor gets called, which sort of detracts from the automatic cleanup that destructors are intended to offer.

Thoughts?

2 Router-Dealer proxies do not work

Simply put, whichever proxy is first then the related processes work fine. The second one is ignored. For instance:

zmq::proxy(m_frontEnd, m_backEnd, 0); zmq::proxy(m_frontEnd2, m_backEnd2, 0);

I have a server with multiple threads listening for different connection points. All services listening for messages on the proxy m_frontEnd and m_backEnd work fine. The one using m_frontEnd2, m_backEnd2 never receive messages.

If I swap the order:

zmq::proxy(m_frontEnd2, m_backEnd2, 0); zmq::proxy(m_frontEnd, m_backEnd, 0);

Now all services not recieving messages before are working fine and the services listening for messages on the proxy m_frontEnd and m_backEnd never receive messages.

The docs imply that we can have multiple proxies. This isn't the case or is a bug.

I'm unsure which cppzmq version I'm using and the headers show support for cmake zmq over 4.1.0 and I'm using czmq 4.0.2 if that helps

expression result unused from macro 'ZMQ_ASSERT'

We met this issue while building an application with cppzmq using clang 3.1:

./build_tools/third_party/include/zmq.hpp:144:28: error: expression result unused [-Werror,-Wunused-value]
ZMQ_ASSERT (rc == 0);
~~~~~~~~~~~~~~~^~~~~
./build_tools/third_party/include/zmq.hpp:55:36: note: expanded from macro 'ZMQ_ASSERT'

define ZMQ_ASSERT(expression) (expression)
                                                                 ^

Here is the version of the clang we are using:

...$ clang++ --version
clang version 3.1 (trunk 154757)
Target: x86_64-unknown-linux-gnu
Thread model: posix

Is there a workaround for this?

Thanks,
Gang

Please tag a release for the c++ bindings

Would it be possible to also add a version to the c++ bindings and cut a release of it?

Then one can better package it for a distribution and maybe find a matching version with the main zeromq library and it always stays in sync with the latest zeromq.

With just one header file, why was it split into a separate git repository without any version information at all? Having it along the zeromq library would simplify this also.

segfault when testing the cppzmq implementation of the zguide Majordomo Protocol example

When executing the zguide C++ implementation of the Majordomo example in verbose and debug mode, using the cppzmq header file zmq.hpp; a segfault occurs in process ./mdworker,
just after a client has requested a first service to the broker.

It's a bit hard to debug with breakpoints because the code excerpt pointed to by valgrind is frequently executed. So I thank you in advance for your help.

When tracing the memory usage with valgrind,
I get the following memory error message (extracted from stdout)
Uninitialised value was created by a heap allocation, as listed in more details here below:

[004] echo

15-03-22 11:03:12 I: sending HEARTBEAT to broker

[000]
[006] MDPW01
[001] 04
==4784== Conditional jump or move depends on uninitialised value(s)
==4784== at 0x804ABF1: std::basic_string<unsigned char, std::char_traits, std::allocator >::basic_string(unsigned char const_, std::allocator const&) (char_traits.h:161)
==4784== by 0x804FE1B: mdwrk::recv(zmsg_&) (zmsg.hpp:71)
==4784== by 0x8049E9F: main (mdworker.cpp:18)
==4784== Uninitialised value was created by a heap allocation
==4784== at 0x402A9AC: operator new(unsigned int, std::nothrow_t const&) (in /usr/lib/valgrind/vgpreload_memcheck-x86-linux.so)
==4784== by 0x40A766A: zmq::stream_engine_t::handshake() (stream_engine.cpp:600)
==4784== by 0x40A840F: zmq::stream_engine_t::in_event() (stream_engine.cpp:257)
==4784== by 0x40837A8: zmq::epoll_t::loop() (epoll.cpp:166)
==4784== by 0x40AF90F: thread_routine (thread.cpp:86)
==4784== by 0x4058F6F: start_thread (pthread_create.c:312)
==4784== by 0x42C350D: clone (clone.S:129)
==4784==
15-03-22 11:03:12 I: received message from broker:

lsb_release -a
: Distributor ID: Ubuntu
: Description: Ubuntu 14.04.1 LTS
: Release: 14.04
: Codename: trusty
cppzmq version: git commit 67b2166
zeromq version: zeromq-4.1.0-rc1
zguide version: git commit 67a79791ab082b942e1c8ccb2953b10b0727fa66
mdworker.cpp version: commit 0b5b12b9c7dc5c30853adf10e75c500c43258c6e

mdclient log trace: > ./mdclient -v
15-03-22 11:03:17 I: connecting to broker at tcp://localhost:5555...

15-03-22 11:03:17 I: send request to 'echo' service:

[006] MDPC01
[004] echo
[011] Hello world

mdworker log traces: > ./valgrind --track-origins=yes ./mdworker -v 2>&1 | tee debug.mdworker.log

==4784== Memcheck, a memory error detector
==4784== Copyright (C) 2002-2013, and GNU GPL'd, by Julian Seward et al.
==4784== Using Valgrind-3.10.0.SVN and LibVEX; rerun with -h for copyright info
==4784== Command: ./mdworker -v
==4784==
15-03-22 11:03:09 I: connecting to broker at tcp://localhost:5555...

15-03-22 11:03:09 I: sending READY to broker

[000]
[006] MDPW01
[001] 01
[004] echo

15-03-22 11:03:12 I: sending HEARTBEAT to broker

15-03-22 11:03:12 I: received message from broker:

[000]
[006] MDPW01
[001] 04

15-03-22 11:03:14 I: received message from broker:

[000]
[006] MDPW01
[001] 04

15-03-22 11:03:14 I: sending HEARTBEAT to broker

[000]
[006] MDPW01
[001] 04

15-03-22 11:03:17 I: received message from broker:

[000]
[006] MDPW01
[001] 04

15-03-22 11:03:17 I: sending HEARTBEAT to broker

[000]
[006] MDPW01
[001] 04

15-03-22 11:03:17 I: received message from broker:

[000]
[006] MDPW01
[001] 02
[000]
[000]
[00b] Hello world
mdworker: mdwrkapi.hpp:124: zmsg* mdwrk::recv(zmsg*&): Assertion `m_reply_to.size()!=0' failed.
==4784==
==4784== HEAP SUMMARY:
==4784== in use at exit: 55,634 bytes in 61 blocks
==4784== total heap usage: 261 allocs, 200 frees, 71,059 bytes allocated
==4784==
==4784== LEAK SUMMARY:
==4784== definitely lost: 0 bytes in 0 blocks
==4784== indirectly lost: 0 bytes in 0 blocks
==4784== possibly lost: 590 bytes in 14 blocks
==4784== still reachable: 55,044 bytes in 47 blocks
==4784== suppressed: 0 bytes in 0 blocks
==4784== Rerun with --leak-check=full to see details of leaked memory
==4784==
==4784== For counts of detected and suppressed errors, rerun with: -v
==4784== ERROR SUMMARY: 15 errors from 1 contexts (suppressed: 0 from 0)

mdbroker log trace: > ./mdbroker -v
15-03-22 11:03:09 I: MDP broker/0.1.1 is active at tcp://*:5555

15-03-22 11:03:10 I: received message:

[009] 64F6-D716
[000]
[006] MDPW01
[001] 01
[004] echo
15-03-22 11:03:10 I: registering new worker: 64F6-D716
15-03-22 11:03:10 I: received message: