EAGAIN in ZMQ extended request reply - c++

Question

I'm trying to create a REQ <--> Router <--> Dealer <--> REP communication in C++. The child process binds the router and dealer, proxies between router and dealer, connects the REP to the dealer and waits for a message with zmq_recv.
The parent process connects a REQ to the router and tries to send a message, however I'm getting a zmq_send error in parent: Resource temporarily unavailable (which is EAGAIN). According to zmq_send docs, EAGAIN means:
Non-blocking mode was requested and the message cannot be sent at the moment.
However the message does get sent since it is received in the child process. Why does it return that errno?
Here is the MCVE:
#include <zmq.h>
#include <iostream>
#include <sys/types.h>
#include <unistd.h>
#include <assert.h>
#include <thread>
#include <stdio.h>
int main() {
char connect_path[35];
int rc;
int msg;
pid_t child_pid = fork();
if (child_pid == 0) {
// Child
void* child_context = zmq_ctx_new ();
if (child_context == NULL) {
std::cerr << "\nChild context error\n";
}
void* router = zmq_socket(child_context, ZMQ_ROUTER);
if (router == NULL) {
perror("zmq_socket of type router error");
}
char bind_path[35];
snprintf(bind_path, sizeof(bind_path), "ipc:///tmp/zmqtest%d-router", getpid());
rc = zmq_bind(router, bind_path);
assert (rc == 0);
void* dealer = zmq_socket(child_context, ZMQ_DEALER);
if (dealer == NULL) {
perror("zmq_socket of type dealer error");
}
snprintf(bind_path, sizeof(bind_path), "ipc:///tmp/zmqtest%d-dealer", getpid());
rc = zmq_bind(dealer, bind_path);
assert (rc == 0);
std::thread z_proxy (zmq_proxy, router, dealer, nullptr);
z_proxy.detach();
void* rep_socket = zmq_socket (child_context, ZMQ_REP);
if (rep_socket == NULL) {
perror("zmq_socket of type rep error");
}
snprintf(connect_path, sizeof(connect_path), "ipc:///tmp/zmqtest%d-dealer", getpid());
rc = zmq_connect(rep_socket, connect_path);
assert (rc == 0);
while(1) {
if (zmq_recv (rep_socket, &msg, sizeof(msg), 0) != 0) {
perror("zmq_recv error");
}
printf("\nReceived msg %d in process %d\n", msg, getpid());
break;
}
if (zmq_close(rep_socket) != 0) {
perror("zmq_close of rep_socket in child error");
}
if (zmq_ctx_term(child_context) != 0) {
perror("zmq_ctx_term of child_context error");
}
} else {
// Parent
sleep(1);
void* parent_context = zmq_ctx_new ();
if (parent_context == NULL) {
std::cerr << "\nParent ctx error\n";
}
void* req_socket = zmq_socket (parent_context, ZMQ_REQ);
if (req_socket == NULL) {
perror("zmq_socket of type req error in parent");
}
snprintf(connect_path, sizeof(connect_path), "ipc:///tmp/zmqtest%d-router", child_pid);
rc = zmq_connect(req_socket, connect_path);
assert (rc == 0);
msg = 30;
if (zmq_send (req_socket, &msg, sizeof(msg), 0) != 0) {
perror("zmq_send error in parent");
}
if (zmq_close(req_socket) != 0) {
perror("zmq_close of req_socket in parent error");
}
if (zmq_ctx_term(parent_context) != 0) {
perror("zmq_ctx_term of parent_context error");
}
}
}

Answer 1

Step 1: Make a trivial test:
Well, as a minimum point, there ought be this sort of test-en-Queueing first:
rc = zmq_send ( req_socket, "A_TEST_BLOCK", 12, ZMQ_DONTWAIT );
printf ( "INF: zmq_send ( req_socket, "A_TEST_BLOCK", 12, ZMQ_DONTWAIT )\nZMQ: returned rc == %d\nZMQ: zmq_errno ~ %s\n",
rc,
zmq_strerror ( zmq_errno() )
);
.
Step 2: post the printed outputs
Next, if there are any "missed" shots, the error-analysis may advise on potential reason(s)
( if and only if the parent_ctx indeed rejected to even accept the data from a simplest ever zmq_send() call into it's internal queueing facility with an explicit reason for having done so ).
Otherwise we know nothing ( and the ZMQ_DONTWAIT flag is not the reason here ).
As the test was run, it yielded:
INF: zmq_send ( req_socket, 'A_TEST_BLOCK', 12, ZMQ_DONTWAIT )
ZMQ: returned rc == 12
ZMQ: zmq_errno ~ Resource temporarily unavailable
Step 3:
The test has confirmed, as per documentation:
The zmq_send() function shall return number of bytes in the message if successful.
So, let's dig a step deeper:
int major, minor, patch;
zmq_version ( &major, &minor, &patch );
printf ( "INF: current ØMQ version is %d.%d.%d\nZMQ: zmq_errno ~ %s\n",
major, minor, patch,
zmq_strerror ( zmq_errno() )
);
Step 4:
In case the bleeding-edge API-updates do not conform to the published API-specification, document the incident:
printf ( "EXPECT( NO ERROR, ON START ): zmq_errno ~ %s\n",
zmq_strerror ( zmq_errno() )
);
printf ( "EXPECT( <major>.<minor>.<patch> ): zmq_version ~\n" );
int major, minor, patch
zmq_version ( &major, &minor, &patch );
printf ( "INF: current ØMQ version is %d.%d.%d\nZMQ: zmq_errno ~ %s\n",
major, minor, patch
)
printf ( "EXPECT( NO ERROR ): zmq_errno ~ %s\n",
zmq_strerror ( zmq_errno() )
);
printf ( "EXPECT( NO ERROR ): zmq_send() ~ %s\n" );
rc = zmq_send ( req_socket, "A_TEST_BLOCK", 12, ZMQ_DONTWAIT );
printf ( "INF: zmq_send ( req_socket, "A_TEST_BLOCK", 12, ZMQ_DONTWAIT )\nZMQ: returned rc == %d which ouhgt be == 12, is it?\n",
rc
);
printf ( "EXPECT( NO ERROR ): zmq_errno ~ %s\n",
zmq_strerror ( zmq_errno() )
);
and feel free to file an issue, if unexpected results appear.