API_ZmqFailed exception thrown in ZmqEventSupplier::push_heartbeat_event leads to program termination

We got a core file generated on one of our Tango device servers (cppTango 9.3.3 running on Debian Stretch):

Here is the backtrace:

#0  __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:51
#1  0x00007fb8eb92942a in __GI_abort () at abort.c:89
#2  0x00007fb8ec2400ad in __gnu_cxx::__verbose_terminate_handler() () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#3  0x00007fb8ec23e066 in ?? () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#4  0x00007fb8ec23e0b1 in std::terminate() () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#5  0x00007fb8ec23e2c9 in __cxa_throw () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#6  0x000055ef49d87f19 in Tango::Except::throw_exception (reason=0x7fb8ee3a13d4 "API_ZmqFailed", 
    desc="Can't push ZMQ heartbeat event for event tango://mytangohost:10000/dserver/hostinfo/my_instance.heartbeat\nZmq error: Interrupted system call", origin=0x7fb8ee3a1a20 "ZmqEventSupplier::push_heartbeat_event", sever=Tango::ERR)
    at /segfs/tango/release/debian9/include/tango/except.h:595
#7  0x00007fb8ee3137cd in Tango::ZmqEventSupplier::push_heartbeat_event() () from /opt/os/lib/libtango.so.9
#8  0x00007fb8ee20dde8 in Tango::PollThread::eve_heartbeat() () from /opt/os/lib/libtango.so.9
#9  0x00007fb8ee208cd7 in Tango::PollThread::one_more_poll() () from /opt/os/lib/libtango.so.9
#10 0x00007fb8ee206acc in Tango::PollThread::run_undetached(void*) () from /opt/os/lib/libtango.so.9
#11 0x00007fb8ecba2731 in omni_thread_wrapper () from /opt/os/lib/libomnithread.so.4
#12 0x00007fb8ec538494 in start_thread (arg=0x7fb8e2ffd700) at pthread_create.c:333
#13 0x00007fb8eb9ddacf in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:97

It looks like an interrupted system call raised a ZMQ error and an exception got thrown in ZmqEventSupplier::push_heartbeat_event method. This exception is not caught at the upper levels (Polling Thread) and seems to lead to a crash of the device server.

terminate called after throwing an instance of 'Tango::DevFailed'

I have no idea why we got this system interrupted call in this specific case.

Maybe we could make the polling thread more robust for this kind of issues?