zmq_ctx_destroy() hangs with no way to find out why

[mrvn]

On exit I'm unregistering all sockets from the poller, closing all sockets and finally destroying the context. The last part frequently blocks. I looked at the code over and over and everything seems to be in order and nothing forgotten. I even set LINGER to 0 for sockets.

So how can one find out why zmq_ctx_destroy() blocks? Or how can one destroy the context anyway without blocking?

[bluca]

What kind of sockets? What transport? Have you tried using GDB to find out where it's blocked?

[mrvn]

ROUTER sockets with tcp, inproc for ZAP and monitor socket on the ROUTER. The hangs started when I added the monitor socket.

gdb isn't helpfull without debug infos and I don't have them here at the moment:

#0 0x00007ffff78e412d in poll () at ../sysdeps/unix/syscall-template.S:81
#1 0x00007ffff62d447a in ?? () from /usr/lib/x86_64-linux-gnu/libzmq.so.4
#2 0x00007ffff62bfb90 in ?? () from /usr/lib/x86_64-linux-gnu/libzmq.so.4
#3 0x00007ffff62b0c1c in ?? () from /usr/lib/x86_64-linux-gnu/libzmq.so.4
#4 0x00007ffff550ac4b in ?? ()
from /usr/lib/python3/dist-packages/zmq/backend/cython/context.cpython-34m-x86_64-linux-gnu.so
#5 0x00007ffff550b137 in ?? ()
from /usr/lib/python3/dist-packages/zmq/backend/cython/context.cpython-34m-x86_64-linux-gnu.so

My thoughts now are that closing the sockets generates some message on the monitoring socket that it waits to deliver. But nobody every fetches it.

[bluca]

libzmq.so.4 - that is quite old, version 4.0.x I think?

This looks like a race condition with the monitor socket that was fixed a while ago, can you try with the latest version?

[mrvn]

Will do so after this release. Don't want to chance introducing new bugs by changing the lib now.

[bjovke]

@mrvn Any news about this? Did you try the newer version?

[lab313ru]

This problem happens if you didn't close sockets that were opened at this context.

[stale]

This issue has been automatically marked as stale because it has not had activity for 365 days. It will be closed if no further activity occurs within 56 days. Thank you for your contributions.

[maixuanhan]

This problem happens if you didn't close sockets that were opened at this context.

And maybe because some data lingering. Setting ZMQ_LINGER may help

Read the description of this for more detail
https://linux.die.net/man/3/zmq_ctx_destroy