ZMQ sockets (zsock_t) do not work in child process after fork

[santupari]

Hi All,

We are observing an issue where in sockets created in child process do not work after forking.
Below is the sequence of code and I have also attached test code that reproduces issue and as well as working code:

1. Our design has a server which listens to multiple clients and responds
2. Clients uses wrapper API (example : test_tx_rx()) that exchanges messages with server
3. the issue observed when a client calls test_tx_rx() from parent and then fork and if you call test_tx_rx() from child it fails to communicate with server
4. The fix is to call zsys_shutdown() from child to clean up all initialization done by parent and assume that ZSYS is already initialized. However, as part of this fix we also noticed that there is a mutex that should be re-initialized to avoid deadlock. Hence, we are also attaching a patch that defines new API for re-initializing the mutex.

Please let us know if the patch is valid or how to avoid deadlock in case if the parents threads are accessing the mutex at the same when child is clearing mutex in zsys_shutdown().

Thanks,
Santosh Bidaralli

zsys_patch_wtih_test_code.zip

[bluca]

Adding a new API is not good for usability.

A better fix is to record the PID on init, and check it on shutdown - if it's different, then a fork happened and the mutex can be reinitialised.

Could you please send a pull request for this change, after testing that it works for your use case? Thanks!

[santupari]

Thank you @bluca!
I will make the suggested change in zsys_shutdown(). If my understanding is correct, defining a new static variable in zsys.c to store the parent pid should be sufficient to check it during the init and shutdown.
Please send me the instructions to submit a pull request.
Thanks,
Santosh Bidaralli

[bluca]

https://help.github.com/en/github/collaborating-with-issues-and-pull-requests

[santupari]

Thanks @bluca! I found better way of handling this issue and i have attached the patch with changes with which there is no need for application to call zsys_shutdown(). Please review these changes and let me know if i can submit a pull request for the same.
czmq_zsys_fork_issue_2038.zip

[bluca]

That looks fine, thanks

[santupari]

Hi @bluca,
We did thorough testing of the earlier patch in a multi-threaded environment and we were facing intermittent crashes. We worked to come up with a different patch altogether to make zsys_init() thread-safe and as well as handling fork conditions. I have attached the patch for review. Please review the same and let me know for any comments.
czmq_zsys_fork_issue_2038.zip

[raveendrapadala]

@bluca : We are facing similar Crash issue when an application is doing fork inside fork with ZMQ sockets in the parent process.
So is this patch going to be available in the next release? please confirm.

[bluca]

If someone sends a pull request, yes

[stale]

This issue has been automatically marked as stale because it has not had recent activity for 90 days. It will be closed if no further activity occurs within 21 days. Thank you for your contributions.