Cluster module with pauseOnConnect failed transactions

[zhonked]

Hi,
I'm attempting to use the cluster module to implement sticky sessions for use with socket.io based on the recommendations from socket.io:
http://socket.io/docs/using-multiple-nodes/

and examples found here:
https://github.com/indutny/sticky-session
https://github.com/elad/node-cluster-socket.io

It appears however that something in the passing of sockets/fds between master and worker is causing failed transactions when run against siege.

This issue was discussed here:
elad/node-cluster-socket.io#10

Test code has been pushed here, unfortunately not in the form of a real unit test yet:
https://github.com/zhonked/node-cluster-sticky-test

Thanks for taking a look!

[cjihrig]

@zhonked are you able to reduce the test case?

If this is a manifestation of the same original problem, then this might need to be solved in the C++ layer. @bnoordhuis the problem is that accepted connections are read immediately. Is it possible to optionally forgo the uv_accept() call in TCPWrap and still pass the socket to another process. This would require exposing uv_accept() to JavaScript as well.