[zeromq-dev] Unexpected routing ID with ZMQ_ROUTER_HANDOVER

[Eric Badger]

Hi zeromq-dev,

I'm running some (pyzmq) client/server code with a server ROUTER
socket which sets ROUTER_HANDOVER. The clients are DEALERs which all
set their ROUTING_ID manually to a string value. However, I find that
in some cases during handover, the ROUTER side will see a message with
a routing ID that is not what the client set. It seems to be an ID
generated internally by ZMQ. Is this an expected case that a ROUTER
using ROUTER_HANDOVER must handle? Or is something misbehaving here?

Here's a simple client/server example that I find can generally show
this behavior within a couple of tries:

https://pastebin.com/WVJF3sxj

To run it, start a server, and then 2 batches of 4 clients like so:

$ ./serv.py
$ for i in $(seq 1 2); do for j in $(seq 1 4); do (./client.py $j &) ;
done ; sleep 3; done

Usually, the server will see the routing ID as set, printing something
like this:

raw routing id: b'myroutingid4'

But sometimes, it will get a non-string value like so:

raw routing id: b'\x00k\x8bEi'
Traceback (most recent call last):
  File "./serv.py", line 21, in <module>
    print("decoded routing id: {}".format(wid.decode()))
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x8b in position
2: invalid start byte

I've found this behavior on these test systems:
- CentOS 7.7, Python 3.6.8, libzmq 4.3.2, pyzmq 18.1.1
- Fedora 30, Python 3.7.5, libzmq 4.3.2, pyzmq 18.1.1
- Ubuntu 16.0.4, Python 3.5.2, libzmq 4.3.2, pyzmq 18.1.0

Any insights appreciated.

Thanks,
Eric