[ovs-discuss] lib/rconn.c:567: assertion version >= 0 && version <= 0xff failed in run_ACTIVE()

[张东亚]

Hi list,

We are now using unix stream socket (using openflowj as encoder/decoder of
OpenFlow message) to emulate ovs-ofctl behavior,  we occasionally trigger
the assertion above, and I have found other one have reported the same bug
in June this year.

After I have read the code and I can now reproduce this bug manually.

The reason is that our program may not send hello packet (due to we may
debugging which hang the JVM) quickly after the unix stream connection is
accepted by ovs-vswitchd, then the default max timeout 5s is trigger, since
no version is negotiated because ovs-vswitchd did not received hello yet,
it will raise the assertion.

My question is that assertion will make ovs-vswitchd aborts and exits, is
this behavior too strict ? should we handle this more gracefully like just
disconnect the connection and log it? Besides, if ovs-vswitchd accept a
connection and the other peer's connectivity is lost due to network, this
might also trigger the assertion due to the reason above.

Thanks a lot.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openvswitch.org/pipermail/ovs-discuss/attachments/20161205/6d489d49/attachment-0001.html>