[ovs-discuss] When ovn-controller restarts, packets was dropped during recomputing flows.

zhihui.ding at easystack.cn zhihui.ding at easystack.cn
Sat Oct 23 09:17:05 UTC 2021


Dear All

I'm newer and using ovn for openstack, and we has a problem. when ovn-controller
restarts, the ping has 2+ losed packets from a vm to other vm on different nodes. 

And I see ovn-kubernetes already discussed it in this link 
https://groups.google.com/g/ovn-kubernetes/c/PHf8KvWenfs 

This has some difficulty to resolve it,
1. How to detemine which time to clear older flows after recomputing is completely.
2. After restart, a meter or group table entry may use a different table-id. If not to clear 
extend tables, ovs will receive the worng flow action that not we excepted.


We try fix it by the following step:

1. When ovn-controller starts, not to clear flows and recomputing flows will replace olders.
2. Try to fetch existed meter and group by *_STATS_REQUEST before ofctrl state machine
changing to S_FLOW_CLEAR. 
3. When add-meter in function ofctrl-put, need check table-id, if meter-id already existed,
using OFPMC13_MODIFY to update it, otherwise, OFPMC13_ADD is still used.


I not sure the above whether has some issues,  and quite look forward to your comments.


Best regarids,
ZhiHui Ding



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openvswitch.org/pipermail/ovs-discuss/attachments/20211023/ac00ec85/attachment.html>


More information about the discuss mailing list