Hi guys,

We experienced some problems with CMS when watching nb_cfg/sb_cfg up.
A problem is that inappropriate time delay happens between nb_cfg bump 
and sb_cfg bump.
A research shows that the reason is double run of northd engine combined 
with improper sequence of calls
engine <-> counter update.

Exactly, making the change in logical network structure combined with 
nb_cfg bump results in the following
execution sequence:

1. 'update3' notification of changed made in ovn-nbctl
2. engine recompute run, (long time here)
3. 'transact' request to NB DB + SB DB, send results of engine's work
4. 'update3' + 'reply' received. this update3 fires back our changes we 
just sent to SB DB.
5. engine recompute second run, (long time here, because change handles 
are not implemented in branch-22)
6. Call update_sequence_numbers(), sb_cfg is sent to NB DB.

Thus, time delay between p4 - p6 need to be eliminated.

For main branch, where all change handlers are implemented, p.5 takes a 
short time so, problem is not so visible.


I have prepared a patch that is going to fix this problem in our 
environment, made over branch-22.09, see in attachment.

So guys, if you want to have it in official ovn repo, let me know, I 
will post the patch to patchwork.

best regards,

Alexander




_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to