On Thu, Nov 19, 2020 at 12:53:45PM +0530, Numan Siddique wrote: > I have few comments w.r.t to scaling. > > 1. In the last OVN weekly irc meeting, I had mentioned about > ovn-northd-ddlog taking a lot of time in the loop. I also shared the > scale test results with ddlog > (https://imgur.com/JJpsMhv) and without ddlog (https://imgur.com/gmbhfoa). > > I was able to come up with a dirty script which you can run locally. > This script creates OVN resources > 250 times. As the scale increases you will notice the warning messages > in the ovn-northd-ddlog log file > > *** > timeval|WARN|Unreasonably long 1160ms poll interval (8ms user, 0ms system) > *** > > As the test progresses the time reported in the warning also increases > gradually and it reaches around 3500ms > in my setup. > > I do not observe such warning messages with the c version of northd. > > Can you please test it out locally with this script - > https://gist.github.com/numansiddique/1adf22cef57f6b125981b3581085694d
Thank you so much for the script. I didn't manage to work it in for v8, because it was so much work rewriting the whole database access layer and cross-porting all the changes in the meantime, but I will work it in and figure out what's going on for v9. > 2. ovn-northd-ddlog is missing support for connecting to a raft cluster. This should be fixed in v8. _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
