DescriptionLucas Alvares Gomes
2019-06-24 14:57:01 UTC
Description of problem:
For some deployments the default ovn-remove-probe-interval of 5000ms (5s) [0] is not enough.
If the ovn-controller takes more time recomputing the flows than the probe interval, chances are that it will try to reconnected to the SB OVSDB causing a full re-sync of the DB with it's in-memory replica (costy).
In order to change this default one can run the following command where the ovn-controller is running:
$ ovs-vsctl set open . external_ids:ovn-remote-probe-interval=<TIME IN MS>
We need to make it either configurable via Tripleo-Heat-Templates.
Perhaps even thing about a "saner" default to accommodate for bigger deployments which requires ovn-controller to process a lot of flows (e.g having a lot of SG and SG rules would result in a lot of flows being created for ACLs, specially in OSP 13 where we do not have Port Groups).
[0] https://github.com/openvswitch/ovs/blob/88d2ac50aa4e3383e185b698a1b3a44a6f7b4f80/ovn/controller/ovn-controller.c#L75
Version-Release number of selected component (if applicable):
OSP-13 and forward
Comment 1Lucas Alvares Gomes
2019-07-09 14:13:57 UTC
*** Bug 1716394 has been marked as a duplicate of this bug. ***