fre. 8. apr. 2022, 14:41 skrev Christian Stelter <[email protected]>:
> Hi! > > We are experiencing high latencies from time to time in one of our > openstack environments. While looking into the issue we found that > the southdb has frequent leader changes. > > We are running the Xena release and ovsdb-server (Open vSwitch) 2.16.0 > deployed via kolla-ansible. > > > We also noticed that the leader changes are caused by snapshots > (database compaction?). > Since this commit [0] introduced in OVS 2.16, transfer of leadership before snapshot is the intended behaviour. Before the change the leader would stop processing requests until the snapshot completed. Do these changes/snapshots have an impact on the performance of the southdb? > > Is there anything to improve the performance in case of leadership changes? > Until this patch [1], Neutron would require a connection to the leader for all threads talking to the SB DB. That lead to unesecary reconnection on leader change, which may be the source of the issues you see. Try it out and see if it helps. 0: https://github.com/openvswitch/ovs/commit/3c2d6274bceecb65ec8f2f93f2aac26897a7ddfe 1: https://review.opendev.org/q/I7442170d015f195a5430e71567fbc7d67b81d385 -- Frode Nordahl Kind regards, > > Christian Stelter > _______________________________________________ > discuss mailing list > [email protected] > https://mail.openvswitch.org/mailman/listinfo/ovs-discuss >
_______________________________________________ discuss mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
