[RFE] Distributed processing of OVSDB events in networking-ovn
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
networking-ovn |
Fix Released
|
High
|
Lucas Alvares Gomes |
Bug Description
Reported at: https:/
In networking-ovn, the OVSDB Monitor is responsible for listening the events OVSDB and performing certain actions on them. We do use it extensively on various tasks including critical ones such as monitor port binding events (in order to notify Neutron/Nova that a port has been bound to a certain chassis).
Currently, the OVSDB Monitor class uses a distributed OVSDB lock to make sure that only one instance is going to handle those events at a time. The problem with this approach is that it creates a bottle neck because, even tho we could have many neutron workers running only 1 is handling those OVSDB events.
The bottleneck problem is highlighted even more when working with other technologies such as containers which relies on creating batch of ports and waiting them to be bound to chassis in a performant fashion.
For this RFE we need to think about a mechanism that would allow multiple events from OVSDB to be handled in a distributed fashion to improve the overall performance of those event actions.
Changed in networking-ovn: | |
importance: | Undecided → High |
assignee: | nobody → Lucas Alvares Gomes (lucasagomes) |
tags: | added: networking-ovn-proactive-backport-potential |
tags: | added: networking-ovn-easy-proactive-backport-potential |
tags: | removed: networking-ovn-easy-proactive-backport-potential networking-ovn-proactive-backport-potential |
Related fix proposed to branch: master /review. openstack. org/652040
Review: https:/