Public bug reported:
Our in-tree openvswitch agent now process the port as a serial way. This
may cause some delay operation and even effect the whole work, such as
the real vm startup..
Could we let l2 agent process more quickly or make it process the port
in concurrent mode? And I know some feature or framework need the
current way to process the info they concerned. But I think it is
necessary to promote the l2 agent performance.
Cases effected:
At first, we make the assumption that the openstack is running as public could
services. There are many uncontrolled api calles, including migration
instances, instances creation, etc..
Now we can focus on 1 compute node. If there are many requests from
neutron server, then server rpc to l2 agent and add the updated ports to
the process list(set). The target port we concerned is at the end of the
process list, make the assumption that there are 10 ports in the list,
and each port cost 2 seconds to process. The target port will be done
after 20s. So the whole process may delay 20s..This is just a example,
please don't care about the cost time.
Also, the instance migration, nova side is more quick than neutron, if
the target port also at the end of the process list, the whole time of
vm ready is very long.
** Affects: neutron
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1683566
Title:
[RFE] L2 agent need concurrent processing port
Status in neutron:
New
Bug description:
Our in-tree openvswitch agent now process the port as a serial way.
This may cause some delay operation and even effect the whole work,
such as the real vm startup..
Could we let l2 agent process more quickly or make it process the port
in concurrent mode? And I know some feature or framework need the
current way to process the info they concerned. But I think it is
necessary to promote the l2 agent performance.
Cases effected:
At first, we make the assumption that the openstack is running as public
could services. There are many uncontrolled api calles, including migration
instances, instances creation, etc..
Now we can focus on 1 compute node. If there are many requests from
neutron server, then server rpc to l2 agent and add the updated ports
to the process list(set). The target port we concerned is at the end
of the process list, make the assumption that there are 10 ports in
the list, and each port cost 2 seconds to process. The target port
will be done after 20s. So the whole process may delay 20s..This is
just a example, please don't care about the cost time.
Also, the instance migration, nova side is more quick than neutron, if
the target port also at the end of the process list, the whole time of
vm ready is very long.
To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1683566/+subscriptions
--
Mailing list: https://launchpad.net/~yahoo-eng-team
Post to : yahoo-eng-team@lists.launchpad.net
Unsubscribe : https://launchpad.net/~yahoo-eng-team
More help : https://help.launchpad.net/ListHelp
