VM send DHCP request before openflow is created in br-int
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Compute (nova) |
Expired
|
Low
|
Unassigned |
Bug Description
I am using OpenStack queens set up by OpenStack-Ansible.
When I create a VM, it seems to be started and sends DHCP request just before openflow is created.
This will result in a two minutes delay for VM to be accessible.
Test is carried out with Cirros image, as pasted below, VM can't get IP in the first round "sending discover".
Is there a way to add some delay when booting a VM ? Nova will not check openflow before starting the VM ?
info: initramfs: up at 0.64GROWROOT: CHANGED: partition=1 start=16065 old: size=64260 end=80325 new: size=2072385,
info: /etc/init.
modprobe: module virtio_net not found in modules.dep
WARN: /etc/rc3.
Initializing random number generator... done.
Starting acpid: OK
cirros-ds 'local' up at 0.76
no results found for mode=local. up 0.78. searched: nocloud configdrive ec2
Starting network...
udhcpc (v1.20.1) started
Sending discover...
Sending discover...
Sending select for 192.168.11.113...
description: | updated |
can you provide more information on what neutron network backend you are using.
nova has had a feature where we wait for vif-plugged event from neutron on boot before /blueprints. launchpad. net/nova/ +spec/admin- event-callback- api
starting the guest. that feature was implemneted in icehouse as part of this bluepint
https:/
some ml2 backends such as odl send the vif-plugged event on port bind instead of when they have completed
wireing up the port. .i.e odl send the vif-plugged event before it installs the required openflow rules.
i belive later versions of odl and networking odl actully send the vif-plugged after the openflow rules are installed but for queens
i do not think this is the case.
if you are use ml2/ovs or ml2/linux-bridge you should see this behavior.
i this is likely not a nova bug but rather a case of the ml2 driver sending the vif-plugged event before it has acutlly
finished wiring up the port intoducing a race.
if you are using ml2/ovs and can provide nova-compute and nutron agent logs for the event we can reasses.
for now i will mark this as incomplete and recommend that we close as in valid if futher info cannot be provided
as there is insuffent info currently to fully triage.