Merge "Macvtap: Add more details restrictions with migration"

This commit is contained in:
Jenkins
2016-03-10 02:57:04 +00:00
committed by Gerrit Code Review

View File

@@ -54,9 +54,28 @@ Known limitations
resources like DHCP, Routers and others is not supported. Therefore run
either OVS or linux bridge in VLAN or flat mode on the controller node.
* Live Migration requires the same physical_interface_mapping configuration on
each host. It is not possible that physnet1 uses eth1 on host1, but eth2
on host2.
* Migration requires the same ``physical_interface_mapping`` configuration on
each host. It is not recommended to use different mappings, like
node1 uses ``physnet1:eth1`` but node2 uses ``physnet1:eth2``. Having
different mappings could
* cause migration to fail, if the interface configured on the source node
does not exist on the target node
* result in an instance placed on the wrong physical network, if the
interface used on the source node exists on the target node, but is used
by another physical network or not used at all by OpenStack. Such an
instance does not have access to its configured networks anymore.
It then has layer 2 connectivity to either another OpenStack network, or
one of the hosts networks.
.. note::
To get around those problems, make sure that your
``physical_interface_mapping`` is in sync between all nodes using the
macvtap agent. This `bug
<https://bugs.launchpad.net/neutron/+bug/1550400>`_ tracks the progress
on overcoming this limitation.
* Only centralized routing on the network node is supported (using either
the Open vSwitch or the Linux bridge agent). DVR is NOT supported.
@@ -104,12 +123,6 @@ require an IP address range because it only handles layer-2 connectivity.
.. image:: figures/scenario-classic-mt-services.png
:alt: Service layout
.. note::
For VLAN external and project networks, the physical network infrastructure
must support VLAN tagging.
OpenStack services - controller node
------------------------------------