Description of problem: Customers who install Neutron with VXLAN or GRE get poor network performance until they set the MTU of their guests to 1400 as described in KCS 786743 (https://access.redhat.com/solutions/786743). Could the puppet modules check if VXLAN or GRE is going to be installed and then implement the resolution found in KCS 786743? Version-Release number of selected component (if applicable): openstack-puppet-modules-2014.1-19.3.el7ost.noarch How reproducible: Deterministic. Steps to Reproduce: * Perform a packstack installation with a compute node separate from its network node which includes VXLAN or GRE. * Send network traffic from an instance on the compute node and observe log messages like "dropped over-mtu packet: 1550 > 1500" Actual results: * Guests on the compute node will have VXLAN or GRE overhead and some packets will be dropped because of the default MTU of 1500 on their NICs; resulting in poor network performance. Expected results: * Guests on the compute node will have VXLAN or GRE overhead but packets will not be dropped for this reason because the extra byes will fit within default MTU of 1500 on the compute node's NICs because the guest MTU will be automatically set 1400; resulting in normal network performance. Additional info: * This will save customers from having to deal with this problem after an installation and possibly reduce calls to GSS.
*** Bug 1138954 has been marked as a duplicate of this bug. ***
The neutron puppet module has a network_device_mtu parameter i think this change should be done in packstack and the other installers that use the module to install and configure neutron. Is this acceptable?
Having packstack and other installers set the network_device_mtu parameter in puppet seems reasonable to me. I will update bugs.
(In reply to Ivan Chavero from comment #3) > The neutron puppet module has a network_device_mtu parameter i think this > change should be done in packstack and the other installers that use the > module to install and configure neutron. > > Is this acceptable? I have done a test with network_device_mtu=1400. In this case, all ports created inside the network node netns use mtu=1400, however the guest mtu is still 1500. Would this be enough to fix the performance issue?
Just confirmed it is not enough, we need to set the guest MTU as specified in the KB article.
Proposed patch https://review.openstack.org/132888 for review.
-- Fedora Bugzappers volunteer triage team https://fedoraproject.org/wiki/BugZappers
Checked with version: openstack-packstack-puppet-2014.1.1-0.47.dev1288.el7ost.noarch openstack-packstack-2014.1.1-0.47.dev1288.el7ost.noarch
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://rhn.redhat.com/errata/RHBA-2015-1746.html