yahoo-eng-team team mailing list archive

[Rodolfo Alonso <1946456@xxxxxxxxxxxxxxxxxx>]

Hello:

In [1] we removed the artifact "OVN_GATEWAY_INVALID_CHASSIS" used for
the LRP scheduler. The goal was not to create any scheduling for a LRP
if there are no valid GW chassis.

The answer for this proposal is the same: if there are no GW chassis
(that means, chassis with external connectivity) there is no reason to
define any HA_Chassis_Group to this external port, because the OVN
cluster won't be able to receive this external port traffic.

I'm closing this bug. Please feel free to reopen if needed.

Regards.

[1]https://review.opendev.org/c/openstack/neutron/+/909305

** Changed in: neutron
       Status: New => Won't Fix

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1946456

Title:
  [OVN] Scheduling of HA Chassis Group for external port does not work
  when no chassis has 'enable-chassis-as-gw' option set

Status in neutron:
  Won't Fix

Bug description:
  The 'enable-chassis-as-gw' CMS option allows the operator to influence
  where L3 routers and HA Chassis Groups for external ports are
  scheduled.

  When no chassis has the option set all chassis will be eligible for
  scheduling of L3 gateways [0].

  This is a powerful concept, and it gives operators fine grained
  control when needed, and lets the system decide when the fine grained
  control is not needed.

  However, this logic is not applied for HA Chassis Groups for use with
  the external port for servicing of SR-IOV instances.  If no chassis
  has the 'enable-chassis-as-gw' option set, no HA Chassis Group for use
  with external ports are created, and subsequently SR-IOV instances
  will not be serviced for DHCP or Metadata.

  While I understand the desire to require the use of the 'enable-
  chassis-as-gw' option and the simplification of the code that entails,
  it does introduce a unexpected difference in behavior from the
  scheduling of L3 gateways as documented in [0].  It also removes some
  of the power of the option.

  For example, when you have a deployment in a L3-only DC fabric and do
  not use features such as EVPN, you most likely want to schedule
  gateways only to chassis in physical vicinity of your border gateways.
  If you have a deployment in a L3-only DC fabric with EVPN, you would
  most likely not care where the gateways are scheduled and would like
  the system to schedule the gateways for you.  Other deployment
  scenarios are where you have a set of hypervisors you for some reason
  don't want any centralized services scheduled to them.

  The current behavior of the HA Chassis Group scheduling only caters
  for the latter of the above scenarios, and makes it more cumbersome
  for the user to support the two former.

  Would it be more consistent if Neutron chose among all chassis when no
  chassis has the 'enable-chassis-as-gw' option set also for the HA
  Chassis Group for external port feature?

  0:
  https://docs.openstack.org/neutron/latest/admin/ovn/routing.html#l3ha-
  support

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1946456/+subscriptions