Loadbalancer missing logical switch association

[Luis Tomas Bolivar]

When a port (P1) in a Subnet (S1) tries to reach a member (P2 on S2) of a loadbalancer (LB1) through the loadbalancer VIP (P3 on S3), the traffic gets lost in it's way back from the member (P2) to the caller (P1).

The problem raises from the loadbalancer not being associated to the caller's subnet switch. For instance, the next command with make the request successful:

ovn-nbctl set logical_switch S1_logical_switch_id load_balancer=LB1_id

[Reedip (reedip-banerjee-deactivatedaccount)]

As per the discussion with Numan yesterday, an ideal way to handle the above issue might be to track the changes in a router.

- Use case #1: Members, VIPs and user are in the same subnet
- Use case #2: Network 1 and Network 2 are created , followed by router.After that LB is created on Network 1, Network 1 has VIP, Network 2 has Member and User putting the Curl Request.
- Use case #3: Network 1 followed by router.After that LB is created on Network 1, Network 1 has VIP. Network 2 is now created, connected to the router . Network 2 now has a Member for the VIP. User can be on Network 1/Network 2.
- Use case #4: Network 1 followed by router.After that LB is created on Network 1, Network 1 has VIP. Network 2 is now created, connected to the router . Network 2 now has a Member for the VIP. Network 3 is now created and connected to the router. Network 3 has the User for the curl request.

In Use Case 2-4 , router will know about all the changes in the network, so it would be easier if we can track all the networks associated to a router. In case a network has a LB and is associated with a new router, the LB can be associated with the router and any NEW network connected to the router.

NOTE: the same process has to be considered if the network interfaces are removed from a router :)

[Daniel Alvarez (dalvarezs)]

This can be tricky at scale. Imagine an scenario where we have a Logical Router with a large number of networks connected to it. One of that networks has an LB which is main use case is that Members, VIPS and user are in the same subnet. Still, because we don't know that this is the use case, we're going to add the load balancer to *all* of the networks connected to that router. Not sure how this can impact at scale but I'd say it's something to keep in mind.

[Daniel Alvarez (dalvarezs)]

So the idea is to tackle this by applying the LB to all LS connected to the LR and if this is a problem at scale we'll look at how to enhance it in core OVN so that the LB is only applied to the LR.

[Daniel Alvarez (dalvarezs)]

Plan is to monitor the Logical_Router table and on every update/delete event:

1. Calculate which LS are connected to it through the lrps present in the router.
2. Make sure that all LBs present in those LS are also added to the router.
3. Make sure that all those LBs are present also in every LS connected to the router.

Note: This ^ has to be also done on CREATE event (at connection/reconnection time because we'll get a dump from OVSDB for all existing LRs).

[Miguel Angel Ajo (mangelajo)]

You can also monitor the LSP/LRP tables to find out when a router port on a network is created.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/667642

[OpenStack Infra (hudson-openstack)]

Change abandoned by Maciej Józefczyk (<email address hidden>) on branch: master
Review: https://review.opendev.org/667642

[OpenStack Infra (hudson-openstack)]

Change abandoned by Reedip (<email address hidden>) on branch: master
Review: https://review.opendev.org/608862
Reason: Abandoning in favor of the new patch

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/668397
Committed: https://git.openstack.org/cgit/openstack/networking-ovn/commit/?id=aa4268a09470aea28f4a8c9f1062f32f057cb151
Submitter: Zuul
Branch: master

commit aa4268a09470aea28f4a8c9f1062f32f057cb151
Author: reedip <email address hidden>
Date: Thu Jun 27 14:12:56 2019 +0000

    Add a new Event Handler for Octavia

    This patch adds a new event handler for Octavia so that in
    the event that a network is added to a router, the loadbalancer
    on the network are also applied to the router and vice versa.

    The process involves:
    - Adding a new connection class for Octavia to communicate with
      OVN NB DB, setting up the connection and communicating with
      the IDL.
    - Setting up events for Router interface addition and deletion
    - Updating Loadbalancer tables and Router/Switch tables based
      upon the generated events.
    - Add network UUID to LRP external_ids

    Co-Authored-By: Terry Wilson <email address hidden>
    Co-Authored-By: Maciej Józefczyk <email address hidden>

    Change-Id: If10016325517e4a1fc5f7cfac7f7bd58deb09b54
    Closes-Bug: #1794260

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/stein
Review: https://review.opendev.org/672645

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/672645
Committed: https://git.openstack.org/cgit/openstack/networking-ovn/commit/?id=c5169327ec13e7fb02ee2346a70f8666c725ab2a
Submitter: Zuul
Branch: stable/stein

commit c5169327ec13e7fb02ee2346a70f8666c725ab2a
Author: reedip <email address hidden>
Date: Thu Jun 27 14:12:56 2019 +0000

    Add a new Event Handler for Octavia

    This patch adds a new event handler for Octavia so that in
    the event that a network is added to a router, the loadbalancer
    on the network are also applied to the router and vice versa.

    The process involves:
    - Adding a new connection class for Octavia to communicate with
      OVN NB DB, setting up the connection and communicating with
      the IDL.
    - Setting up events for Router interface addition and deletion
    - Updating Loadbalancer tables and Router/Switch tables based
      upon the generated events.
    - Add network UUID to LRP external_ids

    Co-Authored-By: Terry Wilson <email address hidden>
    Co-Authored-By: Maciej Józefczyk <email address hidden>

    Change-Id: If10016325517e4a1fc5f7cfac7f7bd58deb09b54
    Closes-Bug: #1794260
    (cherry picked from commit aa4268a09470aea28f4a8c9f1062f32f057cb151)

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/networking-ovn 7.0.0.0b1 development milestone.

[OpenStack Infra (hudson-openstack)]

Related fix proposed to branch: stable/rocky
Review: https://review.opendev.org/697558

[OpenStack Infra (hudson-openstack)]

Related fix proposed to branch: stable/queens
Review: https://review.opendev.org/697559

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/697559
Committed: https://git.openstack.org/cgit/openstack/networking-ovn/commit/?id=eef8fa3933b46c7c0cfe40539c5f1d12db14139f
Submitter: Zuul
Branch: stable/queens

commit eef8fa3933b46c7c0cfe40539c5f1d12db14139f
Author: Flavio Fernandes <email address hidden>
Date: Thu Dec 5 14:43:27 2019 -0500

    Add Neutron network id to OVN LRPs

    This patch is adding an external_id to the Logical_Router_Port
    rows including the Neutron network ID.

    The reason for this is that if we consume Octavia from Stein+ in
    Rocky/Queens, this external id is missing and the OVN Octavia
    driver (octavia_api) will not be able to apply the Load Balancers
    to all the Logical Switches and Logical Routers where needed.

    Note: This is a partial cherry-pick from
    https://review.opendev.org/#/c/668397/ since Octavia itself
    does not exist in stable/queens but can be made to work
    with neutron's stable/queens.

    Change-Id: I4a0614afda8d2c2b8ccbb2583bc339a232fc802e
    Co-Authored-By: Daniel Alvarez <email address hidden>
    Closes-Bug: #1855328
    Related-Bug: #1794260
    (cherry picked from commit aa4268a09470aea28f4a8c9f1062f32f057cb151)

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/697558
Committed: https://git.openstack.org/cgit/openstack/networking-ovn/commit/?id=5fae8f68a033d66838fbecec3796e37d0ccdd60d
Submitter: Zuul
Branch: stable/rocky

commit 5fae8f68a033d66838fbecec3796e37d0ccdd60d
Author: Flavio Fernandes <email address hidden>
Date: Thu Dec 5 14:43:27 2019 -0500

    Add Neutron network id to OVN LRPs

    This patch is adding an external_id to the Logical_Router_Port
    rows including the Neutron network ID.

    The reason for this is that if we consume Octavia from Stein+ in
    Rocky/Queens, this external id is missing and the OVN Octavia
    driver (octavia_api) will not be able to apply the Load Balancers
    to all the Logical Switches and Logical Routers where needed.

    Note: This is a partial cherry-pick from
    https://review.opendev.org/#/c/668397/ since Octavia itself
    does not exist in stable/rocky but can be made to work
    with neutron's stable/rocky.

    Change-Id: I4a0614afda8d2c2b8ccbb2583bc339a232fc802e
    Co-Authored-By: Daniel Alvarez <email address hidden>
    Closes-Bug: #1855328
    Related-Bug: #1794260
    (cherry picked from commit aa4268a09470aea28f4a8c9f1062f32f057cb151)

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/networking-ovn 6.0.1 release.