Wrong ip placement after live migration of instance

[Michel Nederlof]

When using the NB_ovn_bgp_driver, the placement of a floating ip after live migration is not determined properly.

During investigation we found out that the method responsible for determining the primary chassis is primarily looking at neutron:host_id field in external_ids, while the requested-chassis field in options column should overrule the placement of an lsp.

https://mail.openvswitch.org/pipermail/ovs-dev/2017-August/337623.html

I'll add some examples in the comments for live migration sequence

[Michel Nederlof (mnederlof)]

Original lsp (before live migration)

```
_uuid : 0eb19cff-b062-4eee-96b2-8d5119fb3ecf
addresses : ["fa:16:3e:a9:a6:75 2001:db8:1200::148 198.51.100.220"]
dhcpv4_options : b2afab41-f805-4445-bd55-df9271355f40
dhcpv6_options : 5119edfb-ea51-4c4e-81e9-66b56c4f7388
dynamic_addresses : []
enabled : true
external_ids : {"neutron:cidrs"="2001:db8:1200::148/64 198.51.100.220/22", "neutron:device_id"="6a6db2eb-bcc6-4737-9e33-396780e17b96", "neutron:device_owner"="compute:DC1", "neutron:host_id"=compute1, "neutron:network_name"=neutron-64903804-9685-4fda-8014-03b9fa439923, "neutron:port_capabilities"="", "neutron:port_name"="", "neutron:project_id"=d26351fd860647b48930e3be286ff825, "neutron:revision_number"="25", "neutron:security_group_ids"="5ce5434f-2798-4c24-b35e-adcc3e6c4821", "neutron:subnet_pool_addr_scope4"="", "neutron:subnet_pool_addr_scope6"="", "neutron:vnic_type"=normal}
ha_chassis_group : []
mirror_rules : []
name : "b0d84030-d4b9-48d3-8915-29f518b52bb9"
options : {}
parent_name : []
port_security : ["fa:16:3e:a9:a6:75 2001:db8:1200::148 198.51.100.220"]
tag : []
tag_request : []
type : ""
up : true
```

[Michel Nederlof (mnederlof)]

then, during live migration this field changes:

options : {activation-strategy=rarp, requested-chassis="compute1,compute2"}

and afterwards, the following fields have updated:

external_ids : {"neutron:cidrs"="2001:db8:1200::148/64 198.51.100.220/22", "neutron:device_id"="6a6db2eb-bcc6-4737-9e33-396780e17b96", "neutron:device_owner"="compute:DC1", "neutron:host_id"=compute2, "neutron:network_name"=neutron-64903804-9685-4fda-8014-03b9fa439923, "neutron:port_capabilities"="", "neutron:port_name"="", "neutron:project_id"=d26351fd860647b48930e3be286ff825, "neutron:revision_number"="34", "neutron:security_group_ids"="5ce5434f-2798-4c24-b35e-adcc3e6c4821", "neutron:subnet_pool_addr_scope4"="", "neutron:subnet_pool_addr_scope6"="", "neutron:vnic_type"=normal}
options : {requested-chassis=compute2}

[Michel Nederlof (mnederlof)]

https://review.opendev.org/c/openstack/ovn-bgp-agent/+/906111

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/ovn-bgp-agent/+/906111
Committed: https://opendev.org/openstack/ovn-bgp-agent/commit/6e0d5766501ad28dabc84941f84a2edbc976d0c8
Submitter: "Zuul (22348)"
Branch: master

commit 6e0d5766501ad28dabc84941f84a2edbc976d0c8
Author: Michel Nederlof <email address hidden>
Date: Fri Jan 19 14:51:24 2024 +0100

    Fix event handling for LSP and prefer the options.requested-chassis info

    Since the requested-chassis superseeds the placement in external_ids
    (which is managed by neutron), we should preferable use that instead of the
    value set by neutron (which _could_ lag or be wrong in specific scenario's)

    Also update logic for FIP handling to make migrations more efficient.

    Closes-Bug: #2049902

    Change-Id: I7f73a1ba7956f22e58fdde383775e88bf72cba14

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/2023.2
Review: https://review.opendev.org/c/openstack/ovn-bgp-agent/+/910305

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/ovn-bgp-agent/+/910305
Committed: https://opendev.org/openstack/ovn-bgp-agent/commit/16cfb301d1de00262067d73ee606401134aea2e9
Submitter: "Zuul (22348)"
Branch: stable/2023.2

commit 16cfb301d1de00262067d73ee606401134aea2e9
Author: Michel Nederlof <email address hidden>
Date: Fri Jan 19 14:51:24 2024 +0100

    Fix event handling for LSP and prefer the options.requested-chassis info

    Since the requested-chassis superseeds the placement in external_ids
    (which is managed by neutron), we should preferable use that instead of the
    value set by neutron (which _could_ lag or be wrong in specific scenario's)

    Also update logic for FIP handling to make migrations more efficient.

    Closes-Bug: #2049902

    Change-Id: I7f73a1ba7956f22e58fdde383775e88bf72cba14
    (cherry picked from commit 6e0d5766501ad28dabc84941f84a2edbc976d0c8)

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/ovn-bgp-agent 2.0.0.0rc1 release candidate.

[OpenStack Infra (hudson-openstack)]

Related fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/ovn-bgp-agent/+/913654

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/ovn-bgp-agent/+/913654
Committed: https://opendev.org/openstack/ovn-bgp-agent/commit/1bacff1dff825d267c390752fa21592c2e6f6588
Submitter: "Zuul (22348)"
Branch: master

commit 1bacff1dff825d267c390752fa21592c2e6f6588
Author: Michel Nederlof <email address hidden>
Date: Tue Mar 19 09:25:14 2024 +0000

    Fix placement of lsp when external_ids not in sync

    When options.requested-chassis is not in sync with
    external_ids.neutron:host_id it would pick both hosts, causing duplicate
    announcements from more than 1 host.

    This has been fixed in change 910305, but was left unchanged for the
    sync method, causing issues when the sync interval was re-evaluating all
    lsp's on the node.

    The code for determining the chassis of a port has been moved from the
    base_watcher to driver_utils so the logic for the event is the same as the
    logic when fetching the records from the northbound database.

    Related-bug: #2049902
    Change-Id: I545d6b41fd308eb56e5295657260718dc14868f7

[OpenStack Infra (hudson-openstack)]

Related fix proposed to branch: stable/2024.1
Review: https://review.opendev.org/c/openstack/ovn-bgp-agent/+/913508

[OpenStack Infra (hudson-openstack)]

Related fix proposed to branch: stable/2023.2
Review: https://review.opendev.org/c/openstack/ovn-bgp-agent/+/913509

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/ovn-bgp-agent/+/913509
Committed: https://opendev.org/openstack/ovn-bgp-agent/commit/c9b50da5badea46281dad6b7aaf61c41394cf422
Submitter: "Zuul (22348)"
Branch: stable/2023.2

commit c9b50da5badea46281dad6b7aaf61c41394cf422
Author: Michel Nederlof <email address hidden>
Date: Tue Mar 19 09:25:14 2024 +0000

    Fix placement of lsp when external_ids not in sync

    When options.requested-chassis is not in sync with
    external_ids.neutron:host_id it would pick both hosts, causing duplicate
    announcements from more than 1 host.

    This has been fixed in change 910305, but was left unchanged for the
    sync method, causing issues when the sync interval was re-evaluating all
    lsp's on the node.

    The code for determining the chassis of a port has been moved from the
    base_watcher to driver_utils so the logic for the event is the same as the
    logic when fetching the records from the northbound database.

    Related-bug: #2049902
    Change-Id: I545d6b41fd308eb56e5295657260718dc14868f7
    (cherry picked from commit 1bacff1dff825d267c390752fa21592c2e6f6588)

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/ovn-bgp-agent/+/913508
Committed: https://opendev.org/openstack/ovn-bgp-agent/commit/047d261cb7ea8f87cf33fdbce5a245106fcca335
Submitter: "Zuul (22348)"
Branch: stable/2024.1

commit 047d261cb7ea8f87cf33fdbce5a245106fcca335
Author: Michel Nederlof <email address hidden>
Date: Tue Mar 19 09:25:14 2024 +0000

    Fix placement of lsp when external_ids not in sync

    When options.requested-chassis is not in sync with
    external_ids.neutron:host_id it would pick both hosts, causing duplicate
    announcements from more than 1 host.

    This has been fixed in change 910305, but was left unchanged for the
    sync method, causing issues when the sync interval was re-evaluating all
    lsp's on the node.

    The code for determining the chassis of a port has been moved from the
    base_watcher to driver_utils so the logic for the event is the same as the
    logic when fetching the records from the northbound database.

    Related-bug: #2049902
    Change-Id: I545d6b41fd308eb56e5295657260718dc14868f7
    (cherry picked from commit 1bacff1dff825d267c390752fa21592c2e6f6588)