fip will loss when it migrate between dvr-sant agent and dvr_no_external in Rocky

We hava 4 node,3 control node and 1 compute node ,our control node and network node are installed together, we use Rocky.

[TestCase]
Internal ip is 172.16.135.206
Float ip is 13.5.4.113

1,Create a vm with fip in control node
2,Shut down vm
3,Migrate this vm from control node to computer
4,Start up vm

[expect result]
Try to ping internet,fip worked,

[actually]
Fip does not work, fip can't ping internet successfully.

This nat rule is in snat ns

[root@CRH-KZ-3 neutron]# ip netns exec snat-3597ff2f-60c9-4310-b11b-3d808e63c4b9 iptables -t nat -S
-A POSTROUTING -j neutron-postrouting-bottom
-A neutron-l3-agent-OUTPUT -d 13.5.4.113/32 -j DNAT --to-destination 172.16.135.206
-A neutron-l3-agent-PREROUTING -d 13.5.4.113/32 -j DNAT --to-destination 172.16.135.206

I cannot find this fip 13.5.4.113
[root@CRH-KZ-3 neutron]# ip netns exec snat-3597ff2f-60c9-4310-b11b-3d808e63c4b9 ip a
4: qg-ecb0baea-0a@if10454455: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
    link/ether fa:16:3e:d7:bd:fb brd ff:ff:ff:ff:ff:ff link-netnsid 0
    inet 13.5.4.232/32 brd 13.5.4.232 scope global qg-ecb0baea-0a
       valid_lft forever preferred_lft forever
    inet 13.5.4.32/32 brd 13.5.4.32 scope global qg-ecb0baea-0a
       valid_lft forever preferred_lft forever
    inet 13.5.4.4/32 scope global qg-ecb0baea-0a
       valid_lft forever preferred_lft forever
    inet6 2022:419:1710:eeee::389/64 scope global nodad
       valid_lft forever preferred_lft forever
    inet6 fe80::f816:3eff:fed7:bdfb/64 scope link
       valid_lft forever preferred_lft forever

It seems fip loss during migration.If I add addr fip to qg-ecb0baea-0a,fip will work normally

I try to debug it.

Before migration,our vm with fip is in control node, all traffic will go through interface fg-5ff577fd-8c(mac addr fa:16:3e:ec:cc:f0) in fip namespace.
Below is our sw info.

ARM-R3-14-45U-SPINE-98.1>show arp | inc 13.5.4
Internet 13.5.4.113 1 fa16.3eec.ccf0(a interface in fip ns) ARPA vlan205 te0/7

Below is interface fg-5ff577fd-8c in fip ns
[root@CRH-KZ-3 ~]# ip netns exec fip-d3840bac-d92c-4fa7-beb3-6e39c403af84 ip a
2: fg-5ff577fd-8c@if10454106: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
    link/ether fa:16:3e:ec:cc:f0 brd ff:ff:ff:ff:ff:ff link-netnsid 0
    inet 13.5.4.70/24 brd 13.5.4.255 scope global fg-5ff577fd-8c
       valid_lft forever preferred_lft forever
    inet6 fe80::f816:3eff:feec:ccf0/64 scope link
       valid_lft forever preferred_lft forever

After migration,the same vm is in compute node,all traffic will go through qg-ecb0baea-0a(mac addr fa:16:3e:d7:bd:fb) in snat namespace.Because fip is not added into snat ns,I ping to internet ,this request traffic will go through snat ns, but relay through fip ns.
Below is our sw info.mac in sw does not change
ARM-R3-14-45U-SPINE-98.1>show arp | inc 13.5.4
Internet 13.5.4.113 1 fa16.3eec.ccf0(a interface in fip ns) ARPA vlan205 te0/7

Below is our cofiguration.

l3_agent.ini in control node

[root@CRH-KZ-3 neutron]# cat l3_agent.ini
[DEFAULT]
debug = True
interface_driver = neutron.agent.linux.interface.OVSInterfaceDriver
external_network_bridge =
ha_vrrp_auth_password = xxxx
interface_driver = openvswitch
agent_mode = dvr_snat
enable_metadata_proxy = false
ovs_use_veth = True
[agent]
extensions=fwaas,fip_qos
[ovs]

l3_agent.ini in compute node
[root@CRH-JS-7 ~]# cat /etc/neutron/l3_agent.ini
[DEFAULT]
debug = True
interface_driver = openvswitch
external_network_bridge =
ha_vrrp_health_check_interval = 30
agent_mode = dvr_no_external
enable_metadata_proxy = false
ovs_use_veth = True
[agent]
extensions=fip_qos
[ovs]
ovsdb_debug = true

[root@CRH-KZ-3 ~]# neutron agent-list | grep CRH-KZ-3
neutron CLI is deprecated and will be removed in the future. Use openstack CLI instead.
| 1119b4f8-b60a-47a2-9550-c0ae9837d73a | DHCP agent | CRH-KZ-3 | nova | :-) | True | neutron-dhcp-agent |
| 7051de00-be68-4f85-81d6-78c73ece87ef | Open vSwitch agent | CRH-KZ-3 | | :-) | True | neutron-openvswitch-agent |
| 91fcfb25-f38c-4e75-a203-ea46ec797781 | L3 agent | CRH-KZ-3 | nova | :-) | True | neutron-l3-agent |
| 943429e6-e88c-4ea7-923f-91e36684d24d | Metadata agent | CRH-KZ-3 | | :-) | True | neutron-metadata-agent |
[root@CRH-KZ-3 ~]#
[root@CRH-KZ-3 ~]#
[root@CRH-KZ-3 ~]#
[root@CRH-KZ-3 ~]# nova service-list | grep CRH-KZ-3
| 5491120c-da2b-41ee-91f7-896cc918a637 | nova-conductor | CRH-KZ-3 | internal | enabled | up | 2022-10-12T03:11:31.000000 | - | False |
| 036adf98-aa04-4532-ab92-e168b273469c | nova-consoleauth | CRH-KZ-3 | internal | enabled | up | 2022-10-12T03:11:23.000000 | - | False |
| 11b8a94c-5171-48b3-a782-fcf0b63ba621 | nova-scheduler | CRH-KZ-3 | internal | enabled | up | 2022-10-12T03:11:31.000000 | - | False |
| f13fef0a-65c6-4a8a-9903-de85c13e7671 | nova-compute | CRH-KZ-3 | sfdev-az-02 | enabled | up | 2022-10-12T03:11:24.000000 | - | False |
[root@CRH-KZ-3 ~]#

[root@CRH-KZ-3 ~]# neutron agent-list | grep CRH-JS-7
neutron CLI is deprecated and will be removed in the future. Use openstack CLI instead.
| 01ab3e58-693b-448e-823c-3dd08355bdcf | Metadata agent | CRH-JS-7 | | :-) | True | neutron-metadata-agent |
| 042cf399-ae41-4a27-9810-2fdd9f6d787f | Open vSwitch agent | CRH-JS-7 | | :-) | True | neutron-openvswitch-agent |
| 5b2d6292-b249-4e2a-b0a2-2980c773b3a4 | DHCP agent | CRH-JS-7 | nova | :-) | True | neutron-dhcp-agent |
| aa176740-f463-417a-b12e-7be2ba6e2e00 | L3 agent | CRH-JS-7 | nova | :-) | True | neutron-l3-agent |
[root@CRH-KZ-3 ~]#
[root@CRH-KZ-3 ~]#
[root@CRH-KZ-3 ~]# nova service-list | grep CRH-JS-7
| b797cf8f-8037-47a8-b9fe-479c4d04a4ca | nova-compute | CRH-JS-7 | sfdev-az-02 | enabled | up | 2022-10-12T03:12:21.000000 | - | False |
[root@CRH-KZ-3 ~]#