I'm using Devstack to bring up one control+compute and another compute.
Devstack configures the external network and router before the second compute (and it's OVS) is configured

When the OVS on the second compute goes up, it is missing the group entry responsible for external network. (group id is 200000, external port is OF port 1)

OVS1

OFPST_GROUP_DESC reply (OF1.3) (xid=0x2):
group_id=200000,type=all,bucket=actions=set_field:00:1c:73:4e:d3:31->eth_dst,load:0x200->NXM_NX_REG6[],resubmit(,220)
group_id=209999,type=all,bucket=actions=set_field:0x1->tun_id,resubmit(,55),bucket=actions=set_field:0x4->tun_id,resubmit(,55)
group_id=150002,type=all,bucket=actions=set_field:fa:16:3e:3e:e9:1e->eth_dst,load:0x400->NXM_NX_REG6[],resubmit(,220)
group_id=150000,type=all,bucket=actions=set_field:fa:16:3e:65:08:3f->eth_dst,load:0x100->NXM_NX_REG6[],resubmit(,220)
group_id=210000,type=all,bucket=actions=group:209999,bucket=actions=set_field:0x1388->tun_id,output:4
group_id=210002,type=all,bucket=actions=group:210001,bucket=actions=load:0x200->NXM_NX_REG6[],resubmit(,220)
group_id=210001,type=all

cookie=0x8000007, duration=1191.945s, table=220, n_packets=890, n_bytes=210, priority=7,reg6=0x200 actions=output:1
cookie=0x8000007, duration=1191.564s, table=220, n_packets=50, n_bytes=5204, priority=7,reg6=0x100 actions=output:2
cookie=0x8000007, duration=720.422s, table=220, n_packets=69, n_bytes=7093, priority=7,reg6=0xe0000400 actions=output:3
cookie=0x6900000, duration=720.422s, table=220, n_packets=69, n_bytes=7093, priority=6,reg6=0x400 actions=load:0xe0000400->NXM_NX_REG6[],write_metadata:0xe000040000000000/0xfffffffffffffffe,goto_table:251

OVS2

OFPST_GROUP_DESC reply (OF1.3) (xid=0x2):
group_id=200001,type=all,bucket=actions=output:3
group_id=150001,type=all,bucket=actions=set_field:fa:16:3e:9c:c2:be->eth_dst,load:0x300->NXM_NX_REG6[],resubmit(,220)
group_id=209999,type=all,bucket=actions=set_field:0x3->tun_id,resubmit(,55)
group_id=210000,type=all,bucket=actions=group:209999,bucket=actions=set_field:0x1388->tun_id,output:3

cookie=0x6900000, duration=659.040s, table=220, n_packets=70, n_bytes=7163, priority=6,reg6=0x300 actions=load:0xe0000300->NXM_NX_REG6[],write_metadata:0xe000030000000000/0xfffffffffffffffe,goto_table:251
cookie=0x8000007, duration=659.039s, table=220, n_packets=70, n_bytes=7163, priority=7,reg6=0xe0000300 actions=output:2

Analysis of the class ExternalNetworkGroupInstaller shows that it is triggered by one of two events:
1. gw mac changed
2. subnet (of the external network, i guess) changed

My assumption is that the first OVS is configured properly because it is up before configuring the external network.

Couldn't find any code that handles new OVS after external network is configured.

This issue blocks VMs on this OVS from using floating IP, and may also block all VMs using NAPT if this OVS is chosen as the "NAPT SWITCH".