GARP request from Fabric interface is not getting flooded to VM's
Affects | Status | Importance | Assigned to | Milestone | ||
---|---|---|---|---|---|---|
Juniper Openstack | Status tracked in Trunk | |||||
R2.20 |
Fix Committed
|
High
|
Divakar Dharanalakota | |||
Trunk |
Fix Committed
|
High
|
Divakar Dharanalakota |
Bug Description
Setup:
AAP associated to two VMs which runs keepalived with garp
The VMs are distributed across two computes.
Agent listens to vrrp mcast and adjusts the local preference accordingly however the garp request to update the arp table of the other vms in the network is lost. Looks like the garp broadcast pkts gets dual encapsulated (vxlan over mplsogre)
The preferred encap mode is set to mplsogre.
Hi Praveen, Anand,
Got a chance to try the same (garp on aap) and it looks like we encap it twice (vxlan over mplsogre).
GARP is sent for 77.7.7.6 from 02:ad:a8:06:1c:ce on compute host 5.5.5.16 to compute host 5.5.5.15
00:24:09.602302 IP (tos 0x0, ttl 64, id 47912, offset 0, flags [none], proto GRE (47), length 110)
5.5.5.16 > 5.5.5.15: GREv0, Flags [none], length 90
MPLS (label 4103, exp 0, [S], ttl 64)
mplsogre(5.5.5.16 - 5.5.5.15):
0x0000: 4500 006e bb28 0000 402f ab10 0505 0510
0x0010: 0505 050f 0000 8847 0100 7140 0000 0000
vxlan:
0x0020: 4500 004e bb27 0000 4011 ab4e 0505 0510
0x0030: 0505 0510 c2cd 12b5 003a 0000 0800 0000
garp:
0x0040: 0000 0700 ffff ffff ffff 02ad a806 1cce
0x0050: 0806 0001 0800 0604 0001 02ad a806 1cce
0x0060: 4d07 0706 ffff ffff ffff 4d07 0706
In case if you wanna take a look at the setup.
Mgmt IP: 10.102.42.67 - 5.5.5.15
Metadata IP: vrrp master: 169.254.0.6 on 5.5.5.16
- Senthil
From: Senthilnathan Murugappan
Sent: Thursday, July 23, 2015 11:22 AM
To: Praveen K V
Subject: RE: Contrail and GARP
Hi Praveen,
I tried vrrp+garp with aap a while back and had observed that garp is consumed by vrouter and never sent to all hosts in the VN.
Hope its not expected, will check the same again and update you.
- Senthil
summary: |
- GARP request gets dual encapsulated (vxlan over mplsogre) + GARP request from Fabric interface is not getting flooded to VM's |
gcore of the agent is at /cs-shared/ bugs/1478662/