The vm can not access the vip of load balancer under DVR enviroment

Bug #1504726 reported by Zhou Zhihong on 2015-10-10
8
This bug affects 1 person
Affects Status Importance Assigned to Milestone
neutron
High
Swaminathan Vasudevan
Kilo
High
Unassigned

Bug Description

Version
=======
Kilo

Description
===========
The vip is on 192.168.1.0/24 subnet,and the vm is on 192.168.2.0/24 subnet. There is a router connected to the two subnets. For the computer node which the vm belong to, it's DVR l3-agent don't have ARP to the vip address. So the vm can not access the vip by route.

@Swami: can you triage please?

tags: added: l3-dvr-backlog
Changed in neutron:
assignee: nobody → Swaminathan Vasudevan (swaminathan-vasudevan)

will triage it.

This seems to have been reported in Kilo, there was bug fix that went in for the LBaaS dvr serviced port to update the arp. That fix went in the Liberty.
Was that fix back ported to kilo.

I did verify the fix that I mentioned was for LBaaS v2 and has nothing to do with the issue reported here.
Zhihong could you include the steps to reproduce this problem.

Zhou Zhihong (shuihushen) wrote :

The steps is below:
Step1: Create a subnet1 192.168.1.0/24
Step2: Create a vm1 on the subnet1 which ip is 192.168.1.2
Step3: Create a subnet2 192.168.2.0/24
Step4: Create a vm2 on the subnet2 which ip is 192.168.2.2
Step5: Create a router1 connect to the subnet1 and subnet2
Step6: Create a load balancer pool1(LBaaS v1) on subnet2
Step7: Pool1 add a member vm2, and add a vip1 which ip is 192.168.2.3
Step8: vm1(192.168.1.2) ping vm2(192.168.2.2) is OK
Step9: vm1(192.168.1.2) ping vip1(192.168.2.3) is fail ======> Here should be OK.

zhihong thanks for the adding in the steps to reproduce the problem.
I will take a look at it.

Changed in neutron:
importance: Undecided → High
Adolfo Duarte (adolfo-duarte) wrote :

Hi Zhihong,
could you upload your local.conf file from devstack configuration.
I am having problems reproducing the error. it would be easier if you just post your local.conf
Thankjs.

Zhou Zhihong (shuihushen) wrote :

Hi Adolfo,
I didn't use devstack. I build the openstack enviroment by manual and there're many nodes.
To reproduce the problem must use muilti-node which computer node and network node is indepent.
Thanks.

Carl Baldwin (carl-baldwin) wrote :

I had marked this as High. That importance was questioned in the L3 meeting today. It seems like interoperability between LBaaS and DVR is important and it is currently broken. If my understanding is correct, then I think it should be High. But, I'm open to discussion about lowering the priority to Medium.

I was not able to reproduce this problem in the master branch.
Can you check once again in the master branch.

Is this only seen in Kilo.

Zhou Zhihong (shuihushen) wrote :

Sorry, I only have Kilo enviroment. I don't know if the master branch has the problem.

@Swami: Kilo is still a maintained branch, we should try and identify how the issue got solved.

Yes we are looking into the problem in Kilo as well. I will update the status once I triage it.

ZongKai LI (zongkai) wrote :

Is this related to https://bugs.launchpad.net/neutron/+bug/1481613 ? This should be fixed, and its patch has been cherry picked to stable/kilo https://review.openstack.org/#/c/211383/ .

Hi Zhou Zhihong (shuihushen) can you confirm if you are using the Kilo/stable code post september 17th just to confirm that the fix is in your repo.

Kyle Mestery (mestery) wrote :

This is not valid in Kilo, as the issue is addressed per Swami in comment #10.

Kilo is under security only patches now, so it's not clear we can merge this there.

Changed in neutron:
status: New → Invalid
To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers