fg- device is not deleted after the deletion of the last VM on the compute node
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
neutron |
Fix Released
|
Undecided
|
Stephen Ma | ||
Juno |
Fix Released
|
Undecided
|
Unassigned |
Bug Description
The external gateway port in the fip- namespace on a compute node is not removed after the user deleted the last VM running on the node.
How to reproduce the problem:
1. SETUP:
* Use devstack to start up the controller node. In local.conf, Q_DVR_MODE=
* Use devstack to setup a compute node. In local.conf, Q_DVR_MODE=dvr.
At the start, there are no VMs hosted on the compute node. The fip namespace hasn't been created yet.
1. Create a network and subnet
2. Create a router and dd the subnet to the router
3. Tie the router to the external network
4. Boot up a VM using the network, and assign it a floatingip
5. Ping the floating IP (make sure you open up your SG)
6. Note the fg- device in the fip namespace on the compute node
7. Now delete the VM
Expected results:
- The VM is deleted.
- Neutron port-list shows the gateway port is also deleted.
- The FIP namespace is also cleared
Experienced results:
- The fg- device still remains in the fip namespace on the compute node and the fip namespace isn't removed.
For detailed command sequence, see:
Changed in neutron: | |
assignee: | nobody → Stephen Ma (stephen-ma) |
description: | updated |
description: | updated |
tags: | added: juno-backport-potential |
Changed in neutron: | |
milestone: | none → kilo-1 |
status: | Fix Committed → Fix Released |
Changed in neutron: | |
milestone: | kilo-1 → 2015.1.0 |
I did look at this very test case in the past, and I recall these related bug reports and fixes:
https:/ /bugs.launchpad .net/neutron/ +bug/1351066 /bugs.launchpad .net/neutron/ +bug/1367588
https:/
https:/ /review. openstack. org/#/c/ 120917/ /review. openstack. org/#/c/ 111421/
https:/
In my experience, namespaces were cleaned up correctly, but for that I needed router_ delete_ namespaces= True set for the L3 Agent.
Can you confirm that you have that too? That said, I believe that no more fixes should be filed against these issues if we don't have a functional test that demonstrated that the namespaces are indeed cleared. Clearly just unit coverage is not enough.