[calico] Birdc configuration isn't changed after modification of environment
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| Fuel Plugins |
Fix Released
|
Undecided
|
Emma Gordon | ||
Bug Description
Fuel ISO# MirantisOpenSta
Description:
Plugin doesn't install birdc on nodes without controller or compute role. However if we have a node with other role (CephOSD for example), birdc is configured to establish peering with this node.
Enviroment:
OS: Ubuntu
Compute: QEMU
Networking Setup: Neutron with vlan segmentation
Ceph storage: No
Additional services: all by default
Steps to reproduce:
1. Install calico plugin
2. Deploy env with calico and following nodes: 1 controller, 1 compute + cinder.
3. Login to controller node and check birdc parameters.
4. Add one compute node to env and deploy changes.
5. Login to controller node and check birdc parameters.
Observed behaviour:
Birdc configuration on controller node doesn't contain information about new compute node:
Controller:
bird> show protocols
name proto table state since info
device1 Device master up 10:20:45
bgp1 BGP master up 10:20:49 Established <---- This is first compute node. We don't have configuration for the added node.
Added compute node:
bird> show protocols
name proto table state since info
kernel1 Kernel master up 12:21:36
device1 Device master up 12:21:36
direct1 Direct master up 12:21:36
N1 BGP master start 12:21:36 Active Socket: Connection closed
| tags: | added: calico |
| Changed in fuel-plugins: | |
| status: | New → Confirmed |
| assignee: | nobody → Emma Gordon (emma-0) |
| Changed in fuel-plugins: | |
| milestone: | none → 6.1 |
| Changed in fuel-plugins: | |
| status: | Confirmed → In Progress |
| Emma Gordon (emma-0) wrote | #1 |
| Changed in fuel-plugins: | |
| status: | In Progress → Fix Released |
Fixed, pending code review - https:/
Note that the bird configuration will be updated immediately when a new compute node is added, but when a compute node is deleted this will not be picked up until the next time that a node (of any role) is added to the deployment. This is a result of a fuel bug whereby /etc/hiera/
This behaviour will resolve itself when that Fuel bug is fixed and I propose that it is liveable with until then (the only side effect being that bird will have timed out bgp sessions to deleted compute nodes - note this should not cause any problems with running out of tcp sockets since the next time a node is added to the deployment the deleted nodes will be noticed and the connections to them will be closed).