Race condition in deleting VPN routes learnt from BGP Peer
Affects | Status | Importance | Assigned to | Milestone | ||
---|---|---|---|---|---|---|
Juniper Openstack | Status tracked in Trunk | |||||
R2.20 |
Fix Committed
|
Medium
|
Nischal Sheth | |||
Trunk |
Fix Committed
|
Medium
|
Nischal Sheth |
Bug Description
There's a corner case during BgpServer shutdown which could
result in stale/leftover BGP VPN routes, which ultimately causes
an assertion when the BGP Peer which advertised those routes is
deleted.
The problem happens only if family route-target is negotiated on
the BGP Peer. If the BgpServer is shut down after the BGP Peer is
established, but before the EndOfRibMarker for the route-target
family is received, then the VPN routes received from that peer do
not get deleted.
This happens because the BGP Peer is not properly registered (in
the membership manager) to any VPN tables before EndOfRib is
received. However, we do process and install BGP routes that we
receive from the peer. During peer close, membership manager
only deletes routes from tables to which the peer is registered.
summary: |
- Race condition in deleting VPN routes from BGP Peers + Race condition in deleting VPN routes learnt from BGP Peer |
Review in progress for https:/ /review. opencontrail. org/11440
Submitter: Nischal Sheth (<email address hidden>)