Details
-
Bug
-
Status: Resolved
-
Resolution: Done
-
None
-
None
-
None
-
Operating System: All
Platform: All
-
4957
Description
This problem generaly comes from architecture issue. Openflow Lithium plugin was basicaly designed without cluster.
In this context we are able to see a missing relationship between Transaction Chain (Distributed DataStore) and Role Context (especially OpenflowOwnershipListener). Actual implementation ignores fact that only MASTER should have active TransactionChain and only MASTER should write data to DataStore (in close scenario we have to be the last one who could clean Node from DataStore and that event has to come from Entity unregistration process).
This leads to unexpected faulty states of TxChain infrastructure and might cause connection being constantly rejected and it bring transaction submit failures due to concurent modification exception.
So solution needs architecutal redesing of TransactionChainManager lifecycle.
Attachments
Issue Links
- blocks
-
OPNFLWPLUG-579 flows not removed from operational ds when all switches disconnect
- Resolved
-
OPNFLWPLUG-591 [Clustering]: Openflow connections unstable with Lithium plugin
- Resolved
-
OPNFLWPLUG-595 Cluster Li plugin: controller does not show correct OF topology after owner recovers from failure
- Resolved
-
OPNFLWPLUG-596 Cluster Li plugin: controller does not clean operational inventory after a fail and recover cycle
- Resolved
-
OPNFLWPLUG-620 Li plugin: does not work with OF1.0
- Resolved
- is blocked by
-
CONTROLLER-1482 Missing notification request after registration
- Resolved