He Plugin - Disconnecting and Re-connection of network causes flow loss in the network

In this use case, 100k flows are programmed into a simulated network (mininet) with 15 nodes. Then, mininet is closed down and started again. This should result in all 100k flows programmed into the network. However, I only see between 55k and 65k programmed into the network, never 100k.

There seem to be two issues here:

• There is one or two switches that do not have any flows at all - this seems to be related to session re-establishment; this problem also happens in the Li OF plugin

• All other switches do not have all their flows programmed; this only happens with the He OF plugin.

To reproduce, use the following steps:

1. Start mininet using the following command:

"sudo mn --controller=remote,ip=192.168.162.1:6633 --topo tree,4 --switch ovsk,protocols=OpenFlow13"

Let mininet connect to the controller

2. From another shell, use the flow_config_blaster script (located in the integration test project under 'test/tools/odl-mdsal-clustering-tests/clustering-performance-test') to program 100k flows into the controller:

./flow_config_blaster.py --threads 8 --flows 12500 --auth --no-delete --fpr 200

3. In the mininet test VM, use a script to make sure that all 100k flows have been programmed into the network:

/bin/get-total-found.sh

(or something similar); make sure all 100k flows have been injected into mininet. You may also check the controller's REST API, but collecting all the 100k flows takes a long time (> 10 minutes).

4. Stop mininet

5. Start mininet again:

"sudo mn --controller=remote,ip=192.168.162.1:6633 --topo tree,4 --switch ovsk,protocols=OpenFlow13"

6. In the mininet test VM, use a script to check whether all flows have been programmed from the switches:

/bin/get-total-found.sh

The sum of all flow counts on all switches should now be 100k. However, I see a much lower number - around 55-65k flows in total.