Flows and nodes are not cleaned up from the operational data store when mininet disconnects from the controller.
This might be related to this : https://bugs.opendaylight.org/show_bug.cgi?id=3085
This is likely fixed - correct Kamal?
Too early to say abhijit, the patch i submitted might fix this, as i have encountered the same issue many times.
There is one scenario, which vaclav mentioned, which might need some reconsideration to the patch i submitted. Still in progress.
Jan,
Is this still happening and is it still a blocker?
Thanks,
Abhijit
Hi Jan,
we have a job that tries to scale using mininet, and it's able to get to 300 switches before it runs in to trouble at 400. Part of the test is checking that all switches are gone after mininet is stopped, so at least with 300 we are seeing that they are removed.
just fyi.
JamO
and, as I looked closer the reason the test did not scale to 400 switches was because there were
some that did not get removed after stopping mininet when using a linear topology of 400.
just more fyi 
Hi Jamo,
Are you running your tests in a 3 node cluster?
On a 3 node cluster, i have seen a single mininet not performing with more than 250+ nodes.
I divided the nodes into 3 mininet vms and things were considerably smooth
Patch to be reviewed: https://git.opendaylight.org/gerrit/#/c/21915/
The above patch might solve this issue. Please re-test.
I have seen that duplicate NodeRemoved notifications causes the transaction chain to fail, which would cause other deserving node removals to fail well.
I tried with 300 nodes and all the nodes got cleaned up.
Can you please re-test with the lithium plugin?
can we mark this fixed/resolved?
the switch scalability test in CI is passing for stable/lithium:
https://jenkins.opendaylight.org/releng/view/openflowplugin/job/openflowplugin-csit-periodic-1node-cds-scalability-daily-only-stable-lithium/plot/Inventory%20Scalability/
that loops from 100,200..500 switch mininet topos and tears the network down each iteration. If nodes
are left over after stopping mininet, the test would fail. so, even after a network of 500, it was ok.
the same job is there for the lithium-redesign plugin and that made it past 400 ok. Unfortunately,
I think mininet was too slow with doing 500 and the job aborted. But, 400 was ok.
Good to know.