PDT/Scale: Took 20 minutes to converge traffic after restart ovsdb-server/Contrail2.2

[Anoop Kumar Sahu]

Were told with 2.2, this problem will go away..But the problem is still seen

Gnats:
1087763-1 edit PDT/Scale: Took 20 minutes to converge traffic after restart ovsdb-server/Contrail2.2

=========copied from Gnats ====

Please refer to the topology and spirent chart attached. Restarted ovsdb-server on Leaf L1 that is running bidirectional traffic with S4 (VNI 50-950).

      S4 ....OVSDB ....L1

Mac *03:05* *05:01*

Traffic stopped completely, took bouncy rides before converging after 20 minutes approx. Detail logs,ovsdbdump can be found at /volume/labcores/PR/PR-#

While no traffic Recv on L1, it was observed S4 didn't had ovsdb remote mac

root@vdc-vcf-s4# run show ovsdb mac remote | match 05:01 | count
Count: 3 lines

{master:0}[edit]
root@vdc-vcf-s4# run show ethernet-switching table | match 05:01 | count
Count: 5 lines

[chhandak (chhandak)]

Tried restarting ovsdb-server by cli as well as killing ovsdb-process by kill -9 command in QFX.
Mesured Broadcast and Unicast traffic loss. Observed following loss respectively

Brodcast 14-17 Sec (ARP Traffic. Tried 3 times)
Unicast 6-7 Sec (ICMP Traffic between BMS and Contrail VM. Tried 3 times)

[Hari Prasad Killi (haripk)]

Chhandak's observations above were with build 29.

Following were fixed by build 29:
1. SSL link between TOR-Agent and TOR was flapping due to errored packet send from TOR-Agent. This was fixed.
2. Few TOR-Agent crash fixes

With these fixes, the connection was stable and convergence is seen to be in the order of a few seconds.

[Anoop Kumar Sahu (anoops)]

The issue is still there and we see traffic converges after 4-5 minutes

[OpenContrail Admin (ci-admin-f)]

The convergence time will remain at this level for high scale. For lower scale this has been fixed as part of
https://bugs.launchpad.net/juniperopenstack/+bug/1456284

Pl retry.