Errors occur when run dragonflow with redis pub/sub enabled

[duankebo]

The following message is recorded periodically in the log:

2016-09-08 20:13:51.148 121698 INFO dragonflow.db.api_nb [-] Pushing Update to Queue: Action:dbrestart, Table:None, Key:None Value:False Topic:None
2016-09-08 20:14:01.156 121698 WARNING dragonflow.db.pubsub_drivers.redis_db_pubsub_driver [-] subscriber listening task lost connection string indices must be integers

[feipeng (feipeng1)]

yuli update redis jason process, so you need to update df code and restart neutron and df,to make them run same code for redis pub sub driver

[Hong Hui Xiao (xiaohhui)]

I have seen this error before when patch [1] just merged. I update the code, and restart the df-controller, the error show up. However, after restart neutron-server, the error has gone. And I never see the error again. So I think it should not be a bug now, unless more details are provided.

[1] 2f81efdf21557b810867b5eaa3bf477cab9b1b52

[duankebo (duankebo)]

It's due to inconsistency of dragonflow version of neutron server and compute node.
Pub/sub message sent by neutron server can't be parsed successfully by compute node.
Compute node then throw an exception like this:

dragonflow.db.pubsub_drivers.redis_db_pubsub_driver [-] subscriber listening task lost connection string indices must be integers

And generate a "dbstart" event and push it to northbound message queue:

dragonflow.db.api_nb [-] Pushing Update to Queue: Action:dbrestart, Table:None, Key:None Value:False Topic:None

Every time when receive a pub/sub mesg, these logs will be printed.

By updating to one same version, this problem can be solved.
So I think this bug can be closed.

[Omer Anson (omer-anson)]

In that case, this is not a bug. Neutron and Dragonflow have to be from the same branch, and both from the tip of branch.