data inconsistent in subscribe info in topology module

[hujie]

In current topology implementation, when a new VM is created and the ovs-interface port was notified to topology module, if it could not find the corresponding lport object from DF DB (for example, the db connection breaks down), topology will not invoke the _add_to_topic_subscribed(self, topic, lport_id) method, so that the subscribe info is inconsistent with the VMs on the host, if the VM is a new tenant VM on the host, the new tenant info will not be subscribed.

So we need a mechanism to make the subscribe info consistent(for example when the db connection is recovered and established again)

[hujie (mike-hu)]

I will add the bug fix implement in my new db consistency patch for local controller

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/336377

[Li Ma (nick-ma-z)]

I find it a little hard to reproduce the problem to demonstrate the code you submitted. Do you have any approaches?

[hujie (mike-hu)]

1. you can deploy the redis server independent and make the connection between the db and compute node break down, and then launch the vm;
2. you can break down the connection between redis and compute node, and then start a vm by libvirt and insert the corresponding lport info into df db.

Just a simple thought, and I agree it's a difficult situation to reproduce: )

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/336377
Committed: https://git.openstack.org/cgit/openstack/dragonflow/commit/?id=611e23c2833c6b9747ff99656ce79648d9426aa8
Submitter: Jenkins
Branch: master

commit 611e23c2833c6b9747ff99656ce79648d9426aa8
Author: hujie <email address hidden>
Date: Fri Nov 25 19:12:37 2016 +0800

    Implement db consistency logic in df local controller

    Change-Id: Iafe306116d6ad738a75cae2a640135411169ac2a
    Closes-bug: #1596455
    Partially-implements: blueprint keep-db-consistency