haproxy-lbaas-agent goes out of sync with db after being down

[Simon]

How to reproduce:

1) create a pool with a VIP
2) check which lbaas-agent that pool is scheduled to
3) shut down lbaas-agent on the node
4) remove the pool/VIP via the API
5) restart lbaas-agent

It turns out the VIP would remain on the lbaas-agent host forever.

Looks like the instance_mapping variable is reset to empty restart the agent restarted, so it essentially lost track of that removed pool/vip.

[Eugene Nikanorov (enikanorov)]

Apparently you mean that haproxy process will remain. That's correct.
Agent has no knowledge of running haproxy processes unless it knows pool id, but it doesn't know it because the pool was deleted during the downtime.

The only option would be to scan state path for certain directories, parse ids and kill corresponding processes.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/81242

[Openstack Gerrit (openstack-gerrit)]

Reviewed: https://review.openstack.org/81242
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=d00e0b839378aed4c337efa1a33f6abbb384c149
Submitter: Jenkins
Branch: master

commit d00e0b839378aed4c337efa1a33f6abbb384c149
Author: Oleg Bondarev <email address hidden>
Date: Tue Mar 18 12:22:05 2014 +0400

    LBaaS: remove orphan haproxy instances on agent start

    This change adds remove_orphans() handling to the haproxy
    namespace driver. remove_orphans() is already called by
    lbaas agent on start for all drivers

    Closes-Bug: #1262885
    Change-Id: I5deae8e56c2cd2deb1667e9646633fd59a94b34e