DHCP Agent scheduled on locked host

[Nimalini Rasa]

Brief Description
-----------------
STX: some Vms not reachable after DOR

The issue seems to be in the Neuton scheduling of dhcp agent.
Expected:
controller-0:~$ sudo cat /etc/neutron/neutron.conf |grep network_scheduler_driver
Password:
# in combination with network_scheduler_driver =
# network_scheduler_driver is WeightScheduler, dhcp_load_type can be configured
#network_scheduler_driver = neutron.scheduler.dhcp_agent_scheduler.WeightScheduler
network_scheduler_driver=neutron.scheduler.dhcp_host_agent_scheduler.HostBasedScheduler

in STX:

controller-1:~$ sudo cat /etc/neutron/neutron.conf |grep network_scheduler_driver
Password:
# in combination with network_scheduler_driver =
# network_scheduler_driver is WeightScheduler, dhcp_load_type can be configured
#network_scheduler_driver = neutron.scheduler.dhcp_agent_scheduler.WeightScheduler
network_scheduler_driver=neutron.scheduler.dhcp_agent_scheduler.WeightScheduler

host-based scheduler prevents scheduling on locked hosts which is missing in STX.

Severity
--------
Major

Steps to Reproduce
------------------
Install STX load
Launch an instance
Lock a host
perform DOR

Expected Behavior
------------------
All VMs become reachable after powering up all host.

Actual Behavior
----------------
Some VMs were not reachable.

Reproducibility
---------------
Only done once.

System Configuration
--------------------
2+4+20 ( any 2+2 should be fine)

Branch/Pull Time/Commit
-----------------------
master as of 2018-10-10_20-18-00
The same behavior would happen in stx.2018.10 builds as well

Timestamp/Logs
--------------
Mon Oct 15 10:42:15 EDT 2018 (power on)
Mon Oct 15 14:55:28 2018 (1st VM pingable)

[Ghada Khalil (gkhalil)]

It appears that some code was removed from the puppet neutron config in stx as part of the initial starlingx submission. It's not clear why that was done.

This line was changed
from:
neutron::server::network_scheduler_driver: 'neutron.scheduler.dhcp_host_agent_scheduler.HostBasedScheduler'
to:
neutron::server::network_scheduler_driver: 'neutron.scheduler.dhcp_agent_scheduler.WeightScheduler'

File Reference: https://github.com/openstack/stx-config/blame/master/puppet-manifests/src/hieradata/controller.yaml#L273

The plan is to put it back.
Note: The code required in stx-neutron to support this config is still there.

[Ghada Khalil (gkhalil)]

Targeting stx.2018.10 - required code was removed, resulting in issues with dhcp agent scheduling

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/611411

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/611411
Committed: https://git.openstack.org/cgit/openstack/stx-config/commit/?id=ae87d881cf88ba3513d781d608fd68ff268ff340
Submitter: Zuul
Branch: master

commit ae87d881cf88ba3513d781d608fd68ff268ff340
Author: Teresa Ho <email address hidden>
Date: Wed Oct 17 13:41:00 2018 -0400

    Set neutron dhcp host agent scheduler to host-based

    The neutron network scheduler driver was set to the default dhcp
    agent weight scheduler. This update is to set it to the host-based
    scheduler.

    Change-Id: I9d7d62935e0c2c26f22879dd1e249ef8eeeadd5a
    Closes-bug: 1798096
    Signed-off-by: Teresa Ho <email address hidden>

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: r/2018.10
Review: https://review.openstack.org/611684