live-migration abort parameters are not honored

[Joseph Lanoux]

The live_migration_downtime, live_migration_downtime_steps and live_migration_downtime_delay default values are not honored.

Those parameters tune the live-migration abortion and must be higher than the LIVE_MIGRATION_DOWNTIME_MIN, LIVE_MIGRATION_DOWNTIME_STEPS_MIN and LIVE_MIGRATION_DOWNTIME_DELAY_MIN minimum values.

However, those parameters have default values that are higher than the minimum values but they are overridden by the minimum ones.

Steps to reproduce:
- live-migrate an instance
- simulate a downtime
- observe that the minimum values are used instead of the default ones.

[Augustina Ragwitz (auggy)]

live_migration_downtime, live_migration_downtime_steps and live_migration_downtime_delay represent maximum values. I don't see in the code where those values are overridden by the LIVE_MIGRATION_*_MIN values unless they are less than those minimum values. I only see the LIVE_MIGRATION_*_MIN values referenced in the files I've linked below:

Initializes the config values, sets defaults:
https://github.com/openstack/nova/blob/master/nova/conf/libvirt.py#L29
https://github.com/openstack/nova/blob/master/nova/conf/libvirt.py#L268

Checks that the max values aren't less than the min values:
https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L5985

In addition, I ran the test_live_migration_downtime_steps unit test and verified that the downtime values specified in the test weren't being overwritten. I also changed the values in the test to be less than the MIN values and they were overwritten as expected.

Can you provide more insight into when you are seeing this behavior? Could you perhaps attach your nova conf and some logs for when you see this behavior?

Or maybe someone else with more knowledge of live migration and libvirt can offer some insight on what could be happening here?

[Augustina Ragwitz (auggy)]

I'm going to mark this as Invalid for now. I asked in the nova irc channel and it looks like my previous comment is accurate. If you have further information per my comment above, please feel free to add it and reopen this bug.