Please do cell mapping without periodic task or active admin involvment

[Attila Fazekas]

'nova-manage cell_v2 discover_hosts' usage should not be needed,
the compute nodes at start up time could ask for registration
 and not repeat it during the n-cpu process lifetime.

In case of you have 30k compute node and restarting
them every two week in average, it would generate
~ 0.03 avg. request per sec.

I do not see why it is mandatory to use

nova-manage cell_v2 discover_hosts or
[scheduler]
discover_hosts_in_cells_interval = 300

even in a small deployment.

[Matt Riedemann (mriedem)]

discover_hosts_in_cells_interval is not mandatory, and discover_hosts can be run in deployment tooling when a new compute node is deployed. One of the goals of cells v2 is isolation from the cell to the API DB, so for the cell to be able to register itself with the API DB would require an "up-call" to the API DB to do so which violates that isolation.

[Dan Smith (danms)]

We've discussed this in the past and I think that this is unlikely to change in the future. For the reasons Matt has already said, this would require us to reverse course on something else we consider to be more important as it affects the security and stability at runtime, as opposed to operator convenience at the relatively infrequent point of adding compute nodes to the system.

[Attila Fazekas (afazekas)]

We have ~ 10k nova deployment per week in CI, so the
nova-manage cell_v2 discover_hosts is frequently used call.

I am working on PoC for faster CI installer and ATM moment nova
deployment is the longest (parallel, so I am waiting for the slowest one).

Is there any way to make the n-cpu first status report guaranteed to be fast and leaving
the later ones on the ~20 sec periodic ?

Is there a way to add multiple n-cpu nodes
to the DB without discovery and without relying on direct
database update ?