creating volume successfully after disabling cinder-scheduler service

Not able to verify it with devstack master. The volume remains in creating state. Marking Invalid.

Hi Swapnil Kulkarni, I have reproduced this issue in my devstack environment which has updated code and restart. it still has the problem, please recheck :)

Current behavior is expected.

'disabled' flag is only a tag in DB for services. The tricky part is, the service itself doesn't actually do anything when its status is being changed (via admin API). Only scheduler honors 'disabled' flag of OTHER services (in Cinder context, cinder-volume service). I think the purpose of introducing 'disabled' flag was to allow admin to easily put a service (cinder-volume, and similarly nova-compute) into/out of available resource pool. Scheduler will not consider disabled service when making resource placement decision.

It's funny when you try to disable scheduler itself. It updates the flag in DB, but no one actually looks at the flag for scheduler, which I think is fine. To me, there's not much value to 'disable' a schedule, it may help a bit when you are debugging multi-scheduler environment but to make it work like that way you want it, there is quite a bit of work to do (e.g. have c-api to notify a service that it's been disabled and to make service react such notification).

For the reason above, I'd suggest we invalid this bug or make it to wishlist.

Maybe just disallowing at the API level a disable of anything other than a cinder-volume service would be a good change for now?

@Duncan Yes, I agree.

Hi Huang Zhiteng, there is an scenario, if we have deployed multi-scheduler on multi-hosts, when we want to disable a scheduler service of a specified host, the disabling of cinder-scheduler may be valuable.

I agree with Duncan and Winston-D. I see how this might be valuable in the future but this is the type of thing that should have a Blueprint and some discussion around the implementation. Lets maybe look at Duncan's proposal for implementation and address the change for disabling the scheduler in the future.

Automatically unassigning due to inactivity.

This may have other implications now for active/active HA. Assigning to Gorka for now to get input on whether this kind of scenario is covered with the recent changed.

Unassigning due to no activity for > 6 months.