Confusing Error status if deleting ceph osd node

[Roman Sokolkov]

MOS 9.0

While removing node with Ceph OSD user gets "Error" saying that some manual procedure should be done, before removing Ceph OSD node.

See this bug: https://bugs.launchpad.net/fuel/+bug/1424060

1) This should be a Warning not Error. (This is very confusing, user will decide that smth already wrong with his env)
2) Manual procedure should be documented in User guide.

Steps to reproduce:
1) Install MOS9.0 with 3 Ceph nodes
2) Try to remove 1 Ceph node and Deploy changes

Actual result:
Error
Deployment has failed. Check these nodes:
'Untitled (94:7c)', 'Untitled (43:42)', 'Untitled (f3:90)'
Ceph data still exists on: node-4. You must manually remove the OSDs from the cluster and allow Ceph to rebalance before deleting these nodes.

Expected result:
Warning, cluster should be in Operational state.
Ability to find manual procedure in Mirantis documentation.

[Dmitry Pyzhov (dpyzhov)]

In Newton release we'll be able to remove OSD automatically with LCM tools. But for previous releases we should add pre-removal warning for user with explanation that removal will fail without extra manual changes.

[Roman Sokolkov (rsokolkov)]

My bad regarding document, it was available

https://docs.mirantis.com/openstack/fuel/fuel-9.0/operations.html#how-to-safely-remove-a-ceph-osd-node

[Dmitry Ukov (dukov)]

Any updates here?
Is it going to be fixed?
This bug affects scale down ceph cluster in Fuel which means that this feature does not work.

[Vitaly Kramskikh (vkramskikh)]

Decided not to introduce extra warning on the UI side - it won't improve the UX much without removing the block on nailgun/library side

[Oleksiy Molchanov (omolchanov)]

Right now we need feature to be implemented to nailgun, as far as I know it is ready, but not sure it goes to 9.x

[Rodion Tikunov (rtikunov)]

Blueprint related to this bug: https://blueprints.launchpad.net/fuel/+spec/ceph-osd-remove

[Vladimir Kuklin (vkuklin)]

This change would actually require an essential change of Astute code piece, although it does not affect cluster functionality directly. I would mark it as Medium and fix with a proper data-driven node deletion mechanism in the upcoming releases

[Oleksiy Molchanov (omolchanov)]

Won't fix because of data-driven deletion mechanism is not going to be implemented.