thx Paul - I suspect these shadow trees causing this misbehaviour.
I have a second luminous cluster where these balancer settings work as expected
- this working one has hdd+ssd osds
i cannot use the upmap balancer because of some jewel-krbd clients - at least
they are being reported as jewel
Device classes are implemented with magic invisible crush trees; you've got
two completely independent trees internally: one for crush rules mapping to
HDDs, one to legacy crush rules not specifying a device class.
The balancer *should* be aware of this and ignore it, but I'm not sure
about the
Hi,
tried to enable the ceph balancer on a 12.2.12 cluster and got this:
mgr[balancer] Some osds belong to multiple subtrees: [0, 1, 2, 3, 4, 5, 6, 7,
8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27,
28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43,
