DeepSea icon indicating copy to clipboard operation
DeepSea copied to clipboard

filestore to bluestore migration resets reweight

Open Martin-Weiss opened this issue 5 years ago • 0 comments

Description of Issue/Question

We have a cluster where we had to do

ceph osd reweight-by-utilization

due to OSDs filling up "not 100 equal".

In this cluster we are now doing a filestore to bluestore migration using migrate.osds.

We had to realize that for each migrated OSD the reweight setting for the given OSD got lost where the expectation would have been to preserve it.

Is there any chance to get this fixed so that the filestore to bluestore migration does not "reset" the reweight?

Before the fs->bs migration for osd.0:

ceph osd tree ID CLASS WEIGHT TYPE NAME STATUS REWEIGHT PRI-AFF -1 361.52435 root default -2 54.17982 host ceph-p-01 0 hdd 0 osd.0 up 0.90002 1.00000

After the fs->bs migration for osd.0:

ID CLASS WEIGHT TYPE NAME STATUS REWEIGHT PRI-AFF -1 365.15631 root default -2 57.81181 host ceph-p-01 0 hdd 3.63199 osd.0 up 1.00000 1.00000

--> after the migration we should also see 0.90002

Setup

ses 5.5 with deepsea-0.8.6-2.21.1.noarch

Steps to Reproduce Issue

have data in a filestore based cluster run ceph osd reweight-by-utilization or adjust the reweight of the OSDs manually run filestore to bluestore migration

Martin-Weiss avatar Jan 11 '19 15:01 Martin-Weiss