Feature #3101
closed
allow "ceph osd crush set" without overwriting weight
Added by Anonymous over 11 years ago.
Updated over 11 years ago.
Description
Currently, a ceph-osd coming up gets its weight reset by src/upstart/ceph-osd.conf. This is not good. A node reboot should not override admin decisions about OSD weights.
There should be a way to adjust the OSD location in CRUSH without changing the weight.
Perhaps use a placeholder value like "nochange", "-" or "keep", to signal no change.
Related, why do we need to pass in both name and id?
ceph \
--cluster="${cluster:-ceph}" \
--name="osd.$id" \
--keyring="/var/lib/ceph/osd/${cluster:-ceph}-$id/keyring" \
osd crush set \
-- \
"$id" "osd.$id" "${weight:-1}" \
root=default \
...
Sage proposed this would be "ceph osd crush create-or-move".. I think the args could be "ID INITIAL_WEIGHT VAR=VAL.."
Sage, I'm taking back my "why don't we put the mon in charge of initial weight configurability" -- I see the upstart script is already looking at ceph.conf, we might as well use that: $osd_crush_weight. Perhaps we should rename that $osd_crush_initial_weight, though, and change the default embedded in src/upstart/ceph-osd.conf to 0 (and remove $osd_crush_weight).
- Tracker changed from Bug to Feature
- Category set to Monitor
- Status changed from New to Fix Under Review
- Target version set to v0.53c
- Assignee set to Sage Weil
- Translation missing: en.field_story_points set to 1
- Translation missing: en.field_position set to 1
- Translation missing: en.field_position changed from 1 to 1394
- Status changed from Fix Under Review to Resolved
Also available in: Atom
PDF