Feature #3480: ceph-deploy: remove osd - Ceph-deploy - Ceph

Feature #3480

ceph-deploy: remove osd

Added by Sage Weil about 11 years ago. Updated almost 4 years ago.

Status:

New

Priority:

High

Assignee:

Target version:

% Done:

Source:

Tags:

Backport:

Reviewed:

Affected Versions:

History

#1 Updated by Sage Weil about 11 years ago

Subject changed from ceph-deploy: remove osd, mon to ceph-deploy: remove osd
translation missing: en.field_position deleted (1)
translation missing: en.field_position set to 1

#2 Updated by Sage Weil almost 11 years ago

Category set to ceph-deploy

#3 Updated by Sage Weil almost 11 years ago

translation missing: en.field_position deleted (21)
translation missing: en.field_position set to 20

#4 Updated by Sage Weil over 10 years ago

Target version set to v0.61 - Cuttlefish

#5 Updated by Neil Levine over 10 years ago

Status changed from New to 12

#6 Updated by Ian Colle over 10 years ago

Target version changed from v0.61 - Cuttlefish to v0.62a

#7 Updated by Ian Colle over 10 years ago

Status changed from 12 to Need More Info

#8 Updated by Ian Colle over 10 years ago

translation missing: en.field_story_points set to 5.00

#9 Updated by Tamilarasi muthamizhan over 10 years ago

Assignee set to Dan Mick

#10 Updated by Ian Colle over 10 years ago

Priority changed from Normal to Urgent

#11 Updated by Dan Mick over 10 years ago

Why is this marked "need more info"?

#12 Updated by Ian Colle over 10 years ago

Because Neil wanted to know more about the user story. Are you working on this?

#13 Updated by Dan Mick over 10 years ago

It's mine, and Urgent, so I assume it's my priority 1...

#14 Updated by Dan Mick over 10 years ago

Talking with Sam, I think we agree this ought to be two commands (names possibly fluid)

ceph-deploy remove-begin osd
ceph-deploy remove-finish osd

The difficulty is that we want to wait for the cluster to rebalance between 'osd out' and
'osd dead'; having only one command sorta implies that it will wait, and that would mean
progress messages, and that wouldn't be very script-friendly. So the proposal is to do
"remove-begin osd", followed (perhaps) by some sort of cluster monitoring to see when it appears the osd is empty, followed by "remove-finish osd". "remove-finish" would block if the cluster appears not to have finished its recovery. (Today, "appears to have finished its recovery" will be defined as "all pgs reporting active+clean"; tomorrow, that might be "there are no PGs left on the specific osd that's shutting down".)

Presumably "remove-begin" will basically do ceph osd out, and
"remove-finish" will kill the OSD, remove it from CRUSH, remove the osd key from the cluster, remove any 'I'm ready' marker temp files from the OSD FS, and osd rm.

#15 Updated by Sage Weil over 10 years ago

how about

ceph-deploy osd drain HOST:DISK

and then

ceph-deploy osd remove HOST:DISK

i think the trick is that we should make 'remove' leave the disk in a state where it won't mount itself, and the osd id is freed up, but the data and cluster fsid is still there. and you could re-add it with