I would still always recommend having at least having n+1 failure domain in
any production cluster where n is your replica size.

On Tue, Jul 18, 2017, 11:20 PM David Turner <[email protected]> wrote:

> You do not need to empty the host before moving it in the crush map.  It
> will just cause data movement because you are removing an item under root
> and changing the crush weight of the rack.  There is no way I am aware of
> to really ease into this data movement other than to stare it head on and
> utilize osd_max_backfills to control disk io in your cluster.
>
> Are you changing you're failure domain to rack from host after this is
> done? Changing that in the crush map will cause everything to peer at once
> and then instigate a lot of data movement. You can do both moving the hosts
> into their racks and change the failure domain in the same update to only
> move data once. You would do that by downloading the crush map, modifying
> it, and then uploading it back into the cluster. It would be smart to test
> this on a test cluster. You could even do it on a 3 node cluster by
> changing each node to its own rack and setting the failure domain to rack.
>
> On Tue, Jul 18, 2017, 7:06 PM Mike Cave <[email protected]> wrote:
>
>> Greetings,
>>
>>
>>
>> I’m trying to figure out the best way to move our hosts from the
>> root=default bucket into their rack buckets.
>>
>>
>>
>> Our crush map has the notion of three racks which will hold all of our
>> osd nodes.
>>
>>
>>
>> As we have added new nodes, we have assigned them to their correct rack
>> location in the map. However, when the cluster was first conceived, the
>> majority of the nodes were left in the default bucket.
>>
>>
>>
>> Now I would like to move them into their correct rack buckets.
>>
>>
>>
>> I have a feeling I know the answer to this question, but I thought I’d
>> ask and hopefully be pleasantly surprised.
>>
>>
>>
>> Can I move a host from the root bucket into the correct rack without
>> draining it and then refilling it or do I need to reweight the host to 0,
>> move the host to the correct bucket, and then reweight it back to it’s
>> correct value?
>>
>>
>>
>> Any insights here will be appreciated.
>>
>>
>>
>> Thank you for your time,
>>
>> Mike Cave
>> _______________________________________________
>> ceph-users mailing list
>> [email protected]
>> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
>>
>
_______________________________________________
ceph-users mailing list
[email protected]
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

Reply via email to