You'll have to check your crush rule to determine that.
ceph osd getcrushmap -o crushmap
crushtool -d crushmap -o crushmap.txt
vi crushmap.txt
check the rules near the end of that file. Rule 0 shows placement by host, and
rule 1 shows placemeny by osd. You can add another rule to your config and
then change any pool to use the other rule.
# rules
rule sandbox_host {
ruleset 0
type replicated
min_size 2
max_size 10
step take sandbox
step chooseleaf firstn 0 type host
step emit
}
rule sandbox_osd {
ruleset 1
type replicated
min_size 1
max_size 10
step take sandbox
step chooseleaf firstn 0 type osd
step emit
}
To put new rule into use:
crushtool -c crushmap.txt -o crushmap
ceph osd setcrushmap -i crushmap
example of how to change "rbd" pool to "rule 0"
ceph osd pool set rbd crush_ruleset 0
Hope this helps!
-----Original Message-----
From: ceph-users [mailto:[email protected]] On Behalf Of Rene
Hadler
Sent: Wednesday, November 26, 2014 7:15 AM
To: [email protected]
Subject: [ceph-users] Many OSDs on one node and replica distribution
Hi dear list,
i have a question about distribution of replicas on hosts with multiple
OSDs.
For example this configuration:
4x nodes
each node has 4 OSDs
replica set to 3
When i save now an object to the pool, how it is replicated? Is there a
chance that the original object and the 2 replicas are stored on the
same node?
I cant believe that but it was never clear for me. If yes and the node
fails this means the object is destroyed. Or should i set the replica
count to 5 to be sure?
Thanks.
--
Mit freundlichen Grüßen / Best regards
Rene Hadler
_______________________________________________
ceph-users mailing list
[email protected]
https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.ceph.com_listinfo.cgi_ceph-2Dusers-2Dceph.com&d=AAIGaQ&c=lXkdEK1PC7UK9oKA-BBSI8p1AamzLOSncm6Vfn0C_UQ&r=CSYA9OS6Qd7fQySI2LDvlQ&m=A5TPfHXvfTto86R6BmKuTG4NxfEeW9mYjqg3_xdw35E&s=HIbW9b0YvSjOvTgo2tj_JgEzkXkHUr8TYdm0BG4o1Qc&e=
_______________________________________________
ceph-users mailing list
[email protected]
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
