On 01/29/2018 01:14 PM, Niklas wrote:
Ceph luminous 12.2.2
$: ceph osd pool create hybrid 1024 1024 replicated hybrid
$: ceph -s
cluster:
id: e07f568d-056c-4e01-9292-732c64ab4f8e
health: HEALTH_WARN
Degraded data redundancy: 431 pgs unclean, 431 pgs
degraded, 431 pgs undersized
services:
mon: 3 daemons, quorum s11,s12,s13
mgr: s11(active), standbys: s12, s13
osd: 54 osds: 54 up, 54 in
data:
pools: 1 pools, 1024 pgs
objects: 0 objects, 0 bytes
usage: 61749 MB used, 707 GB / 767 GB avail
pgs: 593 active+clean
431 active+undersized+degraded
This can be solved if I set same weight 1.000 (or 50.000, or 100.000 or
..) manually on all items (hosts) in datacenter vDC, vDC1, vDC2 and vDC3
(see below). Problem is when adding new OSDs, the crush map is getting
new weights in these datacenters resulting in a broken cluster until
fixed manually.
*My question is, in a datacenter where it exist only 3 hosts, why is
ceph not mapping pgs to these 3 hosts? Clearly it has something to do
with the big weight difference between hosts but why?*
------------------------
Below is a simplified ceph setup of a Hybrid solution with NVMe and HDD
drivers. 1 copy on NVMe and 2 copies on HDD. Advantage is great read
performance and cost savings. Disadvantages is low write performance.
Still the write performance is good thanks to rockdb on Intel Optane
disks in HDD servers.
I have six servers in this virtualized lab ceph cluster.
Only NVMe drives on s11, s12 and s13.
Only HDD drives on s21, s22 and s23.
# buckets
host s11 {
# weight 0.016
alg straw2
hash 0
item osd.0 weight 0.002
item osd.1 weight 0.002
item osd.2 weight 0.006
item osd.3 weight 0.002
item osd.4 weight 0.002
item osd.18 weight 0.002
}
host s12 {
# weight 0.016
alg straw2
hash 0
item osd.5 weight 0.006
item osd.6 weight 0.002
item osd.7 weight 0.002
item osd.8 weight 0.002
item osd.9 weight 0.002
item osd.53 weight 0.002
}
host s13 {
# weight 0.016
alg straw2
hash 0
item osd.10 weight 0.006
item osd.11 weight 0.002
item osd.12 weight 0.002
item osd.13 weight 0.002
item osd.14 weight 0.002
item osd.54 weight 0.002
}
host s21 {
# weight 0.228
alg straw2
hash 0
item osd.15 weight 0.019
item osd.16 weight 0.019
item osd.17 weight 0.019
item osd.19 weight 0.019
item osd.20 weight 0.019
item osd.21 weight 0.019
item osd.22 weight 0.019
item osd.23 weight 0.019
item osd.24 weight 0.019
item osd.25 weight 0.019
item osd.26 weight 0.019
item osd.51 weight 0.019
}
host s22 {
# weight 0.228
alg straw2
hash 0
item osd.27 weight 0.019
item osd.28 weight 0.019
item osd.29 weight 0.019
item osd.30 weight 0.019
item osd.31 weight 0.019
item osd.32 weight 0.019
item osd.33 weight 0.019
item osd.34 weight 0.019
item osd.35 weight 0.019
item osd.36 weight 0.019
item osd.37 weight 0.019
item osd.38 weight 0.019
}
host s23 {
# weight 0.228
alg straw2
hash 0
item osd.39 weight 0.019
item osd.40 weight 0.019
item osd.41 weight 0.019
item osd.42 weight 0.019
item osd.43 weight 0.019
item osd.44 weight 0.019
item osd.45 weight 0.019
item osd.46 weight 0.019
item osd.47 weight 0.019
item osd.48 weight 0.019
item osd.49 weight 0.019
item osd.50 weight 0.019
}
datacenter vDC1 {
# weight 0.472
alg straw2
hash 0
item s11 weight 0.016
item s22 weight 10.000
item s23 weight 10.000
}
datacenter vDC2 {
# weight 0.472
alg straw2
hash 0
item s12 weight 0.016
item s21 weight 0.228
item s23 weight 0.228
}
datacenter vDC3 {
# weight 0.472
alg straw2
hash 0
item s13 weight 0.016
item s21 weight 0.228
item s22 weight 0.228
}
datacenter vDC {
# weight 1,416
alg straw2
hash 0
item vDC1 weight 0.472
item vDC2 weight 0.472
item vDC3 weight 0.472
}
# rules
rule hybrid {
id 1
type replicated
min_size 1
max_size 10
step take vDC
step choose firstn 1 type datacenter
step chooseleaf firstn 0 type host
step emit
}
Is it your intention to put all copies of a object in only one DC?
What is your exact idea behind this rule? What's the purpose?
Wido
Again, setting same weight like 1.000 OR 100.000 on all items in
datacenters vDC, vDC1, vDC2 and vDC3 makes the cluster work.
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
