Thank you Tyler,
That looks like exactly what I was looking for (now to test it in a test
rig) :-)
Cheers
Dulux-Oz
On 28/09/2022 07:16, Tyler Brekke wrote:
Hi Matthew,
You just have to take two steps when writing your crush rule. First
you want to get 3 different hosts, then you need 2 osd from each host.
ceph osd getcrushmap -o /tmp/crush
crushtool -d /tmp/crush -o /tmp/crush.txt
#edit it / make new rule
rule custom-ec-ruleset {
id 3
type erasure
min_size 4
max_size 6
step take your-root
step choose indep 3 type host
step chooseleaf indep 2 type osd
step emit
}
crushtool -c /tmp/crush.txt -o /tmp/crush.new
You can use the crushtool to test the mappings and make sure they are
working as you expect.
crushtool -i /tmp/crush.new --test --show-mappings --rule 3 --num-rep 6
You can then compare the OSD id and make sure its exactly what you are
looking for.
You can set the crushmap with
ceph osd setcrushmap -i /tmp/crush.new
Note: If you are overwriting your current rule, your data will need to
rebalance as soon as your set the crushmap, close to 100% of your
objects will move. If you create a new rule, you can set your pool to
use the new pool id anytime you are ready.
On Sun, Sep 25, 2022 at 12:49 AM duluxoz <duluxoz@xxxxxxxxx> wrote:
Hi Everybody (Hi Dr. Nick),
TL/DR: Is is possible to have a "2-Layer" Crush Map?
I think it is (although I'm not sure about how to set it up).
My issue is that we're using 4-2 Erasure coding on our OSDs, with
7-OSDs
per OSD-Node (yes, the Cluster is handling things AOK - we're
running at
about 65-70% utilisation of CPU, RAM, etc, so no problem there).
However, we only have 3 OSD-Nodes, and I'd like to ensure that
each Node
has 2 of each pool's OSDs so that if we loose a Node the other 2 can
take up the slack. I know(?) that with 4-2 EC we can loose 2 out
of the
6 OSDs, but I'm worried that if we loose a Node it'll take more
than 2
OSDs with it, rendering us "stuffed" (stuffed: a technical term
which is
used as a substitute for a four-letter word rhyming with "truck") 😁
Anyone have any pointers?
Cheers
Dulux-Oz
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
