Hi,
You can increase the number of placement groups (PG) per pool (should be possible to do online) - keep in mind that this results in a rebalance (the data will get shifted around and increase the load on the network).
Choosing the appropriate number of PGs per pool can be quite tricky and depends on a few factors - number of pools, whether all OSDs are equal (if you mix SSDs and HDDs in a cluster and distribute the pools per device class you need to calculate the PGs per device class)....
The Ceph-documentation provides quite a thorough discussion, and you can use the PG-calculator to get a good overview of a sensible target value:
http://docs.ceph.com/docs/luminous/rados/operations/placement-groups/
https://ceph.com/pgcalc/
Hope this helps
You can increase the number of placement groups (PG) per pool (should be possible to do online) - keep in mind that this results in a rebalance (the data will get shifted around and increase the load on the network).
Choosing the appropriate number of PGs per pool can be quite tricky and depends on a few factors - number of pools, whether all OSDs are equal (if you mix SSDs and HDDs in a cluster and distribute the pools per device class you need to calculate the PGs per device class)....
The Ceph-documentation provides quite a thorough discussion, and you can use the PG-calculator to get a good overview of a sensible target value:
http://docs.ceph.com/docs/luminous/rados/operations/placement-groups/
https://ceph.com/pgcalc/
Hope this helps
It depends more on your CEPH Pools not Crush Map, but on the numbers of OSD and if you want to increase the capacity in the future.
As @Stoiko Ivanov already wrote: https://ceph.com/pgcalc/
There you can calculate how many PGs you need.
As @Stoiko Ivanov already wrote: https://ceph.com/pgcalc/
There you can calculate how many PGs you need.
It depends on your current pools, use the CEPH Calculator and check if its enough. You can increase the PGs without any Problems, but not decrease. If you want to decrease the PGs you need to add an new Pool with the correct PGs and move all of your Data into the new Pool.
Hello,
Thanks for your help so far, i am still a little concerned because looking at the script that calculator generates it makes a new pool
ceph osd pool create STORAGE 256
ceph osd pool set STORAGE size 3
while [ $(ceph -s | grep creating -c) -gt 0 ]; do echo -n .;sleep 1; done
It doesnt adjust the current one. - would this not break it and i would lose our entire production of VM's
Thanks for your help so far, i am still a little concerned because looking at the script that calculator generates it makes a new pool
ceph osd pool create STORAGE 256
ceph osd pool set STORAGE size 3
while [ $(ceph -s | grep creating -c) -gt 0 ]; do echo -n .;sleep 1; done
It doesnt adjust the current one. - would this not break it and i would lose our entire production of VM's
I also see on here that those commands are different then what is said on the CEPH page.
http://docs.ceph.com/docs/master/ra...nt-groups/#set-the-number-of-placement-groups
and i also have to adjust the pgp_num as well.
http://docs.ceph.com/docs/master/ra...nt-groups/#set-the-number-of-placement-groups
and i also have to adjust the pgp_num as well.
This script is meant for new installations (no pools exist).
Please check that you always use the docs for the installed version.
I assume Ceph luminous (12.2.10).
http://docs.ceph.com/docs/luminous/...nt-groups/#set-the-number-of-placement-groups
Yes, both need to be set. An increase in PGs will split the current PGs to produce new ones. If you want to do an alternative, then you could create a new pool and migrate your VM/CT to the new one (one by one).
>>or other way round (;-), why is it not used in the current version of PVE (no critisim, just want to know)?
you can use it with ceph nautilus on proxmox6
#ceph mgr module enable pg_autoscaler
you can use it with ceph nautilus on proxmox6
#ceph mgr module enable pg_autoscaler
i am aware of this. so i am wondering, if there may be "good reason" to not using it ...
It depends on your cluster setup, eg. resource consumption (data will be moved).
https://ceph.io/rados/new-in-nautilus-pg-merging-and-autotuning/
thanks for the link ... i "tried" it out and because of my "very small" system directly fall in same "warning" bug (ie. "overcommitted...") .. which will be fixed in 14.2.5 .... however .. because of my very small system and already fitting PGs "nothing really" happend