Ceph PG tuning

[Gastondc]

I ask for your help in trying to understand. Always a pleasure to come to this forum.

I've been reading the forum. I saw that in general all the answers paste the link: https://docs.ceph.com/en/latest/rados/operations/placement-groups/

But there is something I do not understand about the calculation of PG

in the documentation they say that by default they are 100 PG for each OSD. but i have 225 active and i have the warning


"1 pools have too many placement groups"



the ceph web calculator gives me much higher values, if I assign those my cluster is in warinng.

https://ceph.io/pgcalc/




In the image are all the pools. use 100% for the first 2 because a rule applies. The other pool has its own rule.


My configuration:


I have 3 ceph nodes.

I have 2 rules created, by disk types.

root @ pve01: ~ # ceph osd crush rule ls
replicated_rule
rule_1
rule_2

I have 18 OSD evenly distributed

root @ pve01: ~ # ceph osd ls | wc -l
18

root@pve01:~# cat /etc/pve/ceph.conf
[global]
auth_client_required = cephx
auth_cluster_required = cephx
auth_service_required = cephx
cluster_network = 192.168.30.3/24
fsid = 80e7521d-57fb-4683-9d67-943eef4a91b5
mon_allow_pool_delete = true
mon_host = 192.168.30.5 192.168.30.3 192.168.30.4
osd_pool_default_min_size = 2
osd_pool_default_size = 3
public_network = 192.168.30.3/24


┌───────────────────────┬──────┬──────────┬────────┬───────────────────┬──────────────────┬──────────────────────┬──────────────┐
│ Name │ Size │ Min Size │ PG Num │ PG Autoscale Mode │ Crush Rule Name │ %-Used │ Used │
╞═══════════════════════╪══════╪══════════╪════════╪═══════════════════╪══════════════════╪══════════════════════╪══════════════╡
│ ceph-rbd1 │ 3 │ 2 │ 32 │ warn │ rule_1 │ 0.0120256366208196 │ 129596165718 │
├───────────────────────┼──────┼──────────┼────────┼───────────────────┼──────────────────┼──────────────────────┼──────────────┤
│ ceph_rbd2 │ 3 │ 2 │ 128 │ warn │ rule_2 │ 0.00215974287129939 │ 39375153024 │
├───────────────────────┼──────┼──────────┼────────┼───────────────────┼──────────────────┼──────────────────────┼──────────────┤
│ cephfs01_data │ 3 │ 2 │ 32 │ on │ rule_1 │ 0.0104792825877666 │ 112755146742 │
├───────────────────────┼──────┼──────────┼────────┼───────────────────┼──────────────────┼──────────────────────┼──────────────┤
│ cephfs01_metadata │ 3 │ 2 │ 32 │ on │ rule_1 │ 6.70525651003118e-07 │ 7139132 │
├───────────────────────┼──────┼──────────┼────────┼───────────────────┼──────────────────┼──────────────────────┼──────────────┤
│ device_health_metrics │ 3 │ 2 │ 1 │ on │ replicated_rule │ 4.79321187185633e-08 │ 1360904 │
└───────────────────────┴──────┴──────────┴────────┴───────────────────┴──────────────────┴──────────────────────┴──────────────┘

 

[Alwin]

"1 pools have too many placement groups"

On of these pools has more PGs than the autoscaler thinks is necessary. Hence the warning. This is regardless of the actual amount of PGs on an OSD. Though 225 PGs is already a very high number.

 

[Gastondc]

On of these pools has more PGs than the autoscaler thinks is necessary. Hence the warning. This is regardless of the actual amount of PGs on an OSD. Though 225 PGs is already a very high number.

Thank you very much for the quick answer.

What I don't understand, in ceph's pg calculator. the result is 256 for the pool.

But I configure half (128) and throw the warning.

Some of this I do not understand. They are my first steps with Ceph.

 

[Alwin]

The autoscaler uses a factor of 3x by wich it will warn about an over-/under provisioning of PGs in a pool. The factor is caused by fill level of the pool.
https://docs.ceph.com/en/octopus/ra...nt-groups/#viewing-pg-scaling-recommendations