PG数计算

原地址:http://xiaqunfeng.cc/2017/09/15/too-many-PGs-per-OSD/

ceph告警问题:”too many PGs per OSD” 的解决方法,以及pg数量的合理设定

现象

1
2
3
4
5
6
7
8
9
10
11
12
13
# ceph -s
cluster 4c7ec5af-cbd3-40fd-8c96-0615c77660d4
health HEALTH_WARN
too many PGs per OSD (412 > max 300)
monmap e2: 3 mons at {ceph0=172.21.1.21:6789/0,ceph1=172.21.1.22:6789/0,ceph2=172.21.1.23:6789/0}
election epoch 1780, quorum 0,1,2 ceph0,ceph1,ceph2
mgr active: ceph0 standbys: ceph1, ceph2
osdmap e94: 6 osds: 6 up, 6 in
flags sortbitwise,require_jewel_osds,require_kraken_osds
pgmap v161317: 824 pgs, 10 pools, 30201 MB data, 8642 objects
90831 MB used, 181 GB / 269 GB avail
824 active+clean
client io 34800 B/s wr, 0 op/s rd, 9 op/s wr

原因

  • 集群osd 数量较少
  • 搭建rgw网关、OpenStack、容器组件等,pool创建较多,每个pool默认需要占用一些pg,pool中pg数目设置不合理,导致集群 total pg 数过多

解决

方法

调整每个osd默认pg数,参数为 mon_pg_warn_max_per_osd ,当前默认参数如下:

1
2
# ceph --show-config | grep mon_pg_warn_max_per_osd
mon_pg_warn_max_per_osd = 300

步骤

1、修改ceph配置文件

1
2
3
4
5
# cd /etc/ceph
# vim ceph.conf
[global]
.......
mon_pg_warn_max_per_osd = 500

2、将配置文件推到mon所在的其他节点

1
# ceph-deploy --overwrite-conf config push ceph1 ceph2

3、重启mon进程

1
# systemctl restart ceph-mon.target

重启成功后,再次查看配置项

1
2
# ceph --show-config | grep mon_pg_warn_max_per_osd
mon_pg_warn_max_per_osd = 500

此时集群状态ok

1
2
3
4
5
6
7
8
9
10
11
12
# ceph -s
cluster 4c7ec5af-cbd3-40fd-8c96-0615c77660d4
health HEALTH_OK
monmap e2: 3 mons at {ceph0=172.21.1.21:6789/0,ceph1=172.21.1.22:6789/0,ceph2=172.21.1.23:6789/0}
election epoch 1780, quorum 0,1,2 ceph0,ceph1,ceph2
mgr active: ceph0 standbys: ceph1, ceph2
osdmap e94: 6 osds: 6 up, 6 in
flags sortbitwise,require_jewel_osds,require_kraken_osds
pgmap v161317: 824 pgs, 10 pools, 30201 MB data, 8642 objects
90831 MB used, 181 GB / 269 GB avail
824 active+clean
client io 34800 B/s wr, 0 op/s rd, 9 op/s wr

pg数目的设定

信息查询

1、查看当前osd数目

1
2
# ceph osd ls | wc -l
6

2、查看当前有多少个pool

1
2
# ceph osd pool ls | wc -l
10

3、查看 replication pool 的数量

1
2
3
4
5
6
7
8
9
10
11
# ceph osd dump | grep repli
pool 0 'rbd' replicated size 3 min_size 2 crush_ruleset 0 object_hash rjenkins pg_num 256 pgp_num 256 last_change 31 flags hashpspool stripe_width 0
pool 1 '.rgw.root' replicated size 3 min_size 2 crush_ruleset 0 object_hash rjenkins pg_num 8 pgp_num 8 last_change 14 flags hashpspool stripe_width 0
pool 2 'default.rgw.control' replicated size 3 min_size 2 crush_ruleset 0 object_hash rjenkins pg_num 8 pgp_num 8 last_change 16 flags hashpspool stripe_width 0
pool 3 'default.rgw.data.root' replicated size 3 min_size 2 crush_ruleset 0 object_hash rjenkins pg_num 8 pgp_num 8 last_change 19 flags hashpspool stripe_width 0
pool 4 'default.rgw.gc' replicated size 3 min_size 2 crush_ruleset 0 object_hash rjenkins pg_num 8 pgp_num 8 last_change 20 flags hashpspool stripe_width 0
pool 5 'default.rgw.lc' replicated size 3 min_size 2 crush_ruleset 0 object_hash rjenkins pg_num 8 pgp_num 8 last_change 21 flags hashpspool stripe_width 0
pool 6 'default.rgw.log' replicated size 3 min_size 2 crush_ruleset 0 object_hash rjenkins pg_num 8 pgp_num 8 last_change 23 flags hashpspool stripe_width 0
pool 7 'default.rgw.users.uid' replicated size 3 min_size 2 crush_ruleset 0 object_hash rjenkins pg_num 8 pgp_num 8 last_change 26 flags hashpspool stripe_width 0
pool 8 'kube' replicated size 3 min_size 2 crush_ruleset 0 object_hash rjenkins pg_num 256 pgp_num 256 last_change 36 flags hashpspool stripe_width 0
pool 9 'stage' replicated size 3 min_size 2 crush_ruleset 0 object_hash rjenkins pg_num 256 pgp_num 256 last_change 54 flags hashpspool stripe_width 0

可以看到,每个pool都是 3 副本(replicated size 3

total pg num

公式如下:

1
Total PGs = (Total_number_of_OSD * 100) / max_replication_count

结果必须取最接近该数的 2 的幂

比如,根据以上信息:

1
2
3
Total_number_of_OSD = 6
max_replication_count = 3
Total PGs = 200

最接近 200 的 2 的幂是 256。所以推荐的集群最大 total PGs 数为 256。

pool pg num

每个 pool 的 pg 数目计算:

1
Total PGs = ((Total_number_of_OSD * 100) / max_replication_count) / pool_count

结果同样要取最接近的 2 的幂。

对应该例,每个 pool 的 pg num 为:

1
2
pool_count = 10
Total PGs = 200 / 10 = 20

所以每个 pool 的平均分配 pg num 为 16。

pg num command

得到和设置指定 pool 中的 pg_num 和 pgp_num

ceph osd pool create <pool-name> <pg-number> <pgp-number>To create a new pool
ceph osd pool get <pool-name> <pg_num>To get number of PG in a pool
ceph osd pool get <pool-name> <pgp_num>To get number of PGP in a pool
ceph osd pool set <pool-name> <pg_num number>To increase number of PG in a pool
ceph osd pool set <pool-name> <pgp_num number>To increase number of PGP in a pool

创建pool时如果不指定 pg_num,默认为8

1
2
# ceph --show-config | grep osd_pool_default_pg_num
osd_pool_default_pg_num = 8
05-23 05:07