ceph分布式存储-数据不均衡调整
2018-12-26 本文已影响6人
lihanglucien
1. 查看数据分布是否均衡
#查看osd使用情况
$ ceph osd df tree
ID CLASS WEIGHT REWEIGHT SIZE USE AVAIL %USE VAR PGS TYPE NAME
-1 196.21051 - 190T 347G 190T 0.18 1.00 - root default
-3 65.40106 - 61390G 115G 61274G 0.19 1.06 - host ceph-xxx-osd01
0 hdd 5.45009 0 0 0 0 0 0 0 osd.0
1 hdd 5.45009 1.00000 5580G 8640M 5572G 0.15 0.85 125 osd.1
2 hdd 5.45009 1.00000 5580G 7797M 5573G 0.14 0.77 128 osd.2
3 hdd 5.45009 1.00000 5580G 10811M 5570G 0.19 1.06 131 osd.3
4 hdd 5.45009 1.00000 5580G 11150M 5570G 0.20 1.10 124 osd.4
5 hdd 5.45009 0.95001 5580G 13175M 5568G 0.23 1.29 114 osd.5
6 hdd 5.45009 1.00000 5580G 9769M 5571G 0.17 0.96 141 osd.6
7 hdd 5.45009 0.90002 5580G 11773M 5569G 0.21 1.16 102 osd.7
8 hdd 5.45009 1.00000 5580G 11805M 5569G 0.21 1.16 118 osd.8
9 hdd 5.45009 1.00000 5580G 12964M 5568G 0.23 1.27 123 osd.9
10 hdd 5.45009 0.90002 5580G 12315M 5568G 0.22 1.21 123 osd.10
11 hdd 5.45009 1.00000 5580G 8549M 5572G 0.15 0.84 115 osd.11
-5 65.40106 - 66971G 115G 66855G 0.17 0.97 - host ceph-xxx-osd02
12 hdd 5.45009 1.00000 5580G 8975M 5572G 0.16 0.88 115 osd.12
13 hdd 5.45009 1.00000 5580G 10832M 5570G 0.19 1.06 106 osd.13
14 hdd 5.45009 1.00000 5580G 10183M 5570G 0.18 1.00 111 osd.14
15 hdd 5.45009 1.00000 5580G 9337M 5571G 0.16 0.92 93 osd.15
16 hdd 5.45009 1.00000 5580G 11508M 5569G 0.20 1.13 111 osd.16
17 hdd 5.45009 1.00000 5580G 6237M 5574G 0.11 0.61 118 osd.17
18 hdd 5.45009 1.00000 5580G 9822M 5571G 0.17 0.97 116 osd.18
19 hdd 5.45009 0.95001 5580G 13169M 5568G 0.23 1.29 122 osd.19
20 hdd 5.45009 1.00000 5580G 7996M 5573G 0.14 0.79 132 osd.20
21 hdd 5.45009 1.00000 5580G 7671M 5573G 0.13 0.75 97 osd.21
22 hdd 5.45009 1.00000 5580G 7959M 5573G 0.14 0.78 121 osd.22
23 hdd 5.45009 0.85004 5580G 15036M 5566G 0.26 1.48 102 osd.23
-7 65.40839 - 66978G 115G 66862G 0.17 0.97 - host ceph-xxx-osd03
24 hdd 5.45070 1.00000 5581G 8716M 5572G 0.15 0.86 128 osd.24
25 hdd 5.45070 1.00000 5581G 11036M 5570G 0.19 1.08 113 osd.25
26 hdd 5.45070 1.00000 5581G 9507M 5572G 0.17 0.93 113 osd.26
27 hdd 5.45070 0.85004 5581G 12854M 5568G 0.22 1.26 105 osd.27
28 hdd 5.45070 1.00000 5581G 9550M 5572G 0.17 0.94 102 osd.28
29 hdd 5.45070 1.00000 5581G 9189M 5572G 0.16 0.90 109 osd.29
30 hdd 5.45070 1.00000 5581G 7804M 5573G 0.14 0.77 105 osd.30
31 hdd 5.45070 1.00000 5581G 11366M 5570G 0.20 1.12 116 osd.31
32 hdd 5.45070 1.00000 5581G 9286M 5572G 0.16 0.91 110 osd.32
33 hdd 5.45070 1.00000 5581G 10233M 5571G 0.18 1.01 115 osd.33
34 hdd 5.45070 1.00000 5581G 8223M 5573G 0.14 0.81 107 osd.34
35 hdd 5.45070 1.00000 5581G 10989M 5570G 0.19 1.08 121 osd.35
TOTAL 190T 347G 190T 0.18
MIN/MAX VAR: 0.61/1.48 STDDEV: 0.03
#查看osd_num,PGS, %USE
ceph osd df tree | awk '/osd\./{print $NF" "$(NF-1)" "$(NF-3) }'
osd.1 125 0.15
osd.2 128 0.14
osd.3 131 0.19
osd.4 124 0.20
osd.5 114 0.23
osd.6 141 0.17
osd.7 102 0.21
osd.8 118 0.21
osd.9 123 0.23
osd.10 123 0.22
osd.11 115 0.15
osd.12 115 0.16
osd.13 106 0.19
osd.14 111 0.18
osd.15 93 0.16
osd.16 111 0.20
osd.17 118 0.11
osd.18 116 0.17
osd.19 122 0.23
osd.20 132 0.14
osd.21 97 0.13
osd.22 121 0.14
osd.23 102 0.26
osd.24 128 0.15
osd.25 113 0.19
osd.26 113 0.17
osd.27 105 0.22
osd.28 102 0.17
osd.29 109 0.16
osd.30 105 0.14
osd.31 116 0.20
osd.32 110 0.16
osd.33 115 0.18
osd.34 107 0.14
osd.35 121 0.19
2. reweight-by-pg 按归置组分布情况调整 OSD 的权重
$ceph osd reweight-by-pg
moved 35 / 4032 (0.868056%) #35个PG发送迁移
avg 115.2 #每个OSD承载的平均PG数目为115.2
stddev 10.378 -> 9.47418 (expected baseline 10.5787) #执行本次调整后， 标准方差将由10.378变为9.47418
min osd.15 with 93 -> 92 pgs (0.807292 -> 0.798611 * mean) #当前负载最轻的OSD为osd.15，只承载了93个PG, 执行本次调整后，将承载92个PG
max osd.6 with 141 -> 132 pgs (1.22396 -> 1.14583 * mean) #当前负载最重的OSD为osd.6, 承载了141个PG, 执行本次调整后，讲承载132个PG
oload 120
max_change 0.05
max_change_osds 4
average_utilization 21.1365
overload_utilization 25.3638
osd.6 weight 1.0000 -> 0.9500 #执行本次调整后，对osd.6,osd.23,osd.7,osd.27的reweight进行调整
osd.23 weight 0.8500 -> 0.9000
osd.7 weight 0.9000 -> 0.9500
osd.27 weight 0.8500 -> 0.9000
3. reweight-by-utilization 按利用率调整 OSD 的权重
$ceph osd reweight-by-utilization
moved 35 / 4032 (0.868056%) #35个PG发送迁移
avg 115.2 #每个OSD承载的平均PG数目为115.2
stddev 10.378 -> 9.47418 (expected baseline 10.5787) #执行本次调整后， 标准方差将由10.378变为9.47418
min osd.15 with 93 -> 92 pgs (0.807292 -> 0.798611 * mean) #当前负载最轻的OSD为osd.15，只承载了93个PG, 执行本次调整后，将承载92个PG
max osd.6 with 141 -> 132 pgs (1.22396 -> 1.14583 * mean) #当前负载最重的OSD为osd.6, 承载了141个PG, 执行本次调整后，讲承载132个PG
oload 120
max_change 0.05
max_change_osds 4
average_utilization 21.1365
overload_utilization 25.3638
osd.6 weight 1.0000 -> 0.9500 #执行本次调整后，对osd.6,osd.23,osd.7,osd.27的reweight进行调整
osd.23 weight 0.8500 -> 0.9000
osd.7 weight 0.9000 -> 0.9500
osd.27 weight 0.8500 -> 0.9000
4. 数据均衡后还原权重
#统计osd_num, REWEIGHT
$ceph osd df tree | awk '/osd\./{print $NF" "$4 }'
osd.1 1.00000
osd.2 1.00000
osd.3 1.00000
osd.4 1.00000
osd.5 0.90002
osd.6 0.95001
osd.7 0.90002
osd.8 1.00000
osd.9 1.00000
osd.10 0.90002
osd.11 1.00000
osd.12 1.00000
osd.13 1.00000
osd.14 1.00000
osd.15 1.00000
osd.16 1.00000
osd.17 1.00000
osd.18 1.00000
osd.19 0.95001
osd.20 1.00000
osd.21 1.00000
osd.22 1.00000
osd.23 0.85004
osd.24 1.00000
osd.25 1.00000
osd.26 1.00000
osd.27 0.85004
osd.28 1.00000
osd.29 1.00000
osd.30 1.00000
osd.31 1.00000
osd.32 1.00000
osd.33 1.00000
osd.34 1.00000
osd.35 1.00000
#依次设置osd权重为默认值，1.0
#ceph osd reweight {id} {weight}
#说明：osd weight的取值为0~1
$ ceph osd reweight 5 1.0
