在恢复上下文中,新的擦除PG保持不活动+不完整

时间:2018-06-07 14:08:07

标签: ceph

问题(如果您想查看'状态'首先,请查看帖子末尾的输出)

在我的生产群集中,缓慢恢复的擦除编码池,

新的擦除池无法使用,因为他们所有的pgs仍然存在"创建+不完整"看似永远。

我现有的游泳池恢复需要很长时间(几个月),但我希望能够创建新的' clean'立即汇集以有弹性的方式接收新数据。

我尝试了什么:

ceph osd pool create newpool 128 128 erasure myprofile

rados --pool newpool put anobject afile ==>这会阻止

ceph pg ls-by-pool newpool incomplete ==>列出我所有的pgs

ceph pg 15.1查询 ==>国家"创建+不完全"    "向上"和"表演"仅包含osd' 1'作为第一个元素,并且在所有其他位置都是' null'(2147483647)。

请注意osd' 1'在我的平台上是最多的(它的PG数量几乎是其他OSD的两倍)

完整上下文

  • Luminous 12.2.0(Ubuntu 16.04)从jewel迁移
  • 现有的擦除池存在非常恶化的情况(所有OSD现在都已启动,但83%的对象放错地方,13%的对象降级)
  • 恢复时间估计为一年
  • 我的删除配置文件是12 + 3

==>我想开始写一个新的" clean"游泳池,而现有的游泳池将慢慢恢复。

===================================== ceph状态

群集:     id:b5ee2a02-b92c-4829-8d43-0eb17314c0f6     健康:HEALTH_WARN             1314118857/1577308005物品放错地图(83.314%)             数据可用性降低:10 pgs不活跃,10 pgs不完整             降级的数据冗余:203997294/1577308005对象降级(12.933%),492 pgs不洁,342 pgs降级,279 pgs小型化

服务:     mon:28个守护进程,法定人数1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22, 23,24,25,26,27,28     mgr:37(有效),备用:38,12,45,33,29,10,11,22,31,47,15,36,40,32,41,24,44,34,27,28,43 ,35,39,16,25,26,2,9,3,4,5,23,19,17,5,42,6,21,7,20,30,13,18,14,46,1     osd:47 osds:47 up,47 in; 512个重新映射的pgs

数据:     游泳池:3个游泳池,522 pgs     对象:100M对象,165 TB     用量:191 TB,466 TB / 657 TB     pgs:1.916%pgs无效              203997294/1577308005对象降级(12.933%)              1314118857/1577308005物品放错地图(83.314%)              155 active + undersized + degraded + remapped + backfill_wait              140 active + remapped + backfill_wait              114活动+小型+降级+重新映射              63 active + recovery_wait + degraded + remapped              30活动+清洁+重新映射              10创建+不完整              9 active + recovery_wait + undersized + degraded + remapped              1个活跃+小型+降级+重新映射+回填

IO:     客户端:291 kB / s rd,0 B / s wr,14 op / s rd,16 op / s wr     恢复:6309 kB / s,2个对象/ s

ceph健康细节

HEALTH_WARN 1314114780/1577303100错位的对象(83.314%);数据可用性降低:10 pgs不活跃,10 pgs不完整;降级数据冗余:203992956/1577303100对象降级(12.933%),492 pgs不洁,342 pgs降级,279 pgs缩小 OBJECT_MISPLACED 1314114780/1577303100错位的对象(83.314%) PG_AVAILABILITY降低数据可用性:10 pgs不活跃,10 pgs不完整     PG 15.0 +创建不完整的,从12作用[1,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647](减少池测试器MIN_SIZE可能有助于;搜索ceph.com/docs为'不完整')     PG 15.1 +创建不完整的,从12作用[1,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647](减少池测试器MIN_SIZE可能有助于;搜索ceph.com/docs为'不完整')     PG 15.2是+创建不完整的,从12作用[1,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647](减少池测试器MIN_SIZE可能有助于;搜索ceph.com/docs为'不完整')     PG 15.3是+创建不完整的,从12作用[1,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647](减少池测试器MIN_SIZE可能有助于;搜索ceph.com/docs为'不完整')     PG 15.4是+创建不完整的,从12作用[1,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647](减少池测试器MIN_SIZE可能有助于;搜索ceph.com/docs为'不完整')     PG 15.5 +创建不完整的,从12作用[1,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647](减少池测试器MIN_SIZE可能有助于;搜索ceph.com/docs为'不完整')     PG 15.6是+创建不完整的,从12作用[1,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647](减少池测试器MIN_SIZE可能有助于;搜索ceph.com/docs为'不完整')     PG 15.7是+创建不完整的,从12作用[1,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647](减少池测试器MIN_SIZE可能有助于;搜索ceph.com/docs为'不完整')     PG 15.8是+创建不完整的,从12作用[1,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647](减少池测试器MIN_SIZE可能有助于;搜索ceph.com/docs为'不完整')     PG 15.9是+创建不完整的,从12作用[1,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647,2147483647](减少池测试器MIN_SIZE可能有助于;搜索ceph.com/docs为'不完整') PG_DEGRADED降级数据冗余:203992956/1577303100对象降级(12.933%),492 pgs不洁净,342 pgs降级,279 pgs缩小     pg 1.e2被卡住不洁,为9096318.249662,当前状态有效+超小+退化+重新映射,最后代理[1,2147483647,2147483647,5,40,8,28,47,13,12,29,10,23,2147483647 ,35]     pg 1.e3被卡在11585.111340,当前状态有效+低于+降级+重新映射,最后一次[1,3,2147483647,47,11,21,32,46,28,23,2147483647,13,2147483647,19 ,26]     pg 1.e4卡在11588.194871,当前状态有效+尺寸不足+降级+重映射+回填_等,最后作用[26,6,23,46,18,30,2147483647,25,38,29,13,45,9] ,35.20]     pg 1.e5被卡住尺寸为11588.374341,当前状态有效+尺寸不足+退化+重新映射+反向填充_等,最后代理[14,40,2147483647,22,18,17,29,31,28,43,34,19,33 ,15,32]     pg 1.e6被锁定为11584.602668,当前状态有效+超出+降级+重新映射,最后一次[1,38,40,2147483647,46,14,2147483647,23,7,44,15,39,8,21] ,28]     pg 1.e7被锁定为11578.574380,当前状态有效+超出+降级+重新映射,最后一次[1,13,2147483647,37,29,33,18,2147483647,9,38,23,16,42,2147483647 ,3]     pg 1.e8被压缩为11571.385848,当前状态有效+超小+退化+重新映射,最后一次[1,23,2147483647,7,36,26,6,39,38,2147483647,29,11,15,2147483647 ,19]     pg 1.e9被锁定为11588.254477,当前状态有效+超小+降级+重映射+回填_等,最后一次行动[13,44,16,11,9,2147483647,32,37,45,17,20,21,40 ,46,2147483647]     pg 1.ea被卡住11580.242417,当前状态有效+超小+降级+重映射+回填_等,最后一次[25,19,30,33,2147483647,44,20,39,17,45,43,24,2147483647 ,10,21]     pg 1.eb卡在11588.329063,当前状态有效+尺寸不足+降级+重新映射+回填_等,最后一次[29,39,2147483647,2147483647,40,18,4,33,24,38,32,36,15] ,47,12]     pg 1.ec被锁定为11587.781353,当前状态有效+超小+降级+重映射+回填_等,最后一次[35,37,42,11,2147483647,2147483647,30,15,39,44,43,46,17 ,4,7]

[...]

pg 3.fa is stuck unclean for 11649.009911, current state active+remapped+backfill_wait, last acting [36,15,42,4,21,14,34,16,17,8,39,3,2,7,19]
pg 3.fb is stuck undersized for 11580.419670, current state active+undersized+degraded+remapped, last acting [1,7,16,9,19,39,2147483647,33,26,23,20,8,35,40,29]
pg 3.fd is stuck unclean for 11649.040651, current state active+remapped+backfill_wait, last acting [17,21,8,26,15,42,46,27,7,39,14,35,4,29,25]
pg 3.fe is active+recovery_wait+degraded+remapped, acting [22,8,45,18,10,46,33,36,16,7,17,34,43,1,23]
pg 3.ff is stuck unclean for 11649.056722, current state active+remapped+backfill_wait, last acting [33,46,47,17,37,4,40,34,28,43,3,44,13,2,11]

1 个答案:

答案 0 :(得分:0)

随着宝石到光亮的迁移群集大小增加,我遇到了同样的问题。在这个过程中的某个地方,挤压图权重已经被损坏(不是通过" reweight"命令定位的那个)并被归零。这导致我的集群做出奇怪的放置决策(溢出其中一个pgs太多的节点),并且无法分配新的pgs(至少在这个节点不负责时)。

当我们编辑crushmap(解压缩,反编译,手动编译,重新编译,推送)并重新启动我们的集群时,问题就解决了。我们只是手动重置所有设备的权重为1.0(因为我们所有的磁盘都是相同的)。

重新启动后,可以分配新的pgs / pool,并且分配行为似乎恢复正常(重新平衡事情)。

希望有所帮助,

Soltiz。