Ceph 出现pg object unfound的处理经过

Ceph 出现pg object unfound的处理经过

原文链接：https://my.oschina.net/xiaozhublog/blog/3075404

1、背景

集群中的一个节点损坏，同时另外一个节点坏了一块盘

2、问题

查看ceph集群的状态，看到归置组pg 4.210丢了一个块

# ceph health detail
HEALTH_WARN 481/5647596 objects misplaced (0.009%); 1/1882532 objects unfound (0.000%); Degraded data redundancy: 965/5647596 objects degraded (0.017%), 1 pg degraded, 1 pg undersized
OBJECT_MISPLACED 481/5647596 objects misplaced (0.009%)
OBJECT_UNFOUND 1/1882532 objects unfound (0.000%)
    pg 4.210 has 1 unfound objects
PG_DEGRADED Degraded data redundancy: 965/5647596 objects degraded (0.017%), 1 pg degraded, 1 pg undersized
    pg 4.210 is stuck undersized for 38159.843116, current state active+recovery_wait+undersized+degraded+remapped, last acting [2]

3、处理过程

3.1、先让集群可以正常使用

查看pg 4.210，可以看到它现在只有一个副本

# ceph pg dump_json pools |grep 4.210
dumped all
4.210       482                  1      965       481       1  2013720576 3461     3461 active+recovery_wait+undersized+degraded+remapped 2019-07-10 09:34:53.693724   9027'1835435   9027:1937140  [6,17,20]          6        [2]              2   6368'1830618 2019-07-07 01:36:16.289885    6368'1830618 2019-07-07 01:36:16.289885             2
# ceph pg map 4.210
osdmap e9181 pg 4.210 (4.210) -> up [26,20,2] acting [2]
丢了两个副本，而且最主要的是主副本也丢了…

因为默认指定的pool的min_size为2，这就导致4.210所在的池vms不能正常使用

# ceph osd pool stats vms
pool vms id 4
  965/1478433 objects degraded (0.065%)
  481/1478433 objects misplaced (0.033%)
  1/492811 objects unfound (0.000%)
  client io 680 B/s rd, 399 kB/s wr, 0 op/s rd, 25 op/s wr
# ceph osd pool ls detail|grep vms
pool 4 'vms' replicated size 3 min_size 2 crush_rule 0 object_hash rjenkins pg_num 1024 pgp_num 1024 last_change 10312 lfor 0/874 flags hashpspool stripe_width 0 application rbd

直接影响了部分虚拟机，导致部分虚拟机夯住了，执行命令无回应

为了可以正常使用，先见vms池的min_size调整为1

# ceph osd pool set vms min_size 1
set pool 4 min_size to 1

3.2、尝试恢复pg4.210丢失的块

查看pg4.210

# ceph pg 4.210 query 
    "recovery_state": [
        {
            "name": "Started/Primary/Active",
            "enter_time": "2019-07-09 23:04:31.718033",
            "might_have_unfound": [
                {
                    "osd": "4",
                    "status": "already probed"
                },
                {
                    "osd": "6",
                    "status": "already probed"
                },
                {
                    "osd": "15",
                    "status": "already probed"
                },
                {
                    "osd": "17",
                    "status": "already probed"
                },
                {
                    "osd": "20",
                    "status": "already probed"
                },
                {
                    "osd": "22",
                    "status": "osd is down"
                },
                {
                    "osd": "23",
                    "status": "already probed"
                },
                {
                    "osd": "26",
                    "status": "osd is down"
                }
            ]

字面上理解，pg 4.210的自我恢复状态，它已经探查了osd4、6、15、17、20、23,osd22和26已经down了，而我这里的osd22和26都已经移出了集群

根据官网了解到此处might_have_unfound的osd有以下四种状态

already probed
querying
OSD is down
not queried (yet)

两种解决方案，回退旧版或者直接删除

# ceph pg 4.210 mark_unfound_lost revert
Error EINVAL: pg has 1 unfound objects but we haven't probed all sources,not marking lost
# ceph pg 4.210 mark_unfound_lost delete
Error EINVAL: pg has 1 unfound objects but we haven't probed all sources,not marking lost

提示报错，pg那个未发现的块还没有探查所有的资源，不能标记为丢失，也就是不会回退也不可以删除

猜测可能是已经down的osd22和26未探查，刚好坏的节点也重装完成，重新添加osd

osd的删除添加过程此处不赘述了。

添加完成后，再次查看pg 4.210

"recovery_state": [
        {
            "name": "Started/Primary/Active",
            "enter_time": "2019-07-15 15:24:32.277667",
            "might_have_unfound": [
                {
                    "osd": "4",
                    "status": "already probed"
                },
                {
                    "osd": "6",
                    "status": "already probed"
                },
                {
                    "osd": "15",
                    "status": "already probed"
                },
                {
                    "osd": "17",
                    "status": "already probed"
                },
                {
                    "osd": "20",
                    "status": "already probed"
                },
                {
                    "osd": "22",
                    "status": "already probed"
                },
                {
                    "osd": "23",
                    "status": "already probed"
                },
                {
                    "osd": "24",
                    "status": "already probed"
                },
                {
                    "osd": "26",
                    "status": "already probed"
                }
            ],
            "recovery_progress": {
                "backfill_targets": [
                    "20",
                    "26"

                ],

可以看到所有的资源都probed了，此时执行回退命令

# ceph pg  4.210  mark_unfound_lost revert
pg has 1 objects unfound and apparently lost marking

查看集群状态

# ceph health detail
HEALTH_OK

恢复池vms的min_size为2

# ceph osd pool set vms min_size 2
set pool 4 min_size to 2

参考链接

http://www.itzhoulin.com/2016/04/20/object-unfound-when-wrong-steps-in-ceph-full/http://docs.ceph.com/docs/master/rados/troubleshooting/troubleshooting-pg/?highlight=troubleshoot