Hm I'm not so sure, because we did have a disk outage indeed.
When we added many new disks, 50% of objects were misplaced.
Then the disk failed and ~2% of objects were degraded.
The recovery went on fine, but I would expect that fixing the degraded objects should have a priority over data
migrations, which it seems they don't have. The ratio (then 50%/2% = 20%/0.8% now) remained the same.
Maybe there's just some configuration option I haven't found yet?
-- Jonas
On 01/11/2018 08.47, Janne Johansson wrote:
> I think that all the misplaced PGs that are in the queue that get
> writes _while_ waiting for backfill will get the "degraded" status,
> meaning that before they were just on the wrong place, now they are on
> the wrong place, AND the newly made PG they should backfill into will
> get an old dump made first, then an incremental with all the changes
> that came in while waiting or while finishing the first backfill, then
> become active+clean.
> Nothing to worry about, that is how recovery looks on all clusters.
>
> Den ons 31 okt. 2018 kl 22:29 skrev Jonas Jelten <jelten@xxxxxxxxx>:
>>
>> Hello!
>>
>> My cluster currently has this health state:
>>
>> 2018-10-31 21:20:13.694633 mon.lol [WRN] Health check update: 39010709/192173470 objects misplaced (20.300%)
>> (OBJECT_MISPLACED)
>> 2018-10-31 21:20:13.694684 mon.lol [WRN] Health check update: Degraded data redundancy: 1624786/192173470 objects
>> degraded (0.845%), 49 pgs degraded, 57 pgs undersized (PG_DEGRADED)
>> [...]
>> 2018-10-31 21:39:24.113440 mon.lol [WRN] Health check update: 38897646/192173470 objects misplaced (20.241%)
>> (OBJECT_MISPLACED)
>> 2018-10-31 21:39:24.113526 mon.lol [WRN] Health check update: Degraded data redundancy: 1613658/192173470 objects
>> degraded (0.840%), 49 pgs degraded, 57 pgs undersized (PG_DEGRADED)
>>
>>
>> It is recovering slowly, but apparenly does not recover the 0.8% degraded objects first. Instead it recovers both at the
>> same relative rate, which even means that the misplaced objects are recovered way slower than the misplaced objects!
>>
>> Is there a way to recover the degraded objects first?
>>
>>
>> Cheers
>> -- Jonas
>> _______________________________________________
>> ceph-users mailing list
>> ceph-users@xxxxxxxxxxxxxx
>> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
>
>
>
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
