Re: will crush rule be used during object relocation in OSD failure ?

[Maged Mokhtar <mmokhtar@xxxxxxxxxxx>]

On 23/11/18 18:00, ST Wong (ITSC) wrote:

Hi all,

We've 8 osd hosts, 4 in room 1 and 4 in room2.   

A pool with size = 3 using following crush map is created, to cater for room failure.

rule multiroom {
        id 0
        type replicated
        min_size 2
        max_size 4
        step take default
        step choose firstn 2 type room
        step chooseleaf firstn 2 type host
        step emit
}

We're expecting:

1.for each object, there are always 2 replicas in one room and 1 replica in other room making size=3.  But we can't control which room has 1 or 2 replicas.

2.in case an osd host fails, ceph will assign remaining osds to the same PG to hold replicas on the failed osd host.  Selection is based on crush rule of the pool, thus maintaining the same failure domain - won't make all replicas in the same room.  

  

3.in case of entire room with 1 replica fails, the pool will remain degraded but won't do any replica relocation.

4. in case of entire room with 2 replicas fails, ceph will make use of osds in the surviving room and making 2 replicas.  Pool will not be writeable before all objects are made 2 copies (unless we make pool size=4?).  Then when recovery is complete, pool will remain in degraded state until the failed room recover.

Is our understanding correct?  Thanks a lot.

Will do some simulation later to verify.

Regards,

/stwong

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

I think this is correct. To re-phrase 2) : all PGs on the failed node will be re-distributed on  several other hosts within the same room.

Since some PGs will have 2 replicas in 1 room whereas some other PGs will have 2 replicas in the other room, i tend to dis-like such setup as it is not symmetric,some PGs will suffer more than others in case on room failure, you failure domain is not symmetric. Besides more importantly, as you stated in 4, you cluster will be down while these unfortunate PGs recover ( statistically that is half your data ). I would prefer in such case you would use a size=4 min_size=2 setup.

/Maged

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com