On 11/03/2016 05:52 PM, wido@xxxxxxxx wrote:
Op 3 nov. 2016 om 16:44 heeft Joao Eduardo Luis <joao@xxxxxxx> het volgende geschreven:
On 11/03/2016 01:24 PM, Wido den Hollander wrote:
Op 3 november 2016 om 13:09 schreef Joao Eduardo Luis <joao@xxxxxxx>:
On 11/03/2016 09:40 AM, Wido den Hollander wrote:
root@mon3:/var/lib/ceph/mon# ceph-monstore-tool ceph-mon3 dump-keys|awk '{print $1}'|uniq -c
96 auth
1143 logm
3 mdsmap
1 mkfs
1 mon_sync
6 monitor
3 monmap
1158 osdmap
358364 paxos
656 pgmap
6 pgmap_meta
168 pgmap_osd
6144 pgmap_pg
root@mon3:/var/lib/ceph/mon#
So there are 358k Paxos entries in the Mon store.
Any suggestions on how to trim those from the MON store(s)?
Can you check the value of paxos:first_committed in the store?
Here you go:
root@mon3:~# ceph-monstore-tool /var/lib/ceph/mon/ceph-mon3 show-versions --map-type paxos
first committed: 174349108
last committed: 174349609
root@mon3:~#
Doesn't seem like a lot of keys in there?
I have this annoying feeling that this relates to some really old bug I can't really recall the ticket number, but in essence was not trimming the maps even though first_committed was updated.
The one way out of this I can think of is to change the value of 'first_committed' in the store to the very first paxos epoch you have. You will have to use 'ceph_kvstore_tool' to do that.
Is one mon enough? So stop that mon, change the value and start it again?
Or do I need to do this on a mon which will become primary?
Sorry for not being clear. This needs to be done on the one that will
become the leader.
Personally, I don't like this solution one bit, but I can't see any
other way without a patched monitor, or maybe ceph_monstore_tool.
If you are willing to wait till tomorrow, I'll be happy to kludge a
sanitation feature onto ceph_monstore_tool that will clean those
versions for you (latency being due to coding + testing + building).
-Joao
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
