Hello all,
recently I get to the conclusion that of a 40 TB of physical space I
could use only 16TB before seeing pg stick because osd was too full.
The data space used seems to be for ever growing.
using ceph osd reweight-by-utilization 103 seems at first to rebalance
the osd pg use. Then the problem is solve for a time. But then the
problems appears again with more PGs stuck_too_full. and the problem for
ever grows. Sure the solution should be to add more disk space but for
that enhancement to be significant and solving the problem it should be
at least of a 25% which means growing the ceph cluster of 10 TB (5 disks
of 2TB or 3 disks of 4TB) that has a cost, and the problem will only be
solved for a moment until the replicas that are never freed fills again
the added data.
In the end I really can count on using a rbd image of 16 TB out of a 37
TB of global ceph cluster disk. Which means I can really use a 40% and
over the time that ratio will drop constantly.
So It is requiered of that the replicas and data can be overwriten so
that the hidden data will not keep growing. Or that I can clean them
when I need to.
Alphe Salas.
_______________________________________________
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
