In your situation the JJ Balancer might help. > > On 2024-11-27 17:53, Anthony D'Atri wrote: >>> Hi, >>> My Ceph cluster is out-of-balance. The amount of PG's per OSD ranges from about 50 up to 100 PG's per OSD. This is far from balanced. >> Do you have multiple CRUSH roots or device classes? Are all OSDs the same weight? > > Yes, I have 2 CRUSH roots and 2 device classes. 1 CRUSH root is using the ssd device class, another CRUSH root is using the hdd device class. I am only talking about the HDD's. > >>> My disk sizes differs from 1.6T up to 2.4T. >> Ah. The number of PG replicas as reported by `ceph osd df` should be proportional to the OSD capacities. With some of your OSDs double the size of others, it is natural that some will have double the number of PGs. >> That said, 2.4T is an unusual size for a drive. What specific storage drives are you using? Chances are that you’d benefit from bumping pg_num on at least some of your pools. > > Sorry, I was wrong. It is 1.6T and 2.2T (not 2.4T). > The 2.2T OSD's are ranging between 65-104 PG's, while the 1.6T OSD's are ranging between 47-80 PG's. > > Bumping the PG's per pool might balance it a bit, but without bumping the PG's the balancer module should balance it as well, right? _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |