Sorry for not making it clear, we are using upmap. Just saw this from the code and wondering about the usage. For the OSDs, we do not have any OSD weight < 1.00 until one OSD reaches the 85% near full ratio. Before I reweight the OSD, our mgr/balancer/upmap_max_deviation is set to 5 and the PG distribution is around +/- 5 on each OSD. What's more, I checked the OSD usage and found the usage varies from about 50% to 70% when the average usage is 60%, is this distribution OK? We also enabled the compression and used snappy, will the compression affect the OSD usage? On Wed, Nov 8, 2023 at 7:24 AM <bryansoong21@gmail.com> wrote:

Hello,

We are using a Ceph Pacific (16.2.10) cluster and enabled the balancer module, but the usage of some OSDs keeps growing and reached up to mon_osd_nearfull_ratio, which we use 85% by default, and we think the balancer module should do some balancer work.

So I checked our balancer configuration and found that our "crush_compat_metrics" is set to "pgs,objects,bytes", and this three values are used in src.pybind.mgr.balancer.module.Module.calc_eval. However, when doing the actual balance task, only the first key is used to do the auto balance, in src.pybind.mgr.balancer.module.Module.do_crush_compat: metrics = self.get_module_option('crush_compat_metrics').split(',') key = metrics[0] # balancing using the first score metric

My concern is, any reason why we calculate the balancing using the three items but only do the balance using the first one?

Thanks. _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io

-- Thanks & best regards... Bryan (Longchao Song)