On Fri, Sep 2, 2011 at 4:35 AM, huang jun <hjwsm1989@xxxxxxxxx> wrote: > hi,all > we test osd adding > at first there is only one osd , and we add an osd online, at the step: > $ osdmaptool --createsimple 2 --clobber /tmp/osdmap.junk > --export-crush /tmp/crush.new > we look into the source code, and a few things confused us. > the pg_num is based on osd_num, but now we have 2 osds, i think pg_num > should be 128. > but the "ceph osd dump -o -" outputs: > pg_pool 0 'data' pg_pool(rep pg_size 2 crush_ruleset 0 object_hash > rjenkins pg_num 64 pgp_num 64 lpg_num 2 lpgp_num 2 last_change 1 owner > 0) > pg_pool 1 'metadata' pg_pool(rep pg_size 2 crush_ruleset 1 object_hash > rjenkins pg_num 64 pgp_num 64 lpg_num 2 lpgp_num 2 last_change 1 owner > 0) > pg_pool 2 'rbd' pg_pool(rep pg_size 2 crush_ruleset 2 object_hash > rjenkins pg_num 64 pgp_num 64 lpg_num 2 lpgp_num 2 last_change 1 owner > 0) > > max_osd 2 > osd0 up in weight 1 up_from 2 up_thru 5 down_at 0 last_clean_interval 0-0 > osd1 up in weight 1 up_from 5 up_thru 6 down_at 0 last_clean_interval 0-0 > we test many other situations, and a common result we got is: > 1) if we add osds on line the total pg don't change, only move some > new pgs to the new osd. > 2) if we remove osds, total pg don't change too, just move the pg from > the dead osd to the the exist osds in current osdmap. > so, did the "osd dump -o -" depends on the very first time when we > start the ceph cluster? > then, we are interesting in situation that add 130 osds online one by > one, the pg_num doesn't change too?always stay 64? > and the most confusing question is: when should we split pg?? > thank you ! > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html > Hi Huang, The number of PGs is set when you run mkcephfs. The number of PGs can be changed later by the administrator. The code for changing the number of PGs is under development, and it's unstable at the moment. sincerely, Colin -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
