Thanks for your reply. But I still get some wired results. I remember that min_alloc_size cannot be changed after OSD creation, but I can’t find the source now. Searching results in this PR[1], which state: “min_alloc_size is now fixed at mkfs time”. Is it true that we can only change the min_alloc_size by destroying and recreating the OSD? If it is true, then all the provided methods does not give real effective min_alloc_size. Assuming min_alloc_size cannot be changed can’t be changed without OSD recreation: * Output of “ceph daemon osd.0 config ...” shows old value (64k), but after I restart the OSD daemon, it shows new value (32k), which can’t be correct. * Output of “ceph config get ...” always shows new value, which can’t be the effective value. Or maybe my knowledge is out of date. Now we can really change the min_alloc_size by just setting the new value and restarting the OSD daemon? [1]: https://github.com/ceph/ceph/pull/17610 在 2020年12月2日，15:56，Eugen Block <eblock(a)nde.ag> 写道： Hi, there're several ways to retrieve config information: osd-host:~ # ceph daemon osd.0 config show | grep bluestore_min_alloc_size "bluestore_min_alloc_size": "0", "bluestore_min_alloc_size_hdd": "65536", "bluestore_min_alloc_size_ssd": "4096", osd-host:~ # ceph daemon osd.0 config get bluestore_min_alloc_size { "bluestore_min_alloc_size": "0" } osd-host:~ # ceph config get osd bluestore_min_alloc_size osd-host:~ # ceph config get osd bluestore_min_alloc_size_hdd 65536 osd-host:~ # ceph config get osd.0 bluestore_min_alloc_size_hdd 65536 Regards, Eugen Zitat von 胡 玮文 <huww98(a)outlook.com>om>: Hi all, I’ve read from this mail list that too high bluestore_min_alloc_size will result in too much space wasted if I have many small objects. But too low bluestore_min_alloc_size will reduce performance. I’ve also read that this config can’t be changed after OSD creation. Now I want to tune this config myself. I change bluestore_min_alloc_size_hdd from default 64k to 32k, then deployed some more OSDs with cephadm. I assume old OSD will continue to use the old setting, and new OSDs will use the new setting. But how can I verify that? To be specific, how can I query the effective min_alloc_size for a specific OSD? I’ve tried to use “ceph daemon osd.X ...” but cannot find the appropriate command. I also searched the OSD logs. Thanks. _______________________________________________ ceph-users mailing list -- ceph-users(a)ceph.io To unsubscribe send an email to ceph-users-leave(a)ceph.io _______________________________________________ ceph-users mailing list -- ceph-users(a)ceph.io To unsubscribe send an email to ceph-users-leave(a)ceph.io

Effective min_alloc_size stays the same after mkfs. Config modifications don't impact it after OSD is created. On 12/2/2020 2:51 PM, 胡 玮文 wrote: > Thanks for your reply. But I still get some wired results. > > I remember that min_alloc_size cannot be changed after OSD creation, but I can’t find the source now. Searching results in this PR[1], which state: “min_alloc_size is now fixed at mkfs time”. Is it true that we can only change the min_alloc_size by destroying and recreating the OSD? > > If it is true, then all the provided methods does not give real effective min_alloc_size. Assuming min_alloc_size cannot be changed can’t be changed without OSD recreation: > * Output of “ceph daemon osd.0 config ...” shows old value (64k), but after I restart the OSD daemon, it shows new value (32k), which can’t be correct. > * Output of “ceph config get ...” always shows new value, which can’t be the effective value. > > Or maybe my knowledge is out of date. Now we can really change the min_alloc_size by just setting the new value and restarting the OSD daemon? > > [1]: https://github.com/ceph/ceph/pull/17610 > > 在 2020年12月2日，15:56，Eugen Block <eblock(a)nde.ag> 写道： > > Hi, > > there're several ways to retrieve config information: > > osd-host:~ # ceph daemon osd.0 config show | grep bluestore_min_alloc_size > "bluestore_min_alloc_size": "0", > "bluestore_min_alloc_size_hdd": "65536", > "bluestore_min_alloc_size_ssd": "4096", > > osd-host:~ # ceph daemon osd.0 config get bluestore_min_alloc_size > { > "bluestore_min_alloc_size": "0" > } > > osd-host:~ # ceph config get osd bluestore_min_alloc_size > > > osd-host:~ # ceph config get osd bluestore_min_alloc_size_hdd > 65536 > > osd-host:~ # ceph config get osd.0 bluestore_min_alloc_size_hdd > 65536 > > > Regards, > Eugen > > > Zitat von 胡 玮文 <huww98(a)outlook.com>om>: > > Hi all, > > I’ve read from this mail list that too high bluestore_min_alloc_size will result in too much space wasted if I have many small objects. But too low bluestore_min_alloc_size will reduce performance. I’ve also read that this config can’t be changed after OSD creation. > > Now I want to tune this config myself. I change bluestore_min_alloc_size_hdd from default 64k to 32k, then deployed some more OSDs with cephadm. I assume old OSD will continue to use the old setting, and new OSDs will use the new setting. But how can I verify that? To be specific, how can I query the effective min_alloc_size for a specific OSD? > > I’ve tried to use “ceph daemon osd.X ...” but cannot find the appropriate command. I also searched the OSD logs. > > Thanks. > _______________________________________________ > ceph-users mailing list -- ceph-users(a)ceph.io > To unsubscribe send an email to ceph-users-leave(a)ceph.io > > > _______________________________________________ > ceph-users mailing list -- ceph-users(a)ceph.io > To unsubscribe send an email to ceph-users-leave(a)ceph.io > _______________________________________________ > ceph-users mailing list -- ceph-users(a)ceph.io > To unsubscribe send an email to ceph-users-leave(a)ceph.io