Hi Reto,
On Wed, Apr 19, 2023 at 9:34 PM Ilya Dryomov <idryomov(a)gmail.com> wrote:
On Wed, Apr 19, 2023 at 5:57 PM Reto Gysi <rlgysi(a)gmail.com> wrote:
Hi,
Am Mi., 19. Apr. 2023 um 11:02 Uhr schrieb Ilya Dryomov <idryomov(a)gmail.com>om>:
On Wed, Apr 19, 2023 at 10:29 AM Reto Gysi <rlgysi(a)gmail.com> wrote:
yes, I used the same ecpool_hdd also for cephfs file systems. The new pool ecpool_test
I've created for a test, I've also created it with application profile
'cephfs', but there aren't any cephfs filesystem attached to it.
This is not and has never been supported.
Do you mean 1) using the same erasure coded pool for both rbd and cephfs, or 2) multiple
cephfs filesystem using the same erasure coded pool via
ceph.dir.layout.pool="ecpool_hdd"?
(1), using the same EC pool for both RBD and CephFS.
1)
2)
rgysi cephfs filesystem
rgysi - 5 clients
=====
RANK STATE MDS ACTIVITY DNS INOS DIRS CAPS
0 active rgysi.debian.uhgqen Reqs: 0 /s 409k 408k 40.8k 16.5k
POOL TYPE USED AVAIL
cephfs.rgysi.meta metadata 1454M 2114G
cephfs.rgysi.data data 4898G 17.6T
ecpool_hdd data 29.3T 29.6T
root@zephir:~# getfattr -n ceph.dir.layout /home/rgysi/am/ecpool/
getfattr: Removing leading '/' from absolute path names
# file: home/rgysi/am/ecpool/
ceph.dir.layout="stripe_unit=4194304 stripe_count=1 object_size=4194304
pool=ecpool_hdd"
root@zephir:~#
backups cephfs filesystem
backups - 2 clients
=======
RANK STATE MDS ACTIVITY DNS INOS DIRS CAPS
0 active backups.debian.runngh Reqs: 0 /s 253k 253k 21.3k 899
POOL TYPE USED AVAIL
cephfs.backups.meta metadata 1364M 2114G
cephfs.backups.data data 16.7T 16.4T
ecpool_hdd data 29.3T 29.6T
root@zephir:~# getfattr -n ceph.dir.layout /mnt/backups/windows/windows-drives/
getfattr: Removing leading '/' from absolute path names
# file: mnt/backups/windows/windows-drives/
ceph.dir.layout="stripe_unit=4194304 stripe_count=1 object_size=4194304
pool=ecpool_hdd"
root@zephir:~#
So I guess I should use a different ec datapool for rbd and for each of the cephfs
filesystems in the future, correct?
Definitely a different EC pool for RBD (i.e. don't mix with CephFS).
Not sure about the _each_ of the filesystems bit -- Venky or Patrick can
comment on whether sharing an EC pool between filesystems is OK.
That's true for CephFS too -- different pools for each ceph file
system is recommended.
You can use the `--allow-dangerous-metadata-overlay` option when
creating a ceph file system to reuse metadata and data pools if those
are already in use, however, it's only to be used during emergency
situations.