ceph-osd does not start correctly OSDs if journal are directly-linked partition
I know that mine is a corner case, but...
I've setup a little ceph cluster (PVE based), where journal devices are shared (partition on) system (OS) disks; servers are old, so system/boot disk are DOS (non-GPT), and so journal partition are DOS too.
In boot phase, ceph-osd relay on udev to change (chown) journal partition user to ceph, and udev rules relay on GPT tags, so this schema does not work for non-GPT partition.
I've build a little patch for ceph-osd initialization script, ceph-osd-prestart.sh, that work also for direct-symplinked journal partition. The patch does not disrupt the script, simply add a condition and an action (the chown), and (but i'm a ceph-ignorant!) seems to me it is not dangerous at all.
I hope will be applied upstream. Thanks.