Oftern in ceph you may want to change the default Disk schedular to ensure you can deliver the best expereience for your users. For my usecase we needed to ensure that client I/O takes prioirty over Recovery So I set the below options in ceph.conf on the osd servers.

1
2

osd_disk_thread_ioprio_class = idle
osd_disk_thread_ioprio_priority = 7

As the above onyl work with CFQ and the default schedular was deadline I added the below to the grub config to change this.

1

GRUB_CMDLINE_LINUX="elevator=cfq"

However as I use SSD’s for the Journals I wanted to ensure these disks use NOOP so I used the below udev rule for this:

1
2

# cat /etc/udev/rules.d/90-noop.rules
ACTION=="add|change", KERNEL=="sd[a-z]", ATTR{queue/rotational}=="0",ATTR{queue/scheduler}="noop"