Den sön 17 maj 2020 kl 14:33 skrev Marc Roos <M.Roos@f1-outsourcing.eu>:
outs, and you are more likely to shoot yourself in the foot. At least ask first. Eg this bcache, I am not 100% sure what it is, but if it is sitting between the osd process and the disk, it could be getting nasty with a reset/power outage, when ceph thinks data is written to disk, while it is not.
Bcache is a neat way in linux to stack devices, working basically like the ceph layered pools, but on a per-host basis of course. The main advantage, and why some ceph admins use if instead of putting WAL/DB on the fast device, is that a single bcache device can be placed in front of several other devices, and the kernel will decide which writes are going to use the write cache, so in effect you are sharing the full size of your fast device speed on all OSDs, instead of having to split your NVME into pieces or preset a journal size. If you at any point expect to add more drives to a box, this becomes even more interesting, since knowing beforehand how many preallocations to leave room for is very hard and leaves out potential gains until you get it right. -- May the most significant bit of your life be positive.