shared WAL between CEPH OSD's?

[Magneto]

Is it possible to share a CEPH WAL between all the OSD's, instead of having to partition the WAL?

If I have 12 drives, I have to create 12 equal partition on the WAL, and assign each partition to an OSD. Is there a better way to assign the WAL?

 

[alexskysilk]

Possible- yes. Advisable- NO. Think about the consequences of what you're doing- you are taking 12 disks worth of PGs and tying it to a SPOF.

 

[Magneto]

What would happen if the WAL disk fails?

 

[spirit]

What would happen if the WAL disk fails?

you loose all your osd of this node. (so, do a raid1 for the wal if you want to avoid this, or simply rebuild all yours osd).

 

[Magneto]

Do I need to use one WAL per OSD if I use spinning disks?

 

[alexskysilk]

Do I need to use one WAL per OSD if I use spinning disks?

No. the question to ask is what happens when a something in the failure domain fails. since in the case of a WAL device fault it will take out ALL associated OSDs, as long as those OSDs will not kill your cluster you could consider it as a solution.

IF you had 100 OSDs in 10 nodes- concievably you could have a single WAL device back 10 disks with the understanding that WAL failure will take out the whole node- which is a nuisance but not critical. If you had 12 OSDs in 3 nodes I would not recommend sharing the WAL.