Ceph, and fstab discard

[NKarnel]

Hello everyone,

I'm thinking of using the "discard" flag in my guest VMs fstabs.

Is there any downside to it, apart from maybe a performance hit?
What about data integrity?

Thanks!

 

[Alwin]

This shouldn't make a difference for CephFS, as a client is writing/reading objects to/from a pool and the metadata is kept by the MDS in another. An object deletion should directly release the occupied space.

 

[NKarnel]

I'm using RDB mainly, and it does make a difference in the total "used" space seen by the cluster, thus helps me oversubscribe space.

 

[Alwin]

using RDB mainly,

An RBD image has a filesystem on top, Ceph is not aware of the FS state.

 

[spirit]

Yes, using discard in /etc/fstab or fstrim in crontab is working fine on rbd to release space.
(just enable discard too on the vm disk option).

Just be carefull of discard in /etc/fstab, before kernel 4.7, it was not async, so you need to wait when you delete a file.
with kernel >= 4.7, it's async, so no performance degradation.

 

[Alwin]

@spirit, thanks for clarifying. It seems I completely misread the post.

 

[NKarnel]

Thank you for your help! All is clear now