[SOLVED] Ceph luminous required for PVE 5 if Ceph is external

[Hans van den Bogert]

We are planning an upgrade of PVE 4.x to 5.x. Is the requirement[1] on Ceph Luminous also in the case of using an external Ceph cluster?

[1] https://pve.proxmox.com/wiki/Upgrad...he_basic_system_to_Debian_Stretch_and_PVE_5.0

 

[fabian]

no. storage access is done using plain librbd / krbd and should work fine using the stock Debian Ceph packages and an external cluster (which I hope is a t least Jewel, since Hammer is EOL upstream )

 

[Hans van den Bogert]

@fabian That's quite a relief. My reasoning was that even when just using plain librbd/krbd, it could still be the case that PVE 5 expects to be able to enable certain RBD functionality which is not in Jewel.
So, to get a better picture of the referenced requirement in my opening post, why *is* it necessary for a proxmox-hosted Ceph cluster?

 

[fabian]

@fabian That's quite a relief. My reasoning was that even when just using plain librbd/krbd, it could still be the case that PVE 5 expects to be able to enable certain RBD functionality which is not in Jewel.
So, to get a better picture of the referenced requirement in my opening post, why *is* it necessary for a proxmox-hosted Ceph cluster?

because supporting multiple versions is a lot more work, both for maintaining the packages (which we now do ourselves for PVE 5) as well as for integrating it into our tools. e.g., Ceph Luminous breaks a lot of the status and monitoring output compared to Jewel, introduced the new Ceph manager service, and introduced a new default backing storage for OSDs. for the latter we obviously have to support the older FileStore as well (as does Ceph itself of course), but testing and supporting all the other stuff in both variants would require a lot of extra effort.

basically the same reason why we don't support 3 different kernel branches in one release, or multiple Qemu versions.