FC LUNs for CTs + migration

[zeha]

Hi,

I've found some topics on this, but they did not really answer my question so far, so have to ask again:

We are planning to have a three node PVE 4.3 cluster, with LXC CTs, where most of the CTs will have "normal" LVM storage on top of FC exported LUNs.

Now some CTs should also get dedicated LUNs, without any LVM. Unfortunately I did not find a way to add these LUNs (appearing as /dev/disk/by-id/dm-uuid-mpath-<WWN> on the PVE hosts) to the PVE web UI.

I -can- add those LUNs directly with pct set, but then migration fails (cannot determine storage ... kinda obvious).

Example:
pct set 40001 -mp0 volume=/dev/disk/by-id/dm-uuid-mpath-360002ac0000000000300017b00004b3c,mp=/srv/db

What is the correct way of adding such 'raw' LUNs?

Thanks,
Chris

 

[LnxBil]

Unfortunately for you, there is none.

Why do you want to use a plain LUN for this? LVM is cluster aware and if you do all your work with the shared LVM, you will be safe concerning split brain and stuff like that.

 

[zeha]

It is my understanding "cluster awareness" comes from PVE handling it correctly, not strictly from the LVM storage plugin.

Why we want to use this: these LUNs happen to be (storage-level) snapshots of other LUNs; having LVM metadata on them can only confuse the PVE hosts. We have existing tooling to update these snapshots (and taking care of stopping the related containers, etc.)

 

[LnxBil]

It is my understanding "cluster awareness" comes from PVE handling it correctly, not strictly from the LVM storage plugin.

There is also cluster LVM, but it is not enabled per default, because it is not needed if everything is handled via PVE calls.

Why we want to use this: these LUNs happen to be (storage-level) snapshots of other LUNs; having LVM metadata on them can only confuse the PVE hosts. We have existing tooling to update these snapshots (and taking care of stopping the related containers, etc.)

Ah yes, if your storage supports this I see.
You can directly put the multipathed device name in your <vmid>.conf, but this is totally manual and there is no GUI option for this. Internally, the disk is passthroughed to the VM.

 

[zeha]

You can directly put the multipathed device name in your <vmid>.conf, but this is totally manual and there is no GUI option for this. Internally, the disk is passthroughed to the VM.

Well yes, but migration complains about the unknown storage type.

 

[LnxBil]

Oh yes. That'll happen. Maybe there will be an option for direct LUN mapping in the future. Should not be that hard to code that.

Can you create a feature-request in bugzilla for that?

 

[zeha]

This is now https://bugzilla.proxmox.com/show_bug.cgi?id=1169

 

[LnxBil]

And someone is already working on a similar issue. Nice!