[pve-devel] nexent questions

Dietmar Maurer dietmar at proxmox.com
Mon Sep 17 20:06:37 CEST 2012


> We currently pass the following url to kvm:
> 
> iscsi:$portal/$target/lun
> 
> >>but the nexenta server does not always expose iscsi on ‘$portal’.
> >>Instead, there are target groups. Any idea how to fix that?
> maybe can we replace  the portal option by "management" ip.
> and add target ip option ?

no, we should query the API to get the correct IPs. But I can't get that to work - and
query to object 'iscsitgt' fails for me - any ideas?

> >>Also, libiscsi does not support multipath, so maybe we are better off not
> using it?
> >>Can we use standard open-iscsi + nexenta api?
> Indeed, with libiscsi, you need to do multipath inside the guest. 

how should that work?

> I'm not sure
> you can do true iscsi passthrough with multipath on host)
> 
> 
> for open-iscsi,It's possible, I have done some work some month ago, But
> having a lot of devices on host is difficult (but no impossible) to maintain.
> We really need to add/remove the device from the host when the vm is
> starting.
> some problem if not : when we delete/unmap a volume, we need to rescan
> the iscsi + multipath layer to remove the device.
> 
> The main problem was that open-iscsi scan all devices at iscsi login.
> (I have done some tricks with udev settle, deleting all devices after the iscsi
> login)
> 
> Also when snapshot rollback, we need to unmap the iscsi lun, and the device
> wwid is changing.
> Maybe some other problem, but I don't remember.
> 
> 
> Maybe can we implement both, libiscsi + open-iscsi ?

Yes, maybe a boolean option 'libiscsi: yes|no'?



More information about the pve-devel mailing list