Status of iSER support.
After searching the official documentation, wiki and the code on gitlab, i still can not find any useful information about how to handle iSER targets with libvirt. I did find some RFCs for patches on this mailing list, but it seems none of them have been merged. There is also a somewhat related issue open on gitlab (https://gitlab.com/libvirt/libvirt/-/issues/16). So effectively my question is the following: Is libvirt currently capable of supporting iSER targets as either storage pools or as virtio device and if not, is an implementation of such support planned? Thanks in advance for your responses, David Schweizer
On Sun, May 10, 2020 at 16:09:35 +0200, David Schweizer wrote:
After searching the official documentation, wiki and the code on gitlab, i still can not find any useful information about how to handle iSER targets with libvirt. I did find some RFCs for patches on this mailing list, but it seems none of them have been merged. There is also a somewhat related issue open on gitlab (https://gitlab.com/libvirt/libvirt/-/issues/16).
So effectively my question is the following: Is libvirt currently capable of supporting iSER targets as either storage pools or as virtio device and if not, is an implementation of such support planned?
The following patchset attempts to implement iser transport for disk source specification: https://www.redhat.com/archives/libvir-list/2020-April/msg01247.html It's fairly recent so you can give it a try and ideally provide a 'Tested-by:'. I don't have a RDMA setup handy so I can't test it and thus can provide a code-only review. In case of iSCSI storage pools I don't think that anybody is working on iser support.
Hello Peter, Thank you for your answer. The problem is that i would like to use the feature in a production environment and can not rely on out of tree patches. In regard to my question i conclude the following: - There is currently no support for iSER storage objects in libvirt. - There will not be support for iSER storage objects in the next few libvirt releases. Which for me, at least at the moment, means dropping iSER/IB, using iSCSI/TCP/IPoIB and loosing some performance or using a commercially available hypervisor. Kind regards, David On 5/11/20 9:03 AM, Peter Krempa wrote:
The following patchset attempts to implement iser transport for disk source specification:
https://www.redhat.com/archives/libvir-list/2020-April/msg01247.html
It's fairly recent so you can give it a try and ideally provide a 'Tested-by:'. I don't have a RDMA setup handy so I can't test it and thus can provide a code-only review.
In case of iSCSI storage pools I don't think that anybody is working on iser support.
On Tue, May 12, 2020 at 08:08:28 +0200, David Schweizer wrote:
Hello Peter,
Thank you for your answer.
The problem is that i would like to use the feature in a production environment and can not rely on out of tree patches.
Well, you certainly can contribute in having them merged. Even testing is a contribution.
In regard to my question i conclude the following: - There is currently no support for iSER storage objects in libvirt.
No.
- There will not be support for iSER storage objects in the next few libvirt releases.
Depends on the status of the patches. I've reviewed them yesterday and I don't agree with the design. If you have more knowledge of how those things work, even explaining them is a contribution to possbily adding the feature.
Which for me, at least at the moment, means dropping iSER/IB, using iSCSI/TCP/IPoIB and loosing some performance or using a commercially available hypervisor.
Well, for us that means that we don't have anybody willing to give input on the feature. I don't have experience nor hardware for testing it and don't have knowledge on how e.g. addressing works.
Hello Peter, On 5/12/20 8:33 AM, Peter Krempa wrote:
Well, you certainly can contribute in having them merged. Even testing is a contribution. [..] Well, for us that means that we don't have anybody willing to give input on the feature. I don't have experience nor hardware for testing it [..].
I also have only one capable host left, but no counterpart to test with. Other hardware is either running in production environment or reserved for emergencies.
Depends on the status of the patches. I've reviewed them yesterday and I don't agree with the design. If you have more knowledge of how those things work, even explaining them is a contribution to possbily adding the feature.
I mean it is kind of tricky to implement, since iSER sits somewhat between the iSCSI protocol and different transport methods. Maybe implementing it as its own protocol capable of communicating over different transport methods (RoCE, Infiniband, etc. (NOT: RDMA)) is the way to go. I think the author simply tried to adapt it to the implementation in qemu as best as he could to avoid self inflicting head wounds with his keyboard during alteration of the libvirt qemu driver. Kind regards, David
participants (2)
-
David Schweizer -
Peter Krempa