[manila] What's the status of the virtiofs driver?
Hi, I've been waiting for this promising feature for a long time already, so I went to check, and saw that in launchpad [1], the blueprint seemed implemented. Though I didn't see anything in Manila's docs about it. So what's the status of this driver? Is it soon ready? Cheers, Thomas Goirand (zigo) [1] https://blueprints.launchpad.net/nova/+spec/libvirt-virtiofs-attach-manila-s...
As I'm waiting it as well, I've checked the status of patches [1] no later that yesterday, and as you might see - it's heavily in progress. Reviews for some patches were left just couple of days ago. But 1 thing is clear - this won't happen for 2024.1 [1] https://review.opendev.org/q/topic:%22bp/manila_shares_attachments_v7%22 On Fri, Mar 1, 2024, 23:46 Thomas Goirand <zigo@debian.org> wrote:
Hi,
I've been waiting for this promising feature for a long time already, so I went to check, and saw that in launchpad [1], the blueprint seemed implemented. Though I didn't see anything in Manila's docs about it.
So what's the status of this driver? Is it soon ready?
Cheers,
Thomas Goirand (zigo)
[1]
https://blueprints.launchpad.net/nova/+spec/libvirt-virtiofs-attach-manila-s...
+1 I am on same boat On Sat, Mar 2, 2024 at 1:33 AM Dmitriy Rabotyagov <noonedeadpunk@gmail.com> wrote:
As I'm waiting it as well, I've checked the status of patches [1] no later that yesterday, and as you might see - it's heavily in progress. Reviews for some patches were left just couple of days ago. But 1 thing is clear - this won't happen for 2024.1
[1] https://review.opendev.org/q/topic:%22bp/manila_shares_attachments_v7%22
On Fri, Mar 1, 2024, 23:46 Thomas Goirand <zigo@debian.org> wrote:
Hi,
I've been waiting for this promising feature for a long time already, so I went to check, and saw that in launchpad [1], the blueprint seemed implemented. Though I didn't see anything in Manila's docs about it.
So what's the status of this driver? Is it soon ready?
Cheers,
Thomas Goirand (zigo)
[1]
https://blueprints.launchpad.net/nova/+spec/libvirt-virtiofs-attach-manila-s...
we expect this functionality to merge early in 2024.2 over the course of 2024.1 the required sdk/manial apis that were not aviable in 2023.2 have all been merged. 3 weeks ago when reviewing the nova implementation we discovered the implementation had diverged form the approved spec changing the semantics of when the shares would be attached which altered the failure modes and security aspects of the functionality. while we suspect that had happened a cycle or two ago due to the lack of some functionality in manila (the locking apis) or similar justified reasons it was not documented and there were a number of hard to solve problems with the divergence so we ask for the implementation to be refactored to conform to what we had approved. René has been trying to accommodate the review feedback but was not able to compelte the refactoring in time for the feature freeze. As a core team nova's intent is to continue to review this work and aim to merge it early in the next cycle. for me that means aiming for before milestone 1. partically specaking that means when RC 1 is released in 2 weeks the master branch will reopen and we could start merging the serise again if Rene has completed the refactoring. pragmatically we may wait 4 weeks until the actull 2024.1 release has happened to merge this work. i have not been directly involed in the feature review to any great detail as there were a number of other features i had signed up to review ahead of this but this is my understanding of where the manila share functionality lies. Code review velocity has been an problem for a number of releases for complext features and while we have context loaded on this complex feature we do not want to loose the momentem on this as we do belive this to be a valuable improvement. im glad to hear that ye are both interested in this functionality. if yes have time and inclination feel free to test out the work in fling and give us feedback in the nova irc channel or on gerrit. regards sean On Sun, 2024-03-03 at 10:06 -0500, Satish Patel wrote:
+1 I am on same boat
On Sat, Mar 2, 2024 at 1:33 AM Dmitriy Rabotyagov <noonedeadpunk@gmail.com> wrote:
As I'm waiting it as well, I've checked the status of patches [1] no later that yesterday, and as you might see - it's heavily in progress. Reviews for some patches were left just couple of days ago. But 1 thing is clear - this won't happen for 2024.1
[1] https://review.opendev.org/q/topic:%22bp/manila_shares_attachments_v7%22
On Fri, Mar 1, 2024, 23:46 Thomas Goirand <zigo@debian.org> wrote:
Hi,
I've been waiting for this promising feature for a long time already, so I went to check, and saw that in launchpad [1], the blueprint seemed implemented. Though I didn't see anything in Manila's docs about it.
So what's the status of this driver? Is it soon ready? it would not be in the manila docs as its a nova feature. it was functional this cycle with bugs/spec divergance but not ready for production or merging as a result.
Cheers,
Thomas Goirand (zigo)
[1]
https://blueprints.launchpad.net/nova/+spec/libvirt-virtiofs-attach-manila-s...
On 3/4/24 12:56, smooney@redhat.com wrote:
we expect this functionality to merge early in 2024.2 over the course of 2024.1 the required sdk/manial apis that were not aviable in 2023.2 have all been merged. 3 weeks ago when reviewing the nova implementation we discovered the implementation had diverged form the approved spec changing the semantics of when the shares would be attached which altered the failure modes and security aspects of the functionality.
while we suspect that had happened a cycle or two ago due to the lack of some functionality in manila (the locking apis) or similar justified reasons it was not documented and there were a number of hard to solve problems with the divergence so we ask for the implementation to be refactored to conform to what we had approved.
René has been trying to accommodate the review feedback but was not able to compelte the refactoring in time for the feature freeze. As a core team nova's intent is to continue to review this work and aim to merge it early in the next cycle. for me that means aiming for before milestone 1.
partically specaking that means when RC 1 is released in 2 weeks the master branch will reopen and we could start merging the serise again if Rene has completed the refactoring. pragmatically we may wait 4 weeks until the actull 2024.1 release has happened to merge this work. i have not been directly involed in the feature review to any great detail as there were a number of other features i had signed up to review ahead of this but this is my understanding of where the manila share functionality lies. Code review velocity has been an problem for a number of releases for complext features and while we have context loaded on this complex feature we do not want to loose the momentem on this as we do belive this to be a valuable improvement. im glad to hear that ye are both interested in this functionality.
if yes have time and inclination feel free to test out the work in fling and give us feedback in the nova irc channel or on gerrit. regards sean
Hi Sean, Thanks a lot for this detailed status update. That's very helpful for planning my future work. Indeed, we're kind of excited and impatient to have this new feature, because, as we see it, the CephFS driver requires the customers to access the Ceph cluster, and the Generic driver isn't providing any type of redundancy (the VM hosting the cinder block + NFS storage, plus the single worker are both points of failure without redundancy). We see the virtiofs as the only real option for seriously providing manila support in a public cloud environment. If I follow you well, there's a chance to have a set of patch early after the Caracal release. If that's the case, then I probably can easily gather the patches on top of Caracal, at least in my (nested virtualization) lab, to test-proof it, and probably add support for it in puppet-openstack, so it may be ready this summer before the Dalmatian release (hopefully, I'll find enough time for it). That's a very good thing already, and I wasn't expected much more (I mean: 6 months development is quite fast for such a feature). So, thanks to everyone involved! :) Am I then right to say that the only missing parts are in Nova? Or is there more to do in Manila itself as well? Cheers, Thomas Goirand (zigo)
I’m replying to zigo’s mail, but the same goes for the others here.
Am I then right to say that the only missing parts are in Nova? Or is there more to do in Manila itself as well?
Just curious, but have the people who are interested in this tried pulling the series and testing it for their intended use-case? This is a big thorny series (as evidenced by the long lead-time and still lots of review/refactoring). It would be very helpful and informational (IMHO) if people that want to deploy this in production could help confirm that it works as they expect and help highlight any issues they may see. Confirming the cost/benefit of something like this and confirming that the design works for the intended use is helpful in getting comfortable with merging. —Dan
participants (5)
-
Dan Smith
-
Dmitriy Rabotyagov
-
Satish Patel
-
smooney@redhat.com
-
Thomas Goirand