A classical setup for a large number of workstations sharing the same applications is to have / installed locally on the hard drive of each workstation, but have /usr mounted by NFS (or other network filesystem) from some central server. Documentation regarding how to do so with "early userspace mounting" is lacking.
Also, apparently (and I didn't do this myself) dracut does not support mounting an nfs /usr out-of-the-box right now; something to do with the network not being started. This was reported by a fellow dev a few days ago, I don't know if he opened a bug or not. As such, there may not only be a general lack-of-documentation issue but also operational issues outstanding (the kind for which proper and specific documentation of workarounds is the only solution)
Do we have official documentation for this setup without an initramfs? If we do not, I question why it should be a blocker on this tracker. Thanks, William
(In reply to Ian Stakenvicius from comment #1) > Also, apparently (and I didn't do this myself) dracut does not support > mounting an nfs /usr out-of-the-box right now; something to do with the > network not being started. This was reported by a fellow dev a few days > ago, I don't know if he opened a bug or not. Don't highjack this bug with vague comments about bugs that might exist in specific initramfs tools. > As such, there may not only be a general lack-of-documentation issue but > also operational issues outstanding (the kind for which proper and specific > documentation of workarounds is the only solution) If there are specific documentation issues, they should be issues that are documented for non-initramfs setups but not documented for initramfs setups.
I think the concern on this bug is about whether the initramfs generators such as Dracut and genkernel support mounting /usr early over nfs. I submit that although that may be an issue, it isn't a blocker for bug #481202, because you can still roll your own initramfs and make sure to include support for it.
All, I am removing the docs team from this bug, because after talking to AxS, it seems the concern is about the specific tools, dracut/genkernel, not supporting an nfs-mounted /usr correctly. If that's the case, like I said above, I'm not sure that constitutes making this bug block #481202.
A review of genkernel code indicates that a genkernel initramfs will support mounting /usr on NFS , including both the case when root is an nfsroot and when it isn't. Although other methods are probably not so straight-forward, a simple 'genkernel --install initramfs' seems to handle this and is simple enough to not require any special documentation. (Please note I haven't done a runtime test of any of this, but the code is all there and everything seems enabled by default)
@blueness: Based on comment #6, can we close this bug since there seems to be no need for special documentation to cover this issue?
Can somebody confirm that nfs mounting of /usr actually works. In my attempt to do this in a VM I get a protocol not supported error. I'll keep plugging away at it - at least it seems closer than dracut. Let's not debate here whether this is a show-stopper and just keep to the facts. If anybody has any suggestions for what is causing the protocol not supported error feel free to send them to me via email/irc/etc and we can report back status here. I'm not an NFS guru, but seems like most kernel-related errors in that it is pretty vague.
(In reply to Richard Freeman from comment #8) > Can somebody confirm that nfs mounting of /usr actually works. > > In my attempt to do this in a VM I get a protocol not supported error. I'll > keep plugging away at it - at least it seems closer than dracut. Let's not > debate here whether this is a show-stopper and just keep to the facts. > > If anybody has any suggestions for what is causing the protocol not > supported error feel free to send them to me via email/irc/etc and we can > report back status here. I'm not an NFS guru, but seems like most > kernel-related errors in that it is pretty vague. I trust Ian's judgment on this but a test would help. This is a classic setup so its hard to imagine that genkernel doesn't generate nfs-friendly code. @comment 4. Of course users can always do things for themselves, but we would like to make it easy if possible. So if they did need to generate their own initramfs, then they should have some guide. However, this doesn't seem to be the case since genkernel does it for them. I'm cc-ing the genkernel folks to confirm Ian's observation.
Seems like this "just works", so closing the bug. Please re-open if you disagree.
