On Apr 9, 2013, at 2:20 PM, "J. Bruce Fields" <[email protected]> wrote:
> On Tue, Apr 09, 2013 at 02:01:09PM -0400, Chuck Lever wrote: >> Asking the NFS server gurus.... >> >> As part of the next release of fedfs-utils, I'd like to provide more tools >> that can hide the details of setting up a FedFS domain. One of the first >> tasks when setting up a domain is to create a FedFS domain root directory. >> Here are the instructions I provide for fedfs-utils 0.9 (the latest release): >> >> http://wiki.linux-nfs.org/wiki/index.php/FedFsNfsDomainRoot0.9 >> >> I'm kind of brainstorming about this right now, not necessarily attached to >> any particular solution or to the naive way we are doing it now. >> >> It would be nice if we had a tool that would ensure that all the NFS-related >> infrastructure was in place: >> >> o Starting and enabling the NFS service as needed >> o Verifying the junction resolution plug-in is installed >> o Setting up the /.domainroot export if it doesn't exist >> >> The tool would have the administrator simply specify the name of new domain. >> The outcome would be a directory like "/.domainroot/example.net" that would >> be automatically exported with the correct security flavors and other >> settings. The NFS server that shares a domain root can be used for more than >> one domain root, so this process could be done more than once on a >> particular NFS server. >> >> Afterwards, an administrator would use nfsref or mkdir to customize the >> contents of the domain root directory. We could have the tool create >> junctions in the domain root directory, no files or directories. Not sure >> if that's useful: could be a simplification for our administrative >> interface, and we could continue to allow arbitrary "mkdir" and "nfsref" in >> this directory, like any other exported directory, but those would not be >> managed with the setup tool. >> >> On NFS servers I've set up for this purpose, I create a separate logical >> volume with a filesystem mounted at /.domainroot. This avoids exporting a >> piece of / on the server. But maybe there's a better way to go about this. > > Sounds OK to me. Some kind of in-memory filesystem would work, I guess? My concern with tmpfs is that when the server reboots, wouldn't the file handles change? Another alternative would be to allocate a 20MB file in / or /var, create a filesystem on that, then loopback mount it. Creating a physical /.domainroot directory on the server is a little sneaky too. Is there a guru-approved way I can have the domain root mounted, say, under /var/lib/fedfs and then export it as /.domainroot? -- Chuck Lever chuck[dot]lever[at]oracle[dot]com _______________________________________________ fedfs-utils-devel mailing list [email protected] https://oss.oracle.com/mailman/listinfo/fedfs-utils-devel
