When IP setup is done by the kernel (using DHCP), the hostname is already
set by the kernel before rc.sysinit is started. rc.sysinit should detect it
at leave it alone.
Created attachment 2216 [details]
Patch to fix it. Also prevents rc.sysinit from creating /etc/HOSTNAME - see my comments on BUG #3906
Out of curiousity, what are you doing with a bunch of NFS root
machines that's causing this spate of bugs? :)
We have a 16x2 cluster running RH 6.2. We've set the slave nodes to mount
root-over-NFS from the master node (they all mount the same partition withe the
same config files and because of that we do all the network configuration
through kernel-DHCP). Our latest achievment is mounting the root FS read-only
(but with no_root_squash) - it prints a buch of warnings about being unable to
change things like /etc/issue, but boots fine. The slave nodes have their own
HDs that we use for things like /tmp, /var and swap, but this root-on-NFS
makes administering much easier than giving each of the slaves its own copy of
The only thing that we had to change and that I haven't submitted as a bug is
that we had to make /etc/mtab a symlink to /proc/mounts - works surprizingly
well (even when root FS was RW, nobody tried to replace the symlink with
something else). Of course the better solution would be for all the
mount-related programs to know to leave /etc/mtab alone when /proc is available
(at least not to consider failure to change /etc/mtab to be critical), but I am
not sure if it's worth the effort.
Neat. I'm not sure about the mtab vs. /proc/mounts, but
I doubt we'd make one a symlink to another.
Fixed in CVS, will be in 5.40-somethingorother.
Commit pushed to master at https://github.com/openshift/origin
Merge pull request #15786 from jim-minter/issue15666
Automatic merge from submit-queue
fix swagger version reporting
It was the change in serialisation of "Version" to "version" under "x-kubernetes-group-version-kind" that made us hit the rogue sed command.