[Linux-cluster] Installing on FC4

[Birger Wathne]

Now that FC4 has been released (finally), I thought I'd post a note with 
the challenges I met when upgrading my 1-node cluster from FC3.

I did a full reinstall of the operating system.

I tried to install the cluster packages using yum (both from the base 
and the develop repositories), but found that old bugs that have been 
fixed in cvs were still present in those packages.

In the end I kept device-mapper from 'base', but reinstalled both 
'cluster' and 'LVM2' from source. My reason for reinstalling lvm2 was 
simply that trying to add the lvm2-cluster package wanted to drag in ccs 
and some others as rpm's. I didn't want that.

To fetch sources:
cvs -d :pserver:cvs at sources.redhat.com:/cvs/cluster login cvs
cvs -d :pserver:cvs at sources.redhat.com:/cvs/cluster checkout -r FC4 cluster
cvs -d :pserver:cvs at sources.redhat.com:/cvs/lvm2 checkout LVM2

To build I did

cd cluster
./configure --kernel_src=/lib/modules/`uname -r`/build/
make
make install

cd ../LVM2
./configure --with-clvmd --with-cluster=shared
make
make install

But it wasn't really that simple. :-(
Compilation of cluster failed in 2 places.

I removed -Werror from the options to gcc in magma/lib/Makefile to get 
it to compile with a  warning.
I also removed the 'static int' declaration of loglevel in 
rgmanager/src/clulib/clulog.c
leavin gonly the initialisation of the variable.

With these 2 changes I got the thing installed and started.

My next problem was that I got a lot of
clurgmgrd[11283]: <notice> status on nfsclient "XXX" returned 1 (generic 
error)
and I know from previous experience that this can give I/O errors on the 
clients when rgmanager decides to unexport and reexport the file system, 
so it has to get fixed.
My temporary fix is to edit /usr/share/cluster/nfsclient.sh and replace 
this line
exportfs | grep -q "^${OCF_RESKEY_path}\ .*${OCF_RESKEY_target}"
with the line
grep -q "^${OCF_RESKEY_path}[   ]*${OCF_RESKEY_target}" /var/lib/nfs/etab

IMPORTANT NOTE: The characters between [] are a space and a tab.

For some strange reason exportfs doesn't list exported file systems...
I have one file system defined in /etc/exports, just to get nfs up at 
boot. Just 'exportfs' doesn't list anything. 'exportfs|cat' sometimes 
lists this one export, sometimes nothing. 'cat /var/lib/nfs/etab' shows 
all exports, including those added by rgmanager.

Another note is that hosts may not show up in etab with the same name 
you used in the nfsclient line in the config. I guess it uses the result 
from a reverse lookup in DNS? Check messages, and if you get errors, 
find the name used in etab and change your nfsclient line to use the 
same name.

The only other issues I have run into (not cluster related) are:
The TSM backup client README states I need to have 
compat-gcc-c++-7.3-2.96.122.i386.rpm to install the tsm rpm's. I found 
that on FC4 I needed compat-libstdc++-33 instead. Not tested yet, just 
installed.
Some problem getting the samba server to join the AD domain. It creates 
the host in AD, but seems to hang forever afterwards. Perhaps time I 
double checked that I have all the iptables openings with me from FC3 
(where it used to work).

-- 
birger