Re: [OSR-devel] OSR on DRBD+GFS

[Gordan B. <go...@bo...>]

Marc Grimme wrote:
> Hi Gordan,
> On Saturday 09 February 2008 01:56:18 Gordan Bobic wrote:
>> Hi,
>>
>> I'm trying to add support for DRBDs, and trying to stick with doing it
>> in a way consistent with iSCSI support as far as it is possible. I'm
>> seeing a few weird things, though.
>>
>> In linuxrc.generic.sh:
>> [...]
>> # start iscsi if apropriate
>> isISCSIRootsource $rootsource
>> if [ $? -eq 0 ]; then
>>          loadISCSI
>>          startISCSI $rootsource $nodename
>> fi
>>
>> # start drbd if appropriate
>> isDRBDRootSource $rootsource
>> if [ $? -eq 0 ]; then
>>          loadDRBD
>>          startDRBD $rootsource $nodename
>> fi
>> [...]
>>
>> and in cluster.conf:
>> [...]
>> <clusternodes>
>>                  <clusternode name="sentinel1c" nodeid="1" votes="2">
>>                          <com_info>
>>                                  <rootsource name="drbd"/>
>>
>> [...]
> Agreed. That's the place.You also have a bootoption rootsource to overwrite 
> it. Normally we would suspect rootsource to be some kind of url like for 
> example iscsi://iscsi-host/export-name or it might be drbd://slave/export or 
> something. I cannot remember the semantics there too long ago ;-) .
> 
> Are you also aware of the com-expert bootoption? This one is quite handy as it 
> brings you to a shell before the process of the initrd is started. And you 
> can change some things to see if they work or not.
> 
> Answer to the iscsi questioning stuff.
> [marc@mobilix-07 boot-scripts]$ source etc/iscsi-lib.sh
> [marc@mobilix-07 boot-scripts]$ isISCSIRootsource drbd
> [marc@mobilix-07 boot-scripts]$ echo $?
> 1
> [marc@mobilix-07 boot-scripts]$ isISCSIRootsource iscsi
> [marc@mobilix-07 boot-scripts]$ echo $?
> 0
> [marc@mobilix-07 boot-scripts]$
> 
> So if we are going iscsi we'll return a '0'. As a successfully executed 
> shellscript does.
>>  From this, one would assume that isISCSIRootsource would return false
>> and iSCSI not get initialized, and isDRBDRootSource would return true
>> and get initialized.
>>
>> This is the opposite of what I'm seeing. iSCSI gets started, and the
>> modules loaded. DRBD doesn't.
>>
>> Am I misunderstanding where $nodename is coming from? I'm not
>> particularly bothered by iSCSI trying to start (and failing because the
>> iscsi packages aren't in the rpm list), but that makes the drbd not
>> starting all the more puzzling. :-/
 >
> $nodename is the nodename that belongs to the node in bootprocess. But it is 
> not necessarily the hostname. The function 
> cluster-lib.sh/getClusterFSParameters returns it. It's a little bit calling 
> other functions to make it more independent from the cluster.conf (that was 
> the idea in the first place).

Ah, I see. I think it may be necessary to do
hostname $nodename
in that case, since DRBD requires hostname to be matching the node name 
in the DRBD configuration.

>> Further on (unrelated to this), I get cman starting, but ccsd failing,
>> so the boot-up aborts. But ps -aux | grep ccsd shows that ccsd is in
>> fact already running. I haven't seen this behaviour before. The only
>> thing I can think of that is different is that this is a 2-node cluster,
>> which is much smaller than what I usually work with. (Yes I did set
>> <cman two_node="1" expected_votes="2"/>)
 >
> Enable syslogging to a syslog server. <syslog name="server"/> does the trick 
> and see what it tells you.

I'm feeling a bit silly now. The answer was in the line I pasted. For 
two_node=1, expected_votes needs to be 1 as well. :-/

On a separate note, is there a configurable way to disable LVM/CLVM, 
even if only at mkinitrd time? All it is achieving for me is throwing up 
ugly "FAILED" startup messages because I'm not using LVM. I could 
comment it out of the boot scripts, but that seems a bit nasty.

I'm about 99% there with getting a single node to boot of the 
OSR/GFS/DRBD. I just need to write a quick wrapper and it'll be ready. 
Hoping to post a patch tomorrow.

Gordan