[Users] Installing a lab setup from scratch using F18

[noc]

Sorry for the late reply, see comments inline.

> 2013-1-22 2:11, Joop:
>> As promised on IRC (jvandewege) I'll post my findings of setting up 
>> an ovirt lab environment from scratch using F18.
>> First some background:
>> - 2 hosts for testing storage cluster with replicated gluster data 
>> and iso domains (HP ML110 G5)
> gluster data or cluster data?
Gluster
>
>> - 2 hosts for VMs (HP DL360 G5?)
>> - 1 managment server (HP ML110 G5)
>> All physical servers have atleast 1Gb connection and they also have 2 
>> 10Gb ethernet ports connected to two Arista switches.
>> Complete setup (except for the managment srv) is redundant. Using 
>> F18-x64 DVD and using minimal server with extra tools, after install 
>> the ovirt.repo and the beta gluster repo is activated.
>> This serves as a proof of concept for a bigger setup.
>>
>> Problems sofar:
>> - looks like F18 uses a different path to access video since using 
>> the defaults leads to garbled video, need to use nomodeset as a 
>> kernel option
> Do you mean the VDSM host installed with F18 or the guest in the host 
> with FC 18 here? I suppose you don't have much chance to access the 
> graphics screen of the VDSM host.
Talking about the hosts.
>
>> - upgrading the minimal install (yum upgrade) gives me 
>> kernel-3.7.2-204 and the boot process halts with soft locks on 
>> different cpus, reverting to 3.6.10-4.fc18.x86_64 fixes that. 
>> Managment is using 3.7.2 kernel without problems BUT it doesn't use 
>> libvirt/qemu-kvm/vdsm, my guess its related.
>> - need to disable NetworkManager and enable network (and ifcfg-xxx) 
>> to get network going
>> - adding the storage hosts from de webui works but after reboot vdsm 
>> is not starting, reason seems to be that network isn't initialised 
>> until after all interfaces are done with their dhcp requests. There 
>> are 4 interfaces which use dhcp, setting those to bootprotocol=none 
>> seems the help.
> Usually, I use the static IP for each server because the engine add 
> the VDSM host by IPs.
I set a static address too but still NetworkManager runs and tries to do 
its best.
The DL360 machines run 3.7.2 without problems BUT they don't run the 
latest gluster-3.4, that might be the difference or else they are immune 
for this specific problem between 3.7.2 and ML110 hardware.

>
>> - during deploy their is a warning about 'cannot set tuned profile', 
>> seems harmless but hadn't seen that one until now.
>> - the deployment script discovers during deployment that the ID of 
>> the second storage server is identical to the first one and abort the 
>> deployment (Blame HP!) shouldn't it generate a unique one using 
>> uuidgen??
>>
>>
Host deployment went OK but for one thing and that the storage domain 
wouldn't attach. Putting the host in maintenance mode and activating did 
not help. It turnes out there are two problem of which one is already 
solved:
- ntpd/chronyc SOLVED thanks to Gianluca
- host needs to have SELinux set to Permissive and then attach storage 
domain works.
- on the managment host I still need NFSv3 settings in/etc/nfsmount.conf 
or else iso-upload doesn't work


>> Things that are OK sofar:
>> - ovirt-engine setup (no problems with postgresql)
>> - creating/activating gluster volumes (no more deadlocks)
>>
>> Adding virt hosts has to wait til tomorrow, got problems getting the 
>> dvd iso onto an usb stick, will probably burn a DVD to keep going.
Migration works OK, powermanagment through ipmilan/ilo2 works, 
export/import VMs, v2v test from esxi to ovirt, all OK ;-)


Setting up a storage network still needs to be tested

Joop
--
irc:jvandewege