[Users] Post-Install Engine VM Changes Feasible?

[Giuseppe Ragusa]

Hi Joshua,

Date: Sat, 15 Mar 2014 02:32:59 -0400
From: josh at wrale.com
To: users at ovirt.org
Subject: [Users] Post-Install Engine VM Changes Feasible?

Hi,

I'm in the process of installing 3.4 RC(2?) on Fedora 19.  I'm using hosted engine, introspective GlusterFS+keepalived+NFS ala [1], across six nodes.

I have a layered networking topology ((V)LANs for public, internal, storage, compute and ipmi).  I am comfortable doing the bridging for each interface myself via /etc/sysconfig/network-scripts/ifcfg-*.  


Here's my desired topology: http://www.asciiflow.com/#Draw6325992559863447154

Here's my keepalived setup: https://gist.github.com/josh-at-knoesis/98618a16418101225726


I'm writing a lot of documentation of the many steps I'm taking.  I hope to eventually release a distributed introspective all-in-one (including distributed storage) guide.  


Looking at vm.conf.in, it looks like I'd by default end up with one interface on my engine, probably on my internal VLAN, as that's where I'd like the control traffic to flow.  I definitely could do NAT, but I'd be most happy to see the engine have a presence on all of the LANs, if for no other reason than because I want to send backups directly over the storage VLAN.  


I'll cut to it:  I believe I could successfully alter the vdsm template (vm.conf.in) to give me the extra interfaces I require.  It hit me, however, that I could just take the defaults for the initial install.  Later, I think I'll be able to come back with virsh and make my changes to the gracefully disabled VM.  Is this true? 


[1] http://www.andrewklau.com/ovirt-hosted-engine-with-3-4-0-nightly/

Thanks,
Joshua



I started from the same reference[1] and ended up "statically" modifying vm.conf.in before launching setup, like this:

cp -a /usr/share/ovirt-hosted-engine-setup/templates/vm.conf.in /usr/share/ovirt-hosted-engine-setup/templates/vm.conf.in.orig
cat << EOM > /usr/share/ovirt-hosted-engine-setup/templates/vm.conf.in
vmId=@VM_UUID@
memSize=@MEM_SIZE@
display=@CONSOLE_TYPE@
devices={index:2,iface:ide,address:{ controller:0, target:0,unit:0, bus:1, type:drive},specParams:{},readonly:true,deviceId:@CDROM_UUID@,path:@CDROM@,device:cdrom,shared:false,type:disk at BOOT_CDROM@}
devices={index:0,iface:virtio,format:raw,poolID:@SP_UUID@,volumeID:@VOL_UUID@,imageID:@IMG_UUID@,specParams:{},readonly:false,domainID:@SD_UUID@,optional:false,deviceId:@IMG_UUID@,address:{bus:0x00, slot:0x06, domain:0x0000, type:pci, function:0x0},device:disk,shared:exclusive,propagateErrors:off,type:disk at BOOT_DISK@}
devices={device:scsi,model:virtio-scsi,type:controller}
devices={index:4,nicModel:pv,macAddr:@MAC_ADDR@,linkActive:true,network:@BRIDGE@,filter:vdsm-no-mac-spoofing,specParams:{},deviceId:@NIC_UUID@,address:{bus:0x00, slot:0x03, domain:0x0000, type:pci, function:0x0},device:bridge,type:interface at BOOT_PXE@}
devices={index:8,nicModel:pv,macAddr:02:16:3e:4f:c4:b0,linkActive:true,network:lan,filter:vdsm-no-mac-spoofing,specParams:{},address:{bus:0x00, slot:0x09, domain:0x0000, type:pci, function:0x0},device:bridge,type:interface at BOOT_PXE@}
devices={device:console,specParams:{},type:console,deviceId:@CONSOLE_UUID@,alias:console0}
vmName=@NAME@
spiceSecureChannels=smain,sdisplay,sinputs,scursor,splayback,srecord,ssmartcard,susbredir
smp=@VCPUS@
cpuType=@CPU_TYPE@
emulatedMachine=@EMULATED_MACHINE@
EOM

I simply added a second nic (with a fixed MAC address from the locally-administered pool, since I didn't know how to auto-generate one) and added an index for nics too (mimicking the the storage devices setup already present).

My network setup is much simpler than yours: ovirtmgmt bridge is on an isolated oVirt-management-only network without gateway, my actual LAN with gateway and Internet access (for package updates/installation) is connected to lan bridge and the SAN/migration LAN is a further (not bridged) 10 Gib/s isolated network for which I do not expect to need Engine/VMs reachability (so no third interface for Engine) since all actions should be performed from Engine but only through vdsm hosts (I use a "split-DNS" setup by means of carefully crafted hosts files on Engine and vdsm hosts)

I can confirm that the engine vm gets created as expected and that network connectivity works.

Unfortunately I cannot validate the whole design yet, since I'm still debugging HA-agent problems that prevent a reliable Engine/SD startup.

Hope it helps.

Greetings,
Giuseppe

 		 	   		  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20140315/ff4bce9b/attachment.html>