Hello everyone,
This is a bit of a strange problem that I've been chasing after for some time now. I'm at the point where I'm totally stumped and so I turn to this wonderful community.
Bit of background, I'm running 16 HP BL460c's (14 x G7, 2 x Gen8). I use all of these as Xen hosts and they are managed through Salt. All of these blades use software bridging to allow the VMs to reach the network: the host blade creates the bridge, adds the physical ethernet devices and as VMs are created their virtual interfaces are added to the bridge. This has worked on all of my G7s.
On the Gen8s, our VMs have no access to the network. I can see the VM dhcp request leave the bridge and the answer comes back to the bridge but doesn't make it back to the virtual interface.
So the thinking process went (these aren't in order of occurrence)
- Assume it was a bad VM. Moved it to a G7 and it's fine
- Maybe it's iptables. Disabled/Enabled and no difference
- Maybe chassis is having issues. Move Gen8 to new slot, same issue. Move G7 to Gen8 slot, G7 is fine
- Maybe it's a bad NIC on the Gen8? Unlikely since it's happening on two of them. Got a new mezz card from HP and swapped that out to no avail
- Maybe it's something that's because of the automation? Let's do this manually. No luck
ARP requests from the VM to: vif > bridge > bond > world
but replies are: world > bond > bridge > black hole of doom
- Maybe it's because of the ethernet bonding? That'd be weird since we're doing it on our G7s too
One thing makes this setup work -- if I make the bridge a hub, everything is okay
[root@claymore ~]# cat /etc/sysconfig/network-scripts/ifcfg-xenbr0
DEVICE=xenbr0
TYPE=Bridge
ONBOOT=yes
BOOTPROTO=dhcp
NM_CONTROLLED=no
DELAY=0
BRIDGING_OPTS="ageing_time=0"
So maybe it's something having to do with the bridge and how it interacts with the FlexFabric? I'm honestly completely stuck with this one and so any help would be appreciated.
[link][4 comments]