After install or upgrade to the Oracle VM Server 3.1.1 connected more than 8 multipathed (4 path) volumes of 2* Sun ST2540 disk array
disappears from the hypervisor interface eth0 (all 8 interfaces) (?) in 3.0.3 does not have this problem. Kernel options?
Pardon? You updated OVM server from 3.0.3 to 3.1.1 and lost your NICs? Could you elaborate more on the setup/HW of your host? Otherwise it's nearly impossible to judge what might have been going wrong.
Unfortunately, but it's possible. As I wrote (regardless whether it's an install, upgrade or update) after adding more than ~8 volumes (4 paths each) after reboot Server 3.1.1 eth0 disappears (bond0 does not work). You can reduce the number of volumes or disconnect hypervisor from the SAN to restore correct operation of eth0 (without altering the configuration).
Hypervisor without SAN mappded volumes boot time ~2min, with all mapped volumes ~120min!!!
I didn't have this problem with any previous version (2.2.2, 3.0.3).
My hardware configuration:
+4* Dell 910 (8* eth, 4* fc)+
+2* Broadcom SAN switch (2 fabrick)+
+2* Storage Sun ST2540+
Sounds like a bug for sure. Probably want to open a ticket. Make sure you have your FC zoning right.
I had something wierd like this happen on a OEL/Redhat server one time and it was due the fact that the Ethernet connections where actually renumbering themselves after configuring bonding interfaces. Just a thought.
SR 3-5508905681 : bnx2 driver fail to load when disks are attached through lpfc driver
we experience a problem with bnx2 when apron. more than 4 volumes are mounted on boot time. Unfortunately no response since may on this SR. It is really pity since I was expecting OVS 3.1.1 new kernel would fix it
Verified that both fabricks are configured correctly.
There was no renumbering of ethernet ports, only the first port in the running configuration is missing, even though configuration files are correct (with the right MAC address) for the port in ifcfg-eth0.
Network configuration parameters for the VM Server and VM Manager are definitely properly set (in OVS+OVM-3.0.3 and OVS-3.0.3+OVM-3.1.1 works flawlessly)
Kirill, I think you're right. my problem is almost the same as observed by you.
I have the impression that either the problem lies in the newer modules for Broadcom (bnx2i) or Emulex (lpfc) or in the kernel compile options in version 3.1.1.
We use IBM V7000 (svc) to virtualize storage. One possible workaround in our environment for such issue is to map only few volumes on boot time and after attach them in hot mode. It works for us, as we have not a lot of reboots in production, but we definitely expect Oracle to fix this. Just a few days ago I received a response on this SR, so I will update this thread as I will progress
PS Actually I am really frustrated :) as one of my target for small upgrade project from 3.0.3 to 3.1.1 was to eliminate this problem. I was thinking that it would be fixed.