Upstream switches may detect bridging loop from compute nodes
Symptom
After booting up two or more virtual machines on the same compute node, if there is a Virtual Network misconfiguration, networking may appear to "drop" from the compute node until the upstream network ports are re-enabled. On an HP blade chassis, the only way to re-enable the internal network ports to a blade is to reapply the blade profile.
Conditions
The following conditions must be true: - Multiple Virtual Machines are being deployed to the same compute node. - The VMs must reside on separate Virtual Networks. - Network traffic appears to drop to the compute node after booting the VMs. - Upstream switch logs may show bridging loop detection events.
Root Cause
This issue is caused by using the same "BRIDGE" attribute across multiple virtual networks. BRIDGE should ONLY be the same if sharing the same VLAN ID. HyperCloud will allow this configuration to be applied since it's a potentially valid configuration. Take very special care when creating Virtual Networks to set the "BRIDGE" and "VLAN_ID" attributes properly!
Workaround
To remediate an environment where this misconfiguration has occurred, multiple steps
must be taken:
- Update all Virtual Networks to have properly set "BRIDGE" names.
- For all VMs that were attached to the Virtual Networks, detach and reattach the NICs.
- If a rogue u-br0 device (or whatever the offending names was) exists, it may need
to be manually removed using a command similar to ip link del dev u-br0.