Issue

• After upgrading two of our hypervisors in a 4 node cluster last night from kernel 2.6.32-431.5.1 to 2.6.32-431.17.1, alerts were received from monitoring system.
• For virtual machines attached to an untagged logical network in the cluster (vcp_aa) in which the virtual machines would lose connectivity or stop responding to our monitor's probes for roughly 1 - 3 minutes, after which affected virtual machines would come back online.
• The virtual machines themselves are running fine, and after doing some testing on a specific virtual machine after a receiving an alert for it, on the VM the default gateway could be pingable, but nothing beyond that, for 1 - 3 minutes, after which connectivity/network visibility is restored.
• On the messages file on the hypervisor :

Jun 19 20:32:45 AA kernel: __ratelimit: 50 callbacks suppressed
Jun 19 20:32:45 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=ab!
Jun 19 20:32:45 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=ac!
Jun 19 20:32:46 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=ad!
Jun 19 20:32:47 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=ae!
Jun 19 20:32:48 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=af!
Jun 19 20:32:52 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=b0!
Jun 19 20:32:58 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=b1!
Jun 19 20:33:11 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=b2!
Jun 19 20:33:21 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=0!
Jun 19 20:33:28 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=72!
Jun 19 20:33:28 AA kernel: ixgbe 0000:0b:00.1: partial checksum but l4 proto=fe!