yahoo-eng-team team mailing list archive

[Robert Pothier <rpothier@xxxxxxxxx>]

This issue is in the switches in the datapath, and not neutron/ML2


** Changed in: neutron
       Status: New => Invalid

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1412900

Title:
  Cisco Nexus VxLAN ML2: UDP large packets dropped in L3 data path

Status in OpenStack Neutron (virtual network service):
  Invalid

Bug description:
  UDP packets larger than 1428 bytes are being dropped (100% packet loss) in an L3 configuration.
  Testbed configuration to reproduce:
  - 1 TOR N9K with 1 compute node
  - 1 TOR N9K with 1 network node and 1 compute node
  - 1 Spine connecting the 2 TOR
  Run any UDP test between VM hosted on each compute node, on different virtual networks.
  The MTU has been forced to 1400 on all VMs.

  Data path: VM1->TOR1->Spine->TOR2->router->TOR2->VM2

  The UDP test with any packet size <= 1428 passes always. Over 1428 bytes it always fails.
  Can be reproduced manually or in an automated way using the VMTP test tool.

   In an alternate configuration, where the 2 VMs reside on the same compute node (still different L3 network) and same rack as the Network node, the UDP traffic works fine regardless of packet size.
  VM1->TOR->router->TOR->VM2

  This seems to indicate the packet drop is happening only when the VLAN
  tag/VNI mapping is used.

  Versioning:
  https://github.com/cisco-openstack/neutron.git (staging/junoplus branch)
  openstack and devstack: stable/juno

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1412900/+subscriptions