1

I am trying to test out active/active LACP bonding between CentOS 6.4 and Arista. One of the tests I'm doing is to shut down an interface on the host and see what happens.

On the switch:

  • Link failure counter is incremented
  • No more LACP pdu's come in. (causing the port-channel to go down).
  • Interface status still shows connected.

From the host I have tried (with the same results) to shut the port like so:

  • ifdown em3
  • ip link set em3 down

I haven't tried ifconfig yet, and do not have access to try it at the moment.

The end result is that an "ifdown" on the interface causes the host to become unavailable on the network for about 20 seconds. On the other hand, if I shutdown the port from the switch, downtime is under 1 second.

Details:

# cat /etc/redhat-release
CentOS release 6.4 (Final)

# uname -a
Linux hostnameRemoved 2.6.32-358.23.2.el6.x86_64 #1 SMP Wed Oct 16 18:37:12 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux

Dell R720xd (latest firmware including the nic)

# lspci |grep Broadcom | head -1
01:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM5720 Gigabit Ethernet PCIe


# ethtool -i em3
driver: tg3
version: 3.124
firmware-version: FFV7.10.18 bc 5720-v1.34
bus-info: 0000:02:00.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: no
Improve this question
3
  • 1
    Why is your system so far out of date?
    – Michael Hampton
    Oct 20, 2014 at 22:36
  • 1
    (pull the plug?)
    – ewwhite
    Oct 20, 2014 at 23:13
  • 1
    I knew someone would jump on the cent version without providing any useful input. Its a sore topic, but the long story short is that it stays at 6.4 for now. If you can say yes its bug resolved in 6.x, thats another story.
    – ArtimusCF
    Oct 21, 2014 at 1:31

1 Answer 1

Reset to default
1

Don't test bonding link failure with commands, pull the cable.

If you read ifdown, it actually unenslaves an interface from a bond before setting it down. This does nothing to test bonding failover under the conditions you want to survive a failover, this just tests the bonding driver's ability to have its active slave changed.

You're lucky enough that your NICs consider PHY to be down when the switchport is shut, hence the working fast failover when you shut the switchport.

Not all NICs work like this, some actually need their electrical connection to the switch to be broken, so you can shut the switchport but the NIC doesn't consider the link as failed enough to fail over a bond.

Bonding with miimon is high availability against physical link failure. The way to test against physical link failure is to physically fail the link, no other way.

Pull the cable.

Improve this answer
2
  • 1
    This is the right advice. Test the failure the right way.
    – ewwhite
    Nov 1, 2014 at 22:39
  • Yes I agree. Being that the server is on the other side of the country and remote hands is $225/hr, im just going to have faith that it will work. Worst case scenario, 20 seconds for a failover is still much better than human intervention.
    – ArtimusCF
    Nov 3, 2014 at 0:13

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .