Summit Stack hitless upgrade

  • 0
  • 1
  • Question
  • Updated 3 years ago
  • Answered
Hi guys,
I have a stack with two x480 running XOS 15.3.1.4
Is it possible to upgrade the two switches in the stack one after another so that one of the switches is allways forwarding traffic?
Thanks,
Thomas
Photo of Thomas Hilber

Thomas Hilber

  • 168 Points 100 badge 2x thumb

Posted 4 years ago

  • 0
  • 1
Photo of PARTHIBAN CHINNAYA

PARTHIBAN CHINNAYA, Alum

  • 4,382 Points 4k badge 2x thumb
Hi Thomas,
This is not possible.
Also with hitless extreme may not support
Sync of bgp, ospf routes.
From which exos version you are upgrading code ?

As it is x480 I assume you must be using bgp/ospf ... There is lot of code changes between 12.6 to 15.x
Photo of Paul Russo

Paul Russo, Alum

  • 9,694 Points 5k badge 2x thumb
Hello Thomas

We do not support hitless upgrade using ISSU within the stack.  I have come up with a partially manual way to perform a hitless upgrade to the stack.  This assumes that all devices that are connected to the stack are connected to each of the nodes.

If you are interested in how you can do this please see the document at this link https://www.dropbox.com/sh/f66b3kfoii2t573/AAAWE5Idb4W01b-SEWnrgda9a?dl=0


I hope this provides a solution for you.

P
Photo of Thomas Hilber

Thomas Hilber

  • 168 Points 100 badge 2x thumb
Current release is XOS 15.3.1.4 patch 10 and we have to upgrade to 15.3.1.4 patch 41.
yes, we are using ospf.
Photo of Thomas Hilber

Thomas Hilber

  • 168 Points 100 badge 2x thumb
Thank you, Paul. We will give this a try!
Photo of Paul Russo

Paul Russo, Alum

  • 9,694 Points 5k badge 2x thumb
Welcome Thomas  This document was created for a customer that had a DC in an EAPS ring. If you don't have an EAPS ring then you may not need to disable the port as indicated in the document.

Just download the code to the opposite image container. Tell the stack to use the new code then follow the process for powering done the node and disconnecting the stacking cable. The whole process should only take approx 5 min.

Good luck and let us know how it works for you.

P
Photo of Grosjean, Stephane

Grosjean, Stephane, Employee

  • 13,676 Points 10k badge 2x thumb
Hi Thomas,

If you stay on a patch level upgrade (15.3.1p1-10 to 15.3.1p1-41), you should be able to upgrade one stack member at a time. That's not ISSU, but more like Rolling Upgrade. Put the image on one member, reboot it. When it's back and operational, do the same for the other. So really, slot by slot (each command should offer you the slot option).
Photo of Ted

Ted

  • 1,174 Points 1k badge 2x thumb
I'm on version 15.3.1.4 patch1-36 and moving to 16.1.2.14 patch1-1.  I want to upgrade a stack one at a time but I'm not clear on this process.  Please let me know if I understand correctly. I can install the new image to the non active partitions but reboot to the new install, one stack at a time?  Meaning the switches will be running at the same time on two different images ( I just have a 2 stack with both stack ports connected).  Or will I need to un-stack them to keep them connected.  This situation I'm not using MLAG on the upstream, but I have others that are using sharing to a MLAG connection on my cores.  I'll  need to work that scenario out too.

Thank you very much.
Photo of Drew C.

Drew C., Community Manager

  • 40,316 Points 20k badge 2x thumb
During the upgrade installation, network traffic will continue as usual.  The only time your traffic should need to "failover" is when you decide to reboot the stack to complete the upgrade.
Photo of Ted

Ted

  • 1,174 Points 1k badge 2x thumb
So there's no scenario where one switch can stay active while the other one reboots and installs the new image?  I've worked with Extreme for quite sometime but we are starting to use stacking more often.  We stack to keep our uplinks or downlinks active when a switch in the stack fails traffic isn't re-routed.  Its seems like I should be able to unstack one while we upgrade then re stack it.  Might be too much trouble though.  The ideal situation is a hitless upgrade.  Otherwise I've got to notify members who terminate to this stack that there will in fact be downtime.
Photo of Drew C.

Drew C., Community Manager

  • 40,316 Points 20k badge 2x thumb
Technically, you can reboot one node at a time, but as each ones comes back up, it won't be able to communicate with the master node and you'd end up with a mess.
"Breaking" the stack would cause you to lose your configuration.  Your safest bet is to schedule a reboot of the stack.
Photo of Drew C.

Drew C., Community Manager

  • 40,316 Points 20k badge 2x thumb
Further, the hitless/rolling upgrade described above would not be supported with the large jump in software version that you're planning.
Photo of Ted

Ted

  • 1,174 Points 1k badge 2x thumb
I appreciate all the info.  I'll schedule the downtime.

Thank you,
Ted