Arista MLAG ISSU upgrade - primary/secondary switch?

Question

When upgrading Arista switches which are part of an MLAG cluster, is it necessary to start with the secondary switch?

In "Arista Warrior", from Gary A. Donahue, he specifies on page 123 to start by upgrading the secondary switch. But there is nothing like that on the official documentation, page 710 (for version 4.13.2F).

Also, is there a way (even if it's not necessary) to force one of the members to become primary? I found the command primary-priority here, but the command does not exist on my switch.

Answer 1

By default MLAG chooses primary/secondary based on switch MAC address - lowest will always become the primary.

Ref: https://eos.arista.com/forum/mlag-priorities/

Why do you want to force the priority? MLAG was developed to achieve to active/active uplinks to different chassis.

I believe Arista Warrior might be stale - but to be honest I would be making use of your support contract with Arista. At the end of the day they will be able to give you the definitive answer.

Searching on the Arista site brings up nothing regarding explicitly having to upgrade the secondary first. End hosts see the Sys-ID in MLAG so convergence should be sub-second.

Answer 2

From the 14.4.6M guide:

10.3.3.4 Performing an MLAG ISSU Upgrade The following procedure performs an MLAG ISSU upgrade:

Step 1 Verfy configuration consistency on each peer (Section 10.3.3.1).

Step 2 Verify version compatibility between the new and existing images (Section 10.3.3.2).

Step 3 Configure reload-delay mlag (Section 10.3.2). Recommended delay period varies by switch type.

Step 4 Install the new image onto one of the peers:

• Step a Upload the new image to the switch (Section 2.4.1.2).
• Step b Set the boot path to the new image (Section 2.4.1.3).
• Step c Enter the reload command. (Section 2.4.1.4).

Step 5 Resolve all reload warnings (Section 10.3.3.3).

Step 6 Confirm the reload.

Step 7 Wait for MLAG peers to renegotiate to the active state and reload-delay expiry on rebooted peer. Avoid configuration changes on both peers after this step.

Step 8 Repeat the upgrade process for the other peer. When upgrading modular switches with dual supervisors, upgrade the standby supervisors first, then upgrade the active supervisors.

Answer 3

The Primary-priority will not cause the switch to pre-emptively take the primary role back from the other switch. This is why the command is hidden. MLAG peers are just that, one is not a master.

So, what that means is when MLAG is negotiated and you have modified the priority, the switch with the superior value will become Primary. However, If the Primary were to go down, be reloaded or mlag is shutdown, the secondary will become primary. Once the original primary switch comes back, it does NOT and will not take over. It remains Secondary until another event occurs to cause it to take the Primary role. Therefore the role(s) are non-deterministic and you must interrogate the switches to know which one is secondary before starting your MLAG ISSU process.

Also note, the latest (4.18.x) EOS manual is on the arista.com website and is not behind a login. It is true that the MLAG ISSU procedure does not explicitly say to start with the Secondary MLAG Peer but it is recommended in my experience with Arista SEs.

https://www.arista.com/en/um-eos/eos-section-12-3-mlag-maintenance?searchword=mlag%20maintenance

Answer 4

When upgrading Arista switches which are part of an MLAG cluster, is it necessary to start with the secondary switch?

No, there's no difference between starting an upgrade with a secondary switch or a primary switch.

Also, is there a way (even if it's not necessary) to force one of the members to become primary? I found the command primary-priority here, but the command does not exist on my switch.

The primary-priority command is a hidden command that will still take effect even though it looks like it doesn't exist on your switch.

You just have to type out the whole command and press Enter. You can confirm that the command has taken effect by performing a show active (as you can see in the picture)