You want to gracefully take a device out of service to perform an OS upgrade.

In Apstra 5.1, the correct operational method to gracefully remove a switch from service for maintenance is to set its Deploy Mode to Drain. Drain is a day-2 operational control that tells Apstra to adjust intent so the fabric can continue operating while the targeted device is logically taken out of service as much as the design allows. This is especially relevant in EVPN-VXLAN leaf-spine fabrics where taking down a spine or a leaf can disrupt underlay BGP adjacencies and overlay reachability if traffic is not shifted first.

This action is performed from the blueprint’s Active view because Drain affects the currently deployed, running fabric state (not a staged design change). Selecting the device under Active and changing Deploy Mode to Drain initiates the workflow that prepares the device for maintenance by reducing its role in forwarding and/or withdrawing dependent services according to the blueprint’s modeled redundancy (for example, shifting server-facing traffic to an MLAG/ESI peer where applicable, or reducing reliance on the device for transit). After the device is drained, an OS upgrade can be performed with less impact, and the device can later be returned to service by switching Deploy Mode back to Deploy and committing the change.

The “Upgrade” action is not the deploy-mode mechanism described for graceful removal; the key is Deploy Mode → Drain from Active, which is explicitly intended for maintenance and decommissioning scenarios.

Verified Juniper sources (URLs):

https://www.juniper.net/documentation/us/en/software/apstra5.1/apstra-user-guide/topics/task/device-drain.html

https://www.juniper.net/documentation/us/en/software/apstra4.2/apstra-drain-mode/apstra-drain-mode-guide/topics/concept/apstra-drain-mode-activate-or-disable-drain.html