We operate an EVPN/VXLAN fabric. One of our switches lost BGP peering following a code upgrade on neighboring switches.

The switch then shutdown all ESI-LAG interfaces while the physical interfaces maintained up. Why did the AE shutdown? Is this expected behavior?

This behavior happens due to feature in Juniper’s EVPN implementation known as Core Isolation. When a device in an EVPN network loses all of its EVPN BGP peering sessions, the core isolation feature, working in conjunction with LACP, automatically brings down all ESI-LAG interfaces on the device.

This is designed to prevent traffic loss during route convergence when a PE device becomes isolated from the EVPN core. The shutdown of the ESI-LAG interfaces triggers the connected devices to reroute their traffic, avoiding the isolated device. This behavior might not always be desirable as it could lead to traffic halt and in some specific cases, it's recommended to have it disabled.

The following link explains how Core Isolation feature works:

https://www.juniper.net/documentation/us/en/software/junos/evpn-vxlan/topics/concept/evpn-vxlan-core-isolation-disabling.html