Hi Felipe, glad to hear this is fixed. That's The HUB! 🙂
The ISC link is responsible to keep both MLAG peers working as just 1 logical box (by exchanging health-check packets) and must by active all the time.
"If the ISC link alone goes down when the remote peer is alive, both the MLAG peers forward the south- bound traffic, resulting in duplication of traffic. However, this does not result in traffic loops. This is because the remote node load shares to both the MLAG peers and does not forward the traffic received on one of the load shared member ports to other member ports of the same load shared group."
To minimize ISC fail you can use 2 or more links between MLAG peers (ISC link) and enable sharing (LAG with LACP).
Another option would be using an alternate ipaddress (available in EXOS 15.5 and above) for MLAG peer health-check, but according to the diagram provided, the physical path is the same.