We tried both ways.
First with stopping the failover on the secondary node which is obviously the correct way.
And the next time without stopping it to check how the two servers behave.