Slave server don't use Master which one send NOTIFY
In https://kb.isc.org/docs/aa-01467，It say The servers are queried in turn - named moves on to the next server in the list. But is it unreasonable. Usually, Slave use more than two master for HA. If first master happend some error such as kernel panic, and second master send a NOTIFY to slave. now the slave will wait 15s(udptimeout time) + 15s(try again without EDNS) + tcp_timeout(about 120s). it may cause some error.
When Slave receive a NOTIFY, the Slave should choose master who send the NOTIFY rathen than queried the masters list.
Links / references