My understanding is that with Azure, you get a single endpoint to target redis. As such, your redis client (tyk in this case) does not need to point to all the redis targets. This would indicate that Azure promotes the slave to master if the master fails automatically - and the client wouldn’t know any different.
If you are using a Hybrid Tyk installation, a Master Redis cluster is in our data-centre. As such the redis deployed within your “slaved DCs” can pretty much be ephemeral. Upon restart, the local redis cache will be populated with required session tokens as the tokens are used
Obviously, if you treat your redis as ephemeral - you would lose quota and rate-limit data when redis restarts and if is not configured for persistence.