Thanks.
It is not important. All of them return the same result to you because sentinels are synced. But you should use a “round robin” connection to handle the fail scenarios. If you look out replication topic in below documentation(predis), you can see.

nrk/predis