SentinelConnectionPool does not retrieve updated master address from Sentinel on connection retries

[SamuelGD]

Description:
When the Redis master address changes (e.g., after a Sentinel failover), the Redis client using a SentinelConnectionPool does not fetch the updated master address during connection retries.

Currently, only the connect method in SentinelManagedConnection retrieves the master address from Sentinel https://github.com/redis/redis-py/blob/v7.1.0/redis/asyncio/sentinel.py#L53.

However, on retries, connect_check_health call the internal _connect method, which does not update the master address: https://github.com/redis/redis-py/blob/v7.1.0/redis/asyncio/connection.py#L306-L311.

As a result, the client may continue attempting to connect to the old master.

Expected behavior:
Retries should retrieve the latest master address from Sentinel, ensuring the client reconnects to the correct master after a failover.

[petyaslavova]

Hi @SamuelGD, thank you for reporting this! We will have a look at it.

[SamuelGD]

Thank you @petyaslavova, I also opened a PR for this #3875

[SamuelGD]

The bug may have been introduced in this PR https://github.com/redis/redis-py/pull/3557/files.

In send_packed_command, self.connect was swapped with self.connect_check_health but SentinelConnectionPool only overrides connect.