Issue with enterprise proxy API and using standard client class

[tylerhutcherson]

LangGraph today expects either a Redis or RedisCluster client. This is due to the differences in how clustered Redis works for OSS/dev and cloud like AMR. Enterprise proxy allows for chatter over a single/standard endpoint and thus no RedisCluster client class is needed. Fix the internal discovery logic to handle this case. The expected workaround is the set cluster_mode = True on the checkpoint saver class manually for now.

[eses-wk]

I am using an Azure Cache for Redis (with RedisSearch and RedisJSON module on) and passed the configured client to the RedisSaver class. checkpoint.setup() works fine, but when we do agent.invoke(), there will be a cluster cross slot error.. I suspect it is a similar issue as described in this issue.

ClusterCrossSlotError: Command # 2 (EXPIRE...) of pipeline caused error: Keys in request don't hash to the same slot (context='within MULTI', command='expire', original-slot='5958', wrong-slot='8725', first-key='checkpoint:XXXXX:__empty__:00000000-0000-0000-0000-000000000000', violating-key='checkpoint_blob:XXXXX:__empty__:__start__:00000000000000000000000000000001.0.6xxxxxxxxx')