READONLY command issued each time a command is performed against a slave node with read_from_replicas=True?

[rtkcthomson]

We are using AWS ElastiCache Redis configured in cluster mode with master and slave node pairs. Using a RedisCluster instance configured with read_from_replicas=True resulted in a noticeably higher CPU load on the slave nodes than on the master nodes while connection count, bytes transferred, GET command count and other metrics appeared to be fairly even between the two. In investigating further we believe that

redis-py-cluster/rediscluster/client.py

Line 626 in 58bafe4

connection.send_command('READONLY')

is resulting in a READONLY command being issued to the slave nodes prior to every read command that is performed. This uneven distribution of CPU load limits the scalability of the Redis cluster.

Could the READONLY command be performed less often or is it essential to be performed for every read command? For example could it be done once for a connection? Alternatively, could it be pipelined with the originally requested command instead of paying the overhead of having two requests to the slave nodes for each read operation?

[Grokzen]

There is a open PR #403 for this to reduce this to only be sent once per connection instead of once per command/pipeline but it has yet to be merged.

[rtkcthomson]

Thank you Grokzen, I'll watch the PR.