We fail to send the READONLY command to replicas when running commands inside a pipeline, which results in an increased latency as the server will respond with an unneeded MOVED message forcing us to retry the command.
We move the sending of the READONLY command to the connection creation itself, since at that point we already know if we're connecting to a replica or not, so we can ensure consistency on sending this command.
We fail to send the
READONLYcommand to replicas when running commands inside a pipeline, which results in an increased latency as the server will respond with an unneededMOVEDmessage forcing us to retry the command.We move the sending of the
READONLYcommand to the connection creation itself, since at that point we already know if we're connecting to a replica or not, so we can ensure consistency on sending this command.Fixes https://github.com/Grokzen/redis-py-cluster/issues/470