Mention `ReadBufferEachConn` and `WriteBufferEachConn` on README

[zhaozekunssss]

Hello,

I try to use the ping command regularly to check if the rueidis client is disconnected or reconnected, but I found that if there are a lot of redis servers, ping will cause the memory to keep rising. Through the go pprof tool, I found that bufio.NewWriteSize and bufio.NewReadSize occupy a lot of memory. What is the reason for this?

I know rueidis has already implemented an automatic detection method for reconnection internally, and if the ping command for periodic detection is incorrect, how should we know if redis client is disconnected for the application layer?

[rueian]

Hi @zhaozekunssss, a PING request will be send to one random redis server and cause one connection be established. One connection will have one reader buffer and one writer buffer which are both 0.5MiB by default.

I guess that’s why you saw your memory usage kept raising, but the raising should stop once all server connections were established.

You can change the default sizes by setting ReadBufferEachConn and WriteBufferEachConn:

rueidis/rueidis.go

Lines 161 to 164 in 37dfd47

	// ReadBufferEachConn is the size of the bufio.NewReaderSize for each connection, default to DefaultReadBuffer (0.5 MiB).
	ReadBufferEachConn int
	// WriteBufferEachConn is the size of the bufio.NewWriterSize for each connection, default to DefaultWriteBuffer (0.5 MiB).
	WriteBufferEachConn int

how should we know if redis client is disconnected for the application layer?

Applications can only notice disconnections when they get network errors from trying to interact with connections. So, doing periodic PINGs is correct, but rueidis already does that for you. If you want to do something else when there are connection issues, you can also monitor each redis response to see if it is a network error.

[zhaozekunssss]

Hi @rueian , thanks for your reply, I have modified the values of ReadBufferEachConn and WriteBufferEachConn, and the memory usage has been reduced to a very low value：）
However, I would like to know why the default values of these two parameters need to be so large, and what their specific functions are? Will this memory be cleared?

[rueian]

Those buffers help the client avoid unnecessary read and write system calls to the TCP connections when handling a large amount of concurrent requests and large pipelines. Therefore, the default values are relatively big. Those buffers will only be cleared after disconnections.

[rueian]

A follow-up to this issue could be adding a description of ReadBufferEachConn and WriteBufferEachConn to the https://github.com/redis/rueidis?tab=readme-ov-file#memory-consumption-consideration section.

[zhaozekunssss]

Hi @rueian, another question, since we deployed in a memory sensitive environment, I set the ClientOption.RiscaleEachConn to 8 and found that newRing() still occupies some memory, about 8M.
How about setting it to 6? Will it cause any other problems? We are willing to reduce some throughput.

[rueian]

Hi @zhaozekunssss, it will only affect the upper bound of the throughput. Setting to 5 or lower is ok if you are willing to reduce some throughput.