when redisReaderFeed big buffer, redisReaderGetReply will cause high cpu usage

[hongliuliao]

when redis reader feed 40m buffer, the get reply will cause high cpu usage like as follow:

Because in redisReaderGetReply method, after a reply parsed, it will use "sdsrange" to move mem and when the buffer is big, the memmove() will cost many time & cpu!

[mnunberg]

Perhaps we should make this setting configurable. I am assuming most of our users are using buffers in the size of several K than in the size of several M?

…

On Apr 8, 2019, at 5:18 AM, harleyliao ***@***.***> wrote: when redis reader feed 40m buffer, the get reply will cause many cpu like as follow: <https://user-images.githubusercontent.com/1174006/55712758-ea219200-5a21-11e9-976d-0c051be35852.png> Because in redisReaderGetReply method, after a reply parsed, it will use sdsrange to move mem and when the buffer is big, the memmove will cause many time & cpu! <https://user-images.githubusercontent.com/1174006/55712839-09b8ba80-5a22-11e9-9bd6-4c6a8dc8299e.png> — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#656>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAanUXoSAjaOjXCJR9BupyraZ6dEjbfhks5vewligaJpZM4chsTJ>.

[hongliuliao]

Thanks for your reply! I think "this setting configurable" may better. I use redisReaderFeed & redisReaderGetReply to read a big aof file and parse, and then i think when the read buffer is bigger, the performance will better.

[michael-grunder]

It would be a non-trivial change, but we could look into using a ring buffer.

[michael-grunder]

Closing in favor of #866