Redis broker pool size about centrifuge

[FZambia]

Hello, @anhnguyensgu!

There are 3 modes in current Redis broker implementation:

1. Standalone Redis
2. Redis with Sentinel
3. Redis Cluster

For cases 1 and 2 pool size does not really matter - we utilize Redis pipelining and only several active connections from Centrifuge to Redis are active at any point of time.

I think pool size only matters in Redis Cluster case - where we are using pool to issue most of commands except publish without history (which is pipelined).

I think in the near future Centrifuge Redis implementation may migrate from redigo to go-redis or possibly to rueidis library - and those allow using pipelining for communication with Redis Cluster too. Pool size should not be too important in that case I believe.

So I think we should not expose pool size at this point. Do you have issues with current default pool size (which I think may happen only in Redis Cluster scenario)? If yes - we can address it more precisely, possibly looking how go-redis from the mentioned pull request behaves - I think it should provide better overall throughput.

[anhnguyensgu]

Thanks for your support. At this time, I have no issues with the pool, I will let you know if having any.

[FZambia]

We migrated to rueidis library in #262 – where connection pool does not have much sense anymore, Centrifuge will use pipelining over several connections for all operations. So I suppose this issue can be closed.

[FZambia]

The draft of blog post about rueidis migration: centrifugal/centrifugal.dev#18 - @anhnguyensgu probably you will be interested to check it