Questions about the acquire of boundAsyncPool

[soafftt]

Project is using boundAsyncPool for Redis Pipeline functionality.

There are two AWS Elasticache Redis Slaves, and we are using DirDnsResolver to disable DnsCache as much as possible.

It seems like a traffic that doesn't require the use of a pool, but I'm using a pipeline because there are many concurrent commands.

There is a phenomenon in which the acqure of boundAsyncPool provides connection in the FIFO, so it is crowded to one of the two Slaves.

Can you give me some advice to solve these issues?

[tishun]

@soafftt , please allow the team some time to get to this question.

[tishun]

It seems like a traffic that doesn't require the use of a pool, but I'm using a pipeline because there are many concurrent commands.

Concurrency itself is not a good reason to use connection pooling. Typically you would use connection pooling when there are blocking commands (e.g. BLPOP) or you are using transactions.

There is a phenomenon in which the acqure of boundAsyncPool provides connection in the FIFO, so it is crowded to one of the two Slaves.

I assume you want to split the load evenly between the two AWS Elasticache Redis Slaves?

The choice of node is made based on the ReadFrom setting. There is currently no load-balancing mechanism, but you can specify random nodes to be used, see #834.

Does that help?

[soafftt]

In our case, we do not use MasterReplica. because we use each RedisClient to strictly distinguish between Master and Replica when using pipeline.

(only using Master/Replica in cases where pipeline is not used.)

it seems that the distribution of elasticache's reader endpoint is not accurate due to aws dns cache. I will ask AWS about this.

Thanks for the advice.

[tishun]

Happy to help. Please keep us posted in case this turns out to be some other problem.