Regression in v9 due to idle connection behavior change

[bduffany]

With go-redis v9, idle connections are now closed lazily, instead of proactively in the background like they were in v8. As a result, we're hitting redis client connection limits that we weren't hitting before, because we now have tons of idle connections (thousands of idle connections for each of our server instances that are connected to redis).

Some questions:

1. What was the reasoning for this change to close idle conns lazily instead of eagerly?
2. Is it possible to have ConnMaxIdleTime close the idle conns eagerly instead of lazily? The docs on MaxIdleConns say "The idle connections are not closed by default." but "by default" seems to imply there should be a way to get non-lazy behavior?

[ndyakov]

Hello @bduffany and thank you for opening this discussion. I cannot share about the reasoning, will dig deeper in the git history and get back to you if I find something.

[haunt98]

Is this PR #3418 cause this?
I also encounter this regression when I deploy version v9.16.0 recently.

I use metric pool_conn_idle_current from https://github.com/redis/go-redis/blob/master/extra/redisprometheus/README.md

Before:

After:

[ndyakov]

@haunt98 the original issue is for something else, feel free to open a new issue with detailed description what you are experiencing.Please include information about the "before" version as well