why pool_max_idle_per_host(0) can solve cpu usage

[chenqimiao]

let download_client = reqwest::Client::builder()
    .user_agent(UA)
    .default_headers(headers)
    // OSS closes idle connections after 0 seconds,
    .pool_idle_timeout(Duration::from_secs(0))
    .pool_max_idle_per_host(0)
    .connect_timeout(Duration::from_secs(10))
    .timeout(Duration::from_secs(60))
    .build()?;

I'm encountering an issue with the pool_max_idle_per_host(0) parameter. I understand this controls connection closure behavior.

##Problem: When I omit this line, high-intensity usage of download_client results in sustained high CPU usage even after all requests complete.

My investigation:

I suspected lingering TCP connections might be the cause, but when I checked using:
sudo lsof -nP -iTCP -sTCP:ESTABLISHED | grep 'quarkdr'
I observed that all TCP processes were actually released.

Question:

If TCP connections are confirmed released, what else could be causing the persistent CPU usage?

Looking forward to your answer, Thanks !

[seanmonstar]

Thanks for the report! I tested it out locally with some logging, and found that the pool's idle checker was basically triggering constantly because it would wait in a loop for 0 seconds and check for idle connections. I've filed hyperium/hyper-util#215 as the immediate fix, but I'll look into refactoring it a bit more to improve other ways too.

[chenqimiao]

Thanks for your reply. Looking forward to the optimized version.