On 01/13/2026 14:11, Vardin. Angel wrote:
Hello,
Thank you for the provided information. My implementation contains logic that
cleans expired connections every 10 seconds. To verify the correct timeout
value, I checked the default logic in the HttpClientBuilder, which cleans both
expired and idle connections. As far as I can see, this is implemented in the
IdleConnectionEvictor thread (see links below):
https://github.com/apache/httpcomponents-client/blob/master/httpclient5/src/main/java/org/apache/hc/client5/http/impl/IdleConnectionEvictor.java#L56
https://github.com/apache/httpcomponents-client/blob/master/httpclient5/src/main/java/org/apache/hc/client5/http/impl/classic/HttpClientBuilder.java#L1118
You mentioned in your response that the timeout for expited connections
cleaning should not be less than one minute. However, in the
IdleConnectionEvictor, the default timeout is 5 seconds if no value is
specified. Additionally, if evictIdleConnections is not configured, the timeout
defaults seems to be 1 second.
We had extreme cases with people trying to call the damn method every
few milliseconds. A safe-guard has been introduced to ensure the sleep
interval could not be less than a second. The patch had a bug in it that
I had overlooked. The default should have never been 1 sec.
In the normal course of operations when the pool is reasonably busy it
will evict expired connections as it encounters them. The idle evictor
thread is needed only if there _long_ periods of inactivity. One minute
sweep should be perfectly sufficient, Or 30 seconds. Anything below 5
seconds is outright crazy but we still allow it in case someone really
has reasons for that.
I will address the problem with IdleConnectionEvictor defaults, but
there unlikely to be changes to the logging granularity.
Oleg
Could you please provide more details on what the timeout value should be?
Best regards, Angel
________________________________
From: Oleg Kalnichevski <[email protected]>
Sent: Saturday, January 10, 2026 1:18 PM
To: HttpClient User Discussion <[email protected]>
Subject: Re: Request for enhancement of Apache HTTP Client debug logs
On Thu, 2026-01-08 at 15:48 +0000, Vardin. Angel wrote:
Hello,
I enabled the HttpComponents Client 5.5 logs to show DEBUG and TRACE
output for the package "org.apache.hc.client5.http.impl.io" in order
to get detailed HTTP/1.1 I/O information.
During testing, I noticed that when the log severity is set to DEBUG,
many log entries appear that would ideally belong at the TRACE level.
For example, the logs were flooded with the following entry, which
provides very limited information:
https://github.com/apache/httpcomponents-client/blob/master/httpclient5/src/main/java/org/apache/hc/client5/http/impl/io/PoolingHttpClientConnectionManager.java#L606
I also reviewed the codebase and could not find many TRACE-level log
statements.
Is there any plan to improve the logging granularity in the future?
No, there are no such plans. The verbosity and granularity of logging
is subjective and we cannot please everyone.
TRACE priority is not being used mostly because it is not consistently
supported by all logging frameworks and back-ends.
Regarding the example referenced above, this is actually a pretty bad
case because if your code calls #closeExpired method more often than
once a minute, something is very wrong.
Oleg
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
