[
https://issues.apache.org/jira/browse/HTTPCLIENT-2404?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18033675#comment-18033675
]
Oleg Kalnichevski commented on HTTPCLIENT-2404:
-----------------------------------------------
[~jaypi75] Just a few comments in random order:
There are 5 distinct configuration levels: the transport level (applies to
specific i/o models), the connection management level, the TLS level, the
protocol level and the request execution level.
The socket timeout presently can be set at 4 levels, each layer being able to
override the setting of lower levels.
1. Transport level: `SocketConfig` (applies to the classic transport);
`IOReactorConfig` (applies to the async transport)
2. Connection management level: `ConnectionConfig`
3. TLS level: `TLSConfig`
4. Request execution: `RequestConfig`
All other timeouts apply to the connection management only.
ConnectionEndpoint exposes the socket timeout of the underlying connection at
runtime but it is not a distinct layer.
This is it. There is nothing more to the timeout configuration.
Now I actually think we need a proper configuration guide, not timeout
documentation. Timeouts are just a small subset of configurable settings with
the socket timeout only being a bit special.
Oleg
> Provide comprehensive timeout configuration documentation with interaction
> matrix and override behavior
> -------------------------------------------------------------------------------------------------------
>
> Key: HTTPCLIENT-2404
> URL: https://issues.apache.org/jira/browse/HTTPCLIENT-2404
> Project: HttpComponents HttpClient
> Issue Type: Wish
> Components: HttpClient (async), HttpClient (classic)
> Affects Versions: 5.5, 5.5.1
> Reporter: Jens Popp
> Priority: Major
> Labels: volunteers-wanted
> Fix For: Future
>
>
> Apache HTTP Client currently exposes numerous timeout settings across
> different configuration points (connection manager, request config, socket
> config, etc.). Right now they are documented (as far as I was able to find)
> only on the point, where they can be set. The lack of comprehensive
> documentation about their interactions, precedence rules, and override
> behavior makes it difficult to configure clients correctly—especially in
> distributed systems where timeout handling is critical for failover and
> performance.
> *Problem Statement:*
> As demonstrated my comment in HTTPCLIENT-2386, seemingly minor implementation
> changes can have severe production impacts. In our case, upgrading from 5.5
> to 5.5.1 changed timeout behavior during TLS handshake scenarios. When
> servers under heavy load (100% CPU) could establish socket connections but
> failed TLS handshakes, the increased timeout (defaulted back to request
> timeout instead of connection timeout) prevented proper failover to backup
> servers, causing service degradation.
> *Requested Documentation:*
> # *Complete timeout inventory* - All configurable timeouts with their
> default values and scope
> # *Interaction matrix* - How timeouts interact when multiple apply to the
> same operation
> # *Override rules* - Precedence order when timeouts are set at different
> levels (global, connection manager, request)
> # *Lifecycle coverage* - Which timeout applies to each phase (DNS, connect,
> TLS handshake, data transfer)
> # *System best practices* - Recommended patterns for different scenarios
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
