[
https://issues.apache.org/jira/browse/CXF-9109?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17930637#comment-17930637
]
Radek Kraus commented on CXF-9109:
----------------------------------
IMHO this problem was caused by CXF-8678, where per-HTTPClientPolicy caching
was introduced ([https://github.com/apache/cxf/pull/923]). AFAIU
"per-HTTPClientPolicy caching" means that each instance of {{AsyncHTTPConduit}}
has own instance of {{AsyncClient}} (own instance of
{{PoolingAsyncClientConnectionManager}} included), regardless on
{{{}TlsStrategy{}}}. I am sorry, but I don't understand why
{{HTTPClientPolicy}} is used as map key (even without redefined
{{{}equals(){}}}/{{{}hashCode(){}}} methods).
In addition, there is a comment ({{{}AsyncHTTPConduitFactory{}}}), {_}... the
context propagation is necessary to remove per-HTTPClientPolicy caching{_}, but
HTTPCLIENT-2209 has already introduced {{HttpContext}} propagation.
Does it mean, that now, per-HTTPClientPolicy caching can be somehow removed
(because HTTPCLIENT-2209 is done)?
Is there some plan to remove per-HTTPClientPolicy?
Or I am completelly wrong?
When I tried to use {{cxf-rt-transports-http-hc}} (instead of
{{{}cxf-rt-transports-http-hc5{}}}), it seems, that attached test/example works
like is expected.
Thank you for any answer.
> MAX_PER_HOST_CONNECTIONS does not work
> --------------------------------------
>
> Key: CXF-9109
> URL: https://issues.apache.org/jira/browse/CXF-9109
> Project: CXF
> Issue Type: Bug
> Components: JAX-RS
> Affects Versions: 4.1.0
> Reporter: shpelda
> Priority: Major
> Attachments: ConnectionPoolingTest.java
>
>
> After migration from 3.4.4 to 4.1.0, one of my tests started crashing.
> In v3, AsyncHTTPConduitFactory used shared ConnectionManager, so that
> bus-level properties
> _org.apache.cxf.transport.http.async.MAX_CONNECTIONS_ and
> _org.apache.cxf.transport.http.async.MAX_PER_HOST_CONNECTIONS_
> worked across clients.
> This properties were the reason why I used the AsyncHTTPConduitFactory, as i
> had to limit count of created connections.
>
> v3 implementation of AsyncHTTPConduitFactory used shared instance of client
> and connectionManager, which finally caused above properties to work properly.
> hc5.AsyncHTTPConduitFactory(correctly) maintains cache of
> httpclientPolicy->httpClient.
> However _PoolingAsyncClientConnectionManager_ must not created with each
> client, but must be shared. Should be owned by the {_}Bus{_}.
> Otherwise connection pool cannot work correctly.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
