hc-httpclient-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Pellerin, Clement" <Clement_Pelle...@ibi.com>
Subject RE: Controlling releaseConnection
Date Tue, 18 Oct 2016 14:12:32 GMT
I agree that code is correct. I never said there was a bug.
I am asking how to override the behavior.
We are porting our product from HttpClient 4.1 to 4.5.2
and we need to preserve that feature because it is used by our customers.
It would be a shame to duplicate all of MainClientExec because
of a lack of modularity in that class.

We tried to subclass the PoolingHttpClientConnectionManager
but that does not work because the connection state is actually
stored in the ConnectionHolder created directly in MainClientExec.execute().
        final ConnectionHolder connHolder = new ConnectionHolder(this.log, this.connManager,
managedConn);

Basically, I'm asking if the maintainers could consider a method like
	protected boolean isEarlyRelease(HttpEntity entity) {
		return entity == null || !entity.isStreaming();
	}

Or a release strategy interface if this is a common requirement.
I am also interested in any other work-around.

-----Original Message-----
From: Shawn Heisey [mailto:apache@elyograg.org] 
Sent: Monday, October 17, 2016 9:24 PM
To: HttpClient User Discussion
Subject: Re: Controlling releaseConnection

On 10/17/2016 3:22 PM, Pellerin, Clement wrote:
> Our customer needs to delay the release of the connection until the response is fully
processed.
> They want to turn off the early automatic release of the connection and do it manually
later.
>
> This is the problematic code in MainClientExec
>             // check for entity, release connection if possible
>             final HttpEntity entity = response.getEntity();
>             if (entity == null || !entity.isStreaming()) {
>                 // connection not needed and (assumed to be) in re-usable state
>                 connHolder.releaseConnection();
>                 return new HttpResponseProxy(response, null);
>             } else {
>                 return new HttpResponseProxy(response, connHolder);
>             }

Mostly an end-user here, with no status to speak of in this project.  I do have status on
another Apache project that utilizes HttpClient, but I don't know much about that part of
the code.  I have written some HttpClient code for a completely unrelated project of my own,
but that code is VERY simple.

When I read the code above, what I see is this: It only releases the connection if the entity
is nonexistent (null) or the entity is NOT a type that uses streaming.

I will fully admit that my experience with HttpClient is limited, but I think the chance is
very small that the HttpComponents committers have made a mistake here.  I think this particular
code has probably been discussed and examined, then ultimately validated as correct.  Here's
why I think they didn't make a mistake:

If the entity object is null, then the response probably doesn't HAVE an entity (response
body), so it will be entirely self-contained, consisting of headers only, and the connection
doesn't have anything further to send.  If the entity exists but doesn't utilize streaming,
then I think it's likely that the entity was received in its entirety and has been incorporated
into the response object already, and once again, the connection isn't needed.  If my limited
understanding of non-streaming entities is correct, they have the potential to be very dangerous
from a memory consumption perspective, and my own usage of HttpClient (where I did not set
anything related to the entity type) suggests that streaming entities are used by default.

Restating in another way:  In the first situation that results in a released connection, there's
nothing to consume, you just need the response object that you already have.  In the second
situation, the entity you will consume is probably already available within the response object
and doesn't need the connection.  The comment on the release call in the code quoted above
implies that this is how things work.

In these situations, why do you need the connection to stick around?  I think it can't do
anything else that's useful for that request.  I would imagine that if the connection utilizes
keepalive/pipelining, that it will typically remain open after release and can be utilized
again for a different request.

Someone with more direct knowledge of HttpClient's internal implementation will need to confirm
whether or not I'm correct in what I've written.  My understanding could be wrong.

Thanks,
Shawn


---------------------------------------------------------------------
To unsubscribe, e-mail: httpclient-users-unsubscribe@hc.apache.org
For additional commands, e-mail: httpclient-users-help@hc.apache.org

Mime
View raw message