On 30/10/2014 08:46, Rémy Maucherat wrote:
> 2014-10-30 9:42 GMT+01:00 <r...@apache.org>:
> 
>> Author: remm
>> Date: Thu Oct 30 08:42:17 2014
>> New Revision: 1635431
>>
>> URL: http://svn.apache.org/r1635431
>> Log:
>> Disable chunking when there's no keepalive (for example when the request
>> specified connection close, which is detected).
>>
> IMO although chunking can be disabled and can be considered more costly,
> it removes the possibility to detect truncated responses (or requests, but
> it's up to the client to use it or not). So using a system property to
> disable this could be worthwhile.
Was it just the cost of the chunking that triggered this change or was
it something else? If it was performance, how much difference does this
change make?

Looking at the code, I can see a couple of places where this changes
behaviour:
- If the client requests the connection is closed
- If the request is the last one before the connection is closed due to
  reaching the maximum number of keep-alive requests

We have had bug reports in the past about not being able to detect a
truncated response. I suspect the current unit tests for those miss the
second case above which is my main concern right now.

I don't think a system property is the right way to handle configuration
of this feature (if such configuration is required). There is no reason
it could not be a connector attribute.

Mark


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org

Reply via email to