loub wrote:
Hi -,
We are using Apache HttpClient 3.1, with the Multi Threaded Http Connection
Manager to send >1000 HTTP POST messages per second, to 4 test servers (in
our test environment) using trusted persistent connections. This setup has
been soaked for over 72 hours and we found it to function as expected.
The Connection Manager is configured with the following:
closeIdleConnections = 180 seconds
MaxConnectionsPerHost= 1000
MaxConnections= 2100
ConnectionTimeout= 500 milli seconds
SoTimeout=2000 milli seconds
The HTTP POST messages are distributed to each test server using a simple
round robin algorithm. In reasonably steady state conditions, the client
establishes about 100 separate connections to each test server.
If while repeating the above test, one of the 4 test servers is taken
offline for say one second; we observe that all of the connections managed
by the Connection Manager tend to fail within just a few seconds.
I personally find this somewhat hard to believe and see no reason why
connections to remaining servers should fail.
If you are absolutely sure all of the connections managed by HttpClient
fail, and no suspiciously looking little green men have been spotted
messing with your computer, please provide a wire log of the HTTP
session exhibiting the problem.
Please also consider upgrading to HttpClient 4.0, as HttpClient 3.1 is
nearing end of life and is effectively unsupported.
3.1:
http://hc.apache.org/httpclient-3.x/logging.html
4.0:
http://hc.apache.org/httpcomponents-client/logging.html
Oleg
At 1000
messages per second, I appreciate that flow control is an issue after just 2
seconds but this is not our case.
>
Our understanding, is that the messages being sent to the 3 working test
servers should continue to receive messages irrespective of what is
happening to the fourth.
Any suggestions?
Regards
Lou
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
