For the client side
--------------------------
The policies will be decided at the stub generation time only (stub will be generated using the wsdl of the service which has the policies). The way this information is available for the module is only as a property of the message context . The thing is, client will be talking to different endpoints or different opeartions of the same endpoint. In both these cases Policies will vary from Message Context to Message Context (ok may be its from op. context to op. context but any way the idea is the same - it varies). So it should be a property of some context and the Sender should use this information to decide the things like retransmission of each message.
For the server side
--------------------------
Only for consistency. Each message context will be filled with the same policy infromation if they are from the same service and the same operation. (So what u say is right for the server side :) )
This way the RM Sender code can be generalized. Sender will work at the message context level (just like Sandesha1). It will use the policy information of each Message Context to adjust the next retransmission time etc.
thanx,
Chamikara
On 11/11/05, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
Hi Chamikara,
When we design Axis2 the model that we discussed to handle policies was as
follows.
Axis2 handle the policies and each module can access the policy elements,
which will be there in some context (service context)
Please clarify why do we need to pass them around, if we can access them
from anywhere.
Thanks,
Jaliya
----- Original Message -----
From: Chamikara Jayalath
To: [email protected]
Sent: Thursday, November 10, 2005 8:30 PM
Subject: [Sandesha2] The Policy model
Hi All,
I'm working on implementing the policy model for sandesha2. Sanka has
already created a RMPolicyBean and a PolicyUtil class which can be used
to extract a RMPolicyBean out of a large Policy OM Element. The problem is
how to pass this extracted information around the Sandesha2 framework.
This is what I have in mind.
Each messageContext should have an associated RMPolicyBean which contains
its policy information. In the client side this information will be set by
the user (probably wrapped by the stub). In the server side, a fixed set
of policy values (given in the module.xml) will be loaded to the
RMPolicyBean. This infoamtion will be later set to each messageContext.
Here are how these policies can be used.
Rretransmission interval, exponention backoff
-----------------------------------------------------------------
The RetransmitterBean will adjust the nextTimeToSend after each
retransmission. To calculate the nextTimeToSend, the sender will use these
two policy values (with the sentCount property which gives the no. of
times the message has been retransmitted) which can be obtained from the
associated RMPolicyBean of the current message Context.
Acknowledgement interval
-------------------------------------
When adding a ack message the timeToSend will alsays be set to
currentTime+Acknowledgement Interval so that the ack message will not be
send before the interval get expired. When the sender detect an
applicationMessage it will look for an ack of the same sequence. If it
find one, the ack will be piggybacked. otherwise the ack will be sent as a
standalone message.
Inactivity timeout
------------------------
Each sequence has an associated lastRespondedTime property which gives the
last time the other end responsed. Sender thread in it's each loop will
look at the the lastRespondedTime of each sequence, if it find tha
currentTime>lastRespondedTime+inactivityTimeout it will end that sequence.
Comments ???
Thanx,
Chamikara
