I agree with David. Payload of the control tuple is in the userObject
and operators/ports don't need to be exposed to the implementation of
the ControlTuple class. With the proposed interface operators developers
are free to extend ControlTuple further and I don't think that such
capability needs to be provided. The wrapping in the ControlTuple class
is necessary and most likely ControlTuple needs to be extended from the
buffer server Tuple. It may be good to have a common parent other than
Object for all user payloads, but it may be a marker interface as well.
Thank you,
Vlad
On 12/16/16 09:59, Bhupesh Chawda wrote:
Hi David,
Actually, I was thinking of another API class called ControlTuple,
different from the actual tuple class in buffer server or stram.
This could serve as a way for the Buffer server publisher to understand
that it is a control tuple and needs to be wrapped differently.
~ Bhupesh
On Dec 16, 2016 22:28, "David Yan" <[email protected]> wrote:
// DefaultInputPort
public void processControl(ControlTuple tuple)
{
// Default Implementation to avoid need to implement it in all
implementations
}
{code}
{code}
// DefaultOutputPort
public void emitControl(ControlTuple tuple)
{
}
I think we don't need to expose the ControlTuple class to the operator
developers because the window ID is just the current window ID when these
methods are called. How about making them just Object? We also need to
provide the way for the user to specify custom serializer for the control
tuple.
David
On Thu, Dec 15, 2016 at 12:43 AM, Bhupesh Chawda <[email protected]>
wrote:
Hi All,
Here are the initial interfaces:
{code}
// DefaultInputPort
public void processControl(ControlTuple tuple)
{
// Default Implementation to avoid need to implement it in all
implementations
}
{code}
{code}
// DefaultOutputPort
public void emitControl(ControlTuple tuple)
{
}
{code}
We have an option to add these methods to the interfaces - InputPort and
OutputPort; But these would not be backward compatible and also not
consistent with the current implementation of basic data tuple flow (as
with process() and emit()).
We also need to expose an interface / class for users to wrap their object
and emit downstream. This should be part of API.
{code}
public class ControlTuple extends Tuple
{
Object userObject;
public ControlTuple(long windowId, Object userObject)
{
//
}
}
{code}
The emitted tuples would traverse the same flow as with other control
tuples. The plan is to intercept the control tuples in GenericNode and use
the Reservior to emit the control tuples at the end of the window.
GenericNode seems to be the best place to buffer incoming custom control
tuples without delivering them immediately to the operator port. Once the
end of the window is reached, we plan to use the reservoir sink to push
them to the port. This is different behavior than any other control tuple
where we are changing the order of tuples in the stream. The custom
control
tuples will be buffered and not delivered to the ports until the end of
the
window.
To accomplish this, we need to have a public method in SweepableReservoir
which allows to put a tuple back in the sink of the reservoir.
~ Bhupesh
