Streams Versus Windows
Both streams and windows process events. The difference is that windows have state, meaning they can retain and store data, while streams are stateless and cannot.
Streams process incoming events and produce output events according to the continuous query that is attached to the stream, but no data is retained.
By contrast, a window consists of a table where incoming events can add rows, update existing rows, or delete rows. You can set the size of the window based on time, or on the number of events recorded. For example, a window might retain all events over the past 20 minutes, or the most recent 1,000 events. A window can also retain all events. In this case, the incoming event stream must be self-managing in that it contains events that both insert rows into the window and delete rows from the window, so that the window does not grow infinitely large. Windows are needed for performing aggregate operations, as this cannot be done on streams.
Streams and windows can be designated as input, output, or local. Input streams are the point at which data enters the project from external sources via adapters. A project may have any number of input streams. Input streams do not have continuous queries attached to them, although you can define filters for them.
Local and output streams and windows take their input from other streams or windows, rather than from adapters, and they apply a continuous query to produce their output. Local streams and windows are identical to output streams and windows, except that local streams and windows are hidden from outside subscribers. Thus, when a subscriber selects which stream or window to subscribe to, only output streams and windows are available.