Github user uce commented on a diff in the pull request:
https://github.com/apache/flink/pull/2154#discussion_r69152891
--- Diff: docs/apis/streaming/windows.md ---
@@ -24,1023 +24,608 @@ specific language governing permissions and
limitations
under the License.
-->
+Flink uses a concept called *windows* to divide a (potentially) infinite
`DataStream` into finite
+slices based on the timestamps of elements or other criteria. This
division is required when working
+with infinite streams of data and performing transformations that
aggregate elements.
+
+<span class="label label-info">Info</span> We will mostly talk about
*keyed windowing* here, i.e.
+windows that are applied on a `KeyedStream`. Keyed windows have the
advantage that elements are
+subdivided based on both window and key before being given to
+a user function. The work can thus be distributed across the cluster
+because the elements for different keys can be processed independently. If
you absolutely have to,
+you can check out [non-keyed windowing](#non-keyed-windowing) where we
describe how non-keyed
+windows work.
+
* This will be replaced by the TOC
{:toc}
-## Windows on Keyed Data Streams
-
-Flink offers a variety of methods for defining windows on a `KeyedStream`.
All of these group elements *per key*,
-i.e., each window will contain elements with the same key value.
+## Basics
-### Basic Window Constructs
+For a windowed transformations you must at least specify a *key*
+(see [specifying keys](apis/common/index.html#specifying-keys))
+a *window assigner* and a *window function*. The *key* divides the
infinite, non-keyed, stream
+into logical keyed streams while the *window assigner* assigns elements to
finite per-key windows.
+Finally, the *window function* is used to process the elements of each
window.
-Flink offers a general window mechanism that provides flexibility, as well
as a number of pre-defined windows
-for common use cases. See first if your use case can be served by the
pre-defined windows below before moving
-to defining your own windows.
+The basic structure of a windowed transformation is thus as follows:
<div class="codetabs" markdown="1">
<div data-lang="java" markdown="1">
+{% highlight java %}
+DataStream<T> input = ...;
-<br />
-
-<table class="table table-bordered">
- <thead>
- <tr>
- <th class="text-left" style="width: 25%">Transformation</th>
- <th class="text-center">Description</th>
- </tr>
- </thead>
- <tbody>
- <tr>
- <td><strong>Tumbling time window</strong><br>KeyedStream →
WindowedStream</td>
- <td>
- <p>
- Defines a window of 5 seconds, that "tumbles". This means that
elements are
- grouped according to their timestamp in groups of 5 second
duration, and every element belongs to exactly one window.
- The notion of time is specified by the selected TimeCharacteristic
(see <a href="{{ site.baseurl }}/apis/streaming/event_time.html">time</a>).
- {% highlight java %}
-keyedStream.timeWindow(Time.seconds(5));
- {% endhighlight %}
- </p>
- </td>
- </tr>
- <tr>
- <td><strong>Sliding time window</strong><br>KeyedStream →
WindowedStream</td>
- <td>
- <p>
- Defines a window of 5 seconds, that "slides" by 1 second.
This means that elements are
- grouped according to their timestamp in groups of 5 second
duration, and elements can belong to more than
- one window (since windows overlap by at most 4 seconds)
- The notion of time is specified by the selected
TimeCharacteristic (see <a href="{{ site.baseurl
}}/apis/streaming/event_time.html">time</a>).
- {% highlight java %}
-keyedStream.timeWindow(Time.seconds(5), Time.seconds(1));
- {% endhighlight %}
- </p>
- </td>
- </tr>
- <tr>
- <td><strong>Tumbling count window</strong><br>KeyedStream →
WindowedStream</td>
- <td>
- <p>
- Defines a window of 1000 elements, that "tumbles". This means
that elements are
- grouped according to their arrival time (equivalent to
processing time) in groups of 1000 elements,
- and every element belongs to exactly one window.
- {% highlight java %}
-keyedStream.countWindow(1000);
- {% endhighlight %}
- </p>
- </td>
- </tr>
- <tr>
- <td><strong>Sliding count window</strong><br>KeyedStream →
WindowedStream</td>
- <td>
- <p>
- Defines a window of 1000 elements, that "slides" every 100
elements. This means that elements are
- grouped according to their arrival time (equivalent to
processing time) in groups of 1000 elements,
- and every element can belong to more than one window (as windows
overlap by at most 900 elements).
- {% highlight java %}
-keyedStream.countWindow(1000, 100)
- {% endhighlight %}
- </p>
- </td>
- </tr>
- </tbody>
-</table>
-
+input
+ .keyBy(<key selector>)
+ .window(<window assigner>)
+ .<windowed transformation>(<window function>);
+{% endhighlight %}
</div>
<div data-lang="scala" markdown="1">
+{% highlight scala %}
+val input: DataStream[T] = ...
-<br />
-
-<table class="table table-bordered">
- <thead>
- <tr>
- <th class="text-left" style="width: 25%">Transformation</th>
- <th class="text-center">Description</th>
- </tr>
- </thead>
- <tbody>
- <tr>
- <td><strong>Tumbling time window</strong><br>KeyedStream →
WindowedStream</td>
- <td>
- <p>
- Defines a window of 5 seconds, that "tumbles". This means that
elements are
- grouped according to their timestamp in groups of 5 second
duration, and every element belongs to exactly one window.
- The notion of time is specified by the selected
TimeCharacteristic (see <a href="{{ site.baseurl
}}/apis/streaming/event_time.html">time</a>).
- {% highlight scala %}
-keyedStream.timeWindow(Time.seconds(5))
- {% endhighlight %}
- </p>
- </td>
- </tr>
- <tr>
- <td><strong>Sliding time window</strong><br>KeyedStream →
WindowedStream</td>
- <td>
- <p>
- Defines a window of 5 seconds, that "slides" by 1 second.
This means that elements are
- grouped according to their timestamp in groups of 5 second
duration, and elements can belong to more than
- one window (since windows overlap by at most 4 seconds)
- The notion of time is specified by the selected
TimeCharacteristic (see <a href="{{ site.baseurl
}}/apis/streaming/event_time.html">time</a>).
- {% highlight scala %}
-keyedStream.timeWindow(Time.seconds(5), Time.seconds(1))
- {% endhighlight %}
- </p>
- </td>
- </tr>
- <tr>
- <td><strong>Tumbling count window</strong><br>KeyedStream →
WindowedStream</td>
- <td>
- <p>
- Defines a window of 1000 elements, that "tumbles". This means
that elements are
- grouped according to their arrival time (equivalent to
processing time) in groups of 1000 elements,
- and every element belongs to exactly one window.
- {% highlight scala %}
-keyedStream.countWindow(1000)
- {% endhighlight %}
- </p>
- </td>
- </tr>
- <tr>
- <td><strong>Sliding count window</strong><br>KeyedStream →
WindowedStream</td>
- <td>
- <p>
- Defines a window of 1000 elements, that "slides" every 100
elements. This means that elements are
- grouped according to their arrival time (equivalent to
processing time) in groups of 1000 elements,
- and every element can belong to more than one window (as windows
overlap by at most 900 elements).
- {% highlight scala %}
-keyedStream.countWindow(1000, 100)
- {% endhighlight %}
- </p>
- </td>
- </tr>
- </tbody>
-</table>
-
+input
+ .keyBy(<key selector>)
+ .window(<window assigner>)
+ .<windowed transformation>(<window function>)
+{% endhighlight %}
</div>
</div>
-### Advanced Window Constructs
+We will cover [window assigners](#window-assigners) in a separate section
below.
+
+The window transformation can be one of `reduce()`, `fold()` or `apply()`.
Which respectively
+takes a `ReduceFunction`, `FoldFunction` or `WindowFunction`. We describe
each of these ways
+of specifying a windowed transformation in detail below: [window
functions](#window-functions).
+
+For more advanced use cases you can also specify a `Trigger` that
determines when exactly a window
+is being considered as *ready for processing*. These will be covered in
more detail in
+[triggers](#triggers).
+
+## Window Assigners
+
+The window assigner specifies how elements of the stream are divided into
finite slices. Flink comes
+with pre-implemented window assigners for the most typical use cases,
namely *tumbling windows*,
+*sliding windows*, *session windows* and *global windows* but you can
implement your own by
+extending the `WindowAssigner` class. All the built-in window assigners,
except for the global
+windows one, assign elements to windows based on time, which can either be
processing time or event
+time.
+
+Let's first look at how each of these window assigners works before
looking at how they can be used
--- End diff --
I'm not sure but should it be "how each of these window assigners
work~~s~~" (without s)?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
