Github user uce commented on a diff in the pull request:
https://github.com/apache/flink/pull/774#discussion_r31703637
--- Diff: docs/apis/programming_guide.md ---
@@ -394,26 +382,66 @@ def write(outputFormat: FileOutputFormat[T],
writeMode: WriteMode = WriteMode.NO_OVERWRITE)
def print()
-{% endhighlight %}
-The last method is only useful for developing/debugging on a local machine,
-it will output the contents of the DataSet to standard output. (Note that
in
-a cluster, the result goes to the standard out stream of the cluster nodes
and ends
-up in the *.out* files of the workers).
-The first two do as the name suggests, the third one can be used to
specify a
-custom data output format. Please refer
-to [Data Sinks](#data-sinks) for more information on writing to files and
also
-about custom data output formats.
-
-Once you specified the complete program you need to call `execute` on
-the `ExecutionEnvironment`. This will either execute on your local
-machine or submit your program for execution on a cluster, depending on
-how you created the execution environment.
+def collect()
+{% endhighlight %}
</div>
</div>
+The first two methods (`writeAsText()` and `writeAsCsv()`) do as the name
suggests, the third one
+can be used to specify a custom data output format. Please refer to [Data
Sinks](#data-sinks) for
+more information on writing to files and also about custom data output
formats.
+
+The `print()` method is useful for developing/debugging. It will output
the contents of the DataSet
+to standard output (on the JVM starting the Flink execution). **NOTE** The
behavior of the `print()`
+method changed with Flink 0.9.x. Before it was printing to the log file of
the workers, now its
+sending the DataSet results to the client and printing the results there.
+
+`collect()` allows to retrieve the DataSet from the cluster to the local
JVM. The `collect()` method
+will return a `List` containing the elements.
+
+Both `print()` and `collect()` will trigger the execution of the program.
+
+
+**NOTE** `print()` and `collect()` retrieve the data from the cluster to
the client. Currently,
+the data sizes you can retrieve with `collect()` are limited due to our
RPC system. It is not advised
+to collect DataSets larger than 10MBs.
+
+
+Once you specified the complete program you need to **trigger the program
execution**. You can call
+`execute()` directly on the `ExecutionEnviroment` or you implicitly
trigger the execution with
+`collect()` or `print()`.
+Depending on the type of the `ExecutionEnvironment` the execution will be
triggered on your local
+machine or submit your program for execution on a cluster.
+
+Note that you can not call both `print()` (or `collect()`) and `execute()`
at the end of program.
+
+The `execute()` method is returning the `JobExecutionResult`, including
execution times and
+accumulator results. `print()` and `collect()` are not returning the
result, but it can be
+accessed from the `getLastJobExecutionResult()` method.
+
+
+[Back to top](#top)
+
+
+DataSet abstraction
+---------------
+
+The batch processing APIs of Flink are centered around the `DataSet`
abstraction. A `DataSet` is only
+an abstract representation of a set of data that can contain duplicates.
+
+Also note that Flink is not always physically creating (materializing)
each DataSet at runtime. This
+depends on the used runtime, the configuration and optimizer decisions.
+
+The Flink runtime is usually not materializing the DataSets because it is
using a streaming runtime model.
--- End diff --
We could formulate this more positive: The Flink runtime does not need to
always materialize...
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
