Github user srowen commented on a diff in the pull request:
https://github.com/apache/spark/pull/5074#discussion_r26751786
--- Diff: docs/programming-guide.md ---
@@ -1022,6 +1022,27 @@ for details.
</tr>
</table>
+### Shuffle operations
+
+Certain operations within Spark trigger an operation known as the shuffle.
The shuffle is Spark's mechanism for re-distributing data so data with the same
key becomes co-located after a shuffle.
+
+#### Background
+
+To understand what happens during the shuffle we can consider the example
of the [`groupByKey`](#GroupByLink) operation. The `groupByKey` operation
generates a new RDD where all values for a single key are combined into a
2-tuple - the key and an Iterable object containing all the associated values.
If we think of the map and reduce steps for `groupByKey()` then we can see that
to generate the list of all values associated with a key, all of the values
must reside on the same reducer, since the output of the reduce step is the
complete array.
+
+In Spark, by default, the way data is distributed across partitions is
undefined. During computations, a single task will operate on a single
partition - thus, to organize all the data for a single `groupByKey` reduce
task to execute, Spark needs to perform an all-to-all operation. It must read
from all partitions to find all the values for all keys, and then organize
those such that all values for any key lie within the same partition - this is
called the **shuffle**.
+
+Although the set of elements in each partition of newly shuffled data will
be deterministic, the ordering of these elements is not. If one desires
predictably ordered data following shuffle operations,
[`mapPartitions`](#MapPartLink) can be used to sort each partition. A similar
operation, [`repartitionAndSortWithinPartitions`](#Repartition2Link`) coupled
with `mapPartitions`, may be used to enact a `Hadoop` style shuffle.
+
+Operations, which cause a shuffle include [`groupByKey`](#GroupByLink),
[`sortByKey`](#SortByLink), [`reduceByKey`](#ReduceByLink),
[`aggregateByKey`](#AggregateByLink), [`repartition`](#RepartitionLink),
[`repartitionAndSortWithinPartitions`](#Repartition2Link`),
[`coalesce`](#CoalesceLink), and [`countByKey`](#CountByLink).
--- End diff --
Nit: extra comma after Operations
I think it's fair to say *can* cause a shuffle, since some of these won't
shuffle in some cases.
`combineByKey` too? and all the join operations including `cogroup`? and
`subtractByKey`? It seems like all the "by key" operations do with the
exception of the counting ones, I think? It might be an easier rule of thumb to
state.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
