[
https://issues.apache.org/jira/browse/FLINK-1105?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14301080#comment-14301080
]
ASF GitHub Bot commented on FLINK-1105:
---------------------------------------
Github user rmetzger commented on a diff in the pull request:
https://github.com/apache/flink/pull/347#discussion_r23914605
--- Diff:
flink-java/src/main/java/org/apache/flink/api/java/operators/DataSink.java ---
@@ -83,6 +93,107 @@ public DataSink(DataSet<T> data, OutputFormat<T>
format, TypeInformation<T> type
}
/**
+ * Sorts each local partition of a {@link
org.apache.flink.api.java.tuple.Tuple} data set
+ * on the specified field in the specified {@link Order} before it is
emitted by the output format.</br>
+ * <b>Note: Only tuple data sets can be sorted using integer field
indices.</b><br/>
+ * The tuple data set can be sorted on multiple fields in different
orders
+ * by chaining {@link #sortLocalOutput(int, Order)} calls.
+ *
+ * @param field The Tuple field on which the data set is locally sorted.
+ * @param order The Order in which the specified Tuple field is locally
sorted.
+ * @return This data sink operator with specified output order.
+ *
+ * @see org.apache.flink.api.java.tuple.Tuple
+ * @see Order
+ */
+ public DataSink<T> sortLocalOutput(int field, Order order) {
+
+ if (!this.type.isTupleType()) {
+ throw new InvalidProgramException("Specifying order
keys via field positions is only valid for tuple data types");
+ }
+ if (field >= this.type.getArity()) {
+ throw new InvalidProgramException("Order key out of
tuple bounds.");
+ }
+
+ if(this.sortKeyPositions == null) {
+ // set sorting info
+ this.sortKeyPositions = new int[] {field};
+ this.sortOrders = new Order[] {order};
+ } else {
+ // append sorting info to exising info
+ int newLength = this.sortKeyPositions.length + 1;
+ this.sortKeyPositions =
Arrays.copyOf(this.sortKeyPositions, newLength);
+ this.sortOrders = Arrays.copyOf(this.sortOrders,
newLength);
+ this.sortKeyPositions[newLength-1] = field;
+ this.sortOrders[newLength-1] = order;
+ }
+ return this;
+ }
+
+ /**
+ * Sorts each local partition of a data set on the field(s) specified
by the field expression
+ * in the specified {@link Order} before it is emitted by the output
format.</br>
+ * <b>Note: Non-composite types can only be sorted on the full element
which is specified by
+ * a wildcard expression ("*" or "_").</b><br/>
+ * Data sets of composite types (Tuple or Pojo) can be sorted on
multiple fields in different orders
+ * by chaining {@link #sortLocalOutput(String, Order)} calls.
+ *
+ * @param fieldExpression The field expression for the field(s) on
which the data set is locally sorted.
+ * @param order The Order in which the specified field(s) are locally
sorted.
+ * @return This data sink operator with specified output order.
+ *
+ * @see Order
+ */
+ public DataSink<T> sortLocalOutput(String fieldExpression, Order order)
{
+
+ int numFields;
+ int[] fields;
+ Order[] orders;
+
+ if(this.type instanceof CompositeType) {
+ // compute flat field positions for (nested) sorting
fields
+ Keys.ExpressionKeys<T> ek;
+ try {
+ ek = new Keys.ExpressionKeys<T>(new
String[]{fieldExpression}, this.type);
+ } catch(IllegalArgumentException iae) {
+ throw new
InvalidProgramException(iae.getMessage());
--- End diff --
Why are you creating a new exception with the error message instead of
forwarding the illegal argument exception?
I personally like it very much when I can find the exact location where the
exception was thrown.
> Add support for locally sorted output
> -------------------------------------
>
> Key: FLINK-1105
> URL: https://issues.apache.org/jira/browse/FLINK-1105
> Project: Flink
> Issue Type: Sub-task
> Components: Java API
> Reporter: Fabian Hueske
> Assignee: Fabian Hueske
> Priority: Minor
>
> This feature will make it possible to sort the output which is sent to an
> OutputFormat to obtain a locally sorted result.
> This feature was available in the "old" Java API and has not be ported to the
> new Java API yet. Hence optimizer and runtime should already have support for
> this feature. However, the API and job generation part is missing.
> It is also a subfeature of FLINK-598 which will provide also globally sorted
> results.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
