Github user fhueske commented on a diff in the pull request:
https://github.com/apache/flink/pull/4471#discussion_r132238828
--- Diff:
flink-libraries/flink-table/src/main/scala/org/apache/flink/table/plan/util/UpdatingPlanChecker.scala
---
@@ -90,40 +96,86 @@ object UpdatingPlanChecker {
// resolve names of input fields
.map(io => (inNames.get(io._1), io._2))
- // filter by input keys
- val outKeys = inOutNames.filter(io =>
keys.get.contains(io._1)).map(_._2)
- // check if all keys have been preserved
- if (outKeys.nonEmpty && outKeys.length == keys.get.length) {
+ // filter by input keyAncestors
+ val outKeyAncesters = inOutNames
+ .filter(io => keyAncestors.get.map(e =>
e._1).contains(io._1))
+ .map(io => (io._2, keyAncestors.get.find(ka => ka._1 ==
io._1).get._2))
+
+ // check if all keyAncestors have been preserved
+ if (outKeyAncesters.nonEmpty &&
+ outKeyAncesters.map(ka => ka._2).distinct.length ==
+ keyAncestors.get.map(ka => ka._2).distinct.length) {
// all key have been preserved (but possibly renamed)
- keys = Some(outKeys.toArray)
+ Some(outKeyAncesters.toList)
} else {
// some (or all) keys have been removed. Keys are no longer
unique and removed
- keys = None
+ None
}
+ } else {
+ None
}
+
case _: DataStreamOverAggregate =>
- super.visit(node, ordinal, parent)
- // keys are always forwarded by Over aggregate
+ // keyAncestors are always forwarded by Over aggregate
+ visit(node.getInput(0))
case a: DataStreamGroupAggregate =>
- // get grouping keys
+ // get grouping keyAncestors
val groupKeys =
a.getRowType.getFieldNames.asScala.take(a.getGroupings.length)
- keys = Some(groupKeys.toArray)
+ Some(groupKeys.map(e => (e, e)).toList)
case w: DataStreamGroupWindowAggregate =>
- // get grouping keys
+ // get grouping keyAncestors
val groupKeys =
w.getRowType.getFieldNames.asScala.take(w.getGroupings.length).toArray
// get window start and end time
val windowStartEnd = w.getWindowProperties.map(_.name)
// we have only a unique key if at least one window property is
selected
if (windowStartEnd.nonEmpty) {
- keys = Some(groupKeys ++ windowStartEnd)
+ Some((groupKeys ++ windowStartEnd).map(e => (e, e)).toList)
--- End diff --
I think that window properties should be treated as keys derived from the
window. If we have start and end one of both could be removed and we would
still have a unique key.
So I would add them as follows:
```
Some((groupKeys.map(e => (e, e)) ++ windowStartEnd.map((_,
w.getWindowAlias))).toList)
```
(you have to add `def getWindowAlias: String =
window.aliasAttribute.toString` to `DataStreamGroupWindowAggregate`).
With this we treat the window (by its alias) as the ancestor of the key.
---
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.
---
