IMPALA-2018: Where clause does not propagate to joins inside nested

views This commit fixes an issue where during predicate propagation a predicate from the where clause is not properly assigned at the join node that outer joins the generated predicate. Change-Id: Ifccc1b0e0a0579c3baa48f0fb3dedcbd44941b53 Reviewed-on: http://gerrit.cloudera.org:8080/476 Reviewed-by: Alex Behm <alex.behm@cloudera.com> Tested-by: Internal Jenkins
cloudera · Jun 26, 2015 · 07cd53c · 07cd53c
1 parent d4c8050
commit 07cd53c
Show file tree

Hide file tree

Showing 2 changed files with 54 additions and 3 deletions.
diff --git a/fe/src/main/java/com/cloudera/impala/analysis/Analyzer.java b/fe/src/main/java/com/cloudera/impala/analysis/Analyzer.java
@@ -1386,9 +1386,15 @@ public ArrayList<Expr> getBoundPredicates(TupleId destTid, Set<SlotId> ignoreSlo
             }
           }
 
-          boolean evalByJoin = evalByJoin(srcConjunct) &&
-              (globalState_.ojClauseByConjunct.get(srcConjunct.getId())
-                  != globalState_.outerJoinedTupleIds.get(srcTid));
+          // Check if either srcConjunct or the generated predicate needs to be evaluated
+          // at a join node (IMPALA-2018).
+          boolean evalByJoin =
+              (evalByJoin(srcConjunct)
+               && (globalState_.ojClauseByConjunct.get(srcConjunct.getId())
+                != globalState_.outerJoinedTupleIds.get(srcTid)))
+              || (evalByJoin(p)
+                  && (globalState_.ojClauseByConjunct.get(p.getId())
+                   != globalState_.outerJoinedTupleIds.get(destTid)));
 
           // mark all bound predicates including duplicate ones
           if (reverseValueTransfer && !evalByJoin) markConjunctAssigned(srcConjunct);

diff --git a/testdata/workloads/functional-planner/queries/PlannerTest/predicate-propagation.test b/testdata/workloads/functional-planner/queries/PlannerTest/predicate-propagation.test
@@ -1120,3 +1120,48 @@ where b.id < 10
    partitions=24/24 files=24 size=478.45KB
    predicates: id > 20, functional.alltypes.id < 10
 ====
+# Test proper predicate assignment with predicate propagation when the
+# generated predicate is bound by an outer joined tuple (IMPALA-2018)
+select * from
+  (select j.int_col, sum(j.tinyint_col) as total from
+   (select a.id, a.tinyint_col, b.int_col from
+    functional.alltypestiny a left outer join functional.alltypesagg b
+    on a.id = b.id) j
+   group by j.int_col) v
+where v.int_col = 10
+---- PLAN
+03:AGGREGATE [FINALIZE]
+|  output: sum(a.tinyint_col)
+|  group by: b.int_col
+|  having: j.int_col = 10
+|
+02:HASH JOIN [RIGHT OUTER JOIN]
+|  hash predicates: b.id = a.id
+|
+|--00:SCAN HDFS [functional.alltypestiny a]
+|     partitions=4/4 files=4 size=460B
+|
+01:SCAN HDFS [functional.alltypesagg b]
+   partitions=11/11 files=11 size=814.73KB
+   predicates: b.int_col = 10
+====
+# Test proper predicate assignment with predicate propagation when the
+# generated predicate is bound by an outer joined tuple (IMPALA-2018)
+select * from
+  (select j.int_col, j.tinyint_col from
+   (select a.id, a.tinyint_col, b.int_col from
+    functional.alltypestiny a left outer join functional.alltypesagg b
+    on a.id = b.id) j) v
+where v.int_col = 10
+---- PLAN
+02:HASH JOIN [RIGHT OUTER JOIN]
+|  hash predicates: b.id = a.id
+|  other predicates: b.int_col = 10
+|
+|--00:SCAN HDFS [functional.alltypestiny a]
+|     partitions=4/4 files=4 size=460B
+|
+01:SCAN HDFS [functional.alltypesagg b]
+   partitions=11/11 files=11 size=814.73KB
+   predicates: b.int_col = 10
+====