jcamachor commented on a change in pull request #1531:
URL: https://github.com/apache/hive/pull/1531#discussion_r497866947



##########
File path: 
ql/src/java/org/apache/hadoop/hive/ql/optimizer/stats/annotation/StatsRulesProcFactory.java
##########
@@ -2921,6 +2920,97 @@ public Object process(Node nd, Stack<Node> stack, 
NodeProcessorCtx procCtx,
     }
   }
 
+  /**
+   * LateralViewJoinOperator changes the data size and column level statistics.
+   *
+   * A diagram of LATERAL VIEW.
+   *
+   *   [Lateral View Forward]
+   *          /     \
+   *    [Select]  [Select]
+   *        |        |
+   *        |     [UDTF]
+   *        \       /
+   *   [Lateral View Join]
+   *
+   * For each row of the source, the left branch just picks columns and the 
right branch processes UDTF.
+   * And then LVJ joins a row from the left branch with rows from the right 
branch.
+   * The join has one-to-many relationship since UDTF can generate multiple 
rows.
+   *
+   * This rule multiplies the stats from the left branch by T(right) / T(left) 
and sums up the both sides.
+   */
+  public static class LateralViewJoinStatsRule extends DefaultStatsRule 
implements SemanticNodeProcessor {
+    @Override
+    public Object process(Node nd, Stack<Node> stack, NodeProcessorCtx procCtx,
+                          Object... nodeOutputs) throws SemanticException {
+      final LateralViewJoinOperator lop = (LateralViewJoinOperator) nd;
+      final AnnotateStatsProcCtx aspCtx = (AnnotateStatsProcCtx) procCtx;
+      final HiveConf conf = aspCtx.getConf();
+
+      if (!isAllParentsContainStatistics(lop)) {
+        return null;
+      }
+
+      final List<Operator<? extends OperatorDesc>> parents = 
lop.getParentOperators();
+      if (parents.size() != 2) {
+        LOG.warn("LateralViewJoinOperator should have just two parents but 
actually has "
+                + parents.size() + " parents.");
+        return null;
+      }
+
+      final Statistics selectStats = 
parents.get(LateralViewJoinOperator.SELECT_TAG).getStatistics().clone();

Review comment:
       Do you need to clone them? Are you modifying them? (Same for next line)

##########
File path: 
ql/src/java/org/apache/hadoop/hive/ql/optimizer/stats/annotation/StatsRulesProcFactory.java
##########
@@ -2921,6 +2920,97 @@ public Object process(Node nd, Stack<Node> stack, 
NodeProcessorCtx procCtx,
     }
   }
 
+  /**
+   * LateralViewJoinOperator changes the data size and column level statistics.
+   *
+   * A diagram of LATERAL VIEW.
+   *
+   *   [Lateral View Forward]
+   *          /     \
+   *    [Select]  [Select]
+   *        |        |
+   *        |     [UDTF]
+   *        \       /
+   *   [Lateral View Join]
+   *
+   * For each row of the source, the left branch just picks columns and the 
right branch processes UDTF.
+   * And then LVJ joins a row from the left branch with rows from the right 
branch.
+   * The join has one-to-many relationship since UDTF can generate multiple 
rows.

Review comment:
       Just leaving a note. I took a quick look at the UDTF logic and it seems 
the selectivity is hardcoded via config. It seems the outer flag is not taken 
into account either, which could be a straightforward improvement for the 
estimates, i.e., UDFT will produce at least as many rows as it receives.
   




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: gitbox-unsubscr...@hive.apache.org
For additional commands, e-mail: gitbox-h...@hive.apache.org

Reply via email to