HIVE-29556: do not adjust "unknown" NDVs in extractNDVGroupingColumns

ql/src/java/org/apache/hadoop/hive/ql/plan/ColStatistics.java

@@ -31,6 +31,7 @@ public class ColStatistics {
   private boolean isPrimaryKey;
   private boolean isEstimated;
   private boolean isFilteredColumn;
+  private boolean isConst;
   private byte[] bitVectors;
   private byte[] histogram;
 
@@ -155,6 +156,8 @@ public String toString() {
 
     sb.append(" isEstimated: ");
     sb.append(isEstimated);
+    sb.append(" isConst: ");
+    sb.append(isConst);
     return sb.toString();
   }
 
@@ -171,6 +174,7 @@ public ColStatistics clone() {
     clone.setPrimaryKey(isPrimaryKey);
     clone.setIsEstimated(isEstimated);
     clone.setIsFilteredColumn(isFilteredColumn);
+    clone.setConst(isConst);
     if (range != null ) {
       clone.setRange(range.clone());
     }
@@ -232,4 +236,12 @@ public boolean isFilteredColumn() {
     return isFilteredColumn;
   }
 
+  public boolean isConst() {
+    return isConst;
+  }
+
+  public void setConst(boolean isConst) {
+    this.isConst = isConst;
+  }
+
 }

ql/src/java/org/apache/hadoop/hive/ql/stats/StatsUtils.java

@@ -823,9 +823,11 @@ public static ColStatistics getColStatistics(ColumnStatisticsObj cso, String col
       if (numTrues == 0 && numFalses == 0) {
         // All NULL column - no non-null distinct values
         cs.setCountDistint(0);
+        cs.setConst(true);
       } else if (numTrues == 0 || numFalses == 0) {
         // One value type confirmed absent (=0), other is present (>0) or unknown (<0)
         cs.setCountDistint(1);
+        cs.setConst(csd.getBooleanStats().getNumNulls() == 0 && (numTrues > 0 || numFalses > 0));
       } else {
         // Both != 0: either both present (>0), both unknown (<0), or one present + one unknown
         cs.setCountDistint(2);
@@ -1646,6 +1648,7 @@ private static ColStatistics buildColStatForConstant(HiveConf conf, long numRows
     colStats.setAvgColLen(avgColSize);
     colStats.setCountDistint(countDistincts);
     colStats.setNumNulls(numNulls);
+    colStats.setConst(true);
 
     Optional<Number> value = getConstValue(encd);
     value.ifPresent(number -> colStats.setRange(number, number));
@@ -2112,7 +2115,7 @@ private static List<Long> extractNDVGroupingColumns(List<ColStatistics> colStats
     for (ColStatistics cs : colStats) {
       if (cs != null) {
         long ndv = cs.getCountDistint();
-        if (cs.getNumNulls() > 0) {
+        if (cs.getNumNulls() > 0 && (ndv > 0 || cs.isConst())) {
           ndv = StatsUtils.safeAdd(ndv, 1);
         }
         ndvValues.add(ndv);

ql/src/java/org/apache/hadoop/hive/ql/stats/estimator/PessimisticStatCombiner.java

@@ -62,6 +62,7 @@ public void add(ColStatistics stat) {
     if (stat.isFilteredColumn()) {
       result.setFilterColumn();
     }
+    result.setConst(result.isConst() && stat.isConst());
   }
   public Optional<ColStatistics> getResult() {
     return Optional.of(result);

ql/src/test/org/apache/hadoop/hive/ql/stats/TestStatsUtils.java

@@ -367,6 +367,46 @@ void testGetColStatisticsBooleanAllNull() {
 
     assertNotNull(cs);
     assertEquals(0, cs.getCountDistint(), "Boolean NDV should be 0 for all-NULL column");
+    assertEquals(true, cs.isConst(), "all-NULL boolean column must be marked isConst");
+  }
+
+  @ParameterizedTest(name = "isConst=true for verified single-value boolean (numTrues={0}, numFalses={1}, numNulls=0)")
+  @org.junit.jupiter.params.provider.CsvSource({"100, 0", "0, 100"})
+  void testGetColStatisticsBooleanIsConstForVerifiedSingleValue(long numTrues, long numFalses) {
+    ColumnStatisticsObj cso = new ColumnStatisticsObj();
+    cso.setColName("bool_col");
+    cso.setColType(serdeConstants.BOOLEAN_TYPE_NAME);
+    BooleanColumnStatsData boolStats = new BooleanColumnStatsData();
+    boolStats.setNumTrues(numTrues);
+    boolStats.setNumFalses(numFalses);
+    boolStats.setNumNulls(0);
+    ColumnStatisticsData data = new ColumnStatisticsData();
+    data.setBooleanStats(boolStats);
+    cso.setStatsData(data);
+
+    ColStatistics cs = StatsUtils.getColStatistics(cso, "bool_col");
+
+    assertEquals(true, cs.isConst());
+  }
+
+  @ParameterizedTest(name = "isConst stays false for boolean (numTrues={0}, numFalses={1})")
+  @org.junit.jupiter.params.provider.CsvSource({
+      "100, 0", "0, 100", "-1, 100", "100, -1", "50, 50"})
+  void testGetColStatisticsBooleanIsConstNotSetForNonAllNullCases(long numTrues, long numFalses) {
+    ColumnStatisticsObj cso = new ColumnStatisticsObj();
+    cso.setColName("bool_col");
+    cso.setColType(serdeConstants.BOOLEAN_TYPE_NAME);
+    BooleanColumnStatsData boolStats = new BooleanColumnStatsData();
+    boolStats.setNumTrues(numTrues);
+    boolStats.setNumFalses(numFalses);
+    boolStats.setNumNulls(10);
+    ColumnStatisticsData data = new ColumnStatisticsData();
+    data.setBooleanStats(boolStats);
+    cso.setStatsData(data);
+
+    ColStatistics cs = StatsUtils.getColStatistics(cso, "bool_col");
+
+    assertEquals(false, cs.isConst());
   }
 
   @Test
@@ -565,4 +605,24 @@ void testGetColStatisticsTimestampType() {
     assertEquals(1700000000L, range.maxValue.longValue(), "maxValue mismatch for TIMESTAMP");
   }
 
+  @Test
+  void testColStatisticsIsConstDefaultsFalse() {
+    ColStatistics cs = new ColStatistics("c", "int");
+    assertEquals(false, cs.isConst(), "isConst should default to false");
+  }
+
+  @Test
+  void testColStatisticsIsConstSetterAndClone() {
+    ColStatistics cs = new ColStatistics("c", "int");
+    cs.setConst(true);
+    assertEquals(true, cs.isConst(), "setConst(true) should be observable via isConst()");
+
+    ColStatistics clone = cs.clone();
+    assertEquals(true, clone.isConst(), "clone() must propagate isConst");
+
+    cs.setConst(false);
+    assertEquals(false, cs.isConst(), "setConst(false) should clear the flag");
+    assertEquals(true, clone.isConst(), "Clone should be independent of source after mutation");
+  }
+
 }

ql/src/test/org/apache/hadoop/hive/ql/stats/estimator/TestPessimisticStatCombiner.java

@@ -22,6 +22,8 @@
 
 import org.apache.hadoop.hive.ql.plan.ColStatistics;
 import org.junit.jupiter.api.Test;
+import org.junit.jupiter.params.ParameterizedTest;
+import org.junit.jupiter.params.provider.CsvSource;
 
 class TestPessimisticStatCombiner {
 
@@ -155,6 +157,21 @@ void testCombineBothUnknownNumTruesAndNumFalses() {
     assertEquals(-1, combined.getNumFalses(), "Both unknown should result in unknown (-1)");
   }
 
+  @ParameterizedTest(name = "combine isConst({0}, {1}) = {2}")
+  @CsvSource({"true, true, true", "true, false, false", "false, true, false", "false, false, false"})
+  void testCombineIsConstAndSemantics(boolean stat1Const, boolean stat2Const, boolean expected) {
+    ColStatistics stat1 = createStat("col1", "int", 0, 100, 4.0);
+    stat1.setConst(stat1Const);
+    ColStatistics stat2 = createStat("col2", "int", 0, 100, 4.0);
+    stat2.setConst(stat2Const);
+
+    PessimisticStatCombiner combiner = new PessimisticStatCombiner();
+    combiner.add(stat1);
+    combiner.add(stat2);
+
+    assertEquals(expected, combiner.getResult().get().isConst());
+  }
+
   private ColStatistics createStat(String name, String type, long ndv, long numNulls, double avgColLen) {
     ColStatistics stat = new ColStatistics(name, type);
     stat.setCountDistint(ndv);

ql/src/test/queries/clientpositive/groupby_unknown_ndv.q

@@ -0,0 +1,30 @@
+-- HIVE-29556: GROUP BY of a column with unknown NDV (NDV=0 sentinel + numNulls>0)
+-- must not collapse to "1 row" estimate. Both probes feed a join sized so that
+-- master's bogus 1-row estimate triggers Map Join while the heuristic fallback
+-- forces Merge Join.
+
+SET hive.auto.convert.join=true;
+
+CREATE TABLE big (k BIGINT);
+CREATE TABLE small (id BIGINT, name STRING);
+
+ALTER TABLE big UPDATE STATISTICS SET('numRows'='100000000');
+ALTER TABLE big UPDATE STATISTICS for column k SET ('numDVs'='0','numNulls'='100000000');
+
+ALTER TABLE small UPDATE STATISTICS SET('numRows'='1000000');
+ALTER TABLE small UPDATE STATISTICS for column id   SET ('numDVs'='1000000','numNulls'='0');
+ALTER TABLE small UPDATE STATISTICS for column name SET ('numDVs'='1000000','numNulls'='0','avgColLen'='100','maxColLen'='100');
+
+-- U1: direct column with unknown NDV.
+EXPLAIN
+SELECT s.name, g.cnt
+FROM (SELECT k, COUNT(*) AS cnt FROM big GROUP BY k) g
+JOIN small s ON g.cnt = s.id;
+
+-- U2: PessimisticStatCombiner output for CASE WHEN with one NULL branch.
+EXPLAIN
+SELECT s.name, g.cnt
+FROM (SELECT x, COUNT(*) AS cnt
+      FROM (SELECT CASE WHEN k > 0 THEN k ELSE cast(NULL AS BIGINT) END AS x FROM big) t
+      GROUP BY x) g
+JOIN small s ON g.cnt = s.id;