sql/stats: skip over histogram buckets for dropped enum values

While decoding histogram upper bounds, if we discover that a physical
representation of an enum value cannot be found in the enum type, assume
that the value has been dropped and skip over the bucket. This allows us
to continue using the most recent statistics after altering an enum
type.

Fixes: #67050

Release note (bug fix): Fix a bug which causes the optimizer to use
stale table statistics after altering an enum type used in the table.

pkg/sql/logictest/testdata/logic_test/distsql_stats

@@ -3374,3 +3374,43 @@ upper_bound   range_rows  distinct_range_rows  equal_rows
 'paid'        0           0                    1
 'dispatched'  0           0                    1
 'delivered'   0           0                    1
+
+# Regression test for #67050: make sure we skip over enum values that have been
+# dropped when decoding histograms.
+
+statement ok
+CREATE TYPE e67050 AS ENUM ('a', 'b', 'c')
+
+statement ok
+CREATE TABLE t67050 (x e67050 PRIMARY KEY)
+
+statement ok
+INSERT INTO t67050 VALUES ('a'), ('b'), ('c')
+
+statement ok
+ANALYZE t67050
+
+statement ok
+DELETE FROM t67050 WHERE x = 'a'
+
+statement ok
+ALTER TYPE e67050 DROP VALUE 'a'
+
+query T
+SELECT jsonb_pretty(statistics->0->'histo_buckets') FROM
+[SHOW STATISTICS USING JSON FOR TABLE t67050]
+----
+[
+    {
+        "distinct_range": 0,
+        "num_eq": 1,
+        "num_range": 0,
+        "upper_bound": "b"
+    },
+    {
+        "distinct_range": 0,
+        "num_eq": 1,
+        "num_range": 0,
+        "upper_bound": "c"
+    }
+]

pkg/sql/stats/json.go

@@ -63,29 +63,29 @@ func (js *JSONStatistic) SetHistogram(h *HistogramData) error {
 	// done across databases. If it is a user-defined type, we need the type name
 	// resolution to be for the correct database.
 	js.HistogramColumnType = typ.SQLStringFullyQualified()
-	js.HistogramBuckets = make([]JSONHistoBucket, len(h.Buckets))
+	js.HistogramBuckets = make([]JSONHistoBucket, 0, len(h.Buckets))
 	js.HistogramVersion = h.Version
 	var a tree.DatumAlloc
 	for i := range h.Buckets {
 		b := &h.Buckets[i]
-		js.HistogramBuckets[i].NumEq = b.NumEq
-		js.HistogramBuckets[i].NumRange = b.NumRange
-		js.HistogramBuckets[i].DistinctRange = b.DistinctRange
-
 		if b.UpperBound == nil {
 			return fmt.Errorf("histogram bucket upper bound is unset")
 		}
 		datum, err := DecodeUpperBound(h.Version, typ, &a, b.UpperBound)
 		if err != nil {
+			if h.ColumnType.Family() == types.EnumFamily && errors.Is(err, types.EnumValueNotFound) {
+				// Skip over buckets for enum values that were dropped.
+				continue
+			}
 			return err
 		}
 
-		js.HistogramBuckets[i] = JSONHistoBucket{
+		js.HistogramBuckets = append(js.HistogramBuckets, JSONHistoBucket{
 			NumEq:         b.NumEq,
 			NumRange:      b.NumRange,
 			DistinctRange: b.DistinctRange,
 			UpperBound:    tree.AsStringWithFlags(datum, tree.FmtExport),
-		}
+		})
 	}
 	return nil
 }

pkg/sql/stats/stats_cache.go

@@ -681,41 +681,42 @@ func (sc *TableStatisticsCache) parseStats(
 // the resulting buckets into tabStat.Histogram.
 func DecodeHistogramBuckets(tabStat *TableStatistic) error {
 	h := tabStat.HistogramData
-	var offset int
 	if tabStat.NullCount > 0 {
 		// A bucket for NULL is not persisted, but we create a fake one to
 		// make histograms easier to work with. The length of res.Histogram
 		// is therefore 1 greater than the length of the histogram data
 		// buckets.
 		// TODO(michae2): Combine this with setHistogramBuckets, especially if we
 		// need to change both after #6224 is fixed (NULLS LAST in index ordering).
-		tabStat.Histogram = make([]cat.HistogramBucket, len(h.Buckets)+1)
+		tabStat.Histogram = make([]cat.HistogramBucket, 1, len(h.Buckets)+1)
 		tabStat.Histogram[0] = cat.HistogramBucket{
 			NumEq:         float64(tabStat.NullCount),
 			NumRange:      0,
 			DistinctRange: 0,
 			UpperBound:    tree.DNull,
 		}
-		offset = 1
 	} else {
-		tabStat.Histogram = make([]cat.HistogramBucket, len(h.Buckets))
-		offset = 0
+		tabStat.Histogram = make([]cat.HistogramBucket, 0, len(h.Buckets))
 	}
 
 	// Decode the histogram data so that it's usable by the opt catalog.
 	var a tree.DatumAlloc
-	for i := offset; i < len(tabStat.Histogram); i++ {
-		bucket := &h.Buckets[i-offset]
+	for i := range h.Buckets {
+		bucket := &h.Buckets[i]
 		datum, err := DecodeUpperBound(h.Version, h.ColumnType, &a, bucket.UpperBound)
 		if err != nil {
+			if h.ColumnType.Family() == types.EnumFamily && errors.Is(err, types.EnumValueNotFound) {
+				// Skip over buckets for enum values that were dropped.
+				continue
+			}
 			return err
 		}
-		tabStat.Histogram[i] = cat.HistogramBucket{
+		tabStat.Histogram = append(tabStat.Histogram, cat.HistogramBucket{
 			NumEq:         float64(bucket.NumEq),
 			NumRange:      float64(bucket.NumRange),
 			DistinctRange: bucket.DistinctRange,
 			UpperBound:    datum,
-		}
+		})
 	}
 	return nil
 }

pkg/sql/types/types.go

@@ -2787,6 +2787,8 @@ func (t *T) IsNumeric() bool {
 	}
 }
 
+var EnumValueNotFound = errors.New("could not find enum value")
+
 // EnumGetIdxOfPhysical returns the index within the TypeMeta's slice of
 // enum physical representations that matches the input byte slice.
 func (t *T) EnumGetIdxOfPhysical(phys []byte) (int, error) {
@@ -2799,7 +2801,7 @@ func (t *T) EnumGetIdxOfPhysical(phys []byte) (int, error) {
 			return i, nil
 		}
 	}
-	err := errors.Newf(
+	err := errors.Wrapf(EnumValueNotFound,
 		"could not find %v in enum %q representation %s %s",
 		phys,
 		t.TypeMeta.Name.FQName(true /* explicitCatalog */),