Implement SortSupport for macaddr data type

Introduces a scheme to produce abbreviated keys for the macaddr type.
Bump catalog version.

Author: Brandur Leach
Reviewed-by: Julien Rouhaud, Peter Geoghegan

https://commitfest.postgresql.org/13/743/

src/backend/utils/adt/mac.c

@@ -14,9 +14,13 @@
 #include "postgres.h"
 
 #include "access/hash.h"
+#include "lib/hyperloglog.h"
 #include "libpq/pqformat.h"
+#include "port/pg_bswap.h"
 #include "utils/builtins.h"
+#include "utils/guc.h"
 #include "utils/inet.h"
+#include "utils/sortsupport.h"
 
 
 /*
@@ -29,6 +33,21 @@
 #define lobits(addr) \
   ((unsigned long)(((addr)->d<<16)|((addr)->e<<8)|((addr)->f)))
 
+/* sortsupport for macaddr */
+typedef struct
+{
+	int64		input_count;	/* number of non-null values seen */
+	bool		estimating;		/* true if estimating cardinality */
+
+	hyperLogLogState abbr_card; /* cardinality estimator */
+}	macaddr_sortsupport_state;
+
+static int	macaddr_cmp_internal(macaddr *a1, macaddr *a2);
+static int	macaddr_fast_cmp(Datum x, Datum y, SortSupport ssup);
+static int	macaddr_cmp_abbrev(Datum x, Datum y, SortSupport ssup);
+static bool macaddr_abbrev_abort(int memtupcount, SortSupport ssup);
+static Datum macaddr_abbrev_convert(Datum original, SortSupport ssup);
+
 /*
  *	MAC address reader.  Accepts several common notations.
  */
@@ -159,7 +178,7 @@ macaddr_send(PG_FUNCTION_ARGS)
  *	Comparison function for sorting:
  */
 
-static int32
+static int
 macaddr_cmp_internal(macaddr *a1, macaddr *a2)
 {
 	if (hibits(a1) < hibits(a2))
@@ -326,3 +345,194 @@ macaddr_trunc(PG_FUNCTION_ARGS)
 
 	PG_RETURN_MACADDR_P(result);
 }
+
+/*
+ * SortSupport strategy function. Populates a SortSupport struct with the
+ * information necessary to use comparison by abbreviated keys.
+ */
+Datum
+macaddr_sortsupport(PG_FUNCTION_ARGS)
+{
+	SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
+
+	ssup->comparator = macaddr_fast_cmp;
+	ssup->ssup_extra = NULL;
+
+	if (ssup->abbreviate)
+	{
+		macaddr_sortsupport_state *uss;
+		MemoryContext oldcontext;
+
+		oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
+
+		uss = palloc(sizeof(macaddr_sortsupport_state));
+		uss->input_count = 0;
+		uss->estimating = true;
+		initHyperLogLog(&uss->abbr_card, 10);
+
+		ssup->ssup_extra = uss;
+
+		ssup->comparator = macaddr_cmp_abbrev;
+		ssup->abbrev_converter = macaddr_abbrev_convert;
+		ssup->abbrev_abort = macaddr_abbrev_abort;
+		ssup->abbrev_full_comparator = macaddr_fast_cmp;
+
+		MemoryContextSwitchTo(oldcontext);
+	}
+
+	PG_RETURN_VOID();
+}
+
+/*
+ * SortSupport "traditional" comparison function. Pulls two MAC addresses from
+ * the heap and runs a standard comparison on them.
+ */
+static int
+macaddr_fast_cmp(Datum x, Datum y, SortSupport ssup)
+{
+	macaddr    *arg1 = DatumGetMacaddrP(x);
+	macaddr    *arg2 = DatumGetMacaddrP(y);
+
+	return macaddr_cmp_internal(arg1, arg2);
+}
+
+/*
+ * SortSupport abbreviated key comparison function. Compares two MAC addresses
+ * quickly by treating them like integers, and without having to go the heap.
+ */
+static int
+macaddr_cmp_abbrev(Datum x, Datum y, SortSupport ssup)
+{
+	if (x > y)
+		return 1;
+	else if (x == y)
+		return 0;
+	else
+		return -1;
+}
+
+/*
+ * Callback for estimating effectiveness of abbreviated key optimization.
+ *
+ * We pay no attention to the cardinality of the non-abbreviated data, because
+ * there is no equality fast-path within authoritative macaddr comparator.
+ */
+static bool
+macaddr_abbrev_abort(int memtupcount, SortSupport ssup)
+{
+	macaddr_sortsupport_state *uss = ssup->ssup_extra;
+	double		abbr_card;
+
+	if (memtupcount < 10000 || uss->input_count < 10000 || !uss->estimating)
+		return false;
+
+	abbr_card = estimateHyperLogLog(&uss->abbr_card);
+
+	/*
+	 * If we have >100k distinct values, then even if we were sorting many
+	 * billion rows we'd likely still break even, and the penalty of undoing
+	 * that many rows of abbrevs would probably not be worth it. At this point
+	 * we stop counting because we know that we're now fully committed.
+	 */
+	if (abbr_card > 100000.0)
+	{
+#ifdef TRACE_SORT
+		if (trace_sort)
+			elog(LOG,
+				 "macaddr_abbrev: estimation ends at cardinality %f"
+				 " after " INT64_FORMAT " values (%d rows)",
+				 abbr_card, uss->input_count, memtupcount);
+#endif
+		uss->estimating = false;
+		return false;
+	}
+
+	/*
+	 * Target minimum cardinality is 1 per ~2k of non-null inputs. 0.5 row
+	 * fudge factor allows us to abort earlier on genuinely pathological data
+	 * where we've had exactly one abbreviated value in the first 2k
+	 * (non-null) rows.
+	 */
+	if (abbr_card < uss->input_count / 2000.0 + 0.5)
+	{
+#ifdef TRACE_SORT
+		if (trace_sort)
+			elog(LOG,
+				 "macaddr_abbrev: aborting abbreviation at cardinality %f"
+			   " below threshold %f after " INT64_FORMAT " values (%d rows)",
+				 abbr_card, uss->input_count / 2000.0 + 0.5, uss->input_count,
+				 memtupcount);
+#endif
+		return true;
+	}
+
+#ifdef TRACE_SORT
+	if (trace_sort)
+		elog(LOG,
+			 "macaddr_abbrev: cardinality %f after " INT64_FORMAT
+			 " values (%d rows)", abbr_card, uss->input_count, memtupcount);
+#endif
+
+	return false;
+}
+
+/*
+ * SortSupport converstion routine. Converts original macaddr representation
+ * to abbreviated key representation.
+ *
+ * Packs the bytes of a 6-byte MAC address into a Datum and treats it as an
+ * unsigned integer for purposes of comparison. On a 64-bit machine, there
+ * will be two zeroed bytes of padding. The integer is converted to native
+ * endianness to facilitate easy comparison.
+ */
+static Datum
+macaddr_abbrev_convert(Datum original, SortSupport ssup)
+{
+	macaddr_sortsupport_state *uss = ssup->ssup_extra;
+	macaddr    *authoritative = DatumGetMacaddrP(original);
+	Datum		res;
+
+	/*
+	 * On a 64-bit machine, zero out the 8-byte datum and copy the 6 bytes of
+	 * the MAC address in. There will be two bytes of zero padding on the end
+	 * of the least significant bits.
+	 */
+#if SIZEOF_DATUM == 8
+	memset(&res, 0, SIZEOF_DATUM);
+	memcpy(&res, authoritative, sizeof(macaddr));
+#else							/* SIZEOF_DATUM != 8 */
+	memcpy(&res, authoritative, SIZEOF_DATUM);
+#endif
+	uss->input_count += 1;
+
+	/*
+	 * Cardinality estimation. The estimate uses uint32, so on a 64-bit
+	 * architecture, XOR the two 32-bit halves together to produce slightly
+	 * more entropy. The two zeroed bytes won't have any practical impact on
+	 * this operation.
+	 */
+	if (uss->estimating)
+	{
+		uint32		tmp;
+
+#if SIZEOF_DATUM == 8
+		tmp = (uint32) res ^ (uint32) ((uint64) res >> 32);
+#else							/* SIZEOF_DATUM != 8 */
+		tmp = (uint32) res;
+#endif
+
+		addHyperLogLog(&uss->abbr_card, DatumGetUInt32(hash_uint32(tmp)));
+	}
+
+	/*
+	 * Byteswap on little-endian machines.
+	 *
+	 * This is needed so that macaddr_cmp_abbrev() (an unsigned integer 3-way
+	 * comparator) works correctly on all platforms. Without this, the
+	 * comparator would have to call memcmp() with a pair of pointers to the
+	 * first byte of each abbreviated key, which is slower.
+	 */
+	res = DatumBigEndianToNative(res);
+
+	return res;
+}

src/include/catalog/catversion.h

@@ -53,6 +53,6 @@
  */
 
 /*							yyyymmddN */
-#define CATALOG_VERSION_NO	201703291
+#define CATALOG_VERSION_NO	201703292
 
 #endif

src/include/catalog/pg_amproc.h

@@ -117,6 +117,7 @@ DATA(insert (	1976   20 23 1 2189 ));
 DATA(insert (	1976   20 21 1 2193 ));
 DATA(insert (	1982   1186 1186 1 1315 ));
 DATA(insert (	1984   829 829 1 836 ));
+DATA(insert (	1984   829 829 2 3359 ));
 DATA(insert (	1986   19 19 1 359 ));
 DATA(insert (	1986   19 19 2 3135 ));
 DATA(insert (	1988   1700 1700 1 1769 ));

src/include/catalog/pg_proc.h

@@ -2125,6 +2125,8 @@ DESCR("less-equal-greater");
 DATA(insert OID = 3144 (  macaddr_not		PGNSP PGUID 12 1 0 0 0 f f f f t f i s 1 0 829 "829" _null_ _null_ _null_ _null_ _null_ macaddr_not _null_ _null_ _null_ ));
 DATA(insert OID = 3145 (  macaddr_and		PGNSP PGUID 12 1 0 0 0 f f f f t f i s 2 0 829 "829 829" _null_ _null_ _null_ _null_ _null_ macaddr_and _null_ _null_ _null_ ));
 DATA(insert OID = 3146 (  macaddr_or		PGNSP PGUID 12 1 0 0 0 f f f f t f i s 2 0 829 "829 829" _null_ _null_ _null_ _null_ _null_ macaddr_or _null_ _null_ _null_ ));
+DATA(insert OID = 3359 (  macaddr_sortsupport PGNSP PGUID 12 1 0 0 0 f f f f t f i s 1 0 2278 "2281" _null_ _null_ _null_ _null_ _null_ macaddr_sortsupport _null_ _null_ _null_ ));
+DESCR("sort support");
 
 /* for macaddr8 type support */
 DATA(insert OID = 4110 (  macaddr8_in		PGNSP PGUID 12 1 0 0 0 f f f f t f i s 1 0 774 "2275" _null_ _null_ _null_ _null_ _null_ macaddr8_in _null_ _null_ _null_ ));