@@ -786,7 +786,7 @@ ____________________________________________________________________________-->
786786 <secondary>multivariate</secondary>
787787 </indexterm>
788788
789- <sect2>
789+ <sect2 id="functional-dependencies" >
790790<!--==========================orignal english content==========================
791791 <title>Functional Dependencies</title>
792792____________________________________________________________________________-->
@@ -949,7 +949,7 @@ EXPLAIN (ANALYZE, TIMING OFF) SELECT * FROM t WHERE a = 1 AND b = 1;
949949 </para>
950950 </sect2>
951951
952- <sect2>
952+ <sect2 id="multivariate-ndistinct-counts" >
953953<!--==========================orignal english content==========================
954954 <title>Multivariate N-Distinct Counts</title>
955955____________________________________________________________________________-->
@@ -1034,6 +1034,216 @@ EXPLAIN (ANALYZE, TIMING OFF) SELECT COUNT(*) FROM t GROUP BY a, b;
10341034 </para>
10351035
10361036 </sect2>
1037+
1038+ <sect2 id="mcv-lists">
1039+ <!--==========================orignal english content==========================
1040+ <title>MCV Lists</title>
1041+ ____________________________________________________________________________-->
1042+ <title>MCV 列表</title>
1043+
1044+ <!--==========================orignal english content==========================
1045+ <para>
1046+ As explained in <xref linkend="functional-dependencies"/>, functional
1047+ dependencies are very cheap and efficient type of statistics, but their
1048+ main limitation is their global nature (only tracking dependencies at
1049+ the column level, not between individual column values).
1050+ </para>
1051+ ____________________________________________________________________________-->
1052+ <para>
1053+ 如 <xref linkend="functional-dependencies"/>中所述，函数依赖是非常廉价和高效的统计类型，但它们的主要限制是其全局特性（仅跟踪列级别的依赖项，而不是在单个列值之间）。
1054+ </para>
1055+
1056+ <!--==========================orignal english content==========================
1057+ <para>
1058+ This section introduces multivariate variant of <acronym>MCV</acronym>
1059+ (most-common values) lists, a straightforward extension of the per-column
1060+ statistics described in <xref linkend="row-estimation-examples"/>. These
1061+ statistics address the limitation by storing individual values, but it is
1062+ naturally more expensive, both in terms of building the statistics in
1063+ <command>ANALYZE</command>, storage and planning time.
1064+ </para>
1065+ ____________________________________________________________________________-->
1066+ <para>
1067+ 本节介绍<acronym>MCV</acronym>（最常见值）列表的多变量变体, <xref linkend="row-estimation-examples"/> 中描述的每列统计数据的简单扩展。
1068+ 这些统计数据通过存储单独的值来解决这个限制，但是就构建<command>ANALYZE</command>中的统计数据、存储和规划时间而言，它的成本自然更高。
1069+ </para>
1070+
1071+ <!--==========================orignal english content==========================
1072+ <para>
1073+ Let's look at the query from <xref linkend="functional-dependencies"/>
1074+ again, but this time with a <acronym>MCV</acronym> list created on the
1075+ same set of columns (be sure to drop the functional dependencies, to
1076+ make sure the planner uses the newly created statistics).
1077+
1078+ <programlisting>
1079+ DROP STATISTICS stts;
1080+ CREATE STATISTICS stts2 (mcv) ON a, b FROM t;
1081+ ANALYZE t;
1082+ EXPLAIN (ANALYZE, TIMING OFF) SELECT * FROM t WHERE a = 1 AND b = 1;
1083+ QUERY PLAN
1084+ -−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-
1085+ Seq Scan on t (cost=0.00..195.00 rows=100 width=8) (actual rows=100 loops=1)
1086+ Filter: ((a = 1) AND (b = 1))
1087+ Rows Removed by Filter: 9900
1088+ </programlisting>
1089+
1090+ The estimate is as accurate as with the functional dependencies, mostly
1091+ thanks to the table being fairly small and having a simple distribution
1092+ with a low number of distinct values. Before looking at the second query,
1093+ which was not handled by functional dependencies particularly well,
1094+ let's inspect the <acronym>MCV</acronym> list a bit.
1095+ </para>
1096+ ____________________________________________________________________________-->
1097+ <para>
1098+ 让我们再看看来自<xref linkend="functional-dependencies"/>的查询，但这次在相同列集上创建了<acronym>MCV</acronym>列表（请确保删除函数依赖，以确保规划器使用新创建的统计数据）。
1099+
1100+ <programlisting>
1101+ DROP STATISTICS stts;
1102+ CREATE STATISTICS stts2 (mcv) ON a, b FROM t;
1103+ ANALYZE t;
1104+ EXPLAIN (ANALYZE, TIMING OFF) SELECT * FROM t WHERE a = 1 AND b = 1;
1105+ QUERY PLAN
1106+ -------------------------------------------------------------------------------
1107+ Seq Scan on t (cost=0.00..195.00 rows=100 width=8) (actual rows=100 loops=1)
1108+ Filter: ((a = 1) AND (b = 1))
1109+ Rows Removed by Filter: 9900
1110+ </programlisting>
1111+
1112+ The estimate is as accurate as with the functional dependencies, mostly
1113+ thanks to the table being fairly small and having a simple distribution
1114+ with a low number of distinct values. Before looking at the second query,
1115+ which was not handled by functional dependencies particularly well,
1116+ let's inspect the <acronym>MCV</acronym> list a bit.
1117+ 估计值与函数依赖一样准确，这主要是由于表相当小而且具有少量不同值的简单分布。
1118+ 在查看第二个查询之前，这个函数依赖处理得不是很好，让我们先检查一下<acronym>MCV</acronym>列表。
1119+ </para>
1120+
1121+ <!--==========================orignal english content==========================
1122+ <para>
1123+ Inspecting the <acronym>MCV</acronym> list is possible using
1124+ <function>pg_mcv_list_items</function> set-returning function.
1125+
1126+ <programlisting>
1127+ SELECT m.* FROM pg_statistic_ext join pg_statistic_ext_data on (oid = stxoid),
1128+ pg_mcv_list_items(stxdmcv) m WHERE stxname = 'stts2';
1129+ index | values | nulls | frequency | base_frequency
1130+ -−-−-−-+-−-−-−-−-−+-−-−-−-+-−-−-−-−-−-+-−-−-−-−-−-−-−-−
1131+ 0 | {0, 0} | {f,f} | 0.01 | 0.0001
1132+ 1 | {1, 1} | {f,f} | 0.01 | 0.0001
1133+ ...
1134+ 49 | {49, 49} | {f,f} | 0.01 | 0.0001
1135+ 50 | {50, 50} | {f,f} | 0.01 | 0.0001
1136+ ...
1137+ 97 | {97, 97} | {f,f} | 0.01 | 0.0001
1138+ 98 | {98, 98} | {f,f} | 0.01 | 0.0001
1139+ 99 | {99, 99} | {f,f} | 0.01 | 0.0001
1140+ (100 rows)
1141+ </programlisting>
1142+
1143+ This confirms there are 100 distinct combinations in the two columns, and
1144+ all of them are about equally likely (1% frequency for each one). The
1145+ base frequency is the frequency computed from per-column statistics, as if
1146+ there were no multi-column statistics. Had there been any null values in
1147+ either of the columns, this would be identified in the
1148+ <structfield>nulls</structfield> column.
1149+ </para>
1150+ ____________________________________________________________________________-->
1151+ <para>
1152+ 可以使用<function>pg_mcv_list_items</function>集返回函数检查<acronym>MCV</acronym>列表。
1153+
1154+ <programlisting>
1155+ SELECT m.* FROM pg_statistic_ext join pg_statistic_ext_data on (oid = stxoid),
1156+ pg_mcv_list_items(stxdmcv) m WHERE stxname = 'stts2';
1157+ index | values | nulls | frequency | base_frequency
1158+ -------+----------+-------+-----------+----------------
1159+ 0 | {0, 0} | {f,f} | 0.01 | 0.0001
1160+ 1 | {1, 1} | {f,f} | 0.01 | 0.0001
1161+ ...
1162+ 49 | {49, 49} | {f,f} | 0.01 | 0.0001
1163+ 50 | {50, 50} | {f,f} | 0.01 | 0.0001
1164+ ...
1165+ 97 | {97, 97} | {f,f} | 0.01 | 0.0001
1166+ 98 | {98, 98} | {f,f} | 0.01 | 0.0001
1167+ 99 | {99, 99} | {f,f} | 0.01 | 0.0001
1168+ (100 rows)
1169+ </programlisting>
1170+
1171+ 这确认了在这两列中有100种不同的组合，并且它们的概率都差不多(每个的频率为1%)。
1172+ 基本频率是从每列统计数据中计算的频率，好像没有多列统计数据。如果任一列中有任何空值，将在 <structfield>nulls</structfield>列中标识。
1173+ </para>
1174+
1175+ <!--==========================orignal english content==========================
1176+ <para>
1177+ When estimating the selectivity, the planner applies all the conditions
1178+ on items in the <acronym>MCV</acronym> list, and then sums the frequencies
1179+ of the matching ones. See <function>mcv_clauselist_selectivity</function>
1180+ in <filename>src/backend/statistics/mcv.c</filename> for details.
1181+ </para>
1182+ ____________________________________________________________________________-->
1183+ <para>
1184+ 在估计选择性时，规划器对<acronym>MCV</acronym>列表中的项目应用所有条件，然后对匹配项的频率求和。
1185+ 详情请参阅<filename>src/backend/statistics/mcv.c</filename>中的<function>mcv_clauselist_selectivity</function>。
1186+ </para>
1187+
1188+ <!--==========================orignal english content==========================
1189+ <para>
1190+ Compared to functional dependencies, <acronym>MCV</acronym> lists have two
1191+ major advantages. Firstly, the list stores actual values, making it possible
1192+ to decide which combinations are compatible.
1193+
1194+ <programlisting>
1195+ EXPLAIN (ANALYZE, TIMING OFF) SELECT * FROM t WHERE a = 1 AND b = 10;
1196+ QUERY PLAN
1197+ -−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-
1198+ Seq Scan on t (cost=0.00..195.00 rows=1 width=8) (actual rows=0 loops=1)
1199+ Filter: ((a = 1) AND (b = 10))
1200+ Rows Removed by Filter: 10000
1201+ </programlisting>
1202+
1203+ Secondly, <acronym>MCV</acronym> lists handle a wider range of clause types,
1204+ not just equality clauses like functional dependencies. For example,
1205+ consider the following range query for the same table:
1206+
1207+ <programlisting>
1208+ EXPLAIN (ANALYZE, TIMING OFF) SELECT * FROM t WHERE a <= 49 AND b > 49;
1209+ QUERY PLAN
1210+ -−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-−-
1211+ Seq Scan on t (cost=0.00..195.00 rows=1 width=8) (actual rows=0 loops=1)
1212+ Filter: ((a <= 49) AND (b > 49))
1213+ Rows Removed by Filter: 10000
1214+ </programlisting>
1215+
1216+ </para>
1217+ ____________________________________________________________________________-->
1218+ <para>
1219+ 与函数依赖相比，<acronym>MCV</acronym>列表有两大主要优点。
1220+ 首先，列表存储实际值，从而可以决定哪些组合是兼容的。
1221+
1222+ <programlisting>
1223+ EXPLAIN (ANALYZE, TIMING OFF) SELECT * FROM t WHERE a = 1 AND b = 10;
1224+ QUERY PLAN
1225+ ---------------------------------------------------------------------------
1226+ Seq Scan on t (cost=0.00..195.00 rows=1 width=8) (actual rows=0 loops=1)
1227+ Filter: ((a = 1) AND (b = 10))
1228+ Rows Removed by Filter: 10000
1229+ </programlisting>
1230+
1231+ 第二，<acronym>MCV</acronym> 列表处理更广泛的子句类型，而不仅仅是类似函数依赖的相等子句。
1232+ 例如，请考虑对同一表的以下范围查询：
1233+
1234+ <programlisting>
1235+ EXPLAIN (ANALYZE, TIMING OFF) SELECT * FROM t WHERE a <= 49 AND b > 49;
1236+ QUERY PLAN
1237+ ---------------------------------------------------------------------------
1238+ Seq Scan on t (cost=0.00..195.00 rows=1 width=8) (actual rows=0 loops=1)
1239+ Filter: ((a <= 49) AND (b > 49))
1240+ Rows Removed by Filter: 10000
1241+ </programlisting>
1242+
1243+ </para>
1244+
1245+ </sect2>
1246+
10371247 </sect1>
10381248
10391249 <sect1 id="planner-stats-security">
0 commit comments