New variable and behavior: fast_analyze_table for, well, faster ANALYZE TABLE

Introducing a new global/session variable: fast_analyze_table, a boolean that
defaults false. When the value is true (or 1), then an ANALYZE TABLE only
analyzes and persists stats for the clustering index (normally the PRIMARY KEY)
and skips the secondary keys.

The purpose of this change is to support faster ANALYZE TABLE operations, where
all we want is really to get an updated estimation of the number of rows in the
table. That information is extracted from the clustering index, not from the
secondary keys, and thus they can be skipped.

Signed-off-by: Dirkjan Bussink <[email protected]>

Author: shlomi-noach

mysql-test/r/fast_analyze_table.result

@@ -0,0 +1,26 @@
+drop table if exists t0;
+create table t0 (id int primary key);
+show global variables like 'fast_analyze_table';
+Variable_name	Value
+fast_analyze_table	OFF
+show session variables like 'fast_analyze_table';
+Variable_name	Value
+fast_analyze_table	OFF
+select @@global.fast_analyze_table;
+@@global.fast_analyze_table
+0
+select @@session.fast_analyze_table;
+@@session.fast_analyze_table
+0
+set @@fast_analyze_table=1;
+show global variables like 'fast_analyze_table';
+Variable_name	Value
+fast_analyze_table	OFF
+show session variables like 'fast_analyze_table';
+Variable_name	Value
+fast_analyze_table	ON
+analyze table t0;
+Table	Op	Msg_type	Msg_text
+test.t0	analyze	status	OK
+set @@fast_analyze_table=0;
+drop table if exists t0;

mysql-test/t/fast_analyze_table.test

@@ -0,0 +1,25 @@
+#
+# Test of fast_analyze_table variable
+#
+
+--source include/count_sessions.inc
+
+--disable_warnings
+drop table if exists t0;
+--enable_warnings
+
+create table t0 (id int primary key);
+
+show global variables like 'fast_analyze_table';
+show session variables like 'fast_analyze_table';
+
+select @@global.fast_analyze_table;
+select @@session.fast_analyze_table;
+
+set @@fast_analyze_table=1;
+show global variables like 'fast_analyze_table';
+show session variables like 'fast_analyze_table';
+
+analyze table t0;
+set @@fast_analyze_table=0;
+drop table if exists t0;

sql/query_options.h

@@ -129,4 +129,6 @@
 */
 #define OPTION_NO_SUBQUERY_DURING_OPTIMIZATION (1ULL << 39)  // intern
 
+#define OPTION_FAST_ANALYZE_TABLE (1ULL << 61)  // THD, user, binlog
+
 #endif /* QUERY_OPTIONS_INCLUDED */

sql/sys_vars.cc

@@ -5601,6 +5601,12 @@ static Sys_var_bit Sys_unique_checks("unique_checks", "unique_checks",
                                      REVERSE(OPTION_RELAXED_UNIQUE_CHECKS),
                                      DEFAULT(true), NO_MUTEX_GUARD, IN_BINLOG);
 
+static Sys_var_bit Sys_fast_analyze_table("fast_analyze_table", "fast_analyze_table",
+    HINT_UPDATEABLE SESSION_VAR(option_bits),
+    NO_CMD_LINE, OPTION_FAST_ANALYZE_TABLE,
+    DEFAULT(false), NO_MUTEX_GUARD,
+    IN_BINLOG);
+
 #ifdef ENABLED_PROFILING
 static Sys_var_bit Sys_profiling("profiling", "profiling",
                                  SESSION_VAR(option_bits), NO_CMD_LINE,

storage/innobase/dict/dict0stats.cc

@@ -2006,7 +2006,7 @@ static void dict_stats_analyze_index(
  will be saved on disk.
  @return DB_SUCCESS or error code */
 static dberr_t dict_stats_update_persistent(
-    dict_table_t *table) /*!< in/out: table */
+    dict_table_t *table, bool only_clustered_index) /*!< in/out: table */
 {
   dict_index_t *index;
 
@@ -2050,7 +2050,7 @@ static dberr_t dict_stats_update_persistent(
 
     dict_stats_empty_index(index);
 
-    if (dict_stats_should_ignore_index(index)) {
+    if (only_clustered_index || dict_stats_should_ignore_index(index)) {
       continue;
     }
 
@@ -2841,6 +2841,7 @@ storage */
   switch (stats_upd_option) {
     dberr_t err;
 
+    case DICT_STATS_RECALC_PERSISTENT_FAST:
     case DICT_STATS_RECALC_PERSISTENT:
 
       if (srv_read_only_mode) {
@@ -2862,7 +2863,7 @@ storage */
       persistent stats enabled */
       ut_a(strchr(table->name.m_name, '/') != nullptr);
 
-      err = dict_stats_update_persistent(table);
+      err = dict_stats_update_persistent(table, stats_upd_option == DICT_STATS_RECALC_PERSISTENT_FAST);
 
       if (err != DB_SUCCESS) {
         return (err);

storage/innobase/handler/ha_innodb.cc

@@ -17209,6 +17209,10 @@ int ha_innobase::info_low(uint flag, bool is_analyze) {
       if (dict_stats_is_persistent_enabled(ib_table)) {
         if (is_analyze) {
           opt = DICT_STATS_RECALC_PERSISTENT;
+
+          if (m_prebuilt->trx->mysql_thd != nullptr && thd_test_options(m_prebuilt->trx->mysql_thd, OPTION_FAST_ANALYZE_TABLE)) {
+            opt = DICT_STATS_RECALC_PERSISTENT_FAST;
+          }
         } else {
           /* This is e.g. 'SHOW INDEXES', fetch
           the persistent stats from disk. */

storage/innobase/include/dict0stats.h

@@ -46,6 +46,9 @@ enum dict_stats_upd_option_t {
                            storage, if the persistent storage is
                            not present then emit a warning and
                            fall back to transient stats */
+  DICT_STATS_RECALC_PERSISTENT_FAST, /* like DICT_STATS_RECALC_PERSISTENT,
+                            but only computes statistics for the clustering index
+                            (PRIMARY KEY) */
   DICT_STATS_RECALC_TRANSIENT,  /* (re) calculate the statistics
                             using an imprecise quick algo
                             without saving the results