Re: [Tarantool-patches] [PATCH v3 1/1] sql: enable autoindex optimization

[Nikita Pettik <korablev@tarantool.org>]

On 11 Oct 13:48, imeevma@tarantool.org wrote:
> From: Kirill Yukhin <kyukhin@tarantool.org>
> 
> This patch enables "autoindex" optimization. This optimization uses an
> ephemeral space that contains all the fields used in the query. This

Не очень понимаю, что значит "все поля в запросе"..? Это как?:)
Ты имеешь ввиду все поля result-set'a? Или объединение всех полей
всех таблиц используемых в запросе?

> ephemeral space is called "autoindex". Autoindex may have fewer fields
> than the original space, and the order of these fields in autoindex may

Опять: original space - это какой спейс? В запросе может участвовать
много таблиц..

Еще имхо какая-то путаница выходит: называем эфемерные спейсы автоиндексами.
Я бы обозвал это как-нибудь по-другому. Это же просто еще одно использование
эфемерных таблиц, не более.

> be different from their order in the original space. In addition to the
> mentioned fields, autoindex contains a rowid.
> 
> The primary key of autoindex is a covering index, i.e. it contains all
> of its fields.

В мемтиксе все индексы покрывающие. И 'covering' - свойство для каждого
запроса отдельно взятого, а не индекса как такового.

> 
> Currently this optimization mainly used in cases when values SELECTED
> from more than one space and WHERE specified.
> 
> At the moment, this optimization does not take into account the number
> of tuples in spaces. This should be fixed.
> 
> Closes #4933
> 
> Co-authored-by: Mergen Imeev <imeevma@gmail.com>
> Co-authored-by: Timur Safin <tsafin@tarantool.org>
> ---
> https://github.com/tarantool/tarantool/issues/4933
> https://github.com/tarantool/tarantool/tree/imeevma/gh-4933-autoindex
> 
> @ChangeLog
>  - "autoindex" optimization is now enabled (gh-4933).

МНе кажется, можно чуть больше написать про такую важную штуку
 
>  src/box/CMakeLists.txt           |   2 +-
>  src/box/sql.c                    |   2 +-
>  src/box/sql/delete.c             |  28 ----
>  src/box/sql/sqlInt.h             |  35 -----
>  src/box/sql/where.c              | 238 +++++++++++++++++++++----------
>  src/box/sql/wherecode.c          |  13 +-
>  test/sql-tap/autoindex1.test.lua | 115 +++++++++++++++
>  test/sql-tap/autoindex4.test.lua |  14 +-
>  test/sql-tap/whereF.test.lua     |  16 ++-
>  test/sql/misc.result             |  35 +++++
>  test/sql/misc.test.lua           |  14 ++
>  11 files changed, 358 insertions(+), 154 deletions(-)
>  create mode 100755 test/sql-tap/autoindex1.test.lua
> 
> diff --git a/src/box/sql.c b/src/box/sql.c
> index 000246ee9..a551bffc3 100644
> --- a/src/box/sql.c
> +++ b/src/box/sql.c
> @@ -719,7 +719,7 @@ int
>  tarantoolsqlIdxKeyCompare(struct BtCursor *cursor,
>  			      struct UnpackedRecord *unpacked)
>  {
> -	assert(cursor->curFlags & BTCF_TaCursor);
> +	assert(cursor->curFlags & (BTCF_TaCursor | BTCF_TEphemCursor));
>  	assert(cursor->iter != NULL);
>  	assert(cursor->last_tuple != NULL);
>  
> diff --git a/src/box/sql/where.c b/src/box/sql/where.c
> index e9e936856..6ab2be3ba 100644
> --- a/src/box/sql/where.c
> +++ b/src/box/sql/where.c
> @@ -683,7 +683,6 @@ translateColumnToCopy(Vdbe * v,		/* The VDBE containing code to translate */
>  	}
> +/**
> + * Generate a code that will create a tuple, which will be inserted in the
> + * autoindex. The created tuple consists of rowid and fields described in the
> + * index key description.
> + *
> + * @param parse Parsing context.
> + * @param key_def The index key description.
> + * @param cursor Cursor of space where we will get values for tuple.
> + * @param reg_out Register to which the created tuple will be placed.
> + * @param reg_eph Register holding pointer to autoindex.
> + *
> + * @return Return a register number which is the first in a block of registers
> + * that holds the elements of the created tuple. The block of registers has
> + * already been deallocated by the time this routine returns.

Не очень понял, зачем возвращать уже деаллоцированные регистры..

> + */
> +static int
> +vdbe_emit_create_autoindex_tuple(struct Parse *parse,
> +				 const struct key_def *key_def,
> +				 int cursor, int reg_out, int reg_eph)
> +{
> +	assert(reg_out != 0);
> +	struct Vdbe *v = parse->pVdbe;
> +	int col_cnt = key_def->part_count;
> +	int reg_base = sqlGetTempRange(parse, col_cnt + 1);
> +	for (int j = 0; j < col_cnt; j++) {
> +		uint32_t tabl_col = key_def->parts[j].fieldno;
> +		sqlVdbeAddOp3(v, OP_Column, cursor, tabl_col, reg_base + j);
> +	}
> +	sqlVdbeAddOp2(v, OP_NextIdEphemeral, reg_eph, reg_base + col_cnt);
> +	sqlVdbeAddOp3(v, OP_MakeRecord, reg_base, col_cnt + 1, reg_out);
> +	sqlReleaseTempRange(parse, reg_base, col_cnt + 1);
> +	return reg_base;
> +}
> +
>  /*
> - * Generate code to construct the Index object for an automatic index
> - * and to set up the WhereLevel object pLevel so that the code generator
> - * makes use of the automatic index.
> + * Generate code to construct the ephemeral space that contains all used in
> + * query fields of one of the tables. This ephemeral space will be known as an

Опять: какой таблицы..Очень запутывающий коммент имхо

> + * "autoindex". PK of this ephemeral space is a covering index. Also, this
> + * functions set up the WhereLevel object pLevel so that the code generator
> + * makes use of the auto-index.
>   */
>  static void
>  constructAutomaticIndex(Parse * pParse,			/* The parsing context */
>  		}
>   * Estimate the location of a particular key among all keys in an
> -#ifndef SQL_OMIT_AUTOMATIC_INDEX
>  	/* Automatic indexes */
> -	LogEst rSize = pTab->nRowLogEst;
> +	rSize = DEFAULT_TUPLE_LOG_COUNT;
> +	/* Increase cost of autoindex if number of tuples in space < ~10000. */

10k как-то маловато. В скулайте, если я правильно помню, какие-то подвижки
начинаются со 100к affected rows.

> +	if (!space->def->opts.is_view && sql_space_tuple_log_count(space) < 133)
> +		rSize += DEFAULT_TUPLE_LOG_COUNT;

Почему для представления это не работает?

>  	LogEst rLogSize = estLog(rSize);
>  	if (!pBuilder->pOrSet && /* Not pqart of an OR optimization */
>  	    (pWInfo->wctrlFlags & WHERE_OR_SUBCLAUSE) == 0 &&
>  	    (pWInfo->pParse->sql_flags & SQL_AutoIndex) != 0 &&
>  	    pSrc->pIBIndex == 0	/* Has no INDEXED BY clause */
>  	    && !pSrc->fg.notIndexed	/* Has no NOT INDEXED clause */
>  	    &&!pSrc->fg.isCorrelated	/* Not a correlated subquery */
>  	    && !pSrc->fg.isRecursive	/* Not a recursive common table expression. */
>  	    ) {
> @@ -2829,24 +2910,10 @@ tnt_error:
>  			if (termCanDriveIndex(pTerm, pSrc, 0)) {
>  				pNew->nEq = 1;
>  				pNew->nSkip = 0;
> -				pNew->pIndex = 0;
> +				pNew->index_def = NULL;
>  				pNew->nLTerm = 1;
>  				pNew->aLTerm[0] = pTerm;
> -				/* TUNING: One-time cost for computing the automatic index is
> -				 * estimated to be X*N*log2(N) where N is the number of rows in
> -				 * the table being indexed and where X is 7 (LogEst=28) for normal
> -				 * tables or 1.375 (LogEst=4) for views and subqueries.  The value
> -				 * of X is smaller for views and subqueries so that the query planner
> -				 * will be more aggressive about generating automatic indexes for
> -				 * those objects, since there is no opportunity to add schema
> -				 * indexes on subqueries and views.
> -				 */
> -				pNew->rSetup = rLogSize + rSize + 4;
> -				if (!pTab->def->opts.is_view &&
> -				    pTab->def->id == 0)
> -					pNew->rSetup += 24;
> -				if (pNew->rSetup < 0)
> -					pNew->rSetup = 0;
> +				pNew->rSetup = rLogSize + rSize;

Можешь объяснить, почему тут произошли изменения? То есть почему оптимизация
стала более агрессивной?

>  				/* TUNING: Each index lookup yields 20 rows in the table.  This
>  				 * is more than the usual guess of 10 rows, since we have no way
>  				 * of knowing how selective the index will ultimately be.  It would
> diff --git a/src/box/sql/wherecode.c b/src/box/sql/wherecode.c
> index 6d8768865..0d58b4701 100644
> --- a/src/box/sql/wherecode.c
> +++ b/src/box/sql/wherecode.c
> @@ -219,12 +219,12 @@ sqlWhereExplainOneScan(Parse * pParse,	/* Parse context */
>  
>  			assert(!(flags & WHERE_AUTO_INDEX)
>  			       || (flags & WHERE_IDX_ONLY));
> -			if (idx_def->iid == 0) {
> +			if ((flags & WHERE_AUTO_INDEX) != 0) {
> +				zFmt = "AUTOMATIC COVERING INDEX";

Я бы оставил просто automatic index

> +			} else if (idx_def->iid == 0) {
>  				if (isSearch) {
>  					zFmt = "PRIMARY KEY";
>  				}
> -			} else if (flags & WHERE_AUTO_INDEX) {
> -				zFmt = "AUTOMATIC COVERING INDEX";
>  			} else if (flags & WHERE_IDX_ONLY) {
>  				zFmt = "COVERING INDEX %s";
>  			} else {
> @@ -1047,8 +1048,6 @@ sqlWhereCodeOneLoopStart(WhereInfo * pWInfo,	/* Complete information about the W
>  			startEq = 0;
>  			start_constraints = 1;
>  		}
> -		struct index_def *idx_pk = space->index[0]->def;
> -		uint32_t pk_part_count = idx_pk->key_def->part_count;
>  		/*
>  		 * Tarantool's iterator over integer fields doesn't
>  		 * tolerate floating point values. Hence, if term
> @@ -1234,6 +1233,8 @@ sqlWhereCodeOneLoopStart(WhereInfo * pWInfo,	/* Complete information about the W
>  		if (omitTable) {
>  			/* pIdx is a covering index.  No need to access the main table. */
>  		}  else if (iCur != iIdxCur) {
> +			struct index_def *idx_pk = space->index[0]->def;
> +			uint32_t pk_part_count = idx_pk->key_def->part_count;

Этот дифф точно нужен для патча?

>  			int iKeyReg = sqlGetTempRange(pParse, pk_part_count);
>  			for (j = 0; j < (int) pk_part_count; j++) {
>  				k = idx_pk->key_def->parts[j].fieldno;
> diff --git a/test/sql-tap/autoindex1.test.lua b/test/sql-tap/autoindex1.test.lua
> new file mode 100755
> index 000000000..42b995dd3
> --- /dev/null
> +++ b/test/sql-tap/autoindex1.test.lua
> @@ -0,0 +1,115 @@
> +#!/usr/bin/env tarantool
> +test = require("sqltester")
> +test:plan(7)
> +
> +--  2010 April 07
> +
> +--  The author disclaims copyright to this source code.  In place of
> +--  a legal notice, here is a blessing:
> +
> +--     May you do good and not evil.
> +--     May you find forgiveness for yourself and forgive others.
> +--     May you share freely, never taking more than you give.
> +

Необязательно включать эти странные хидеры в наши тесты..

> +---------------------------------------------------------------------------
> +--  This file implements regression tests for sql library.  The focus of this
> +-- script is testing automatic index creation logic.
> +
> +test:execsql([[
> +    CREATE TABLE t1(a INT, b INT PRIMARY KEY);
> +    INSERT INTO t1 VALUES(1, 11);
> +    INSERT INTO t1 VALUES(2, 22);
> +    INSERT INTO t1 SELECT a + 2, b + 22 FROM t1;
> +    INSERT INTO t1 SELECT a + 4, b + 44 FROM t1;
> +    CREATE TABLE t2(c INT, d INT PRIMARY KEY);
> +]])
> +
> +for i = 1, 11000 do test:execsql("INSERT INTO t2 VALUES ("..i..", "..i..");") end

Это быстро работает? Может сделать тест long-run?
Было бы еще неплохо к каждой группе тестов коммент, поясняющий почему
с эфмереной табличкой будет быстрее.

> +test:do_execsql_test(
> +    "autoindex-1.1", [[
> +        EXPLAIN QUERY PLAN SELECT b, (SELECT d FROM t2 WHERE c = a) FROM t1;
> +    ]], {
> +        0,0,0,"SCAN TABLE T1 (~1048576 rows)",
> +        0,0,0,"EXECUTE CORRELATED SCALAR SUBQUERY 1",
> +        1,0,0,"SEARCH TABLE T2 USING AUTOMATIC COVERING INDEX (C=?) (~20 rows)"
> +    })
> +
> +result = test:execsql([[SELECT b, (SELECT d FROM t2 WHERE c = a) FROM t1;]])
> +
> +test:do_execsql_test(
> +    "autoindex-1.2", [[
> +        EXPLAIN QUERY PLAN SELECT b, d FROM t1 JOIN t2 ON a = c ORDER BY b;
> +    ]], {
> +        0,0,0,"SCAN TABLE T1 (~1048576 rows)",
> +        0,1,1,"SEARCH TABLE T2 USING AUTOMATIC COVERING INDEX (C=?) (~20 rows)"
> +    })
> +
> +test:do_execsql_test(
> +    "autoindex-1.3", [[
> +        SELECT b, d FROM t1 JOIN t2 ON a = c ORDER BY b;
> +    ]], result)
> +
> +test:do_execsql_test(
> +    "autoindex-1.4", [[
> +        EXPLAIN QUERY PLAN SELECT b, d FROM t1 CROSS JOIN t2 ON (c = a);
> +    ]], {
> +        0,0,0,"SCAN TABLE T1 (~1048576 rows)",
> +        0,1,1,"SEARCH TABLE T2 USING AUTOMATIC COVERING INDEX (C=?) (~20 rows)"
> +    })
> +
> +test:do_execsql_test(
> +    "autoindex-1.5", [[
> +        SELECT b, d FROM t1 CROSS JOIN t2 ON (c = a);
> +    ]], result)
> +
> +test:execsql([[
> +    CREATE TABLE t3(i INT PRIMARY KEY, a INT, b INT);
> +]])
> +
> +for i = 1, 11000 do test:execsql("INSERT INTO t3 VALUES ("..i..", "..i..", "..(i + 1)..");") end
> +
> +test:do_execsql_test(
> +    "autoindex-1.6", [[
> +        SELECT count(*)
> +          FROM t3 AS x1
> +          JOIN t3 AS x2 ON x2.a=x1.b
> +          JOIN t3 AS x3 ON x3.a=x2.b
> +          JOIN t3 AS x4 ON x4.a=x3.b
> +          JOIN t3 AS x5 ON x5.a=x4.b
> +          JOIN t3 AS x6 ON x6.a=x5.b
> +          JOIN t3 AS x7 ON x7.a=x6.b
> +          JOIN t3 AS x8 ON x8.a=x7.b
> +          JOIN t3 AS x9 ON x9.a=x8.b
> +          JOIN t3 AS x10 ON x10.a=x9.b;
> +    ]], {
> +        10991
> +    })
> +
> +test:do_execsql_test(
> +    "autoindex-1.7", [[
> +        EXPLAIN QUERY PLAN SELECT count(*)
> +          FROM t3 AS x1
> +          JOIN t3 AS x2 ON x2.a=x1.b
> +          JOIN t3 AS x3 ON x3.a=x2.b
> +          JOIN t3 AS x4 ON x4.a=x3.b
> +          JOIN t3 AS x5 ON x5.a=x4.b
> +          JOIN t3 AS x6 ON x6.a=x5.b
> +          JOIN t3 AS x7 ON x7.a=x6.b
> +          JOIN t3 AS x8 ON x8.a=x7.b
> +          JOIN t3 AS x9 ON x9.a=x8.b
> +          JOIN t3 AS x10 ON x10.a=x9.b;
> +    ]], {
> +        0,0,0,"SCAN TABLE T3 AS X1 (~1048576 rows)",
> +        0,1,1,"SEARCH TABLE T3 AS X2 USING AUTOMATIC COVERING INDEX (A=?) (~20 rows)",
> +        0,2,2,"SEARCH TABLE T3 AS X3 USING AUTOMATIC COVERING INDEX (A=?) (~20 rows)",
> +        0,3,3,"SEARCH TABLE T3 AS X4 USING AUTOMATIC COVERING INDEX (A=?) (~20 rows)",
> +        0,4,4,"SEARCH TABLE T3 AS X5 USING AUTOMATIC COVERING INDEX (A=?) (~20 rows)",
> +        0,5,5,"SEARCH TABLE T3 AS X6 USING AUTOMATIC COVERING INDEX (A=?) (~20 rows)",
> +        0,6,6,"SEARCH TABLE T3 AS X7 USING AUTOMATIC COVERING INDEX (A=?) (~20 rows)",
> +        0,7,7,"SEARCH TABLE T3 AS X8 USING AUTOMATIC COVERING INDEX (A=?) (~20 rows)",
> +        0,8,8,"SEARCH TABLE T3 AS X9 USING AUTOMATIC COVERING INDEX (A=?) (~20 rows)",
> +        0,9,9,"SEARCH TABLE T3 AS X10 USING AUTOMATIC COVERING INDEX (A=?) (~20 rows)"
> +    })
> +
> +test:finish_test()
> diff --git a/test/sql-tap/autoindex4.test.lua b/test/sql-tap/autoindex4.test.lua
> index 46488f13a..e22f49c67 100755
> --- a/test/sql-tap/autoindex4.test.lua
> +++ b/test/sql-tap/autoindex4.test.lua
> @@ -1,6 +1,6 @@
>  #!/usr/bin/env tarantool
>  test = require("sqltester")
> -test:plan(7)
> +test:plan(8)
>  
>  --!./tcltestrunner.lua
>  -- 2014-10-24
> @@ -75,6 +75,18 @@ test:do_execsql_test(
>          -- </autoindex4-1.4>
>      })
>  
> +test:do_execsql_test(
> +    "autoindex4-2.0",
> +    [[
> +        CREATE TABLE t3(i INT PRIMARY KEY, e INT, f INT);
> +        INSERT INTO t3 VALUES(1, 123,654), (2, 555,444), (3, 234,987);
> +        SELECT (SELECT count(*) FROM t1, t2 WHERE a=e AND x=f), e, f, '|' FROM t3 ORDER BY i;
> +    ]], {
> +        -- <autoindex4-2.0>
> +        1, 123, 654, "|", 0, 555, 444, "|", 4, 234, 987, "|"
> +        -- </autoindex4-2.0>
> +    })
> +
>  -- do_execsql_test autoindex4-2.0 {
>  --   CREATE TABLE t3(e INT,f INT);
>  --   INSERT INTO t3 VALUES(123,654),(555,444),(234,987);
> diff --git a/test/sql-tap/whereF.test.lua b/test/sql-tap/whereF.test.lua
> index 5a894b748..3235df437 100755
> --- a/test/sql-tap/whereF.test.lua
> +++ b/test/sql-tap/whereF.test.lua
> @@ -90,10 +90,20 @@ test:do_execsql_test(
>  
>  -- for _ in X(0, "X!foreach", [=[["tn sql","\n  1 \"SELECT * FROM t1,           t2 WHERE t1.a>? AND t2.d>t1.c AND t1.b=t2.e\"\n  2 \"SELECT * FROM t2,           t1 WHERE t1.a>? AND t2.d>t1.c AND t1.b=t2.e\"\n  3 \"SELECT * FROM t2 CROSS JOIN t1 WHERE t1.a>? AND t2.d>t1.c AND t1.b=t2.e\"\n"]]=]) do
>  for tn, sql in ipairs({"SELECT * FROM t1,           t2 WHERE t1.a>? AND t2.d>t1.c AND t1.b=t2.e",
> -                       "SELECT * FROM t2,           t1 WHERE t1.a>? AND t2.d>t1.c AND t1.b=t2.e",
> -                       "SELECT * FROM t2 CROSS JOIN t1 WHERE t1.a>? AND t2.d>t1.c AND t1.b=t2.e"}) do
> +                       "SELECT * FROM t2,           t1 WHERE t1.a>? AND t2.d>t1.c AND t1.b=t2.e"}) do
>      test:do_test(
> -        "2."..tn,
> +        "2.1."..tn,
> +        function()
> +            return test:execsql("EXPLAIN QUERY PLAN "..sql)
> +        end, {
> +            '/SEARCH TABLE T1/',
> +            '/SEARCH TABLE T2/'
> +        })
> +end
> +
> +for tn, sql in ipairs({"SELECT * FROM t2 CROSS JOIN t1 WHERE t1.a>? AND t2.d>t1.c AND t1.b=t2.e"}) do
> +    test:do_test(
> +        "2.2."..tn,
>          function()
>              return test:execsql("EXPLAIN QUERY PLAN "..sql)
>          end, {
> diff --git a/test/sql/misc.result b/test/sql/misc.result
> index 6af11bfba..b52a83bda 100644
> --- a/test/sql/misc.result
> +++ b/test/sql/misc.result
> @@ -204,3 +204,38 @@ box.execute('SELECT field66, field68, field70 FROM test')
>  box.space.TEST:drop()
>  ---
>  ...
> +-- gh-4933: Make sure that autoindex optimization is used.
> +box.execute('CREATE TABLE t1(i INT PRIMARY KEY, a INT);')
> +---
> +- row_count: 1
> +...
> +box.execute('CREATE TABLE t2(i INT PRIMARY KEY, b INT);')
> +---
> +- row_count: 1
> +...
> +for i = 1, 11000 do\
> +	box.execute('INSERT INTO t1 VALUES ($1, $1);', {i})\
> +	box.execute('INSERT INTO t2 VALUES ($1, $1);', {i})\
> +end
> +---
> +...
> +--
> +-- There is no need to insert values in the tables since planner assumes a
> +-- default number of tuples for each space, regardless of how many tuples there
> +-- actually are in those spaces. The default value is 1048576 (== 2^20).
> +--
> +box.execute('EXPLAIN QUERY PLAN SELECT a, b FROM t1, t2 WHERE a = b;')
> +---
> +- metadata:
> +  - name: selectid
> +    type: integer
> +  - name: order
> +    type: integer
> +  - name: from
> +    type: integer
> +  - name: detail
> +    type: text
> +  rows:
> +  - [0, 0, 0, 'SCAN TABLE T1 (~1048576 rows)']
> +  - [0, 1, 1, 'SEARCH TABLE T2 USING AUTOMATIC COVERING INDEX (B=?) (~20 rows)']
> +...
> diff --git a/test/sql/misc.test.lua b/test/sql/misc.test.lua
> index 63244228c..d7673bb49 100644
> --- a/test/sql/misc.test.lua
> +++ b/test/sql/misc.test.lua
> @@ -64,3 +64,17 @@ box.execute('SELECT field70, field64 FROM test')
>  pk:alter({parts = {66}})
>  box.execute('SELECT field66, field68, field70 FROM test')
>  box.space.TEST:drop()
> +
> +-- gh-4933: Make sure that autoindex optimization is used.
> +box.execute('CREATE TABLE t1(i INT PRIMARY KEY, a INT);')
> +box.execute('CREATE TABLE t2(i INT PRIMARY KEY, b INT);')
> +for i = 1, 11000 do\
> +	box.execute('INSERT INTO t1 VALUES ($1, $1);', {i})\
> +	box.execute('INSERT INTO t2 VALUES ($1, $1);', {i})\
> +end
> +--
> +-- There is no need to insert values in the tables since planner assumes a
> +-- default number of tuples for each space, regardless of how many tuples there
> +-- actually are in those spaces. The default value is 1048576 (== 2^20).
> +--
> +box.execute('EXPLAIN QUERY PLAN SELECT a, b FROM t1, t2 WHERE a = b;')
> -- 
> 2.25.1
>