From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from smtp29.i.mail.ru (smtp29.i.mail.ru [94.100.177.89]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by dev.tarantool.org (Postfix) with ESMTPS id BB7BF4696C3 for ; Thu, 16 Apr 2020 11:33:10 +0300 (MSK) Date: Thu, 16 Apr 2020 11:33:08 +0300 From: Mergen Imeev Message-ID: <20200416083307.GA4278@tarantool.org> References: <2542c80348de25f26cac73e101b394a1873156da.1586933931.git.imeevma@gmail.com> <20200416000957.GB2581@tarantool.org> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline In-Reply-To: <20200416000957.GB2581@tarantool.org> Subject: Re: [Tarantool-patches] [PATCH v3 3/3] sql: do not change order of inserted values List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Nikita Pettik Cc: tarantool-patches@dev.tarantool.org Thank you for review! My answer, diff and new patch below. On Thu, Apr 16, 2020 at 12:09:57AM +0000, Nikita Pettik wrote: > On 15 Apr 10:09, imeevma@tarantool.org wrote: > > My answers and new patch below. > > > > On 4/14/20 1:59 AM, Nikita Pettik wrote: > > > > diff --git a/src/box/sql/insert.c b/src/box/sql/insert.c > > index 43a0de5..787855e 100644 > > --- a/src/box/sql/insert.c > > +++ b/src/box/sql/insert.c > > @@ -455,8 +455,31 @@ sqlInsert(Parse * pParse, /* Parser context */ > > reg_eph = ++pParse->nMem; > > regRec = sqlGetTempReg(pParse); > > regCopy = sqlGetTempRange(pParse, nColumn + 1); > > - sqlVdbeAddOp2(v, OP_OpenTEphemeral, reg_eph, > > - nColumn + 1); > > + /* > > + * This key_info is used to show that > > + * rowid should be the first part of PK in > > + * case we used AUTOINCREMENT feature. > > + * This way we will save initial order of > > + * the inserted values. The order is > > + * important if we use the AUTOINCREMENT > > + * feature, since changing the order can > > + * change the number inserted instead of > > + * NULL. > > + */ > > + if (space->sequence != NULL) { > > + struct sql_key_info *key_info = > > + sql_key_info_new(pParse->db, > > + nColumn + 1); > > + key_info->parts[nColumn].type = > > + FIELD_TYPE_UNSIGNED; > > + key_info->is_pk_rowid = true; > > + sqlVdbeAddOp4(v, OP_OpenTEphemeral, reg_eph, > > + nColumn + 1, 0, (char *)key_info, > > + P4_KEYINFO); > > + } else { > > + sqlVdbeAddOp2(v, OP_OpenTEphemeral, reg_eph, > > + nColumn + 1); > > + } > > You can remove 'else' branch and always add OP_OPenTEphemeral with > 2 operands. Then inside 'if' branch invoke sqlVdbeChangeP4(). > Otherwise LGTM. > Fixed. Diff: diff --git a/src/box/sql/insert.c b/src/box/sql/insert.c index 787855e..588e142 100644 --- a/src/box/sql/insert.c +++ b/src/box/sql/insert.c @@ -455,6 +455,8 @@ sqlInsert(Parse * pParse, /* Parser context */ reg_eph = ++pParse->nMem; regRec = sqlGetTempReg(pParse); regCopy = sqlGetTempRange(pParse, nColumn + 1); + sqlVdbeAddOp2(v, OP_OpenTEphemeral, reg_eph, + nColumn + 1); /* * This key_info is used to show that * rowid should be the first part of PK in @@ -473,12 +475,8 @@ sqlInsert(Parse * pParse, /* Parser context */ key_info->parts[nColumn].type = FIELD_TYPE_UNSIGNED; key_info->is_pk_rowid = true; - sqlVdbeAddOp4(v, OP_OpenTEphemeral, reg_eph, - nColumn + 1, 0, (char *)key_info, - P4_KEYINFO); - } else { - sqlVdbeAddOp2(v, OP_OpenTEphemeral, reg_eph, - nColumn + 1); + sqlVdbeChangeP4(v, -1, (void *)key_info, + P4_KEYINFO); } addrL = sqlVdbeAddOp1(v, OP_Yield, dest.iSDParm); VdbeCoverage(v); New patch: >From 167e3e17005b859d0fb2fc9ae09d5a251f54f9f0 Mon Sep 17 00:00:00 2001 From: Mergen Imeev Date: Thu, 2 Apr 2020 11:15:24 +0300 Subject: [PATCH] sql: do not change order of inserted values Before this patch, if an ephemeral space was used during INSERT or REPLACE, the inserted values were sorted by the first column, since this was the first part of the index. This can lead to an error when using the AUTOINCREMENT feature, since changing the order of the inserted value can change the value inserted instead of NULL. To avoid this, the patch makes the rowid of the inserted row in the ephemeral space the only part of the ephemeral space index. Closes #4256 diff --git a/src/box/sql.c b/src/box/sql.c index 6436f64..fc1386f 100644 --- a/src/box/sql.c +++ b/src/box/sql.c @@ -318,10 +318,17 @@ struct space * sql_ephemeral_space_create(uint32_t field_count, struct sql_key_info *key_info) { struct key_def *def = NULL; + uint32_t part_count = field_count; if (key_info != NULL) { def = sql_key_info_to_key_def(key_info); if (def == NULL) return NULL; + /* + * In case is_pk_rowid is true we can use rowid + * as the only part of the key. + */ + if (key_info->is_pk_rowid) + part_count = 1; } struct region *region = &fiber()->gc; @@ -332,8 +339,8 @@ sql_ephemeral_space_create(uint32_t field_count, struct sql_key_info *key_info) * turned to string, which is 10 and plus 1 for \0. */ uint32_t name_len = strlen("_COLUMN_") + 11; - uint32_t size = field_count * (sizeof(struct field_def) + name_len + - sizeof(struct key_part_def)); + uint32_t size = field_count * (sizeof(struct field_def) + name_len) + + part_count * sizeof(struct key_part_def); struct field_def *fields = region_alloc(region, size); if (fields == NULL) { diag_set(OutOfMemory, size, "region_alloc", "fields"); @@ -342,7 +349,7 @@ sql_ephemeral_space_create(uint32_t field_count, struct sql_key_info *key_info) struct key_part_def *ephemer_key_parts = (void *)fields + field_count * sizeof(struct field_def); char *names = (char *)ephemer_key_parts + - field_count * sizeof(struct key_part_def); + part_count * sizeof(struct key_part_def); for (uint32_t i = 0; i < field_count; ++i) { struct field_def *field = &fields[i]; field->name = names; @@ -362,18 +369,27 @@ sql_ephemeral_space_create(uint32_t field_count, struct sql_key_info *key_info) } } - for (uint32_t i = 0; i < field_count; ++i) { + for (uint32_t i = 0; i < part_count; ++i) { struct key_part_def *part = &ephemer_key_parts[i]; - part->fieldno = i; + /* + * In case we need to save initial order of + * inserted into ephemeral space rows we use rowid + * as the only part of PK. If ephemeral space has + * a rowid, it is always the last column. + */ + uint32_t j = i; + if (key_info != NULL && key_info->is_pk_rowid) + j = field_count - 1; + part->fieldno = j; part->nullable_action = ON_CONFLICT_ACTION_NONE; part->is_nullable = true; part->sort_order = SORT_ORDER_ASC; part->path = NULL; - part->type = fields[i].type; - part->coll_id = fields[i].coll_id; + part->type = fields[j].type; + part->coll_id = fields[j].coll_id; } struct key_def *ephemer_key_def = key_def_new(ephemer_key_parts, - field_count, false); + part_count, false); if (ephemer_key_def == NULL) return NULL; diff --git a/src/box/sql/insert.c b/src/box/sql/insert.c index 43a0de5..588e142 100644 --- a/src/box/sql/insert.c +++ b/src/box/sql/insert.c @@ -457,6 +457,27 @@ sqlInsert(Parse * pParse, /* Parser context */ regCopy = sqlGetTempRange(pParse, nColumn + 1); sqlVdbeAddOp2(v, OP_OpenTEphemeral, reg_eph, nColumn + 1); + /* + * This key_info is used to show that + * rowid should be the first part of PK in + * case we used AUTOINCREMENT feature. + * This way we will save initial order of + * the inserted values. The order is + * important if we use the AUTOINCREMENT + * feature, since changing the order can + * change the number inserted instead of + * NULL. + */ + if (space->sequence != NULL) { + struct sql_key_info *key_info = + sql_key_info_new(pParse->db, + nColumn + 1); + key_info->parts[nColumn].type = + FIELD_TYPE_UNSIGNED; + key_info->is_pk_rowid = true; + sqlVdbeChangeP4(v, -1, (void *)key_info, + P4_KEYINFO); + } addrL = sqlVdbeAddOp1(v, OP_Yield, dest.iSDParm); VdbeCoverage(v); sqlVdbeAddOp2(v, OP_NextIdEphemeral, reg_eph, diff --git a/src/box/sql/select.c b/src/box/sql/select.c index 65e41f2..f394840 100644 --- a/src/box/sql/select.c +++ b/src/box/sql/select.c @@ -1422,6 +1422,7 @@ sql_key_info_new(sql *db, uint32_t part_count) key_info->key_def = NULL; key_info->refs = 1; key_info->part_count = part_count; + key_info->is_pk_rowid = false; for (uint32_t i = 0; i < part_count; i++) { struct key_part_def *part = &key_info->parts[i]; part->fieldno = i; @@ -1448,6 +1449,7 @@ sql_key_info_new_from_key_def(sql *db, const struct key_def *key_def) key_info->key_def = NULL; key_info->refs = 1; key_info->part_count = key_def->part_count; + key_info->is_pk_rowid = false; key_def_dump_parts(key_def, key_info->parts, NULL); return key_info; } diff --git a/src/box/sql/sqlInt.h b/src/box/sql/sqlInt.h index e45a767..37283e5 100644 --- a/src/box/sql/sqlInt.h +++ b/src/box/sql/sqlInt.h @@ -4118,6 +4118,8 @@ struct sql_key_info { struct key_def *key_def; /** Reference counter. */ uint32_t refs; + /** Rowid should be the only part of PK, if true. */ + bool is_pk_rowid; /** Number of parts in the key. */ uint32_t part_count; /** Definition of the key parts. */ diff --git a/src/box/sql/vdbe.c b/src/box/sql/vdbe.c index 5ab9779..724bc18 100644 --- a/src/box/sql/vdbe.c +++ b/src/box/sql/vdbe.c @@ -2702,23 +2702,10 @@ case OP_Column: { pC->cacheStatus = p->cacheCtr; } enum field_type field_type = field_type_MAX; - if (pC->eCurType == CURTYPE_TARANTOOL) { - /* - * Ephemeral spaces feature only one index - * covering all fields, but ephemeral spaces - * lack format. So, we can fetch type from - * key parts. - */ - if (pC->uc.pCursor->curFlags & BTCF_TEphemCursor) { - field_type = pC->uc.pCursor->index->def-> - key_def->parts[p2].type; - } else if (pC->uc.pCursor->curFlags & BTCF_TaCursor) { - field_type = pC->uc.pCursor->space->def-> - fields[p2].type; - } - } else if (pC->eCurType == CURTYPE_SORTER) { + if (pC->eCurType == CURTYPE_TARANTOOL) + field_type = pC->uc.pCursor->space->def->fields[p2].type; + else if (pC->eCurType == CURTYPE_SORTER) field_type = vdbe_sorter_get_field_type(pC->uc.pSorter, p2); - } struct Mem *default_val_mem = pOp->p4type == P4_MEM ? pOp->p4.pMem : NULL; if (vdbe_field_ref_fetch(&pC->field_ref, p2, pDest) != 0) diff --git a/test/sql/gh-4256-do-not-change-order-during-insertion.result b/test/sql/gh-4256-do-not-change-order-during-insertion.result new file mode 100644 index 0000000..8a585b3 --- /dev/null +++ b/test/sql/gh-4256-do-not-change-order-during-insertion.result @@ -0,0 +1,44 @@ +-- test-run result file version 2 +-- +-- Make sure that when inserting, values are inserted in the given +-- order when ephemeral space is used. +-- +CREATE TABLE t (i INT PRIMARY KEY AUTOINCREMENT); + | --- + | - row_count: 1 + | ... +-- +-- In order for this INSERT to use the ephemeral space, we created +-- this trigger. +-- +CREATE TRIGGER r AFTER INSERT ON t FOR EACH ROW BEGIN SELECT 1; END + | --- + | - row_count: 1 + | ... +INSERT INTO t VALUES (1), (NULL), (10), (NULL), (NULL), (3), (NULL); + | --- + | - autoincrement_ids: + | - 2 + | - 11 + | - 12 + | - 13 + | row_count: 7 + | ... +SELECT * FROM t; + | --- + | - metadata: + | - name: I + | type: integer + | rows: + | - [1] + | - [2] + | - [3] + | - [10] + | - [11] + | - [12] + | - [13] + | ... +DROP TABLE t; + | --- + | - row_count: 1 + | ... diff --git a/test/sql/gh-4256-do-not-change-order-during-insertion.test.sql b/test/sql/gh-4256-do-not-change-order-during-insertion.test.sql new file mode 100644 index 0000000..9b5afa0 --- /dev/null +++ b/test/sql/gh-4256-do-not-change-order-during-insertion.test.sql @@ -0,0 +1,13 @@ +-- +-- Make sure that when inserting, values are inserted in the given +-- order when ephemeral space is used. +-- +CREATE TABLE t (i INT PRIMARY KEY AUTOINCREMENT); +-- +-- In order for this INSERT to use the ephemeral space, we created +-- this trigger. +-- +CREATE TRIGGER r AFTER INSERT ON t FOR EACH ROW BEGIN SELECT 1; END +INSERT INTO t VALUES (1), (NULL), (10), (NULL), (NULL), (3), (NULL); +SELECT * FROM t; +DROP TABLE t;