Tarantool development patches archive
 help / color / mirror / Atom feed
From: "n.pettik" <korablev@tarantool.org>
To: tarantool-patches@freelists.org
Cc: Kirill Yukhin <kyukhin@tarantool.org>
Subject: [tarantool-patches] Re: [PATCH 1/2] sql: after table rename properly update indexes
Date: Tue, 21 Aug 2018 13:26:38 +0300	[thread overview]
Message-ID: <8A23ACEC-DCAA-47A2-8311-F46D75C9F9C0@tarantool.org> (raw)
In-Reply-To: <cf29cff0e305973e0b01c89bdb97d2feb4b4d9f2.1534783275.git.kyukhin@tarantool.org>


> diff --git a/src/box/sql.c b/src/box/sql.c
> index ae12cae..9485899 100644
> --- a/src/box/sql.c
> +++ b/src/box/sql.c
> @@ -840,6 +840,46 @@ rename_fail:
> 	return SQL_TARANTOOL_ERROR;
> }
> 
> +int
> +sql_update_index_table_name(struct index_def *def, const char *new_tbl_name,
> +			    char **sql_stmt)

I guess name should be like (according to naming convention):
sql_index_update_table_name()..

> +{
> +	assert(new_tbl_name != NULL);
> +	uint32_t iid = def->iid;

Do you really need this shortcut for single (or two) usage(s)?

> +
> +	bool is_quoted = false;
> +	*sql_stmt = rename_table(db, def->opts.sql, new_tbl_name, &is_quoted);

You need here check: rename_table() may fail and return NULL.

> +
> +	uint32_t key_len = mp_sizeof_uint(def->space_id) + mp_sizeof_uint(iid) +
> +			   mp_sizeof_array(2);
> +	uint32_t new_opts_sz = tarantoolSqlite3MakeIdxOpts(def->opts.is_unique,
> +							   *sql_stmt, NULL);
> +	uint32_t op_sz = mp_sizeof_array(1) + mp_sizeof_array(3) +
> +			 mp_sizeof_str(1) + mp_sizeof_uint(4) + new_opts_sz;
> +
> +	char *key_begin = (char*) region_alloc(&fiber()->gc, key_len + op_sz);
> +	if (key_begin == NULL) {
> +		diag_set(OutOfMemory, key_len, "region_alloc", "key_begin”);

In diag_set you pass key_len, but allocate key_len + op_sz.

> +		return SQL_TARANTOOL_ERROR;
> +	}
> +	char *key = mp_encode_array(key_begin, 2);
> +	key = mp_encode_uint(key, def->space_id);
> +	key = mp_encode_uint(key, iid);
> +
> +	char *op_begin = key;
> +	char *op = mp_encode_array(op_begin, 1);
> +	op = mp_encode_array(op, 3);
> +	op = mp_encode_str(op, "=", 1);
> +	op = mp_encode_uint(op, 4);

Mb it would be better to use BOX_INDEX_FIELD_OPTS?
At least it would be easier to understand what happen here.

> +	op += tarantoolSqlite3MakeIdxOpts(def->opts.is_unique, *sql_stmt, op);
> +
> +	if (box_update(BOX_INDEX_ID, 0, key_begin, key, op_begin, op,
> +		       0, NULL) != 0)
> +		return SQL_TARANTOOL_ERROR;
> +
> +	return SQLITE_OK;

AFAIR we decided to avoid using SQLITE_OK macros.
Lets use instead simple return 0/-1.

> +}
> +
> int
> tarantoolSqlite3IdxKeyCompare(struct BtCursor *cursor,
> 			      struct UnpackedRecord *unpacked)
> @@ -1484,23 +1524,12 @@ int tarantoolSqlite3MakeIdxParts(SqliteIndex *pIndex, void *buf)
> 	return p - base;
> }
> 
> -/*
> - * Format "opts" dictionary for _index entry.
> - * Returns result size.
> - * If buf==NULL estimate result size.
> - *
> - * Ex: {
> - *   "unique": "true",
> - *   "sql": "CREATE INDEX student_by_name ON students(name)"
> - * }
> - */
> -int tarantoolSqlite3MakeIdxOpts(SqliteIndex *index, const char *zSql, void *buf)
> +int
> +tarantoolSqlite3MakeIdxOpts(bool is_unique, const char *zSql, void *buf)
> {

You can pass here struct index_def (or struct index_opts)
instead of two args:
is_unique ---> def->opst.is_unique and zSql ---> def->opts.sql.

I see that you pass new zSql, but you still can reassign def->opts.sql.
It is up to you, thou.

> 	const struct Enc *enc = get_enc(buf);
> 	char *base = buf, *p;
> 
> -	(void)index;
> -
> 	p = enc->encode_map(base, 2);
> 	/* Mark as unique pk and unique indexes */
> 	p = enc->encode_str(p, "unique", 6);
> @@ -1511,7 +1540,7 @@ int tarantoolSqlite3MakeIdxOpts(SqliteIndex *index, const char *zSql, void *buf)
> 	 * INSERT OR REPLACE/IGNORE uniqueness checks will be also done by
> 	 * Tarantool.
> 	 */
> -	p = enc->encode_bool(p, IsUniqueIndex(index));
> +	p = enc->encode_bool(p, is_unique);
> 	p = enc->encode_str(p, "sql", 3);
> 	p = enc->encode_str(p, zSql, zSql ? strlen(zSql) : 0);
> 	return (int)(p - base);
> diff --git a/src/box/sql/build.c b/src/box/sql/build.c
> index dddeb12..05b729d 100644
> --- a/src/box/sql/build.c
> +++ b/src/box/sql/build.c
> @@ -1246,13 +1246,14 @@ createIndex(Parse * pParse, Index * pIndex, int iSpaceId, int iIndexId,
> 
> 	/* Format "opts" and "parts" for _index entry. */
> 	zOpts = sqlite3DbMallocRaw(pParse->db,
> -				   tarantoolSqlite3MakeIdxOpts(pIndex, zSql,
> -							       NULL) +
> +				   tarantoolSqlite3MakeIdxOpts(IsUniqueIndex(pIndex),
> +							       zSql, NULL) +
> 				   tarantoolSqlite3MakeIdxParts(pIndex,
> 								NULL) + 2);
> 	if (!zOpts)
> 		return;
> -	zOptsSz = tarantoolSqlite3MakeIdxOpts(pIndex, zSql, zOpts);
> +	zOptsSz = tarantoolSqlite3MakeIdxOpts(IsUniqueIndex(pIndex), zSql,
> +					      zOpts);
> 	zParts = zOpts + zOptsSz + 1;
> 	zPartsSz = tarantoolSqlite3MakeIdxParts(pIndex, zParts);
> #if SQLITE_DEBUG
> diff --git a/src/box/sql/tarantoolInt.h b/src/box/sql/tarantoolInt.h
> index 94517f6..536b0e3 100644
> --- a/src/box/sql/tarantoolInt.h
> +++ b/src/box/sql/tarantoolInt.h
> @@ -89,6 +89,20 @@ int tarantoolSqlite3ClearTable(struct space *space);
> int
> sql_rename_table(uint32_t space_id, const char *new_name, char **sql_stmt);
> 
> +/**
> + * Update CREATE INDEX field (def->opt.sql) replacing table name
> + * w/ new one in _index space.
> + *
> + * @param idef Index definition.
> + * @param new_tbl_name new name of table

Nitpicking: put dot at the end of sentence and start sentence with capital letter.

> + * @param[out] sql_stmt New CREATE INDEX statement.
> + *
> + * @retval SQLITE_OK on success, SQLITE_TARANTOOL_ERROR otherwise.
> + */
> +int
> +sql_update_index_table_name(struct index_def *idef, const char *new_tbl_name,
> +			    char **sql_stmt);
> +
> /* Alter trigger statement after rename table. */
> int tarantoolSqlite3RenameTrigger(const char *zTriggerName,
> 				  const char *zOldName, const char *zNewName);
> @@ -171,12 +185,17 @@ fkey_encode_links(const struct fkey_def *def, int type, char *buf);
>  */
> int tarantoolSqlite3MakeIdxParts(Index * index, void *buf);
> 
> -/*
> +/**
>  * Format "opts" dictionary for _index entry.
>  * Returns result size.
>  * If buf==NULL estimate result size.
> + *
> + * @param is_unique Flag if index is unique.
> + * @param zSql SQL create stmt.
> + * @param[out] buf destination buffer.

You didn’t specify @retval for this function.

> /**
>  * Extract next id from _sequence space.
> diff --git a/src/box/sql/vdbe.c b/src/box/sql/vdbe.c
> index dc5146f..4516ee9 100644
> --- a/src/box/sql/vdbe.c
> +++ b/src/box/sql/vdbe.c
> @@ -4625,13 +4625,35 @@ case OP_RenameTable: {
> 	db->init.busy = 1;
> 	init.rc = SQLITE_OK;
> 	sql_init_callback(&init, zNewTableName, space_id, 0, zSqlStmt);
> -	db->init.busy = 0;
> 	rc = init.rc;
> -	if (rc) {
> +	if (rc != SQLITE_OK) {
> 		sqlite3CommitInternalChanges();
> +		db->init.busy = 0;
> 		goto abort_due_to_error;
> 	}
> 
> +	/* Space was altered, refetch the pointer. */
> +	space = space_by_id(space_id);
> +	for (uint32_t i = 0; i < space->index_count; ++i) {
> +		struct index_def *def = space->index[i]->def;
> +		if (def->opts.sql == NULL)
> +			continue;
> +		char *sql_stmt;
> +		rc = sql_update_index_table_name(def, zNewTableName, &sql_stmt);
> +		if (rc != SQLITE_OK)
> +			goto abort_due_to_error;

Again the same problem with RENAME: in case of fail part of indexes would
remain with old ‘create index …’ statement and after server’s restart it is likely
to result in segfault/assertion.

> +		space = space_by_id(space_id);

Does opts.sql update can lead to renewal of space ptr?
If so, it looks quite weird: string with ‘CREATE INDEX …’ statement
seems to be sort of static attribute which may be useful only during
instance restart...

> +		sql_init_callback(&init, zNewTableName, space_id,
> +				  space->index[i]->def->iid, sql_stmt);
> +		rc = init.rc;
> +		if (rc != SQLITE_OK) {
> +			sqlite3CommitInternalChanges();
> +			db->init.busy = 0;
> +			goto abort_due_to_error;
> +		}
> +	}
> +	db->init.busy = 0;
> +
> 	/*
> 	 * Rebuild 'CREATE TRIGGER' expressions of all triggers
> 	 * created on this table. Sure, this action is not atomic
> 
> diff --git a/test/sql/gh-3613-idx-alter-update.test.lua b/test/sql/gh-3613-idx-alter-update.test.lua
> new file mode 100644
> index 0000000..287096b
> --- /dev/null
> +++ b/test/sql/gh-3613-idx-alter-update.test.lua
> @@ -0,0 +1,21 @@
> +test_run = require('test_run').new()
> +engine = test_run:get_cfg('engine')
> +box.sql.execute('pragma sql_default_engine=\''..engine..'\'')
> +
> +box.sql.execute('CREATE TABLE t (s1 INT PRIMARY KEY)')
> +box.sql.execute('CREATE INDEX i ON t (s1)')
> +box.sql.execute('ALTER TABLE t RENAME TO j3')
> +
> +-- Due to gh-3613, next stmt caused segfault
> +box.sql.execute('DROP INDEX i ON j3')
> +
> +box.sql.execute('CREATE INDEX i ON j3 (s1)’)

Well, actually I would also add test when instance is restarted
right after dropping to make sure that no artefacts are left after rename.
Recreation may be misleading in this case.

  reply	other threads:[~2018-08-21 10:26 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-08-20 16:49 [tarantool-patches] [PATCH 0/2] sql: update inexes after table rename Kirill Yukhin
2018-08-20 16:49 ` [tarantool-patches] [PATCH 1/2] sql: after table rename properly update indexes Kirill Yukhin
2018-08-21 10:26   ` n.pettik [this message]
2018-08-21 12:20     ` [tarantool-patches] " Kirill Yukhin
2018-08-21 20:38       ` n.pettik
2018-08-22  6:39         ` Kirill Yukhin
2018-08-20 16:49 ` [tarantool-patches] [PATCH 2/2] sql: take sql field in index_opts_cmp Kirill Yukhin
2018-08-21 10:26   ` [tarantool-patches] " n.pettik
2018-08-21 12:30     ` Kirill Yukhin
2018-08-21 20:41       ` n.pettik
2018-08-22  6:47 ` [tarantool-patches] Re: [PATCH 0/2] sql: update inexes after table rename Kirill Yukhin

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=8A23ACEC-DCAA-47A2-8311-F46D75C9F9C0@tarantool.org \
    --to=korablev@tarantool.org \
    --cc=kyukhin@tarantool.org \
    --cc=tarantool-patches@freelists.org \
    --subject='[tarantool-patches] Re: [PATCH 1/2] sql: after table rename properly update indexes' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox