a real table can be created even read_only = 1

[yinfeng]

this bug has been reported here: http://bugs.mysql.com/bug.php?id=64992

I've written my analyses there (user: zhai weixiang) based on percona5.5.18, here just copy it

test case:

xx@test 09:59:54>show variables like 'read_only%';
+---------------+-------+
| Variable_name | Value |
+---------------+-------+
| read_only | ON |
+---------------+-------+
1 row in set (0.00 sec)

xx@test 09:59:58>select @@version;
+------------+
| @@version |
+------------+
| 5.5.22-log |
+------------+
1 row in set (0.00 sec)

xx@test 10:00:22>create table t(foo varchar(12), bar varchar(24));
ERROR 1290 (HY000): The MySQL server is running with the --read-only option so it cannot execute this statement
xx@test 10:02:38>create temporary table t(id bigint unsigned not null primary key);
Query OK, 0 rows affected (0.04 sec)

xx@test 10:02:43>create table t(foo varchar(12), bar varchar(24));
Query OK, 0 rows affected (0.00 sec)

------------------------

deny_updates_if_read_only_option() was used to check if a query was allowed to excute. if we first create a temporary table and then create a real table, the later operation will be allowed.

i guess there must be some problem with the following code in deny_updates_if_read_only_option()

  const my_bool update_real_tables=
    some_non_temp_table_to_be_updated(thd, all_tables) &&
    !(create_temp_tables || drop_temp_tables);

Generally some_non_temp_table_to_be_updated() should return true and create_temp_tables equals zero while we are creating a real table.

but actully the return value of some_non_temp_table_to_be_updated() is 0, because the same table name was found in temporary table list.

function some_non_temp_table_to_be_updated:

static bool some_non_temp_table_to_be_updated(THD *thd, TABLE_LIST *tables)
{
  for (TABLE_LIST *table= tables; table; table= table->next_global)
  {
    DBUG_ASSERT(table->db && table->table_name);
    if (table->updating && !find_temporary_table(thd, table))
      return 1;
  }
  return 0;
}

so i think the value of update_real_tables should be calculated like this :

  const my_bool update_real_tables=
    (lex->sql_command == SQLCOM_CREATE_TABLE || some_non_temp_table_to_be_updated(thd, all_tables)) &&
    !(create_temp_tables || drop_temp_tables);

I don't know my guess is correct , but this small change really works

[yinfeng (yinfeng-zwx)]

from the comment of http://bugs.mysql.com/bug.php?id=64992, this bug was fixed in 5.7.0.
But I can't find any information about 5.7.0 and I don't know how they fix this bug.

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-1914