mariadb/sql/sql_cursor.cc
unknown 526798dbb5 A fix and a test case for Bug#12713 "Error in a stored function called from
a SELECT doesn't cause ROLLBACK of statem".

The idea of the fix is to ensure that we always commit the current
statement at the end of dispatch_command(). In order to not issue
redundant disc syncs, an optimization of the two-phase commit
protocol is implemented to bypass the two phase commit if
the transaction is read-only.


mysql-test/suite/binlog/r/binlog_row_mix_innodb_myisam.result:
  Update test results.
mysql-test/suite/binlog/r/binlog_stm_mix_innodb_myisam.result:
  Update test results.
mysql-test/suite/rpl_ndb/t/disabled.def:
  Disable the tests, for which this changeset reveals a bug:
  the injector thread does not always add 'statement commit' to the
  rows injected in circular replication set up.
  To be investigated separately.
sql/ha_ndbcluster_binlog.cc:
  Add close_thread_tables() to run_query: this ensures
  that all tables are closed and there is no pending statement transaction.
sql/handler.cc:
  Implement optimisation of read-only transactions.
  If a transaction consists only of DML statements that do not change
  data, we do not perform a two-phase commit for it 
  (run one phase commit only).
sql/handler.h:
  Implement optimisation of read-only transactions.
  If a transaction consists only of DML statements that do not change
  data, we do not perform a two-phase commit for it 
  (run one phase commit only).
sql/log.cc:
  Mark the binlog transaction read-write whenever it's started.
  We never read from binlog, so it's safe and least intrusive to add
  this mark up here.
sql/log_event.cc:
  Update to the new layout of thd->transaction.
sql/rpl_injector.cc:
  Always commit statement transaction before committing the global one.
sql/sp.cc:
  Ad comments.
sql/sp_head.cc:
  Add comments.
sql/sql_base.cc:
  Commit transaction at the end of the statement. Always.
sql/sql_class.cc:
  Update thd_ha_data to return the right pointer in the new layout.
  
  Fix select_dumpvar::send_data to properly return operation status.
  A test case from commit.inc would lead to an assertion failure in the 
  diagnostics area (double assignment). Not test otherwise by the test suite.
sql/sql_class.h:
  Implement a new layout of storage engine transaction info in which 
  it is easy to access all members related to the handlerton only
  based on ht->slot.
sql/sql_cursor.cc:
  Update to the new layout of thd->transaction.
sql/sql_delete.cc:
  Remove wrong and now redundant calls to ha_autocommit_or_rollback.
  The transaction is committed in one place, at the end of the statement.
  Remove calls to mysql_unlock_tables, since some engines count locks
  and commit statement transaction in unlock_tables(), which essentially
  equates mysql_unlock_tables to ha_autocommit_or_rollback.
  Previously it was necessary to unlock tables soon because we wanted
  to avoid sending of 'ok' packet to the client under locked tables.
  This is no longer necessary, since OK packet is also sent from one place
  at the end of transaction.
sql/sql_do.cc:
  Add DO always clears the error, we must rollback the current
  statement before this happens. Otherwise the statement will be committed,
  and not rolled back in the end.
sql/sql_insert.cc:
  Remove wrong and now redundant calls to ha_autocommit_or_rollback.
  The transaction is committed in one place, at the end of the statement.
  Remove calls to mysql_unlock_tables, since some engines count locks
  and commit statement transaction in unlock_tables(), which essentially
  equates mysql_unlock_tables to ha_autocommit_or_rollback.
  Previously it was necessary to unlock tables soon because we wanted
  to avoid sending of 'ok' packet to the client under locked tables.
  This is no longer necessary, since OK packet is also sent from one place
  at the end of transaction.
sql/sql_load.cc:
  Remove wrong and now redundant calls to ha_autocommit_or_rollback.
  The transaction is committed in one place, at the end of the statement.
  Remove calls to mysql_unlock_tables, since some engines count locks
  and commit statement transaction in unlock_tables(), which essentially
  equates mysql_unlock_tables to ha_autocommit_or_rollback.
  Previously it was necessary to unlock tables soon because we wanted
  to avoid sending of 'ok' packet to the client under locked tables.
  This is no longer necessary, since OK packet is also sent from one place
  at the end of transaction.
sql/sql_parse.cc:
  Implement optimisation of read-only transactions: bypass 2-phase
  commit for them.
  Always commit statement transaction before commiting the global one.
  Fix an unrelated crash in check_table_access, when called from 
  information_schema.
sql/sql_partition.cc:
  Partitions commit at the end of a DDL operation.
  Make sure that send_ok() is done only if the commit has succeeded.
sql/sql_table.cc:
  Use ha_autocommit_or_rollback and end_active_trans everywhere.
  Add end_trans to mysql_admin_table, so that it leaves no pending
  transaction.
sql/sql_udf.cc:
  Remvove a redundant call to close_thread_tables()
sql/sql_update.cc:
  Remove wrong and now redundant calls to ha_autocommit_or_rollback.
  The transaction is committed in one place, at the end of the statement.
  Remove calls to mysql_unlock_tables, since some engines count locks
  and commit statement transaction in unlock_tables(), which essentially
  equates mysql_unlock_tables to ha_autocommit_or_rollback.
  Previously it was necessary to unlock tables soon because we wanted
  to avoid sending of 'ok' packet to the client under locked tables.
  This is no longer necessary, since OK packet is also sent from one place
  at the end of transaction.
mysql-test/include/commit.inc:
  New BitKeeper file ``mysql-test/include/commit.inc''
mysql-test/r/commit_1innodb.result:
  New BitKeeper file ``mysql-test/r/commit_1innodb.result''
mysql-test/t/commit_1innodb.test:
  New BitKeeper file ``mysql-test/t/commit_1innodb.test''
2008-02-19 14:43:01 +03:00

674 lines
19 KiB
C++

/* Copyright (C) 2005-2006 MySQL AB
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
#ifdef USE_PRAGMA_IMPLEMENTATION
#pragma implementation /* gcc class implementation */
#endif
#include "mysql_priv.h"
#include "sql_cursor.h"
#include "sql_select.h"
/****************************************************************************
Declarations.
****************************************************************************/
/**
Sensitive_cursor -- a sensitive non-materialized server side
cursor. An instance of this class cursor has its own runtime
state -- list of used items and memory root for runtime memory,
open and locked tables, change list for the changes of the
parsed tree. This state is freed when the cursor is closed.
*/
class Sensitive_cursor: public Server_side_cursor
{
MEM_ROOT main_mem_root;
Query_arena *stmt_arena;
JOIN *join;
TABLE *open_tables;
MYSQL_LOCK *lock;
TABLE *derived_tables;
/* List of items created during execution */
query_id_t query_id;
struct Engine_info
{
handlerton *ht;
void *read_view;
};
Engine_info ht_info[MAX_HA];
Item_change_list change_list;
my_bool close_at_commit;
THR_LOCK_OWNER lock_id;
private:
/* bzero cursor state in THD */
void reset_thd(THD *thd);
public:
Sensitive_cursor(THD *thd, select_result *result_arg);
THR_LOCK_OWNER *get_lock_id() { return &lock_id; }
/* Save THD state into cursor */
void post_open(THD *thd);
virtual bool is_open() const { return join != 0; }
virtual int open(JOIN *join);
virtual void fetch(ulong num_rows);
virtual void close();
virtual ~Sensitive_cursor();
};
/**
Materialized_cursor -- an insensitive materialized server-side
cursor. The result set of this cursor is saved in a temporary
table at open. The cursor itself is simply an interface for the
handler of the temporary table.
*/
class Materialized_cursor: public Server_side_cursor
{
MEM_ROOT main_mem_root;
/* A fake unit to supply to select_send when fetching */
SELECT_LEX_UNIT fake_unit;
TABLE *table;
List<Item> item_list;
ulong fetch_limit;
ulong fetch_count;
public:
Materialized_cursor(select_result *result, TABLE *table);
virtual bool is_open() const { return table != 0; }
virtual int open(JOIN *join __attribute__((unused)));
virtual void fetch(ulong num_rows);
virtual void close();
virtual ~Materialized_cursor();
};
/**
Select_materialize -- a mediator between a cursor query and the
protocol. In case we were not able to open a non-materialzed
cursor, it creates an internal temporary HEAP table, and insert
all rows into it. When the table reaches max_heap_table_size,
it's converted to a MyISAM table. Later this table is used to
create a Materialized_cursor.
*/
class Select_materialize: public select_union
{
select_result *result; /**< the result object of the caller (PS or SP) */
public:
Select_materialize(select_result *result_arg) :result(result_arg) {}
virtual bool send_fields(List<Item> &list, uint flags);
};
/**************************************************************************/
/**
Attempt to open a materialized or non-materialized cursor.
@param thd thread handle
@param[in] flags create a materialized cursor or not
@param[in] result result class of the caller used as a destination
for the rows fetched from the cursor
@param[out] pcursor a pointer to store a pointer to cursor in
@retval
0 the query has been successfully executed; in this
case pcursor may or may not contain
a pointer to an open cursor.
@retval
non-zero an error, 'pcursor' has been left intact.
*/
int mysql_open_cursor(THD *thd, uint flags, select_result *result,
Server_side_cursor **pcursor)
{
Sensitive_cursor *sensitive_cursor;
select_result *save_result;
Select_materialize *result_materialize;
LEX *lex= thd->lex;
int rc;
/*
The lifetime of the sensitive cursor is the same or less as the
lifetime of the runtime memory of the statement it's opened for.
*/
if (! (result_materialize= new (thd->mem_root) Select_materialize(result)))
return 1;
if (! (sensitive_cursor= new (thd->mem_root) Sensitive_cursor(thd, result)))
{
delete result;
return 1;
}
save_result= lex->result;
lex->result= result_materialize;
if (! (flags & (uint) ALWAYS_MATERIALIZED_CURSOR))
{
thd->lock_id= sensitive_cursor->get_lock_id();
thd->cursor= sensitive_cursor;
}
rc= mysql_execute_command(thd);
lex->result= save_result;
thd->lock_id= &thd->main_lock_id;
thd->cursor= 0;
/*
Possible options here:
- a sensitive cursor is open. In this case rc is 0 and
result_materialize->table is NULL, or
- a materialized cursor is open. In this case rc is 0 and
result_materialize->table is not NULL
- an error occured during materializaton.
result_materialize->table is not NULL, but rc != 0
- successful completion of mysql_execute_command without
a cursor: rc is 0, result_materialize->table is NULL,
sensitive_cursor is not open.
This is possible if some command writes directly to the
network, bypassing select_result mechanism. An example of
such command is SHOW VARIABLES or SHOW STATUS.
*/
if (rc)
goto err_open;
if (sensitive_cursor->is_open())
{
DBUG_ASSERT(!result_materialize->table);
/*
It's safer if we grab THD state after mysql_execute_command
is finished and not in Sensitive_cursor::open(), because
currently the call to Sensitive_cursor::open is buried deep
in JOIN::exec of the top level join.
*/
sensitive_cursor->post_open(thd);
*pcursor= sensitive_cursor;
goto end;
}
else if (result_materialize->table)
{
Materialized_cursor *materialized_cursor;
TABLE *table= result_materialize->table;
MEM_ROOT *mem_root= &table->mem_root;
if (!(materialized_cursor= new (mem_root)
Materialized_cursor(result, table)))
{
rc= 1;
goto err_open;
}
if ((rc= materialized_cursor->open(0)))
{
delete materialized_cursor;
goto err_open;
}
*pcursor= materialized_cursor;
thd->stmt_arena->cleanup_stmt();
goto end;
}
err_open:
DBUG_ASSERT(! (sensitive_cursor && sensitive_cursor->is_open()));
delete sensitive_cursor;
if (result_materialize->table)
free_tmp_table(thd, result_materialize->table);
end:
delete result_materialize;
return rc;
}
/****************************************************************************
Server_side_cursor
****************************************************************************/
Server_side_cursor::~Server_side_cursor()
{
}
void Server_side_cursor::operator delete(void *ptr, size_t size)
{
Server_side_cursor *cursor= (Server_side_cursor*) ptr;
MEM_ROOT own_root= *cursor->mem_root;
DBUG_ENTER("Server_side_cursor::operator delete");
TRASH(ptr, size);
/*
If this cursor has never been opened mem_root is empty. Otherwise
mem_root points to the memory the cursor object was allocated in.
In this case it's important to call free_root last, and free a copy
instead of *mem_root to avoid writing into freed memory.
*/
free_root(&own_root, MYF(0));
DBUG_VOID_RETURN;
}
/****************************************************************************
Sensitive_cursor
****************************************************************************/
Sensitive_cursor::Sensitive_cursor(THD *thd, select_result *result_arg)
:Server_side_cursor(&main_mem_root, result_arg),
stmt_arena(0),
join(0),
close_at_commit(FALSE)
{
/* We will overwrite it at open anyway. */
init_sql_alloc(&main_mem_root, ALLOC_ROOT_MIN_BLOCK_SIZE, 0);
thr_lock_owner_init(&lock_id, &thd->lock_info);
bzero((void*) ht_info, sizeof(ht_info));
}
/**
Save THD state into cursor.
@todo
- XXX: thd->locked_tables is not changed.
- What problems can we have with it if cursor is open?
- TODO: must be fixed because of the prelocked mode.
*/
void
Sensitive_cursor::post_open(THD *thd)
{
Engine_info *info;
/*
We need to save and reset thd->mem_root, otherwise it'll be
freed later in mysql_parse.
We can't just change thd->mem_root here as we want to keep the
things that are already allocated in thd->mem_root for
Sensitive_cursor::fetch()
*/
*mem_root= *thd->mem_root;
stmt_arena= thd->stmt_arena;
state= stmt_arena->state;
/* Allocate a new memory root for thd */
init_sql_alloc(thd->mem_root,
thd->variables.query_alloc_block_size,
thd->variables.query_prealloc_size);
/*
Save tables and zero THD pointers to prevent table close in
close_thread_tables.
*/
derived_tables= thd->derived_tables;
open_tables= thd->open_tables;
lock= thd->lock;
query_id= thd->query_id;
free_list= thd->free_list;
change_list= thd->change_list;
reset_thd(thd);
/* Now we have an active cursor and can cause a deadlock */
thd->lock_info.n_cursors++;
close_at_commit= FALSE; /* reset in case we're reusing the cursor */
info= &ht_info[0];
for (Ha_trx_info *ha_trx_info= thd->transaction.stmt.ha_list;
ha_trx_info; ha_trx_info= ha_trx_info->next())
{
handlerton *ht= ha_trx_info->ht();
close_at_commit|= test(ht->flags & HTON_CLOSE_CURSORS_AT_COMMIT);
if (ht->create_cursor_read_view)
{
info->ht= ht;
info->read_view= (ht->create_cursor_read_view)(ht, thd);
++info;
}
}
/*
XXX: thd->locked_tables is not changed.
What problems can we have with it if cursor is open?
TODO: must be fixed because of the prelocked mode.
*/
}
/**
bzero cursor state in THD.
*/
void
Sensitive_cursor::reset_thd(THD *thd)
{
thd->derived_tables= 0;
thd->open_tables= 0;
thd->lock= 0;
thd->free_list= 0;
thd->change_list.empty();
}
int
Sensitive_cursor::open(JOIN *join_arg)
{
join= join_arg;
THD *thd= join->thd;
/* First non-constant table */
JOIN_TAB *join_tab= join->join_tab + join->const_tables;
DBUG_ENTER("Sensitive_cursor::open");
join->change_result(result);
/*
Send fields description to the client; server_status is sent
in 'EOF' packet, which follows send_fields().
We don't simply use SEND_EOF flag of send_fields because we also
want to flush the network buffer, which is done only in a standalone
send_eof().
*/
result->send_fields(*join->fields, Protocol::SEND_NUM_ROWS);
thd->server_status|= SERVER_STATUS_CURSOR_EXISTS;
result->send_eof();
thd->server_status&= ~SERVER_STATUS_CURSOR_EXISTS;
/* Prepare JOIN for reading rows. */
join->tmp_table= 0;
join->join_tab[join->tables-1].next_select= setup_end_select_func(join);
join->send_records= 0;
join->fetch_limit= join->unit->offset_limit_cnt;
/* Disable JOIN CACHE as it is not working with cursors yet */
for (JOIN_TAB *tab= join_tab;
tab != join->join_tab + join->tables - 1;
tab++)
{
if (tab->next_select == sub_select_cache)
tab->next_select= sub_select;
}
DBUG_ASSERT(join_tab->table->reginfo.not_exists_optimize == 0);
DBUG_ASSERT(join_tab->not_used_in_distinct == 0);
/*
null_row is set only if row not found and it's outer join: should never
happen for the first table in join_tab list
*/
DBUG_ASSERT(join_tab->table->null_row == 0);
DBUG_RETURN(0);
}
/**
Fetch next num_rows rows from the cursor and send them to the client.
Precondition:
- Sensitive_cursor is open
@param num_rows fetch up to this number of rows (maybe less)
*/
void
Sensitive_cursor::fetch(ulong num_rows)
{
THD *thd= join->thd;
JOIN_TAB *join_tab= join->join_tab + join->const_tables;
enum_nested_loop_state error= NESTED_LOOP_OK;
Query_arena backup_arena;
Engine_info *info;
DBUG_ENTER("Sensitive_cursor::fetch");
DBUG_PRINT("enter",("rows: %lu", num_rows));
DBUG_ASSERT(thd->derived_tables == 0 && thd->open_tables == 0 &&
thd->lock == 0);
thd->derived_tables= derived_tables;
thd->open_tables= open_tables;
thd->lock= lock;
thd->query_id= query_id;
thd->change_list= change_list;
/* save references to memory allocated during fetch */
thd->set_n_backup_active_arena(this, &backup_arena);
for (info= ht_info; info->read_view ; info++)
(info->ht->set_cursor_read_view)(info->ht, thd, info->read_view);
join->fetch_limit+= num_rows;
error= sub_select(join, join_tab, 0);
if (error == NESTED_LOOP_OK || error == NESTED_LOOP_NO_MORE_ROWS)
error= sub_select(join,join_tab,1);
if (error == NESTED_LOOP_QUERY_LIMIT)
error= NESTED_LOOP_OK; /* select_limit used */
if (error == NESTED_LOOP_CURSOR_LIMIT)
join->resume_nested_loop= TRUE;
ha_release_temporary_latches(thd);
/* Grab free_list here to correctly free it in close */
thd->restore_active_arena(this, &backup_arena);
change_list= thd->change_list;
reset_thd(thd);
for (info= ht_info; info->read_view; info++)
(info->ht->set_cursor_read_view)(info->ht, thd, 0);
if (error == NESTED_LOOP_CURSOR_LIMIT)
{
/* Fetch limit worked, possibly more rows are there */
thd->server_status|= SERVER_STATUS_CURSOR_EXISTS;
result->send_eof();
thd->server_status&= ~SERVER_STATUS_CURSOR_EXISTS;
}
else
{
close();
if (error == NESTED_LOOP_OK)
{
thd->server_status|= SERVER_STATUS_LAST_ROW_SENT;
result->send_eof();
thd->server_status&= ~SERVER_STATUS_LAST_ROW_SENT;
}
else if (error != NESTED_LOOP_KILLED)
my_message(ER_OUT_OF_RESOURCES, ER(ER_OUT_OF_RESOURCES), MYF(0));
}
DBUG_VOID_RETURN;
}
/**
@todo
Another hack: we need to set THD state as if in a fetch to be
able to call stmt close.
*/
void
Sensitive_cursor::close()
{
THD *thd= join->thd;
DBUG_ENTER("Sensitive_cursor::close");
for (Engine_info *info= ht_info; info->read_view; info++)
{
(info->ht->close_cursor_read_view)(info->ht, thd, info->read_view);
info->read_view= 0;
info->ht= 0;
}
thd->change_list= change_list;
{
/*
XXX: Another hack: we need to set THD state as if in a fetch to be
able to call stmt close.
*/
DBUG_ASSERT(lock || open_tables || derived_tables);
TABLE *tmp_derived_tables= thd->derived_tables;
MYSQL_LOCK *tmp_lock= thd->lock;
thd->open_tables= open_tables;
thd->derived_tables= derived_tables;
thd->lock= lock;
/* Is expected to at least close tables and empty thd->change_list */
stmt_arena->cleanup_stmt();
thd->open_tables= tmp_derived_tables;
thd->derived_tables= tmp_derived_tables;
thd->lock= tmp_lock;
}
thd->lock_info.n_cursors--; /* Decrease the number of active cursors */
join= 0;
stmt_arena= 0;
free_items();
change_list.empty();
DBUG_VOID_RETURN;
}
Sensitive_cursor::~Sensitive_cursor()
{
if (is_open())
close();
}
/***************************************************************************
Materialized_cursor
****************************************************************************/
Materialized_cursor::Materialized_cursor(select_result *result_arg,
TABLE *table_arg)
:Server_side_cursor(&table_arg->mem_root, result_arg),
table(table_arg),
fetch_limit(0),
fetch_count(0)
{
fake_unit.init_query();
fake_unit.thd= table->in_use;
}
int Materialized_cursor::open(JOIN *join __attribute__((unused)))
{
THD *thd= fake_unit.thd;
int rc;
Query_arena backup_arena;
thd->set_n_backup_active_arena(this, &backup_arena);
/* Create a list of fields and start sequential scan */
rc= (table->fill_item_list(&item_list) ||
result->prepare(item_list, &fake_unit) ||
table->file->ha_rnd_init(TRUE));
thd->restore_active_arena(this, &backup_arena);
if (rc == 0)
{
/*
Now send the result set metadata to the client. We need to
do it here, as in Select_materialize::send_fields the items
for column types are not yet created (send_fields requires
a list of items). The new types may differ from the original
ones sent at prepare if some of them were altered by MySQL
HEAP tables mechanism -- used when create_tmp_field_from_item
may alter the original column type.
We can't simply supply SEND_EOF flag to send_fields, because
send_fields doesn't flush the network buffer.
*/
rc= result->send_fields(item_list, Protocol::SEND_NUM_ROWS);
thd->server_status|= SERVER_STATUS_CURSOR_EXISTS;
result->send_eof();
thd->server_status&= ~SERVER_STATUS_CURSOR_EXISTS;
}
return rc;
}
/**
Fetch up to the given number of rows from a materialized cursor.
Precondition: the cursor is open.
If the cursor points after the last row, the fetch will automatically
close the cursor and not send any data (except the 'EOF' packet
with SERVER_STATUS_LAST_ROW_SENT). This is an extra round trip
and probably should be improved to return
SERVER_STATUS_LAST_ROW_SENT along with the last row.
*/
void Materialized_cursor::fetch(ulong num_rows)
{
THD *thd= table->in_use;
int res= 0;
result->begin_dataset();
for (fetch_limit+= num_rows; fetch_count < fetch_limit; fetch_count++)
{
if ((res= table->file->rnd_next(table->record[0])))
break;
/* Send data only if the read was successful. */
result->send_data(item_list);
}
switch (res) {
case 0:
thd->server_status|= SERVER_STATUS_CURSOR_EXISTS;
result->send_eof();
thd->server_status&= ~SERVER_STATUS_CURSOR_EXISTS;
break;
case HA_ERR_END_OF_FILE:
thd->server_status|= SERVER_STATUS_LAST_ROW_SENT;
result->send_eof();
thd->server_status&= ~SERVER_STATUS_LAST_ROW_SENT;
close();
break;
default:
table->file->print_error(res, MYF(0));
close();
break;
}
}
void Materialized_cursor::close()
{
/* Free item_list items */
free_items();
(void) table->file->ha_rnd_end();
/*
We need to grab table->mem_root to prevent free_tmp_table from freeing:
the cursor object was allocated in this memory.
*/
main_mem_root= table->mem_root;
mem_root= &main_mem_root;
clear_alloc_root(&table->mem_root);
free_tmp_table(table->in_use, table);
table= 0;
}
Materialized_cursor::~Materialized_cursor()
{
if (is_open())
close();
}
/***************************************************************************
Select_materialize
****************************************************************************/
bool Select_materialize::send_fields(List<Item> &list, uint flags)
{
DBUG_ASSERT(table == 0);
if (create_result_table(unit->thd, unit->get_unit_column_types(),
FALSE, thd->options | TMP_TABLE_ALL_COLUMNS, ""))
return TRUE;
return FALSE;
}