mariadb/sql/records.cc
Konstantin Osipov 06c9d62a9f A fix and a test case for
Bug#41756 "Strange error messages about locks from InnoDB".

In JT_EQ_REF (join_read_key()) access method,
don't try to unlock rows in the handler, unless certain that
a) they were locked
b) they are not used.

Unlocking of rows is done by the logic of the nested join loop,
and is unaware of the possible caching that the access method may
have. This could lead to double unlocking, when a row
was unlocked first after reading into the cache, and then
when taken from cache, as well as to unlocking of rows which
were actually used (but taken from cache).

Delegate part of the unlocking logic to the access method,
and in JT_EQ_REF count how many times a record was actually
used in the join. Unlock it only if it's usage count is 0.

Implemented review comments.


mysql-test/r/bug41756.result:
  Add result file (Bug#41756)
mysql-test/t/bug41756-master.opt:
  Use --innodb-locks-unsafe-for-binlog, as in 5.0 just
  using read_committed isolation is not sufficient to 
  reproduce the bug.
mysql-test/t/bug41756.test:
  Add a test file (Bug#41756)
sql/item_subselect.cc:
  Complete struct READ_RECORD initialization with a new
  member to unlock records.
sql/records.cc:
  Extend READ_RECORD API with a method to unlock read records.
sql/sql_select.cc:
  In JT_EQ_REF (join_read_key()) access method,
  don't try to unlock rows in the handler, unless certain that
  a) they were locked
  b) they are not used.
sql/sql_select.h:
  Add members to TABLE_REF to count TABLE_REF buffer usage count.
sql/structs.h:
  Update declarations.
2009-11-03 19:58:54 +03:00

583 lines
16 KiB
C++

/* Copyright (C) 2000-2006 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 */
/* Functions for easy reading of records, possible through a cache */
#include "mysql_priv.h"
static int rr_quick(READ_RECORD *info);
int rr_sequential(READ_RECORD *info);
static int rr_from_tempfile(READ_RECORD *info);
static int rr_unpack_from_tempfile(READ_RECORD *info);
static int rr_unpack_from_buffer(READ_RECORD *info);
static int rr_from_pointers(READ_RECORD *info);
static int rr_from_cache(READ_RECORD *info);
static int init_rr_cache(THD *thd, READ_RECORD *info);
static int rr_cmp(uchar *a,uchar *b);
static int rr_index_first(READ_RECORD *info);
static int rr_index(READ_RECORD *info);
/*
Initialize READ_RECORD structure to perform full index scan
SYNOPSIS
init_read_record_idx()
info READ_RECORD structure to initialize.
thd Thread handle
table Table to be accessed
print_error If true, call table->file->print_error() if an error
occurs (except for end-of-records error)
idx index to scan
DESCRIPTION
Initialize READ_RECORD structure to perform full index scan (in forward
direction) using read_record.read_record() interface.
This function has been added at late stage and is used only by
UPDATE/DELETE. Other statements perform index scans using
join_read_first/next functions.
*/
void init_read_record_idx(READ_RECORD *info, THD *thd, TABLE *table,
bool print_error, uint idx)
{
bzero((char*) info,sizeof(*info));
info->thd= thd;
info->table= table;
info->file= table->file;
info->record= table->record[0];
info->print_error= print_error;
info->unlock_row= rr_unlock_row;
table->status=0; /* And it's always found */
if (!table->file->inited)
{
table->file->ha_index_init(idx);
table->file->extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY);
}
/* read_record will be changed to rr_index in rr_index_first */
info->read_record= rr_index_first;
}
/*
init struct for read with info->read_record
SYNOPSIS
init_read_record()
info OUT read structure
thd Thread handle
table Table the data [originally] comes from.
select SQL_SELECT structure. We may select->quick or
select->file as data source
use_record_cache Call file->extra_opt(HA_EXTRA_CACHE,...)
if we're going to do sequential read and some
additional conditions are satisfied.
print_error Copy this to info->print_error
disable_rr_cache Don't use rr_from_cache (used by sort-union
index-merge which produces rowid sequences that
are already ordered)
DESCRIPTION
This function sets up reading data via one of the methods:
rr_unpack_from_tempfile Unpack full records from sequential file
rr_unpack_from_buffer ... or from buffer
rr_from_tempfile Read rowids from tempfile and get full records
with handler->rnd_pos() calls.
rr_from_pointers ... or get rowids from buffer
rr_from_cache Read a bunch of rowids from file, sort them,
get records in rowid order, return, repeat.
rr_quick Get data from QUICK_*_SELECT
rr_sequential Sequentially scan the table using
handler->rnd_next() calls
*/
void init_read_record(READ_RECORD *info,THD *thd, TABLE *table,
SQL_SELECT *select,
int use_record_cache, bool print_error,
bool disable_rr_cache)
{
IO_CACHE *tempfile;
DBUG_ENTER("init_read_record");
bzero((char*) info,sizeof(*info));
info->thd=thd;
info->table=table;
info->file= table->file;
info->forms= &info->table; /* Only one table */
if (table->sort.addon_field)
{
info->rec_buf= table->sort.addon_buf;
info->ref_length= table->sort.addon_length;
}
else
{
info->record= table->record[0];
info->ref_length= table->file->ref_length;
}
info->select=select;
info->print_error=print_error;
info->unlock_row= rr_unlock_row;
info->ignore_not_found_rows= 0;
table->status=0; /* And it's always found */
if (select && my_b_inited(&select->file))
tempfile= &select->file;
else
tempfile= table->sort.io_cache;
if (tempfile && my_b_inited(tempfile)) // Test if ref-records was used
{
DBUG_PRINT("info",("using rr_from_tempfile"));
info->read_record= (table->sort.addon_field ?
rr_unpack_from_tempfile : rr_from_tempfile);
info->io_cache=tempfile;
reinit_io_cache(info->io_cache,READ_CACHE,0L,0,0);
info->ref_pos=table->file->ref;
if (!table->file->inited)
table->file->ha_rnd_init(0);
/*
table->sort.addon_field is checked because if we use addon fields,
it doesn't make sense to use cache - we don't read from the table
and table->sort.io_cache is read sequentially
*/
if (!disable_rr_cache &&
!table->sort.addon_field &&
! (specialflag & SPECIAL_SAFE_MODE) &&
thd->variables.read_rnd_buff_size &&
!(table->file->table_flags() & HA_FAST_KEY_READ) &&
(table->db_stat & HA_READ_ONLY ||
table->reginfo.lock_type <= TL_READ_NO_INSERT) &&
(ulonglong) table->s->reclength* (table->file->records+
table->file->deleted) >
(ulonglong) MIN_FILE_LENGTH_TO_USE_ROW_CACHE &&
info->io_cache->end_of_file/info->ref_length * table->s->reclength >
(my_off_t) MIN_ROWS_TO_USE_TABLE_CACHE &&
!table->s->blob_fields &&
info->ref_length <= MAX_REFLENGTH)
{
if (! init_rr_cache(thd, info))
{
DBUG_PRINT("info",("using rr_from_cache"));
info->read_record=rr_from_cache;
}
}
}
else if (select && select->quick)
{
DBUG_PRINT("info",("using rr_quick"));
info->read_record=rr_quick;
}
else if (table->sort.record_pointers)
{
DBUG_PRINT("info",("using record_pointers"));
table->file->ha_rnd_init(0);
info->cache_pos=table->sort.record_pointers;
info->cache_end=info->cache_pos+
table->sort.found_records*info->ref_length;
info->read_record= (table->sort.addon_field ?
rr_unpack_from_buffer : rr_from_pointers);
}
else
{
DBUG_PRINT("info",("using rr_sequential"));
info->read_record=rr_sequential;
table->file->ha_rnd_init(1);
/* We can use record cache if we don't update dynamic length tables */
if (!table->no_cache &&
(use_record_cache > 0 ||
(int) table->reginfo.lock_type <= (int) TL_READ_HIGH_PRIORITY ||
!(table->s->db_options_in_use & HA_OPTION_PACK_RECORD) ||
(use_record_cache < 0 &&
!(table->file->table_flags() & HA_NOT_DELETE_WITH_CACHE))))
VOID(table->file->extra_opt(HA_EXTRA_CACHE,
thd->variables.read_buff_size));
}
/* Condition pushdown to storage engine */
if (thd->variables.engine_condition_pushdown &&
select && select->cond &&
(select->cond->used_tables() & table->map) &&
!table->file->pushed_cond)
table->file->cond_push(select->cond);
DBUG_VOID_RETURN;
} /* init_read_record */
void end_read_record(READ_RECORD *info)
{ /* free cache if used */
if (info->cache)
{
my_free_lock((char*) info->cache,MYF(0));
info->cache=0;
}
if (info->table)
{
filesort_free_buffers(info->table,0);
(void) info->file->extra(HA_EXTRA_NO_CACHE);
if (info->read_record != rr_quick) // otherwise quick_range does it
(void) info->file->ha_index_or_rnd_end();
info->table=0;
}
}
static int rr_handle_error(READ_RECORD *info, int error)
{
if (info->thd->killed)
{
info->thd->send_kill_message();
return 1;
}
if (error == HA_ERR_END_OF_FILE)
error= -1;
else
{
if (info->print_error)
info->table->file->print_error(error, MYF(0));
if (error < 0) // Fix negative BDB errno
error= 1;
}
return error;
}
/* Read a record from head-database */
static int rr_quick(READ_RECORD *info)
{
int tmp;
while ((tmp= info->select->quick->get_next()))
{
if (info->thd->killed || (tmp != HA_ERR_RECORD_DELETED))
{
tmp= rr_handle_error(info, tmp);
break;
}
}
return tmp;
}
/*
Reads first row in an index scan
SYNOPSIS
rr_index_first()
info Scan info
RETURN
0 Ok
-1 End of records
1 Error
*/
static int rr_index_first(READ_RECORD *info)
{
int tmp= info->file->index_first(info->record);
info->read_record= rr_index;
if (tmp)
tmp= rr_handle_error(info, tmp);
return tmp;
}
/*
Reads index sequentially after first row
SYNOPSIS
rr_index()
info Scan info
DESCRIPTION
Read the next index record (in forward direction) and translate return
value.
RETURN
0 Ok
-1 End of records
1 Error
*/
static int rr_index(READ_RECORD *info)
{
int tmp= info->file->index_next(info->record);
if (tmp)
tmp= rr_handle_error(info, tmp);
return tmp;
}
int rr_sequential(READ_RECORD *info)
{
int tmp;
while ((tmp=info->file->rnd_next(info->record)))
{
/*
rnd_next can return RECORD_DELETED for MyISAM when one thread is
reading and another deleting without locks.
*/
if (info->thd->killed || (tmp != HA_ERR_RECORD_DELETED))
{
tmp= rr_handle_error(info, tmp);
break;
}
}
return tmp;
}
static int rr_from_tempfile(READ_RECORD *info)
{
int tmp;
for (;;)
{
if (my_b_read(info->io_cache,info->ref_pos,info->ref_length))
return -1; /* End of file */
if (!(tmp=info->file->rnd_pos(info->record,info->ref_pos)))
break;
/* The following is extremely unlikely to happen */
if (tmp == HA_ERR_RECORD_DELETED ||
(tmp == HA_ERR_KEY_NOT_FOUND && info->ignore_not_found_rows))
continue;
tmp= rr_handle_error(info, tmp);
break;
}
return tmp;
} /* rr_from_tempfile */
/*
Read a result set record from a temporary file after sorting
SYNOPSIS
rr_unpack_from_tempfile()
info Reference to the context including record descriptors
DESCRIPTION
The function first reads the next sorted record from the temporary file.
into a buffer. If a success it calls a callback function that unpacks
the fields values use in the result set from this buffer into their
positions in the regular record buffer.
RETURN
0 - Record successfully read.
-1 - There is no record to be read anymore.
*/
static int rr_unpack_from_tempfile(READ_RECORD *info)
{
if (my_b_read(info->io_cache, info->rec_buf, info->ref_length))
return -1;
TABLE *table= info->table;
(*table->sort.unpack)(table->sort.addon_field, info->rec_buf);
return 0;
}
static int rr_from_pointers(READ_RECORD *info)
{
int tmp;
byte *cache_pos;
for (;;)
{
if (info->cache_pos == info->cache_end)
return -1; /* End of file */
cache_pos= info->cache_pos;
info->cache_pos+= info->ref_length;
if (!(tmp=info->file->rnd_pos(info->record,cache_pos)))
break;
/* The following is extremely unlikely to happen */
if (tmp == HA_ERR_RECORD_DELETED ||
(tmp == HA_ERR_KEY_NOT_FOUND && info->ignore_not_found_rows))
continue;
tmp= rr_handle_error(info, tmp);
break;
}
return tmp;
}
/*
Read a result set record from a buffer after sorting
SYNOPSIS
rr_unpack_from_buffer()
info Reference to the context including record descriptors
DESCRIPTION
The function first reads the next sorted record from the sort buffer.
If a success it calls a callback function that unpacks
the fields values use in the result set from this buffer into their
positions in the regular record buffer.
RETURN
0 - Record successfully read.
-1 - There is no record to be read anymore.
*/
static int rr_unpack_from_buffer(READ_RECORD *info)
{
if (info->cache_pos == info->cache_end)
return -1; /* End of buffer */
TABLE *table= info->table;
(*table->sort.unpack)(table->sort.addon_field, info->cache_pos);
info->cache_pos+= info->ref_length;
return 0;
}
/* cacheing of records from a database */
static int init_rr_cache(THD *thd, READ_RECORD *info)
{
uint rec_cache_size;
DBUG_ENTER("init_rr_cache");
info->struct_length= 3+MAX_REFLENGTH;
info->reclength= ALIGN_SIZE(info->table->s->reclength+1);
if (info->reclength < info->struct_length)
info->reclength= ALIGN_SIZE(info->struct_length);
info->error_offset= info->table->s->reclength;
info->cache_records= (thd->variables.read_rnd_buff_size /
(info->reclength+info->struct_length));
rec_cache_size= info->cache_records*info->reclength;
info->rec_cache_size= info->cache_records*info->ref_length;
// We have to allocate one more byte to use uint3korr (see comments for it)
if (info->cache_records <= 2 ||
!(info->cache=(byte*) my_malloc_lock(rec_cache_size+info->cache_records*
info->struct_length+1,
MYF(0))))
DBUG_RETURN(1);
#ifdef HAVE_purify
// Avoid warnings in qsort
bzero(info->cache,rec_cache_size+info->cache_records* info->struct_length+1);
#endif
DBUG_PRINT("info",("Allocated buffert for %d records",info->cache_records));
info->read_positions=info->cache+rec_cache_size;
info->cache_pos=info->cache_end=info->cache;
DBUG_RETURN(0);
} /* init_rr_cache */
static int rr_from_cache(READ_RECORD *info)
{
reg1 uint i;
ulong length;
my_off_t rest_of_file;
int16 error;
byte *position,*ref_position,*record_pos;
ulong record;
for (;;)
{
if (info->cache_pos != info->cache_end)
{
if (info->cache_pos[info->error_offset])
{
shortget(error,info->cache_pos);
if (info->print_error)
info->table->file->print_error(error,MYF(0));
}
else
{
error=0;
memcpy(info->record,info->cache_pos,
(size_t) info->table->s->reclength);
}
info->cache_pos+=info->reclength;
return ((int) error);
}
length=info->rec_cache_size;
rest_of_file=info->io_cache->end_of_file - my_b_tell(info->io_cache);
if ((my_off_t) length > rest_of_file)
length= (ulong) rest_of_file;
if (!length || my_b_read(info->io_cache,info->cache,length))
{
DBUG_PRINT("info",("Found end of file"));
return -1; /* End of file */
}
length/=info->ref_length;
position=info->cache;
ref_position=info->read_positions;
for (i=0 ; i < length ; i++,position+=info->ref_length)
{
memcpy(ref_position,position,(size_s) info->ref_length);
ref_position+=MAX_REFLENGTH;
int3store(ref_position,(long) i);
ref_position+=3;
}
my_qsort(info->read_positions, length, info->struct_length,
(qsort_cmp) rr_cmp);
position=info->read_positions;
for (i=0 ; i < length ; i++)
{
memcpy(info->ref_pos,position,(size_s) info->ref_length);
position+=MAX_REFLENGTH;
record=uint3korr(position);
position+=3;
record_pos=info->cache+record*info->reclength;
if ((error=(int16) info->file->rnd_pos(record_pos,info->ref_pos)))
{
record_pos[info->error_offset]=1;
shortstore(record_pos,error);
DBUG_PRINT("error",("Got error: %d:%d when reading row",
my_errno, error));
}
else
record_pos[info->error_offset]=0;
}
info->cache_end=(info->cache_pos=info->cache)+length*info->reclength;
}
} /* rr_from_cache */
static int rr_cmp(uchar *a,uchar *b)
{
if (a[0] != b[0])
return (int) a[0] - (int) b[0];
if (a[1] != b[1])
return (int) a[1] - (int) b[1];
if (a[2] != b[2])
return (int) a[2] - (int) b[2];
#if MAX_REFLENGTH == 4
return (int) a[3] - (int) b[3];
#else
if (a[3] != b[3])
return (int) a[3] - (int) b[3];
if (a[4] != b[4])
return (int) a[4] - (int) b[4];
if (a[5] != b[5])
return (int) a[1] - (int) b[5];
if (a[6] != b[6])
return (int) a[6] - (int) b[6];
return (int) a[7] - (int) b[7];
#endif
}