mariadb/sql/structs.h
unknown 60272e750e WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements. 
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully 
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.

Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().

Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.


mysql-test/extra/binlog_tests/binlog.test:
  Testing that if INSERT_ID is set to a value too big for the
  column's type, the binlogged INSERT_ID is the truncated value
  (important if slave has a column of a "wider" numeric type).
  Testing binlogging of INSERT_ID with INSERT DELAYED, to be sure that 
  we binlog an INSERT_ID event only for the delayed rows which use one.
mysql-test/extra/rpl_tests/rpl_insert_id.test:
  Testcase for BUG#20339 "stored procedure using
  LAST_INSERT_ID() does not replicate statement-based".
  Testcase for BUG#20341 "stored function inserting into one
  auto_increment puts bad data in slave".
mysql-test/extra/rpl_tests/rpl_loaddata.test:
  Test that LOAD DATA INFILE sets a value for a future LAST_INSERT_ID().
mysql-test/r/auto_increment.result:
  behaviour change: when INSERT totally fails (not even succeeds
  partially and then rolls back), don't change last_insert_id().
  Behaviour change: LAST_INSERT_ID() is now the first successfully inserted,
  autogenerated, id.
  Behaviour change: if INSERT ON DUPLICATE KEY UPDATE, if the table has auto_increment
  and a row is updated, then LAST_INSERT_ID() returns the id of this row.
mysql-test/r/binlog_row_binlog.result:
  result update
mysql-test/r/binlog_stm_binlog.result:
  result update
mysql-test/r/insert.result:
  result update
mysql-test/r/rpl_insert_id.result:
  result update
mysql-test/r/rpl_loaddata.result:
  result update
mysql-test/r/rpl_ndb_auto_inc.result:
  ndb's behaviour is now like other engines wrt SET INSERT_ID
  in a multi-row INSERT:
  - with INSERT IGNORE: the id passed in SET INSERT_ID is re-used until
  a row succeeds.
  - generally, SET INSERT_ID sets the first value and other values are
  simply computed from this first value, instead of previously where
  the 2nd and subsequent values where not influenced by SET INSERT_ID;
  this good change is due to the removal of "thd->next_insert_id=0"
  from ha_ndbcluster.
mysql-test/t/auto_increment.test:
  A testcase of BUG#19243: if ON DUPLICATE KEY UPDATE updates a row,
  LAST_INSERT_ID() now returns the id of the row.
  Test of new behaviour of last_insert_id() when no autogenerated value was
  inserted, or when only some autogenerated value (not the first of them) was
  inserted.
mysql-test/t/insert.test:
  testing INSERT IGNORE re-using generated values
sql/ha_federated.cc:
  update for new variables.
sql/ha_ndbcluster.cc:
  handler::auto_increment_column_changed not needed, equivalent to
  (insert_id_for_cur_row > 0).
  thd->next_insert_id=0 not needed anymore; it was used to force
  handler::update_auto_increment() to call ha_ndbcluster::get_auto_increment()
  for each row of a multi-row INSERT, now this happens naturally
  because NDB says "I have reserved you *one* value" in get_auto_increment(),
  so handler::update_auto_increment() calls again for next row.
sql/handler.cc:
  More comments, use of new methods and variables. Hopes to be clearer
  than current code.
  thd->prev_insert_id not in THD anymore: it is managed locally by inserters
  (like mysql_insert()).
  THD::clear_next_insert_id is now equivalent to
  handler::next_insert_id > 0.
  get_auto_increment() reserves an interval of values from the engine,
  uses this interval for next rows of the statement, until interval
  is exhausted then it asks for another interval (of a bigger size
  than the first one; size doubles until reaching 65535 then it stays constant).
  If doing statement-based binlogging, intervals are remembered in a list
  for storage in the binlog.
  For "forced" insert_id values (SET INSERT_ID or replication slave),
  forced_auto_inc_intervals is non-empty and the handler takes its intervals
  from there, without calling get_auto_increment().
  ha_release_auto_increment() resets the handler's auto_increment variables;
  it calls release_auto_increment() which is handler-dependent and
  serves to return to the engine any unused tail of the last used
  interval.
  If ending a statement, next_insert_id>0 means that autoinc values have been
  generated or taken from the master's binlog (in a replication slave) so
  we clear those values read from binlog, so that next top- or sub-
  statement does not use them.
sql/handler.h:
  handler::auto_increment_changed can be replaced by
  (handler::insert_id_for_cur_row > 0).
  THD::next_insert_id moves into handler (more natural, and prepares
  for the day when we'll support a single statement inserting into
  two tables - "multi-table INSERT" like we have UPDATE - will this
  happen?).
  This move makes the backup/restore of THD::next_insert_id when entering
  a substatement unneeded, as each substatement has its own handler
  objects.
sql/item_func.cc:
  new names for variables.
  For the setting of what mysql_insert_id() will return to the client,
  LAST_INSERT_ID(X) used to simply pretend that the generated autoinc
  value for the current row was X, but this led to having no reliable
  way to know the really generated value, so we now have a bool:
  thd->arg_of_last_insert_id_function which enables us to know that
  LAST_INSERT_ID(X) was called (and then X can be found in
  thd->first_successful_insert_id_in_prev_stmt).
sql/log.cc:
  new variable names for insert_ids. Removing some unused variables in the slow
  log.
sql/log_event.cc:
  new variable names, comments. Preparing for when master's won't binlog
  LAST_INSERT_ID if it was 0.
sql/set_var.cc:
  new variable names.
  The last change repeats how Bar fixed BUG#20392
  "INSERT_ID session variable has weird value" in 5.0.
sql/sql_class.cc:
  new variables for insert_id. In THD::cleanup_after_query() we fix
  BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
  statement-based" (will one want to fix it in 5.0?). Many comments
  about what stored functions do to auto_increment.
  In reset|restore_sub_statement_state(), we need to backup less
  auto_inc variables as some of them have moved to the handler;
  we backup/restore those which are about the current top- or sub-
  statement, *not* those about the statement-based binlog
  (which evolve as the top- and sub-statement execute).
  Because we split THD::last_insert_id into 
  THD::first_successful_insert_id_in_prev_stmt and
  THD::auto_inc_intervals_for_binlog (among others), we fix
  BUG#20341 "stored function inserting into one auto_increment
  puts bad data in slave": indeed we can afford to not backup/restore
  THD::auto_inc_intervals_for_binlog (which fixes the bug) while still
  backing up / restoring THD::first_successful_insert_id_in_prev_stmt
  (ensuring that the top-level LAST_INSERT_ID() is not affected by INSERTs
  done by sub-statements, as is desirable and tested in rpl_insert_id.test).
sql/sql_class.h:
  new variables and methods for auto_increment.
  Some THD members move into handler (those which are really about
  the table being inserted), some stay in THD (those which are
  about what a future LAST_INSERT_ID() should return, or about
  what should be stored into the statement-based binlog).
  THD::next_insert_id moves to handler::.
  THD::clear_next_insert_id removed (had become equivalent
  to next_insert_id > 0).
  THD::last_insert_id becomes four:
  THD::first_successful_insert_id_in_cur_stmt,
  THD::auto_inc_intervals_for_binlog,
  handler::insert_id_for_cur_row,
  THD::first_successful_insert_id_in_prev_stmt.
  THD::current_insert_id becomes:
  THD::first_successful_insert_id_in_prev_stmt_for_binlog
  THD::prev_insert_id is removed, handler can just use
  handler::insert_id_for_cur_row instead (which is more accurate:
  for the first row, prev_insert_id was set before get_auto_increment
  was called, so was 0, causing a call to
  get_auto_increment() for the 2nd row if the 1st row fails;
  here we don't need the call as insert_id_for_cur_row has
  the value of the first row).
  THD::last_insert_id_used becomes: stmt_depends_on_first_row_in_prev_stmt
  THD::insert_id_used is removed (equivalent to
  auto_inc_intervals_for_binlog non empty).
  The interval returned by get_auto_increment() and currently being
  consumed is handler::auto_inc_interval_for_cur_row.
  Comments to explain each of them.
  select_insert::last_insert_id becomes autoinc_value_of_last_inserted_row.
sql/sql_insert.cc:
  the "id" variable is not changed for each row now; it used to compensate for
  this contradiction:
  - thd->last_insert_id supposed job was to keep the id of the first row
  - but it was updated for every row
  - so mysql_insert() made sure to catch its first value and restore it at the end of stmt.
  Now THD keeps the first value in first_successful_insert_id_in_cur_stmt,
  and value of the row in insert_id_for_cur_row. So "id" only serves to fill
  mysql_insert_id(), as depending on some conditions, "id" must be different
  values.
  Prev_insert_id moves from THD to write_record().
  We now set LAST_INSERT_ID() in ON DUPLICATE KEY UPDATE too (BUG#19243).
  In an INSERT DELAYED, we still "reset auto-increment caching" but differently
  (by calling ha_release_auto_increment()).
sql/sql_load.cc:
  no need to fiddle with "id", THD maintains
  THD::first_successful_insert_id_in_cur_stmt by itself and correctly now.
  ha_release_auto_increment() is now (logically) called before we unlock
  the table.
sql/sql_parse.cc:
  update to new variable names.
  Assertion that reset_thd_for_next_command() is not called for every
  substatement of a routine (I'm not against it, but if we do this change,
  statement-based binlogging needs some adjustments).
sql/sql_select.cc:
  update for new variable names
sql/sql_table.cc:
  next_insert_id not needed in mysql_alter_table(), THD manages.
sql/sql_update.cc:
  update for new variable names.
  Even though this is UPDATE, an insert id can be generated (by
  LAST_INSERT_ID(X)) and should be recorded because mysql_insert_id() wants
  to know about it.
sql/structs.h:
  A class for "discrete" intervals (intervals of integer numbers with a certain
  increment between them): Discrete_interval, and a class for a list of such
  intervals: Discrete_intervals_list
tests/mysql_client_test.c:
  tests of behaviour of mysql_insert_id(): there were no such tests, while in
  our manual we document its behaviour. In comments you'll notice the behaviour
  changes introduced (there are 5).
2006-07-09 17:52:19 +02:00

348 lines
11 KiB
C++

/* Copyright (C) 2000 MySQL AB & MySQL Finland AB & TCX DataKonsult 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; either version 2 of the License, or
(at your option) any later version.
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 */
/* The old structures from unireg */
struct st_table;
class Field;
typedef struct st_date_time_format {
uchar positions[8];
char time_separator; /* Separator between hour and minute */
uint flag; /* For future */
LEX_STRING format;
} DATE_TIME_FORMAT;
typedef struct st_keyfile_info { /* used with ha_info() */
byte ref[MAX_REFLENGTH]; /* Pointer to current row */
byte dupp_ref[MAX_REFLENGTH]; /* Pointer to dupp row */
uint ref_length; /* Length of ref (1-8) */
uint block_size; /* index block size */
File filenr; /* (uniq) filenr for table */
ha_rows records; /* Records i datafilen */
ha_rows deleted; /* Deleted records */
ulonglong data_file_length; /* Length off data file */
ulonglong max_data_file_length; /* Length off data file */
ulonglong index_file_length;
ulonglong max_index_file_length;
ulonglong delete_length; /* Free bytes */
ulonglong auto_increment_value;
int errkey,sortkey; /* Last errorkey and sorted by */
time_t create_time; /* When table was created */
time_t check_time;
time_t update_time;
ulong mean_rec_length; /* physical reclength */
} KEYFILE_INFO;
typedef struct st_key_part_info { /* Info about a key part */
Field *field;
uint offset; /* offset in record (from 0) */
uint null_offset; /* Offset to null_bit in record */
uint16 length; /* Length of keypart value in bytes */
/*
Number of bytes required to store the keypart value. This may be
different from the "length" field as it also counts
- possible NULL-flag byte (see HA_KEY_NULL_LENGTH)
- possible HA_KEY_BLOB_LENGTH bytes needed to store actual value length.
*/
uint16 store_length;
uint16 key_type;
uint16 fieldnr; /* Fieldnum in UNIREG */
uint16 key_part_flag; /* 0 or HA_REVERSE_SORT */
uint8 type;
uint8 null_bit; /* Position to null_bit */
} KEY_PART_INFO ;
typedef struct st_key {
uint key_length; /* Tot length of key */
uint flags; /* dupp key and pack flags */
uint key_parts; /* How many key_parts */
uint extra_length;
uint usable_key_parts; /* Should normally be = key_parts */
uint block_size;
enum ha_key_alg algorithm;
/*
Note that parser is used when the table is opened for use, and
parser_name is used when the table is being created.
*/
union
{
struct st_plugin_int *parser; /* Fulltext [pre]parser */
LEX_STRING *parser_name; /* Fulltext [pre]parser name */
};
KEY_PART_INFO *key_part;
char *name; /* Name of key */
/*
Array of AVG(#records with the same field value) for 1st ... Nth key part.
0 means 'not known'.
For temporary heap tables this member is NULL.
*/
ulong *rec_per_key;
union {
int bdb_return_if_eq;
} handler;
struct st_table *table;
} KEY;
struct st_join_table;
typedef struct st_reginfo { /* Extra info about reg */
struct st_join_table *join_tab; /* Used by SELECT() */
enum thr_lock_type lock_type; /* How database is used */
bool not_exists_optimize;
bool impossible_range;
} REGINFO;
struct st_read_record; /* For referense later */
class SQL_SELECT;
class THD;
class handler;
typedef struct st_read_record { /* Parameter to read_record */
struct st_table *table; /* Head-form */
handler *file;
struct st_table **forms; /* head and ref forms */
int (*read_record)(struct st_read_record *);
THD *thd;
SQL_SELECT *select;
uint cache_records;
uint ref_length,struct_length,reclength,rec_cache_size,error_offset;
uint index;
byte *ref_pos; /* pointer to form->refpos */
byte *record;
byte *rec_buf; /* to read field values after filesort */
byte *cache,*cache_pos,*cache_end,*read_positions;
IO_CACHE *io_cache;
bool print_error, ignore_not_found_rows;
} READ_RECORD;
/*
Originally MySQL used TIME structure inside server only, but since
4.1 it's exported to user in the new client API. Define aliases for
new names to keep existing code simple.
*/
typedef struct st_mysql_time TIME;
typedef enum enum_mysql_timestamp_type timestamp_type;
typedef struct {
ulong year,month,day,hour;
ulonglong minute,second,second_part;
bool neg;
} INTERVAL;
typedef struct st_known_date_time_format {
const char *format_name;
const char *date_format;
const char *datetime_format;
const char *time_format;
} KNOWN_DATE_TIME_FORMAT;
enum SHOW_COMP_OPTION { SHOW_OPTION_YES, SHOW_OPTION_NO, SHOW_OPTION_DISABLED};
extern const char *show_comp_option_name[];
typedef int *(*update_var)(THD *, struct st_mysql_show_var *);
typedef struct st_lex_user {
LEX_STRING user, host, password;
} LEX_USER;
/*
This structure specifies the maximum amount of resources which
can be consumed by each account. Zero value of a member means
there is no limit.
*/
typedef struct user_resources {
/* Maximum number of queries/statements per hour. */
uint questions;
/*
Maximum number of updating statements per hour (which statements are
updating is defined by sql_command_flags array).
*/
uint updates;
/* Maximum number of connections established per hour. */
uint conn_per_hour;
/* Maximum number of concurrent connections. */
uint user_conn;
/*
Values of this enum and specified_limits member are used by the
parser to store which user limits were specified in GRANT statement.
*/
enum {QUERIES_PER_HOUR= 1, UPDATES_PER_HOUR= 2, CONNECTIONS_PER_HOUR= 4,
USER_CONNECTIONS= 8};
uint specified_limits;
} USER_RESOURCES;
/*
This structure is used for counting resources consumed and for checking
them against specified user limits.
*/
typedef struct user_conn {
/*
Pointer to user+host key (pair separated by '\0') defining the entity
for which resources are counted (By default it is user account thus
priv_user/priv_host pair is used. If --old-style-user-limits option
is enabled, resources are counted for each user+host separately).
*/
char *user;
/* Pointer to host part of the key. */
char *host;
/* Total length of the key. */
uint len;
/* Current amount of concurrent connections for this account. */
uint connections;
/*
Current number of connections per hour, number of updating statements
per hour and total number of statements per hour for this account.
*/
uint conn_per_hour, updates, questions;
/* Maximum amount of resources which account is allowed to consume. */
USER_RESOURCES user_resources;
/*
The moment of time when per hour counters were reset last time
(i.e. start of "hour" for conn_per_hour, updates, questions counters).
*/
time_t intime;
} USER_CONN;
/* Bits in form->update */
#define REG_MAKE_DUPP 1 /* Make a copy of record when read */
#define REG_NEW_RECORD 2 /* Write a new record if not found */
#define REG_UPDATE 4 /* Uppdate record */
#define REG_DELETE 8 /* Delete found record */
#define REG_PROG 16 /* User is updating database */
#define REG_CLEAR_AFTER_WRITE 32
#define REG_MAY_BE_UPDATED 64
#define REG_AUTO_UPDATE 64 /* Used in D-forms for scroll-tables */
#define REG_OVERWRITE 128
#define REG_SKIP_DUP 256
/* Bits in form->status */
#define STATUS_NO_RECORD (1+2) /* Record isn't usably */
#define STATUS_GARBAGE 1
#define STATUS_NOT_FOUND 2 /* No record in database when needed */
#define STATUS_NO_PARENT 4 /* Parent record wasn't found */
#define STATUS_NOT_READ 8 /* Record isn't read */
#define STATUS_UPDATED 16 /* Record is updated by formula */
#define STATUS_NULL_ROW 32 /* table->null_row is set */
#define STATUS_DELETED 64
/*
Such interval is "discrete": it is the set of
{ auto_inc_interval_min + k * increment,
0 <= k <= (auto_inc_interval_values-1) }
Where "increment" is maintained separately by the user of this class (and is
currently only thd->variables.auto_increment_increment).
It mustn't derive from Sql_alloc, because SET INSERT_ID needs to
allocate memory which must stay allocated for use by the next statement.
*/
class Discrete_interval {
private:
ulonglong interval_min;
ulonglong interval_values;
ulonglong interval_max; // excluded bound. Redundant.
public:
Discrete_interval *next; // used when linked into Discrete_intervals_list
void replace(ulonglong start, ulonglong val, ulonglong incr)
{
interval_min= start;
interval_values= val;
interval_max= (val == ULONGLONG_MAX) ? val : start + val * incr;
}
Discrete_interval(ulonglong start, ulonglong val, ulonglong incr) :
next(NULL) { replace(start, val, incr); };
Discrete_interval() : next(NULL) { replace(0, 0, 0); };
ulonglong minimum() const { return interval_min; };
ulonglong values() const { return interval_values; };
ulonglong maximum() const { return interval_max; };
/*
If appending [3,5] to [1,2], we merge both in [1,5] (they should have the
same increment for that, user of the class has to ensure that). That is
just a space optimization. Returns 0 if merge succeeded.
*/
bool merge_if_contiguous(ulonglong start, ulonglong val, ulonglong incr)
{
if (interval_max == start)
{
if (val == ULONGLONG_MAX)
{
interval_values= interval_max= val;
}
else
{
interval_values+= val;
interval_max= start + val * incr;
}
return 0;
}
return 1;
};
};
/* List of Discrete_interval objects */
class Discrete_intervals_list {
private:
Discrete_interval *head;
Discrete_interval *tail;
/*
When many intervals are provided at the beginning of the execution of a
statement (in a replication slave or SET INSERT_ID), "current" points to
the interval being consumed by the thread now (so "current" goes from
"head" to "tail" then to NULL).
*/
Discrete_interval *current;
uint elements; // number of elements
public:
Discrete_intervals_list() : head(NULL), current(NULL), elements(0) {};
void empty_no_free()
{
head= current= NULL;
elements= 0;
}
void empty()
{
for (Discrete_interval *i= head; i;)
{
Discrete_interval *next= i->next;
delete i;
i= next;
}
empty_no_free();
}
const Discrete_interval* get_next()
{
Discrete_interval *tmp= current;
if (current != NULL)
current= current->next;
return tmp;
}
~Discrete_intervals_list() { empty(); };
bool append(ulonglong start, ulonglong val, ulonglong incr);
ulonglong minimum() const { return (head ? head->minimum() : 0); };
ulonglong maximum() const { return (head ? tail->maximum() : 0); };
uint nb_elements() const { return elements; }
};