------------------------------------------------------------------------
r2537 | inaam | 2008-07-15 20:46:03 +0300 (Tue, 15 Jul 2008) | 12 lines
branches/5.1 issue# 4
Fixed a timing hole where a thread dropping an index can free the
in-memory index struct while another thread is still using
that structure to remove entries from adaptive hash index belonging
to one of the pages that belongs to the index being dropped.
The fix is to have a reference counter in the index struct and to
wait for this counter to drop to zero beforing freeing the struct.
Reviewed by: Heikki
------------------------------------------------------------------------
Add a space in a warning message. No functional change to the code.
The original message is "invalid innodb_file_format_checkvalue;...".
The new one is "invalid innodb_file_format_check value;..."
Fixes issue# 51 - by removing the negative test cases that produce
warnings. The warnings cause mysql-test-run to fail.
Those negative test cases will be moved to separate test/result files.
Approved by: Sunny
Disable part of innodb-index test because MySQL changed its behavior and
is not calling ::add_index() anymore in the following ALTER TABLE:
CREATE TABLE t (a INT) ENGINE=INNODB;
INSERT INTO t VALUES (NULL);
ALTER TABLE t ADD PRIMARY KEY (a); -- adding primary index on non-NULL column
Previously, in ALTER TABLE, MySQL would call ::add_index() which would
fail with a "primary key cannot contain NULL" error.
The change occured in:
mysql-5.1$ bzr log -v -r2667
------------------------------------------------------------
revno: 2667
committer: Davi Arnaut <davi@mysql.com>
branch nick: 33873-5.1
timestamp: Tue 2008-06-17 11:12:21 -0300
message:
Bug#33873: Fast ALTER TABLE doesn't work with multibyte character sets
The problem was that when comparing tables for a possible
fast alter table, the comparison was being performed using
the parsed information and not the final definition.
The solution is to use the possible final table layout to
compare if a fast alter is possible or not.
modified:
mysql-test/include/mix1.inc
mysql-test/r/alter_table.result
mysql-test/r/innodb_mysql.result
mysql-test/t/alter_table.test
sql/sql_table.cc
mysql-5.1$
------------------------------------------------------------------------
r2519 | sunny | 2008-06-26 16:55:43 +0300 (Thu, 26 Jun 2008) | 5 lines
branches/5.1: Add test cases and fix a bug where the last AUTOINC cached value
was not reset to 0 when the table was truncated.
Bug #37531 : After truncate, auto_increment behaves incorrectly for InnoDB
------------------------------------------------------------------------
r2520 | vasil | 2008-06-26 17:38:02 +0300 (Thu, 26 Jun 2008) | 7 lines
branches/5.1:
Fix Bug#36941 Performance problem in ha_print_info (SHOW INNODB STATUS)
by disabling some of the code in ha_print_info() in production builds.
Approved by: Heikki (via IM)
------------------------------------------------------------------------
r2521 | vasil | 2008-06-26 17:39:01 +0300 (Thu, 26 Jun 2008) | 8 lines
branches/5.1:
Fix Bug#36942 Performance problem in lock_get_n_rec_locks (SHOW INNODB STATUS)
by not calling lock_get_n_rec_locks() from lock_print_info_summary() on
production builds.
Approved by: Heikki (via IM)
------------------------------------------------------------------------
r2524 | vasil | 2008-07-01 10:37:34 +0300 (Tue, 01 Jul 2008) | 29 lines
branches/5.1:
Merge a change from MySQL (this fixes the failing innodb-replace test):
revno: 2659
committer: Mattias Jonsson <mattiasj@mysql.com>
branch nick: b31210-51-bugteam
timestamp: Tue 2008-06-03 13:25:41 +0200
message:
Bug#31210: INSERT DELAYED crashes server when used on partitioned tables
Problem was an unclear error message since it could suggest that
MyISAM did not support INSERT DELAYED.
Changed the error message to say that DELAYED is not supported by the
table, instead of the table's storage engine.
The confusion is that a partitioned table is in somewhat sense using
the partitioning storage engine, which in turn uses the ordinary
storage engine. By saying that the table does not support DELAYED we
do not give any extra informantion about the storage engine or if it
is partitioned.
modified:
mysql-test/r/innodb-replace.result
mysql-test/t/innodb-replace.test
mysql-test/t/merge.test
mysql-test/t/partition_hash.test
sql/share/errmsg.txt
sql/sql_insert.cc
------------------------------------------------------------------------
Rewrite the function innodb_plugin_init() to support parameters in
different order (in static and dynamic InnoDB) and to support more
parameters in the static InnoDB.
The previous implementation traversed both lists at the same time,
skipping parameters in the dynamic list that do not exist in the static
list. E.g. both lists were allowed to be static=(a, b, c),
dynamic=(a, b, c) or static=(a, b, c), dynamic=(a, b, x, c).
With the new implementation they are allowed to be
static=(a, b, c), dynamic=(b, a, c) or
static=(a, b, x, c), dynamic=(b, a, c) in addition.
The new implementation has complexity O(N^2) while the old one was O(N),
but this is acceptable sacrifice provided that innodb_plugin_init() is
called once per InnoDB lifetime and that N=39 currently, and N is not
going to increase much in the future, N=number of the InnoDB parameters.
Suggested by: Sunny
Approved by: Sunny
Fix the rpl_ddl mysql-test that produces a warnings:
master.err: rpl.rpl_ddl: 080611 16:27:01 [Warning] MySQL is closing a connection that has an active InnoDB transaction. 13 row modifications will roll back.
the bug was introduced in r2256, which adds "prebuilt->trx->active_trans = 1;"
in ::add_index() and ::final_drop_index() but never calls innobase_commit() or
innobase_rollback() to set back active_trans to 0.
Also, per Heikki's suggestion we remove the calls to trans_register_ha() from
::add_index() and ::final_drop_index(). Following is the IM discussion that
took place about this:
--- cut ---
(17:59:05) vasil:
unsigned active_trans:2; /* 1 - if a transaction in MySQL
is active. 2 - if prepare_commit_mutex
was taken */
(17:59:13) vasil: what about 0?
(17:59:26) Heikki: Hmm
(17:59:42) Heikki: Maybe in the other cases :)
(18:01:00) vasil:
2 1369 handler/ha_innodb.cc <<<unknown>>>
if (trx->active_trans == 0) {
3 1372 handler/ha_innodb.cc <<<unknown>>>
trx->active_trans = 1;
(18:01:05) vasil: yes, it can be 0
(18:01:38) Heikki: Maybe no active transaction => the value is 0
(18:02:03) vasil: probably yes
(18:09:09) vasil:
/* The flag trx->active_trans is set to 1 in
1. ::external_lock(),
2. ::start_stmt(),
3. innobase_query_caching_of_table_permitted(),
4. innobase_savepoint(),
5. ::init_table_handle_for_HANDLER(),
6. innobase_start_trx_and_assign_read_view(),
7. ::transactional_table_lock()
and it is only set to 0 in a commit or a rollback.
(18:09:14) vasil: this is no longer true
(18:09:35) vasil: it is also set to 1 in ::final_drop_index
(18:09:56) vasil: and in ::add_index()
(18:12:31) Heikki: Please correct the comment :(
(18:13:01) vasil: it is not only the comment, I am trying to understand what's wrong
(18:13:28) Heikki: What is the symptom?
(18:14:37) vasil: can you find this email and the subsequent one:
From: Vasil Dimov <vasil.dimov@oracle.com>
To: InnoDB Dev <innodb_dev_ww@oracle.com>
Subject: Re: innodb - marko - r2256 - in branches/zip: handler include lock
trx
Date: Wed, 11 Jun 2008 16:33:47 +0300
(18:15:02) vasil: the symptom is
master.err: rpl.rpl_ddl: 080611 16:27:01 [Warning] MySQL is closing a connection that has
+an active InnoDB transaction. 13 row modifications will roll back.
(18:16:22) vasil: My understanding till now is that active_trans is set to 1 in ::add_index() and forgotten like this, never set to 0; I wonder if it should be set to 0 somewhere at the end or not set to 1 at all... or there is more...
(18:16:31) Heikki: Vasil, ok the problem is that Marko forgets to commit the transaction
(18:16:55) Heikki: An active transaction has to be committed or rolled back
(18:17:09) vasil: but i can find this in ::add_index():
/* Commit the data dictionary transaction in order to release
the table locks on the system tables. Unfortunately, this
means that if MySQL crashes while creating a new primary key
inside row_merge_build_indexes(), indexed_table will not be
dropped on crash recovery. Thus, it will become orphaned. */
trx_commit_for_mysql(trx);
(18:17:27) vasil: but this function does not set active_trans to 0
(18:17:43) Heikki: Let me look
(18:19:20) vasil: it is only set to 0 in innobase_commit() and in innobase_rollback()
(18:20:10) Heikki: Ok
(18:20:59) Heikki: Hmm...
(18:21:08) vasil: innobase_commit() calls innobase_commit_low() calls trx_commit_for_mysql()
innobase_commit() sets active_trans=0
(18:21:26) Heikki: active_trans is strictly a ha_innodb.cc variable
(18:21:53) Heikki: It is intended to be used in transactions visible to MySQL
(18:22:25) Heikki: I think Marko should call innobase_commit()
(18:23:51) vasil: but still set active_trans=1 inside :add_index() inside handler/handler0alter.cc?
(18:24:12) Heikki: Hmm
(18:24:56) Heikki: I think the transaction is NOT really visible to MySQL
(18:24:57) vasil:
> Author: marko
> Date: 2008-01-25 16:26:07 +0200 (Fri, 25 Jan 2008)
> New Revision: 2256
>
> Log:
> branches/zip: Fast index creation: Release locks on system tables before
> creating indexes. Lock the user table inside the user transaction.
>
> enum trx_dict_op: Remove TRX_OP_INDEX_MAY_WAIT.
>
> ha_innobase::add_index(): Lock the user tables within prebuilt->trx.
> Commit the data dictionary transaction before creating indexes.
>
> ha_innobase::final_drop_index(): Lock the user table within prebuilt->trx.
(18:26:41) Heikki: Let me look at the old ::create_table()
(18:27:25) Heikki: Ok, in the old code, the transactions are internal to InnoDB, not visible to MySQL
(18:27:46) Heikki: I think it is an error to set active_trans=1 inside :add_index() i
(18:28:11) Heikki: MySQL is not really interested in HOW the engine adds the index. In a transaction or not.
(18:28:26) vasil: this is the change inside drop_index(), that is supposed to "Lock the user table within prebuilt->trx":
(18:28:37) vasil:
@@ -1093,27 +1101,25 @@ ha_innobase::final_drop_index(
/* Create a background transaction for the operations on
the data dictionary tables. */
trx = trx_allocate_for_mysql();
trx_start_if_not_started(trx);
trans_register_ha(user_thd, FALSE, ht);
+ prebuilt->trx->active_trans = 1;
trx->mysql_thd = user_thd;
trx->mysql_query_str = thd_query(user_thd);
/* Flag this transaction as a dictionary operation, so that
- the data dictionary will be locked in crash recovery. Prevent
- warnings if row_merge_lock_table() results in a lock wait,
- i.e., when another transaction is holding a conflicting lock
- on the table, e.g., because of SELECT ... FOR UPDATE. */
- trx_set_dict_operation(trx, TRX_DICT_OP_INDEX_MAY_WAIT);
+ the data dictionary will be locked in crash recovery. */
+ trx_set_dict_operation(trx, TRX_DICT_OP_INDEX);
/* Lock the table exclusively, to ensure that no active
transaction depends on an index that is being dropped. */
err = convert_error_code_to_mysql(
- row_merge_lock_table(trx, prebuilt->table, LOCK_X),
+ row_merge_lock_table(prebuilt->trx, prebuilt->table, LOCK_X),
user_thd);
if (UNIV_UNLIKELY(err)) {
/* Unmark the indexes to be dropped. */
row_mysql_lock_data_dictionary(trx);
@@ -1125,14 +1131,12 @@ ha_innobase::final_drop_index(
}
row_mysql_unlock_data_dictionary(trx);
goto func_exit;
}
- trx_set_dict_operation(trx, TRX_DICT_OP_INDEX);
-
/* Drop indexes marked to be dropped */
row_mysql_lock_data_dictionary(trx);
index = dict_table_get_first_index(prebuilt->table);
@@ -1159,12 +1163,13 @@ ha_innobase::final_drop_index(
dict_table_check_for_dup_indexes(prebuilt->table);
#endif
row_mysql_unlock_data_dictionary(trx);
func_exit:
trx_commit_for_mysql(trx);
+ trx_commit_for_mysql(prebuilt->trx);
/* Flush the log to reduce probability that the .frm files and
the InnoDB data dictionary get out-of-sync if the user runs
with innodb_flush_log_at_trx_commit = 0 */
log_buffer_flush_to_disk();
(18:28:43) Heikki: Let me look
(18:29:05) vasil: this is the change inside drop_index(), that is supposed to "Lock the user table within prebuilt->trx"
(18:29:26) Heikki: trans_register_ha(user_thd, FALSE, ht);
+ prebuilt->trx->active_trans = 1;
(18:29:34) Heikki: I think the above code should be removed
(18:29:53) Heikki: MySQL is not interested in InnoDB's transaction in this case
(18:30:33) Heikki: Alternatively, we should use innobase_commit() instead of trx_commit_...
(18:31:01) vasil: but there are many places in ::add_index() where it could return before reaching the trx_commit...
(18:31:12) vasil: leaving active_traans=1
(18:31:56) Heikki: Does Marko handle the errors with trx_rollback()?
(18:32:08) vasil: let me see..
(18:32:30) Heikki: It is risky to leave error handling to MySQL :(
(18:32:35) vasil:
651 if (UNIV_UNLIKELY(error)) {
652 err_exit:
653 mem_heap_free(heap);
654 trx_general_rollback_for_mysql(trx, FALSE, NULL);
655 trx_free_for_mysql(trx);
656 trx_commit_for_mysql(prebuilt->trx);
657 DBUG_RETURN(error);
658 }
(18:32:43) Heikki: Ok
(18:33:25) Heikki: Please remove the ha_register... and active_trx =1 ....
(18:33:42) Heikki: Then check if it prints something
(18:34:08) Heikki: I believe MySQL is not interested in how the engine does the index build
(18:34:11) vasil: why remove ha_register?
(18:34:28) Heikki: Because there is no transaction visible to MySQL
(18:34:53) vasil: the ha_register call was there from before
(18:35:09) Heikki: Registers that InnoDB takes part in an SQL statement, so that MySQL knows to
roll back the statement if the statement results in an error. This MUST be
called for every SQL statement that may be rolled back by MySQL. Calling this
several times to register the same statement is allowed, too.
(18:36:01) Heikki: Hmm
(18:36:17) Heikki: Obviously, MySQL does NOT call commit at the end
(18:36:36) Heikki: Because then innobase_commit() would set active_trx=0!
(18:37:00) Heikki: Ok, maybe MySQL thinks CREATE INDEX is NOT a transactional statement
(18:37:21) Heikki: Then, .._register_.. should be removed
(18:37:51) Heikki: It is logical this way. If it also passes test, even better
(18:38:21) vasil: ha_registrer in add_index() was added in r1845 and in drop_index() in r2128, let me read the comments..
(18:38:55) vasil:
------------------------------------------------------------------------
r1845 | marko | 2007-09-13 13:31:54 +0300 (Thu, 13 Sep 2007) | 3 lines
branches/zip: Move the code related to fast index creation
(smart ALTER TABLE) from ha_innodb.cc to a separate module, handler0alter.cc.
(18:39:05) Heikki: One thing you must check is that MySQL does write the CREATE INDEX in the binlog, even if we remove the _register_
(18:39:41) Heikki: I assume Marko did not know if MySQL thinks CREATE INDEX is transactional or not
(18:39:49) Heikki: He thought it is transactional
(18:39:53) vasil:
------------------------------------------------------------------------
r2128 | marko | 2007-11-29 12:34:55 +0200 (Thu, 29 Nov 2007) | 5 lines
branches/zip: ha_innobase::final_drop_index(): Allocate a separate transaction
for dropping the index trees, and set the dictionary operation flag, similar
to what ha_innobase::add_index() does. This should ensure correct crash
recovery.
(18:40:10) Heikki: Ok
(18:40:17) vasil: how do you know CREATE INDEX is not transactional?
(18:40:33) Heikki: You cannot roll it back
(18:40:44) vasil: ok
(18:41:03) Heikki: MySQL devs probably thought this way
(18:44:15) vasil: hmm
(18:44:21) vasil: about r1845
(18:45:00) vasil: the comments says "move the code" but it is actually "move and change the code"
(18:45:28) Heikki: Well
(18:45:29) vasil: the code that is deleted from ha_innodb is this:
(18:45:32) vasil:
- update_thd(ha_thd());
-
- heap = mem_heap_create(1024);
-
- /* In case MySQL calls this in the middle of a SELECT query, release
- possible adaptive hash latch to avoid deadlocks of threads. */
- trx_search_latch_release_if_reserved(check_trx_exists(user_thd));
-
- trx = trx_allocate_for_mysql();
- trx_start_if_not_started(trx);
-
- innobase_register_stmt(ht, user_thd);
-
- trx->mysql_thd = user_thd;
(18:45:38) vasil: (in add_index)
(18:45:48) inaam left the room.
(18:45:52) vasil: but the new one in handler0alter is this:
(18:46:16) vasil:
+ update_thd(ha_thd());
+
+ heap = mem_heap_create(1024);
+
+ /* In case MySQL calls this in the middle of a SELECT query, release
+ possible adaptive hash latch to avoid deadlocks of threads. */
+ trx_search_latch_release_if_reserved(check_trx_exists(user_thd));
+
+ trx = trx_allocate_for_mysql();
+ trx_start_if_not_started(trx);
+
+ trans_register_ha(user_thd, FALSE, ht);
+
+ trx->mysql_thd = user_thd;
(18:46:48) vasil: innobase_register_stmt(ht, user_thd); was substituted with trans_register_ha(user_thd, FALSE, ht); during the move
(18:46:49) Heikki: I do not see how MySQL could call it in the middle of a SELECT query!
(18:46:59) Heikki: Yes
(18:47:21) Heikki: Did Marko change innobase_register_stmt() code too?
(18:48:05) vasil: not in this commit r1845
(18:48:54) Heikki: Anyway, please test a version where you remove the register thing
(18:49:08) vasil: and active_trans=1?
(18:49:15) Heikki: Remove that, too
(18:49:35) Heikki: innobase_register_stmt(
/*===================*/
handlerton* hton, /* in: Innobase hton */
THD* thd) /* in: MySQL thd (connection) object */
{
DBUG_ASSERT(hton == innodb_hton_ptr);
/* Register the statement */
trans_register_ha(thd, FALSE, hton);
}
(18:50:23) Heikki: looks pretty much the same as what Marko has in the later version
(18:50:43) vasil: innobase_register_stmt() was added in add_index() in r1584
(18:50:47) vasil:
r1584 | marko | 2007-06-18 15:46:42 +0300 (Mon, 18 Jun 2007) | 12 lines
branches/zip: ha_innobase::add_index(): Split some assertions.
Remove the variable parent_trx.
Call innobase_register_stmt() in order to work around a MySQL bug
in mysql_alter_table(), which, as of
ChangeSet@1.2482.61.2, 2007-06-07 16:37:15+02:00, joerg@trift2. +8 -0
commits the transaction before calling ha_innobase::add_index().
Without re-registering the statement, the ha_commit_stmt(thd)
in mysql_alter_table() would not invoke innobase_commit.
(18:51:56) Heikki: But it does not seem to invoke it now, because active_trans is left 1!
(18:54:15) Heikki: Apparently, there has been confusion also among MySQL devs whether CREATE INDEX is transactional or not
(18:54:45) vasil: :-(
(18:54:53) Heikki: The most robust way is to handle all commits and rollbacks internally inside InnoDB and not tell MySQL anything about InnoDB's internal transactions
(18:55:01) Heikki: That is my idea
(18:55:20) Heikki: Then MySQL cannot spoil the logic of the code
(18:55:59) vasil: mysql_alter_table() is ~1100 lines
(18:56:08) vasil: and you suggest what?
(18:56:43) Heikki: I suggest removing the ..._register_... thing and active_trans = 1
(18:57:08) Heikki: Those things tell MySQL that there is a transaction going on
(18:57:14) Heikki: Better not to tell
(18:57:29) vasil: and possibly ignoring the above comment in r1584, "Call innobase_register_stmt() in order to work around a MySQL bug
in mysql_alter_table()........."
(18:57:32) vasil: ?
(18:57:36) Heikki: Yes
(18:57:38) vasil: ok
(18:57:44) vasil: same for drop_index?
(18:57:48) Heikki: Yes
(18:58:51) vasil: testing
Index: handler/handler0alter.cc
===================================================================
--- handler/handler0alter.cc (revision 2498)
+++ handler/handler0alter.cc (working copy)
@@ -632,14 +632,14 @@ ha_innobase::add_index(
/* Create a background transaction for the operations on
the data dictionary tables. */
trx = trx_allocate_for_mysql();
trx_start_if_not_started(trx);
- trans_register_ha(user_thd, FALSE, ht);
- prebuilt->trx->active_trans = 1;
+ //trans_register_ha(user_thd, FALSE, ht);
+ //prebuilt->trx->active_trans = 1;
trx->mysql_thd = user_thd;
trx->mysql_query_str = thd_query(user_thd);
innodb_table = indexed_table
= dict_table_get(prebuilt->table->name, FALSE);
@@ -1081,14 +1081,14 @@ ha_innobase::final_drop_index(
/* Create a background transaction for the operations on
the data dictionary tables. */
trx = trx_allocate_for_mysql();
trx_start_if_not_started(trx);
- trans_register_ha(user_thd, FALSE, ht);
- prebuilt->trx->active_trans = 1;
+ //trans_register_ha(user_thd, FALSE, ht);
+ //prebuilt->trx->active_trans = 1;
trx->mysql_thd = user_thd;
trx->mysql_query_str = thd_query(user_thd);
/* Flag this transaction as a dictionary operation, so that
the data dictionary will be locked in crash recovery. */
(18:59:07) vasil: not warnings
(18:59:22) Heikki: Did it create the indexes
(18:59:36) vasil: I did test only removing active_trans=1 and it removed the warnings too
(18:59:38) vasil: dunno
(18:59:54) Heikki: That is the most important thing to check :)
(19:00:10) vasil: :)
(19:00:38) vasil: it is some mysqltest that is 610 lines
(19:00:48) vasil: and it happens at the end
(19:01:02) vasil: lets see how may add index statements...
(19:01:11) vasil: 0
(19:01:26) Heikki: Mostly I am concerned that MySQL might work wrong without that _register_ thing
(19:03:28) vasil: could be
(19:03:49) Heikki: You need to check it still calls ::add_index()
(19:05:15) vasil: well, the change we made is inside ::add_index() how could it result in mysql stopping to call ::add_index()?
(19:05:39) vasil: if it calls it twice...
(19:05:43) Heikki: :-D
(19:08:34) vasil: I guess it must have created the index, because the test is doing SHOW INDEX after CREATE INDEX and after DROP INDEX
(19:08:43) vasil: the test would have failed if the index was not created
(19:09:10) Heikki: You need to check that a query using that index will have the right EXPLAIN and result
(19:15:04) vasil: this is with the patch:
+CREATE TABLE t (a INT) ENGINE=innodb;
+CREATE INDEX ti ON t(a);
+INSERT INTO t VALUES (1), (2), (3);
+EXPLAIN SELECT * FROM t WHERE a=2;
+id select_type table type possible_keys key key_len ref rows Extra
+1 SIMPLE t ref ti ti 5 const 1 Using where; Using index
+SELECT * FROM t WHERE a=2;
+a
+2
(19:15:15) Heikki: :)
(19:15:46) vasil: this is with the original code:
+CREATE TABLE t (a INT) ENGINE=innodb;
+CREATE INDEX ti ON t(a);
+INSERT INTO t VALUES (1), (2), (3);
+EXPLAIN SELECT * FROM t WHERE a=2;
+id select_type table type possible_keys key key_len ref rows Extra
+1 SIMPLE t ref ti ti 5 const 1 Using where; Using index
+SELECT * FROM t WHERE a=2;
+a
+2
(19:15:59) vasil: it is the same
(19:16:06) vasil: do you think this is sufficient?
(19:16:20) Heikki: No, you need to do some manual tests, too
(19:16:28) vasil: this is manual test
(19:16:37) Heikki: Several of them
(19:16:46) vasil: what should they do?
(19:17:13) Heikki: Test tables of different sizes, different indexes
(19:17:27) Heikki: CREATE several indexes at a time
(19:19:14) Heikki: Dinner time -->
--- cut ---
Approved by: Heikki
Fixed the issue where an unchanged blob column that is part of the
primary key as well is lost if it is externally stored.
The fix is provided by Marko.
Reviewed by: Sunny
This is to fix an unintended side effect of file_format_check changes.
We were reading the trx system page (TRX_SYS_PAGE_NO) before starting
recovery and that resulted in redo logs not being applied to the page.
The fix is to force a reread by calling buf_pool_invalidate().
This, however, made necessary that any writes to page are deferred to
until after the redo log application.
Reviewed by: Sunny
------------------------------------------------------------------------
r2488 | vasil | 2008-06-09 13:53:12 +0300 (Mon, 09 Jun 2008) | 15 lines
Changed paths:
M /branches/5.1/mysql-test/innodb.result
branches/5.1:
Merge change from MySQL (this almost fixes the failing innodb test):
ChangeSet@1.2613, 2008-05-07 09:58:21+04:00, sergefp@mysql.com +5 -0
BUG#35850 "Performance regression in 5.1.23/5.1.24"
- Disable the "prefer full scan on clustered primary key over full scan
of any secondary key" rule introduced by BUG#35850.
- Update test results accordingly
(bk trigger: file this for BUG#35850)
mysql-test/r/innodb.result@1.214, 2008-05-07 09:57:31+04:00, sergefp@mysql.com +15 -15
BUG#35850 "Performance regression in 5.1.23/5.1.24"
- Update test results
------------------------------------------------------------------------
r2489 | vasil | 2008-06-09 14:16:51 +0300 (Mon, 09 Jun 2008) | 20 lines
Changed paths:
M /branches/5.1/mysql-test/innodb.result
branches/5.1:
Merge change from MySQL (this fixes the failing innodb test):
ChangeSet@1.2628 2008-05-13 17:01:02+05:00 gshchepa@host.loc
Merge host.loc:/work/bugs/5.0-bugteam-30059
into host.loc:/work/bk/5.1-bugteam
NOTE: the diff for mysql-test/r/innodb.result in this ChangeSet looks
empty but the line:
Data truncated for column 'c' at row 1
is present in annotated mysql-test/r/innodb.result@1.215 [1]
and not present in annotated mysql-test/r/innodb.result@1.216 [2] so it
must have been deleted in the above ChangeSet.
[1] http://mysql.bkbits.net:8080/mysql-5.1/mysql-test/r/innodb.result?PAGE=anno&REV=4825e033JNz_463Ha9pwInSjqLvIQg
[2] http://mysql.bkbits.net:8080/mysql-5.1/mysql-test/r/innodb.result?PAGE=anno&REV=482982fcUoskWjhU2xMH2BZ4Gyukfg
------------------------------------------------------------------------
(r2478 was skipped for the obvious reason)
------------------------------------------------------------------------
r2464 | vasil | 2008-05-19 17:59:42 +0300 (Mon, 19 May 2008) | 9 lines
branches/5.1:
Fix Bug#36600 SHOW STATUS takes a lot of CPU in buf_get_latched_pages_number
by removing the Innodb_buffer_pool_pages_latched variable from
SHOW STATUS output in non-UNIV_DEBUG compilation.
Approved by: Heikki
------------------------------------------------------------------------
r2466 | calvin | 2008-05-20 01:37:14 +0300 (Tue, 20 May 2008) | 12 lines
branches/5.1:
Fix Bug#11894 innodb_file_per_table crashes w/ Windows .sym symbolic
link hack
The crash was due to un-handled error 3 (path not found). In the case
of file per table, change the call to os_file_handle_error_no_exit()
from os_file_handle_error(). Also, checks for full path pattern during
table create (Windows only), which is used in symbolic link and temp
table creation.
Approved by: Heikki
------------------------------------------------------------------------
r2478 | sunny | 2008-05-23 08:29:08 +0300 (Fri, 23 May 2008) | 3 lines
branches/5.1: Fix for bug# 36793. This is a back port from branches/zip. This
code has been tested on a big-endian machine too.
------------------------------------------------------------------------
r2480 | vasil | 2008-05-27 11:40:07 +0300 (Tue, 27 May 2008) | 11 lines
branches/5.1:
Fix Bug#36819 ut_usectime does not handle errors from gettimeofday
by retrying gettimeofday() several times if it fails in ut_usectime().
If it fails on all calls then return error to the caller to be handled
at higher level.
Update the variable innodb_row_lock_time_max in SHOW STATUS output only
if ut_usectime() was successful.
------------------------------------------------------------------------
r2482 | sunny | 2008-05-28 12:18:35 +0300 (Wed, 28 May 2008) | 5 lines
branches/5.1: Fix for Bug#35602, "Failed to read auto-increment value from
storage engine". The test for REPLACE was an error of ommission since it's
classified as a simple INSERT. For REPLACE statements we don't acquire
the special AUTOINC lock for AUTOINC_NEW_STYLE_LOCKING with this fix.
------------------------------------------------------------------------
r2485 | vasil | 2008-05-28 16:01:14 +0300 (Wed, 28 May 2008) | 9 lines
branches/5.1:
Fix Bug#36149 Read buffer overflow in srv0start.c found during "make test"
Use strncmp(3) instead of memcmp(3) to avoid reading past end of the string
if it is empty (*str == '\0'). This bug is _not_ a buffer overflow.
Discussed with: Sunny (via IM)
------------------------------------------------------------------------
page must be big enough to store two records, this is done because compression
later may very well result in two records residing on the same page. This
change handles the case where only one record fits on a page. We don't
split the page in the middle by default if there is only record on the page.
We only split the page if the tuple to be inserted is less than existing
record. That way the existing record is copied over to the right page
during the split and the new tuple is inserted to the left.
Implement the system tablespace tagging described on the wiki:
https://svn.innodb.com/innobase/InnoDB_version_and_feature_compatibility
A brief description of the changes:
* The file format tag will be saved in the trx system page, starting at
(UNIV_PAGE_SIZE - 16) for 8 bytes.
* The configuration parameter innodb_file_format_check is introduced.
This variable can be set to on/off and any of the supported file
formats in the configuration file, but can only be set to any of
the supported file formats during runtime. The default is on.
* During table create/open, check the current file format against
the max in file_format_max. If the current file format is newer,
update file_format_max and tag the system tablespace with the
newer one in a normal mtr.
* During startup, write the tag to the error log and check it against
DICT_TF_FORMAT_MAX. Refuse to start with error, if
-- DICT_TF_FORMAT_MAX < the tag, and
-- innodb_file_format_check is ON
Print out a warning , if
-- DICT_TF_FORMAT_MAX < the tag, but
-- innodb_file_format_check is off
* The system tablespace tag is re-settable using:
set innodb_file_format_check = <file_format>
Approved by: Sunny
Add InnoDB version in these places:
* In INFORMATION_SCHEMA.PLUGINS.PLUGIN_VERSION, only 1.2 out of 1.2.3
because MySQL supports only one dot there
* At startup:
"080501 12:28:06 InnoDB Plugin 1.0.1 started; log sequence number 46509"
* In a server variable innodb_version;
mysql> select @@innodb_version;
+------------------+
| @@innodb_version |
+------------------+
| 1.0.1 |
+------------------+
Approved by: Sunny
Allow InnoDB I_S plugins to be installed when InnoDB storage engine is
not installed but show empty tables when SELECTing from them in this
case. Set srv_was_started to FALSE at the end of
innobase_shutdown_for_mysql() so the I_S plugins can detect if InnoDB is
uninstalled.
This is needed to workaround the MySQL
Bug#36399 Double free bug when INFORMATION_SCHEMA plugin init function fails
and is more flexible as it allows I_S plugins to be installed before the
InnoDB plugin and detects if the latter is uninstalled after this.
Approved by: Heikki
SYNC_LEVEL_VARYING would cause an assertion failure when a thread tried
to release it. We still have to check because it's a varying lock and
could have been added later, so if the lookup fails then we simply ignore
it and return TRUE.
------------------------------------------------------------------------
r2429 | vasil | 2008-04-30 11:19:06 +0300 (Wed, 30 Apr 2008) | 10 lines
branches/5.1:
* Use INNODB_CFLAGS insead of cluttering CFLAGS with InnoDB specific
flags. CFLAGS are used to compile every file in the MySQL source tree.
* Add INNODB_DYNAMIC_CFLAGS to the flags of the dynamic plugin and use
-prefer-non-pic to make the dynamic plugin faster on i386.
Approved by: Sunny
------------------------------------------------------------------------
r2430 | vasil | 2008-04-30 11:48:35 +0300 (Wed, 30 Apr 2008) | 8 lines
branches/5.1:
Use CFLAGS instead of INNODB_CFLAGS to avoid having incompatible flags
inserted in CFLAGS and INNODB_CFLAGS and to avoid compiling mysql and
innodb with different flags.
Discussed with: Sunny
------------------------------------------------------------------------
r2431 | vasil | 2008-04-30 11:54:49 +0300 (Wed, 30 Apr 2008) | 4 lines
branches/5.1:
Fix r2430, it should be CFLAGS="$CFLAGS ..." not CFLAGS="..."
------------------------------------------------------------------------
r2432 | vasil | 2008-04-30 11:58:38 +0300 (Wed, 30 Apr 2008) | 4 lines
branches/5.1:
Non-functional change: use tabs for indentation in plug.in.
------------------------------------------------------------------------
r2433 | vasil | 2008-04-30 12:02:35 +0300 (Wed, 30 Apr 2008) | 5 lines
branches/5.1:
Add vim modeline to hint it that plug.in is a config file
so it can be colorized.
------------------------------------------------------------------------
r2434 | vasil | 2008-04-30 18:45:44 +0300 (Wed, 30 Apr 2008) | 17 lines
branches/5.1:
Merge changes from MySQL:
ChangeSet@1.2645, 2007-12-19 13:24:43+00:00, jperkin@chorlton.adsl.perkin.org.uk +14 -0
Add new pkgplugindir handling to seperate plugins from libraries,
and allow override for binary distributions. Extend mysql_config
to print compiled-in plugin location for third-party plugins to
use. Resolves bug#31736.
and
ChangeSet@1.2646, 2008-01-02 13:00:46+00:00, jperkin@chorlton.adsl.perkin.org.uk +11 -0
Use pkglibdir to simplify pkgplugindir, and fix the path in a
couple of Makefiles. Continuation of the fix for bug#31736.
------------------------------------------------------------------------
r2435 | vasil | 2008-04-30 19:04:36 +0300 (Wed, 30 Apr 2008) | 10 lines
branches/5.1:
Merge change from MySQL:
ChangeSet@1.2563, 2008-03-18 19:42:04+04:00, gluh@mysql.com +1 -0
Bug#35406 5.1-opt crashes on select from I_S.REFERENTIAL_CONSTRAINTS
added intialization of f_key_info.referenced_key_name for the case when
referenced table is dropped
------------------------------------------------------------------------
r2436 | vasil | 2008-04-30 19:15:46 +0300 (Wed, 30 Apr 2008) | 7 lines
branches/5.1:
Non-functional white space change in Makefile.am:
Use tabs for indentation and be consistent about spaces around the
equal sign.
------------------------------------------------------------------------
r2437 | vasil | 2008-04-30 20:36:11 +0300 (Wed, 30 Apr 2008) | 8 lines
branches/5.1:
Fix Bug#36434 ha_innodb.so is installed in the wrong directory
Change pkglib_LTLIBRARIES with pkgplugin_LTLIBRARIES which has been
forgotten in this commit: http://lists.mysql.com/commits/40206
------------------------------------------------------------------------
------------------------------------------------------------------------
r2386 | vasil | 2008-03-27 07:45:02 +0200 (Thu, 27 Mar 2008) | 22 lines
branches/5.1:
Merge change from MySQL (this fixes the failing innodb test):
ChangeSet@1.1810.3601.4, 2008-02-07 02:33:21+04:00, gshchepa@host.loc +9 -0
Fixed bug#30059.
Server handles truncation for assignment of too-long values
into CHAR/VARCHAR/TEXT columns in a different ways when the
truncated characters are spaces:
1. CHAR(N) columns silently ignore end-space truncation;
2. TEXT columns post a truncation warning/error in the
non-strict/strict mode.
3. VARCHAR columns always post a truncation note in
any mode.
Space truncation processing has been synchronised over
CHAR/VARCHAR/TEXT columns: current behavior of VARCHAR
columns has been propagated as standard.
Binary-encoded string/BLOB columns are not affected.
------------------------------------------------------------------------
r2387 | vasil | 2008-03-27 08:49:05 +0200 (Thu, 27 Mar 2008) | 8 lines
branches/5.1:
Check whether *trx->mysql_query_str is != NULL in addition to
trx->mysql_query_str. This adds more safety.
This may or may not fix Bug#35226 RBR event crashes slave.
------------------------------------------------------------------------
r2388 | vasil | 2008-03-27 14:02:34 +0200 (Thu, 27 Mar 2008) | 7 lines
branches/5.1:
Swap the order in which mysql_thd, mysql_query_str and *mysql_query_str
are checked for non-NULL.
Suggested by: Marko
------------------------------------------------------------------------
r2419 | vasil | 2008-04-23 19:08:06 +0300 (Wed, 23 Apr 2008) | 9 lines
branches/5.1:
Change the fix for Bug#32440 to show bytes instead of kilobytes in
INFORMATION_SCHEMA.TABLES.DATA_FREE.
Suggested by: Domas Mituzas <domas@mysql.com>
Approved by: Heikki
------------------------------------------------------------------------
r2420 | calvin | 2008-04-24 15:25:30 +0300 (Thu, 24 Apr 2008) | 4 lines
branches/5.1: Fix bug#29507 TRUNCATE shows to many rows effected
In InnoDB, the row count is only a rough estimate used by SQL
optimization. InnoDB is now return row count 0 for TRUNCATE operation.
------------------------------------------------------------------------
r2421 | calvin | 2008-04-24 15:32:30 +0300 (Thu, 24 Apr 2008) | 6 lines
branches/5.1: Fix bug#35537 - Innodb doesn't increment handler_update
and handler_delete
Add the calls to ha_statistic_increment() in ha_innobase::delete_row()
and ha_innobase::update_row().
------------------------------------------------------------------------
r2422 | vasil | 2008-04-24 16:00:30 +0300 (Thu, 24 Apr 2008) | 11 lines
branches/5.1:
Fix Bug#36169 create innodb compressed table with too large row size crashed
Sometimes it is possible that
row_drop_table_for_mysql(index->table_name, trx, FALSE); is invoked in
row_create_index_for_mysql() when the index object is freed so copy the
table name to a safe place beforehand and use the copy.
Approved by: Sunny
------------------------------------------------------------------------
Fix Mantis issue#34 https://svn.innodb.com/mantis/view.php?id=34
"I_S plugins should refuse to load if the storage engine plugin is not loaded"
and MySQL Bug#36310 http://bugs.mysql.com/36310
"InnoDB plugin crash"
* Make srv_was_started public
* Set srv_was_started at the end of innobase_start_or_create_for_mysql()
so it is not set to TRUE if innodb encounters some startup error
* Refuse to start any of the I_S plugins if srv_was_started is not TRUE
Approved by: Heikki
Fix Bug#36169 create innodb compressed table with too large row size crashed
Sometimes it is possible that
row_drop_table_for_mysql(index->table_name, trx, FALSE); is invoked in
row_create_index_for_mysql() when the index object is freed, so copy the
table name to a safe place beforehand and use the copy.
Approved by: Sunny
that in the fix but the earlier check fails because it tries to be too
strict and assumes that if the block is on the clean list then it can
only be in state io fix NONE.
described in Mantis#30. Specifically:
- Allow innodb_file_format to take string arguments
- Make innodb_file_format system variable a string instead of a number
- Implement the callback functions
- Update warning messages
Three new functions are implemented:
file_format_name_lookup(): Validate the file format name and return
its corresponding id.
innodb_file_format_check(): Check if it is a valid file format. This
function is registered as a callback with MySQL.
innodb_file_format_update(): Update the global variable using the
"saved" value. This functions is registered as a callback with MySQL.
