Skip to content

M
mariadb
Commit 228de3b0 authored by monty@tik.mysql.fi's avatar monty@tik.mysql.fi
Browse files
Options

Merge work:/my/mysql into tik.mysql.fi:/home/my/mysql

parents 599eee3e e508ad16
Show whitespace changes
Inline Side-by-side
Showing with 178 additions and 74 deletions
Docs/manual.texi
View file @ 228de3b0
......@@ -5467,9 +5467,10 @@ something that is of course not true. We could make things even worse
by just taking the test where PostgreSQL performs worst and claim that
MySQL is more than 2000 times faster than PostgreSQL.
The case is that MySQL does a lot of optimizations that PostgreSQL doesn't
do and the other way around. An SQL optimizer is a very complex thing, and
a company could spend years on just making the optimizer faster and faster.
The case is that MySQL does a lot of optimizations that PostgreSQL
doesn't do. This is of course also true the other way around. An SQL
optimizer is a very complex thing, and a company could spend years on
just making the optimizer faster and faster.
When looking at the benchmark results you should look for things that
you do in your application and just use these results to decide which
......@@ -5604,12 +5605,14 @@ MySQL with 2000 simultaneous connections doing 400 queries per second.
It sounded like he was using a Linux kernel that either had some
problems with many threads, such as kernels before 2.4, which had a problem
with this but we have documented how to fix this and Tim should be aware of
this problem. The other possible problem could have been an old glibc
library and that Tim didn't use a MySQL binary from our site, which is
linked with a corrected glibc library, but had compiled a version of his
own with. In any of the above cases, the symptom would have been exactly
what Tim had measured.
with many threads on multi-CPU machines. We have documented in this manual
how to fix this and Tim should be aware of this problem.
The other possible problem could have been an old glibc library and
that Tim didn't use a MySQL binary from our site, which is linked with
a corrected glibc library, but had compiled a version of his own with.
In any of the above cases, the symptom would have been exactly what Tim
had measured.
We asked Tim if we could get access to his data so that we could repeat
the benchmark and if he could check the MySQL version on the machine to
......@@ -5618,6 +5621,16 @@ He has not done that yet.
Because of this we can't put any trust in this benchmark either :(
Over time things also changes and the above benchmarks are not that
relevant anymore. MySQL now have a couple of different table handlers
with different speed/concurrency tradeoffs. @xref{Table types}. It
would be interesting to see how the above tests would run with the
different transactional table types in MySQL. PostgreSQL has of course
also got new features since the test was made. As the above test are
not publicly available there is no way for us to know how the
database would preform in the same tests today.
Conclusion:
The only benchmarks that exist today that anyone can download and run
......@@ -5632,15 +5645,15 @@ The thing we find strange is that every test we have seen about
PostgreSQL, that is impossible to reproduce, claims that PostgreSQL is
better in most cases while our tests, which anyone can reproduce,
clearly shows otherwise. With this we don't want to say that PostgreSQL
isn't good at many things (it is!). We would just like to see a fair test
where they are very good so that we could get some friendly competition
going!
isn't good at many things (it is!) or that it isn't faster than MySQL
under certain conditions. We would just like to see a fair test where
they are very good so that we could get some friendly competition going!
For more information about our benchmarks suite @xref{MySQL Benchmarks}.
We are working on an even better benchmark suite, including much better
documentation of what the individual tests really do, and how to add more
tests to the suite.
We are working on an even better benchmark suite, including multi user
tests, and a better documentation of what the individual tests really
do and how to add more tests to the suite.
@node TODO, , Comparisons, Introduction
......@@ -8202,7 +8215,7 @@ If your client programs are using threads, you need to also compile a
thread-safe version of the MySQL client library with the
@code{--with-thread-safe-client} configure options. This will create a
@code{libmysqlclient_r} library with which you should link your threaded
applications. @xref{Thread-safe clients}.
applications. @xref{Threaded clients}.
@item
Options that pertain to particular systems can be found in the
......@@ -38964,9 +38977,10 @@ likely it is that we can fix the problem!
* C API datatypes:: C API Datatypes
* C API function overview:: C API Function Overview
* C API functions:: C API Function Descriptions
* C Thread functions::
* C API problems:: Common questions and problems when using the C API
* Building clients:: Building Client Programs
* Thread-safe clients:: How to Make a Thread-safe Client
* Threaded clients:: How to Make a Threaded Client
@end menu
The C API code is distributed with MySQL. It is included in the
......@@ -39007,7 +39021,7 @@ the buffer associated with a connection is not decreased until the connection
is closed, at which time client memory is reclaimed.
For programming with threads, consult the 'how to make a thread-safe
client' chapter. @xref{Thread-safe clients}.
client' chapter. @xref{Threaded clients}.
@node C API datatypes, C API function overview, C, C
......@@ -39455,7 +39469,7 @@ recently invoked function that can succeed or fail, allowing you to determine
when an error occurred and what it was.
@node C API functions, C API problems, C API function overview, C
@node C API functions, C Thread functions, C API function overview, C
@subsection C API Function Descriptions
@menu
......@@ -40373,7 +40387,7 @@ of @code{mysql_field_count()} whether or not the statement was a
@code{MYSQL_FIELD_OFFSET mysql_field_seek(MYSQL_RES *result, MYSQL_FIELD_OFFSET offset)}
* Thread-safe clients:: How to Make a Thread-safe Client
* Threaded clients:: How to Make a Threaded Client
@subsubheading Description
Sets the field cursor to the given offset. The next call to
......@@ -41650,8 +41664,71 @@ The connection to the server was lost during the query.
An unknown error occurred.
@end table
@node C Thread functions, C API problems, C API functions, C
@subsection C Threaded Function Descriptions
You need to use the following functions when you want to create a
threaded client. @xref{Threaded clients}.
@menu
* my_init::
* my_thread_init()::
* my_thread_end()::
@end menu
@node my_init, my_thread_init(), C Thread functions, C Thread functions
@subsubsection @code{my_init()}
@findex @code{my_init()}
@subsubheading Description
This function needs to be called once in the program before calling any
MySQL function. This initializes some global variables that MySQL
needs. If you are using a thread safe client library, this will also
call @code{my_thread_init()} for this thread.
This is automaticly called by @code{mysql_init()}
and @code{mysql_connect()}.
@subsubheading Return Values
none.
@node my_thread_init(), my_thread_end(), my_init, C Thread functions
@subsubsection @code{my_thread_init()}
@findex @code{my_thread_init()}
@subsubheading Description
This function needs to be called for each created thread to initialize
thread specific variables.
This is automaticly called by @code{my_init()} and @code{mysql_connect()}.
@subsubheading Return Values
none.
@node my_thread_end(), , my_thread_init(), C Thread functions
@subsubsection @code{my_thread_end()}
@findex @code{my_thread_end()}
@subsubheading Description
@node C API problems, Building clients, C API functions, C
This function needs to be called before calling @code{pthread_exit()} to
freed memory allocated by @code{my_thread_init()}.
Note that this function is NOT invoked automaticly be the client
library!
@subsubheading Return Values
none.
@node C API problems, Building clients, C Thread functions, C
@subsection Common questions and problems when using the C API
@tindex @code{mysql_query()}
......@@ -41806,7 +41883,7 @@ If this happens on your system, you must include the math library by
adding @code{-lm} to the end of the compile/link line.
@node Building clients, Thread-safe clients, C API problems, C
@node Building clients, Threaded clients, C API problems, C
@subsection Building Client Programs
@cindex client programs, building
......@@ -41827,11 +41904,11 @@ For clients that use MySQL header files, you may need to specify a
files.
@node Thread-safe clients, , Building clients, C
@subsection How to Make a Thread-safe Client
@node Threaded clients, , Building clients, C
@subsection How to Make a Threaded Client
@cindex clients, thread-safe
@cindex thread-safe clients
@cindex clients, threaded
@cindex threaded clients
The client library is almost thread safe. The biggest problem is
that the subroutines in @file{net.c} that read from sockets are not
......@@ -41846,20 +41923,21 @@ Windows binaries are by default compiled to be thread safe).
Newer binary distributions should have both a normal and a
thread-safe client library.
To get a really thread-safe client where you can interrupt the client
from other threads and set timeouts when talking with the MySQL
server, you should use the @code{-lmysys}, @code{-lstring}, and @code{-ldbug}
libraries and the @code{net_serv.o} code that the server uses.
To get a threaded client where you can interrupt the client from other
threads and set timeouts when talking with the MySQL server, you should
use the @code{-lmysys}, @code{-lstring}, and @code{-ldbug} libraries and
the @code{net_serv.o} code that the server uses.
If you don't need interrupts or timeouts, you can just compile a thread
safe client library @code{(mysqlclient_r)} and use this. @xref{C,,
MySQL C API}. In this case you don't have to worry about the
@code{net_serv.o} object file or the other MySQL libraries.
When using a threaded client and you want to use timeouts and interrupts,
you can make great use of the routines in the @file{thr_alarm.c} file.
If you are using routines from the @code{mysys} library, the only thing
you must remember is to call @code{my_init()} first!
When using a threaded client and you want to use timeouts and
interrupts, you can make great use of the routines in the
@file{thr_alarm.c} file. If you are using routines from the
@code{mysys} library, the only thing you must remember is to call
@code{my_init()} first! @xref{C Thread functions}.
All functions except @code{mysql_real_connect()} are by default
thread safe. The following notes describe how to compile a thread safe
......@@ -41906,11 +41984,38 @@ If you program with POSIX threads, you can use
establish and release a mutex lock.
@end itemize
You need to know the following if you have a thread that is calling
MySQL functions, but that thread has not created the connection to the
MySQL database:
When you call @code{mysql_init()} or @code{mysql_connect()}, MySQL will
create a thread specific variable for the thread that is used by the
debug library (among other things).
If you have in a thread call a MySQL function, before a thread has
called @code{mysql_init()} or @code{mysql_connect()}, the thread will
not have the necessary thread specific variables in place and you are
likely to end up with a core dump sooner or later.
The get things to work smoothly you have to do the following:
@enumerate
@item
Call @code{my_init()} at the start of your program if it calls
any other MySQL function before calling @code{mysql_real_connect()}.
@item
Call @code{my_thread_init()} in the thread handler before calling
any MySQL function.
@item
In the thread, call @code{my_thread_end()} before calling
@code{pthread_exit()}. This will free the memory used by MySQL thread
specific variables.
@end enumerate
You may get some errors because of undefined symbols when linking your
client with @code{mysqlclient_r}. In most cases this is because you haven't
included the thread libraries on the link/compile line.
@node Cplusplus, Java, C, Clients
@section MySQL C++ APIs
......@@ -44221,34 +44326,6 @@ thread that is waiting on the disk-full condition will allow the other
threads to continue.
@end itemize
You need to know the following if you have a thread that is calling
MySQL functions, but that thread has not created the connection to the
MySQL database:
When you call @code{mysql_init()} or @code{mysql_connect()}, MySQL will
create a thread specific variable for the thread that is used by the
debug library (among other things).
If you have in a thread call a MySQL function, before a thread has
called @code{mysql_init()} or @code{mysql_connect()}, the thread will
not have the necessary thread specific variables in place and you are
likely to end up with a core dump sooner or later.
The get things to work smoothly you have to do the following:
@enumerate
@item
Call @code{my_init()} at the start of your program if it calls
any other MySQL function before calling @code{mysql_real_connect()}.
@item
Call @code{my_thread_init()} in the thread handler before calling
any MySQL function.
@item
In the thread, call @code{my_thread_end()} before calling
@code{pthread_exit()}. This will free the memory used by MySQL thread
specific variables.
@end enumerate
Exceptions to the above behaveour is when you use @code{REPAIR} or
@code{OPTIMIZE} or when the indexes are created in a batch after an
@code{LOAD DATA INFILE} or after an @code{ALTER TABLE} statement.
......@@ -46707,6 +46784,8 @@ not yet 100% confident in this code.
@appendixsubsec Changes in release 3.23.42
@itemize @bullet
@item
Fixed problem when using @code{LOCK TABLES} and @code{BDB} tables.
@item
Fixed problem with @code{REPAIR TABLE} on MyISAM tables with row lengths
between 65517 - 65520 bytes
@item
sql-bench/test-create.sh
View file @ 228de3b0
......@@ -39,13 +39,21 @@ $opt_loop_count=10000; # Change this to make test harder/easier
chomp($pwd = `pwd`); $pwd = "." if ($pwd eq '');
require "$pwd/bench-init.pl" || die "Can't read Configuration file: $!\n";
$create_loop_count=$opt_loop_count;
if ($opt_small_test)
{
$opt_loop_count/=100;
$create_loop_count/=1000;
}
$max_tables=min($limits->{'max_tables'},$opt_loop_count);
if ($opt_small_test)
{
$max_tables=10;
}
print "Testing the speed of creating and droping tables\n";
print "Testing with $max_tables tables and $opt_loop_count loop count\n\n";
......@@ -177,7 +185,7 @@ print "Testing create+drop\n";
$loop_time=new Benchmark;
for ($i=1 ; $i <= $opt_loop_count ; $i++)
for ($i=1 ; $i <= $create_loop_count ; $i++)
{
do_many($dbh,$server->create("bench_$i",
["i int NOT NULL",
......@@ -190,7 +198,7 @@ for ($i=1 ; $i <= $opt_loop_count ; $i++)
}
$end_time=new Benchmark;
print "Time for create+drop ($opt_loop_count): " .
print "Time for create+drop ($create_loop_count): " .
timestr(timediff($end_time, $loop_time),"all") . "\n";
if ($opt_fast && defined($server->{vacuum}))
......
sql/ha_berkeley.cc
View file @ 228de3b0
......@@ -1658,12 +1658,15 @@ int ha_berkeley::external_lock(THD *thd, int lock_type)
{
if (!thd->transaction.bdb_lock_count++)
{
changed_rows=0;
/* First table lock, start transaction */
if ((thd->options & (OPTION_NOT_AUTO_COMMIT | OPTION_BEGIN)) &&
if ((thd->options & (OPTION_NOT_AUTO_COMMIT | OPTION_BEGIN |
OPTION_TABLE_LOCK)) &&
!thd->transaction.all.bdb_tid)
{
DBUG_ASSERT(thd->transaction.stmt.bdb_tid != 0);
/* We have to start a master transaction */
DBUG_PRINT("trans",("starting transaction"));
DBUG_PRINT("trans",("starting transaction all"));
if ((error=txn_begin(db_env, 0,
(DB_TXN**) &thd->transaction.all.bdb_tid,
0)))
......@@ -1671,8 +1674,10 @@ int ha_berkeley::external_lock(THD *thd, int lock_type)
thd->transaction.bdb_lock_count--; // We didn't get the lock /* purecov: inspected */
DBUG_RETURN(error); /* purecov: inspected */
}
if (thd->in_lock_tables)
DBUG_RETURN(0); // Don't create stmt trans
}
DBUG_PRINT("trans",("starting transaction for statement"));
DBUG_PRINT("trans",("starting transaction stmt"));
if ((error=txn_begin(db_env,
(DB_TXN*) thd->transaction.all.bdb_tid,
(DB_TXN**) &thd->transaction.stmt.bdb_tid,
......@@ -1684,7 +1689,6 @@ int ha_berkeley::external_lock(THD *thd, int lock_type)
}
}
transaction= (DB_TXN*) thd->transaction.stmt.bdb_tid;
changed_rows=0;
}
else
{
......@@ -1722,6 +1726,7 @@ int ha_berkeley::start_stmt(THD *thd)
DBUG_ENTER("ha_berkeley::start_stmt");
if (!thd->transaction.stmt.bdb_tid)
{
DBUG_PRINT("trans",("starting transaction stmt"));
error=txn_begin(db_env, (DB_TXN*) thd->transaction.all.bdb_tid,
(DB_TXN**) &thd->transaction.stmt.bdb_tid,
0);
......
sql/mysql_priv.h
View file @ 228de3b0
......@@ -163,7 +163,8 @@ void kill_one_thread(THD *thd, ulong id);
#define OPTION_BIN_LOG OPTION_BUFFER_RESULT*2
#define OPTION_NOT_AUTO_COMMIT OPTION_BIN_LOG*2
#define OPTION_BEGIN OPTION_NOT_AUTO_COMMIT*2
#define OPTION_QUICK OPTION_BEGIN*2
#define OPTION_TABLE_LOCK OPTION_BEGIN*2
#define OPTION_QUICK OPTION_TABLE_LOCK*2
#define OPTION_QUOTE_SHOW_CREATE OPTION_QUICK*2
#define OPTION_INTERNAL_SUBTRANSACTIONS OPTION_QUOTE_SHOW_CREATE*2
......
sql/sql_base.cc
View file @ 228de3b0
......@@ -413,7 +413,10 @@ void close_thread_tables(THD *thd, bool locked)
DBUG_ENTER("close_thread_tables");
if (thd->locked_tables)
{
ha_commit_stmt(thd); // If select statement
DBUG_VOID_RETURN; // LOCK TABLES in use
}
TABLE *table,*next;
bool found_old_table=0;
......
sql/sql_parse.cc
View file @ 228de3b0
......@@ -81,7 +81,8 @@ static void init_signals(void)
inline bool end_active_trans(THD *thd)
{
int error=0;
if (thd->options & (OPTION_NOT_AUTO_COMMIT | OPTION_BEGIN))
if (thd->options & (OPTION_NOT_AUTO_COMMIT | OPTION_BEGIN |
OPTION_TABLE_LOCK))
{
thd->options&= ~(ulong) (OPTION_BEGIN | OPTION_STATUS_NO_TRANS_UPDATE);
thd->server_status&= ~SERVER_STATUS_IN_TRANS;
......@@ -1825,7 +1826,11 @@ mysql_execute_command(void)
{
thd->lock=thd->locked_tables;
thd->locked_tables=0; // Will be automaticly closed
}
if (thd->options & OPTION_TABLE_LOCK)
{
end_active_trans(thd);
thd->options&= ~(ulong) (OPTION_TABLE_LOCK);
}
if (thd->global_read_lock)
{
......@@ -1847,12 +1852,15 @@ mysql_execute_command(void)
if (check_db_used(thd,tables) || end_active_trans(thd))
goto error;
thd->in_lock_tables=1;
thd->options|= OPTION_TABLE_LOCK;
if (!(res=open_and_lock_tables(thd,tables)))
{
thd->locked_tables=thd->lock;
thd->lock=0;
send_ok(&thd->net);
}
else
thd->options&= ~(ulong) (OPTION_TABLE_LOCK);
thd->in_lock_tables=0;
break;
case SQLCOM_CREATE_DB:
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7