~drizzle-trunk/drizzle/development

/*****************************************************************************

Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.

Copyright (c) 2008, 2009 Google Inc.

Portions of this file contain modifications contributed and copyrighted by

Google, Inc. Those modifications are gratefully acknowledged and are described

briefly in the InnoDB documentation. The contributions by Google are

incorporated with their permission, and subject to the conditions contained in

the file COPYING.Google.

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

*****************************************************************************/

/***********************************************************************

Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.

Copyright (c) 2009, Percona Inc.

Portions of this file contain modifications contributed and copyrighted

by Percona Inc.. Those modifications are

gratefully acknowledged and are described briefly in the InnoDB

documentation. The contributions by Percona Inc. are incorporated with

their permission, and subject to the conditions contained in the file

COPYING.Percona.

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

***********************************************************************/

/**************************************************//**

@file srv/srv0srv.c

The database server main program

NOTE: SQL Server 7 uses something which the documentation

calls user mode scheduled threads (UMS threads). One such

thread is usually allocated per processor. Win32

documentation does not know any UMS threads, which suggests

that the concept is internal to SQL Server 7. It may mean that

SQL Server 7 does all the scheduling of threads itself, even

in i/o waits. We should maybe modify InnoDB to use the same

technique, because thread switches within NT may be too slow.

SQL Server 7 also mentions fibers, which are cooperatively

scheduled threads. They can boost performance by 5 %,

according to the Delaney and Soukup's book.

Windows 2000 will have something called thread pooling

(see msdn website), which we could possibly use.

Another possibility could be to use some very fast user space

thread library. This might confuse NT though.

Created 10/8/1995 Heikki Tuuri

*******************************************************/

/* Dummy comment */

#include "srv0srv.h"

#include "ut0mem.h"

#include "ut0ut.h"

#include "os0proc.h"

#include "mem0mem.h"

#include "mem0pool.h"

#include "sync0sync.h"

#include "thr0loc.h"

#include "que0que.h"

#include "srv0que.h"

#include "log0recv.h"

#include "pars0pars.h"

#include "usr0sess.h"

#include "lock0lock.h"

#include "trx0purge.h"

#include "ibuf0ibuf.h"

#include "buf0flu.h"

#include "buf0lru.h"

#include "btr0sea.h"

#include "dict0load.h"

#include "dict0boot.h"

#include "srv0start.h"

#include "row0mysql.h"

#include "ha_prototypes.h"

#include "trx0i_s.h"

/* This is set to TRUE if the MySQL user has set it in MySQL; currently

affects only FOREIGN KEY definition parsing */

UNIV_INTERN ibool	srv_lower_case_table_names	= FALSE;

/* The following counter is incremented whenever there is some user activity

in the server */

UNIV_INTERN ulint	srv_activity_count	= 0;

/* The following is the maximum allowed duration of a lock wait. */

UNIV_INTERN ulint	srv_fatal_semaphore_wait_threshold = 600;

/* How much data manipulation language (DML) statements need to be delayed,

in microseconds, in order to reduce the lagging of the purge thread. */

UNIV_INTERN ulint	srv_dml_needed_delay = 0;

UNIV_INTERN ibool	srv_lock_timeout_and_monitor_active = FALSE;

UNIV_INTERN ibool	srv_error_monitor_active = FALSE;

UNIV_INTERN const char*	srv_main_thread_op_info = "";

/** Prefix used by MySQL to indicate pre-5.1 table name encoding */

UNIV_INTERN const char	srv_mysql50_table_name_prefix[9] = "#mysql50#";

/* Server parameters which are read from the initfile */

/* The following three are dir paths which are catenated before file

names, where the file name itself may also contain a path */

UNIV_INTERN char*	srv_data_home	= NULL;

#ifdef UNIV_LOG_ARCHIVE

UNIV_INTERN char*	srv_arch_dir	= NULL;

#endif /* UNIV_LOG_ARCHIVE */

/** store to its own file each table created by an user; data

dictionary tables are in the system tablespace 0 */

UNIV_INTERN my_bool	srv_file_per_table;

/** The file format to use on new *.ibd files. */

UNIV_INTERN ulint	srv_file_format = 0;

/** Whether to check file format during startup.  A value of

DICT_TF_FORMAT_MAX + 1 means no checking ie. FALSE.  The default is to

set it to the highest format we support. */

UNIV_INTERN ulint	srv_check_file_format_at_startup = DICT_TF_FORMAT_MAX;

#if DICT_TF_FORMAT_51

# error "DICT_TF_FORMAT_51 must be 0!"

#endif

/** Place locks to records only i.e. do not use next-key locking except

on duplicate key checking and foreign key checking */

UNIV_INTERN ibool	srv_locks_unsafe_for_binlog = FALSE;

UNIV_INTERN ulint	srv_n_data_files = 0;

UNIV_INTERN char**	srv_data_file_names = NULL;

/* size in database pages */

UNIV_INTERN ulint*	srv_data_file_sizes = NULL;

/* if TRUE, then we auto-extend the last data file */

UNIV_INTERN ibool	srv_auto_extend_last_data_file	= FALSE;

/* if != 0, this tells the max size auto-extending may increase the

last data file size */

UNIV_INTERN ulint	srv_last_file_size_max	= 0;

/* If the last data file is auto-extended, we add this

many pages to it at a time */

UNIV_INTERN unsigned int srv_auto_extend_increment = 8;

UNIV_INTERN ulint*	srv_data_file_is_raw_partition = NULL;

/* If the following is TRUE we do not allow inserts etc. This protects

the user from forgetting the 'newraw' keyword to my.cnf */

UNIV_INTERN ibool	srv_created_new_raw	= FALSE;

UNIV_INTERN char**	srv_log_group_home_dirs = NULL;

UNIV_INTERN ulint	srv_n_log_groups	= ULINT_MAX;

UNIV_INTERN ulint	srv_n_log_files		= ULINT_MAX;

/* size in database pages */

UNIV_INTERN ulint	srv_log_file_size	= ULINT_MAX;

/* size in database pages */

UNIV_INTERN ulint	srv_log_buffer_size	= ULINT_MAX;

UNIV_INTERN ulong	srv_flush_log_at_trx_commit = 1;

/* Try to flush dirty pages so as to avoid IO bursts at

the checkpoints. */

UNIV_INTERN bool	srv_adaptive_flushing	= TRUE;

/* The sort order table of the MySQL latin1_swedish_ci character set

collation */

#if defined(BUILD_DRIZZLE)

UNIV_INTERN const byte	srv_latin1_ordering[256]	/* The sort order table of the latin1

					character set. The following table is

					the MySQL order as of Feb 10th, 2002 */

= {

  0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07

, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F

, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17

, 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F

, 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27

, 0x28, 0x29, 0x2A, 0x2B, 0x2C, 0x2D, 0x2E, 0x2F

, 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37

, 0x38, 0x39, 0x3A, 0x3B, 0x3C, 0x3D, 0x3E, 0x3F

, 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47

, 0x48, 0x49, 0x4A, 0x4B, 0x4C, 0x4D, 0x4E, 0x4F

, 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57

, 0x58, 0x59, 0x5A, 0x5B, 0x5C, 0x5D, 0x5E, 0x5F

, 0x60, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47

, 0x48, 0x49, 0x4A, 0x4B, 0x4C, 0x4D, 0x4E, 0x4F

, 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57

, 0x58, 0x59, 0x5A, 0x7B, 0x7C, 0x7D, 0x7E, 0x7F

, 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87

, 0x88, 0x89, 0x8A, 0x8B, 0x8C, 0x8D, 0x8E, 0x8F

, 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97

, 0x98, 0x99, 0x9A, 0x9B, 0x9C, 0x9D, 0x9E, 0x9F

, 0xA0, 0xA1, 0xA2, 0xA3, 0xA4, 0xA5, 0xA6, 0xA7

, 0xA8, 0xA9, 0xAA, 0xAB, 0xAC, 0xAD, 0xAE, 0xAF

, 0xB0, 0xB1, 0xB2, 0xB3, 0xB4, 0xB5, 0xB6, 0xB7

, 0xB8, 0xB9, 0xBA, 0xBB, 0xBC, 0xBD, 0xBE, 0xBF

, 0x41, 0x41, 0x41, 0x41, 0x5C, 0x5B, 0x5C, 0x43

, 0x45, 0x45, 0x45, 0x45, 0x49, 0x49, 0x49, 0x49

, 0x44, 0x4E, 0x4F, 0x4F, 0x4F, 0x4F, 0x5D, 0xD7

, 0xD8, 0x55, 0x55, 0x55, 0x59, 0x59, 0xDE, 0xDF

, 0x41, 0x41, 0x41, 0x41, 0x5C, 0x5B, 0x5C, 0x43

, 0x45, 0x45, 0x45, 0x45, 0x49, 0x49, 0x49, 0x49

, 0x44, 0x4E, 0x4F, 0x4F, 0x4F, 0x4F, 0x5D, 0xF7

, 0xD8, 0x55, 0x55, 0x55, 0x59, 0x59, 0xDE, 0xFF

};

#else

UNIV_INTERN const byte*        srv_latin1_ordering;

#endif /* BUILD_DRIZZLE */

/* use os/external memory allocator */

UNIV_INTERN my_bool	srv_use_sys_malloc	= TRUE;

/* requested size in kilobytes */

UNIV_INTERN ulint	srv_buf_pool_size	= ULINT_MAX;

/* previously requested size */

UNIV_INTERN ulint	srv_buf_pool_old_size;

/* current size in kilobytes */

UNIV_INTERN ulint	srv_buf_pool_curr_size	= 0;

/* size in bytes */

UNIV_INTERN ulint	srv_mem_pool_size	= ULINT_MAX;

UNIV_INTERN ulint	srv_lock_table_size	= ULINT_MAX;

/* This parameter is deprecated. Use srv_n_io_[read|write]_threads

instead. */

UNIV_INTERN ulint	srv_n_file_io_threads	= ULINT_MAX;

UNIV_INTERN ulint	srv_n_read_io_threads	= ULINT_MAX;

UNIV_INTERN ulint	srv_n_write_io_threads	= ULINT_MAX;

/* User settable value of the number of pages that must be present

in the buffer cache and accessed sequentially for InnoDB to trigger a

readahead request. */

UNIV_INTERN ulong	srv_read_ahead_threshold	= 56;

#ifdef UNIV_LOG_ARCHIVE

UNIV_INTERN ibool		srv_log_archive_on	= FALSE;

UNIV_INTERN ibool		srv_archive_recovery	= 0;

UNIV_INTERN ib_uint64_t	srv_archive_recovery_limit_lsn;

#endif /* UNIV_LOG_ARCHIVE */

/* This parameter is used to throttle the number of insert buffers that are

merged in a batch. By increasing this parameter on a faster disk you can

possibly reduce the number of I/O operations performed to complete the

merge operation. The value of this parameter is used as is by the

background loop when the system is idle (low load), on a busy system

the parameter is scaled down by a factor of 4, this is to avoid putting

a heavier load on the I/O sub system. */

UNIV_INTERN ulong	srv_insert_buffer_batch_size = 20;

UNIV_INTERN char*	srv_file_flush_method_str = NULL;

UNIV_INTERN ulint	srv_unix_file_flush_method = SRV_UNIX_FSYNC;

UNIV_INTERN ulint	srv_win_file_flush_method = SRV_WIN_IO_UNBUFFERED;

UNIV_INTERN ulint	srv_max_n_open_files	  = 300;

/* Number of IO operations per second the server can do */

UNIV_INTERN ulong	srv_io_capacity         = 200;

/* The InnoDB main thread tries to keep the ratio of modified pages

in the buffer pool to all database pages in the buffer pool smaller than

the following number. But it is not guaranteed that the value stays below

that during a time of heavy update/insert activity. */

UNIV_INTERN ulong	srv_max_buf_pool_modified_pct	= 75;

/* variable counts amount of data read in total (in bytes) */

UNIV_INTERN ulint srv_data_read = 0;

/* here we count the amount of data written in total (in bytes) */

UNIV_INTERN ulint srv_data_written = 0;

/* the number of the log write requests done */

UNIV_INTERN ulint srv_log_write_requests = 0;

/* the number of physical writes to the log performed */

UNIV_INTERN ulint srv_log_writes = 0;

/* amount of data written to the log files in bytes */

UNIV_INTERN ulint srv_os_log_written = 0;

/* amount of writes being done to the log files */

UNIV_INTERN ulint srv_os_log_pending_writes = 0;

/* we increase this counter, when there we don't have enough space in the

log buffer and have to flush it */

UNIV_INTERN ulint srv_log_waits = 0;

/* this variable counts the amount of times, when the doublewrite buffer

was flushed */

UNIV_INTERN ulint srv_dblwr_writes = 0;

/* here we store the number of pages that have been flushed to the

doublewrite buffer */

UNIV_INTERN ulint srv_dblwr_pages_written = 0;

/* in this variable we store the number of write requests issued */

UNIV_INTERN ulint srv_buf_pool_write_requests = 0;

/* here we store the number of times when we had to wait for a free page

in the buffer pool. It happens when the buffer pool is full and we need

to make a flush, in order to be able to read or create a page. */

UNIV_INTERN ulint srv_buf_pool_wait_free = 0;

/* variable to count the number of pages that were written from buffer

pool to the disk */

UNIV_INTERN ulint srv_buf_pool_flushed = 0;

/** Number of buffer pool reads that led to the

reading of a disk page */

UNIV_INTERN ulint srv_buf_pool_reads = 0;

/** Number of sequential read-aheads */

UNIV_INTERN ulint srv_read_ahead_seq = 0;

/** Number of random read-aheads */

UNIV_INTERN ulint srv_read_ahead_rnd = 0;

/* structure to pass status variables to MySQL */

UNIV_INTERN export_struc export_vars;

/* If the following is != 0 we do not allow inserts etc. This protects

the user from forgetting the innodb_force_recovery keyword to my.cnf */

UNIV_INTERN ulint	srv_force_recovery	= 0;

/*-----------------------*/

/* We are prepared for a situation that we have this many threads waiting for

a semaphore inside InnoDB. innobase_start_or_create_for_mysql() sets the

value. */

UNIV_INTERN ulint	srv_max_n_threads	= 0;

/* The following controls how many threads we let inside InnoDB concurrently:

threads waiting for locks are not counted into the number because otherwise

we could get a deadlock. MySQL creates a thread for each user session, and

semaphore contention and convoy problems can occur withput this restriction.

Value 10 should be good if there are less than 4 processors + 4 disks in the

computer. Bigger computers need bigger values. Value 0 will disable the

concurrency check. */

UNIV_INTERN ulong	srv_thread_concurrency	= 0;

/* this mutex protects srv_conc data structures */

UNIV_INTERN os_fast_mutex_t	srv_conc_mutex;

/* number of transactions that have declared_to_be_inside_innodb set.

It used to be a non-error for this value to drop below zero temporarily.

This is no longer true. We'll, however, keep the lint datatype to add

assertions to catch any corner cases that we may have missed. */

UNIV_INTERN lint	srv_conc_n_threads	= 0;

/* number of OS threads waiting in the FIFO for a permission to enter

InnoDB */

UNIV_INTERN ulint	srv_conc_n_waiting_threads = 0;

typedef struct srv_conc_slot_struct	srv_conc_slot_t;

struct srv_conc_slot_struct{

	os_event_t			event;		/*!< event to wait */

	ibool				reserved;	/*!< TRUE if slot

							reserved */

	ibool				wait_ended;	/*!< TRUE when another

							thread has already set

							the event and the

							thread in this slot is

							free to proceed; but

							reserved may still be

							TRUE at that point */

	UT_LIST_NODE_T(srv_conc_slot_t)	srv_conc_queue;	/*!< queue node */

};

/* queue of threads waiting to get in */

UNIV_INTERN UT_LIST_BASE_NODE_T(srv_conc_slot_t)	srv_conc_queue;

/* array of wait slots */

UNIV_INTERN srv_conc_slot_t* srv_conc_slots;

/* Number of times a thread is allowed to enter InnoDB within the same

SQL query after it has once got the ticket at srv_conc_enter_innodb */

#define SRV_FREE_TICKETS_TO_ENTER srv_n_free_tickets_to_enter

#define SRV_THREAD_SLEEP_DELAY srv_thread_sleep_delay

/*-----------------------*/

/* If the following is set to 1 then we do not run purge and insert buffer

merge to completion before shutdown. If it is set to 2, do not even flush the

buffer pool to data files at the shutdown: we effectively 'crash'

InnoDB (but lose no committed transactions). */

UNIV_INTERN ulint	srv_fast_shutdown	= 0;

/* Generate a innodb_status.<pid> file */

UNIV_INTERN ibool	srv_innodb_status	= FALSE;

/* When estimating number of different key values in an index, sample

this many index pages */

UNIV_INTERN ib_uint64_t	srv_stats_sample_pages = 8;

UNIV_INTERN ibool	srv_use_doublewrite_buf	= TRUE;

UNIV_INTERN ibool	srv_use_checksums = TRUE;

UNIV_INTERN ulong	srv_replication_delay		= 0;

/*-------------------------------------------*/

UNIV_INTERN ulong	srv_n_spin_wait_rounds	= 30;

UNIV_INTERN ulong	srv_n_free_tickets_to_enter = 500;

UNIV_INTERN ulong	srv_thread_sleep_delay = 10000;

UNIV_INTERN ulong	srv_spin_wait_delay	= 6;

UNIV_INTERN ibool	srv_priority_boost	= TRUE;

#ifdef UNIV_DEBUG

UNIV_INTERN ibool	srv_print_thread_releases	= FALSE;

UNIV_INTERN ibool	srv_print_lock_waits		= FALSE;

UNIV_INTERN ibool	srv_print_buf_io		= FALSE;

UNIV_INTERN ibool	srv_print_log_io		= FALSE;

UNIV_INTERN ibool	srv_print_latch_waits		= FALSE;

#endif /* UNIV_DEBUG */

UNIV_INTERN ulint		srv_n_rows_inserted		= 0;

UNIV_INTERN ulint		srv_n_rows_updated		= 0;

UNIV_INTERN ulint		srv_n_rows_deleted		= 0;

UNIV_INTERN ulint		srv_n_rows_read			= 0;

static ulint	srv_n_rows_inserted_old		= 0;

static ulint	srv_n_rows_updated_old		= 0;

static ulint	srv_n_rows_deleted_old		= 0;

static ulint	srv_n_rows_read_old		= 0;

UNIV_INTERN ulint		srv_n_lock_wait_count		= 0;

UNIV_INTERN ulint		srv_n_lock_wait_current_count	= 0;

UNIV_INTERN ib_int64_t	srv_n_lock_wait_time		= 0;

UNIV_INTERN ulint		srv_n_lock_max_wait_time	= 0;

/*

  Set the following to 0 if you want InnoDB to write messages on

  stderr on startup/shutdown

*/

UNIV_INTERN ibool	srv_print_verbose_log		= TRUE;

UNIV_INTERN ibool	srv_print_innodb_monitor	= FALSE;

UNIV_INTERN ibool	srv_print_innodb_lock_monitor	= FALSE;

UNIV_INTERN ibool	srv_print_innodb_tablespace_monitor = FALSE;

UNIV_INTERN ibool	srv_print_innodb_table_monitor = FALSE;

/* Array of English strings describing the current state of an

i/o handler thread */

UNIV_INTERN const char* srv_io_thread_op_info[SRV_MAX_N_IO_THREADS];

UNIV_INTERN const char* srv_io_thread_function[SRV_MAX_N_IO_THREADS];

UNIV_INTERN time_t	srv_last_monitor_time;

UNIV_INTERN mutex_t	srv_innodb_monitor_mutex;

/* Mutex for locking srv_monitor_file */

UNIV_INTERN mutex_t	srv_monitor_file_mutex;

/* Temporary file for innodb monitor output */

UNIV_INTERN FILE*	srv_monitor_file;

/* Mutex for locking srv_dict_tmpfile.

This mutex has a very high rank; threads reserving it should not

be holding any InnoDB latches. */

UNIV_INTERN mutex_t	srv_dict_tmpfile_mutex;

/* Temporary file for output from the data dictionary */

UNIV_INTERN FILE*	srv_dict_tmpfile;

/* Mutex for locking srv_misc_tmpfile.

This mutex has a very low rank; threads reserving it should not

acquire any further latches or sleep before releasing this one. */

UNIV_INTERN mutex_t	srv_misc_tmpfile_mutex;

/* Temporary file for miscellanous diagnostic output */

UNIV_INTERN FILE*	srv_misc_tmpfile;

UNIV_INTERN ulint	srv_main_thread_process_no	= 0;

UNIV_INTERN ulint	srv_main_thread_id		= 0;

/* The following count work done by srv_master_thread. */

/* Iterations by the 'once per second' loop. */

static ulint   srv_main_1_second_loops		= 0;

/* Calls to sleep by the 'once per second' loop. */

static ulint   srv_main_sleeps			= 0;

/* Iterations by the 'once per 10 seconds' loop. */

static ulint   srv_main_10_second_loops		= 0;

/* Iterations of the loop bounded by the 'background_loop' label. */

static ulint   srv_main_background_loops	= 0;

/* Iterations of the loop bounded by the 'flush_loop' label. */

static ulint   srv_main_flush_loops		= 0;

/* Log writes involving flush. */

static ulint   srv_log_writes_and_flush		= 0;

/* Log writes not including flush. */

static ulint   srv_log_buffer_writes		= 0;

/* This is only ever touched by the master thread. It records the

time when the last flush of log file has happened. The master

thread ensures that we flush the log files at least once per

second. */

static time_t	srv_last_log_flush_time;

/* The master thread performs various tasks based on the current

state of IO activity and the level of IO utilization is past

intervals. Following macros define thresholds for these conditions. */

#define SRV_PEND_IO_THRESHOLD	(PCT_IO(3))

#define SRV_RECENT_IO_ACTIVITY	(PCT_IO(5))

#define SRV_PAST_IO_ACTIVITY	(PCT_IO(200))

/*

	IMPLEMENTATION OF THE SERVER MAIN PROGRAM

	=========================================

There is the following analogue between this database

server and an operating system kernel:

DB concept			equivalent OS concept

----------			---------------------

transaction		--	process;

query thread		--	thread;

lock			--	semaphore;

transaction set to

the rollback state	--	kill signal delivered to a process;

kernel			--	kernel;

query thread execution:

(a) without kernel mutex

reserved		--	process executing in user mode;

(b) with kernel mutex reserved

			--	process executing in kernel mode;

The server is controlled by a master thread which runs at

a priority higher than normal, that is, higher than user threads.

It sleeps most of the time, and wakes up, say, every 300 milliseconds,

to check whether there is anything happening in the server which

requires intervention of the master thread. Such situations may be,

for example, when flushing of dirty blocks is needed in the buffer

pool or old version of database rows have to be cleaned away.

The threads which we call user threads serve the queries of

the clients and input from the console of the server.

They run at normal priority. The server may have several

communications endpoints. A dedicated set of user threads waits

at each of these endpoints ready to receive a client request.

Each request is taken by a single user thread, which then starts

processing and, when the result is ready, sends it to the client

and returns to wait at the same endpoint the thread started from.

So, we do not have dedicated communication threads listening at

the endpoints and dealing the jobs to dedicated worker threads.

Our architecture saves one thread swithch per request, compared

to the solution with dedicated communication threads

which amounts to 15 microseconds on 100 MHz Pentium

running NT. If the client

is communicating over a network, this saving is negligible, but

if the client resides in the same machine, maybe in an SMP machine

on a different processor from the server thread, the saving

can be important as the threads can communicate over shared

memory with an overhead of a few microseconds.

We may later implement a dedicated communication thread solution

for those endpoints which communicate over a network.

Our solution with user threads has two problems: for each endpoint

there has to be a number of listening threads. If there are many

communication endpoints, it may be difficult to set the right number

of concurrent threads in the system, as many of the threads

may always be waiting at less busy endpoints. Another problem

is queuing of the messages, as the server internally does not

offer any queue for jobs.

Another group of user threads is intended for splitting the

queries and processing them in parallel. Let us call these

parallel communication threads. These threads are waiting for

parallelized tasks, suspended on event semaphores.

A single user thread waits for input from the console,

like a command to shut the database.

Utility threads are a different group of threads which takes

care of the buffer pool flushing and other, mainly background

operations, in the server.

Some of these utility threads always run at a lower than normal

priority, so that they are always in background. Some of them

may dynamically boost their priority by the pri_adjust function,

even to higher than normal priority, if their task becomes urgent.

The running of utilities is controlled by high- and low-water marks

of urgency. The urgency may be measured by the number of dirty blocks

in the buffer pool, in the case of the flush thread, for example.

When the high-water mark is exceeded, an utility starts running, until

the urgency drops under the low-water mark. Then the utility thread

suspend itself to wait for an event. The master thread is

responsible of signaling this event when the utility thread is

again needed.

For each individual type of utility, some threads always remain

at lower than normal priority. This is because pri_adjust is implemented

so that the threads at normal or higher priority control their

share of running time by calling sleep. Thus, if the load of the

system sudenly drops, these threads cannot necessarily utilize

the system fully. The background priority threads make up for this,

starting to run when the load drops.

When there is no activity in the system, also the master thread

suspends itself to wait for an event making

the server totally silent. The responsibility to signal this

event is on the user thread which again receives a message

from a client.

There is still one complication in our server design. If a

background utility thread obtains a resource (e.g., mutex) needed by a user

thread, and there is also some other user activity in the system,

the user thread may have to wait indefinitely long for the

resource, as the OS does not schedule a background thread if

there is some other runnable user thread. This problem is called

priority inversion in real-time programming.

One solution to the priority inversion problem would be to

keep record of which thread owns which resource and

in the above case boost the priority of the background thread

so that it will be scheduled and it can release the resource.

This solution is called priority inheritance in real-time programming.

A drawback of this solution is that the overhead of acquiring a mutex

increases slightly, maybe 0.2 microseconds on a 100 MHz Pentium, because

the thread has to call os_thread_get_curr_id.

This may be compared to 0.5 microsecond overhead for a mutex lock-unlock

pair. Note that the thread

cannot store the information in the resource, say mutex, itself,

because competing threads could wipe out the information if it is

stored before acquiring the mutex, and if it stored afterwards,

the information is outdated for the time of one machine instruction,

at least. (To be precise, the information could be stored to

lock_word in mutex if the machine supports atomic swap.)

The above solution with priority inheritance may become actual in the

future, but at the moment we plan to implement a more coarse solution,

which could be called a global priority inheritance. If a thread

has to wait for a long time, say 300 milliseconds, for a resource,

we just guess that it may be waiting for a resource owned by a background

thread, and boost the the priority of all runnable background threads

to the normal level. The background threads then themselves adjust

their fixed priority back to background after releasing all resources

they had (or, at some fixed points in their program code).

What is the performance of the global priority inheritance solution?

We may weigh the length of the wait time 300 milliseconds, during

which the system processes some other thread

to the cost of boosting the priority of each runnable background

thread, rescheduling it, and lowering the priority again.

On 100 MHz Pentium + NT this overhead may be of the order 100

microseconds per thread. So, if the number of runnable background

threads is not very big, say < 100, the cost is tolerable.

Utility threads probably will access resources used by

user threads not very often, so collisions of user threads

to preempted utility threads should not happen very often.

The thread table contains

information of the current status of each thread existing in the system,

and also the event semaphores used in suspending the master thread

and utility and parallel communication threads when they have nothing to do.

The thread table can be seen as an analogue to the process table

in a traditional Unix implementation.

The thread table is also used in the global priority inheritance

scheme. This brings in one additional complication: threads accessing

the thread table must have at least normal fixed priority,

because the priority inheritance solution does not work if a background

thread is preempted while possessing the mutex protecting the thread table.

So, if a thread accesses the thread table, its priority has to be

boosted at least to normal. This priority requirement can be seen similar to

the privileged mode used when processing the kernel calls in traditional

Unix.*/

/* Thread slot in the thread table */

struct srv_slot_struct{

	os_thread_id_t	id;		/*!< thread id */

	os_thread_t	handle;		/*!< thread handle */

	unsigned	type:3;		/*!< thread type: user, utility etc. */

	unsigned	in_use:1;	/*!< TRUE if this slot is in use */

	unsigned	suspended:1;	/*!< TRUE if the thread is waiting

					for the event of this slot */

	ib_time_t	suspend_time;	/*!< time when the thread was

					suspended */

	os_event_t	event;		/*!< event used in suspending the

					thread when it has nothing to do */

	que_thr_t*	thr;		/*!< suspended query thread (only

					used for MySQL threads) */

};

/* Table for MySQL threads where they will be suspended to wait for locks */

UNIV_INTERN srv_slot_t*	srv_mysql_table = NULL;

UNIV_INTERN os_event_t	srv_lock_timeout_thread_event;

UNIV_INTERN srv_sys_t*	srv_sys	= NULL;

/* padding to prevent other memory update hotspots from residing on

the same memory cache line */

UNIV_INTERN byte	srv_pad1[64];

/* mutex protecting the server, trx structs, query threads, and lock table */

UNIV_INTERN mutex_t*	kernel_mutex_temp;

/* padding to prevent other memory update hotspots from residing on

the same memory cache line */

UNIV_INTERN byte	srv_pad2[64];

#if 0

/* The following three values measure the urgency of the jobs of

buffer, version, and insert threads. They may vary from 0 - 1000.

The server mutex protects all these variables. The low-water values

tell that the server can acquiesce the utility when the value

drops below this low-water mark. */

static ulint	srv_meter[SRV_MASTER + 1];

static ulint	srv_meter_low_water[SRV_MASTER + 1];

static ulint	srv_meter_high_water[SRV_MASTER + 1];

static ulint	srv_meter_high_water2[SRV_MASTER + 1];

static ulint	srv_meter_foreground[SRV_MASTER + 1];

#endif

/* The following values give info about the activity going on in

the database. They are protected by the server mutex. The arrays

are indexed by the type of the thread. */

UNIV_INTERN ulint	srv_n_threads_active[SRV_MASTER + 1];

UNIV_INTERN ulint	srv_n_threads[SRV_MASTER + 1];

/***********************************************************************

Prints counters for work done by srv_master_thread. */

static

void

srv_print_master_thread_info(

/*=========================*/

	FILE  *file)    /* in: output stream */

{

	fprintf(file, "srv_master_thread loops: %lu 1_second, %lu sleeps, "

		"%lu 10_second, %lu background, %lu flush\n",

		srv_main_1_second_loops, srv_main_sleeps,

		srv_main_10_second_loops, srv_main_background_loops,

		srv_main_flush_loops);

	fprintf(file, "srv_master_thread log flush and writes: %lu "

		      " log writes only: %lu\n",

		      srv_log_writes_and_flush, srv_log_buffer_writes);

}

/*********************************************************************//**

Sets the info describing an i/o thread current state. */

UNIV_INTERN

void

srv_set_io_thread_op_info(

/*======================*/

	ulint		i,	/*!< in: the 'segment' of the i/o thread */

	const char*	str)	/*!< in: constant char string describing the

				state */

{

	ut_a(i < SRV_MAX_N_IO_THREADS);

	srv_io_thread_op_info[i] = str;

}

/*********************************************************************//**

Accessor function to get pointer to n'th slot in the server thread

table.

@return	pointer to the slot */

static

srv_slot_t*

srv_table_get_nth_slot(

/*===================*/

	ulint	index)		/*!< in: index of the slot */

{

	ut_a(index < OS_THREAD_MAX_N);

	return(srv_sys->threads + index);

}

/*********************************************************************//**

Gets the number of threads in the system.

@return	sum of srv_n_threads[] */

UNIV_INTERN

ulint

srv_get_n_threads(void)

/*===================*/

{

	ulint	i;

	ulint	n_threads	= 0;

	mutex_enter(&kernel_mutex);

	for (i = SRV_COM; i < SRV_MASTER + 1; i++) {

		n_threads += srv_n_threads[i];

	}

	mutex_exit(&kernel_mutex);

	return(n_threads);

}

/*********************************************************************//**

Reserves a slot in the thread table for the current thread. Also creates the

thread local storage struct for the current thread. NOTE! The server mutex

has to be reserved by the caller!

@return	reserved slot index */

static

ulint

srv_table_reserve_slot(

/*===================*/

	enum srv_thread_type	type)	/*!< in: type of the thread */

{

	srv_slot_t*	slot;

	ulint		i;

	ut_a(type > 0);

	ut_a(type <= SRV_MASTER);

	i = 0;

	slot = srv_table_get_nth_slot(i);

	while (slot->in_use) {

		i++;

		slot = srv_table_get_nth_slot(i);

	}

	ut_a(slot->in_use == FALSE);

	slot->in_use = TRUE;

	slot->suspended = FALSE;

	slot->type = type;

	slot->id = os_thread_get_curr_id();

	slot->handle = os_thread_get_curr();

	thr_local_create();

	thr_local_set_slot_no(os_thread_get_curr_id(), i);

	return(i);

}

/*********************************************************************//**

Suspends the calling thread to wait for the event in its thread slot.

NOTE! The server mutex has to be reserved by the caller!

@return	event for the calling thread to wait */

static

os_event_t

srv_suspend_thread(void)

/*====================*/

{

	srv_slot_t*		slot;

	os_event_t		event;

	ulint			slot_no;

	enum srv_thread_type	type;

	ut_ad(mutex_own(&kernel_mutex));

	slot_no = thr_local_get_slot_no(os_thread_get_curr_id());

	if (srv_print_thread_releases) {

		fprintf(stderr,

			"Suspending thread %lu to slot %lu\n",

			(ulong) os_thread_get_curr_id(), (ulong) slot_no);