srv0srv.c 73.9 KB
Newer Older
1 2 3 4 5 6 7 8 9
/******************************************************
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
10
in i/o waits. We should maybe modify InnoDB to use the same
11 12 13 14 15 16 17 18 19 20 21 22
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.

unknown's avatar
unknown committed
23
(c) 1995 Innobase Oy
24 25 26

Created 10/8/1995 Heikki Tuuri
*******************************************************/
unknown's avatar
unknown committed
27
/* Dummy comment */
28 29 30 31 32
#include "srv0srv.h"

#include "ut0mem.h"
#include "os0proc.h"
#include "mem0mem.h"
33
#include "mem0pool.h"
34 35 36
#include "sync0sync.h"
#include "thr0loc.h"
#include "que0que.h"
37
#include "srv0que.h"
38 39 40 41 42 43 44 45
#include "log0recv.h"
#include "pars0pars.h"
#include "usr0sess.h"
#include "lock0lock.h"
#include "trx0purge.h"
#include "ibuf0ibuf.h"
#include "buf0flu.h"
#include "btr0sea.h"
46
#include "dict0load.h"
unknown's avatar
unknown committed
47
#include "dict0boot.h"
unknown's avatar
Merge  
unknown committed
48
#include "srv0start.h"
unknown's avatar
unknown committed
49
#include "row0mysql.h"
50

unknown's avatar
unknown committed
51 52 53 54
/* This is set to TRUE if the MySQL user has set it in MySQL; currently
affects only FOREIGN KEY definition parsing */
ibool	srv_lower_case_table_names	= FALSE;

55 56 57 58
/* The following counter is incremented whenever there is some user activity
in the server */
ulint	srv_activity_count	= 0;

59 60 61
/* The following is the maximum allowed duration of a lock wait. */
ulint	srv_fatal_semaphore_wait_threshold = 600;

62 63 64 65
/* How much data manipulation language (DML) statements need to be delayed,
in microseconds, in order to reduce the lagging of the purge thread. */
ulint	srv_dml_needed_delay = 0;

unknown's avatar
Merge  
unknown committed
66 67 68
ibool	srv_lock_timeout_and_monitor_active = FALSE;
ibool	srv_error_monitor_active = FALSE;

unknown's avatar
unknown committed
69
const char*	srv_main_thread_op_info = "";
70

71 72 73 74 75 76
/* 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 */

char*	srv_data_home 	= NULL;
unknown's avatar
unknown committed
77
#ifdef UNIV_LOG_ARCHIVE
78
char*	srv_arch_dir 	= NULL;
unknown's avatar
unknown committed
79
#endif /* UNIV_LOG_ARCHIVE */
80

unknown's avatar
unknown committed
81 82 83 84
ibool	srv_file_per_table = FALSE;	/* store to its own file each table
					created by an user; data dictionary
					tables are in the system tablespace
					0 */
85 86 87 88
ibool   srv_locks_unsafe_for_binlog = FALSE; /* Place locks to records only 
                                                i.e. do not use next-key locking
                                                except on duplicate key checking and
                                                foreign key checking */
89 90 91 92
ulint	srv_n_data_files = 0;
char**	srv_data_file_names = NULL;
ulint*	srv_data_file_sizes = NULL;	/* size in database pages */ 

unknown's avatar
unknown committed
93 94 95 96 97 98 99
ibool	srv_auto_extend_last_data_file	= FALSE; /* if TRUE, then we
						 auto-extend the last data
						 file */
ulint	srv_last_file_size_max	= 0;		 /* if != 0, this tells
						 the max size auto-extending
						 may increase the last data
						 file size */
100
ulong	srv_auto_extend_increment = 8;		 /* If the last data file is
101 102
						 auto-extended, we add this
						 many pages to it at a time */
103 104
ulint*  srv_data_file_is_raw_partition = NULL;

105 106 107 108 109
/* If the following is TRUE we do not allow inserts etc. This protects
the user from forgetting the 'newraw' keyword to my.cnf */

ibool	srv_created_new_raw	= FALSE;

110 111 112 113 114 115
char**	srv_log_group_home_dirs = NULL; 

ulint	srv_n_log_groups	= ULINT_MAX;
ulint	srv_n_log_files		= ULINT_MAX;
ulint	srv_log_file_size	= ULINT_MAX;	/* size in database pages */ 
ulint	srv_log_buffer_size	= ULINT_MAX;	/* size in database pages */ 
unknown's avatar
unknown committed
116
ulint	srv_flush_log_at_trx_commit = 1;
117

unknown's avatar
unknown committed
118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154
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
};
155
		
unknown's avatar
unknown committed
156 157 158 159 160 161 162 163
ulint	srv_pool_size		= ULINT_MAX;	/* size in pages; MySQL inits
						this to size in kilobytes but
						we normalize this to pages in
						srv_boot() */
ulint	srv_awe_window_size 	= 0;		/* size in pages; MySQL inits
						this to bytes, but we
						normalize it to pages in
						srv_boot() */
164 165 166 167 168
ulint	srv_mem_pool_size	= ULINT_MAX;	/* size in bytes */ 
ulint	srv_lock_table_size	= ULINT_MAX;

ulint	srv_n_file_io_threads	= ULINT_MAX;

unknown's avatar
unknown committed
169 170
#ifdef UNIV_LOG_ARCHIVE
ibool	srv_log_archive_on	= FALSE;
171 172
ibool	srv_archive_recovery	= 0;
dulint	srv_archive_recovery_limit_lsn;
unknown's avatar
unknown committed
173
#endif /* UNIV_LOG_ARCHIVE */
174 175

ulint	srv_lock_wait_timeout	= 1024 * 1024 * 1024;
unknown's avatar
unknown committed
176

unknown's avatar
unknown committed
177 178 179
char*   srv_file_flush_method_str = NULL;
ulint   srv_unix_file_flush_method = SRV_UNIX_FDATASYNC;
ulint   srv_win_file_flush_method = SRV_WIN_IO_UNBUFFERED;
180

unknown's avatar
unknown committed
181 182
ulint	srv_max_n_open_files	  = 300;

183 184 185 186 187
/* 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. */

unknown's avatar
unknown committed
188
ulong	srv_max_buf_pool_modified_pct	= 90;
189

190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244
/* variable counts amount of data read in total (in bytes) */
ulint srv_data_read = 0;

/* here we count the amount of data written in total (in bytes) */
ulint srv_data_written = 0;

/* the number of the log write requests done */
ulint srv_log_write_requests = 0;

/* the number of physical writes to the log performed */
ulint srv_log_writes = 0;

/* amount of data written to the log files in bytes */
ulint srv_os_log_written = 0;

/* amount of writes being done to the log files */
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 */
ulint srv_log_waits = 0;

/* this variable counts the amount of times, when the doublewrite buffer
was flushed */
ulint srv_dblwr_writes = 0;

/* here we store the number of pages that have been flushed to the
doublewrite buffer */
ulint srv_dblwr_pages_written = 0;

/* in this variable we store the number of write requests issued */
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. */
ulint srv_buf_pool_wait_free = 0;

/* variable to count the number of pages that were written from buffer
pool to the disk */
ulint srv_buf_pool_flushed = 0;

/* variable to count the number of buffer pool reads that led to the
reading of a disk page */
ulint srv_buf_pool_reads = 0;

/* variable to count the number of sequential read-aheads */
ulint srv_read_ahead_seq = 0;

/* variable to count the number of random read-aheads */
ulint srv_read_ahead_rnd = 0;

/* structure to pass status variables to MySQL */
export_struc export_vars;

245
/* If the following is != 0 we do not allow inserts etc. This protects
unknown's avatar
Merge  
unknown committed
246
the user from forgetting the innodb_force_recovery keyword to my.cnf */
247 248

ulint	srv_force_recovery	= 0;
unknown's avatar
Merge  
unknown committed
249
/*-----------------------*/
250 251 252 253 254 255
/* 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. */

ulint   srv_max_n_threads       = 0;

unknown's avatar
Merge  
unknown committed
256 257 258 259 260 261 262
/* 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. */

263
ulong	srv_thread_concurrency	= SRV_CONCURRENCY_THRESHOLD;
unknown's avatar
Merge  
unknown committed
264 265 266

os_fast_mutex_t	srv_conc_mutex;		/* this mutex protects srv_conc data
					structures */
unknown's avatar
unknown committed
267 268 269 270 271 272 273
lint	srv_conc_n_threads	= 0;	/* number of OS threads currently
					inside InnoDB; it is not an error
					if this drops temporarily below zero
					because we do not demand that every
					thread increments this, but a thread
					waiting for a lock decrements this
					temporarily */
unknown's avatar
unknown committed
274 275 276
ulint	srv_conc_n_waiting_threads = 0;	/* number of OS threads waiting in the
					FIFO for a permission to enter InnoDB
					*/
unknown's avatar
Merge  
unknown committed
277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294

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 */
};

UT_LIST_BASE_NODE_T(srv_conc_slot_t)	srv_conc_queue;	/* queue of threads
							waiting to get in */
295
srv_conc_slot_t* srv_conc_slots;			/* array of wait
unknown's avatar
Merge  
unknown committed
296
							slots */
unknown's avatar
unknown committed
297 298 299

/* 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 */
unknown's avatar
unknown committed
300 301
#define SRV_FREE_TICKETS_TO_ENTER srv_n_free_tickets_to_enter
#define SRV_THREAD_SLEEP_DELAY srv_thread_sleep_delay
unknown's avatar
Merge  
unknown committed
302
/*-----------------------*/
303 304 305 306 307
/* 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). */
ulint	srv_fast_shutdown	= 0;
unknown's avatar
Merge  
unknown committed
308

309 310 311
/* Generate a innodb_status.<pid> file */
ibool	srv_innodb_status	= FALSE;

312
ibool	srv_use_doublewrite_buf	= TRUE;
313
ibool	srv_use_checksums = TRUE;
314

unknown's avatar
unknown committed
315 316
ibool   srv_set_thread_priorities = TRUE;
int     srv_query_thread_priority = 0;
unknown's avatar
unknown committed
317 318 319 320 321 322

/* TRUE if the Address Windowing Extensions of Windows are used; then we must
disable adaptive hash indexes */
ibool	srv_use_awe			= FALSE;
ibool	srv_use_adaptive_hash_indexes 	= TRUE;

323
/*-------------------------------------------*/
unknown's avatar
unknown committed
324 325 326
ulong	srv_n_spin_wait_rounds	= 20;
ulong	srv_n_free_tickets_to_enter = 500;
ulong	srv_thread_sleep_delay = 10000;
327 328 329 330 331 332 333 334 335
ulint	srv_spin_wait_delay	= 5;
ibool	srv_priority_boost	= TRUE;

ibool	srv_print_thread_releases	= FALSE;
ibool	srv_print_lock_waits		= FALSE;
ibool	srv_print_buf_io		= FALSE;
ibool	srv_print_log_io		= FALSE;
ibool	srv_print_latch_waits		= FALSE;

unknown's avatar
unknown committed
336 337 338 339 340 341 342 343
ulint		srv_n_rows_inserted		= 0;
ulint		srv_n_rows_updated		= 0;
ulint		srv_n_rows_deleted		= 0;
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;
344

345 346 347 348
ulint		srv_n_lock_wait_count		= 0;
ulint		srv_n_lock_wait_current_count	= 0;
ib_longlong	srv_n_lock_wait_time		= 0;
ulint		srv_n_lock_max_wait_time	= 0;
unknown's avatar
unknown committed
349 350


351 352 353 354 355
/*
  Set the following to 0 if you want InnoDB to write messages on
  stderr on startup/shutdown
*/
ibool	srv_print_verbose_log		= TRUE;
356
ibool	srv_print_innodb_monitor	= FALSE;
357 358
ibool   srv_print_innodb_lock_monitor   = FALSE;
ibool   srv_print_innodb_tablespace_monitor = FALSE;
359
ibool   srv_print_innodb_table_monitor = FALSE;
360

361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384
/* The parameters below are obsolete: */

ibool	srv_print_parsed_sql		= FALSE;

ulint	srv_sim_disk_wait_pct		= ULINT_MAX;
ulint	srv_sim_disk_wait_len		= ULINT_MAX;
ibool	srv_sim_disk_wait_by_yield	= FALSE;
ibool	srv_sim_disk_wait_by_wait	= FALSE;

ibool	srv_measure_contention	= FALSE;
ibool	srv_measure_by_spin	= FALSE;
	
ibool	srv_test_extra_mutexes	= FALSE;
ibool	srv_test_nocache	= FALSE;
ibool	srv_test_cache_evict	= FALSE;

ibool	srv_test_sync		= FALSE;
ulint	srv_test_n_threads	= ULINT_MAX;
ulint	srv_test_n_loops	= ULINT_MAX;
ulint	srv_test_n_free_rnds	= ULINT_MAX;
ulint	srv_test_n_reserved_rnds = ULINT_MAX;
ulint	srv_test_array_size	= ULINT_MAX;
ulint	srv_test_n_mutexes	= ULINT_MAX;

385 386 387
/* Array of English strings describing the current state of an
i/o handler thread */

388 389
const char* srv_io_thread_op_info[SRV_MAX_N_IO_THREADS];
const char* srv_io_thread_function[SRV_MAX_N_IO_THREADS];
390

unknown's avatar
unknown committed
391 392
time_t	srv_last_monitor_time;

393 394 395 396 397 398
mutex_t	srv_innodb_monitor_mutex;

/* Mutex for locking srv_monitor_file */
mutex_t	srv_monitor_file_mutex;
/* Temporary file for innodb monitor output */
FILE*	srv_monitor_file;
unknown's avatar
unknown committed
399

unknown's avatar
unknown committed
400 401 402
ulint	srv_main_thread_process_no	= 0;
ulint	srv_main_thread_id		= 0;

403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621
/*
	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 */
	ulint		type;		/* thread type: user, utility etc. */
	ibool		in_use;		/* TRUE if this slot is in use */
	ibool		suspended;	/* 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 */
srv_slot_t*	srv_mysql_table = NULL;

os_event_t	srv_lock_timeout_thread_event;

srv_sys_t*	srv_sys	= NULL;

byte		srv_pad1[64];	/* padding to prevent other memory update
				hotspots from residing on the same memory
				cache line */
mutex_t*	kernel_mutex_temp;/* mutex protecting the server, trx structs,
				query threads, and lock table */
byte		srv_pad2[64];	/* padding to prevent other memory update
				hotspots from residing on the same memory
				cache line */

/* 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. */

ulint	srv_meter[SRV_MASTER + 1];
ulint	srv_meter_low_water[SRV_MASTER + 1];
ulint	srv_meter_high_water[SRV_MASTER + 1];
ulint	srv_meter_high_water2[SRV_MASTER + 1];
ulint	srv_meter_foreground[SRV_MASTER + 1];

/* 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. */

ulint	srv_n_threads_active[SRV_MASTER + 1];
ulint	srv_n_threads[SRV_MASTER + 1];

622 623 624 625 626 627 628 629 630 631 632 633 634 635
/*************************************************************************
Sets the info describing an i/o thread current state. */

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;
}
636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727

/*************************************************************************
Accessor function to get pointer to n'th slot in the server thread
table. */
static
srv_slot_t*
srv_table_get_nth_slot(
/*===================*/
				/* out: pointer to the 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. */

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! */
static
ulint
srv_table_reserve_slot(
/*===================*/
			/* out: reserved slot index */
	ulint	type)	/* in: type of the thread: one of SRV_COM, ... */
{
	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->id = os_thread_get_curr_id();
	slot->handle = os_thread_get_curr();
	slot->type = type;

	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! */
static
os_event_t
srv_suspend_thread(void)
/*====================*/
			/* out: event for the calling thread to wait */
{
	srv_slot_t*	slot;
	os_event_t	event;
	ulint		slot_no;
	ulint		type;

728
#ifdef UNIV_SYNC_DEBUG
729
	ut_ad(mutex_own(&kernel_mutex));
730
#endif /* UNIV_SYNC_DEBUG */
731 732 733 734
	
	slot_no = thr_local_get_slot_no(os_thread_get_curr_id());

	if (srv_print_thread_releases) {
735 736
		fprintf(stderr,
			"Suspending thread %lu to slot %lu meter %lu\n",
unknown's avatar
unknown committed
737 738
			(ulong) os_thread_get_curr_id(), (ulong) slot_no,
			(ulong) srv_meter[SRV_RECOVERY]);
739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780
	}

	slot = srv_table_get_nth_slot(slot_no);

	type = slot->type;

	ut_ad(type >= SRV_WORKER);
	ut_ad(type <= SRV_MASTER);

	event = slot->event;
	
	slot->suspended = TRUE;

	ut_ad(srv_n_threads_active[type] > 0);

	srv_n_threads_active[type]--;

	os_event_reset(event);

	return(event);
}

/*************************************************************************
Releases threads of the type given from suspension in the thread table.
NOTE! The server mutex has to be reserved by the caller! */

ulint
srv_release_threads(
/*================*/
			/* out: number of threads released: this may be
			< n if not enough threads were suspended at the
			moment */
	ulint	type,	/* in: thread type */
	ulint	n)	/* in: number of threads to release */
{
	srv_slot_t*	slot;
	ulint		i;
	ulint		count	= 0;

	ut_ad(type >= SRV_WORKER);
	ut_ad(type <= SRV_MASTER);
	ut_ad(n > 0);
781
#ifdef UNIV_SYNC_DEBUG
782
	ut_ad(mutex_own(&kernel_mutex));
783
#endif /* UNIV_SYNC_DEBUG */
784 785 786 787 788
	
	for (i = 0; i < OS_THREAD_MAX_N; i++) {
	
		slot = srv_table_get_nth_slot(i);

unknown's avatar
unknown committed
789
		if (slot->in_use && slot->type == type && slot->suspended) {
790 791 792 793 794 795 796 797
			
			slot->suspended = FALSE;

			srv_n_threads_active[type]++;

			os_event_set(slot->event);

			if (srv_print_thread_releases) {
798
				fprintf(stderr,
799
		"Releasing thread %lu type %lu from slot %lu meter %lu\n",
800 801
				(ulong) slot->id, (ulong) type, (ulong) i,
		                (ulong) srv_meter[SRV_RECOVERY]);
802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844
			}

			count++;

			if (count == n) {
				break;
			}
		}
	}

	return(count);
}

/*************************************************************************
Returns the calling thread type. */

ulint
srv_get_thread_type(void)
/*=====================*/
			/* out: SRV_COM, ... */
{
	ulint		slot_no;
	srv_slot_t*	slot;
	ulint		type;

	mutex_enter(&kernel_mutex);
	
	slot_no = thr_local_get_slot_no(os_thread_get_curr_id());

	slot = srv_table_get_nth_slot(slot_no);

	type = slot->type;

	ut_ad(type >= SRV_WORKER);
	ut_ad(type <= SRV_MASTER);

	mutex_exit(&kernel_mutex);

	return(type);
}

/*************************************************************************
Initializes the server. */
845

846 847 848 849
void
srv_init(void)
/*==========*/
{
unknown's avatar
Merge  
unknown committed
850 851
	srv_conc_slot_t* 	conc_slot;
	srv_slot_t*		slot;
unknown's avatar
unknown committed
852
	dict_table_t*		table;
unknown's avatar
Merge  
unknown committed
853
	ulint			i;
854 855 856 857 858 859

	srv_sys = mem_alloc(sizeof(srv_sys_t));

	kernel_mutex_temp = mem_alloc(sizeof(mutex_t));
	mutex_create(&kernel_mutex);
	mutex_set_level(&kernel_mutex, SYNC_KERNEL);
unknown's avatar
unknown committed
860 861 862

	mutex_create(&srv_innodb_monitor_mutex);
	mutex_set_level(&srv_innodb_monitor_mutex, SYNC_NO_ORDER_CHECK);
863 864 865 866 867 868
	
	srv_sys->threads = mem_alloc(OS_THREAD_MAX_N * sizeof(srv_slot_t));

	for (i = 0; i < OS_THREAD_MAX_N; i++) {
		slot = srv_table_get_nth_slot(i);
		slot->in_use = FALSE;
unknown's avatar
unknown committed
869
                slot->type=0;	/* Avoid purify errors */
870 871 872 873 874 875 876 877 878
		slot->event = os_event_create(NULL);
		ut_a(slot->event);
	}

	srv_mysql_table = mem_alloc(OS_THREAD_MAX_N * sizeof(srv_slot_t));

	for (i = 0; i < OS_THREAD_MAX_N; i++) {
		slot = srv_mysql_table + i;
		slot->in_use = FALSE;
unknown's avatar
unknown committed
879
		slot->type = 0;
880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900
		slot->event = os_event_create(NULL);
		ut_a(slot->event);
	}

	srv_lock_timeout_thread_event = os_event_create(NULL);
	
	for (i = 0; i < SRV_MASTER + 1; i++) {
		srv_n_threads_active[i] = 0;
		srv_n_threads[i] = 0;
		srv_meter[i] = 30;
		srv_meter_low_water[i] = 50;
		srv_meter_high_water[i] = 100;
		srv_meter_high_water2[i] = 200;
		srv_meter_foreground[i] = 250;
	}
	
	srv_sys->operational = os_event_create(NULL);

	ut_a(srv_sys->operational);

	UT_LIST_INIT(srv_sys->tasks);
unknown's avatar
Merge  
unknown committed
901

unknown's avatar
unknown committed
902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926
	/* create dummy table and index for old-style infimum and supremum */
	table = dict_mem_table_create("SYS_DUMMY1",
						DICT_HDR_SPACE, 1, FALSE);
	dict_mem_table_add_col(table, "DUMMY", DATA_CHAR,
					DATA_ENGLISH | DATA_NOT_NULL, 8, 0);

	srv_sys->dummy_ind1 = dict_mem_index_create("SYS_DUMMY1",
					"SYS_DUMMY1", DICT_HDR_SPACE, 0, 1);
	dict_index_add_col(srv_sys->dummy_ind1,
			dict_table_get_nth_col(table, 0), 0, 0);
	srv_sys->dummy_ind1->table = table;
	/* create dummy table and index for new-style infimum and supremum */
	table = dict_mem_table_create("SYS_DUMMY2",
						DICT_HDR_SPACE, 1, TRUE);
	dict_mem_table_add_col(table, "DUMMY", DATA_CHAR,
					DATA_ENGLISH | DATA_NOT_NULL, 8, 0);
	srv_sys->dummy_ind2 = dict_mem_index_create("SYS_DUMMY2",
					"SYS_DUMMY2", DICT_HDR_SPACE, 0, 1);
	dict_index_add_col(srv_sys->dummy_ind2,
			dict_table_get_nth_col(table, 0), 0, 0);
	srv_sys->dummy_ind2->table = table;

	/* avoid ut_ad(index->cached) in dict_index_get_n_unique_in_tree */
	srv_sys->dummy_ind1->cached = srv_sys->dummy_ind2->cached = TRUE;

unknown's avatar
Merge  
unknown committed
927 928 929 930 931
	/* Init the server concurrency restriction data structures */

	os_fast_mutex_init(&srv_conc_mutex);
	
	UT_LIST_INIT(srv_conc_queue);
932

933 934
	srv_conc_slots = mem_alloc(OS_THREAD_MAX_N * sizeof(srv_conc_slot_t));

unknown's avatar
Merge  
unknown committed
935 936 937 938 939 940
	for (i = 0; i < OS_THREAD_MAX_N; i++) {
		conc_slot = srv_conc_slots + i;
		conc_slot->reserved = FALSE;
		conc_slot->event = os_event_create(NULL);
		ut_a(conc_slot->event);
	}
941
}	
unknown's avatar
unknown committed
942

943
/*************************************************************************
unknown's avatar
unknown committed
944
Frees the OS fast mutex created in srv_init(). */
unknown's avatar
unknown committed
945

946
void
unknown's avatar
unknown committed
947 948
srv_free(void)
/*==========*/
949
{
unknown's avatar
unknown committed
950
	os_fast_mutex_free(&srv_conc_mutex);
951 952
}

unknown's avatar
unknown committed
953
/*************************************************************************
unknown's avatar
unknown committed
954
Initializes the synchronization primitives, memory system, and the thread
unknown's avatar
unknown committed
955 956 957
local storage. */

void
unknown's avatar
unknown committed
958
srv_general_init(void)
unknown's avatar
unknown committed
959 960
/*==================*/
{
unknown's avatar
unknown committed
961 962 963 964
	os_sync_init();
	sync_init();
	mem_init(srv_mem_pool_size);
	thr_local_init();
unknown's avatar
unknown committed
965 966
}

unknown's avatar
unknown committed
967 968
/*======================= InnoDB Server FIFO queue =======================*/

969
/* Maximum allowable purge history length.  <=0 means 'infinite'. */
970
ulong	srv_max_purge_lag		= 0;
unknown's avatar
unknown committed
971

unknown's avatar
Merge  
unknown committed
972 973 974 975 976 977 978 979 980 981
/*************************************************************************
Puts an OS thread to wait if there are too many concurrent threads
(>= srv_thread_concurrency) inside InnoDB. The threads wait in a FIFO queue. */

void
srv_conc_enter_innodb(
/*==================*/
	trx_t*	trx)	/* in: transaction object associated with the
			thread */
{
unknown's avatar
unknown committed
982
	ibool			has_slept = FALSE;
983
	srv_conc_slot_t*	slot	  = NULL;
unknown's avatar
Merge  
unknown committed
984 985
	ulint			i;

unknown's avatar
unknown committed
986 987 988 989 990 991 992 993
	/* If trx has 'free tickets' to enter the engine left, then use one
	such ticket */

	if (trx->n_tickets_to_enter_innodb > 0) {
		trx->n_tickets_to_enter_innodb--;

		return;
	}
unknown's avatar
Merge  
unknown committed
994 995

	os_fast_mutex_lock(&srv_conc_mutex);
unknown's avatar
unknown committed
996
retry:
unknown's avatar
unknown committed
997 998
	if (trx->declared_to_be_inside_innodb) {
	        ut_print_timestamp(stderr);
999
		fputs(
unknown's avatar
unknown committed
1000
"  InnoDB: Error: trying to declare trx to enter InnoDB, but\n"
1001 1002 1003
"InnoDB: it already is declared.\n", stderr);
		trx_print(stderr, trx);
		putc('\n', stderr);
unknown's avatar
unknown committed
1004 1005 1006
		os_fast_mutex_unlock(&srv_conc_mutex);

		return;
unknown's avatar
unknown committed
1007 1008
	}

unknown's avatar
unknown committed
1009
	if (srv_conc_n_threads < (lint)srv_thread_concurrency) {
unknown's avatar
Merge  
unknown committed
1010

unknown's avatar
unknown committed
1011 1012 1013 1014
		srv_conc_n_threads++;
		trx->declared_to_be_inside_innodb = TRUE;
		trx->n_tickets_to_enter_innodb = SRV_FREE_TICKETS_TO_ENTER;
		
unknown's avatar
Merge  
unknown committed
1015 1016 1017 1018
		os_fast_mutex_unlock(&srv_conc_mutex);

		return;
	}
unknown's avatar
unknown committed
1019

unknown's avatar
unknown committed
1020 1021
	/* If the transaction is not holding resources, 
  let it sleep for SRV_THREAD_SLEEP_DELAY microseconds, and try again then */
unknown's avatar
unknown committed
1022
 
unknown's avatar
unknown committed
1023 1024 1025
	if (!has_slept && !trx->has_search_latch
	    && NULL == UT_LIST_GET_FIRST(trx->trx_locks)) {

unknown's avatar
unknown committed
1026 1027 1028 1029 1030 1031 1032
	        has_slept = TRUE; /* We let is sleep only once to avoid
				  starvation */

		srv_conc_n_waiting_threads++;

		os_fast_mutex_unlock(&srv_conc_mutex);

unknown's avatar
unknown committed
1033
		trx->op_info = "sleeping before joining InnoDB queue";
1034

unknown's avatar
unknown committed
1035 1036 1037 1038 1039
		/* Peter Zaitsev suggested that we take the sleep away
		altogether. But the sleep may be good in pathological
		situations of lots of thread switches. Simply put some
		threads aside for a while to reduce the number of thread
		switches. */
unknown's avatar
unknown committed
1040 1041 1042 1043
    if (SRV_THREAD_SLEEP_DELAY > 0)
    {
      os_thread_sleep(SRV_THREAD_SLEEP_DELAY);
    }
unknown's avatar
unknown committed
1044

unknown's avatar
unknown committed
1045
		trx->op_info = "";
unknown's avatar
unknown committed
1046

unknown's avatar
unknown committed
1047 1048 1049
		os_fast_mutex_lock(&srv_conc_mutex);

		srv_conc_n_waiting_threads--;
unknown's avatar
unknown committed
1050 1051

		goto retry;
unknown's avatar
unknown committed
1052
	}   
unknown's avatar
unknown committed
1053

unknown's avatar
unknown committed
1054
	/* Too many threads inside: put the current thread to a queue */
unknown's avatar
Merge  
unknown committed
1055 1056 1057 1058 1059

	for (i = 0; i < OS_THREAD_MAX_N; i++) {
		slot = srv_conc_slots + i;

		if (!slot->reserved) {
1060

unknown's avatar
Merge  
unknown committed
1061 1062 1063 1064 1065 1066 1067 1068 1069
			break;
		}
	}

	if (i == OS_THREAD_MAX_N) {
		/* Could not find a free wait slot, we must let the
		thread enter */

		srv_conc_n_threads++;
unknown's avatar
unknown committed
1070 1071
		trx->declared_to_be_inside_innodb = TRUE;
		trx->n_tickets_to_enter_innodb = 0;
unknown's avatar
Merge  
unknown committed
1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090

		os_fast_mutex_unlock(&srv_conc_mutex);

		return;
	}

	/* Release possible search system latch this thread has */
	if (trx->has_search_latch) {
		trx_search_latch_release_if_reserved(trx);
	}

	/* Add to the queue */
	slot->reserved = TRUE;
	slot->wait_ended = FALSE;
	
	UT_LIST_ADD_LAST(srv_conc_queue, srv_conc_queue, slot);

	os_event_reset(slot->event);

unknown's avatar
unknown committed
1091 1092
	srv_conc_n_waiting_threads++;

unknown's avatar
Merge  
unknown committed
1093 1094 1095 1096 1097
	os_fast_mutex_unlock(&srv_conc_mutex);

	/* Go to wait for the event; when a thread leaves InnoDB it will
	release this thread */

unknown's avatar
unknown committed
1098
	trx->op_info = "waiting in InnoDB queue";
unknown's avatar
unknown committed
1099

unknown's avatar
Merge  
unknown committed
1100 1101
	os_event_wait(slot->event);

unknown's avatar
unknown committed
1102
	trx->op_info = "";
unknown's avatar
unknown committed
1103

unknown's avatar
Merge  
unknown committed
1104 1105
	os_fast_mutex_lock(&srv_conc_mutex);

unknown's avatar
unknown committed
1106 1107
	srv_conc_n_waiting_threads--;

unknown's avatar
Merge  
unknown committed
1108 1109 1110 1111 1112 1113 1114
	/* NOTE that the thread which released this thread already
	incremented the thread counter on behalf of this thread */

	slot->reserved = FALSE;

	UT_LIST_REMOVE(srv_conc_queue, srv_conc_queue, slot);

unknown's avatar
unknown committed
1115 1116 1117
	trx->declared_to_be_inside_innodb = TRUE;
	trx->n_tickets_to_enter_innodb = SRV_FREE_TICKETS_TO_ENTER;

unknown's avatar
Merge  
unknown committed
1118 1119 1120 1121 1122 1123 1124 1125
	os_fast_mutex_unlock(&srv_conc_mutex);
}

/*************************************************************************
This lets a thread enter InnoDB regardless of the number of threads inside
InnoDB. This must be called when a thread ends a lock wait. */

void
unknown's avatar
unknown committed
1126 1127 1128 1129
srv_conc_force_enter_innodb(
/*========================*/
	trx_t*	trx)	/* in: transaction object associated with the
			thread */
unknown's avatar
Merge  
unknown committed
1130
{
1131
	if (srv_thread_concurrency >= SRV_CONCURRENCY_THRESHOLD) {
unknown's avatar
unknown committed
1132 1133 1134 1135
	
		return;
	}

unknown's avatar
Merge  
unknown committed
1136 1137 1138
	os_fast_mutex_lock(&srv_conc_mutex);

	srv_conc_n_threads++;
unknown's avatar
unknown committed
1139 1140
	trx->declared_to_be_inside_innodb = TRUE;
	trx->n_tickets_to_enter_innodb = 0;
unknown's avatar
Merge  
unknown committed
1141 1142 1143 1144 1145

	os_fast_mutex_unlock(&srv_conc_mutex);
}

/*************************************************************************
unknown's avatar
unknown committed
1146 1147
This must be called when a thread exits InnoDB in a lock wait or at the
end of an SQL statement. */
unknown's avatar
Merge  
unknown committed
1148 1149

void
unknown's avatar
unknown committed
1150 1151 1152 1153
srv_conc_force_exit_innodb(
/*=======================*/
	trx_t*	trx)	/* in: transaction object associated with the
			thread */
unknown's avatar
Merge  
unknown committed
1154 1155 1156
{
	srv_conc_slot_t*	slot	= NULL;

1157
	if (srv_thread_concurrency >= SRV_CONCURRENCY_THRESHOLD) {
unknown's avatar
unknown committed
1158 1159 1160 1161 1162 1163 1164 1165 1166
	
		return;
	}

	if (trx->declared_to_be_inside_innodb == FALSE) {
		
		return;
	}

unknown's avatar
Merge  
unknown committed
1167 1168 1169
	os_fast_mutex_lock(&srv_conc_mutex);

	srv_conc_n_threads--;
unknown's avatar
unknown committed
1170 1171
	trx->declared_to_be_inside_innodb = FALSE;
	trx->n_tickets_to_enter_innodb = 0;
unknown's avatar
Merge  
unknown committed
1172

unknown's avatar
unknown committed
1173
	if (srv_conc_n_threads < (lint)srv_thread_concurrency) {
unknown's avatar
Merge  
unknown committed
1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199
		/* Look for a slot where a thread is waiting and no other
		thread has yet released the thread */
	
		slot = UT_LIST_GET_FIRST(srv_conc_queue);

		while (slot && slot->wait_ended == TRUE) {
			slot = UT_LIST_GET_NEXT(srv_conc_queue, slot);
		}

		if (slot != NULL) {
			slot->wait_ended = TRUE;

			/* We increment the count on behalf of the released
			thread */

			srv_conc_n_threads++;
		}
	}

	os_fast_mutex_unlock(&srv_conc_mutex);

	if (slot != NULL) {
		os_event_set(slot->event);
	}
}

unknown's avatar
unknown committed
1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224
/*************************************************************************
This must be called when a thread exits InnoDB. */

void
srv_conc_exit_innodb(
/*=================*/
	trx_t*	trx)	/* in: transaction object associated with the
			thread */
{
	if (trx->n_tickets_to_enter_innodb > 0) {
		/* We will pretend the thread is still inside InnoDB though it
		now leaves the InnoDB engine. In this way we save
		a lot of semaphore operations. srv_conc_force_exit_innodb is
		used to declare the thread definitely outside InnoDB. It
		should be called when there is a lock wait or an SQL statement
		ends. */

		return;
	}

	srv_conc_force_exit_innodb(trx);
}

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

1225
/*************************************************************************
1226
Normalizes init parameter values to use units we use inside InnoDB. */
1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242
static
ulint
srv_normalize_init_values(void)
/*===========================*/
				/* out: DB_SUCCESS or error code */
{
	ulint	n;
	ulint	i;

	n = srv_n_data_files;
	
	for (i = 0; i < n; i++) {
		srv_data_file_sizes[i] = srv_data_file_sizes[i]
					* ((1024 * 1024) / UNIV_PAGE_SIZE);
	}		

unknown's avatar
unknown committed
1243 1244 1245
	srv_last_file_size_max = srv_last_file_size_max
					* ((1024 * 1024) / UNIV_PAGE_SIZE);
		
1246 1247 1248 1249
	srv_log_file_size = srv_log_file_size / UNIV_PAGE_SIZE;

	srv_log_buffer_size = srv_log_buffer_size / UNIV_PAGE_SIZE;

unknown's avatar
unknown committed
1250 1251 1252
	srv_pool_size = srv_pool_size / (UNIV_PAGE_SIZE / 1024);

	srv_awe_window_size = srv_awe_window_size / UNIV_PAGE_SIZE;
1253
	
unknown's avatar
unknown committed
1254 1255 1256 1257 1258 1259 1260
	if (srv_use_awe) {
	        /* If we are using AWE we must save memory in the 32-bit
		address space of the process, and cannot bind the lock
		table size to the real buffer pool size. */

	        srv_lock_table_size = 20 * srv_awe_window_size;
	} else {
1261
	        srv_lock_table_size = 5 * srv_pool_size;
unknown's avatar
unknown committed
1262
	}
1263 1264 1265 1266 1267

	return(DB_SUCCESS);
}

/*************************************************************************
1268
Boots the InnoDB server. */
1269 1270 1271 1272 1273 1274 1275 1276 1277

ulint
srv_boot(void)
/*==========*/
			/* out: DB_SUCCESS or error code */
{
	ulint	err;

	/* Transform the init parameter values given by MySQL to
1278
	use units we use inside InnoDB: */
1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297
	
	err = srv_normalize_init_values();

	if (err != DB_SUCCESS) {
		return(err);
	}
	
	/* Initialize synchronization primitives, memory management, and thread
	local storage */
	
	srv_general_init();

	/* Initialize this module */

	srv_init();

	return(DB_SUCCESS);
}

1298
#ifndef UNIV_HOTBACKUP
1299 1300
/*************************************************************************
Reserves a slot in the thread table for the current MySQL OS thread.
unknown's avatar
unknown committed
1301
NOTE! The kernel mutex has to be reserved by the caller! */
1302 1303 1304 1305 1306 1307 1308 1309 1310
static
srv_slot_t*
srv_table_reserve_slot_for_mysql(void)
/*==================================*/
			/* out: reserved slot */
{
	srv_slot_t*	slot;
	ulint		i;

1311
#ifdef UNIV_SYNC_DEBUG
unknown's avatar
unknown committed
1312
	ut_ad(mutex_own(&kernel_mutex));
1313
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1314

1315 1316 1317 1318 1319
	i = 0;
	slot = srv_mysql_table + i;

	while (slot->in_use) {
		i++;
unknown's avatar
unknown committed
1320 1321 1322 1323 1324 1325 1326 1327 1328

		if (i >= OS_THREAD_MAX_N) {

		        ut_print_timestamp(stderr);

		        fprintf(stderr,
"  InnoDB: There appear to be %lu MySQL threads currently waiting\n"
"InnoDB: inside InnoDB, which is the upper limit. Cannot continue operation.\n"
"InnoDB: We intentionally generate a seg fault to print a stack trace\n"
1329
"InnoDB: on Linux. But first we print a list of waiting threads.\n", (ulong) i);
unknown's avatar
unknown committed
1330 1331 1332 1333 1334 1335 1336

			for (i = 0; i < OS_THREAD_MAX_N; i++) {

			        slot = srv_mysql_table + i;

			        fprintf(stderr,
"Slot %lu: thread id %lu, type %lu, in use %lu, susp %lu, time %lu\n",
1337 1338 1339 1340
				  (ulong) i, (ulong) os_thread_pf(slot->id),
				  (ulong) slot->type, (ulong) slot->in_use,
				  (ulong) slot->suspended,
			  (ulong) difftime(ut_time(), slot->suspend_time));
unknown's avatar
unknown committed
1341 1342
			}

1343
		        ut_error;
unknown's avatar
unknown committed
1344
		}
1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356
		
		slot = srv_mysql_table + i;
	}

	ut_a(slot->in_use == FALSE);
	
	slot->in_use = TRUE;
	slot->id = os_thread_get_curr_id();
	slot->handle = os_thread_get_curr();

	return(slot);
}
1357
#endif /* !UNIV_HOTBACKUP */
1358 1359

/*******************************************************************
unknown's avatar
unknown committed
1360 1361 1362 1363 1364
Puts a MySQL OS thread to wait for a lock to be released. If an error
occurs during the wait trx->error_state associated with thr is
!= DB_SUCCESS when we return. DB_LOCK_WAIT_TIMEOUT and DB_DEADLOCK
are possible errors. DB_DEADLOCK is returned if selective deadlock
resolution chose this transaction as a victim. */
1365

unknown's avatar
unknown committed
1366
void
1367 1368
srv_suspend_mysql_thread(
/*=====================*/
unknown's avatar
unknown committed
1369 1370
	que_thr_t*	thr)	/* in: query thread associated with the MySQL
				OS thread */
1371
{
1372
#ifndef UNIV_HOTBACKUP
1373 1374 1375
	srv_slot_t*	slot;
	os_event_t	event;
	double		wait_time;
unknown's avatar
unknown committed
1376
	trx_t*		trx;
unknown's avatar
unknown committed
1377 1378
	ibool		had_dict_lock			= FALSE;
	ibool		was_declared_inside_innodb	= FALSE;
1379 1380 1381 1382 1383
	ib_longlong	start_time			= 0;
	ib_longlong	finish_time;
	ulint		diff_time;
	ulint		sec;
	ulint		ms;
unknown's avatar
unknown committed
1384

1385
#ifdef UNIV_SYNC_DEBUG
1386
	ut_ad(!mutex_own(&kernel_mutex));
1387
#endif /* UNIV_SYNC_DEBUG */
1388

unknown's avatar
unknown committed
1389 1390
	trx = thr_get_trx(thr);
	
1391 1392 1393 1394
	os_event_set(srv_lock_timeout_thread_event);

	mutex_enter(&kernel_mutex);

unknown's avatar
unknown committed
1395 1396
	trx->error_state = DB_SUCCESS;

1397 1398
	if (thr->state == QUE_THR_RUNNING) {

unknown's avatar
unknown committed
1399 1400 1401 1402 1403 1404 1405 1406 1407 1408
		ut_ad(thr->is_active == TRUE);
	
		/* The lock has already been released or this transaction
		was chosen as a deadlock victim: no need to suspend */

		if (trx->was_chosen_as_deadlock_victim) {

			trx->error_state = DB_DEADLOCK;
			trx->was_chosen_as_deadlock_victim = FALSE;
		}
1409 1410 1411

		mutex_exit(&kernel_mutex);

unknown's avatar
unknown committed
1412
		return;
1413 1414
	}
	
unknown's avatar
unknown committed
1415 1416
	ut_ad(thr->is_active == FALSE);

1417 1418 1419 1420 1421 1422 1423 1424 1425
	slot = srv_table_reserve_slot_for_mysql();

	event = slot->event;
	
	slot->thr = thr;

	os_event_reset(event);	

	slot->suspend_time = ut_time();
unknown's avatar
unknown committed
1426

1427 1428 1429
	if (thr->lock_state == QUE_THR_LOCK_ROW) {
		srv_n_lock_wait_count++;
		srv_n_lock_wait_current_count++;
1430

1431 1432 1433
		ut_usectime(&sec, &ms);
		start_time = (ib_longlong)sec * 1000000 + ms;
	}
1434 1435 1436 1437 1438 1439
	/* Wake the lock timeout monitor thread, if it is suspended */

	os_event_set(srv_lock_timeout_thread_event);
	
	mutex_exit(&kernel_mutex);

unknown's avatar
unknown committed
1440 1441 1442
	if (trx->declared_to_be_inside_innodb) {

		was_declared_inside_innodb = TRUE;
unknown's avatar
Merge  
unknown committed
1443
	
unknown's avatar
unknown committed
1444 1445 1446
		/* We must declare this OS thread to exit InnoDB, since a
		possible other thread holding a lock which this thread waits
		for must be allowed to enter, sooner or later */
unknown's avatar
Merge  
unknown committed
1447
	
unknown's avatar
unknown committed
1448 1449
		srv_conc_force_exit_innodb(trx);
	}
unknown's avatar
Merge  
unknown committed
1450

unknown's avatar
unknown committed
1451
	/* Release possible foreign key check latch */
unknown's avatar
unknown committed
1452 1453 1454
	if (trx->dict_operation_lock_mode == RW_S_LATCH) {

		had_dict_lock = TRUE;
unknown's avatar
unknown committed
1455

unknown's avatar
unknown committed
1456
		row_mysql_unfreeze_data_dictionary(trx);
unknown's avatar
unknown committed
1457 1458
	}

unknown's avatar
unknown committed
1459 1460
	ut_a(trx->dict_operation_lock_mode == 0);

1461 1462 1463 1464
	/* Wait for the release */
	
	os_event_wait(event);

unknown's avatar
unknown committed
1465
	if (had_dict_lock) {
unknown's avatar
unknown committed
1466

unknown's avatar
unknown committed
1467
		row_mysql_freeze_data_dictionary(trx);
unknown's avatar
unknown committed
1468 1469
	}

unknown's avatar
unknown committed
1470 1471 1472
	if (was_declared_inside_innodb) {

		/* Return back inside InnoDB */
unknown's avatar
Merge  
unknown committed
1473
	
unknown's avatar
unknown committed
1474 1475
		srv_conc_force_enter_innodb(trx);
	}
unknown's avatar
Merge  
unknown committed
1476

1477 1478 1479 1480 1481 1482 1483
	mutex_enter(&kernel_mutex);

	/* Release the slot for others to use */
	
	slot->in_use = FALSE;

	wait_time = ut_difftime(ut_time(), slot->suspend_time);
unknown's avatar
unknown committed
1484

1485 1486 1487 1488
	if (thr->lock_state == QUE_THR_LOCK_ROW) {
		ut_usectime(&sec, &ms);
		finish_time = (ib_longlong)sec * 1000000 + ms;

1489
		diff_time = (ulint) (finish_time - start_time);
unknown's avatar
unknown committed
1490
  
1491 1492 1493 1494 1495 1496
		srv_n_lock_wait_current_count--;
		srv_n_lock_wait_time = srv_n_lock_wait_time + diff_time;
		if (diff_time > srv_n_lock_max_wait_time) {
			srv_n_lock_max_wait_time = diff_time;
		}
	}
unknown's avatar
unknown committed
1497

unknown's avatar
unknown committed
1498 1499 1500 1501 1502 1503
	if (trx->was_chosen_as_deadlock_victim) {

		trx->error_state = DB_DEADLOCK;
		trx->was_chosen_as_deadlock_victim = FALSE;
	}

1504 1505 1506 1507 1508
	mutex_exit(&kernel_mutex);

	if (srv_lock_wait_timeout < 100000000 && 
	    			wait_time > (double)srv_lock_wait_timeout) {

unknown's avatar
unknown committed
1509 1510
	    	trx->error_state = DB_LOCK_WAIT_TIMEOUT;
	}
1511 1512 1513 1514 1515 1516
#else /* UNIV_HOTBACKUP */
	/* This function depends on MySQL code that is not included in
	InnoDB Hot Backup builds.  Besides, this function should never
	be called in InnoDB Hot Backup. */
	ut_error;
#endif /* UNIV_HOTBACKUP */
1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528
}

/************************************************************************
Releases a MySQL OS thread waiting for a lock to be released, if the
thread is already suspended. */

void
srv_release_mysql_thread_if_suspended(
/*==================================*/
	que_thr_t*	thr)	/* in: query thread associated with the
				MySQL OS thread  */
{
1529
#ifndef UNIV_HOTBACKUP
1530 1531 1532
	srv_slot_t*	slot;
	ulint		i;
	
1533
#ifdef UNIV_SYNC_DEBUG
1534
	ut_ad(mutex_own(&kernel_mutex));
1535
#endif /* UNIV_SYNC_DEBUG */
1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550

	for (i = 0; i < OS_THREAD_MAX_N; i++) {

		slot = srv_mysql_table + i;

		if (slot->in_use && slot->thr == thr) {
			/* Found */

			os_event_set(slot->event);

			return;
		}
	}

	/* not found */
1551 1552 1553 1554 1555 1556
#else /* UNIV_HOTBACKUP */
	/* This function depends on MySQL code that is not included in
	InnoDB Hot Backup builds.  Besides, this function should never
	be called in InnoDB Hot Backup. */
	ut_error;
#endif /* UNIV_HOTBACKUP */
1557 1558
}

1559
#ifndef UNIV_HOTBACKUP
unknown's avatar
unknown committed
1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587
/**********************************************************************
Refreshes the values used to calculate per-second averages. */
static
void
srv_refresh_innodb_monitor_stats(void)
/*==================================*/
{
	mutex_enter(&srv_innodb_monitor_mutex);

	srv_last_monitor_time = time(NULL);

	os_aio_refresh_stats();

	btr_cur_n_sea_old = btr_cur_n_sea;
	btr_cur_n_non_sea_old = btr_cur_n_non_sea;

	log_refresh_stats();
	
	buf_refresh_io_stats();

	srv_n_rows_inserted_old = srv_n_rows_inserted;
	srv_n_rows_updated_old = srv_n_rows_updated;
	srv_n_rows_deleted_old = srv_n_rows_deleted;
	srv_n_rows_read_old = srv_n_rows_read;

	mutex_exit(&srv_innodb_monitor_mutex);
}

unknown's avatar
unknown committed
1588
/**********************************************************************
1589
Outputs to a file the output of the InnoDB Monitor. */
1590

unknown's avatar
unknown committed
1591
void
1592 1593
srv_printf_innodb_monitor(
/*======================*/
1594 1595 1596 1597 1598
	FILE*	file,		/* in: output stream */
	ulint*	trx_start,	/* out: file position of the start of
				the list of active transactions */
	ulint*	trx_end)	/* out: file position of the end of
				the list of active transactions */
1599
{
unknown's avatar
unknown committed
1600 1601
	double	time_elapsed;
	time_t	current_time;
unknown's avatar
unknown committed
1602
	ulint   n_reserved;
unknown's avatar
Merge  
unknown committed
1603

unknown's avatar
unknown committed
1604
	mutex_enter(&srv_innodb_monitor_mutex);
1605

unknown's avatar
unknown committed
1606
	current_time = time(NULL);
1607

unknown's avatar
unknown committed
1608 1609 1610
	/* We add 0.001 seconds to time_elapsed to prevent division
	by zero if two users happen to call SHOW INNODB STATUS at the same
	time */
1611
	
unknown's avatar
unknown committed
1612 1613
	time_elapsed = difftime(current_time, srv_last_monitor_time)
			+ 0.001;
1614

unknown's avatar
unknown committed
1615
	srv_last_monitor_time = time(NULL);
1616

1617
	fputs("\n=====================================\n", file);
unknown's avatar
unknown committed
1618

1619 1620 1621 1622 1623 1624
	ut_print_timestamp(file);
	fprintf(file,
		" INNODB MONITOR OUTPUT\n"
		"=====================================\n"
		"Per second averages calculated from the last %lu seconds\n",
		(ulong)time_elapsed);
unknown's avatar
unknown committed
1625

1626 1627 1628 1629
	fputs("----------\n"
		"SEMAPHORES\n"
		"----------\n", file);
	sync_print(file);
unknown's avatar
unknown committed
1630

unknown's avatar
unknown committed
1631 1632 1633 1634 1635 1636 1637
	/* Conceptually, srv_innodb_monitor_mutex has a very high latching
	order level in sync0sync.h, while dict_foreign_err_mutex has a very
	low level 135. Therefore we can reserve the latter mutex here without
	a danger of a deadlock of threads. */

	mutex_enter(&dict_foreign_err_mutex);

1638 1639 1640 1641 1642 1643
	if (ftell(dict_foreign_err_file) != 0L) {
		fputs("------------------------\n"
			"LATEST FOREIGN KEY ERROR\n"
			"------------------------\n", file);
		ut_copy_file(file, dict_foreign_err_file);
	}
unknown's avatar
unknown committed
1644

unknown's avatar
unknown committed
1645
	mutex_exit(&dict_foreign_err_mutex);
1646

1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664
	lock_print_info_summary(file);
	if (trx_start) {
		long	t = ftell(file);
		if (t < 0) {
			*trx_start = ULINT_UNDEFINED;
		} else {
			*trx_start = (ulint) t;
		}
	}
	lock_print_info_all_transactions(file);
	if (trx_end) {
		long	t = ftell(file);
		if (t < 0) {
			*trx_end = ULINT_UNDEFINED;
		} else {
			*trx_end = (ulint) t;
		}
	}
1665 1666 1667 1668
	fputs("--------\n"
		"FILE I/O\n"
		"--------\n", file);
	os_aio_print(file);
1669

1670 1671 1672 1673 1674 1675 1676 1677
	fputs("-------------------------------------\n"
		"INSERT BUFFER AND ADAPTIVE HASH INDEX\n"
		"-------------------------------------\n", file);
	ibuf_print(file);

	ha_print_info(file, btr_search_sys->hash_index);

	fprintf(file,
unknown's avatar
unknown committed
1678 1679 1680 1681 1682
		"%.2f hash searches/s, %.2f non-hash searches/s\n",
			(btr_cur_n_sea - btr_cur_n_sea_old)
						/ time_elapsed,
			(btr_cur_n_non_sea - btr_cur_n_non_sea_old)
						/ time_elapsed);
unknown's avatar
unknown committed
1683 1684
	btr_cur_n_sea_old = btr_cur_n_sea;
	btr_cur_n_non_sea_old = btr_cur_n_non_sea;
unknown's avatar
unknown committed
1685

1686
	fputs("---\n"
1687
		       "LOG\n"
1688 1689 1690 1691
		"---\n", file);
	log_print(file);

	fputs("----------------------\n"
1692
		       "BUFFER POOL AND MEMORY\n"
1693 1694
		"----------------------\n", file);
	fprintf(file,
unknown's avatar
unknown committed
1695
	"Total memory allocated " ULINTPF
unknown's avatar
unknown committed
1696
	"; in additional pool allocated " ULINTPF "\n",
unknown's avatar
unknown committed
1697 1698
				ut_total_allocated_memory,
				mem_pool_get_reserved(mem_comm_pool));
1699

unknown's avatar
unknown committed
1700
	if (srv_use_awe) {
unknown's avatar
unknown committed
1701
		fprintf(file,
unknown's avatar
unknown committed
1702
	"In addition to that %lu MB of AWE memory allocated\n",
unknown's avatar
unknown committed
1703
		(ulong) (srv_pool_size / ((1024 * 1024) / UNIV_PAGE_SIZE)));
unknown's avatar
unknown committed
1704 1705
	}
	
1706
	buf_print_io(file);
1707

1708 1709 1710 1711
	fputs("--------------\n"
		"ROW OPERATIONS\n"
		"--------------\n", file);
	fprintf(file, "%ld queries inside InnoDB, %lu queries in queue\n",
unknown's avatar
unknown committed
1712 1713
       		       (long) srv_conc_n_threads,
		       (ulong) srv_conc_n_waiting_threads);
unknown's avatar
unknown committed
1714 1715
        n_reserved = fil_space_get_n_reserved_extents(0);
        if (n_reserved > 0) {
unknown's avatar
unknown committed
1716
                fprintf(file,
unknown's avatar
unknown committed
1717
        "%lu tablespace extents now reserved for B-tree split operations\n",
1718
                                                    (ulong) n_reserved);
unknown's avatar
unknown committed
1719
        }
1720

unknown's avatar
unknown committed
1721
#ifdef UNIV_LINUX
1722
	fprintf(file, "Main thread process no. %lu, id %lu, state: %s\n",
1723 1724 1725
		       (ulong) srv_main_thread_process_no,
		       (ulong) srv_main_thread_id,
		       srv_main_thread_op_info);
unknown's avatar
unknown committed
1726
#else
1727
	fprintf(file, "Main thread id %lu, state: %s\n",
1728
			(ulong) srv_main_thread_id,
unknown's avatar
unknown committed
1729
			srv_main_thread_op_info);
unknown's avatar
unknown committed
1730
#endif
1731
	fprintf(file,
unknown's avatar
unknown committed
1732 1733 1734 1735 1736 1737
	"Number of rows inserted " ULINTPF
	", updated " ULINTPF ", deleted " ULINTPF ", read " ULINTPF "\n",
			srv_n_rows_inserted, 
			srv_n_rows_updated, 
			srv_n_rows_deleted, 
			srv_n_rows_read);
1738
	fprintf(file,
1739 1740 1741 1742 1743 1744 1745 1746 1747 1748
	"%.2f inserts/s, %.2f updates/s, %.2f deletes/s, %.2f reads/s\n",
			(srv_n_rows_inserted - srv_n_rows_inserted_old)
						/ time_elapsed,
			(srv_n_rows_updated - srv_n_rows_updated_old)
						/ time_elapsed,
			(srv_n_rows_deleted - srv_n_rows_deleted_old)
						/ time_elapsed,
			(srv_n_rows_read - srv_n_rows_read_old)
						/ time_elapsed);

unknown's avatar
unknown committed
1749
  srv_n_rows_inserted_old = srv_n_rows_inserted;
unknown's avatar
unknown committed
1750 1751 1752
	srv_n_rows_updated_old = srv_n_rows_updated;
	srv_n_rows_deleted_old = srv_n_rows_deleted;
	srv_n_rows_read_old = srv_n_rows_read;
1753

unknown's avatar
unknown committed
1754
  fputs("----------------------------\n"
1755
		       "END OF INNODB MONITOR OUTPUT\n"
1756
		"============================\n", file);
unknown's avatar
unknown committed
1757
	mutex_exit(&srv_innodb_monitor_mutex);
1758
	fflush(file);
unknown's avatar
unknown committed
1759 1760
}

1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804
/**********************************************************************
Function to pass InnoDB status variables to MySQL */

void
srv_export_innodb_status(void)
{

        mutex_enter(&srv_innodb_monitor_mutex);
        export_vars.innodb_data_pending_reads= os_n_pending_reads;
        export_vars.innodb_data_pending_writes= os_n_pending_writes;
        export_vars.innodb_data_pending_fsyncs= 
                fil_n_pending_log_flushes + fil_n_pending_tablespace_flushes;
        export_vars.innodb_data_fsyncs= os_n_fsyncs;
        export_vars.innodb_data_read= srv_data_read;
        export_vars.innodb_data_reads= os_n_file_reads;
        export_vars.innodb_data_writes= os_n_file_writes;
        export_vars.innodb_data_written= srv_data_written;
        export_vars.innodb_buffer_pool_read_requests= buf_pool->n_page_gets;
        export_vars.innodb_buffer_pool_write_requests= srv_buf_pool_write_requests;
        export_vars.innodb_buffer_pool_wait_free= srv_buf_pool_wait_free;
        export_vars.innodb_buffer_pool_pages_flushed= srv_buf_pool_flushed;
        export_vars.innodb_buffer_pool_reads= srv_buf_pool_reads;
        export_vars.innodb_buffer_pool_read_ahead_rnd= srv_read_ahead_rnd;
        export_vars.innodb_buffer_pool_read_ahead_seq= srv_read_ahead_seq;
        export_vars.innodb_buffer_pool_pages_data= UT_LIST_GET_LEN(buf_pool->LRU);
        export_vars.innodb_buffer_pool_pages_dirty= UT_LIST_GET_LEN(buf_pool->flush_list);
        export_vars.innodb_buffer_pool_pages_free= UT_LIST_GET_LEN(buf_pool->free);
        export_vars.innodb_buffer_pool_pages_latched= buf_get_latched_pages_number();
        export_vars.innodb_buffer_pool_pages_total= buf_pool->curr_size;
        export_vars.innodb_buffer_pool_pages_misc= buf_pool->max_size -
          UT_LIST_GET_LEN(buf_pool->LRU) - UT_LIST_GET_LEN(buf_pool->free);
        export_vars.innodb_page_size= UNIV_PAGE_SIZE;
        export_vars.innodb_log_waits= srv_log_waits;
        export_vars.innodb_os_log_written= srv_os_log_written;
        export_vars.innodb_os_log_fsyncs= fil_n_log_flushes;
        export_vars.innodb_os_log_pending_fsyncs= fil_n_pending_log_flushes;
        export_vars.innodb_os_log_pending_writes= srv_os_log_pending_writes;
        export_vars.innodb_log_write_requests= srv_log_write_requests;
        export_vars.innodb_log_writes= srv_log_writes;
        export_vars.innodb_dblwr_pages_written= srv_dblwr_pages_written;
        export_vars.innodb_dblwr_writes= srv_dblwr_writes;
        export_vars.innodb_pages_created= buf_pool->n_pages_created;
        export_vars.innodb_pages_read= buf_pool->n_pages_read;
        export_vars.innodb_pages_written= buf_pool->n_pages_written;
unknown's avatar
unknown committed
1805 1806 1807
        export_vars.innodb_row_lock_waits= srv_n_lock_wait_count;
        export_vars.innodb_row_lock_current_waits= srv_n_lock_wait_current_count;
        export_vars.innodb_row_lock_time= srv_n_lock_wait_time / 10000;
1808 1809 1810 1811 1812 1813
	if (srv_n_lock_wait_count > 0) {
		export_vars.innodb_row_lock_time_avg = (ulint)
			(srv_n_lock_wait_time / 10000 / srv_n_lock_wait_count);
	} else {
		export_vars.innodb_row_lock_time_avg = 0;
	}
unknown's avatar
unknown committed
1814
        export_vars.innodb_row_lock_time_max= srv_n_lock_max_wait_time / 10000;
1815 1816 1817 1818 1819
        export_vars.innodb_rows_read= srv_n_rows_read;
        export_vars.innodb_rows_inserted= srv_n_rows_inserted;
        export_vars.innodb_rows_updated= srv_n_rows_updated;
        export_vars.innodb_rows_deleted= srv_n_rows_deleted;
        mutex_exit(&srv_innodb_monitor_mutex);
unknown's avatar
unknown committed
1820

1821 1822
}

unknown's avatar
unknown committed
1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834
/*************************************************************************
A thread which wakes up threads whose lock wait may have lasted too long.
This also prints the info output by various InnoDB monitors. */

#ifndef __WIN__
void*
#else
ulint
#endif
srv_lock_timeout_and_monitor_thread(
/*================================*/
			/* out: a dummy parameter */
1835 1836
	void*	arg __attribute__((unused)))
			/* in: a dummy parameter required by
unknown's avatar
unknown committed
1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847
			os_thread_create */
{
	srv_slot_t*	slot;
	double		time_elapsed;
	time_t          current_time;
	time_t		last_table_monitor_time;
	time_t		last_monitor_time;
	ibool		some_waits;
	double		wait_time;
	ulint		i;

1848
#ifdef UNIV_DEBUG_THREAD_CREATION
1849
	fprintf(stderr, "Lock timeout thread starts, id %lu\n",
unknown's avatar
unknown committed
1850
			     os_thread_pf(os_thread_get_curr_id()));
1851
#endif
unknown's avatar
unknown committed
1852 1853 1854 1855 1856 1857 1858 1859 1860 1861
	UT_NOT_USED(arg);
	srv_last_monitor_time = time(NULL);
	last_table_monitor_time = time(NULL);
	last_monitor_time = time(NULL);
loop:
	srv_lock_timeout_and_monitor_active = TRUE;

	/* When someone is waiting for a lock, we wake up every second
	and check if a timeout has passed for a lock wait */

unknown's avatar
unknown committed
1862
	os_thread_sleep(1000000);
unknown's avatar
unknown committed
1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877

	/* In case mutex_exit is not a memory barrier, it is
	theoretically possible some threads are left waiting though
	the semaphore is already released. Wake up those threads: */
	
	sync_arr_wake_threads_if_sema_free();

	current_time = time(NULL);

	time_elapsed = difftime(current_time, last_monitor_time);
	
	if (time_elapsed > 15) {
	    last_monitor_time = time(NULL);

	    if (srv_print_innodb_monitor) {
1878
		srv_printf_innodb_monitor(stderr, NULL, NULL);
1879
	    }
unknown's avatar
unknown committed
1880

unknown's avatar
unknown committed
1881 1882 1883
	    if (srv_innodb_status) {
		mutex_enter(&srv_monitor_file_mutex);
		rewind(srv_monitor_file);
1884
		srv_printf_innodb_monitor(srv_monitor_file, NULL, NULL);
unknown's avatar
unknown committed
1885 1886 1887
		os_file_set_eof(srv_monitor_file);
		mutex_exit(&srv_monitor_file_mutex);
	    }
1888

1889 1890
	    if (srv_print_innodb_tablespace_monitor
		&& difftime(current_time, last_table_monitor_time) > 60) {
unknown's avatar
unknown committed
1891 1892 1893

		last_table_monitor_time = time(NULL);	

1894 1895
		fputs("================================================\n",
			stderr);
1896

1897
		ut_print_timestamp(stderr);
1898

1899 1900 1901
		fputs(" INNODB TABLESPACE MONITOR OUTPUT\n"
			"================================================\n",
			stderr);
1902 1903
	       
		fsp_print(0);
1904
		fputs("Validating tablespace\n", stderr);
1905
		fsp_validate(0);
1906 1907
		fputs("Validation ok\n"
			"---------------------------------------\n"
1908
	       		"END OF INNODB TABLESPACE MONITOR OUTPUT\n"
1909 1910
			"=======================================\n",
			stderr);
1911 1912
	    }

unknown's avatar
unknown committed
1913
	    if (srv_print_innodb_table_monitor
1914
		&& difftime(current_time, last_table_monitor_time) > 60) {
unknown's avatar
unknown committed
1915 1916

		last_table_monitor_time = time(NULL);	
1917

1918
		fputs("===========================================\n", stderr);
1919

1920
		ut_print_timestamp(stderr);
1921

1922 1923 1924
		fputs(" INNODB TABLE MONITOR OUTPUT\n"
			"===========================================\n",
			stderr);
1925 1926
	    	dict_print();

1927
		fputs("-----------------------------------\n"
1928
	       		"END OF INNODB TABLE MONITOR OUTPUT\n"
1929 1930
			"==================================\n",
			stderr);
1931 1932 1933
	    }
	}

1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953
	mutex_enter(&kernel_mutex);

	some_waits = FALSE;

	/* Check of all slots if a thread is waiting there, and if it
	has exceeded the time limit */
	
	for (i = 0; i < OS_THREAD_MAX_N; i++) {

		slot = srv_mysql_table + i;

		if (slot->in_use) {
			some_waits = TRUE;

			wait_time = ut_difftime(ut_time(), slot->suspend_time);
			
			if (srv_lock_wait_timeout < 100000000 && 
	    			(wait_time > (double) srv_lock_wait_timeout
						|| wait_time < 0)) {

unknown's avatar
Merge  
unknown committed
1954
				/* Timeout exceeded or a wrap-around in system
1955
				time counter: cancel the lock request queued
1956
				by the transaction and release possible
unknown's avatar
unknown committed
1957 1958 1959 1960 1961 1962 1963 1964
				other transactions waiting behind; it is
				possible that the lock has already been
				granted: in that case do nothing */

			        if (thr_get_trx(slot->thr)->wait_lock) {
				        lock_cancel_waiting_and_release(
				          thr_get_trx(slot->thr)->wait_lock);
			        }
1965 1966 1967 1968 1969 1970 1971 1972
			}
		}
	}

	os_event_reset(srv_lock_timeout_thread_event);

	mutex_exit(&kernel_mutex);

unknown's avatar
Merge  
unknown committed
1973 1974 1975 1976
	if (srv_shutdown_state >= SRV_SHUTDOWN_CLEANUP) {
		goto exit_func;
	}

1977 1978 1979 1980
	if (some_waits || srv_print_innodb_monitor
			|| srv_print_innodb_lock_monitor
			|| srv_print_innodb_tablespace_monitor
			|| srv_print_innodb_table_monitor) {
1981 1982 1983
		goto loop;
	}

1984 1985
	/* No one was waiting for a lock and no monitor was active:
	suspend this thread */
unknown's avatar
Merge  
unknown committed
1986 1987 1988

	srv_lock_timeout_and_monitor_active = FALSE;

1989 1990 1991
#if 0
	/* The following synchronisation is disabled, since
	the InnoDB monitor output is to be updated every 15 seconds. */
1992
	os_event_wait(srv_lock_timeout_thread_event);
1993
#endif
1994 1995
	goto loop;

unknown's avatar
Merge  
unknown committed
1996 1997 1998
exit_func:
	srv_lock_timeout_and_monitor_active = FALSE;

1999 2000 2001 2002
	/* We count the number of threads in os_thread_exit(). A created
	thread should always use that to exit and not use return() to exit. */

	os_thread_exit(NULL);
2003
#ifndef __WIN__
2004
	return(NULL);
2005
#else
2006
	return(0);
2007
#endif
2008 2009
}

2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
/*************************************************************************
A thread which prints warnings about semaphore waits which have lasted
too long. These can be used to track bugs which cause hangs. */

#ifndef __WIN__
void*
#else
ulint
#endif
srv_error_monitor_thread(
/*=====================*/
			/* out: a dummy parameter */
2022 2023
	void*	arg __attribute__((unused)))
			/* in: a dummy parameter required by
2024 2025
			os_thread_create */
{
2026 2027
	/* number of successive fatal timeouts observed */
	ulint	fatal_cnt	= 0;
unknown's avatar
unknown committed
2028 2029
	dulint	old_lsn;
	dulint	new_lsn;
unknown's avatar
unknown committed
2030

unknown's avatar
unknown committed
2031 2032
	old_lsn = srv_start_lsn;

2033
#ifdef UNIV_DEBUG_THREAD_CREATION
2034
	fprintf(stderr, "Error monitor thread starts, id %lu\n",
unknown's avatar
unknown committed
2035
			      os_thread_pf(os_thread_get_curr_id()));
2036
#endif
2037
loop:
unknown's avatar
Merge  
unknown committed
2038 2039
	srv_error_monitor_active = TRUE;

unknown's avatar
unknown committed
2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050
	/* Try to track a strange bug reported by Harald Fuchs and others,
	where the lsn seems to decrease at times */

	new_lsn = log_get_lsn();

	if (ut_dulint_cmp(new_lsn, old_lsn) < 0) {
		ut_print_timestamp(stderr);
		fprintf(stderr,
"  InnoDB: Error: old log sequence number %lu %lu was greater\n"
"InnoDB: than the new log sequence number %lu %lu!\n"
"InnoDB: Please send a bug report to mysql@lists.mysql.com\n",
2051 2052 2053 2054
		(ulong) ut_dulint_get_high(old_lsn),
		(ulong) ut_dulint_get_low(old_lsn),
		(ulong) ut_dulint_get_high(new_lsn),
		(ulong) ut_dulint_get_low(new_lsn));
unknown's avatar
unknown committed
2055 2056 2057 2058
	}

	old_lsn = new_lsn;

unknown's avatar
unknown committed
2059 2060 2061
	if (difftime(time(NULL), srv_last_monitor_time) > 60) {
		/* We referesh InnoDB Monitor values so that averages are
		printed from at most 60 last seconds */
2062

unknown's avatar
unknown committed
2063 2064 2065
		srv_refresh_innodb_monitor_stats();
	}

2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079
	if (sync_array_print_long_waits()) {
		fatal_cnt++;
		if (fatal_cnt > 5) {

			fprintf(stderr,
"InnoDB: Error: semaphore wait has lasted > %lu seconds\n"
"InnoDB: We intentionally crash the server, because it appears to be hung.\n",
				srv_fatal_semaphore_wait_threshold);

			ut_error;
		}
	} else {
		fatal_cnt = 0;
	}
unknown's avatar
unknown committed
2080

2081
	/* Flush stderr so that a database user gets the output
unknown's avatar
unknown committed
2082 2083 2084 2085
	to possible MySQL error file */

	fflush(stderr);

unknown's avatar
unknown committed
2086 2087
	os_thread_sleep(2000000);

unknown's avatar
Merge  
unknown committed
2088 2089 2090 2091 2092 2093
	if (srv_shutdown_state < SRV_SHUTDOWN_LAST_PHASE) {

		goto loop;
	}

	srv_error_monitor_active = FALSE;
2094

2095 2096 2097
	/* We count the number of threads in os_thread_exit(). A created
	thread should always use that to exit and not use return() to exit. */

unknown's avatar
unknown committed
2098 2099
	os_thread_exit(NULL);

2100
#ifndef __WIN__
2101
	return(NULL);
2102 2103 2104 2105 2106
#else
	return(0);
#endif
}

2107
/***********************************************************************
2108
Tells the InnoDB server that there has been activity in the database
2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129
and wakes up the master thread if it is suspended (not sleeping). Used
in the MySQL interface. Note that there is a small chance that the master
thread stays suspended (we do not protect our operation with the kernel
mutex, for performace reasons). */

void
srv_active_wake_master_thread(void)
/*===============================*/
{
	srv_activity_count++;
			
	if (srv_n_threads_active[SRV_MASTER] == 0) {

		mutex_enter(&kernel_mutex);

		srv_release_threads(SRV_MASTER, 1);

		mutex_exit(&kernel_mutex);
	}
}

unknown's avatar
unknown committed
2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145
/***********************************************************************
Wakes up the master thread if it is suspended or being suspended. */

void
srv_wake_master_thread(void)
/*========================*/
{
	srv_activity_count++;
			
	mutex_enter(&kernel_mutex);

	srv_release_threads(SRV_MASTER, 1);

	mutex_exit(&kernel_mutex);
}

2146 2147 2148
/*************************************************************************
The master thread controlling the server. */

2149 2150 2151
#ifndef __WIN__
void*
#else
2152
ulint
2153
#endif
2154 2155 2156
srv_master_thread(
/*==============*/
			/* out: a dummy parameter */
2157 2158
	void*	arg __attribute__((unused)))
			/* in: a dummy parameter required by
2159 2160 2161
			os_thread_create */
{
	os_event_t	event;
2162 2163
	time_t          last_flush_time;
	time_t          current_time;
2164 2165 2166 2167 2168
	ulint		old_activity_count;
	ulint		n_pages_purged;
	ulint		n_bytes_merged;
	ulint		n_pages_flushed;
	ulint		n_bytes_archived;
unknown's avatar
unknown committed
2169
	ulint		n_tables_to_drop;
2170 2171 2172 2173
	ulint		n_ios;
	ulint		n_ios_old;
	ulint		n_ios_very_old;
	ulint		n_pend_ios;
unknown's avatar
unknown committed
2174
	ibool		skip_sleep	= FALSE;
2175 2176
	ulint		i;
	
2177
#ifdef UNIV_DEBUG_THREAD_CREATION
2178
	fprintf(stderr, "Master thread starts, id %lu\n",
unknown's avatar
unknown committed
2179
			      os_thread_pf(os_thread_get_curr_id()));
2180
#endif
unknown's avatar
unknown committed
2181 2182 2183
	srv_main_thread_process_no = os_proc_get_number();
	srv_main_thread_id = os_thread_pf(os_thread_get_curr_id());
	
2184 2185 2186 2187 2188 2189 2190 2191 2192 2193
	srv_table_reserve_slot(SRV_MASTER);	

	mutex_enter(&kernel_mutex);

	srv_n_threads_active[SRV_MASTER]++;

	mutex_exit(&kernel_mutex);

	os_event_set(srv_sys->operational);
loop:
2194 2195 2196 2197
	/*****************************************************************/
	/* ---- When there is database activity by users, we cycle in this
	loop */

unknown's avatar
unknown committed
2198
	srv_main_thread_op_info = "reserving kernel mutex";
2199 2200 2201

	n_ios_very_old = log_sys->n_log_ios + buf_pool->n_pages_read
						+ buf_pool->n_pages_written;
2202 2203
	mutex_enter(&kernel_mutex);

2204
	/* Store the user activity counter at the start of this loop */
2205 2206 2207 2208
	old_activity_count = srv_activity_count;

	mutex_exit(&kernel_mutex);

2209 2210 2211 2212 2213 2214 2215 2216
	if (srv_force_recovery >= SRV_FORCE_NO_BACKGROUND) {

		goto suspend_thread;
	}

	/* ---- We run the following loop approximately once per second
	when there is database activity */

unknown's avatar
unknown committed
2217
	skip_sleep = FALSE;
2218 2219

	for (i = 0; i < 10; i++) {
2220 2221
		n_ios_old = log_sys->n_log_ios + buf_pool->n_pages_read
						+ buf_pool->n_pages_written;
unknown's avatar
unknown committed
2222
		srv_main_thread_op_info = "sleeping";
unknown's avatar
unknown committed
2223 2224 2225 2226 2227 2228 2229
		
		if (!skip_sleep) {

		        os_thread_sleep(1000000);
		}

		skip_sleep = FALSE;
2230

unknown's avatar
unknown committed
2231 2232 2233 2234
		/* ALTER TABLE in MySQL requires on Unix that the table handler
		can drop tables lazily after there no longer are SELECT
		queries to them. */

unknown's avatar
unknown committed
2235
		srv_main_thread_op_info = "doing background drop tables";
unknown's avatar
unknown committed
2236 2237 2238

		row_drop_tables_for_mysql_in_background();

unknown's avatar
unknown committed
2239
		srv_main_thread_op_info = "";
unknown's avatar
unknown committed
2240

unknown's avatar
unknown committed
2241
		if (srv_fast_shutdown && srv_shutdown_state > 0) {
2242

unknown's avatar
unknown committed
2243
			goto background_loop;
2244 2245
		}

unknown's avatar
unknown committed
2246 2247 2248 2249
		/* We flush the log once in a second even if no commit
		is issued or the we have specified in my.cnf no flush
		at transaction commit */

unknown's avatar
unknown committed
2250
		srv_main_thread_op_info = "flushing log";
unknown's avatar
unknown committed
2251
		log_buffer_flush_to_disk();
unknown's avatar
unknown committed
2252

unknown's avatar
unknown committed
2253
		srv_main_thread_op_info = "making checkpoint";
unknown's avatar
unknown committed
2254 2255
		log_free_check();

2256
		/* If there were less than 5 i/os during the
2257 2258 2259 2260 2261 2262 2263 2264
		one second sleep, we assume that there is free
		disk i/o capacity available, and it makes sense to
		do an insert buffer merge. */

		n_pend_ios = buf_get_n_pending_ios()
						+ log_sys->n_pending_writes;
		n_ios = log_sys->n_log_ios + buf_pool->n_pages_read
						+ buf_pool->n_pages_written;
2265
		if (n_pend_ios < 3 && (n_ios - n_ios_old < 5)) {
unknown's avatar
unknown committed
2266
			srv_main_thread_op_info = "doing insert buffer merge";
2267 2268
			ibuf_contract_for_n_pages(TRUE, 5);

unknown's avatar
unknown committed
2269
			srv_main_thread_op_info = "flushing log";
unknown's avatar
unknown committed
2270 2271

			log_buffer_flush_to_disk();
2272
		}
unknown's avatar
Merge  
unknown committed
2273

2274 2275 2276 2277 2278 2279 2280 2281
		if (buf_get_modified_ratio_pct() >
				             srv_max_buf_pool_modified_pct) {

			/* Try to keep the number of modified pages in the
			buffer pool under the limit wished by the user */
			
			n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, 100,
							  ut_dulint_max);
unknown's avatar
unknown committed
2282 2283 2284 2285 2286 2287 2288

		        /* If we had to do the flush, it may have taken
			even more than 1 second, and also, there may be more
			to flush. Do not sleep 1 second during the next
			iteration of this loop. */
			     
			skip_sleep = TRUE;
unknown's avatar
Merge  
unknown committed
2289 2290
		}

2291 2292
		if (srv_activity_count == old_activity_count) {

2293 2294
			/* There is no user activity at the moment, go to
			the background loop */
2295 2296 2297 2298 2299

			goto background_loop;
		}
	}

2300 2301
	/* ---- We perform the following code approximately once per
	10 seconds when there is database activity */
2302

unknown's avatar
unknown committed
2303 2304 2305 2306 2307
#ifdef MEM_PERIODIC_CHECK
	/* Check magic numbers of every allocated mem block once in 10
	seconds */
	mem_validate_all_blocks();
#endif	
2308 2309
	/* If there were less than 200 i/os during the 10 second period,
	we assume that there is free disk i/o capacity available, and it
2310
	makes sense to flush 100 pages. */
2311 2312 2313 2314 2315 2316

	n_pend_ios = buf_get_n_pending_ios() + log_sys->n_pending_writes;
	n_ios = log_sys->n_log_ios + buf_pool->n_pages_read
						+ buf_pool->n_pages_written;
	if (n_pend_ios < 3 && (n_ios - n_ios_very_old < 200)) {

unknown's avatar
unknown committed
2317
		srv_main_thread_op_info = "flushing buffer pool pages";
unknown's avatar
unknown committed
2318
		buf_flush_batch(BUF_FLUSH_LIST, 100, ut_dulint_max);
2319

unknown's avatar
unknown committed
2320
		srv_main_thread_op_info = "flushing log";
unknown's avatar
unknown committed
2321
		log_buffer_flush_to_disk();
2322 2323 2324 2325 2326
	}

	/* We run a batch of insert buffer merge every 10 seconds,
	even if the server were active */

unknown's avatar
unknown committed
2327
	srv_main_thread_op_info = "doing insert buffer merge";
2328 2329
	ibuf_contract_for_n_pages(TRUE, 5);

unknown's avatar
unknown committed
2330
	srv_main_thread_op_info = "flushing log";
unknown's avatar
unknown committed
2331
	log_buffer_flush_to_disk();
2332 2333 2334 2335

	/* We run a full purge every 10 seconds, even if the server
	were active */
	
2336 2337
	n_pages_purged = 1;

unknown's avatar
unknown committed
2338 2339
	last_flush_time = time(NULL);

2340
	while (n_pages_purged) {
unknown's avatar
unknown committed
2341

unknown's avatar
Merge  
unknown committed
2342 2343 2344 2345 2346
		if (srv_fast_shutdown && srv_shutdown_state > 0) {

			goto background_loop;
		}

unknown's avatar
unknown committed
2347
		srv_main_thread_op_info = "purging";
2348
		n_pages_purged = trx_purge();
2349

unknown's avatar
unknown committed
2350 2351 2352
		current_time = time(NULL);

		if (difftime(current_time, last_flush_time) > 1) {
unknown's avatar
unknown committed
2353
			srv_main_thread_op_info = "flushing log";
2354

unknown's avatar
unknown committed
2355
		        log_buffer_flush_to_disk();
unknown's avatar
unknown committed
2356 2357
			last_flush_time = current_time;
		}
2358
	}
unknown's avatar
unknown committed
2359
	
unknown's avatar
unknown committed
2360
	srv_main_thread_op_info = "flushing buffer pool pages";
2361

2362
	/* Flush a few oldest pages to make a new checkpoint younger */
unknown's avatar
unknown committed
2363

2364
	if (buf_get_modified_ratio_pct() > 70) {
unknown's avatar
unknown committed
2365

2366 2367 2368
		/* If there are lots of modified pages in the buffer pool
		(> 70 %), we assume we can afford reserving the disk(s) for
		the time it requires to flush 100 pages */
unknown's avatar
unknown committed
2369

unknown's avatar
unknown committed
2370 2371 2372
	        n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, 100,
							ut_dulint_max);
	} else {
2373 2374 2375
	        /* Otherwise, we only flush a small number of pages so that
		we do not unnecessarily use much disk i/o capacity from
		other work */
unknown's avatar
unknown committed
2376

unknown's avatar
unknown committed
2377 2378 2379
	        n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, 10,
							ut_dulint_max);
	}
unknown's avatar
unknown committed
2380

unknown's avatar
unknown committed
2381
	srv_main_thread_op_info = "making checkpoint";
unknown's avatar
unknown committed
2382 2383 2384 2385

	/* Make a new checkpoint about once in 10 seconds */

	log_checkpoint(TRUE, FALSE);
2386

unknown's avatar
unknown committed
2387
	srv_main_thread_op_info = "reserving kernel mutex";
2388

2389
	mutex_enter(&kernel_mutex);
2390 2391 2392 2393
	
	/* ---- When there is database activity, we jump from here back to
	the start of loop */

2394 2395 2396 2397
	if (srv_activity_count != old_activity_count) {
		mutex_exit(&kernel_mutex);
		goto loop;
	}
2398
	
2399 2400
	mutex_exit(&kernel_mutex);

2401 2402 2403 2404 2405
	/* If the database is quiet, we enter the background loop */

	/*****************************************************************/
background_loop:
	/* ---- In this loop we run background operations when the server
2406 2407
	is quiet from user activity. Also in the case of a shutdown, we
	loop here, flushing the buffer pool to the data files. */
2408

2409 2410 2411
	/* The server has been quiet for a while: start running background
	operations */
		
unknown's avatar
unknown committed
2412
	srv_main_thread_op_info = "doing background drop tables";
2413 2414 2415

	n_tables_to_drop = row_drop_tables_for_mysql_in_background();

2416 2417 2418 2419 2420 2421 2422 2423 2424
	if (n_tables_to_drop > 0) {
	        /* Do not monopolize the CPU even if there are tables waiting
		in the background drop queue. (It is essentially a bug if
		MySQL tries to drop a table while there are still open handles
		to it and we had to put it to the background drop queue.) */

		os_thread_sleep(100000);
	}
 
unknown's avatar
unknown committed
2425
	srv_main_thread_op_info = "purging";
2426

unknown's avatar
unknown committed
2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438
	/* Run a full purge */
	
	n_pages_purged = 1;

	last_flush_time = time(NULL);

	while (n_pages_purged) {
		if (srv_fast_shutdown && srv_shutdown_state > 0) {

			break;
		}

unknown's avatar
unknown committed
2439
		srv_main_thread_op_info = "purging";
unknown's avatar
unknown committed
2440 2441 2442 2443 2444
		n_pages_purged = trx_purge();

		current_time = time(NULL);

		if (difftime(current_time, last_flush_time) > 1) {
unknown's avatar
unknown committed
2445
			srv_main_thread_op_info = "flushing log";
unknown's avatar
unknown committed
2446 2447 2448 2449

		        log_buffer_flush_to_disk();
			last_flush_time = current_time;
		}
unknown's avatar
unknown committed
2450
	}
2451

unknown's avatar
unknown committed
2452
	srv_main_thread_op_info = "reserving kernel mutex";
2453

2454 2455 2456 2457 2458 2459 2460
	mutex_enter(&kernel_mutex);
	if (srv_activity_count != old_activity_count) {
		mutex_exit(&kernel_mutex);
		goto loop;
	}
	mutex_exit(&kernel_mutex);

unknown's avatar
unknown committed
2461
	srv_main_thread_op_info = "doing insert buffer merge";
unknown's avatar
unknown committed
2462 2463 2464 2465 2466 2467

	if (srv_fast_shutdown && srv_shutdown_state > 0) {
	        n_bytes_merged = 0;
	} else {
	        n_bytes_merged = ibuf_contract_for_n_pages(TRUE, 20);
	}
2468

unknown's avatar
unknown committed
2469
	srv_main_thread_op_info = "reserving kernel mutex";
2470

2471 2472 2473 2474 2475 2476 2477
	mutex_enter(&kernel_mutex);
	if (srv_activity_count != old_activity_count) {
		mutex_exit(&kernel_mutex);
		goto loop;
	}
	mutex_exit(&kernel_mutex);
	
2478
flush_loop:
unknown's avatar
unknown committed
2479
	srv_main_thread_op_info = "flushing buffer pool pages";
2480

2481
	if (srv_fast_shutdown < 2) {
2482 2483 2484
		n_pages_flushed =
			buf_flush_batch(BUF_FLUSH_LIST, 100, ut_dulint_max);
	} else {
2485
		/* In the fastest shutdown we do not flush the buffer pool
2486 2487 2488 2489
		to data files: we set n_pages_flushed to 0 artificially. */

		n_pages_flushed = 0;
	}
2490

unknown's avatar
unknown committed
2491
	srv_main_thread_op_info = "reserving kernel mutex";
2492

2493 2494 2495 2496 2497 2498 2499
	mutex_enter(&kernel_mutex);
	if (srv_activity_count != old_activity_count) {
		mutex_exit(&kernel_mutex);
		goto loop;
	}
	mutex_exit(&kernel_mutex);
	
unknown's avatar
unknown committed
2500
	srv_main_thread_op_info = "waiting for buffer pool flush to end";
2501 2502
	buf_flush_wait_batch_end(BUF_FLUSH_LIST);

unknown's avatar
unknown committed
2503
	srv_main_thread_op_info = "flushing log";
unknown's avatar
unknown committed
2504 2505 2506

	log_buffer_flush_to_disk();

unknown's avatar
unknown committed
2507
	srv_main_thread_op_info = "making checkpoint";
2508

2509 2510
	log_checkpoint(TRUE, FALSE);

2511 2512 2513 2514 2515 2516 2517 2518
	if (buf_get_modified_ratio_pct() > srv_max_buf_pool_modified_pct) {

		/* Try to keep the number of modified pages in the
		buffer pool under the limit wished by the user */
			
		goto flush_loop;
	}

unknown's avatar
unknown committed
2519
	srv_main_thread_op_info = "reserving kernel mutex";
2520

2521 2522 2523 2524 2525 2526
	mutex_enter(&kernel_mutex);
	if (srv_activity_count != old_activity_count) {
		mutex_exit(&kernel_mutex);
		goto loop;
	}
	mutex_exit(&kernel_mutex);
unknown's avatar
unknown committed
2527
/*
unknown's avatar
unknown committed
2528
	srv_main_thread_op_info = "archiving log (if log archive is on)";
2529 2530
	
	log_archive_do(FALSE, &n_bytes_archived);
unknown's avatar
unknown committed
2531 2532
*/
	n_bytes_archived = 0;
2533

2534 2535
	/* Keep looping in the background loop if still work to do */

unknown's avatar
Merge  
unknown committed
2536
	if (srv_fast_shutdown && srv_shutdown_state > 0) {
unknown's avatar
unknown committed
2537 2538
		if (n_tables_to_drop + n_pages_flushed
				+ n_bytes_archived != 0) {
unknown's avatar
Merge  
unknown committed
2539

2540 2541
			/* If we are doing a fast shutdown (= the default)
			we do not do purge or insert buffer merge. But we
2542 2543 2544 2545
			flush the buffer pool completely to disk.
			In a 'very fast' shutdown we do not flush the buffer
			pool to data files: we have set n_pages_flushed to
			0 artificially. */
2546

unknown's avatar
Merge  
unknown committed
2547 2548
			goto background_loop;
		}
unknown's avatar
unknown committed
2549
	} else if (n_tables_to_drop +
2550
		   n_pages_purged + n_bytes_merged + n_pages_flushed
2551
						+ n_bytes_archived != 0) {
2552 2553 2554
		/* In a 'slow' shutdown we run purge and the insert buffer
		merge to completion */

2555 2556 2557 2558 2559 2560
		goto background_loop;
	}
		
	/* There is no work for background operations either: suspend
	master thread to wait for more server activity */
	
unknown's avatar
Merge  
unknown committed
2561
suspend_thread:
unknown's avatar
unknown committed
2562
	srv_main_thread_op_info = "suspending";
2563

2564 2565
	mutex_enter(&kernel_mutex);

unknown's avatar
unknown committed
2566 2567 2568 2569 2570 2571
	if (row_get_background_drop_list_len_low() > 0) {
		mutex_exit(&kernel_mutex);

		goto loop;
	}

2572 2573 2574 2575
	event = srv_suspend_thread();

	mutex_exit(&kernel_mutex);

unknown's avatar
unknown committed
2576
	srv_main_thread_op_info = "waiting for server activity";
2577

2578 2579
	os_event_wait(event);

unknown's avatar
unknown committed
2580 2581 2582 2583 2584 2585 2586
	if (srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS) {
	        /* This is only extra safety, the thread should exit
		already when the event wait ends */

	        os_thread_exit(NULL);
	}

2587 2588 2589
	/* When there is user activity, InnoDB will set the event and the main
	thread goes back to loop: */

2590 2591
	goto loop;

2592 2593 2594 2595 2596 2597 2598
	/* We count the number of threads in os_thread_exit(). A created
	thread should always use that to exit and not use return() to exit.
	The thread actually never comes here because it is exited in an
	os_event_wait(). */
	
	os_thread_exit(NULL);

2599
#ifndef __WIN__
2600
        return(NULL);				/* Not reached */
2601
#else
2602
	return(0);
2603
#endif
2604
}
2605
#endif /* !UNIV_HOTBACKUP */