srv0srv.c 61.9 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 37 38 39 40 41 42 43 44 45 46
#include "sync0sync.h"
#include "sync0ipm.h"
#include "thr0loc.h"
#include "que0que.h"
#include "srv0que.h"
#include "log0recv.h"
#include "pars0pars.h"
#include "usr0sess.h"
#include "lock0lock.h"
#include "trx0purge.h"
#include "ibuf0ibuf.h"
#include "buf0flu.h"
#include "btr0sea.h"
47
#include "dict0load.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;

unknown's avatar
unknown committed
55 56 57
/* Buffer which can be used in printing fatal error messages */
char	srv_fatal_errbuf[5000];

58 59 60 61
/* The following counter is incremented whenever there is some user activity
in the server */
ulint	srv_activity_count	= 0;

unknown's avatar
Merge  
unknown committed
62 63 64
ibool	srv_lock_timeout_and_monitor_active = FALSE;
ibool	srv_error_monitor_active = FALSE;

unknown's avatar
unknown committed
65
char*	srv_main_thread_op_info = (char*) "";
66

67 68 69 70 71 72 73 74 75 76 77 78
/* 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;
char*	srv_arch_dir 	= NULL;

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
79 80 81 82 83 84 85
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 */
86 87
ulint*  srv_data_file_is_raw_partition = NULL;

88 89 90 91 92
/* 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;

93 94 95 96 97
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 */ 
unknown's avatar
unknown committed
98
ibool	srv_log_archive_on	= TRUE;
99
ulint	srv_log_buffer_size	= ULINT_MAX;	/* size in database pages */ 
unknown's avatar
unknown committed
100
ulint	srv_flush_log_at_trx_commit = 1;
101

unknown's avatar
unknown committed
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138
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
};
139 140 141
		
ulint	srv_pool_size		= ULINT_MAX;	/* size in database pages;
						MySQL originally sets this
142
						value in bytes */ 
143 144 145 146 147 148 149 150 151
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;

ibool	srv_archive_recovery	= 0;
dulint	srv_archive_recovery_limit_lsn;

ulint	srv_lock_wait_timeout	= 1024 * 1024 * 1024;
unknown's avatar
unknown committed
152

unknown's avatar
unknown committed
153 154 155
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;
156

157 158 159 160 161
/* 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. */

162
ulint	srv_max_buf_pool_modified_pct	= 90;
163

164
/* If the following is != 0 we do not allow inserts etc. This protects
unknown's avatar
Merge  
unknown committed
165
the user from forgetting the innodb_force_recovery keyword to my.cnf */
166 167

ulint	srv_force_recovery	= 0;
unknown's avatar
Merge  
unknown committed
168
/*-----------------------*/
169 170 171 172 173 174
/* 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
175 176 177 178 179 180 181
/* 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. */

unknown's avatar
unknown committed
182
ulint	srv_thread_concurrency	= 8;
unknown's avatar
Merge  
unknown committed
183 184 185

os_fast_mutex_t	srv_conc_mutex;		/* this mutex protects srv_conc data
					structures */
unknown's avatar
unknown committed
186 187 188 189 190 191 192
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
193 194 195
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
196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213

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 */
214
srv_conc_slot_t*			srv_conc_slots;	/* array of wait
unknown's avatar
Merge  
unknown committed
215
							slots */
unknown's avatar
unknown committed
216 217 218 219 220

/* Number of times a thread is allowed to enter InnoDB within the same
SQL query after it has once got the ticket at srv_conc_enter_innodb */
#define SRV_FREE_TICKETS_TO_ENTER	500

unknown's avatar
Merge  
unknown committed
221 222 223 224 225
/*-----------------------*/
/* If the following is set TRUE then we do not run purge and insert buffer
merge to completion before shutdown */

ibool	srv_fast_shutdown	= FALSE;
226

227 228
ibool	srv_use_doublewrite_buf	= TRUE;

unknown's avatar
unknown committed
229 230
ibool   srv_set_thread_priorities = TRUE;
int     srv_query_thread_priority = 0;
231 232 233 234 235 236 237 238 239 240 241
/*-------------------------------------------*/
ulint	srv_n_spin_wait_rounds	= 20;
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
242 243 244 245 246 247 248 249
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;
250

251 252 253 254 255
/*
  Set the following to 0 if you want InnoDB to write messages on
  stderr on startup/shutdown
*/
ibool	srv_print_verbose_log		= TRUE;
256
ibool	srv_print_innodb_monitor	= FALSE;
257 258
ibool   srv_print_innodb_lock_monitor   = FALSE;
ibool   srv_print_innodb_tablespace_monitor = FALSE;
259
ibool   srv_print_innodb_table_monitor = FALSE;
260

261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284
/* 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;

285 286 287 288
/* Array of English strings describing the current state of an
i/o handler thread */

char* srv_io_thread_op_info[SRV_MAX_N_IO_THREADS];
unknown's avatar
unknown committed
289
char* srv_io_thread_function[SRV_MAX_N_IO_THREADS];
290

unknown's avatar
unknown committed
291 292
time_t	srv_last_monitor_time;

unknown's avatar
unknown committed
293 294
mutex_t srv_innodb_monitor_mutex;

unknown's avatar
unknown committed
295 296 297
ulint	srv_main_thread_process_no	= 0;
ulint	srv_main_thread_id		= 0;

298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608
/*
	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];


/*************************************************************************
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;

609
#ifdef UNIV_SYNC_DEBUG
610
	ut_ad(mutex_own(&kernel_mutex));
611
#endif /* UNIV_SYNC_DEBUG */
612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660
	
	slot_no = thr_local_get_slot_no(os_thread_get_curr_id());

	if (srv_print_thread_releases) {
	
		printf("Suspending thread %lu to slot %lu meter %lu\n",
		os_thread_get_curr_id(), slot_no, srv_meter[SRV_RECOVERY]);
	}

	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);
661
#ifdef UNIV_SYNC_DEBUG
662
	ut_ad(mutex_own(&kernel_mutex));
663
#endif /* UNIV_SYNC_DEBUG */
664 665 666 667 668
	
	for (i = 0; i < OS_THREAD_MAX_N; i++) {
	
		slot = srv_table_get_nth_slot(i);

unknown's avatar
unknown committed
669
		if (slot->in_use && slot->type == type && slot->suspended) {
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
			
			slot->suspended = FALSE;

			srv_n_threads_active[type]++;

			os_event_set(slot->event);

			if (srv_print_thread_releases) {
				printf(
		"Releasing thread %lu type %lu from slot %lu meter %lu\n",
				slot->id, type, i, srv_meter[SRV_RECOVERY]);
			}

			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. */
unknown's avatar
unknown committed
724
static
725 726 727 728
void
srv_init(void)
/*==========*/
{
unknown's avatar
Merge  
unknown committed
729 730 731
	srv_conc_slot_t* 	conc_slot;
	srv_slot_t*		slot;
	ulint			i;
732 733 734 735 736 737

	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
738 739 740

	mutex_create(&srv_innodb_monitor_mutex);
	mutex_set_level(&srv_innodb_monitor_mutex, SYNC_NO_ORDER_CHECK);
741 742 743 744 745 746
	
	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
747
                slot->type=0;	/* Avoid purify errors */
748 749 750 751 752 753 754 755 756
		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
757
		slot->type = 0;
758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778
		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
779 780 781 782 783 784

	/* Init the server concurrency restriction data structures */

	os_fast_mutex_init(&srv_conc_mutex);
	
	UT_LIST_INIT(srv_conc_queue);
785 786

	srv_conc_slots = mem_alloc(OS_THREAD_MAX_N * sizeof(srv_conc_slot_t));
unknown's avatar
Merge  
unknown committed
787 788 789 790 791 792 793
	
	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);
	}
794
}	
unknown's avatar
unknown committed
795

796
/*************************************************************************
unknown's avatar
unknown committed
797
Frees the OS fast mutex created in srv_init(). */
unknown's avatar
unknown committed
798

799
void
unknown's avatar
unknown committed
800 801
srv_free(void)
/*==========*/
802
{
unknown's avatar
unknown committed
803
	os_fast_mutex_free(&srv_conc_mutex);
804 805
}

unknown's avatar
unknown committed
806
/*************************************************************************
unknown's avatar
unknown committed
807
Initializes the synchronization primitives, memory system, and the thread
unknown's avatar
unknown committed
808 809 810
local storage. */

void
unknown's avatar
unknown committed
811
srv_general_init(void)
unknown's avatar
unknown committed
812 813
/*==================*/
{
unknown's avatar
unknown committed
814 815 816 817
	os_sync_init();
	sync_init();
	mem_init(srv_mem_pool_size);
	thr_local_init();
unknown's avatar
unknown committed
818 819
}

unknown's avatar
unknown committed
820 821
/*======================= InnoDB Server FIFO queue =======================*/

unknown's avatar
unknown committed
822

unknown's avatar
Merge  
unknown committed
823 824 825 826 827 828 829 830 831 832
/*************************************************************************
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
833
	ibool			has_slept = FALSE;
834
	srv_conc_slot_t*	slot	  = NULL;
unknown's avatar
Merge  
unknown committed
835
	ulint			i;
unknown's avatar
unknown committed
836
	char                    err_buf[1000];
unknown's avatar
Merge  
unknown committed
837

unknown's avatar
unknown committed
838 839 840 841 842 843 844 845 846 847 848 849 850 851
	if (srv_thread_concurrency >= 500) {
		/* Disable the concurrency check */
	
		return;
	}

	/* 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
unknown committed
852

unknown's avatar
Merge  
unknown committed
853
	os_fast_mutex_lock(&srv_conc_mutex);
unknown's avatar
unknown committed
854
retry:
unknown's avatar
unknown committed
855 856 857 858 859 860 861 862
	if (trx->declared_to_be_inside_innodb) {
	        ut_print_timestamp(stderr);

	        trx_print(err_buf, trx);

	        fprintf(stderr,
"  InnoDB: Error: trying to declare trx to enter InnoDB, but\n"
"InnoDB: it already is declared.\n%s\n", err_buf);
unknown's avatar
unknown committed
863 864 865
		os_fast_mutex_unlock(&srv_conc_mutex);

		return;
unknown's avatar
unknown committed
866 867
	}

unknown's avatar
unknown committed
868
	if (srv_conc_n_threads < (lint)srv_thread_concurrency) {
unknown's avatar
Merge  
unknown committed
869

unknown's avatar
unknown committed
870 871 872 873
		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
874 875 876 877
		os_fast_mutex_unlock(&srv_conc_mutex);

		return;
	}
unknown's avatar
unknown committed
878

unknown's avatar
unknown committed
879 880 881 882 883 884 885 886 887 888 889 890 891
	/* If the transaction is not holding resources, let it sleep for 50
	milliseconds, and try again then */
 
	if (!has_slept && !trx->has_search_latch
	    && NULL == UT_LIST_GET_FIRST(trx->trx_locks)) {

	        has_slept = TRUE; /* We let is sleep only once to avoid
				  starvation */

		srv_conc_n_waiting_threads++;

		os_fast_mutex_unlock(&srv_conc_mutex);

892 893
		trx->op_info = (char*)"sleeping before joining InnoDB queue";

unknown's avatar
unknown committed
894 895
		os_thread_sleep(50000);

896 897
		trx->op_info = (char*)"";

unknown's avatar
unknown committed
898 899 900 901 902 903 904
		os_fast_mutex_lock(&srv_conc_mutex);

		srv_conc_n_waiting_threads--;

		goto retry;
	}   

unknown's avatar
unknown committed
905
	/* Too many threads inside: put the current thread to a queue */
unknown's avatar
Merge  
unknown committed
906 907 908 909 910 911 912 913 914 915 916 917 918 919

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

		if (!slot->reserved) {
			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
920 921
		trx->declared_to_be_inside_innodb = TRUE;
		trx->n_tickets_to_enter_innodb = 0;
unknown's avatar
Merge  
unknown committed
922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940

		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
941 942
	srv_conc_n_waiting_threads++;

unknown's avatar
Merge  
unknown committed
943 944 945 946 947
	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
948
	trx->op_info = (char*)"waiting in InnoDB queue";
unknown's avatar
unknown committed
949

unknown's avatar
Merge  
unknown committed
950 951
	os_event_wait(slot->event);

unknown's avatar
unknown committed
952
	trx->op_info = (char*)"";
unknown's avatar
unknown committed
953

unknown's avatar
Merge  
unknown committed
954 955
	os_fast_mutex_lock(&srv_conc_mutex);

unknown's avatar
unknown committed
956 957
	srv_conc_n_waiting_threads--;

unknown's avatar
Merge  
unknown committed
958 959 960 961 962 963 964
	/* 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
965 966 967
	trx->declared_to_be_inside_innodb = TRUE;
	trx->n_tickets_to_enter_innodb = SRV_FREE_TICKETS_TO_ENTER;

unknown's avatar
Merge  
unknown committed
968 969 970 971 972 973 974 975
	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
976 977 978 979
srv_conc_force_enter_innodb(
/*========================*/
	trx_t*	trx)	/* in: transaction object associated with the
			thread */
unknown's avatar
Merge  
unknown committed
980
{
unknown's avatar
unknown committed
981 982 983 984 985
	if (srv_thread_concurrency >= 500) {
	
		return;
	}

unknown's avatar
Merge  
unknown committed
986 987 988
	os_fast_mutex_lock(&srv_conc_mutex);

	srv_conc_n_threads++;
unknown's avatar
unknown committed
989 990
	trx->declared_to_be_inside_innodb = TRUE;
	trx->n_tickets_to_enter_innodb = 0;
unknown's avatar
Merge  
unknown committed
991 992 993 994 995

	os_fast_mutex_unlock(&srv_conc_mutex);
}

/*************************************************************************
unknown's avatar
unknown committed
996 997
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
998 999

void
unknown's avatar
unknown committed
1000 1001 1002 1003
srv_conc_force_exit_innodb(
/*=======================*/
	trx_t*	trx)	/* in: transaction object associated with the
			thread */
unknown's avatar
Merge  
unknown committed
1004 1005 1006
{
	srv_conc_slot_t*	slot	= NULL;

unknown's avatar
unknown committed
1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
	if (srv_thread_concurrency >= 500) {
	
		return;
	}

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

unknown's avatar
Merge  
unknown committed
1017 1018 1019
	os_fast_mutex_lock(&srv_conc_mutex);

	srv_conc_n_threads--;
unknown's avatar
unknown committed
1020 1021
	trx->declared_to_be_inside_innodb = FALSE;
	trx->n_tickets_to_enter_innodb = 0;
unknown's avatar
Merge  
unknown committed
1022

unknown's avatar
unknown committed
1023
	if (srv_conc_n_threads < (lint)srv_thread_concurrency) {
unknown's avatar
Merge  
unknown committed
1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049
		/* 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
1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079
/*************************************************************************
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 (srv_thread_concurrency >= 500) {
	
		return;
	}

	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);
}

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

1080
/*************************************************************************
1081
Normalizes init parameter values to use units we use inside InnoDB. */
1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097
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
1098 1099 1100
	srv_last_file_size_max = srv_last_file_size_max
					* ((1024 * 1024) / UNIV_PAGE_SIZE);
		
1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112
	srv_log_file_size = srv_log_file_size / UNIV_PAGE_SIZE;

	srv_log_buffer_size = srv_log_buffer_size / UNIV_PAGE_SIZE;

	srv_pool_size = srv_pool_size / UNIV_PAGE_SIZE;
	
	srv_lock_table_size = 20 * srv_pool_size;

	return(DB_SUCCESS);
}

/*************************************************************************
1113
Boots the InnoDB server. */
1114 1115 1116 1117 1118 1119 1120 1121 1122

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

	/* Transform the init parameter values given by MySQL to
1123
	use units we use inside InnoDB: */
1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144
	
	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);
}

/*************************************************************************
Reserves a slot in the thread table for the current MySQL OS thread.
unknown's avatar
unknown committed
1145
NOTE! The kernel mutex has to be reserved by the caller! */
1146 1147 1148 1149 1150 1151 1152 1153 1154
static
srv_slot_t*
srv_table_reserve_slot_for_mysql(void)
/*==================================*/
			/* out: reserved slot */
{
	srv_slot_t*	slot;
	ulint		i;

1155
#ifdef UNIV_SYNC_DEBUG
unknown's avatar
unknown committed
1156
	ut_ad(mutex_own(&kernel_mutex));
1157
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1158

1159 1160 1161 1162 1163
	i = 0;
	slot = srv_mysql_table + i;

	while (slot->in_use) {
		i++;
unknown's avatar
unknown committed
1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180

		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"
"InnoDB: on Linux. But first we print a list of waiting threads.\n", i);

			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",
unknown's avatar
unknown committed
1181
				  i, os_thread_pf(slot->id),
unknown's avatar
unknown committed
1182 1183 1184 1185 1186
				  slot->type, slot->in_use,
				  slot->suspended,
			  (ulint)difftime(ut_time(), slot->suspend_time));
			}

1187
		        ut_error;
unknown's avatar
unknown committed
1188
		}
1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202
		
		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);
}

/*******************************************************************
unknown's avatar
unknown committed
1203 1204 1205 1206 1207
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. */
1208

unknown's avatar
unknown committed
1209
void
1210 1211
srv_suspend_mysql_thread(
/*=====================*/
unknown's avatar
unknown committed
1212 1213
	que_thr_t*	thr)	/* in: query thread associated with the MySQL
				OS thread */
1214 1215 1216 1217
{
	srv_slot_t*	slot;
	os_event_t	event;
	double		wait_time;
unknown's avatar
unknown committed
1218
	trx_t*		trx;
unknown's avatar
unknown committed
1219 1220
	ibool		had_dict_lock			= FALSE;
	ibool		was_declared_inside_innodb	= FALSE;
unknown's avatar
unknown committed
1221
	
1222
#ifdef UNIV_SYNC_DEBUG
1223
	ut_ad(!mutex_own(&kernel_mutex));
1224
#endif /* UNIV_SYNC_DEBUG */
1225

unknown's avatar
unknown committed
1226 1227
	trx = thr_get_trx(thr);
	
1228 1229 1230 1231
	os_event_set(srv_lock_timeout_thread_event);

	mutex_enter(&kernel_mutex);

unknown's avatar
unknown committed
1232 1233
	trx->error_state = DB_SUCCESS;

1234 1235
	if (thr->state == QUE_THR_RUNNING) {

unknown's avatar
unknown committed
1236 1237 1238 1239 1240 1241 1242 1243 1244 1245
		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;
		}
1246 1247 1248

		mutex_exit(&kernel_mutex);

unknown's avatar
unknown committed
1249
		return;
1250 1251
	}
	
unknown's avatar
unknown committed
1252 1253
	ut_ad(thr->is_active == FALSE);

1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269
	slot = srv_table_reserve_slot_for_mysql();

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

	os_event_reset(event);	

	slot->suspend_time = ut_time();

	/* 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
1270 1271 1272
	if (trx->declared_to_be_inside_innodb) {

		was_declared_inside_innodb = TRUE;
unknown's avatar
Merge  
unknown committed
1273
	
unknown's avatar
unknown committed
1274 1275 1276 1277 1278 1279
		/* 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 */
	
		srv_conc_force_exit_innodb(trx);
	}
unknown's avatar
Merge  
unknown committed
1280

unknown's avatar
unknown committed
1281
	/* Release possible foreign key check latch */
unknown's avatar
unknown committed
1282 1283 1284
	if (trx->dict_operation_lock_mode == RW_S_LATCH) {

		had_dict_lock = TRUE;
unknown's avatar
unknown committed
1285

unknown's avatar
unknown committed
1286
		row_mysql_unfreeze_data_dictionary(trx);
unknown's avatar
unknown committed
1287 1288
	}

unknown's avatar
unknown committed
1289 1290
	ut_a(trx->dict_operation_lock_mode == 0);

1291 1292 1293 1294
	/* Wait for the release */
	
	os_event_wait(event);

unknown's avatar
unknown committed
1295
	if (had_dict_lock) {
unknown's avatar
unknown committed
1296

unknown's avatar
unknown committed
1297
		row_mysql_freeze_data_dictionary(trx);
unknown's avatar
unknown committed
1298 1299
	}

unknown's avatar
unknown committed
1300 1301 1302
	if (was_declared_inside_innodb) {

		/* Return back inside InnoDB */
unknown's avatar
Merge  
unknown committed
1303
	
unknown's avatar
unknown committed
1304 1305
		srv_conc_force_enter_innodb(trx);
	}
unknown's avatar
Merge  
unknown committed
1306

1307 1308 1309 1310 1311 1312 1313 1314
	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
1315 1316 1317 1318 1319 1320
	if (trx->was_chosen_as_deadlock_victim) {

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

1321 1322 1323 1324 1325
	mutex_exit(&kernel_mutex);

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

unknown's avatar
unknown committed
1326 1327
	    	trx->error_state = DB_LOCK_WAIT_TIMEOUT;
	}
1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342
}

/************************************************************************
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  */
{
	srv_slot_t*	slot;
	ulint		i;
	
1343
#ifdef UNIV_SYNC_DEBUG
1344
	ut_ad(mutex_own(&kernel_mutex));
1345
#endif /* UNIV_SYNC_DEBUG */
1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362

	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 */
}

unknown's avatar
unknown committed
1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390
/**********************************************************************
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
1391 1392
/**********************************************************************
Sprintfs to a buffer the output of the InnoDB Monitor. */
1393

unknown's avatar
unknown committed
1394 1395 1396 1397 1398
void
srv_sprintf_innodb_monitor(
/*=======================*/
	char*	buf,	/* in/out: buffer which must be at least 4 kB */
	ulint	len)	/* in: length of the buffer */
1399
{
unknown's avatar
unknown committed
1400 1401 1402
	char*	buf_end	= buf + len - 2000;
	double	time_elapsed;
	time_t	current_time;
1403
	ulint	n_reserved;
unknown's avatar
Merge  
unknown committed
1404

unknown's avatar
unknown committed
1405
	mutex_enter(&srv_innodb_monitor_mutex);
1406

unknown's avatar
unknown committed
1407
	current_time = time(NULL);
1408

unknown's avatar
unknown committed
1409 1410 1411
	/* 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 */
1412
	
unknown's avatar
unknown committed
1413 1414
	time_elapsed = difftime(current_time, srv_last_monitor_time)
			+ 0.001;
1415

unknown's avatar
unknown committed
1416
	srv_last_monitor_time = time(NULL);
1417

unknown's avatar
unknown committed
1418
	ut_a(len >= 4096);	
1419

unknown's avatar
unknown committed
1420
	buf += sprintf(buf, "\n=====================================\n");
unknown's avatar
Merge  
unknown committed
1421

unknown's avatar
unknown committed
1422 1423
	ut_sprintf_timestamp(buf);
	buf = buf + strlen(buf);
1424
	ut_a(buf < buf_end + 1500);
1425
	
unknown's avatar
unknown committed
1426
	buf += sprintf(buf, " INNODB MONITOR OUTPUT\n"
1427
	       	       "=====================================\n");
unknown's avatar
unknown committed
1428 1429

	buf += sprintf(buf,
unknown's avatar
unknown committed
1430
"Per second averages calculated from the last %lu seconds\n",
unknown's avatar
unknown committed
1431 1432 1433
					(ulint)time_elapsed);
	       	       
	buf += sprintf(buf, "----------\n"
1434 1435
		       "SEMAPHORES\n"
		       "----------\n");
unknown's avatar
unknown committed
1436 1437 1438
	sync_print(buf, buf_end);

	buf = buf + strlen(buf);
1439
	ut_a(buf < buf_end + 1500);
unknown's avatar
unknown committed
1440

unknown's avatar
unknown committed
1441 1442 1443 1444 1445 1446 1447
	/* 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);

unknown's avatar
unknown committed
1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458
	if (*dict_foreign_err_buf != '\0') {
		buf += sprintf(buf,
			"------------------------\n"
		       	"LATEST FOREIGN KEY ERROR\n"
		       	"------------------------\n");

		if (buf_end - buf > 6000) {
			buf+= sprintf(buf, "%.4000s", dict_foreign_err_buf);
		}
	}	

unknown's avatar
unknown committed
1459
	mutex_exit(&dict_foreign_err_mutex);
1460 1461 1462

	ut_a(buf < buf_end + 1500);

unknown's avatar
unknown committed
1463 1464
	lock_print_info(buf, buf_end);
	buf = buf + strlen(buf);
unknown's avatar
unknown committed
1465
	
unknown's avatar
unknown committed
1466
	buf += sprintf(buf, "--------\n"
1467 1468
		       "FILE I/O\n"
		       "--------\n");
unknown's avatar
unknown committed
1469 1470
	os_aio_print(buf, buf_end);
	buf = buf + strlen(buf);
1471
	ut_a(buf < buf_end + 1500);
unknown's avatar
unknown committed
1472 1473

	buf += sprintf(buf, "-------------------------------------\n"
unknown's avatar
unknown committed
1474 1475
		       "INSERT BUFFER AND ADAPTIVE HASH INDEX\n"
		       "-------------------------------------\n");
unknown's avatar
unknown committed
1476 1477
	ibuf_print(buf, buf_end);
	buf = buf + strlen(buf);
1478
	ut_a(buf < buf_end + 1500);
unknown's avatar
unknown committed
1479 1480 1481

	ha_print_info(buf, buf_end, btr_search_sys->hash_index);
	buf = buf + strlen(buf);
1482
	ut_a(buf < buf_end + 1500);
unknown's avatar
unknown committed
1483 1484 1485 1486 1487 1488 1489

	buf += sprintf(buf,
		"%.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
1490 1491
	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
1492 1493

	buf += sprintf(buf,"---\n"
1494 1495
		       "LOG\n"
		       "---\n");
unknown's avatar
unknown committed
1496 1497
	log_print(buf, buf_end);
	buf = buf + strlen(buf);
1498
	ut_a(buf < buf_end + 1500);
unknown's avatar
unknown committed
1499 1500
	
	buf += sprintf(buf, "----------------------\n"
1501 1502
		       "BUFFER POOL AND MEMORY\n"
		       "----------------------\n");
unknown's avatar
unknown committed
1503
	buf += sprintf(buf,
1504 1505 1506
	"Total memory allocated %lu; in additional pool allocated %lu\n",
				ut_total_allocated_memory,
				mem_pool_get_reserved(mem_comm_pool));
unknown's avatar
unknown committed
1507 1508
	buf_print_io(buf, buf_end);
	buf = buf + strlen(buf);
1509
	ut_a(buf < buf_end + 1500);
unknown's avatar
unknown committed
1510 1511

	buf += sprintf(buf, "--------------\n"
1512 1513
		       "ROW OPERATIONS\n"
		       "--------------\n");
unknown's avatar
unknown committed
1514
	buf += sprintf(buf,
1515
	"%ld queries inside InnoDB, %lu queries in queue\n",
unknown's avatar
unknown committed
1516
			srv_conc_n_threads, srv_conc_n_waiting_threads);
1517 1518 1519 1520 1521 1522 1523 1524

	n_reserved = fil_space_get_n_reserved_extents(0);
	if (n_reserved > 0) {
	        buf += sprintf(buf,
	"%lu tablespace extents now reserved for B-tree split operations\n",
						    n_reserved);
	}

unknown's avatar
unknown committed
1525 1526
#ifdef UNIV_LINUX
	buf += sprintf(buf,
unknown's avatar
unknown committed
1527
	"Main thread process no. %lu, id %lu, state: %.29s\n",
unknown's avatar
unknown committed
1528
			srv_main_thread_process_no,
unknown's avatar
unknown committed
1529
			srv_main_thread_id,
unknown's avatar
unknown committed
1530 1531 1532
			srv_main_thread_op_info);
#else
	buf += sprintf(buf,
unknown's avatar
unknown committed
1533
	"Main thread id %lu, state: %.29s\n",
unknown's avatar
unknown committed
1534
			srv_main_thread_id,
unknown's avatar
unknown committed
1535
			srv_main_thread_op_info);
unknown's avatar
unknown committed
1536
#endif
unknown's avatar
unknown committed
1537
	buf += sprintf(buf,
1538 1539 1540 1541 1542
	"Number of rows inserted %lu, updated %lu, deleted %lu, read %lu\n",
			srv_n_rows_inserted, 
			srv_n_rows_updated, 
			srv_n_rows_deleted, 
			srv_n_rows_read);
unknown's avatar
unknown committed
1543
	buf += sprintf(buf,
1544 1545 1546 1547 1548 1549 1550 1551 1552 1553
	"%.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
1554 1555 1556 1557
	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;
1558

unknown's avatar
unknown committed
1559
	buf += sprintf(buf, "----------------------------\n"
1560 1561
		       "END OF INNODB MONITOR OUTPUT\n"
		       "============================\n");
1562 1563
	ut_a(buf < buf_end + 1900);

unknown's avatar
unknown committed
1564
	mutex_exit(&srv_innodb_monitor_mutex);
unknown's avatar
unknown committed
1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591
}

/*************************************************************************
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 */
	void*	arg)	/* in: a dummy parameter required by
			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;
	char*		buf;
	ulint		i;

1592
#ifdef UNIV_DEBUG_THREAD_CREATION
unknown's avatar
unknown committed
1593 1594
	printf("Lock timeout thread starts, id %lu\n",
			     os_thread_pf(os_thread_get_curr_id()));
1595
#endif
unknown's avatar
unknown committed
1596 1597 1598 1599 1600 1601 1602 1603 1604 1605
	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
1606
	os_thread_sleep(1000000);
unknown's avatar
unknown committed
1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622

	/* 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) {

unknown's avatar
unknown committed
1623
	        buf = mem_alloc(100000);
unknown's avatar
unknown committed
1624

1625 1626 1627
	        srv_sprintf_innodb_monitor(buf, 90000);

		ut_a(strlen(buf) < 99000);
unknown's avatar
unknown committed
1628 1629 1630 1631

	    	printf("%s", buf);

	    	mem_free(buf);
1632 1633
            }

unknown's avatar
unknown committed
1634 1635 1636 1637 1638
            if (srv_print_innodb_tablespace_monitor
                && difftime(current_time, last_table_monitor_time) > 60) {

		last_table_monitor_time = time(NULL);	

1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654
		printf("================================================\n");

		ut_print_timestamp(stdout);

		printf(" INNODB TABLESPACE MONITOR OUTPUT\n"
		       "================================================\n");
	       
		fsp_print(0);
		fprintf(stderr, "Validating tablespace\n");
		fsp_validate(0);
		fprintf(stderr, "Validation ok\n");
		printf("---------------------------------------\n"
	       		"END OF INNODB TABLESPACE MONITOR OUTPUT\n"
	       		"=======================================\n");
	    }

unknown's avatar
unknown committed
1655 1656 1657 1658
	    if (srv_print_innodb_table_monitor
                && difftime(current_time, last_table_monitor_time) > 60) {

		last_table_monitor_time = time(NULL);	
1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673

		printf("===========================================\n");

		ut_print_timestamp(stdout);

		printf(" INNODB TABLE MONITOR OUTPUT\n"
		       "===========================================\n");
	    	dict_print();

		printf("-----------------------------------\n"
	       		"END OF INNODB TABLE MONITOR OUTPUT\n"
	       		"==================================\n");
	    }
	}

1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693
	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
1694
				/* Timeout exceeded or a wrap-around in system
1695
				time counter: cancel the lock request queued
1696
				by the transaction and release possible
unknown's avatar
unknown committed
1697 1698 1699 1700 1701 1702 1703 1704
				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);
			        }
1705 1706 1707 1708 1709 1710 1711 1712
			}
		}
	}

	os_event_reset(srv_lock_timeout_thread_event);

	mutex_exit(&kernel_mutex);

unknown's avatar
Merge  
unknown committed
1713 1714 1715 1716
	if (srv_shutdown_state >= SRV_SHUTDOWN_CLEANUP) {
		goto exit_func;
	}

1717 1718 1719 1720
	if (some_waits || srv_print_innodb_monitor
			|| srv_print_innodb_lock_monitor
			|| srv_print_innodb_tablespace_monitor
			|| srv_print_innodb_table_monitor) {
1721 1722 1723
		goto loop;
	}

1724 1725
	/* No one was waiting for a lock and no monitor was active:
	suspend this thread */
unknown's avatar
Merge  
unknown committed
1726 1727 1728

	srv_lock_timeout_and_monitor_active = FALSE;

1729 1730 1731 1732
	os_event_wait(srv_lock_timeout_thread_event);

	goto loop;

unknown's avatar
Merge  
unknown committed
1733 1734 1735
exit_func:
	srv_lock_timeout_and_monitor_active = FALSE;

1736 1737 1738 1739
	/* 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);
1740 1741 1742
#ifndef __WIN__
        return(NULL);
#else
1743
	return(0);
1744
#endif
1745 1746
}

1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761
/*************************************************************************
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 */
	void*	arg)	/* in: a dummy parameter required by
			os_thread_create */
{
unknown's avatar
unknown committed
1762 1763
	ulint	cnt	= 0;

1764
	UT_NOT_USED(arg);
1765
#ifdef UNIV_DEBUG_THREAD_CREATION
unknown's avatar
unknown committed
1766 1767
	printf("Error monitor thread starts, id %lu\n",
			      os_thread_pf(os_thread_get_curr_id()));
1768
#endif
1769
loop:
unknown's avatar
Merge  
unknown committed
1770 1771
	srv_error_monitor_active = TRUE;

unknown's avatar
unknown committed
1772 1773 1774 1775
	cnt++;

	os_thread_sleep(2000000);

unknown's avatar
unknown committed
1776 1777 1778
	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 */
1779

unknown's avatar
unknown committed
1780 1781 1782
		srv_refresh_innodb_monitor_stats();
	}

1783
	sync_array_print_long_waits();
unknown's avatar
unknown committed
1784 1785 1786 1787 1788 1789 1790

	/* Flush stdout and stderr so that a database user gets their output
	to possible MySQL error file */

	fflush(stderr);
	fflush(stdout);

unknown's avatar
Merge  
unknown committed
1791 1792 1793 1794 1795 1796
	if (srv_shutdown_state < SRV_SHUTDOWN_LAST_PHASE) {

		goto loop;
	}

	srv_error_monitor_active = FALSE;
1797

1798 1799 1800
	/* 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
1801 1802
	os_thread_exit(NULL);

1803 1804 1805 1806 1807 1808 1809
#ifndef __WIN__
        return(NULL);
#else
	return(0);
#endif
}

1810
/***********************************************************************
1811
Tells the InnoDB server that there has been activity in the database
1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832
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
1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848
/***********************************************************************
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);
}

1849 1850 1851
/*************************************************************************
The master thread controlling the server. */

1852 1853 1854
#ifndef __WIN__
void*
#else
1855
ulint
1856
#endif
1857 1858 1859 1860 1861 1862 1863
srv_master_thread(
/*==============*/
			/* out: a dummy parameter */
	void*	arg)	/* in: a dummy parameter required by
			os_thread_create */
{
	os_event_t	event;
1864 1865
	time_t          last_flush_time;
	time_t          current_time;
1866 1867 1868 1869 1870
	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
1871
	ulint		n_tables_to_drop;
1872 1873 1874 1875
	ulint		n_ios;
	ulint		n_ios_old;
	ulint		n_ios_very_old;
	ulint		n_pend_ios;
unknown's avatar
unknown committed
1876
	ibool		skip_sleep	= FALSE;
1877 1878 1879 1880
	ulint		i;
	
	UT_NOT_USED(arg);

1881
#ifdef UNIV_DEBUG_THREAD_CREATION
unknown's avatar
unknown committed
1882 1883
	printf("Master thread starts, id %lu\n",
			      os_thread_pf(os_thread_get_curr_id()));
1884
#endif
unknown's avatar
unknown committed
1885 1886 1887
	srv_main_thread_process_no = os_proc_get_number();
	srv_main_thread_id = os_thread_pf(os_thread_get_curr_id());
	
1888 1889 1890 1891 1892 1893 1894 1895 1896 1897
	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:
1898 1899 1900 1901
	/*****************************************************************/
	/* ---- When there is database activity by users, we cycle in this
	loop */

unknown's avatar
unknown committed
1902
	srv_main_thread_op_info = (char*) "reserving kernel mutex";
1903 1904 1905

	n_ios_very_old = log_sys->n_log_ios + buf_pool->n_pages_read
						+ buf_pool->n_pages_written;
1906 1907
	mutex_enter(&kernel_mutex);

1908
	/* Store the user activity counter at the start of this loop */
1909 1910 1911 1912
	old_activity_count = srv_activity_count;

	mutex_exit(&kernel_mutex);

1913 1914 1915 1916 1917 1918 1919 1920
	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
1921 1922
	skip_sleep = FALSE;

1923
	for (i = 0; i < 10; i++) {
1924 1925
		n_ios_old = log_sys->n_log_ios + buf_pool->n_pages_read
						+ buf_pool->n_pages_written;
unknown's avatar
unknown committed
1926
		srv_main_thread_op_info = (char*)"sleeping";
unknown's avatar
unknown committed
1927 1928 1929 1930 1931 1932 1933
		
		if (!skip_sleep) {

		        os_thread_sleep(1000000);
		}

		skip_sleep = FALSE;
1934

unknown's avatar
unknown committed
1935 1936 1937 1938
		/* 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
1939 1940
		srv_main_thread_op_info =
					(char*)"doing background drop tables";
unknown's avatar
unknown committed
1941 1942 1943

		row_drop_tables_for_mysql_in_background();

unknown's avatar
unknown committed
1944
		srv_main_thread_op_info = (char*)"";
unknown's avatar
unknown committed
1945

unknown's avatar
unknown committed
1946 1947 1948 1949 1950
		if (srv_fast_shutdown && srv_shutdown_state > 0) {

			goto background_loop;
		}

unknown's avatar
unknown committed
1951 1952 1953 1954
		/* 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
1955
		srv_main_thread_op_info = (char*)"flushing log";
unknown's avatar
unknown committed
1956
		log_buffer_flush_to_disk();
unknown's avatar
unknown committed
1957

1958
		/* If there were less than 5 i/os during the
1959 1960 1961 1962 1963 1964 1965 1966
		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;
1967
		if (n_pend_ios < 3 && (n_ios - n_ios_old < 5)) {
unknown's avatar
unknown committed
1968
			srv_main_thread_op_info =
unknown's avatar
unknown committed
1969
					(char*)"doing insert buffer merge";
1970 1971
			ibuf_contract_for_n_pages(TRUE, 5);

unknown's avatar
unknown committed
1972 1973 1974
			srv_main_thread_op_info = (char*)"flushing log";

			log_buffer_flush_to_disk();
1975
		}
unknown's avatar
Merge  
unknown committed
1976

1977 1978 1979 1980 1981 1982 1983 1984
		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
1985 1986 1987 1988 1989 1990 1991

		        /* 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;
1992 1993
		}

1994 1995
		if (srv_activity_count == old_activity_count) {

1996 1997
			/* There is no user activity at the moment, go to
			the background loop */
1998 1999 2000 2001 2002

			goto background_loop;
		}
	}

2003 2004
	/* ---- We perform the following code approximately once per
	10 seconds when there is database activity */
2005

unknown's avatar
unknown committed
2006 2007 2008 2009 2010
#ifdef MEM_PERIODIC_CHECK
	/* Check magic numbers of every allocated mem block once in 10
	seconds */
	mem_validate_all_blocks();
#endif	
2011 2012
	/* 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
2013
	makes sense to flush 100 pages. */
2014 2015 2016 2017 2018 2019

	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
2020
		srv_main_thread_op_info = (char*) "flushing buffer pool pages";
unknown's avatar
unknown committed
2021
		buf_flush_batch(BUF_FLUSH_LIST, 100, ut_dulint_max);
2022

unknown's avatar
unknown committed
2023
		srv_main_thread_op_info = (char*) "flushing log";
unknown's avatar
unknown committed
2024
		log_buffer_flush_to_disk();
2025 2026 2027 2028 2029
	}

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

unknown's avatar
unknown committed
2030
	srv_main_thread_op_info = (char*)"doing insert buffer merge";
2031 2032
	ibuf_contract_for_n_pages(TRUE, 5);

unknown's avatar
unknown committed
2033
	srv_main_thread_op_info = (char*)"flushing log";
unknown's avatar
unknown committed
2034
	log_buffer_flush_to_disk();
2035 2036 2037 2038

	/* We run a full purge every 10 seconds, even if the server
	were active */
	
2039 2040
	n_pages_purged = 1;

unknown's avatar
unknown committed
2041 2042
	last_flush_time = time(NULL);

2043
	while (n_pages_purged) {
unknown's avatar
unknown committed
2044

unknown's avatar
Merge  
unknown committed
2045 2046 2047 2048 2049
		if (srv_fast_shutdown && srv_shutdown_state > 0) {

			goto background_loop;
		}

unknown's avatar
unknown committed
2050
		srv_main_thread_op_info = (char*)"purging";
2051
		n_pages_purged = trx_purge();
2052

unknown's avatar
unknown committed
2053 2054 2055
		current_time = time(NULL);

		if (difftime(current_time, last_flush_time) > 1) {
unknown's avatar
unknown committed
2056
			srv_main_thread_op_info = (char*) "flushing log";
2057

unknown's avatar
unknown committed
2058
		        log_buffer_flush_to_disk();
unknown's avatar
unknown committed
2059 2060
			last_flush_time = current_time;
		}
2061
	}
unknown's avatar
unknown committed
2062
	
unknown's avatar
unknown committed
2063
	srv_main_thread_op_info = (char*)"flushing buffer pool pages";
unknown's avatar
unknown committed
2064

2065 2066 2067 2068 2069 2070 2071
	/* Flush a few oldest pages to make a new checkpoint younger */

	if (buf_get_modified_ratio_pct() > 70) {

		/* 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
2072

unknown's avatar
unknown committed
2073 2074 2075
	        n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, 100,
							ut_dulint_max);
	} else {
2076 2077 2078 2079
	        /* 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
2080 2081 2082
	        n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, 10,
							ut_dulint_max);
	}
unknown's avatar
unknown committed
2083

unknown's avatar
unknown committed
2084
	srv_main_thread_op_info = (char*)"making checkpoint";
unknown's avatar
unknown committed
2085 2086 2087 2088

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

	log_checkpoint(TRUE, FALSE);
2089

unknown's avatar
unknown committed
2090
	srv_main_thread_op_info = (char*)"reserving kernel mutex";
2091

2092
	mutex_enter(&kernel_mutex);
2093 2094 2095 2096
	
	/* ---- When there is database activity, we jump from here back to
	the start of loop */

2097 2098 2099 2100
	if (srv_activity_count != old_activity_count) {
		mutex_exit(&kernel_mutex);
		goto loop;
	}
2101
	
2102 2103
	mutex_exit(&kernel_mutex);

2104 2105 2106 2107 2108 2109 2110
	/* If the database is quiet, we enter the background loop */

	/*****************************************************************/
background_loop:
	/* ---- In this loop we run background operations when the server
	is quiet from user activity */

2111 2112 2113
	/* The server has been quiet for a while: start running background
	operations */
		
2114 2115 2116 2117
	srv_main_thread_op_info = (char*)"doing background drop tables";

	n_tables_to_drop = row_drop_tables_for_mysql_in_background();

2118 2119 2120 2121 2122 2123 2124 2125 2126
	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
2127
	srv_main_thread_op_info = (char*)"purging";
2128

unknown's avatar
unknown committed
2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151
	/* 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;
		}

		srv_main_thread_op_info = (char*)"purging";
		n_pages_purged = trx_purge();

		current_time = time(NULL);

		if (difftime(current_time, last_flush_time) > 1) {
			srv_main_thread_op_info = (char*) "flushing log";

		        log_buffer_flush_to_disk();
			last_flush_time = current_time;
		}
unknown's avatar
unknown committed
2152
	}
2153

unknown's avatar
unknown committed
2154
	srv_main_thread_op_info = (char*)"reserving kernel mutex";
2155

2156 2157 2158 2159 2160 2161 2162
	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
2163
	srv_main_thread_op_info = (char*)"doing insert buffer merge";
unknown's avatar
unknown committed
2164 2165 2166 2167 2168 2169

	if (srv_fast_shutdown && srv_shutdown_state > 0) {
	        n_bytes_merged = 0;
	} else {
	        n_bytes_merged = ibuf_contract_for_n_pages(TRUE, 20);
	}
2170

unknown's avatar
unknown committed
2171
	srv_main_thread_op_info = (char*)"reserving kernel mutex";
2172

2173 2174 2175 2176 2177 2178 2179
	mutex_enter(&kernel_mutex);
	if (srv_activity_count != old_activity_count) {
		mutex_exit(&kernel_mutex);
		goto loop;
	}
	mutex_exit(&kernel_mutex);
	
2180
flush_loop:
unknown's avatar
unknown committed
2181
	srv_main_thread_op_info = (char*)"flushing buffer pool pages";
2182
	n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, 100, ut_dulint_max);
2183

unknown's avatar
unknown committed
2184
	srv_main_thread_op_info = (char*)"reserving kernel mutex";
2185

2186 2187 2188 2189 2190 2191 2192
	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
2193 2194
	srv_main_thread_op_info =
			(char*) "waiting for buffer pool flush to end";
2195 2196
	buf_flush_wait_batch_end(BUF_FLUSH_LIST);

unknown's avatar
unknown committed
2197 2198 2199 2200
	srv_main_thread_op_info = (char*) "flushing log";

	log_buffer_flush_to_disk();

unknown's avatar
unknown committed
2201
	srv_main_thread_op_info = (char*)"making checkpoint";
2202

2203 2204
	log_checkpoint(TRUE, FALSE);

2205 2206 2207 2208 2209 2210 2211 2212
	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
2213
	srv_main_thread_op_info = (char*)"reserving kernel mutex";
2214

2215 2216 2217 2218 2219 2220
	mutex_enter(&kernel_mutex);
	if (srv_activity_count != old_activity_count) {
		mutex_exit(&kernel_mutex);
		goto loop;
	}
	mutex_exit(&kernel_mutex);
2221

unknown's avatar
unknown committed
2222
	srv_main_thread_op_info =
unknown's avatar
unknown committed
2223
				(char*)"archiving log (if log archive is on)";
2224 2225 2226
	
	log_archive_do(FALSE, &n_bytes_archived);

2227 2228
	/* Keep looping in the background loop if still work to do */

unknown's avatar
Merge  
unknown committed
2229
	if (srv_fast_shutdown && srv_shutdown_state > 0) {
unknown's avatar
unknown committed
2230 2231
		if (n_tables_to_drop + n_pages_flushed
				+ n_bytes_archived != 0) {
unknown's avatar
Merge  
unknown committed
2232

2233 2234 2235 2236
			/* If we are doing a fast shutdown (= the default)
			we do not do purge or insert buffer merge. But we
			flush the buffer pool completely to disk. */

unknown's avatar
Merge  
unknown committed
2237 2238
			goto background_loop;
		}
unknown's avatar
unknown committed
2239
	} else if (n_tables_to_drop +
2240
		   n_pages_purged + n_bytes_merged + n_pages_flushed
2241
						+ n_bytes_archived != 0) {
2242 2243 2244
		/* In a 'slow' shutdown we run purge and the insert buffer
		merge to completion */

2245 2246 2247 2248 2249 2250
		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
2251
suspend_thread:
unknown's avatar
unknown committed
2252
	srv_main_thread_op_info = (char*)"suspending";
2253

2254 2255
	mutex_enter(&kernel_mutex);

unknown's avatar
unknown committed
2256 2257 2258 2259 2260 2261
	if (row_get_background_drop_list_len_low() > 0) {
		mutex_exit(&kernel_mutex);

		goto loop;
	}

2262 2263 2264 2265
	event = srv_suspend_thread();

	mutex_exit(&kernel_mutex);

unknown's avatar
unknown committed
2266
	srv_main_thread_op_info = (char*)"waiting for server activity";
2267

2268 2269
	os_event_wait(event);

unknown's avatar
unknown committed
2270 2271 2272 2273 2274 2275 2276
	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);
	}

2277 2278 2279
	/* When there is user activity, InnoDB will set the event and the main
	thread goes back to loop: */

2280 2281
	goto loop;

2282 2283 2284 2285 2286 2287 2288
	/* 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);

2289 2290 2291
#ifndef __WIN__
        return(NULL);
#else
2292
	return(0);
2293
#endif
2294
}