srv0srv.c 63.9 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;

55 56 57 58
/* 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
59 60 61
ibool	srv_lock_timeout_and_monitor_active = FALSE;
ibool	srv_error_monitor_active = FALSE;

unknown's avatar
unknown committed
62
char*	srv_main_thread_op_info = (char*) "";
63

64 65 66 67 68 69 70 71
/* 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;

unknown's avatar
unknown committed
72 73 74 75
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 */
76 77 78 79
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
80 81 82 83 84 85 86
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 */
87 88
ulint*  srv_data_file_is_raw_partition = NULL;

89 90 91 92 93
/* 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;

94 95 96 97 98
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
99
ibool	srv_log_archive_on	= FALSE;
100
ulint	srv_log_buffer_size	= ULINT_MAX;	/* size in database pages */ 
unknown's avatar
unknown committed
101
ulint	srv_flush_log_at_trx_commit = 1;
102

unknown's avatar
unknown committed
103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139
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
};
140
		
unknown's avatar
unknown committed
141 142 143 144 145 146 147 148
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() */
149 150 151 152 153 154 155 156 157
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
158

unknown's avatar
unknown committed
159 160 161
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;
162

unknown's avatar
unknown committed
163 164
ulint	srv_max_n_open_files	  = 300;

165 166 167 168 169
/* 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. */

170
ulint	srv_max_buf_pool_modified_pct	= 90;
171

172
/* If the following is != 0 we do not allow inserts etc. This protects
unknown's avatar
Merge  
unknown committed
173
the user from forgetting the innodb_force_recovery keyword to my.cnf */
174 175

ulint	srv_force_recovery	= 0;
unknown's avatar
Merge  
unknown committed
176
/*-----------------------*/
177 178 179 180 181 182
/* 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
183 184 185 186 187 188 189
/* 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
190
ulint	srv_thread_concurrency	= 8;
unknown's avatar
Merge  
unknown committed
191 192 193

os_fast_mutex_t	srv_conc_mutex;		/* this mutex protects srv_conc data
					structures */
unknown's avatar
unknown committed
194 195 196 197 198 199 200
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
201 202 203
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
204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221

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 */
222
srv_conc_slot_t* srv_conc_slots;			/* array of wait
unknown's avatar
Merge  
unknown committed
223
							slots */
unknown's avatar
unknown committed
224 225 226 227 228

/* 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
229 230 231 232 233
/*-----------------------*/
/* 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;
234

235 236
ibool	srv_use_doublewrite_buf	= TRUE;

unknown's avatar
unknown committed
237 238
ibool   srv_set_thread_priorities = TRUE;
int     srv_query_thread_priority = 0;
unknown's avatar
unknown committed
239 240 241 242 243 244 245

/* 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;


246 247 248 249 250 251 252 253 254 255 256
/*-------------------------------------------*/
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
257 258 259 260 261 262 263 264
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;
265

266 267 268 269 270
/*
  Set the following to 0 if you want InnoDB to write messages on
  stderr on startup/shutdown
*/
ibool	srv_print_verbose_log		= TRUE;
271
ibool	srv_print_innodb_monitor	= FALSE;
272 273
ibool   srv_print_innodb_lock_monitor   = FALSE;
ibool   srv_print_innodb_tablespace_monitor = FALSE;
274
ibool   srv_print_innodb_table_monitor = FALSE;
275

276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299
/* 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;

300 301 302
/* Array of English strings describing the current state of an
i/o handler thread */

303 304
const char* srv_io_thread_op_info[SRV_MAX_N_IO_THREADS];
const char* srv_io_thread_function[SRV_MAX_N_IO_THREADS];
305

unknown's avatar
unknown committed
306 307
time_t	srv_last_monitor_time;

308 309 310 311 312 313
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
314

unknown's avatar
unknown committed
315 316 317
ulint	srv_main_thread_process_no	= 0;
ulint	srv_main_thread_id		= 0;

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
/*
	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];

537 538 539 540 541 542 543 544 545 546 547 548 549 550
/*************************************************************************
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;
}
551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642

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

643
#ifdef UNIV_SYNC_DEBUG
644
	ut_ad(mutex_own(&kernel_mutex));
645
#endif /* UNIV_SYNC_DEBUG */
646 647 648 649
	
	slot_no = thr_local_get_slot_no(os_thread_get_curr_id());

	if (srv_print_thread_releases) {
650 651
		fprintf(stderr,
			"Suspending thread %lu to slot %lu meter %lu\n",
unknown's avatar
unknown committed
652 653
			(ulong) os_thread_get_curr_id(), (ulong) slot_no,
			(ulong) srv_meter[SRV_RECOVERY]);
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
	}

	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);
696
#ifdef UNIV_SYNC_DEBUG
697
	ut_ad(mutex_own(&kernel_mutex));
698
#endif /* UNIV_SYNC_DEBUG */
699 700 701 702 703
	
	for (i = 0; i < OS_THREAD_MAX_N; i++) {
	
		slot = srv_table_get_nth_slot(i);

unknown's avatar
unknown committed
704
		if (slot->in_use && slot->type == type && slot->suspended) {
705 706 707 708 709 710 711 712
			
			slot->suspended = FALSE;

			srv_n_threads_active[type]++;

			os_event_set(slot->event);

			if (srv_print_thread_releases) {
713
				fprintf(stderr,
714
		"Releasing thread %lu type %lu from slot %lu meter %lu\n",
715 716
				(ulong) slot->id, (ulong) type, (ulong) i,
		                (ulong) srv_meter[SRV_RECOVERY]);
717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759
			}

			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
760
static
761 762 763 764
void
srv_init(void)
/*==========*/
{
unknown's avatar
Merge  
unknown committed
765 766 767
	srv_conc_slot_t* 	conc_slot;
	srv_slot_t*		slot;
	ulint			i;
768 769 770 771 772 773

	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
774 775 776

	mutex_create(&srv_innodb_monitor_mutex);
	mutex_set_level(&srv_innodb_monitor_mutex, SYNC_NO_ORDER_CHECK);
777 778 779 780 781 782
	
	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
783
                slot->type=0;	/* Avoid purify errors */
784 785 786 787 788 789 790 791 792
		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
793
		slot->type = 0;
794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814
		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
815 816 817 818 819 820

	/* Init the server concurrency restriction data structures */

	os_fast_mutex_init(&srv_conc_mutex);
	
	UT_LIST_INIT(srv_conc_queue);
821

822 823
	srv_conc_slots = mem_alloc(OS_THREAD_MAX_N * sizeof(srv_conc_slot_t));

unknown's avatar
Merge  
unknown committed
824 825 826 827 828 829
	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);
	}
830
}	
unknown's avatar
unknown committed
831

832
/*************************************************************************
unknown's avatar
unknown committed
833
Frees the OS fast mutex created in srv_init(). */
unknown's avatar
unknown committed
834

835
void
unknown's avatar
unknown committed
836 837
srv_free(void)
/*==========*/
838
{
unknown's avatar
unknown committed
839
	os_fast_mutex_free(&srv_conc_mutex);
840 841
}

unknown's avatar
unknown committed
842
/*************************************************************************
unknown's avatar
unknown committed
843
Initializes the synchronization primitives, memory system, and the thread
unknown's avatar
unknown committed
844 845 846
local storage. */

void
unknown's avatar
unknown committed
847
srv_general_init(void)
unknown's avatar
unknown committed
848 849
/*==================*/
{
unknown's avatar
unknown committed
850 851 852 853
	os_sync_init();
	sync_init();
	mem_init(srv_mem_pool_size);
	thr_local_init();
unknown's avatar
unknown committed
854 855
}

unknown's avatar
unknown committed
856 857
/*======================= InnoDB Server FIFO queue =======================*/

unknown's avatar
unknown committed
858

unknown's avatar
Merge  
unknown committed
859 860 861 862 863 864 865 866 867 868
/*************************************************************************
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
869
	ibool			has_slept = FALSE;
870
	srv_conc_slot_t*	slot	  = NULL;
unknown's avatar
Merge  
unknown committed
871 872
	ulint			i;

unknown's avatar
unknown committed
873 874 875 876 877 878 879 880 881 882 883 884 885 886
	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
Merge  
unknown committed
887 888

	os_fast_mutex_lock(&srv_conc_mutex);
unknown's avatar
unknown committed
889
retry:
unknown's avatar
unknown committed
890 891
	if (trx->declared_to_be_inside_innodb) {
	        ut_print_timestamp(stderr);
892
		fputs(
unknown's avatar
unknown committed
893
"  InnoDB: Error: trying to declare trx to enter InnoDB, but\n"
894 895 896
"InnoDB: it already is declared.\n", stderr);
		trx_print(stderr, trx);
		putc('\n', stderr);
unknown's avatar
unknown committed
897 898 899
		os_fast_mutex_unlock(&srv_conc_mutex);

		return;
unknown's avatar
unknown committed
900 901
	}

unknown's avatar
unknown committed
902
	if (srv_conc_n_threads < (lint)srv_thread_concurrency) {
unknown's avatar
Merge  
unknown committed
903

unknown's avatar
unknown committed
904 905 906 907
		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
908 909 910 911
		os_fast_mutex_unlock(&srv_conc_mutex);

		return;
	}
unknown's avatar
unknown committed
912

unknown's avatar
unknown committed
913 914 915
	/* If the transaction is not holding resources, let it sleep for 50
	milliseconds, and try again then */
 
unknown's avatar
unknown committed
916 917 918
	if (!has_slept && !trx->has_search_latch
	    && NULL == UT_LIST_GET_FIRST(trx->trx_locks)) {

unknown's avatar
unknown committed
919 920 921 922 923 924 925
	        has_slept = TRUE; /* We let is sleep only once to avoid
				  starvation */

		srv_conc_n_waiting_threads++;

		os_fast_mutex_unlock(&srv_conc_mutex);

926 927
		trx->op_info = (char*)"sleeping before joining InnoDB queue";

unknown's avatar
unknown committed
928
		os_thread_sleep(50000);
unknown's avatar
unknown committed
929

930
		trx->op_info = (char*)"";
unknown's avatar
unknown committed
931

unknown's avatar
unknown committed
932 933 934
		os_fast_mutex_lock(&srv_conc_mutex);

		srv_conc_n_waiting_threads--;
unknown's avatar
unknown committed
935 936

		goto retry;
unknown's avatar
unknown committed
937
	}   
unknown's avatar
unknown committed
938

unknown's avatar
unknown committed
939
	/* Too many threads inside: put the current thread to a queue */
unknown's avatar
Merge  
unknown committed
940 941 942 943 944

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

		if (!slot->reserved) {
945

unknown's avatar
Merge  
unknown committed
946 947 948 949 950 951 952 953 954
			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
955 956
		trx->declared_to_be_inside_innodb = TRUE;
		trx->n_tickets_to_enter_innodb = 0;
unknown's avatar
Merge  
unknown committed
957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975

		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
976 977
	srv_conc_n_waiting_threads++;

unknown's avatar
Merge  
unknown committed
978 979 980 981 982
	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
983
	trx->op_info = (char*)"waiting in InnoDB queue";
unknown's avatar
unknown committed
984

unknown's avatar
Merge  
unknown committed
985 986
	os_event_wait(slot->event);

unknown's avatar
unknown committed
987
	trx->op_info = (char*)"";
unknown's avatar
unknown committed
988

unknown's avatar
Merge  
unknown committed
989 990
	os_fast_mutex_lock(&srv_conc_mutex);

unknown's avatar
unknown committed
991 992
	srv_conc_n_waiting_threads--;

unknown's avatar
Merge  
unknown committed
993 994 995 996 997 998 999
	/* 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
1000 1001 1002
	trx->declared_to_be_inside_innodb = TRUE;
	trx->n_tickets_to_enter_innodb = SRV_FREE_TICKETS_TO_ENTER;

unknown's avatar
Merge  
unknown committed
1003 1004 1005 1006 1007 1008 1009 1010
	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
1011 1012 1013 1014
srv_conc_force_enter_innodb(
/*========================*/
	trx_t*	trx)	/* in: transaction object associated with the
			thread */
unknown's avatar
Merge  
unknown committed
1015
{
unknown's avatar
unknown committed
1016 1017 1018 1019 1020
	if (srv_thread_concurrency >= 500) {
	
		return;
	}

unknown's avatar
Merge  
unknown committed
1021 1022 1023
	os_fast_mutex_lock(&srv_conc_mutex);

	srv_conc_n_threads++;
unknown's avatar
unknown committed
1024 1025
	trx->declared_to_be_inside_innodb = TRUE;
	trx->n_tickets_to_enter_innodb = 0;
unknown's avatar
Merge  
unknown committed
1026 1027 1028 1029 1030

	os_fast_mutex_unlock(&srv_conc_mutex);
}

/*************************************************************************
unknown's avatar
unknown committed
1031 1032
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
1033 1034

void
unknown's avatar
unknown committed
1035 1036 1037 1038
srv_conc_force_exit_innodb(
/*=======================*/
	trx_t*	trx)	/* in: transaction object associated with the
			thread */
unknown's avatar
Merge  
unknown committed
1039 1040 1041
{
	srv_conc_slot_t*	slot	= NULL;

unknown's avatar
unknown committed
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051
	if (srv_thread_concurrency >= 500) {
	
		return;
	}

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

unknown's avatar
Merge  
unknown committed
1052 1053 1054
	os_fast_mutex_lock(&srv_conc_mutex);

	srv_conc_n_threads--;
unknown's avatar
unknown committed
1055 1056
	trx->declared_to_be_inside_innodb = FALSE;
	trx->n_tickets_to_enter_innodb = 0;
unknown's avatar
Merge  
unknown committed
1057

unknown's avatar
unknown committed
1058
	if (srv_conc_n_threads < (lint)srv_thread_concurrency) {
unknown's avatar
Merge  
unknown committed
1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084
		/* 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
1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114
/*************************************************************************
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);
}

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

1115
/*************************************************************************
1116
Normalizes init parameter values to use units we use inside InnoDB. */
1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132
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
1133 1134 1135
	srv_last_file_size_max = srv_last_file_size_max
					* ((1024 * 1024) / UNIV_PAGE_SIZE);
		
1136 1137 1138 1139
	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
1140 1141 1142
	srv_pool_size = srv_pool_size / (UNIV_PAGE_SIZE / 1024);

	srv_awe_window_size = srv_awe_window_size / UNIV_PAGE_SIZE;
1143
	
unknown's avatar
unknown committed
1144 1145 1146 1147 1148 1149 1150
	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 {
1151
	        srv_lock_table_size = 5 * srv_pool_size;
unknown's avatar
unknown committed
1152
	}
1153 1154 1155 1156 1157

	return(DB_SUCCESS);
}

/*************************************************************************
1158
Boots the InnoDB server. */
1159 1160 1161 1162 1163 1164 1165 1166 1167

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

	/* Transform the init parameter values given by MySQL to
1168
	use units we use inside InnoDB: */
1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189
	
	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
1190
NOTE! The kernel mutex has to be reserved by the caller! */
1191 1192 1193 1194 1195 1196 1197 1198 1199
static
srv_slot_t*
srv_table_reserve_slot_for_mysql(void)
/*==================================*/
			/* out: reserved slot */
{
	srv_slot_t*	slot;
	ulint		i;

1200
#ifdef UNIV_SYNC_DEBUG
unknown's avatar
unknown committed
1201
	ut_ad(mutex_own(&kernel_mutex));
1202
#endif /* UNIV_SYNC_DEBUG */
unknown's avatar
unknown committed
1203

1204 1205 1206 1207 1208
	i = 0;
	slot = srv_mysql_table + i;

	while (slot->in_use) {
		i++;
unknown's avatar
unknown committed
1209 1210 1211 1212 1213 1214 1215 1216 1217

		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"
1218
"InnoDB: on Linux. But first we print a list of waiting threads.\n", (ulong) i);
unknown's avatar
unknown committed
1219 1220 1221 1222 1223 1224 1225

			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",
1226 1227 1228 1229
				  (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
1230 1231
			}

1232
		        ut_error;
unknown's avatar
unknown committed
1233
		}
1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247
		
		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
1248 1249 1250 1251 1252
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. */
1253

unknown's avatar
unknown committed
1254
void
1255 1256
srv_suspend_mysql_thread(
/*=====================*/
unknown's avatar
unknown committed
1257 1258
	que_thr_t*	thr)	/* in: query thread associated with the MySQL
				OS thread */
1259 1260 1261 1262
{
	srv_slot_t*	slot;
	os_event_t	event;
	double		wait_time;
unknown's avatar
unknown committed
1263
	trx_t*		trx;
unknown's avatar
unknown committed
1264 1265
	ibool		had_dict_lock			= FALSE;
	ibool		was_declared_inside_innodb	= FALSE;
unknown's avatar
unknown committed
1266
	
1267
#ifdef UNIV_SYNC_DEBUG
1268
	ut_ad(!mutex_own(&kernel_mutex));
1269
#endif /* UNIV_SYNC_DEBUG */
1270

unknown's avatar
unknown committed
1271 1272
	trx = thr_get_trx(thr);
	
1273 1274 1275 1276
	os_event_set(srv_lock_timeout_thread_event);

	mutex_enter(&kernel_mutex);

unknown's avatar
unknown committed
1277 1278
	trx->error_state = DB_SUCCESS;

1279 1280
	if (thr->state == QUE_THR_RUNNING) {

unknown's avatar
unknown committed
1281 1282 1283 1284 1285 1286 1287 1288 1289 1290
		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;
		}
1291 1292 1293

		mutex_exit(&kernel_mutex);

unknown's avatar
unknown committed
1294
		return;
1295 1296
	}
	
unknown's avatar
unknown committed
1297 1298
	ut_ad(thr->is_active == FALSE);

1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314
	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
1315 1316 1317
	if (trx->declared_to_be_inside_innodb) {

		was_declared_inside_innodb = TRUE;
unknown's avatar
Merge  
unknown committed
1318
	
unknown's avatar
unknown committed
1319 1320 1321
		/* 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
1322
	
unknown's avatar
unknown committed
1323 1324
		srv_conc_force_exit_innodb(trx);
	}
unknown's avatar
Merge  
unknown committed
1325

unknown's avatar
unknown committed
1326
	/* Release possible foreign key check latch */
unknown's avatar
unknown committed
1327 1328 1329
	if (trx->dict_operation_lock_mode == RW_S_LATCH) {

		had_dict_lock = TRUE;
unknown's avatar
unknown committed
1330

unknown's avatar
unknown committed
1331
		row_mysql_unfreeze_data_dictionary(trx);
unknown's avatar
unknown committed
1332 1333
	}

unknown's avatar
unknown committed
1334 1335
	ut_a(trx->dict_operation_lock_mode == 0);

1336 1337 1338 1339
	/* Wait for the release */
	
	os_event_wait(event);

unknown's avatar
unknown committed
1340
	if (had_dict_lock) {
unknown's avatar
unknown committed
1341

unknown's avatar
unknown committed
1342
		row_mysql_freeze_data_dictionary(trx);
unknown's avatar
unknown committed
1343 1344
	}

unknown's avatar
unknown committed
1345 1346 1347
	if (was_declared_inside_innodb) {

		/* Return back inside InnoDB */
unknown's avatar
Merge  
unknown committed
1348
	
unknown's avatar
unknown committed
1349 1350
		srv_conc_force_enter_innodb(trx);
	}
unknown's avatar
Merge  
unknown committed
1351

1352 1353 1354 1355 1356 1357 1358 1359
	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
1360 1361 1362 1363 1364 1365
	if (trx->was_chosen_as_deadlock_victim) {

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

1366 1367 1368 1369 1370
	mutex_exit(&kernel_mutex);

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

unknown's avatar
unknown committed
1371 1372
	    	trx->error_state = DB_LOCK_WAIT_TIMEOUT;
	}
1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387
}

/************************************************************************
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;
	
1388
#ifdef UNIV_SYNC_DEBUG
1389
	ut_ad(mutex_own(&kernel_mutex));
1390
#endif /* UNIV_SYNC_DEBUG */
1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407

	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
1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435
/**********************************************************************
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
1436
/**********************************************************************
1437
Outputs to a file the output of the InnoDB Monitor. */
1438

unknown's avatar
unknown committed
1439
void
1440 1441 1442
srv_printf_innodb_monitor(
/*======================*/
	FILE*	file)	/* in: output stream */
1443
{
unknown's avatar
unknown committed
1444 1445
	double	time_elapsed;
	time_t	current_time;
unknown's avatar
unknown committed
1446
	ulint   n_reserved;
unknown's avatar
Merge  
unknown committed
1447

unknown's avatar
unknown committed
1448
	mutex_enter(&srv_innodb_monitor_mutex);
1449

unknown's avatar
unknown committed
1450
	current_time = time(NULL);
1451

unknown's avatar
unknown committed
1452 1453 1454
	/* 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 */
1455
	
unknown's avatar
unknown committed
1456 1457
	time_elapsed = difftime(current_time, srv_last_monitor_time)
			+ 0.001;
1458

unknown's avatar
unknown committed
1459
	srv_last_monitor_time = time(NULL);
1460

1461 1462
	rewind(file);
	fputs("\n=====================================\n", file);
unknown's avatar
unknown committed
1463

1464 1465 1466 1467 1468 1469
	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
1470

1471 1472 1473 1474
	fputs("----------\n"
		"SEMAPHORES\n"
		"----------\n", file);
	sync_print(file);
unknown's avatar
unknown committed
1475

unknown's avatar
unknown committed
1476 1477 1478 1479 1480 1481 1482
	/* 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);

1483 1484 1485 1486 1487 1488
	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
1489

unknown's avatar
unknown committed
1490
	mutex_exit(&dict_foreign_err_mutex);
1491

1492 1493 1494 1495 1496
	lock_print_info(file);
	fputs("--------\n"
		"FILE I/O\n"
		"--------\n", file);
	os_aio_print(file);
1497

1498 1499 1500 1501 1502 1503 1504 1505
	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
1506 1507 1508 1509 1510
		"%.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
1511 1512
	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
1513

1514
	fputs("---\n"
1515
		       "LOG\n"
1516 1517 1518 1519
		"---\n", file);
	log_print(file);

	fputs("----------------------\n"
1520
		       "BUFFER POOL AND MEMORY\n"
1521 1522
		"----------------------\n", file);
	fprintf(file,
unknown's avatar
unknown committed
1523
	"Total memory allocated " ULINTPF
unknown's avatar
unknown committed
1524
	"; in additional pool allocated " ULINTPF "\n",
unknown's avatar
unknown committed
1525 1526
				ut_total_allocated_memory,
				mem_pool_get_reserved(mem_comm_pool));
1527 1528

	if (mem_out_of_mem_err_msg_count > 0) {
unknown's avatar
unknown committed
1529
	        fprintf(file,
unknown's avatar
unknown committed
1530 1531 1532
	"Mem allocation has spilled out of additional mem pool" ULINTPF
						"times\n",
					mem_out_of_mem_err_msg_count);
1533 1534
	}

unknown's avatar
unknown committed
1535
	if (srv_use_awe) {
unknown's avatar
unknown committed
1536
		fprintf(file,
unknown's avatar
unknown committed
1537
	"In addition to that %lu MB of AWE memory allocated\n",
unknown's avatar
unknown committed
1538
		(ulong) (srv_pool_size / ((1024 * 1024) / UNIV_PAGE_SIZE)));
unknown's avatar
unknown committed
1539 1540
	}
	
1541
	buf_print_io(file);
1542

1543 1544 1545 1546
	fputs("--------------\n"
		"ROW OPERATIONS\n"
		"--------------\n", file);
	fprintf(file, "%ld queries inside InnoDB, %lu queries in queue\n",
unknown's avatar
unknown committed
1547 1548
       		       (long) srv_conc_n_threads,
		       (ulong) srv_conc_n_waiting_threads);
unknown's avatar
unknown committed
1549 1550
        n_reserved = fil_space_get_n_reserved_extents(0);
        if (n_reserved > 0) {
unknown's avatar
unknown committed
1551
                fprintf(file,
unknown's avatar
unknown committed
1552
        "%lu tablespace extents now reserved for B-tree split operations\n",
1553
                                                    (ulong) n_reserved);
unknown's avatar
unknown committed
1554
        }
1555

unknown's avatar
unknown committed
1556
#ifdef UNIV_LINUX
1557
	fprintf(file, "Main thread process no. %lu, id %lu, state: %s\n",
1558 1559 1560
		       (ulong) srv_main_thread_process_no,
		       (ulong) srv_main_thread_id,
		       srv_main_thread_op_info);
unknown's avatar
unknown committed
1561
#else
1562
	fprintf(file, "Main thread id %lu, state: %s\n",
1563
			(ulong) srv_main_thread_id,
unknown's avatar
unknown committed
1564
			srv_main_thread_op_info);
unknown's avatar
unknown committed
1565
#endif
1566
	fprintf(file,
unknown's avatar
unknown committed
1567 1568 1569 1570 1571 1572
	"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);
1573
	fprintf(file,
1574 1575 1576 1577 1578 1579 1580 1581 1582 1583
	"%.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
1584 1585 1586 1587
	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;
1588

1589
	fputs("----------------------------\n"
1590
		       "END OF INNODB MONITOR OUTPUT\n"
1591
		"============================\n", file);
1592

unknown's avatar
unknown committed
1593
	mutex_exit(&srv_innodb_monitor_mutex);
1594
	fflush(file);
unknown's avatar
unknown committed
1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608
}

/*************************************************************************
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 */
1609 1610
	void*	arg __attribute__((unused)))
			/* in: a dummy parameter required by
unknown's avatar
unknown committed
1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621
			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;

1622
#ifdef UNIV_DEBUG_THREAD_CREATION
1623
	fprintf(stderr, "Lock timeout thread starts, id %lu\n",
unknown's avatar
unknown committed
1624
			     os_thread_pf(os_thread_get_curr_id()));
1625
#endif
unknown's avatar
unknown committed
1626 1627 1628 1629 1630 1631 1632 1633 1634 1635
	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
1636
	os_thread_sleep(1000000);
unknown's avatar
unknown committed
1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651

	/* 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) {
1652 1653
		srv_printf_innodb_monitor(stderr);
	    }
unknown's avatar
unknown committed
1654

1655 1656 1657
	    mutex_enter(&srv_monitor_file_mutex);
	    rewind(srv_monitor_file);
	    srv_printf_innodb_monitor(srv_monitor_file);
1658
	    os_file_set_eof(srv_monitor_file);
1659
	    mutex_exit(&srv_monitor_file_mutex);
1660

1661 1662
	    if (srv_print_innodb_tablespace_monitor
		&& difftime(current_time, last_table_monitor_time) > 60) {
unknown's avatar
unknown committed
1663 1664 1665

		last_table_monitor_time = time(NULL);	

1666 1667
		fputs("================================================\n",
			stderr);
1668

1669
		ut_print_timestamp(stderr);
1670

1671 1672 1673
		fputs(" INNODB TABLESPACE MONITOR OUTPUT\n"
			"================================================\n",
			stderr);
1674 1675
	       
		fsp_print(0);
1676
		fputs("Validating tablespace\n", stderr);
1677
		fsp_validate(0);
1678 1679
		fputs("Validation ok\n"
			"---------------------------------------\n"
1680
	       		"END OF INNODB TABLESPACE MONITOR OUTPUT\n"
1681 1682
			"=======================================\n",
			stderr);
1683 1684
	    }

unknown's avatar
unknown committed
1685
	    if (srv_print_innodb_table_monitor
1686
		&& difftime(current_time, last_table_monitor_time) > 60) {
unknown's avatar
unknown committed
1687 1688

		last_table_monitor_time = time(NULL);	
1689

1690
		fputs("===========================================\n", stderr);
1691

1692
		ut_print_timestamp(stderr);
1693

1694 1695 1696
		fputs(" INNODB TABLE MONITOR OUTPUT\n"
			"===========================================\n",
			stderr);
1697 1698
	    	dict_print();

1699
		fputs("-----------------------------------\n"
1700
	       		"END OF INNODB TABLE MONITOR OUTPUT\n"
1701 1702
			"==================================\n",
			stderr);
1703 1704 1705
	    }
	}

1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725
	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
1726
				/* Timeout exceeded or a wrap-around in system
1727
				time counter: cancel the lock request queued
1728
				by the transaction and release possible
unknown's avatar
unknown committed
1729 1730 1731 1732 1733 1734 1735 1736
				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);
			        }
1737 1738 1739 1740 1741 1742 1743 1744
			}
		}
	}

	os_event_reset(srv_lock_timeout_thread_event);

	mutex_exit(&kernel_mutex);

unknown's avatar
Merge  
unknown committed
1745 1746 1747 1748
	if (srv_shutdown_state >= SRV_SHUTDOWN_CLEANUP) {
		goto exit_func;
	}

1749 1750 1751 1752
	if (some_waits || srv_print_innodb_monitor
			|| srv_print_innodb_lock_monitor
			|| srv_print_innodb_tablespace_monitor
			|| srv_print_innodb_table_monitor) {
1753 1754 1755
		goto loop;
	}

1756 1757
	/* No one was waiting for a lock and no monitor was active:
	suspend this thread */
unknown's avatar
Merge  
unknown committed
1758 1759 1760

	srv_lock_timeout_and_monitor_active = FALSE;

1761 1762 1763
#if 0
	/* The following synchronisation is disabled, since
	the InnoDB monitor output is to be updated every 15 seconds. */
1764
	os_event_wait(srv_lock_timeout_thread_event);
1765
#endif
1766 1767
	goto loop;

unknown's avatar
Merge  
unknown committed
1768 1769 1770
exit_func:
	srv_lock_timeout_and_monitor_active = FALSE;

1771 1772 1773 1774
	/* 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);
1775
#ifndef __WIN__
1776
	return(NULL);
1777
#else
1778
	return(0);
1779
#endif
1780 1781
}

1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793
/*************************************************************************
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 */
1794 1795
	void*	arg __attribute__((unused)))
			/* in: a dummy parameter required by
1796 1797
			os_thread_create */
{
unknown's avatar
unknown committed
1798
	ulint	cnt	= 0;
unknown's avatar
unknown committed
1799 1800
	dulint	old_lsn;
	dulint	new_lsn;
unknown's avatar
unknown committed
1801

unknown's avatar
unknown committed
1802 1803
	old_lsn = srv_start_lsn;

1804
#ifdef UNIV_DEBUG_THREAD_CREATION
1805
	fprintf(stderr, "Error monitor thread starts, id %lu\n",
unknown's avatar
unknown committed
1806
			      os_thread_pf(os_thread_get_curr_id()));
1807
#endif
1808
loop:
unknown's avatar
Merge  
unknown committed
1809 1810
	srv_error_monitor_active = TRUE;

unknown's avatar
unknown committed
1811 1812
	cnt++;

unknown's avatar
unknown committed
1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823
	/* 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",
1824 1825 1826 1827
		(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
1828 1829 1830 1831
	}

	old_lsn = new_lsn;

unknown's avatar
unknown committed
1832 1833 1834
	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 */
1835

unknown's avatar
unknown committed
1836 1837 1838
		srv_refresh_innodb_monitor_stats();
	}

1839
	sync_array_print_long_waits();
unknown's avatar
unknown committed
1840

1841
	/* Flush stderr so that a database user gets the output
unknown's avatar
unknown committed
1842 1843 1844 1845
	to possible MySQL error file */

	fflush(stderr);

unknown's avatar
unknown committed
1846 1847
	os_thread_sleep(2000000);

unknown's avatar
Merge  
unknown committed
1848 1849 1850 1851 1852 1853
	if (srv_shutdown_state < SRV_SHUTDOWN_LAST_PHASE) {

		goto loop;
	}

	srv_error_monitor_active = FALSE;
1854

1855 1856 1857
	/* 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
1858 1859
	os_thread_exit(NULL);

1860
#ifndef __WIN__
1861
	return(NULL);
1862 1863 1864 1865 1866
#else
	return(0);
#endif
}

1867
/***********************************************************************
1868
Tells the InnoDB server that there has been activity in the database
1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889
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
1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905
/***********************************************************************
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);
}

1906 1907 1908
/*************************************************************************
The master thread controlling the server. */

1909 1910 1911
#ifndef __WIN__
void*
#else
1912
ulint
1913
#endif
1914 1915 1916
srv_master_thread(
/*==============*/
			/* out: a dummy parameter */
1917 1918
	void*	arg __attribute__((unused)))
			/* in: a dummy parameter required by
1919 1920 1921
			os_thread_create */
{
	os_event_t	event;
1922 1923
	time_t          last_flush_time;
	time_t          current_time;
1924 1925 1926 1927 1928
	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
1929
	ulint		n_tables_to_drop;
1930 1931 1932 1933
	ulint		n_ios;
	ulint		n_ios_old;
	ulint		n_ios_very_old;
	ulint		n_pend_ios;
unknown's avatar
unknown committed
1934
	ibool		skip_sleep	= FALSE;
1935 1936
	ulint		i;
	
1937
#ifdef UNIV_DEBUG_THREAD_CREATION
1938
	fprintf(stderr, "Master thread starts, id %lu\n",
unknown's avatar
unknown committed
1939
			      os_thread_pf(os_thread_get_curr_id()));
1940
#endif
unknown's avatar
unknown committed
1941 1942 1943
	srv_main_thread_process_no = os_proc_get_number();
	srv_main_thread_id = os_thread_pf(os_thread_get_curr_id());
	
1944 1945 1946 1947 1948 1949 1950 1951 1952 1953
	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:
1954 1955 1956 1957
	/*****************************************************************/
	/* ---- When there is database activity by users, we cycle in this
	loop */

unknown's avatar
unknown committed
1958
	srv_main_thread_op_info = (char*) "reserving kernel mutex";
1959 1960 1961

	n_ios_very_old = log_sys->n_log_ios + buf_pool->n_pages_read
						+ buf_pool->n_pages_written;
1962 1963
	mutex_enter(&kernel_mutex);

1964
	/* Store the user activity counter at the start of this loop */
1965 1966 1967 1968
	old_activity_count = srv_activity_count;

	mutex_exit(&kernel_mutex);

1969 1970 1971 1972 1973 1974 1975 1976
	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
1977
	skip_sleep = FALSE;
1978 1979

	for (i = 0; i < 10; i++) {
1980 1981
		n_ios_old = log_sys->n_log_ios + buf_pool->n_pages_read
						+ buf_pool->n_pages_written;
unknown's avatar
unknown committed
1982
		srv_main_thread_op_info = (char*)"sleeping";
unknown's avatar
unknown committed
1983 1984 1985 1986 1987 1988 1989
		
		if (!skip_sleep) {

		        os_thread_sleep(1000000);
		}

		skip_sleep = FALSE;
1990

unknown's avatar
unknown committed
1991 1992 1993 1994
		/* 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
1995 1996
		srv_main_thread_op_info =
					(char*)"doing background drop tables";
unknown's avatar
unknown committed
1997 1998 1999

		row_drop_tables_for_mysql_in_background();

unknown's avatar
unknown committed
2000
		srv_main_thread_op_info = (char*)"";
unknown's avatar
unknown committed
2001

unknown's avatar
unknown committed
2002
		if (srv_fast_shutdown && srv_shutdown_state > 0) {
2003

unknown's avatar
unknown committed
2004
			goto background_loop;
2005 2006
		}

unknown's avatar
unknown committed
2007 2008 2009 2010
		/* 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
2011
		srv_main_thread_op_info = (char*)"flushing log";
unknown's avatar
unknown committed
2012
		log_buffer_flush_to_disk();
unknown's avatar
unknown committed
2013

unknown's avatar
unknown committed
2014 2015 2016
		srv_main_thread_op_info = (char*)"making checkpoint";
		log_free_check();

2017
		/* If there were less than 5 i/os during the
2018 2019 2020 2021 2022 2023 2024 2025
		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;
2026
		if (n_pend_ios < 3 && (n_ios - n_ios_old < 5)) {
unknown's avatar
unknown committed
2027
			srv_main_thread_op_info =
unknown's avatar
unknown committed
2028
					(char*)"doing insert buffer merge";
2029 2030
			ibuf_contract_for_n_pages(TRUE, 5);

unknown's avatar
unknown committed
2031 2032 2033
			srv_main_thread_op_info = (char*)"flushing log";

			log_buffer_flush_to_disk();
2034
		}
unknown's avatar
Merge  
unknown committed
2035

2036 2037 2038 2039 2040 2041 2042 2043
		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
2044 2045 2046 2047 2048 2049 2050

		        /* 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
2051 2052
		}

2053 2054
		if (srv_activity_count == old_activity_count) {

2055 2056
			/* There is no user activity at the moment, go to
			the background loop */
2057 2058 2059 2060 2061

			goto background_loop;
		}
	}

2062 2063
	/* ---- We perform the following code approximately once per
	10 seconds when there is database activity */
2064

unknown's avatar
unknown committed
2065 2066 2067 2068 2069
#ifdef MEM_PERIODIC_CHECK
	/* Check magic numbers of every allocated mem block once in 10
	seconds */
	mem_validate_all_blocks();
#endif	
2070 2071
	/* 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
2072
	makes sense to flush 100 pages. */
2073 2074 2075 2076 2077 2078

	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
2079
		srv_main_thread_op_info = (char*) "flushing buffer pool pages";
unknown's avatar
unknown committed
2080
		buf_flush_batch(BUF_FLUSH_LIST, 100, ut_dulint_max);
2081

unknown's avatar
unknown committed
2082
		srv_main_thread_op_info = (char*) "flushing log";
unknown's avatar
unknown committed
2083
		log_buffer_flush_to_disk();
2084 2085 2086 2087 2088
	}

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

unknown's avatar
unknown committed
2089
	srv_main_thread_op_info = (char*)"doing insert buffer merge";
2090 2091
	ibuf_contract_for_n_pages(TRUE, 5);

unknown's avatar
unknown committed
2092
	srv_main_thread_op_info = (char*)"flushing log";
unknown's avatar
unknown committed
2093
	log_buffer_flush_to_disk();
2094 2095 2096 2097

	/* We run a full purge every 10 seconds, even if the server
	were active */
	
2098 2099
	n_pages_purged = 1;

unknown's avatar
unknown committed
2100 2101
	last_flush_time = time(NULL);

2102
	while (n_pages_purged) {
unknown's avatar
unknown committed
2103

unknown's avatar
Merge  
unknown committed
2104 2105 2106 2107 2108
		if (srv_fast_shutdown && srv_shutdown_state > 0) {

			goto background_loop;
		}

unknown's avatar
unknown committed
2109
		srv_main_thread_op_info = (char*)"purging";
2110
		n_pages_purged = trx_purge();
2111

unknown's avatar
unknown committed
2112 2113 2114
		current_time = time(NULL);

		if (difftime(current_time, last_flush_time) > 1) {
unknown's avatar
unknown committed
2115
			srv_main_thread_op_info = (char*) "flushing log";
2116

unknown's avatar
unknown committed
2117
		        log_buffer_flush_to_disk();
unknown's avatar
unknown committed
2118 2119
			last_flush_time = current_time;
		}
2120
	}
unknown's avatar
unknown committed
2121
	
unknown's avatar
unknown committed
2122
	srv_main_thread_op_info = (char*)"flushing buffer pool pages";
2123

2124
	/* Flush a few oldest pages to make a new checkpoint younger */
unknown's avatar
unknown committed
2125

2126
	if (buf_get_modified_ratio_pct() > 70) {
unknown's avatar
unknown committed
2127

2128 2129 2130
		/* 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
2131

unknown's avatar
unknown committed
2132 2133 2134
	        n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, 100,
							ut_dulint_max);
	} else {
2135 2136 2137
	        /* 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
2138

unknown's avatar
unknown committed
2139 2140 2141
	        n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, 10,
							ut_dulint_max);
	}
unknown's avatar
unknown committed
2142

unknown's avatar
unknown committed
2143
	srv_main_thread_op_info = (char*)"making checkpoint";
unknown's avatar
unknown committed
2144 2145 2146 2147

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

	log_checkpoint(TRUE, FALSE);
2148

unknown's avatar
unknown committed
2149
	srv_main_thread_op_info = (char*)"reserving kernel mutex";
2150

2151
	mutex_enter(&kernel_mutex);
2152 2153 2154 2155
	
	/* ---- When there is database activity, we jump from here back to
	the start of loop */

2156 2157 2158 2159
	if (srv_activity_count != old_activity_count) {
		mutex_exit(&kernel_mutex);
		goto loop;
	}
2160
	
2161 2162
	mutex_exit(&kernel_mutex);

2163 2164 2165 2166 2167 2168 2169
	/* 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 */

2170 2171 2172
	/* The server has been quiet for a while: start running background
	operations */
		
2173 2174 2175 2176
	srv_main_thread_op_info = (char*)"doing background drop tables";

	n_tables_to_drop = row_drop_tables_for_mysql_in_background();

2177 2178 2179 2180 2181 2182 2183 2184 2185
	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
2186
	srv_main_thread_op_info = (char*)"purging";
2187

unknown's avatar
unknown committed
2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210
	/* 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
2211
	}
2212

unknown's avatar
unknown committed
2213
	srv_main_thread_op_info = (char*)"reserving kernel mutex";
2214

2215 2216 2217 2218 2219 2220 2221
	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
2222
	srv_main_thread_op_info = (char*)"doing insert buffer merge";
unknown's avatar
unknown committed
2223 2224 2225 2226 2227 2228

	if (srv_fast_shutdown && srv_shutdown_state > 0) {
	        n_bytes_merged = 0;
	} else {
	        n_bytes_merged = ibuf_contract_for_n_pages(TRUE, 20);
	}
2229

unknown's avatar
unknown committed
2230
	srv_main_thread_op_info = (char*)"reserving kernel mutex";
2231

2232 2233 2234 2235 2236 2237 2238
	mutex_enter(&kernel_mutex);
	if (srv_activity_count != old_activity_count) {
		mutex_exit(&kernel_mutex);
		goto loop;
	}
	mutex_exit(&kernel_mutex);
	
2239
flush_loop:
unknown's avatar
unknown committed
2240
	srv_main_thread_op_info = (char*)"flushing buffer pool pages";
2241
	n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, 100, ut_dulint_max);
2242

unknown's avatar
unknown committed
2243
	srv_main_thread_op_info = (char*)"reserving kernel mutex";
2244

2245 2246 2247 2248 2249 2250 2251
	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
2252 2253
	srv_main_thread_op_info =
			(char*) "waiting for buffer pool flush to end";
2254 2255
	buf_flush_wait_batch_end(BUF_FLUSH_LIST);

unknown's avatar
unknown committed
2256 2257 2258 2259
	srv_main_thread_op_info = (char*) "flushing log";

	log_buffer_flush_to_disk();

unknown's avatar
unknown committed
2260
	srv_main_thread_op_info = (char*)"making checkpoint";
2261

2262 2263
	log_checkpoint(TRUE, FALSE);

2264 2265 2266 2267 2268 2269 2270 2271
	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
2272
	srv_main_thread_op_info = (char*)"reserving kernel mutex";
2273

2274 2275 2276 2277 2278 2279
	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
2280
/*
unknown's avatar
unknown committed
2281
	srv_main_thread_op_info =
unknown's avatar
unknown committed
2282
				(char*)"archiving log (if log archive is on)";
2283 2284
	
	log_archive_do(FALSE, &n_bytes_archived);
unknown's avatar
unknown committed
2285 2286
*/
	n_bytes_archived = 0;
2287

2288 2289
	/* Keep looping in the background loop if still work to do */

unknown's avatar
Merge  
unknown committed
2290
	if (srv_fast_shutdown && srv_shutdown_state > 0) {
unknown's avatar
unknown committed
2291 2292
		if (n_tables_to_drop + n_pages_flushed
				+ n_bytes_archived != 0) {
unknown's avatar
Merge  
unknown committed
2293

2294 2295 2296 2297
			/* 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
2298 2299
			goto background_loop;
		}
unknown's avatar
unknown committed
2300
	} else if (n_tables_to_drop +
2301
		   n_pages_purged + n_bytes_merged + n_pages_flushed
2302
						+ n_bytes_archived != 0) {
2303 2304 2305
		/* In a 'slow' shutdown we run purge and the insert buffer
		merge to completion */

2306 2307 2308 2309 2310 2311
		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
2312
suspend_thread:
unknown's avatar
unknown committed
2313
	srv_main_thread_op_info = (char*)"suspending";
2314

2315 2316
	mutex_enter(&kernel_mutex);

unknown's avatar
unknown committed
2317 2318 2319 2320 2321 2322
	if (row_get_background_drop_list_len_low() > 0) {
		mutex_exit(&kernel_mutex);

		goto loop;
	}

2323 2324 2325 2326
	event = srv_suspend_thread();

	mutex_exit(&kernel_mutex);

unknown's avatar
unknown committed
2327
	srv_main_thread_op_info = (char*)"waiting for server activity";
2328

2329 2330
	os_event_wait(event);

unknown's avatar
unknown committed
2331 2332 2333 2334 2335 2336 2337
	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);
	}

2338 2339 2340
	/* When there is user activity, InnoDB will set the event and the main
	thread goes back to loop: */

2341 2342
	goto loop;

2343 2344 2345 2346 2347 2348 2349
	/* 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);

2350
#ifndef __WIN__
2351
        return(NULL);				/* Not reached */
2352
#else
2353
	return(0);
2354
#endif
2355
}