granicus.if.org Git - postgresql/blob - src/include/pgstat.h

   1 /* ----------
   2  *      pgstat.h
   3  *
   4  *      Definitions for the PostgreSQL statistics collector daemon.
   5  *
   6  *      Copyright (c) 2001-2007, PostgreSQL Global Development Group
   7  *
   8  *      $PostgreSQL: pgsql/src/include/pgstat.h,v 1.64 2007/07/08 22:23:16 tgl Exp $
   9  * ----------
  10  */
  11 #ifndef PGSTAT_H
  12 #define PGSTAT_H
  13
  14 #include "libpq/pqcomm.h"
  15 #include "utils/hsearch.h"
  16 #include "utils/rel.h"
  17 #include "utils/timestamp.h"
  18
  19
  20 /* ----------
  21  * The types of backend -> collector messages
  22  * ----------
  23  */
  24 typedef enum StatMsgType
  25 {
  26         PGSTAT_MTYPE_DUMMY,
  27         PGSTAT_MTYPE_TABSTAT,
  28         PGSTAT_MTYPE_TABPURGE,
  29         PGSTAT_MTYPE_DROPDB,
  30         PGSTAT_MTYPE_RESETCOUNTER,
  31         PGSTAT_MTYPE_AUTOVAC_START,
  32         PGSTAT_MTYPE_VACUUM,
  33         PGSTAT_MTYPE_ANALYZE,
  34         PGSTAT_MTYPE_BGWRITER
  35 } StatMsgType;
  36
  37 /* ----------
  38  * The data type used for counters.
  39  * ----------
  40  */
  41 typedef int64 PgStat_Counter;
  42
  43 /* ----------
  44  * PgStat_TableCounts                   The actual per-table counts kept by a backend
  45  *
  46  * This struct should contain only actual event counters, because we memcmp
  47  * it against zeroes to detect whether there are any counts to transmit.
  48  * It is a component of PgStat_TableStatus (within-backend state) and
  49  * PgStat_TableEntry (the transmitted message format).
  50  *
  51  * Note: for a table, tuples_returned is the number of tuples successfully
  52  * fetched by heap_getnext, while tuples_fetched is the number of tuples
  53  * successfully fetched by heap_fetch under the control of bitmap indexscans.
  54  * For an index, tuples_returned is the number of index entries returned by
  55  * the index AM, while tuples_fetched is the number of tuples successfully
  56  * fetched by heap_fetch under the control of simple indexscans for this index.
  57  *
  58  * tuples_inserted/tuples_updated/tuples_deleted count attempted actions,
  59  * regardless of whether the transaction committed.  new_live_tuples and
  60  * new_dead_tuples are properly adjusted depending on commit or abort.
  61  * Note that new_live_tuples can be negative!
  62  * ----------
  63  */
  64 typedef struct PgStat_TableCounts
  65 {
  66         PgStat_Counter t_numscans;
  67
  68         PgStat_Counter t_tuples_returned;
  69         PgStat_Counter t_tuples_fetched;
  70
  71         PgStat_Counter t_tuples_inserted;
  72         PgStat_Counter t_tuples_updated;
  73         PgStat_Counter t_tuples_deleted;
  74
  75         PgStat_Counter t_new_live_tuples;
  76         PgStat_Counter t_new_dead_tuples;
  77
  78         PgStat_Counter t_blocks_fetched;
  79         PgStat_Counter t_blocks_hit;
  80 } PgStat_TableCounts;
  81
  82
  83 /* ------------------------------------------------------------
  84  * Structures kept in backend local memory while accumulating counts
  85  * ------------------------------------------------------------
  86  */
  87
  88
  89 /* ----------
  90  * PgStat_TableStatus                   Per-table status within a backend
  91  *
  92  * Most of the event counters are nontransactional, ie, we count events
  93  * in committed and aborted transactions alike.  For these, we just count
  94  * directly in the PgStat_TableStatus.  However, new_live_tuples and
  95  * new_dead_tuples must be derived from tuple insertion and deletion counts
  96  * with awareness of whether the transaction or subtransaction committed or
  97  * aborted.  Hence, we also keep a stack of per-(sub)transaction status
  98  * records for every table modified in the current transaction.  At commit
  99  * or abort, we propagate tuples_inserted and tuples_deleted up to the
 100  * parent subtransaction level, or out to the parent PgStat_TableStatus,
 101  * as appropriate.
 102  * ----------
 103  */
 104 typedef struct PgStat_TableStatus
 105 {
 106         Oid                     t_id;                           /* table's OID */
 107         bool            t_shared;                       /* is it a shared catalog? */
 108         struct PgStat_TableXactStatus *trans;   /* lowest subxact's counts */
 109         PgStat_TableCounts t_counts;    /* event counts to be sent */
 110 } PgStat_TableStatus;
 111
 112 /* ----------
 113  * PgStat_TableXactStatus               Per-table, per-subtransaction status
 114  * ----------
 115  */
 116 typedef struct PgStat_TableXactStatus
 117 {
 118         PgStat_Counter tuples_inserted; /* tuples inserted in (sub)xact */
 119         PgStat_Counter tuples_deleted;  /* tuples deleted in (sub)xact */
 120         int                     nest_level;                     /* subtransaction nest level */
 121         /* links to other structs for same relation: */
 122         struct PgStat_TableXactStatus *upper;   /* next higher subxact if any */
 123         PgStat_TableStatus *parent;                             /* per-table status */
 124         /* structs of same subxact level are linked here: */
 125         struct PgStat_TableXactStatus *next;    /* next of same subxact */
 126 } PgStat_TableXactStatus;
 127
 128
 129 /* ------------------------------------------------------------
 130  * Message formats follow
 131  * ------------------------------------------------------------
 132  */
 133
 134
 135 /* ----------
 136  * PgStat_MsgHdr                                The common message header
 137  * ----------
 138  */
 139 typedef struct PgStat_MsgHdr
 140 {
 141         StatMsgType m_type;
 142         int                     m_size;
 143 } PgStat_MsgHdr;
 144
 145 /* ----------
 146  * Space available in a message.  This will keep the UDP packets below 1K,
 147  * which should fit unfragmented into the MTU of the lo interface on most
 148  * platforms. Does anybody care for platforms where it doesn't?
 149  * ----------
 150  */
 151 #define PGSTAT_MSG_PAYLOAD      (1000 - sizeof(PgStat_MsgHdr))
 152
 153
 154 /* ----------
 155  * PgStat_MsgDummy                              A dummy message, ignored by the collector
 156  * ----------
 157  */
 158 typedef struct PgStat_MsgDummy
 159 {
 160         PgStat_MsgHdr m_hdr;
 161 } PgStat_MsgDummy;
 162
 163
 164 /* ----------
 165  * PgStat_TableEntry                    Per-table info in a MsgTabstat
 166  * ----------
 167  */
 168 typedef struct PgStat_TableEntry
 169 {
 170         Oid                     t_id;
 171         PgStat_TableCounts t_counts;
 172 } PgStat_TableEntry;
 173
 174 /* ----------
 175  * PgStat_MsgTabstat                    Sent by the backend to report table
 176  *                                                              and buffer access statistics.
 177  * ----------
 178  */
 179 #define PGSTAT_NUM_TABENTRIES  \
 180         ((PGSTAT_MSG_PAYLOAD - sizeof(Oid) - 3 * sizeof(int))  \
 181          / sizeof(PgStat_TableEntry))
 182
 183 typedef struct PgStat_MsgTabstat
 184 {
 185         PgStat_MsgHdr m_hdr;
 186         Oid                     m_databaseid;
 187         int                     m_nentries;
 188         int                     m_xact_commit;
 189         int                     m_xact_rollback;
 190         PgStat_TableEntry m_entry[PGSTAT_NUM_TABENTRIES];
 191 } PgStat_MsgTabstat;
 192
 193
 194 /* ----------
 195  * PgStat_MsgTabpurge                   Sent by the backend to tell the collector
 196  *                                                              about dead tables.
 197  * ----------
 198  */
 199 #define PGSTAT_NUM_TABPURGE  \
 200         ((PGSTAT_MSG_PAYLOAD - sizeof(Oid) - sizeof(int))  \
 201          / sizeof(Oid))
 202
 203 typedef struct PgStat_MsgTabpurge
 204 {
 205         PgStat_MsgHdr m_hdr;
 206         Oid                     m_databaseid;
 207         int                     m_nentries;
 208         Oid                     m_tableid[PGSTAT_NUM_TABPURGE];
 209 } PgStat_MsgTabpurge;
 210
 211
 212 /* ----------
 213  * PgStat_MsgDropdb                             Sent by the backend to tell the collector
 214  *                                                              about a dropped database
 215  * ----------
 216  */
 217 typedef struct PgStat_MsgDropdb
 218 {
 219         PgStat_MsgHdr m_hdr;
 220         Oid                     m_databaseid;
 221 } PgStat_MsgDropdb;
 222
 223
 224 /* ----------
 225  * PgStat_MsgResetcounter               Sent by the backend to tell the collector
 226  *                                                              to reset counters
 227  * ----------
 228  */
 229 typedef struct PgStat_MsgResetcounter
 230 {
 231         PgStat_MsgHdr m_hdr;
 232         Oid                     m_databaseid;
 233 } PgStat_MsgResetcounter;
 234
 235
 236 /* ----------
 237  * PgStat_MsgAutovacStart               Sent by the autovacuum daemon to signal
 238  *                                                              that a database is going to be processed
 239  * ----------
 240  */
 241 typedef struct PgStat_MsgAutovacStart
 242 {
 243         PgStat_MsgHdr m_hdr;
 244         Oid                     m_databaseid;
 245         TimestampTz m_start_time;
 246 } PgStat_MsgAutovacStart;
 247
 248
 249 /* ----------
 250  * PgStat_MsgVacuum                             Sent by the backend or autovacuum daemon
 251  *                                                              after VACUUM or VACUUM ANALYZE
 252  * ----------
 253  */
 254 typedef struct PgStat_MsgVacuum
 255 {
 256         PgStat_MsgHdr m_hdr;
 257         Oid                     m_databaseid;
 258         Oid                     m_tableoid;
 259         bool            m_analyze;
 260         bool            m_autovacuum;
 261         TimestampTz m_vacuumtime;
 262         PgStat_Counter m_tuples;
 263 } PgStat_MsgVacuum;
 264
 265
 266 /* ----------
 267  * PgStat_MsgAnalyze                    Sent by the backend or autovacuum daemon
 268  *                                                              after ANALYZE
 269  * ----------
 270  */
 271 typedef struct PgStat_MsgAnalyze
 272 {
 273         PgStat_MsgHdr m_hdr;
 274         Oid                     m_databaseid;
 275         Oid                     m_tableoid;
 276         bool            m_autovacuum;
 277         TimestampTz m_analyzetime;
 278         PgStat_Counter m_live_tuples;
 279         PgStat_Counter m_dead_tuples;
 280 } PgStat_MsgAnalyze;
 281
 282
 283 /* ----------
 284  * PgStat_MsgBgWriter           Sent by the bgwriter to update statistics.
 285  * ----------
 286  */
 287 typedef struct PgStat_MsgBgWriter
 288 {
 289         PgStat_MsgHdr m_hdr;
 290
 291         PgStat_Counter  m_timed_checkpoints;
 292         PgStat_Counter  m_requested_checkpoints;
 293         PgStat_Counter  m_buf_written_checkpoints;
 294         PgStat_Counter  m_buf_written_clean;
 295         PgStat_Counter  m_maxwritten_clean;
 296 } PgStat_MsgBgWriter;
 297
 298
 299 /* ----------
 300  * PgStat_Msg                                   Union over all possible messages.
 301  * ----------
 302  */
 303 typedef union PgStat_Msg
 304 {
 305         PgStat_MsgHdr msg_hdr;
 306         PgStat_MsgDummy msg_dummy;
 307         PgStat_MsgTabstat msg_tabstat;
 308         PgStat_MsgTabpurge msg_tabpurge;
 309         PgStat_MsgDropdb msg_dropdb;
 310         PgStat_MsgResetcounter msg_resetcounter;
 311         PgStat_MsgAutovacStart msg_autovacuum;
 312         PgStat_MsgVacuum msg_vacuum;
 313         PgStat_MsgAnalyze msg_analyze;
 314         PgStat_MsgBgWriter msg_bgwriter;
 315 } PgStat_Msg;
 316
 317
 318 /* ------------------------------------------------------------
 319  * Statistic collector data structures follow
 320  *
 321  * PGSTAT_FILE_FORMAT_ID should be changed whenever any of these
 322  * data structures change.
 323  * ------------------------------------------------------------
 324  */
 325
 326 #define PGSTAT_FILE_FORMAT_ID   0x01A5BC96
 327
 328 /* ----------
 329  * PgStat_StatDBEntry                   The collector's data per database
 330  * ----------
 331  */
 332 typedef struct PgStat_StatDBEntry
 333 {
 334         Oid                     databaseid;
 335         PgStat_Counter n_xact_commit;
 336         PgStat_Counter n_xact_rollback;
 337         PgStat_Counter n_blocks_fetched;
 338         PgStat_Counter n_blocks_hit;
 339         PgStat_Counter n_tuples_returned;
 340         PgStat_Counter n_tuples_fetched;
 341         PgStat_Counter n_tuples_inserted;
 342         PgStat_Counter n_tuples_updated;
 343         PgStat_Counter n_tuples_deleted;
 344         TimestampTz last_autovac_time;
 345
 346         /*
 347          * tables must be last in the struct, because we don't write the pointer
 348          * out to the stats file.
 349          */
 350         HTAB       *tables;
 351 } PgStat_StatDBEntry;
 352
 353
 354 /* ----------
 355  * PgStat_StatTabEntry                  The collector's data per table (or index)
 356  * ----------
 357  */
 358 typedef struct PgStat_StatTabEntry
 359 {
 360         Oid                     tableid;
 361
 362         PgStat_Counter numscans;
 363
 364         PgStat_Counter tuples_returned;
 365         PgStat_Counter tuples_fetched;
 366
 367         PgStat_Counter tuples_inserted;
 368         PgStat_Counter tuples_updated;
 369         PgStat_Counter tuples_deleted;
 370
 371         PgStat_Counter n_live_tuples;
 372         PgStat_Counter n_dead_tuples;
 373         PgStat_Counter last_anl_tuples;
 374
 375         PgStat_Counter blocks_fetched;
 376         PgStat_Counter blocks_hit;
 377
 378         TimestampTz vacuum_timestamp;           /* user initiated vacuum */
 379         TimestampTz autovac_vacuum_timestamp;           /* autovacuum initiated */
 380         TimestampTz analyze_timestamp;          /* user initiated */
 381         TimestampTz autovac_analyze_timestamp;          /* autovacuum initiated */
 382 } PgStat_StatTabEntry;
 383
 384
 385 /*
 386  * Global statistics kept in the stats collector
 387  */
 388 typedef struct PgStat_GlobalStats
 389 {
 390         PgStat_Counter  timed_checkpoints;
 391         PgStat_Counter  requested_checkpoints;
 392         PgStat_Counter  buf_written_checkpoints;
 393         PgStat_Counter  buf_written_clean;
 394         PgStat_Counter  maxwritten_clean;
 395 } PgStat_GlobalStats;
 396
 397
 398 /* ----------
 399  * Shared-memory data structures
 400  * ----------
 401  */
 402
 403 /* Max length of st_activity string ... perhaps replace with a GUC var? */
 404 #define PGBE_ACTIVITY_SIZE      1024
 405
 406 /* ----------
 407  * PgBackendStatus
 408  *
 409  * Each live backend maintains a PgBackendStatus struct in shared memory
 410  * showing its current activity.  (The structs are allocated according to
 411  * BackendId, but that is not critical.)  Note that the collector process
 412  * has no involvement in, or even access to, these structs.
 413  * ----------
 414  */
 415 typedef struct PgBackendStatus
 416 {
 417         /*
 418          * To avoid locking overhead, we use the following protocol: a backend
 419          * increments st_changecount before modifying its entry, and again after
 420          * finishing a modification.  A would-be reader should note the value of
 421          * st_changecount, copy the entry into private memory, then check
 422          * st_changecount again.  If the value hasn't changed, and if it's even,
 423          * the copy is valid; otherwise start over.  This makes updates cheap
 424          * while reads are potentially expensive, but that's the tradeoff we want.
 425          */
 426         int                     st_changecount;
 427
 428         /* The entry is valid iff st_procpid > 0, unused if st_procpid == 0 */
 429         int                     st_procpid;
 430
 431         /* Times when current backend, transaction, and activity started */
 432         TimestampTz st_proc_start_timestamp;
 433         TimestampTz st_txn_start_timestamp;
 434         TimestampTz st_activity_start_timestamp;
 435
 436         /* Database OID, owning user's OID, connection client address */
 437         Oid                     st_databaseid;
 438         Oid                     st_userid;
 439         SockAddr        st_clientaddr;
 440
 441         /* Is backend currently waiting on an lmgr lock? */
 442         bool            st_waiting;
 443
 444         /* current command string; MUST be null-terminated */
 445         char            st_activity[PGBE_ACTIVITY_SIZE];
 446 } PgBackendStatus;
 447
 448
 449 /* ----------
 450  * GUC parameters
 451  * ----------
 452  */
 453 extern bool pgstat_collect_startcollector;
 454 extern bool pgstat_collect_resetonpmstart;
 455 extern bool pgstat_collect_tuplelevel;
 456 extern bool pgstat_collect_blocklevel;
 457 extern bool pgstat_collect_querystring;
 458
 459 /*
 460  * BgWriter statistics counters are updated directly by bgwriter and bufmgr
 461  */
 462 extern PgStat_MsgBgWriter BgWriterStats;
 463
 464 /* ----------
 465  * Functions called from postmaster
 466  * ----------
 467  */
 468 extern Size BackendStatusShmemSize(void);
 469 extern void CreateSharedBackendStatus(void);
 470
 471 extern void pgstat_init(void);
 472 extern int      pgstat_start(void);
 473 extern void pgstat_reset_all(void);
 474 extern void allow_immediate_pgstat_restart(void);
 475 #ifdef EXEC_BACKEND
 476 extern void PgstatCollectorMain(int argc, char *argv[]);
 477 #endif
 478
 479
 480 /* ----------
 481  * Functions called from backends
 482  * ----------
 483  */
 484 extern void pgstat_ping(void);
 485
 486 extern void pgstat_report_tabstat(bool force);
 487 extern void pgstat_vacuum_tabstat(void);
 488 extern void pgstat_drop_database(Oid databaseid);
 489
 490 extern void pgstat_clear_snapshot(void);
 491 extern void pgstat_reset_counters(void);
 492
 493 extern void pgstat_report_autovac(Oid dboid);
 494 extern void pgstat_report_vacuum(Oid tableoid, bool shared,
 495                                          bool analyze, PgStat_Counter tuples);
 496 extern void pgstat_report_analyze(Oid tableoid, bool shared,
 497                                           PgStat_Counter livetuples,
 498                                           PgStat_Counter deadtuples);
 499
 500 extern void pgstat_initialize(void);
 501 extern void pgstat_bestart(void);
 502
 503 extern void pgstat_report_activity(const char *what);
 504 extern void pgstat_report_txn_timestamp(TimestampTz tstamp);
 505 extern void pgstat_report_waiting(bool waiting);
 506
 507 extern void pgstat_initstats(Relation rel);
 508
 509 /* nontransactional event counts are simple enough to inline */
 510
 511 #define pgstat_count_heap_scan(rel)                                                                             \
 512         do {                                                                                                                            \
 513                 if (pgstat_collect_tuplelevel && (rel)->pgstat_info != NULL)    \
 514                         (rel)->pgstat_info->t_counts.t_numscans++;                                      \
 515         } while (0)
 516 #define pgstat_count_heap_getnext(rel)                                                                  \
 517         do {                                                                                                                            \
 518                 if (pgstat_collect_tuplelevel && (rel)->pgstat_info != NULL)    \
 519                         (rel)->pgstat_info->t_counts.t_tuples_returned++;                       \
 520         } while (0)
 521 #define pgstat_count_heap_fetch(rel)                                                                    \
 522         do {                                                                                                                            \
 523                 if (pgstat_collect_tuplelevel && (rel)->pgstat_info != NULL)    \
 524                         (rel)->pgstat_info->t_counts.t_tuples_fetched++;                        \
 525         } while (0)
 526 #define pgstat_count_index_scan(rel)                                                                    \
 527         do {                                                                                                                            \
 528                 if (pgstat_collect_tuplelevel && (rel)->pgstat_info != NULL)    \
 529                         (rel)->pgstat_info->t_counts.t_numscans++;                                      \
 530         } while (0)
 531 #define pgstat_count_index_tuples(rel, n)                                                               \
 532         do {                                                                                                                            \
 533                 if (pgstat_collect_tuplelevel && (rel)->pgstat_info != NULL)    \
 534                         (rel)->pgstat_info->t_counts.t_tuples_returned += (n);          \
 535         } while (0)
 536 #define pgstat_count_buffer_read(rel)                                                                   \
 537         do {                                                                                                                            \
 538                 if (pgstat_collect_blocklevel && (rel)->pgstat_info != NULL)    \
 539                         (rel)->pgstat_info->t_counts.t_blocks_fetched++;                        \
 540         } while (0)
 541 #define pgstat_count_buffer_hit(rel)                                                                    \
 542         do {                                                                                                                            \
 543                 if (pgstat_collect_blocklevel && (rel)->pgstat_info != NULL)    \
 544                         (rel)->pgstat_info->t_counts.t_blocks_hit++;                            \
 545         } while (0)
 546
 547 extern void pgstat_count_heap_insert(Relation rel);
 548 extern void pgstat_count_heap_update(Relation rel);
 549 extern void pgstat_count_heap_delete(Relation rel);
 550
 551 extern void AtEOXact_PgStat(bool isCommit);
 552 extern void AtEOSubXact_PgStat(bool isCommit, int nestDepth);
 553
 554 extern void AtPrepare_PgStat(void);
 555 extern void PostPrepare_PgStat(void);
 556
 557 extern void pgstat_twophase_postcommit(TransactionId xid, uint16 info,
 558                                                                            void *recdata, uint32 len);
 559 extern void pgstat_twophase_postabort(TransactionId xid, uint16 info,
 560                                                                           void *recdata, uint32 len);
 561
 562 extern void pgstat_send_bgwriter(void);
 563
 564 /* ----------
 565  * Support functions for the SQL-callable functions to
 566  * generate the pgstat* views.
 567  * ----------
 568  */
 569 extern PgStat_StatDBEntry *pgstat_fetch_stat_dbentry(Oid dbid);
 570 extern PgStat_StatTabEntry *pgstat_fetch_stat_tabentry(Oid relid);
 571 extern PgBackendStatus *pgstat_fetch_stat_beentry(int beid);
 572 extern int      pgstat_fetch_stat_numbackends(void);
 573 extern PgStat_GlobalStats *pgstat_fetch_global(void);
 574
 575 #endif   /* PGSTAT_H */