PostgreSQL Source Code  git master
pg_stat_statements.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * pg_stat_statements.c
4  * Track statement planning and execution times as well as resource
5  * usage across a whole database cluster.
6  *
7  * Execution costs are totalled for each distinct source query, and kept in
8  * a shared hashtable. (We track only as many distinct queries as will fit
9  * in the designated amount of shared memory.)
10  *
11  * As of Postgres 9.2, this module normalizes query entries. Normalization
12  * is a process whereby similar queries, typically differing only in their
13  * constants (though the exact rules are somewhat more subtle than that) are
14  * recognized as equivalent, and are tracked as a single entry. This is
15  * particularly useful for non-prepared queries.
16  *
17  * Normalization is implemented by fingerprinting queries, selectively
18  * serializing those fields of each query tree's nodes that are judged to be
19  * essential to the query. This is referred to as a query jumble. This is
20  * distinct from a regular serialization in that various extraneous
21  * information is ignored as irrelevant or not essential to the query, such
22  * as the collations of Vars and, most notably, the values of constants.
23  *
24  * This jumble is acquired at the end of parse analysis of each query, and
25  * a 64-bit hash of it is stored into the query's Query.queryId field.
26  * The server then copies this value around, making it available in plan
27  * tree(s) generated from the query. The executor can then use this value
28  * to blame query costs on the proper queryId.
29  *
30  * To facilitate presenting entries to users, we create "representative" query
31  * strings in which constants are replaced with parameter symbols ($n), to
32  * make it clearer what a normalized entry can represent. To save on shared
33  * memory, and to avoid having to truncate oversized query strings, we store
34  * these strings in a temporary external query-texts file. Offsets into this
35  * file are kept in shared memory.
36  *
37  * Note about locking issues: to create or delete an entry in the shared
38  * hashtable, one must hold pgss->lock exclusively. Modifying any field
39  * in an entry except the counters requires the same. To look up an entry,
40  * one must hold the lock shared. To read or update the counters within
41  * an entry, one must hold the lock shared or exclusive (so the entry doesn't
42  * disappear!) and also take the entry's mutex spinlock.
43  * The shared state variable pgss->extent (the next free spot in the external
44  * query-text file) should be accessed only while holding either the
45  * pgss->mutex spinlock, or exclusive lock on pgss->lock. We use the mutex to
46  * allow reserving file space while holding only shared lock on pgss->lock.
47  * Rewriting the entire external query-text file, eg for garbage collection,
48  * requires holding pgss->lock exclusively; this allows individual entries
49  * in the file to be read or written while holding only shared lock.
50  *
51  *
52  * Copyright (c) 2008-2020, PostgreSQL Global Development Group
53  *
54  * IDENTIFICATION
55  * contrib/pg_stat_statements/pg_stat_statements.c
56  *
57  *-------------------------------------------------------------------------
58  */
59 #include "postgres.h"
60 
61 #include <math.h>
62 #include <sys/stat.h>
63 #include <unistd.h>
64 
65 #include "catalog/pg_authid.h"
66 #include "common/hashfn.h"
67 #include "executor/instrument.h"
68 #include "funcapi.h"
69 #include "mb/pg_wchar.h"
70 #include "miscadmin.h"
71 #include "optimizer/planner.h"
72 #include "parser/analyze.h"
73 #include "parser/parsetree.h"
74 #include "parser/scanner.h"
75 #include "parser/scansup.h"
76 #include "pgstat.h"
77 #include "storage/fd.h"
78 #include "storage/ipc.h"
79 #include "storage/spin.h"
80 #include "tcop/utility.h"
81 #include "utils/acl.h"
82 #include "utils/builtins.h"
83 #include "utils/memutils.h"
84 
86 
87 /* Location of permanent stats file (valid when database is shut down) */
88 #define PGSS_DUMP_FILE PGSTAT_STAT_PERMANENT_DIRECTORY "/pg_stat_statements.stat"
89 
90 /*
91  * Location of external query text file. We don't keep it in the core
92  * system's stats_temp_directory. The core system can safely use that GUC
93  * setting, because the statistics collector temp file paths are set only once
94  * as part of changing the GUC, but pg_stat_statements has no way of avoiding
95  * race conditions. Besides, we only expect modest, infrequent I/O for query
96  * strings, so placing the file on a faster filesystem is not compelling.
97  */
98 #define PGSS_TEXT_FILE PG_STAT_TMP_DIR "/pgss_query_texts.stat"
99 
100 /* Magic number identifying the stats file format */
101 static const uint32 PGSS_FILE_HEADER = 0x20171004;
102 
103 /* PostgreSQL major version number, changes in which invalidate all entries */
104 static const uint32 PGSS_PG_MAJOR_VERSION = PG_VERSION_NUM / 100;
105 
106 /* XXX: Should USAGE_EXEC reflect execution time and/or buffer usage? */
107 #define USAGE_EXEC(duration) (1.0)
108 #define USAGE_INIT (1.0) /* including initial planning */
109 #define ASSUMED_MEDIAN_INIT (10.0) /* initial assumed median usage */
110 #define ASSUMED_LENGTH_INIT 1024 /* initial assumed mean query length */
111 #define USAGE_DECREASE_FACTOR (0.99) /* decreased every entry_dealloc */
112 #define STICKY_DECREASE_FACTOR (0.50) /* factor for sticky entries */
113 #define USAGE_DEALLOC_PERCENT 5 /* free this % of entries at once */
114 #define IS_STICKY(c) ((c.calls[PGSS_PLAN] + c.calls[PGSS_EXEC]) == 0)
115 
116 #define JUMBLE_SIZE 1024 /* query serialization buffer size */
117 
118 /*
119  * Extension version number, for supporting older extension versions' objects
120  */
121 typedef enum pgssVersion
122 {
128 } pgssVersion;
129 
130 typedef enum pgssStoreKind
131 {
133 
134  /*
135  * PGSS_PLAN and PGSS_EXEC must be respectively 0 and 1 as they're used to
136  * reference the underlying values in the arrays in the Counters struct,
137  * and this order is required in pg_stat_statements_internal().
138  */
141 
142  PGSS_NUMKIND /* Must be last value of this enum */
143 } pgssStoreKind;
144 
145 /*
146  * Hashtable key that defines the identity of a hashtable entry. We separate
147  * queries by user and by database even if they are otherwise identical.
148  *
149  * Right now, this structure contains no padding. If you add any, make sure
150  * to teach pgss_store() to zero the padding bytes. Otherwise, things will
151  * break, because pgss_hash is created using HASH_BLOBS, and thus tag_hash
152  * is used to hash this.
153  */
154 typedef struct pgssHashKey
155 {
156  Oid userid; /* user OID */
157  Oid dbid; /* database OID */
158  uint64 queryid; /* query identifier */
159 } pgssHashKey;
160 
161 /*
162  * The actual stats counters kept within pgssEntry.
163  */
164 typedef struct Counters
165 {
166  int64 calls[PGSS_NUMKIND]; /* # of times planned/executed */
167  double total_time[PGSS_NUMKIND]; /* total planning/execution time,
168  * in msec */
169  double min_time[PGSS_NUMKIND]; /* minimum planning/execution time in
170  * msec */
171  double max_time[PGSS_NUMKIND]; /* maximum planning/execution time in
172  * msec */
173  double mean_time[PGSS_NUMKIND]; /* mean planning/execution time in
174  * msec */
175  double sum_var_time[PGSS_NUMKIND]; /* sum of variances in
176  * planning/execution time in msec */
177  int64 rows; /* total # of retrieved or affected rows */
178  int64 shared_blks_hit; /* # of shared buffer hits */
179  int64 shared_blks_read; /* # of shared disk blocks read */
180  int64 shared_blks_dirtied; /* # of shared disk blocks dirtied */
181  int64 shared_blks_written; /* # of shared disk blocks written */
182  int64 local_blks_hit; /* # of local buffer hits */
183  int64 local_blks_read; /* # of local disk blocks read */
184  int64 local_blks_dirtied; /* # of local disk blocks dirtied */
185  int64 local_blks_written; /* # of local disk blocks written */
186  int64 temp_blks_read; /* # of temp blocks read */
187  int64 temp_blks_written; /* # of temp blocks written */
188  double blk_read_time; /* time spent reading, in msec */
189  double blk_write_time; /* time spent writing, in msec */
190  double usage; /* usage factor */
191  int64 wal_records; /* # of WAL records generated */
192  int64 wal_fpi; /* # of WAL full page images generated */
193  uint64 wal_bytes; /* total amount of WAL bytes generated */
194 } Counters;
195 
196 /*
197  * Statistics per statement
198  *
199  * Note: in event of a failure in garbage collection of the query text file,
200  * we reset query_offset to zero and query_len to -1. This will be seen as
201  * an invalid state by qtext_fetch().
202  */
203 typedef struct pgssEntry
204 {
205  pgssHashKey key; /* hash key of entry - MUST BE FIRST */
206  Counters counters; /* the statistics for this query */
207  Size query_offset; /* query text offset in external file */
208  int query_len; /* # of valid bytes in query string, or -1 */
209  int encoding; /* query text encoding */
210  slock_t mutex; /* protects the counters only */
211 } pgssEntry;
212 
213 /*
214  * Global shared state
215  */
216 typedef struct pgssSharedState
217 {
218  LWLock *lock; /* protects hashtable search/modification */
219  double cur_median_usage; /* current median usage in hashtable */
220  Size mean_query_len; /* current mean entry text length */
221  slock_t mutex; /* protects following fields only: */
222  Size extent; /* current extent of query file */
223  int n_writers; /* number of active writers to query file */
224  int gc_count; /* query file garbage collection cycle count */
226 
227 /*
228  * Struct for tracking locations/lengths of constants during normalization
229  */
230 typedef struct pgssLocationLen
231 {
232  int location; /* start offset in query text */
233  int length; /* length in bytes, or -1 to ignore */
235 
236 /*
237  * Working state for computing a query jumble and producing a normalized
238  * query string
239  */
240 typedef struct pgssJumbleState
241 {
242  /* Jumble of current query tree */
243  unsigned char *jumble;
244 
245  /* Number of bytes used in jumble[] */
247 
248  /* Array of locations of constants that should be removed */
250 
251  /* Allocated length of clocations array */
253 
254  /* Current number of valid entries in clocations array */
256 
257  /* highest Param id we've seen, in order to start normalization correctly */
260 
261 /*---- Local variables ----*/
262 
263 /* Current nesting depth of ExecutorRun+ProcessUtility calls */
264 static int exec_nested_level = 0;
265 
266 /* Current nesting depth of planner calls */
267 static int plan_nested_level = 0;
268 
269 /* Saved hook values in case of unload */
278 
279 /* Links to shared memory state */
280 static pgssSharedState *pgss = NULL;
281 static HTAB *pgss_hash = NULL;
282 
283 /*---- GUC variables ----*/
284 
285 typedef enum
286 {
287  PGSS_TRACK_NONE, /* track no statements */
288  PGSS_TRACK_TOP, /* only top level statements */
289  PGSS_TRACK_ALL /* all statements, including nested ones */
291 
292 static const struct config_enum_entry track_options[] =
293 {
294  {"none", PGSS_TRACK_NONE, false},
295  {"top", PGSS_TRACK_TOP, false},
296  {"all", PGSS_TRACK_ALL, false},
297  {NULL, 0, false}
298 };
299 
300 static int pgss_max; /* max # statements to track */
301 static int pgss_track; /* tracking level */
302 static bool pgss_track_utility; /* whether to track utility commands */
303 static bool pgss_track_planning; /* whether to track planning duration */
304 static bool pgss_save; /* whether to save stats across shutdown */
305 
306 
307 #define pgss_enabled(level) \
308  (pgss_track == PGSS_TRACK_ALL || \
309  (pgss_track == PGSS_TRACK_TOP && (level) == 0))
310 
311 #define record_gc_qtexts() \
312  do { \
313  volatile pgssSharedState *s = (volatile pgssSharedState *) pgss; \
314  SpinLockAcquire(&s->mutex); \
315  s->gc_count++; \
316  SpinLockRelease(&s->mutex); \
317  } while(0)
318 
319 /*---- Function declarations ----*/
320 
321 void _PG_init(void);
322 void _PG_fini(void);
323 
330 
331 static void pgss_shmem_startup(void);
332 static void pgss_shmem_shutdown(int code, Datum arg);
333 static void pgss_post_parse_analyze(ParseState *pstate, Query *query);
335  const char *query_string,
336  int cursorOptions,
337  ParamListInfo boundParams);
338 static void pgss_ExecutorStart(QueryDesc *queryDesc, int eflags);
339 static void pgss_ExecutorRun(QueryDesc *queryDesc,
340  ScanDirection direction,
341  uint64 count, bool execute_once);
342 static void pgss_ExecutorFinish(QueryDesc *queryDesc);
343 static void pgss_ExecutorEnd(QueryDesc *queryDesc);
344 static void pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
345  ProcessUtilityContext context, ParamListInfo params,
346  QueryEnvironment *queryEnv,
348 static uint64 pgss_hash_string(const char *str, int len);
349 static void pgss_store(const char *query, uint64 queryId,
350  int query_location, int query_len,
351  pgssStoreKind kind,
352  double total_time, uint64 rows,
353  const BufferUsage *bufusage,
354  const WalUsage *walusage,
355  pgssJumbleState *jstate);
357  pgssVersion api_version,
358  bool showtext);
359 static Size pgss_memsize(void);
360 static pgssEntry *entry_alloc(pgssHashKey *key, Size query_offset, int query_len,
361  int encoding, bool sticky);
362 static void entry_dealloc(void);
363 static bool qtext_store(const char *query, int query_len,
364  Size *query_offset, int *gc_count);
365 static char *qtext_load_file(Size *buffer_size);
366 static char *qtext_fetch(Size query_offset, int query_len,
367  char *buffer, Size buffer_size);
368 static bool need_gc_qtexts(void);
369 static void gc_qtexts(void);
370 static void entry_reset(Oid userid, Oid dbid, uint64 queryid);
371 static void AppendJumble(pgssJumbleState *jstate,
372  const unsigned char *item, Size size);
373 static void JumbleQuery(pgssJumbleState *jstate, Query *query);
374 static void JumbleRangeTable(pgssJumbleState *jstate, List *rtable);
375 static void JumbleRowMarks(pgssJumbleState *jstate, List *rowMarks);
376 static void JumbleExpr(pgssJumbleState *jstate, Node *node);
377 static void RecordConstLocation(pgssJumbleState *jstate, int location);
378 static char *generate_normalized_query(pgssJumbleState *jstate, const char *query,
379  int query_loc, int *query_len_p, int encoding);
380 static void fill_in_constant_lengths(pgssJumbleState *jstate, const char *query,
381  int query_loc);
382 static int comp_location(const void *a, const void *b);
383 
384 
385 /*
386  * Module load callback
387  */
388 void
389 _PG_init(void)
390 {
391  /*
392  * In order to create our shared memory area, we have to be loaded via
393  * shared_preload_libraries. If not, fall out without hooking into any of
394  * the main system. (We don't throw error here because it seems useful to
395  * allow the pg_stat_statements functions to be created even when the
396  * module isn't active. The functions must protect themselves against
397  * being called then, however.)
398  */
400  return;
401 
402  /*
403  * Define (or redefine) custom GUC variables.
404  */
405  DefineCustomIntVariable("pg_stat_statements.max",
406  "Sets the maximum number of statements tracked by pg_stat_statements.",
407  NULL,
408  &pgss_max,
409  5000,
410  100,
411  INT_MAX,
413  0,
414  NULL,
415  NULL,
416  NULL);
417 
418  DefineCustomEnumVariable("pg_stat_statements.track",
419  "Selects which statements are tracked by pg_stat_statements.",
420  NULL,
421  &pgss_track,
423  track_options,
424  PGC_SUSET,
425  0,
426  NULL,
427  NULL,
428  NULL);
429 
430  DefineCustomBoolVariable("pg_stat_statements.track_utility",
431  "Selects whether utility commands are tracked by pg_stat_statements.",
432  NULL,
434  true,
435  PGC_SUSET,
436  0,
437  NULL,
438  NULL,
439  NULL);
440 
441  DefineCustomBoolVariable("pg_stat_statements.track_planning",
442  "Selects whether planning duration is tracked by pg_stat_statements.",
443  NULL,
445  true,
446  PGC_SUSET,
447  0,
448  NULL,
449  NULL,
450  NULL);
451 
452  DefineCustomBoolVariable("pg_stat_statements.save",
453  "Save pg_stat_statements statistics across server shutdowns.",
454  NULL,
455  &pgss_save,
456  true,
457  PGC_SIGHUP,
458  0,
459  NULL,
460  NULL,
461  NULL);
462 
463  EmitWarningsOnPlaceholders("pg_stat_statements");
464 
465  /*
466  * Request additional shared resources. (These are no-ops if we're not in
467  * the postmaster process.) We'll allocate or attach to the shared
468  * resources in pgss_shmem_startup().
469  */
471  RequestNamedLWLockTranche("pg_stat_statements", 1);
472 
473  /*
474  * Install hooks.
475  */
492 }
493 
494 /*
495  * Module unload callback
496  */
497 void
498 _PG_fini(void)
499 {
500  /* Uninstall hooks. */
509 }
510 
511 /*
512  * shmem_startup hook: allocate or attach to shared memory,
513  * then load any pre-existing statistics from file.
514  * Also create and load the query-texts file, which is expected to exist
515  * (even if empty) while the module is enabled.
516  */
517 static void
519 {
520  bool found;
521  HASHCTL info;
522  FILE *file = NULL;
523  FILE *qfile = NULL;
524  uint32 header;
525  int32 num;
526  int32 pgver;
527  int32 i;
528  int buffer_size;
529  char *buffer = NULL;
530 
533 
534  /* reset in case this is a restart within the postmaster */
535  pgss = NULL;
536  pgss_hash = NULL;
537 
538  /*
539  * Create or attach to the shared memory state, including hash table
540  */
541  LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
542 
543  pgss = ShmemInitStruct("pg_stat_statements",
544  sizeof(pgssSharedState),
545  &found);
546 
547  if (!found)
548  {
549  /* First time through ... */
550  pgss->lock = &(GetNamedLWLockTranche("pg_stat_statements"))->lock;
553  SpinLockInit(&pgss->mutex);
554  pgss->extent = 0;
555  pgss->n_writers = 0;
556  pgss->gc_count = 0;
557  }
558 
559  memset(&info, 0, sizeof(info));
560  info.keysize = sizeof(pgssHashKey);
561  info.entrysize = sizeof(pgssEntry);
562  pgss_hash = ShmemInitHash("pg_stat_statements hash",
564  &info,
566 
567  LWLockRelease(AddinShmemInitLock);
568 
569  /*
570  * If we're in the postmaster (or a standalone backend...), set up a shmem
571  * exit hook to dump the statistics to disk.
572  */
573  if (!IsUnderPostmaster)
575 
576  /*
577  * Done if some other process already completed our initialization.
578  */
579  if (found)
580  return;
581 
582  /*
583  * Note: we don't bother with locks here, because there should be no other
584  * processes running when this code is reached.
585  */
586 
587  /* Unlink query text file possibly left over from crash */
588  unlink(PGSS_TEXT_FILE);
589 
590  /* Allocate new query text temp file */
592  if (qfile == NULL)
593  goto write_error;
594 
595  /*
596  * If we were told not to load old statistics, we're done. (Note we do
597  * not try to unlink any old dump file in this case. This seems a bit
598  * questionable but it's the historical behavior.)
599  */
600  if (!pgss_save)
601  {
602  FreeFile(qfile);
603  return;
604  }
605 
606  /*
607  * Attempt to load old statistics from the dump file.
608  */
610  if (file == NULL)
611  {
612  if (errno != ENOENT)
613  goto read_error;
614  /* No existing persisted stats file, so we're done */
615  FreeFile(qfile);
616  return;
617  }
618 
619  buffer_size = 2048;
620  buffer = (char *) palloc(buffer_size);
621 
622  if (fread(&header, sizeof(uint32), 1, file) != 1 ||
623  fread(&pgver, sizeof(uint32), 1, file) != 1 ||
624  fread(&num, sizeof(int32), 1, file) != 1)
625  goto read_error;
626 
627  if (header != PGSS_FILE_HEADER ||
628  pgver != PGSS_PG_MAJOR_VERSION)
629  goto data_error;
630 
631  for (i = 0; i < num; i++)
632  {
633  pgssEntry temp;
634  pgssEntry *entry;
635  Size query_offset;
636 
637  if (fread(&temp, sizeof(pgssEntry), 1, file) != 1)
638  goto read_error;
639 
640  /* Encoding is the only field we can easily sanity-check */
641  if (!PG_VALID_BE_ENCODING(temp.encoding))
642  goto data_error;
643 
644  /* Resize buffer as needed */
645  if (temp.query_len >= buffer_size)
646  {
647  buffer_size = Max(buffer_size * 2, temp.query_len + 1);
648  buffer = repalloc(buffer, buffer_size);
649  }
650 
651  if (fread(buffer, 1, temp.query_len + 1, file) != temp.query_len + 1)
652  goto read_error;
653 
654  /* Should have a trailing null, but let's make sure */
655  buffer[temp.query_len] = '\0';
656 
657  /* Skip loading "sticky" entries */
658  if (IS_STICKY(temp.counters))
659  continue;
660 
661  /* Store the query text */
662  query_offset = pgss->extent;
663  if (fwrite(buffer, 1, temp.query_len + 1, qfile) != temp.query_len + 1)
664  goto write_error;
665  pgss->extent += temp.query_len + 1;
666 
667  /* make the hashtable entry (discards old entries if too many) */
668  entry = entry_alloc(&temp.key, query_offset, temp.query_len,
669  temp.encoding,
670  false);
671 
672  /* copy in the actual stats */
673  entry->counters = temp.counters;
674  }
675 
676  pfree(buffer);
677  FreeFile(file);
678  FreeFile(qfile);
679 
680  /*
681  * Remove the persisted stats file so it's not included in
682  * backups/replication standbys, etc. A new file will be written on next
683  * shutdown.
684  *
685  * Note: it's okay if the PGSS_TEXT_FILE is included in a basebackup,
686  * because we remove that file on startup; it acts inversely to
687  * PGSS_DUMP_FILE, in that it is only supposed to be around when the
688  * server is running, whereas PGSS_DUMP_FILE is only supposed to be around
689  * when the server is not running. Leaving the file creates no danger of
690  * a newly restored database having a spurious record of execution costs,
691  * which is what we're really concerned about here.
692  */
693  unlink(PGSS_DUMP_FILE);
694 
695  return;
696 
697 read_error:
698  ereport(LOG,
700  errmsg("could not read file \"%s\": %m",
701  PGSS_DUMP_FILE)));
702  goto fail;
703 data_error:
704  ereport(LOG,
705  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
706  errmsg("ignoring invalid data in file \"%s\"",
707  PGSS_DUMP_FILE)));
708  goto fail;
709 write_error:
710  ereport(LOG,
712  errmsg("could not write file \"%s\": %m",
713  PGSS_TEXT_FILE)));
714 fail:
715  if (buffer)
716  pfree(buffer);
717  if (file)
718  FreeFile(file);
719  if (qfile)
720  FreeFile(qfile);
721  /* If possible, throw away the bogus file; ignore any error */
722  unlink(PGSS_DUMP_FILE);
723 
724  /*
725  * Don't unlink PGSS_TEXT_FILE here; it should always be around while the
726  * server is running with pg_stat_statements enabled
727  */
728 }
729 
730 /*
731  * shmem_shutdown hook: Dump statistics into file.
732  *
733  * Note: we don't bother with acquiring lock, because there should be no
734  * other processes running when this is called.
735  */
736 static void
738 {
739  FILE *file;
740  char *qbuffer = NULL;
741  Size qbuffer_size = 0;
742  HASH_SEQ_STATUS hash_seq;
743  int32 num_entries;
744  pgssEntry *entry;
745 
746  /* Don't try to dump during a crash. */
747  if (code)
748  return;
749 
750  /* Safety check ... shouldn't get here unless shmem is set up. */
751  if (!pgss || !pgss_hash)
752  return;
753 
754  /* Don't dump if told not to. */
755  if (!pgss_save)
756  return;
757 
758  file = AllocateFile(PGSS_DUMP_FILE ".tmp", PG_BINARY_W);
759  if (file == NULL)
760  goto error;
761 
762  if (fwrite(&PGSS_FILE_HEADER, sizeof(uint32), 1, file) != 1)
763  goto error;
764  if (fwrite(&PGSS_PG_MAJOR_VERSION, sizeof(uint32), 1, file) != 1)
765  goto error;
766  num_entries = hash_get_num_entries(pgss_hash);
767  if (fwrite(&num_entries, sizeof(int32), 1, file) != 1)
768  goto error;
769 
770  qbuffer = qtext_load_file(&qbuffer_size);
771  if (qbuffer == NULL)
772  goto error;
773 
774  /*
775  * When serializing to disk, we store query texts immediately after their
776  * entry data. Any orphaned query texts are thereby excluded.
777  */
778  hash_seq_init(&hash_seq, pgss_hash);
779  while ((entry = hash_seq_search(&hash_seq)) != NULL)
780  {
781  int len = entry->query_len;
782  char *qstr = qtext_fetch(entry->query_offset, len,
783  qbuffer, qbuffer_size);
784 
785  if (qstr == NULL)
786  continue; /* Ignore any entries with bogus texts */
787 
788  if (fwrite(entry, sizeof(pgssEntry), 1, file) != 1 ||
789  fwrite(qstr, 1, len + 1, file) != len + 1)
790  {
791  /* note: we assume hash_seq_term won't change errno */
792  hash_seq_term(&hash_seq);
793  goto error;
794  }
795  }
796 
797  free(qbuffer);
798  qbuffer = NULL;
799 
800  if (FreeFile(file))
801  {
802  file = NULL;
803  goto error;
804  }
805 
806  /*
807  * Rename file into place, so we atomically replace any old one.
808  */
810 
811  /* Unlink query-texts file; it's not needed while shutdown */
812  unlink(PGSS_TEXT_FILE);
813 
814  return;
815 
816 error:
817  ereport(LOG,
819  errmsg("could not write file \"%s\": %m",
820  PGSS_DUMP_FILE ".tmp")));
821  if (qbuffer)
822  free(qbuffer);
823  if (file)
824  FreeFile(file);
825  unlink(PGSS_DUMP_FILE ".tmp");
826  unlink(PGSS_TEXT_FILE);
827 }
828 
829 /*
830  * Post-parse-analysis hook: mark query with a queryId
831  */
832 static void
834 {
835  pgssJumbleState jstate;
836 
838  prev_post_parse_analyze_hook(pstate, query);
839 
840  /* Assert we didn't do this already */
841  Assert(query->queryId == UINT64CONST(0));
842 
843  /* Safety check... */
844  if (!pgss || !pgss_hash || !pgss_enabled(exec_nested_level))
845  return;
846 
847  /*
848  * Utility statements get queryId zero. We do this even in cases where
849  * the statement contains an optimizable statement for which a queryId
850  * could be derived (such as EXPLAIN or DECLARE CURSOR). For such cases,
851  * runtime control will first go through ProcessUtility and then the
852  * executor, and we don't want the executor hooks to do anything, since we
853  * are already measuring the statement's costs at the utility level.
854  */
855  if (query->utilityStmt)
856  {
857  query->queryId = UINT64CONST(0);
858  return;
859  }
860 
861  /* Set up workspace for query jumbling */
862  jstate.jumble = (unsigned char *) palloc(JUMBLE_SIZE);
863  jstate.jumble_len = 0;
864  jstate.clocations_buf_size = 32;
865  jstate.clocations = (pgssLocationLen *)
866  palloc(jstate.clocations_buf_size * sizeof(pgssLocationLen));
867  jstate.clocations_count = 0;
868  jstate.highest_extern_param_id = 0;
869 
870  /* Compute query ID and mark the Query node with it */
871  JumbleQuery(&jstate, query);
872  query->queryId =
874 
875  /*
876  * If we are unlucky enough to get a hash of zero, use 1 instead, to
877  * prevent confusion with the utility-statement case.
878  */
879  if (query->queryId == UINT64CONST(0))
880  query->queryId = UINT64CONST(1);
881 
882  /*
883  * If we were able to identify any ignorable constants, we immediately
884  * create a hash table entry for the query, so that we can record the
885  * normalized form of the query string. If there were no such constants,
886  * the normalized string would be the same as the query text anyway, so
887  * there's no need for an early entry.
888  */
889  if (jstate.clocations_count > 0)
890  pgss_store(pstate->p_sourcetext,
891  query->queryId,
892  query->stmt_location,
893  query->stmt_len,
894  PGSS_INVALID,
895  0,
896  0,
897  NULL,
898  NULL,
899  &jstate);
900 }
901 
902 /*
903  * Planner hook: forward to regular planner, but measure planning time
904  * if needed.
905  */
906 static PlannedStmt *
908  const char *query_string,
909  int cursorOptions,
910  ParamListInfo boundParams)
911 {
912  PlannedStmt *result;
913 
914  /*
915  * We can't process the query if no query_string is provided, as
916  * pgss_store needs it. We also ignore query without queryid, as it would
917  * be treated as a utility statement, which may not be the case.
918  *
919  * Note that planner_hook can be called from the planner itself, so we
920  * have a specific nesting level for the planner. However, utility
921  * commands containing optimizable statements can also call the planner,
922  * same for regular DML (for instance for underlying foreign key queries).
923  * So testing the planner nesting level only is not enough to detect real
924  * top level planner call.
925  */
927  && pgss_track_planning && query_string
928  && parse->queryId != UINT64CONST(0))
929  {
930  instr_time start;
932  BufferUsage bufusage_start,
933  bufusage;
934  WalUsage walusage_start,
935  walusage;
936 
937  /* We need to track buffer usage as the planner can access them. */
938  bufusage_start = pgBufferUsage;
939 
940  /*
941  * Similarly the planner could write some WAL records in some cases
942  * (e.g. setting a hint bit with those being WAL-logged)
943  */
944  walusage_start = pgWalUsage;
945  INSTR_TIME_SET_CURRENT(start);
946 
948  PG_TRY();
949  {
950  if (prev_planner_hook)
951  result = prev_planner_hook(parse, query_string, cursorOptions,
952  boundParams);
953  else
954  result = standard_planner(parse, query_string, cursorOptions,
955  boundParams);
956  }
957  PG_FINALLY();
958  {
960  }
961  PG_END_TRY();
962 
963  INSTR_TIME_SET_CURRENT(duration);
964  INSTR_TIME_SUBTRACT(duration, start);
965 
966  /* calc differences of buffer counters. */
967  memset(&bufusage, 0, sizeof(BufferUsage));
968  BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
969 
970  /* calc differences of WAL counters. */
971  memset(&walusage, 0, sizeof(WalUsage));
972  WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
973 
974  pgss_store(query_string,
975  parse->queryId,
976  parse->stmt_location,
977  parse->stmt_len,
978  PGSS_PLAN,
979  INSTR_TIME_GET_MILLISEC(duration),
980  0,
981  &bufusage,
982  &walusage,
983  NULL);
984  }
985  else
986  {
987  if (prev_planner_hook)
988  result = prev_planner_hook(parse, query_string, cursorOptions,
989  boundParams);
990  else
991  result = standard_planner(parse, query_string, cursorOptions,
992  boundParams);
993  }
994 
995  return result;
996 }
997 
998 /*
999  * ExecutorStart hook: start up tracking if needed
1000  */
1001 static void
1002 pgss_ExecutorStart(QueryDesc *queryDesc, int eflags)
1003 {
1004  if (prev_ExecutorStart)
1005  prev_ExecutorStart(queryDesc, eflags);
1006  else
1007  standard_ExecutorStart(queryDesc, eflags);
1008 
1009  /*
1010  * If query has queryId zero, don't track it. This prevents double
1011  * counting of optimizable statements that are directly contained in
1012  * utility statements.
1013  */
1014  if (pgss_enabled(exec_nested_level) && queryDesc->plannedstmt->queryId != UINT64CONST(0))
1015  {
1016  /*
1017  * Set up to track total elapsed time in ExecutorRun. Make sure the
1018  * space is allocated in the per-query context so it will go away at
1019  * ExecutorEnd.
1020  */
1021  if (queryDesc->totaltime == NULL)
1022  {
1023  MemoryContext oldcxt;
1024 
1025  oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
1026  queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL);
1027  MemoryContextSwitchTo(oldcxt);
1028  }
1029  }
1030 }
1031 
1032 /*
1033  * ExecutorRun hook: all we need do is track nesting depth
1034  */
1035 static void
1036 pgss_ExecutorRun(QueryDesc *queryDesc, ScanDirection direction, uint64 count,
1037  bool execute_once)
1038 {
1040  PG_TRY();
1041  {
1042  if (prev_ExecutorRun)
1043  prev_ExecutorRun(queryDesc, direction, count, execute_once);
1044  else
1045  standard_ExecutorRun(queryDesc, direction, count, execute_once);
1046  }
1047  PG_FINALLY();
1048  {
1050  }
1051  PG_END_TRY();
1052 }
1053 
1054 /*
1055  * ExecutorFinish hook: all we need do is track nesting depth
1056  */
1057 static void
1059 {
1061  PG_TRY();
1062  {
1063  if (prev_ExecutorFinish)
1064  prev_ExecutorFinish(queryDesc);
1065  else
1066  standard_ExecutorFinish(queryDesc);
1067  }
1068  PG_FINALLY();
1069  {
1071  }
1072  PG_END_TRY();
1073 }
1074 
1075 /*
1076  * ExecutorEnd hook: store results if needed
1077  */
1078 static void
1080 {
1081  uint64 queryId = queryDesc->plannedstmt->queryId;
1082 
1083  if (queryId != UINT64CONST(0) && queryDesc->totaltime &&
1085  {
1086  /*
1087  * Make sure stats accumulation is done. (Note: it's okay if several
1088  * levels of hook all do this.)
1089  */
1090  InstrEndLoop(queryDesc->totaltime);
1091 
1092  pgss_store(queryDesc->sourceText,
1093  queryId,
1094  queryDesc->plannedstmt->stmt_location,
1095  queryDesc->plannedstmt->stmt_len,
1096  PGSS_EXEC,
1097  queryDesc->totaltime->total * 1000.0, /* convert to msec */
1098  queryDesc->estate->es_processed,
1099  &queryDesc->totaltime->bufusage,
1100  &queryDesc->totaltime->walusage,
1101  NULL);
1102  }
1103 
1104  if (prev_ExecutorEnd)
1105  prev_ExecutorEnd(queryDesc);
1106  else
1107  standard_ExecutorEnd(queryDesc);
1108 }
1109 
1110 /*
1111  * ProcessUtility hook
1112  */
1113 static void
1114 pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
1115  ProcessUtilityContext context,
1116  ParamListInfo params, QueryEnvironment *queryEnv,
1118 {
1119  Node *parsetree = pstmt->utilityStmt;
1120 
1121  /*
1122  * If it's an EXECUTE statement, we don't track it and don't increment the
1123  * nesting level. This allows the cycles to be charged to the underlying
1124  * PREPARE instead (by the Executor hooks), which is much more useful.
1125  *
1126  * We also don't track execution of PREPARE. If we did, we would get one
1127  * hash table entry for the PREPARE (with hash calculated from the query
1128  * string), and then a different one with the same query string (but hash
1129  * calculated from the query tree) would be used to accumulate costs of
1130  * ensuing EXECUTEs. This would be confusing, and inconsistent with other
1131  * cases where planning time is not included at all.
1132  *
1133  * Likewise, we don't track execution of DEALLOCATE.
1134  */
1136  !IsA(parsetree, ExecuteStmt) &&
1137  !IsA(parsetree, PrepareStmt) &&
1138  !IsA(parsetree, DeallocateStmt))
1139  {
1140  instr_time start;
1142  uint64 rows;
1143  BufferUsage bufusage_start,
1144  bufusage;
1145  WalUsage walusage_start,
1146  walusage;
1147 
1148  bufusage_start = pgBufferUsage;
1149  walusage_start = pgWalUsage;
1150  INSTR_TIME_SET_CURRENT(start);
1151 
1153  PG_TRY();
1154  {
1155  if (prev_ProcessUtility)
1156  prev_ProcessUtility(pstmt, queryString,
1157  context, params, queryEnv,
1158  dest, qc);
1159  else
1160  standard_ProcessUtility(pstmt, queryString,
1161  context, params, queryEnv,
1162  dest, qc);
1163  }
1164  PG_FINALLY();
1165  {
1167  }
1168  PG_END_TRY();
1169 
1170  INSTR_TIME_SET_CURRENT(duration);
1171  INSTR_TIME_SUBTRACT(duration, start);
1172 
1173  rows = (qc && qc->commandTag == CMDTAG_COPY) ? qc->nprocessed : 0;
1174 
1175  /* calc differences of buffer counters. */
1176  memset(&bufusage, 0, sizeof(BufferUsage));
1177  BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
1178 
1179  /* calc differences of WAL counters. */
1180  memset(&walusage, 0, sizeof(WalUsage));
1181  WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
1182 
1183  pgss_store(queryString,
1184  0, /* signal that it's a utility stmt */
1185  pstmt->stmt_location,
1186  pstmt->stmt_len,
1187  PGSS_EXEC,
1188  INSTR_TIME_GET_MILLISEC(duration),
1189  rows,
1190  &bufusage,
1191  &walusage,
1192  NULL);
1193  }
1194  else
1195  {
1196  if (prev_ProcessUtility)
1197  prev_ProcessUtility(pstmt, queryString,
1198  context, params, queryEnv,
1199  dest, qc);
1200  else
1201  standard_ProcessUtility(pstmt, queryString,
1202  context, params, queryEnv,
1203  dest, qc);
1204  }
1205 }
1206 
1207 /*
1208  * Given an arbitrarily long query string, produce a hash for the purposes of
1209  * identifying the query, without normalizing constants. Used when hashing
1210  * utility statements.
1211  */
1212 static uint64
1213 pgss_hash_string(const char *str, int len)
1214 {
1215  return DatumGetUInt64(hash_any_extended((const unsigned char *) str,
1216  len, 0));
1217 }
1218 
1219 /*
1220  * Store some statistics for a statement.
1221  *
1222  * If queryId is 0 then this is a utility statement and we should compute
1223  * a suitable queryId internally.
1224  *
1225  * If jstate is not NULL then we're trying to create an entry for which
1226  * we have no statistics as yet; we just want to record the normalized
1227  * query string. total_time, rows, bufusage and walusage are ignored in this
1228  * case.
1229  *
1230  * If kind is PGSS_PLAN or PGSS_EXEC, its value is used as the array position
1231  * for the arrays in the Counters field.
1232  */
1233 static void
1234 pgss_store(const char *query, uint64 queryId,
1235  int query_location, int query_len,
1236  pgssStoreKind kind,
1237  double total_time, uint64 rows,
1238  const BufferUsage *bufusage,
1239  const WalUsage *walusage,
1240  pgssJumbleState *jstate)
1241 {
1242  pgssHashKey key;
1243  pgssEntry *entry;
1244  char *norm_query = NULL;
1245  int encoding = GetDatabaseEncoding();
1246 
1247  Assert(query != NULL);
1248 
1249  /* Safety check... */
1250  if (!pgss || !pgss_hash)
1251  return;
1252 
1253  /*
1254  * Confine our attention to the relevant part of the string, if the query
1255  * is a portion of a multi-statement source string.
1256  *
1257  * First apply starting offset, unless it's -1 (unknown).
1258  */
1259  if (query_location >= 0)
1260  {
1261  Assert(query_location <= strlen(query));
1262  query += query_location;
1263  /* Length of 0 (or -1) means "rest of string" */
1264  if (query_len <= 0)
1265  query_len = strlen(query);
1266  else
1267  Assert(query_len <= strlen(query));
1268  }
1269  else
1270  {
1271  /* If query location is unknown, distrust query_len as well */
1272  query_location = 0;
1273  query_len = strlen(query);
1274  }
1275 
1276  /*
1277  * Discard leading and trailing whitespace, too. Use scanner_isspace()
1278  * not libc's isspace(), because we want to match the lexer's behavior.
1279  */
1280  while (query_len > 0 && scanner_isspace(query[0]))
1281  query++, query_location++, query_len--;
1282  while (query_len > 0 && scanner_isspace(query[query_len - 1]))
1283  query_len--;
1284 
1285  /*
1286  * For utility statements, we just hash the query string to get an ID.
1287  */
1288  if (queryId == UINT64CONST(0))
1289  {
1290  queryId = pgss_hash_string(query, query_len);
1291 
1292  /*
1293  * If we are unlucky enough to get a hash of zero(invalid), use
1294  * queryID as 2 instead, queryID 1 is already in use for normal
1295  * statements.
1296  */
1297  if (queryId == UINT64CONST(0))
1298  queryId = UINT64CONST(2);
1299  }
1300 
1301  /* Set up key for hashtable search */
1302  key.userid = GetUserId();
1303  key.dbid = MyDatabaseId;
1304  key.queryid = queryId;
1305 
1306  /* Lookup the hash table entry with shared lock. */
1307  LWLockAcquire(pgss->lock, LW_SHARED);
1308 
1309  entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
1310 
1311  /* Create new entry, if not present */
1312  if (!entry)
1313  {
1314  Size query_offset;
1315  int gc_count;
1316  bool stored;
1317  bool do_gc;
1318 
1319  /*
1320  * Create a new, normalized query string if caller asked. We don't
1321  * need to hold the lock while doing this work. (Note: in any case,
1322  * it's possible that someone else creates a duplicate hashtable entry
1323  * in the interval where we don't hold the lock below. That case is
1324  * handled by entry_alloc.)
1325  */
1326  if (jstate)
1327  {
1328  LWLockRelease(pgss->lock);
1329  norm_query = generate_normalized_query(jstate, query,
1330  query_location,
1331  &query_len,
1332  encoding);
1333  LWLockAcquire(pgss->lock, LW_SHARED);
1334  }
1335 
1336  /* Append new query text to file with only shared lock held */
1337  stored = qtext_store(norm_query ? norm_query : query, query_len,
1338  &query_offset, &gc_count);
1339 
1340  /*
1341  * Determine whether we need to garbage collect external query texts
1342  * while the shared lock is still held. This micro-optimization
1343  * avoids taking the time to decide this while holding exclusive lock.
1344  */
1345  do_gc = need_gc_qtexts();
1346 
1347  /* Need exclusive lock to make a new hashtable entry - promote */
1348  LWLockRelease(pgss->lock);
1350 
1351  /*
1352  * A garbage collection may have occurred while we weren't holding the
1353  * lock. In the unlikely event that this happens, the query text we
1354  * stored above will have been garbage collected, so write it again.
1355  * This should be infrequent enough that doing it while holding
1356  * exclusive lock isn't a performance problem.
1357  */
1358  if (!stored || pgss->gc_count != gc_count)
1359  stored = qtext_store(norm_query ? norm_query : query, query_len,
1360  &query_offset, NULL);
1361 
1362  /* If we failed to write to the text file, give up */
1363  if (!stored)
1364  goto done;
1365 
1366  /* OK to create a new hashtable entry */
1367  entry = entry_alloc(&key, query_offset, query_len, encoding,
1368  jstate != NULL);
1369 
1370  /* If needed, perform garbage collection while exclusive lock held */
1371  if (do_gc)
1372  gc_qtexts();
1373  }
1374 
1375  /* Increment the counts, except when jstate is not NULL */
1376  if (!jstate)
1377  {
1378  /*
1379  * Grab the spinlock while updating the counters (see comment about
1380  * locking rules at the head of the file)
1381  */
1382  volatile pgssEntry *e = (volatile pgssEntry *) entry;
1383 
1384  Assert(kind == PGSS_PLAN || kind == PGSS_EXEC);
1385 
1386  SpinLockAcquire(&e->mutex);
1387 
1388  /* "Unstick" entry if it was previously sticky */
1389  if (IS_STICKY(e->counters))
1390  e->counters.usage = USAGE_INIT;
1391 
1392  e->counters.calls[kind] += 1;
1393  e->counters.total_time[kind] += total_time;
1394 
1395  if (e->counters.calls[kind] == 1)
1396  {
1397  e->counters.min_time[kind] = total_time;
1398  e->counters.max_time[kind] = total_time;
1399  e->counters.mean_time[kind] = total_time;
1400  }
1401  else
1402  {
1403  /*
1404  * Welford's method for accurately computing variance. See
1405  * <http://www.johndcook.com/blog/standard_deviation/>
1406  */
1407  double old_mean = e->counters.mean_time[kind];
1408 
1409  e->counters.mean_time[kind] +=
1410  (total_time - old_mean) / e->counters.calls[kind];
1411  e->counters.sum_var_time[kind] +=
1412  (total_time - old_mean) * (total_time - e->counters.mean_time[kind]);
1413 
1414  /* calculate min and max time */
1415  if (e->counters.min_time[kind] > total_time)
1416  e->counters.min_time[kind] = total_time;
1417  if (e->counters.max_time[kind] < total_time)
1418  e->counters.max_time[kind] = total_time;
1419  }
1420  e->counters.rows += rows;
1421  e->counters.shared_blks_hit += bufusage->shared_blks_hit;
1422  e->counters.shared_blks_read += bufusage->shared_blks_read;
1425  e->counters.local_blks_hit += bufusage->local_blks_hit;
1426  e->counters.local_blks_read += bufusage->local_blks_read;
1429  e->counters.temp_blks_read += bufusage->temp_blks_read;
1433  e->counters.usage += USAGE_EXEC(total_time);
1434  e->counters.wal_records += walusage->wal_records;
1435  e->counters.wal_fpi += walusage->wal_fpi;
1436  e->counters.wal_bytes += walusage->wal_bytes;
1437 
1438  SpinLockRelease(&e->mutex);
1439  }
1440 
1441 done:
1442  LWLockRelease(pgss->lock);
1443 
1444  /* We postpone this clean-up until we're out of the lock */
1445  if (norm_query)
1446  pfree(norm_query);
1447 }
1448 
1449 /*
1450  * Reset statement statistics corresponding to userid, dbid, and queryid.
1451  */
1452 Datum
1454 {
1455  Oid userid;
1456  Oid dbid;
1457  uint64 queryid;
1458 
1459  userid = PG_GETARG_OID(0);
1460  dbid = PG_GETARG_OID(1);
1461  queryid = (uint64) PG_GETARG_INT64(2);
1462 
1463  entry_reset(userid, dbid, queryid);
1464 
1465  PG_RETURN_VOID();
1466 }
1467 
1468 /*
1469  * Reset statement statistics.
1470  */
1471 Datum
1473 {
1474  entry_reset(0, 0, 0);
1475 
1476  PG_RETURN_VOID();
1477 }
1478 
1479 /* Number of output arguments (columns) for various API versions */
1480 #define PG_STAT_STATEMENTS_COLS_V1_0 14
1481 #define PG_STAT_STATEMENTS_COLS_V1_1 18
1482 #define PG_STAT_STATEMENTS_COLS_V1_2 19
1483 #define PG_STAT_STATEMENTS_COLS_V1_3 23
1484 #define PG_STAT_STATEMENTS_COLS_V1_8 32
1485 #define PG_STAT_STATEMENTS_COLS 32 /* maximum of above */
1486 
1487 /*
1488  * Retrieve statement statistics.
1489  *
1490  * The SQL API of this function has changed multiple times, and will likely
1491  * do so again in future. To support the case where a newer version of this
1492  * loadable module is being used with an old SQL declaration of the function,
1493  * we continue to support the older API versions. For 1.2 and later, the
1494  * expected API version is identified by embedding it in the C name of the
1495  * function. Unfortunately we weren't bright enough to do that for 1.1.
1496  */
1497 Datum
1499 {
1500  bool showtext = PG_GETARG_BOOL(0);
1501 
1502  pg_stat_statements_internal(fcinfo, PGSS_V1_8, showtext);
1503 
1504  return (Datum) 0;
1505 }
1506 
1507 Datum
1509 {
1510  bool showtext = PG_GETARG_BOOL(0);
1511 
1512  pg_stat_statements_internal(fcinfo, PGSS_V1_3, showtext);
1513 
1514  return (Datum) 0;
1515 }
1516 
1517 Datum
1519 {
1520  bool showtext = PG_GETARG_BOOL(0);
1521 
1522  pg_stat_statements_internal(fcinfo, PGSS_V1_2, showtext);
1523 
1524  return (Datum) 0;
1525 }
1526 
1527 /*
1528  * Legacy entry point for pg_stat_statements() API versions 1.0 and 1.1.
1529  * This can be removed someday, perhaps.
1530  */
1531 Datum
1533 {
1534  /* If it's really API 1.1, we'll figure that out below */
1535  pg_stat_statements_internal(fcinfo, PGSS_V1_0, true);
1536 
1537  return (Datum) 0;
1538 }
1539 
1540 /* Common code for all versions of pg_stat_statements() */
1541 static void
1543  pgssVersion api_version,
1544  bool showtext)
1545 {
1546  ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1547  TupleDesc tupdesc;
1548  Tuplestorestate *tupstore;
1549  MemoryContext per_query_ctx;
1550  MemoryContext oldcontext;
1551  Oid userid = GetUserId();
1552  bool is_allowed_role = false;
1553  char *qbuffer = NULL;
1554  Size qbuffer_size = 0;
1555  Size extent = 0;
1556  int gc_count = 0;
1557  HASH_SEQ_STATUS hash_seq;
1558  pgssEntry *entry;
1559 
1560  /* Superusers or members of pg_read_all_stats members are allowed */
1561  is_allowed_role = is_member_of_role(GetUserId(), DEFAULT_ROLE_READ_ALL_STATS);
1562 
1563  /* hash table must exist already */
1564  if (!pgss || !pgss_hash)
1565  ereport(ERROR,
1566  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1567  errmsg("pg_stat_statements must be loaded via shared_preload_libraries")));
1568 
1569  /* check to see if caller supports us returning a tuplestore */
1570  if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
1571  ereport(ERROR,
1572  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1573  errmsg("set-valued function called in context that cannot accept a set")));
1574  if (!(rsinfo->allowedModes & SFRM_Materialize))
1575  ereport(ERROR,
1576  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1577  errmsg("materialize mode required, but it is not allowed in this context")));
1578 
1579  /* Switch into long-lived context to construct returned data structures */
1580  per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
1581  oldcontext = MemoryContextSwitchTo(per_query_ctx);
1582 
1583  /* Build a tuple descriptor for our result type */
1584  if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1585  elog(ERROR, "return type must be a row type");
1586 
1587  /*
1588  * Check we have the expected number of output arguments. Aside from
1589  * being a good safety check, we need a kluge here to detect API version
1590  * 1.1, which was wedged into the code in an ill-considered way.
1591  */
1592  switch (tupdesc->natts)
1593  {
1595  if (api_version != PGSS_V1_0)
1596  elog(ERROR, "incorrect number of output arguments");
1597  break;
1599  /* pg_stat_statements() should have told us 1.0 */
1600  if (api_version != PGSS_V1_0)
1601  elog(ERROR, "incorrect number of output arguments");
1602  api_version = PGSS_V1_1;
1603  break;
1605  if (api_version != PGSS_V1_2)
1606  elog(ERROR, "incorrect number of output arguments");
1607  break;
1609  if (api_version != PGSS_V1_3)
1610  elog(ERROR, "incorrect number of output arguments");
1611  break;
1613  if (api_version != PGSS_V1_8)
1614  elog(ERROR, "incorrect number of output arguments");
1615  break;
1616  default:
1617  elog(ERROR, "incorrect number of output arguments");
1618  }
1619 
1620  tupstore = tuplestore_begin_heap(true, false, work_mem);
1621  rsinfo->returnMode = SFRM_Materialize;
1622  rsinfo->setResult = tupstore;
1623  rsinfo->setDesc = tupdesc;
1624 
1625  MemoryContextSwitchTo(oldcontext);
1626 
1627  /*
1628  * We'd like to load the query text file (if needed) while not holding any
1629  * lock on pgss->lock. In the worst case we'll have to do this again
1630  * after we have the lock, but it's unlikely enough to make this a win
1631  * despite occasional duplicated work. We need to reload if anybody
1632  * writes to the file (either a retail qtext_store(), or a garbage
1633  * collection) between this point and where we've gotten shared lock. If
1634  * a qtext_store is actually in progress when we look, we might as well
1635  * skip the speculative load entirely.
1636  */
1637  if (showtext)
1638  {
1639  int n_writers;
1640 
1641  /* Take the mutex so we can examine variables */
1642  {
1643  volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
1644 
1645  SpinLockAcquire(&s->mutex);
1646  extent = s->extent;
1647  n_writers = s->n_writers;
1648  gc_count = s->gc_count;
1649  SpinLockRelease(&s->mutex);
1650  }
1651 
1652  /* No point in loading file now if there are active writers */
1653  if (n_writers == 0)
1654  qbuffer = qtext_load_file(&qbuffer_size);
1655  }
1656 
1657  /*
1658  * Get shared lock, load or reload the query text file if we must, and
1659  * iterate over the hashtable entries.
1660  *
1661  * With a large hash table, we might be holding the lock rather longer
1662  * than one could wish. However, this only blocks creation of new hash
1663  * table entries, and the larger the hash table the less likely that is to
1664  * be needed. So we can hope this is okay. Perhaps someday we'll decide
1665  * we need to partition the hash table to limit the time spent holding any
1666  * one lock.
1667  */
1668  LWLockAcquire(pgss->lock, LW_SHARED);
1669 
1670  if (showtext)
1671  {
1672  /*
1673  * Here it is safe to examine extent and gc_count without taking the
1674  * mutex. Note that although other processes might change
1675  * pgss->extent just after we look at it, the strings they then write
1676  * into the file cannot yet be referenced in the hashtable, so we
1677  * don't care whether we see them or not.
1678  *
1679  * If qtext_load_file fails, we just press on; we'll return NULL for
1680  * every query text.
1681  */
1682  if (qbuffer == NULL ||
1683  pgss->extent != extent ||
1684  pgss->gc_count != gc_count)
1685  {
1686  if (qbuffer)
1687  free(qbuffer);
1688  qbuffer = qtext_load_file(&qbuffer_size);
1689  }
1690  }
1691 
1692  hash_seq_init(&hash_seq, pgss_hash);
1693  while ((entry = hash_seq_search(&hash_seq)) != NULL)
1694  {
1696  bool nulls[PG_STAT_STATEMENTS_COLS];
1697  int i = 0;
1698  Counters tmp;
1699  double stddev;
1700  int64 queryid = entry->key.queryid;
1701 
1702  memset(values, 0, sizeof(values));
1703  memset(nulls, 0, sizeof(nulls));
1704 
1705  values[i++] = ObjectIdGetDatum(entry->key.userid);
1706  values[i++] = ObjectIdGetDatum(entry->key.dbid);
1707 
1708  if (is_allowed_role || entry->key.userid == userid)
1709  {
1710  if (api_version >= PGSS_V1_2)
1711  values[i++] = Int64GetDatumFast(queryid);
1712 
1713  if (showtext)
1714  {
1715  char *qstr = qtext_fetch(entry->query_offset,
1716  entry->query_len,
1717  qbuffer,
1718  qbuffer_size);
1719 
1720  if (qstr)
1721  {
1722  char *enc;
1723 
1724  enc = pg_any_to_server(qstr,
1725  entry->query_len,
1726  entry->encoding);
1727 
1728  values[i++] = CStringGetTextDatum(enc);
1729 
1730  if (enc != qstr)
1731  pfree(enc);
1732  }
1733  else
1734  {
1735  /* Just return a null if we fail to find the text */
1736  nulls[i++] = true;
1737  }
1738  }
1739  else
1740  {
1741  /* Query text not requested */
1742  nulls[i++] = true;
1743  }
1744  }
1745  else
1746  {
1747  /* Don't show queryid */
1748  if (api_version >= PGSS_V1_2)
1749  nulls[i++] = true;
1750 
1751  /*
1752  * Don't show query text, but hint as to the reason for not doing
1753  * so if it was requested
1754  */
1755  if (showtext)
1756  values[i++] = CStringGetTextDatum("<insufficient privilege>");
1757  else
1758  nulls[i++] = true;
1759  }
1760 
1761  /* copy counters to a local variable to keep locking time short */
1762  {
1763  volatile pgssEntry *e = (volatile pgssEntry *) entry;
1764 
1765  SpinLockAcquire(&e->mutex);
1766  tmp = e->counters;
1767  SpinLockRelease(&e->mutex);
1768  }
1769 
1770  /* Skip entry if unexecuted (ie, it's a pending "sticky" entry) */
1771  if (IS_STICKY(tmp))
1772  continue;
1773 
1774  /* Note that we rely on PGSS_PLAN being 0 and PGSS_EXEC being 1. */
1775  for (int kind = 0; kind < PGSS_NUMKIND; kind++)
1776  {
1777  if (kind == PGSS_EXEC || api_version >= PGSS_V1_8)
1778  {
1779  values[i++] = Int64GetDatumFast(tmp.calls[kind]);
1780  values[i++] = Float8GetDatumFast(tmp.total_time[kind]);
1781  }
1782 
1783  if ((kind == PGSS_EXEC && api_version >= PGSS_V1_3) ||
1784  api_version >= PGSS_V1_8)
1785  {
1786  values[i++] = Float8GetDatumFast(tmp.min_time[kind]);
1787  values[i++] = Float8GetDatumFast(tmp.max_time[kind]);
1788  values[i++] = Float8GetDatumFast(tmp.mean_time[kind]);
1789 
1790  /*
1791  * Note we are calculating the population variance here, not
1792  * the sample variance, as we have data for the whole
1793  * population, so Bessel's correction is not used, and we
1794  * don't divide by tmp.calls - 1.
1795  */
1796  if (tmp.calls[kind] > 1)
1797  stddev = sqrt(tmp.sum_var_time[kind] / tmp.calls[kind]);
1798  else
1799  stddev = 0.0;
1800  values[i++] = Float8GetDatumFast(stddev);
1801  }
1802  }
1803  values[i++] = Int64GetDatumFast(tmp.rows);
1804  values[i++] = Int64GetDatumFast(tmp.shared_blks_hit);
1805  values[i++] = Int64GetDatumFast(tmp.shared_blks_read);
1806  if (api_version >= PGSS_V1_1)
1807  values[i++] = Int64GetDatumFast(tmp.shared_blks_dirtied);
1808  values[i++] = Int64GetDatumFast(tmp.shared_blks_written);
1809  values[i++] = Int64GetDatumFast(tmp.local_blks_hit);
1810  values[i++] = Int64GetDatumFast(tmp.local_blks_read);
1811  if (api_version >= PGSS_V1_1)
1812  values[i++] = Int64GetDatumFast(tmp.local_blks_dirtied);
1813  values[i++] = Int64GetDatumFast(tmp.local_blks_written);
1814  values[i++] = Int64GetDatumFast(tmp.temp_blks_read);
1815  values[i++] = Int64GetDatumFast(tmp.temp_blks_written);
1816  if (api_version >= PGSS_V1_1)
1817  {
1818  values[i++] = Float8GetDatumFast(tmp.blk_read_time);
1819  values[i++] = Float8GetDatumFast(tmp.blk_write_time);
1820  }
1821  if (api_version >= PGSS_V1_8)
1822  {
1823  char buf[256];
1824  Datum wal_bytes;
1825 
1826  values[i++] = Int64GetDatumFast(tmp.wal_records);
1827  values[i++] = Int64GetDatumFast(tmp.wal_fpi);
1828 
1829  snprintf(buf, sizeof buf, UINT64_FORMAT, tmp.wal_bytes);
1830 
1831  /* Convert to numeric. */
1832  wal_bytes = DirectFunctionCall3(numeric_in,
1833  CStringGetDatum(buf),
1834  ObjectIdGetDatum(0),
1835  Int32GetDatum(-1));
1836  values[i++] = wal_bytes;
1837  }
1838 
1839  Assert(i == (api_version == PGSS_V1_0 ? PG_STAT_STATEMENTS_COLS_V1_0 :
1840  api_version == PGSS_V1_1 ? PG_STAT_STATEMENTS_COLS_V1_1 :
1841  api_version == PGSS_V1_2 ? PG_STAT_STATEMENTS_COLS_V1_2 :
1842  api_version == PGSS_V1_3 ? PG_STAT_STATEMENTS_COLS_V1_3 :
1843  api_version == PGSS_V1_8 ? PG_STAT_STATEMENTS_COLS_V1_8 :
1844  -1 /* fail if you forget to update this assert */ ));
1845 
1846  tuplestore_putvalues(tupstore, tupdesc, values, nulls);
1847  }
1848 
1849  /* clean up and return the tuplestore */
1850  LWLockRelease(pgss->lock);
1851 
1852  if (qbuffer)
1853  free(qbuffer);
1854 
1855  tuplestore_donestoring(tupstore);
1856 }
1857 
1858 /*
1859  * Estimate shared memory space needed.
1860  */
1861 static Size
1863 {
1864  Size size;
1865 
1866  size = MAXALIGN(sizeof(pgssSharedState));
1867  size = add_size(size, hash_estimate_size(pgss_max, sizeof(pgssEntry)));
1868 
1869  return size;
1870 }
1871 
1872 /*
1873  * Allocate a new hashtable entry.
1874  * caller must hold an exclusive lock on pgss->lock
1875  *
1876  * "query" need not be null-terminated; we rely on query_len instead
1877  *
1878  * If "sticky" is true, make the new entry artificially sticky so that it will
1879  * probably still be there when the query finishes execution. We do this by
1880  * giving it a median usage value rather than the normal value. (Strictly
1881  * speaking, query strings are normalized on a best effort basis, though it
1882  * would be difficult to demonstrate this even under artificial conditions.)
1883  *
1884  * Note: despite needing exclusive lock, it's not an error for the target
1885  * entry to already exist. This is because pgss_store releases and
1886  * reacquires lock after failing to find a match; so someone else could
1887  * have made the entry while we waited to get exclusive lock.
1888  */
1889 static pgssEntry *
1890 entry_alloc(pgssHashKey *key, Size query_offset, int query_len, int encoding,
1891  bool sticky)
1892 {
1893  pgssEntry *entry;
1894  bool found;
1895 
1896  /* Make space if needed */
1897  while (hash_get_num_entries(pgss_hash) >= pgss_max)
1898  entry_dealloc();
1899 
1900  /* Find or create an entry with desired hash code */
1901  entry = (pgssEntry *) hash_search(pgss_hash, key, HASH_ENTER, &found);
1902 
1903  if (!found)
1904  {
1905  /* New entry, initialize it */
1906 
1907  /* reset the statistics */
1908  memset(&entry->counters, 0, sizeof(Counters));
1909  /* set the appropriate initial usage count */
1910  entry->counters.usage = sticky ? pgss->cur_median_usage : USAGE_INIT;
1911  /* re-initialize the mutex each time ... we assume no one using it */
1912  SpinLockInit(&entry->mutex);
1913  /* ... and don't forget the query text metadata */
1914  Assert(query_len >= 0);
1915  entry->query_offset = query_offset;
1916  entry->query_len = query_len;
1917  entry->encoding = encoding;
1918  }
1919 
1920  return entry;
1921 }
1922 
1923 /*
1924  * qsort comparator for sorting into increasing usage order
1925  */
1926 static int
1927 entry_cmp(const void *lhs, const void *rhs)
1928 {
1929  double l_usage = (*(pgssEntry *const *) lhs)->counters.usage;
1930  double r_usage = (*(pgssEntry *const *) rhs)->counters.usage;
1931 
1932  if (l_usage < r_usage)
1933  return -1;
1934  else if (l_usage > r_usage)
1935  return +1;
1936  else
1937  return 0;
1938 }
1939 
1940 /*
1941  * Deallocate least-used entries.
1942  *
1943  * Caller must hold an exclusive lock on pgss->lock.
1944  */
1945 static void
1947 {
1948  HASH_SEQ_STATUS hash_seq;
1949  pgssEntry **entries;
1950  pgssEntry *entry;
1951  int nvictims;
1952  int i;
1953  Size tottextlen;
1954  int nvalidtexts;
1955 
1956  /*
1957  * Sort entries by usage and deallocate USAGE_DEALLOC_PERCENT of them.
1958  * While we're scanning the table, apply the decay factor to the usage
1959  * values, and update the mean query length.
1960  *
1961  * Note that the mean query length is almost immediately obsolete, since
1962  * we compute it before not after discarding the least-used entries.
1963  * Hopefully, that doesn't affect the mean too much; it doesn't seem worth
1964  * making two passes to get a more current result. Likewise, the new
1965  * cur_median_usage includes the entries we're about to zap.
1966  */
1967 
1968  entries = palloc(hash_get_num_entries(pgss_hash) * sizeof(pgssEntry *));
1969 
1970  i = 0;
1971  tottextlen = 0;
1972  nvalidtexts = 0;
1973 
1974  hash_seq_init(&hash_seq, pgss_hash);
1975  while ((entry = hash_seq_search(&hash_seq)) != NULL)
1976  {
1977  entries[i++] = entry;
1978  /* "Sticky" entries get a different usage decay rate. */
1979  if (IS_STICKY(entry->counters))
1981  else
1983  /* In the mean length computation, ignore dropped texts. */
1984  if (entry->query_len >= 0)
1985  {
1986  tottextlen += entry->query_len + 1;
1987  nvalidtexts++;
1988  }
1989  }
1990 
1991  /* Sort into increasing order by usage */
1992  qsort(entries, i, sizeof(pgssEntry *), entry_cmp);
1993 
1994  /* Record the (approximate) median usage */
1995  if (i > 0)
1996  pgss->cur_median_usage = entries[i / 2]->counters.usage;
1997  /* Record the mean query length */
1998  if (nvalidtexts > 0)
1999  pgss->mean_query_len = tottextlen / nvalidtexts;
2000  else
2002 
2003  /* Now zap an appropriate fraction of lowest-usage entries */
2004  nvictims = Max(10, i * USAGE_DEALLOC_PERCENT / 100);
2005  nvictims = Min(nvictims, i);
2006 
2007  for (i = 0; i < nvictims; i++)
2008  {
2009  hash_search(pgss_hash, &entries[i]->key, HASH_REMOVE, NULL);
2010  }
2011 
2012  pfree(entries);
2013 }
2014 
2015 /*
2016  * Given a query string (not necessarily null-terminated), allocate a new
2017  * entry in the external query text file and store the string there.
2018  *
2019  * If successful, returns true, and stores the new entry's offset in the file
2020  * into *query_offset. Also, if gc_count isn't NULL, *gc_count is set to the
2021  * number of garbage collections that have occurred so far.
2022  *
2023  * On failure, returns false.
2024  *
2025  * At least a shared lock on pgss->lock must be held by the caller, so as
2026  * to prevent a concurrent garbage collection. Share-lock-holding callers
2027  * should pass a gc_count pointer to obtain the number of garbage collections,
2028  * so that they can recheck the count after obtaining exclusive lock to
2029  * detect whether a garbage collection occurred (and removed this entry).
2030  */
2031 static bool
2032 qtext_store(const char *query, int query_len,
2033  Size *query_offset, int *gc_count)
2034 {
2035  Size off;
2036  int fd;
2037 
2038  /*
2039  * We use a spinlock to protect extent/n_writers/gc_count, so that
2040  * multiple processes may execute this function concurrently.
2041  */
2042  {
2043  volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
2044 
2045  SpinLockAcquire(&s->mutex);
2046  off = s->extent;
2047  s->extent += query_len + 1;
2048  s->n_writers++;
2049  if (gc_count)
2050  *gc_count = s->gc_count;
2051  SpinLockRelease(&s->mutex);
2052  }
2053 
2054  *query_offset = off;
2055 
2056  /* Now write the data into the successfully-reserved part of the file */
2057  fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDWR | O_CREAT | PG_BINARY);
2058  if (fd < 0)
2059  goto error;
2060 
2061  if (pg_pwrite(fd, query, query_len, off) != query_len)
2062  goto error;
2063  if (pg_pwrite(fd, "\0", 1, off + query_len) != 1)
2064  goto error;
2065 
2066  CloseTransientFile(fd);
2067 
2068  /* Mark our write complete */
2069  {
2070  volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
2071 
2072  SpinLockAcquire(&s->mutex);
2073  s->n_writers--;
2074  SpinLockRelease(&s->mutex);
2075  }
2076 
2077  return true;
2078 
2079 error:
2080  ereport(LOG,
2082  errmsg("could not write file \"%s\": %m",
2083  PGSS_TEXT_FILE)));
2084 
2085  if (fd >= 0)
2086  CloseTransientFile(fd);
2087 
2088  /* Mark our write complete */
2089  {
2090  volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
2091 
2092  SpinLockAcquire(&s->mutex);
2093  s->n_writers--;
2094  SpinLockRelease(&s->mutex);
2095  }
2096 
2097  return false;
2098 }
2099 
2100 /*
2101  * Read the external query text file into a malloc'd buffer.
2102  *
2103  * Returns NULL (without throwing an error) if unable to read, eg
2104  * file not there or insufficient memory.
2105  *
2106  * On success, the buffer size is also returned into *buffer_size.
2107  *
2108  * This can be called without any lock on pgss->lock, but in that case
2109  * the caller is responsible for verifying that the result is sane.
2110  */
2111 static char *
2112 qtext_load_file(Size *buffer_size)
2113 {
2114  char *buf;
2115  int fd;
2116  struct stat stat;
2117 
2118  fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDONLY | PG_BINARY);
2119  if (fd < 0)
2120  {
2121  if (errno != ENOENT)
2122  ereport(LOG,
2124  errmsg("could not read file \"%s\": %m",
2125  PGSS_TEXT_FILE)));
2126  return NULL;
2127  }
2128 
2129  /* Get file length */
2130  if (fstat(fd, &stat))
2131  {
2132  ereport(LOG,
2134  errmsg("could not stat file \"%s\": %m",
2135  PGSS_TEXT_FILE)));
2136  CloseTransientFile(fd);
2137  return NULL;
2138  }
2139 
2140  /* Allocate buffer; beware that off_t might be wider than size_t */
2141  if (stat.st_size <= MaxAllocHugeSize)
2142  buf = (char *) malloc(stat.st_size);
2143  else
2144  buf = NULL;
2145  if (buf == NULL)
2146  {
2147  ereport(LOG,
2148  (errcode(ERRCODE_OUT_OF_MEMORY),
2149  errmsg("out of memory"),
2150  errdetail("Could not allocate enough memory to read file \"%s\".",
2151  PGSS_TEXT_FILE)));
2152  CloseTransientFile(fd);
2153  return NULL;
2154  }
2155 
2156  /*
2157  * OK, slurp in the file. If we get a short read and errno doesn't get
2158  * set, the reason is probably that garbage collection truncated the file
2159  * since we did the fstat(), so we don't log a complaint --- but we don't
2160  * return the data, either, since it's most likely corrupt due to
2161  * concurrent writes from garbage collection.
2162  */
2163  errno = 0;
2164  if (read(fd, buf, stat.st_size) != stat.st_size)
2165  {
2166  if (errno)
2167  ereport(LOG,
2169  errmsg("could not read file \"%s\": %m",
2170  PGSS_TEXT_FILE)));
2171  free(buf);
2172  CloseTransientFile(fd);
2173  return NULL;
2174  }
2175 
2176  if (CloseTransientFile(fd) != 0)
2177  ereport(LOG,
2179  errmsg("could not close file \"%s\": %m", PGSS_TEXT_FILE)));
2180 
2181  *buffer_size = stat.st_size;
2182  return buf;
2183 }
2184 
2185 /*
2186  * Locate a query text in the file image previously read by qtext_load_file().
2187  *
2188  * We validate the given offset/length, and return NULL if bogus. Otherwise,
2189  * the result points to a null-terminated string within the buffer.
2190  */
2191 static char *
2192 qtext_fetch(Size query_offset, int query_len,
2193  char *buffer, Size buffer_size)
2194 {
2195  /* File read failed? */
2196  if (buffer == NULL)
2197  return NULL;
2198  /* Bogus offset/length? */
2199  if (query_len < 0 ||
2200  query_offset + query_len >= buffer_size)
2201  return NULL;
2202  /* As a further sanity check, make sure there's a trailing null */
2203  if (buffer[query_offset + query_len] != '\0')
2204  return NULL;
2205  /* Looks OK */
2206  return buffer + query_offset;
2207 }
2208 
2209 /*
2210  * Do we need to garbage-collect the external query text file?
2211  *
2212  * Caller should hold at least a shared lock on pgss->lock.
2213  */
2214 static bool
2216 {
2217  Size extent;
2218 
2219  /* Read shared extent pointer */
2220  {
2221  volatile pgssSharedState *s = (volatile pgssSharedState *) pgss;
2222 
2223  SpinLockAcquire(&s->mutex);
2224  extent = s->extent;
2225  SpinLockRelease(&s->mutex);
2226  }
2227 
2228  /* Don't proceed if file does not exceed 512 bytes per possible entry */
2229  if (extent < 512 * pgss_max)
2230  return false;
2231 
2232  /*
2233  * Don't proceed if file is less than about 50% bloat. Nothing can or
2234  * should be done in the event of unusually large query texts accounting
2235  * for file's large size. We go to the trouble of maintaining the mean
2236  * query length in order to prevent garbage collection from thrashing
2237  * uselessly.
2238  */
2239  if (extent < pgss->mean_query_len * pgss_max * 2)
2240  return false;
2241 
2242  return true;
2243 }
2244 
2245 /*
2246  * Garbage-collect orphaned query texts in external file.
2247  *
2248  * This won't be called often in the typical case, since it's likely that
2249  * there won't be too much churn, and besides, a similar compaction process
2250  * occurs when serializing to disk at shutdown or as part of resetting.
2251  * Despite this, it seems prudent to plan for the edge case where the file
2252  * becomes unreasonably large, with no other method of compaction likely to
2253  * occur in the foreseeable future.
2254  *
2255  * The caller must hold an exclusive lock on pgss->lock.
2256  *
2257  * At the first sign of trouble we unlink the query text file to get a clean
2258  * slate (although existing statistics are retained), rather than risk
2259  * thrashing by allowing the same problem case to recur indefinitely.
2260  */
2261 static void
2263 {
2264  char *qbuffer;
2265  Size qbuffer_size;
2266  FILE *qfile = NULL;
2267  HASH_SEQ_STATUS hash_seq;
2268  pgssEntry *entry;
2269  Size extent;
2270  int nentries;
2271 
2272  /*
2273  * When called from pgss_store, some other session might have proceeded
2274  * with garbage collection in the no-lock-held interim of lock strength
2275  * escalation. Check once more that this is actually necessary.
2276  */
2277  if (!need_gc_qtexts())
2278  return;
2279 
2280  /*
2281  * Load the old texts file. If we fail (out of memory, for instance),
2282  * invalidate query texts. Hopefully this is rare. It might seem better
2283  * to leave things alone on an OOM failure, but the problem is that the
2284  * file is only going to get bigger; hoping for a future non-OOM result is
2285  * risky and can easily lead to complete denial of service.
2286  */
2287  qbuffer = qtext_load_file(&qbuffer_size);
2288  if (qbuffer == NULL)
2289  goto gc_fail;
2290 
2291  /*
2292  * We overwrite the query texts file in place, so as to reduce the risk of
2293  * an out-of-disk-space failure. Since the file is guaranteed not to get
2294  * larger, this should always work on traditional filesystems; though we
2295  * could still lose on copy-on-write filesystems.
2296  */
2298  if (qfile == NULL)
2299  {
2300  ereport(LOG,
2302  errmsg("could not write file \"%s\": %m",
2303  PGSS_TEXT_FILE)));
2304  goto gc_fail;
2305  }
2306 
2307  extent = 0;
2308  nentries = 0;
2309 
2310  hash_seq_init(&hash_seq, pgss_hash);
2311  while ((entry = hash_seq_search(&hash_seq)) != NULL)
2312  {
2313  int query_len = entry->query_len;
2314  char *qry = qtext_fetch(entry->query_offset,
2315  query_len,
2316  qbuffer,
2317  qbuffer_size);
2318 
2319  if (qry == NULL)
2320  {
2321  /* Trouble ... drop the text */
2322  entry->query_offset = 0;
2323  entry->query_len = -1;
2324  /* entry will not be counted in mean query length computation */
2325  continue;
2326  }
2327 
2328  if (fwrite(qry, 1, query_len + 1, qfile) != query_len + 1)
2329  {
2330  ereport(LOG,
2332  errmsg("could not write file \"%s\": %m",
2333  PGSS_TEXT_FILE)));
2334  hash_seq_term(&hash_seq);
2335  goto gc_fail;
2336  }
2337 
2338  entry->query_offset = extent;
2339  extent += query_len + 1;
2340  nentries++;
2341  }
2342 
2343  /*
2344  * Truncate away any now-unused space. If this fails for some odd reason,
2345  * we log it, but there's no need to fail.
2346  */
2347  if (ftruncate(fileno(qfile), extent) != 0)
2348  ereport(LOG,
2350  errmsg("could not truncate file \"%s\": %m",
2351  PGSS_TEXT_FILE)));
2352 
2353  if (FreeFile(qfile))
2354  {
2355  ereport(LOG,
2357  errmsg("could not write file \"%s\": %m",
2358  PGSS_TEXT_FILE)));
2359  qfile = NULL;
2360  goto gc_fail;
2361  }
2362 
2363  elog(DEBUG1, "pgss gc of queries file shrunk size from %zu to %zu",
2364  pgss->extent, extent);
2365 
2366  /* Reset the shared extent pointer */
2367  pgss->extent = extent;
2368 
2369  /*
2370  * Also update the mean query length, to be sure that need_gc_qtexts()
2371  * won't still think we have a problem.
2372  */
2373  if (nentries > 0)
2374  pgss->mean_query_len = extent / nentries;
2375  else
2377 
2378  free(qbuffer);
2379 
2380  /*
2381  * OK, count a garbage collection cycle. (Note: even though we have
2382  * exclusive lock on pgss->lock, we must take pgss->mutex for this, since
2383  * other processes may examine gc_count while holding only the mutex.
2384  * Also, we have to advance the count *after* we've rewritten the file,
2385  * else other processes might not realize they read a stale file.)
2386  */
2387  record_gc_qtexts();
2388 
2389  return;
2390 
2391 gc_fail:
2392  /* clean up resources */
2393  if (qfile)
2394  FreeFile(qfile);
2395  if (qbuffer)
2396  free(qbuffer);
2397 
2398  /*
2399  * Since the contents of the external file are now uncertain, mark all
2400  * hashtable entries as having invalid texts.
2401  */
2402  hash_seq_init(&hash_seq, pgss_hash);
2403  while ((entry = hash_seq_search(&hash_seq)) != NULL)
2404  {
2405  entry->query_offset = 0;
2406  entry->query_len = -1;
2407  }
2408 
2409  /*
2410  * Destroy the query text file and create a new, empty one
2411  */
2412  (void) unlink(PGSS_TEXT_FILE);
2414  if (qfile == NULL)
2415  ereport(LOG,
2417  errmsg("could not recreate file \"%s\": %m",
2418  PGSS_TEXT_FILE)));
2419  else
2420  FreeFile(qfile);
2421 
2422  /* Reset the shared extent pointer */
2423  pgss->extent = 0;
2424 
2425  /* Reset mean_query_len to match the new state */
2427 
2428  /*
2429  * Bump the GC count even though we failed.
2430  *
2431  * This is needed to make concurrent readers of file without any lock on
2432  * pgss->lock notice existence of new version of file. Once readers
2433  * subsequently observe a change in GC count with pgss->lock held, that
2434  * forces a safe reopen of file. Writers also require that we bump here,
2435  * of course. (As required by locking protocol, readers and writers don't
2436  * trust earlier file contents until gc_count is found unchanged after
2437  * pgss->lock acquired in shared or exclusive mode respectively.)
2438  */
2439  record_gc_qtexts();
2440 }
2441 
2442 /*
2443  * Release entries corresponding to parameters passed.
2444  */
2445 static void
2447 {
2448  HASH_SEQ_STATUS hash_seq;
2449  pgssEntry *entry;
2450  FILE *qfile;
2451  long num_entries;
2452  long num_remove = 0;
2453  pgssHashKey key;
2454 
2455  if (!pgss || !pgss_hash)
2456  ereport(ERROR,
2457  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2458  errmsg("pg_stat_statements must be loaded via shared_preload_libraries")));
2459 
2461  num_entries = hash_get_num_entries(pgss_hash);
2462 
2463  if (userid != 0 && dbid != 0 && queryid != UINT64CONST(0))
2464  {
2465  /* If all the parameters are available, use the fast path. */
2466  key.userid = userid;
2467  key.dbid = dbid;
2468  key.queryid = queryid;
2469 
2470  /* Remove the key if exists */
2471  entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_REMOVE, NULL);
2472  if (entry) /* found */
2473  num_remove++;
2474  }
2475  else if (userid != 0 || dbid != 0 || queryid != UINT64CONST(0))
2476  {
2477  /* Remove entries corresponding to valid parameters. */
2478  hash_seq_init(&hash_seq, pgss_hash);
2479  while ((entry = hash_seq_search(&hash_seq)) != NULL)
2480  {
2481  if ((!userid || entry->key.userid == userid) &&
2482  (!dbid || entry->key.dbid == dbid) &&
2483  (!queryid || entry->key.queryid == queryid))
2484  {
2485  hash_search(pgss_hash, &entry->key, HASH_REMOVE, NULL);
2486  num_remove++;
2487  }
2488  }
2489  }
2490  else
2491  {
2492  /* Remove all entries. */
2493  hash_seq_init(&hash_seq, pgss_hash);
2494  while ((entry = hash_seq_search(&hash_seq)) != NULL)
2495  {
2496  hash_search(pgss_hash, &entry->key, HASH_REMOVE, NULL);
2497  num_remove++;
2498  }
2499  }
2500 
2501  /* All entries are removed? */
2502  if (num_entries != num_remove)
2503  goto release_lock;
2504 
2505  /*
2506  * Write new empty query file, perhaps even creating a new one to recover
2507  * if the file was missing.
2508  */
2510  if (qfile == NULL)
2511  {
2512  ereport(LOG,
2514  errmsg("could not create file \"%s\": %m",
2515  PGSS_TEXT_FILE)));
2516  goto done;
2517  }
2518 
2519  /* If ftruncate fails, log it, but it's not a fatal problem */
2520  if (ftruncate(fileno(qfile), 0) != 0)
2521  ereport(LOG,
2523  errmsg("could not truncate file \"%s\": %m",
2524  PGSS_TEXT_FILE)));
2525 
2526  FreeFile(qfile);
2527 
2528 done:
2529  pgss->extent = 0;
2530  /* This counts as a query text garbage collection for our purposes */
2531  record_gc_qtexts();
2532 
2533 release_lock:
2534  LWLockRelease(pgss->lock);
2535 }
2536 
2537 /*
2538  * AppendJumble: Append a value that is substantive in a given query to
2539  * the current jumble.
2540  */
2541 static void
2542 AppendJumble(pgssJumbleState *jstate, const unsigned char *item, Size size)
2543 {
2544  unsigned char *jumble = jstate->jumble;
2545  Size jumble_len = jstate->jumble_len;
2546 
2547  /*
2548  * Whenever the jumble buffer is full, we hash the current contents and
2549  * reset the buffer to contain just that hash value, thus relying on the
2550  * hash to summarize everything so far.
2551  */
2552  while (size > 0)
2553  {
2554  Size part_size;
2555 
2556  if (jumble_len >= JUMBLE_SIZE)
2557  {
2558  uint64 start_hash;
2559 
2560  start_hash = DatumGetUInt64(hash_any_extended(jumble,
2561  JUMBLE_SIZE, 0));
2562  memcpy(jumble, &start_hash, sizeof(start_hash));
2563  jumble_len = sizeof(start_hash);
2564  }
2565  part_size = Min(size, JUMBLE_SIZE - jumble_len);
2566  memcpy(jumble + jumble_len, item, part_size);
2567  jumble_len += part_size;
2568  item += part_size;
2569  size -= part_size;
2570  }
2571  jstate->jumble_len = jumble_len;
2572 }
2573 
2574 /*
2575  * Wrappers around AppendJumble to encapsulate details of serialization
2576  * of individual local variable elements.
2577  */
2578 #define APP_JUMB(item) \
2579  AppendJumble(jstate, (const unsigned char *) &(item), sizeof(item))
2580 #define APP_JUMB_STRING(str) \
2581  AppendJumble(jstate, (const unsigned char *) (str), strlen(str) + 1)
2582 
2583 /*
2584  * JumbleQuery: Selectively serialize the query tree, appending significant
2585  * data to the "query jumble" while ignoring nonsignificant data.
2586  *
2587  * Rule of thumb for what to include is that we should ignore anything not
2588  * semantically significant (such as alias names) as well as anything that can
2589  * be deduced from child nodes (else we'd just be double-hashing that piece
2590  * of information).
2591  */
2592 static void
2594 {
2595  Assert(IsA(query, Query));
2596  Assert(query->utilityStmt == NULL);
2597 
2598  APP_JUMB(query->commandType);
2599  /* resultRelation is usually predictable from commandType */
2600  JumbleExpr(jstate, (Node *) query->cteList);
2601  JumbleRangeTable(jstate, query->rtable);
2602  JumbleExpr(jstate, (Node *) query->jointree);
2603  JumbleExpr(jstate, (Node *) query->targetList);
2604  JumbleExpr(jstate, (Node *) query->onConflict);
2605  JumbleExpr(jstate, (Node *) query->returningList);
2606  JumbleExpr(jstate, (Node *) query->groupClause);
2607  JumbleExpr(jstate, (Node *) query->groupingSets);
2608  JumbleExpr(jstate, query->havingQual);
2609  JumbleExpr(jstate, (Node *) query->windowClause);
2610  JumbleExpr(jstate, (Node *) query->distinctClause);
2611  JumbleExpr(jstate, (Node *) query->sortClause);
2612  JumbleExpr(jstate, query->limitOffset);
2613  JumbleExpr(jstate, query->limitCount);
2614  JumbleRowMarks(jstate, query->rowMarks);
2615  JumbleExpr(jstate, query->setOperations);
2616 }
2617 
2618 /*
2619  * Jumble a range table
2620  */
2621 static void
2623 {
2624  ListCell *lc;
2625 
2626  foreach(lc, rtable)
2627  {
2629 
2630  APP_JUMB(rte->rtekind);
2631  switch (rte->rtekind)
2632  {
2633  case RTE_RELATION:
2634  APP_JUMB(rte->relid);
2635  JumbleExpr(jstate, (Node *) rte->tablesample);
2636  break;
2637  case RTE_SUBQUERY:
2638  JumbleQuery(jstate, rte->subquery);
2639  break;
2640  case RTE_JOIN:
2641  APP_JUMB(rte->jointype);
2642  break;
2643  case RTE_FUNCTION:
2644  JumbleExpr(jstate, (Node *) rte->functions);
2645  break;
2646  case RTE_TABLEFUNC:
2647  JumbleExpr(jstate, (Node *) rte->tablefunc);
2648  break;
2649  case RTE_VALUES:
2650  JumbleExpr(jstate, (Node *) rte->values_lists);
2651  break;
2652  case RTE_CTE:
2653 
2654  /*
2655  * Depending on the CTE name here isn't ideal, but it's the
2656  * only info we have to identify the referenced WITH item.
2657  */
2658  APP_JUMB_STRING(rte->ctename);
2659  APP_JUMB(rte->ctelevelsup);
2660  break;
2661  case RTE_NAMEDTUPLESTORE:
2662  APP_JUMB_STRING(rte->enrname);
2663  break;
2664  case RTE_RESULT:
2665  break;
2666  default:
2667  elog(ERROR, "unrecognized RTE kind: %d", (int) rte->rtekind);
2668  break;
2669  }
2670  }
2671 }
2672 
2673 /*
2674  * Jumble a rowMarks list
2675  */
2676 static void
2678 {
2679  ListCell *lc;
2680 
2681  foreach(lc, rowMarks)
2682  {
2683  RowMarkClause *rowmark = lfirst_node(RowMarkClause, lc);
2684 
2685  if (!rowmark->pushedDown)
2686  {
2687  APP_JUMB(rowmark->rti);
2688  APP_JUMB(rowmark->strength);
2689  APP_JUMB(rowmark->waitPolicy);
2690  }
2691  }
2692 }
2693 
2694 /*
2695  * Jumble an expression tree
2696  *
2697  * In general this function should handle all the same node types that
2698  * expression_tree_walker() does, and therefore it's coded to be as parallel
2699  * to that function as possible. However, since we are only invoked on
2700  * queries immediately post-parse-analysis, we need not handle node types
2701  * that only appear in planning.
2702  *
2703  * Note: the reason we don't simply use expression_tree_walker() is that the
2704  * point of that function is to support tree walkers that don't care about
2705  * most tree node types, but here we care about all types. We should complain
2706  * about any unrecognized node type.
2707  */
2708 static void
2710 {
2711  ListCell *temp;
2712 
2713  if (node == NULL)
2714  return;
2715 
2716  /* Guard against stack overflow due to overly complex expressions */
2718 
2719  /*
2720  * We always emit the node's NodeTag, then any additional fields that are
2721  * considered significant, and then we recurse to any child nodes.
2722  */
2723  APP_JUMB(node->type);
2724 
2725  switch (nodeTag(node))
2726  {
2727  case T_Var:
2728  {
2729  Var *var = (Var *) node;
2730 
2731  APP_JUMB(var->varno);
2732  APP_JUMB(var->varattno);
2733  APP_JUMB(var->varlevelsup);
2734  }
2735  break;
2736  case T_Const:
2737  {
2738  Const *c = (Const *) node;
2739 
2740  /* We jumble only the constant's type, not its value */
2741  APP_JUMB(c->consttype);
2742  /* Also, record its parse location for query normalization */
2743  RecordConstLocation(jstate, c->location);
2744  }
2745  break;
2746  case T_Param:
2747  {
2748  Param *p = (Param *) node;
2749 
2750  APP_JUMB(p->paramkind);
2751  APP_JUMB(p->paramid);
2752  APP_JUMB(p->paramtype);
2753  /* Also, track the highest external Param id */
2754  if (p->paramkind == PARAM_EXTERN &&
2755  p->paramid > jstate->highest_extern_param_id)
2756  jstate->highest_extern_param_id = p->paramid;
2757  }
2758  break;
2759  case T_Aggref:
2760  {
2761  Aggref *expr = (Aggref *) node;
2762 
2763  APP_JUMB(expr->aggfnoid);
2764  JumbleExpr(jstate, (Node *) expr->aggdirectargs);
2765  JumbleExpr(jstate, (Node *) expr->args);
2766  JumbleExpr(jstate, (Node *) expr->aggorder);
2767  JumbleExpr(jstate, (Node *) expr->aggdistinct);
2768  JumbleExpr(jstate, (Node *) expr->aggfilter);
2769  }
2770  break;
2771  case T_GroupingFunc:
2772  {
2773  GroupingFunc *grpnode = (GroupingFunc *) node;
2774 
2775  JumbleExpr(jstate, (Node *) grpnode->refs);
2776  }
2777  break;
2778  case T_WindowFunc:
2779  {
2780  WindowFunc *expr = (WindowFunc *) node;
2781 
2782  APP_JUMB(expr->winfnoid);
2783  APP_JUMB(expr->winref);
2784  JumbleExpr(jstate, (Node *) expr->args);
2785  JumbleExpr(jstate, (Node *) expr->aggfilter);
2786  }
2787  break;
2788  case T_SubscriptingRef:
2789  {
2790  SubscriptingRef *sbsref = (SubscriptingRef *) node;
2791 
2792  JumbleExpr(jstate, (Node *) sbsref->refupperindexpr);
2793  JumbleExpr(jstate, (Node *) sbsref->reflowerindexpr);
2794  JumbleExpr(jstate, (Node *) sbsref->refexpr);
2795  JumbleExpr(jstate, (Node *) sbsref->refassgnexpr);
2796  }
2797  break;
2798  case T_FuncExpr:
2799  {
2800  FuncExpr *expr = (FuncExpr *) node;
2801 
2802  APP_JUMB(expr->funcid);
2803  JumbleExpr(jstate, (Node *) expr->args);
2804  }
2805  break;
2806  case T_NamedArgExpr:
2807  {
2808  NamedArgExpr *nae = (NamedArgExpr *) node;
2809 
2810  APP_JUMB(nae->argnumber);
2811  JumbleExpr(jstate, (Node *) nae->arg);
2812  }
2813  break;
2814  case T_OpExpr:
2815  case T_DistinctExpr: /* struct-equivalent to OpExpr */
2816  case T_NullIfExpr: /* struct-equivalent to OpExpr */
2817  {
2818  OpExpr *expr = (OpExpr *) node;
2819 
2820  APP_JUMB(expr->opno);
2821  JumbleExpr(jstate, (Node *) expr->args);
2822  }
2823  break;
2824  case T_ScalarArrayOpExpr:
2825  {
2826  ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node;
2827 
2828  APP_JUMB(expr->opno);
2829  APP_JUMB(expr->useOr);
2830  JumbleExpr(jstate, (Node *) expr->args);
2831  }
2832  break;
2833  case T_BoolExpr:
2834  {
2835  BoolExpr *expr = (BoolExpr *) node;
2836 
2837  APP_JUMB(expr->boolop);
2838  JumbleExpr(jstate, (Node *) expr->args);
2839  }
2840  break;
2841  case T_SubLink:
2842  {
2843  SubLink *sublink = (SubLink *) node;
2844 
2845  APP_JUMB(sublink->subLinkType);
2846  APP_JUMB(sublink->subLinkId);
2847  JumbleExpr(jstate, (Node *) sublink->testexpr);
2848  JumbleQuery(jstate, castNode(Query, sublink->subselect));
2849  }
2850  break;
2851  case T_FieldSelect:
2852  {
2853  FieldSelect *fs = (FieldSelect *) node;
2854 
2855  APP_JUMB(fs->fieldnum);
2856  JumbleExpr(jstate, (Node *) fs->arg);
2857  }
2858  break;
2859  case T_FieldStore:
2860  {
2861  FieldStore *fstore = (FieldStore *) node;
2862 
2863  JumbleExpr(jstate, (Node *) fstore->arg);
2864  JumbleExpr(jstate, (Node *) fstore->newvals);
2865  }
2866  break;
2867  case T_RelabelType:
2868  {
2869  RelabelType *rt = (RelabelType *) node;
2870 
2871  APP_JUMB(rt->resulttype);
2872  JumbleExpr(jstate, (Node *) rt->arg);
2873  }
2874  break;
2875  case T_CoerceViaIO:
2876  {
2877  CoerceViaIO *cio = (CoerceViaIO *) node;
2878 
2879  APP_JUMB(cio->resulttype);
2880  JumbleExpr(jstate, (Node *) cio->arg);
2881  }
2882  break;
2883  case T_ArrayCoerceExpr:
2884  {
2885  ArrayCoerceExpr *acexpr = (ArrayCoerceExpr *) node;
2886 
2887  APP_JUMB(acexpr->resulttype);
2888  JumbleExpr(jstate, (Node *) acexpr->arg);
2889  JumbleExpr(jstate, (Node *) acexpr->elemexpr);
2890  }
2891  break;
2892  case T_ConvertRowtypeExpr:
2893  {
2894  ConvertRowtypeExpr *crexpr = (ConvertRowtypeExpr *) node;
2895 
2896  APP_JUMB(crexpr->resulttype);
2897  JumbleExpr(jstate, (Node *) crexpr->arg);
2898  }
2899  break;
2900  case T_CollateExpr:
2901  {
2902  CollateExpr *ce = (CollateExpr *) node;
2903 
2904  APP_JUMB(ce->collOid);
2905  JumbleExpr(jstate, (Node *) ce->arg);
2906  }
2907  break;
2908  case T_CaseExpr:
2909  {
2910  CaseExpr *caseexpr = (CaseExpr *) node;
2911 
2912  JumbleExpr(jstate, (Node *) caseexpr->arg);
2913  foreach(temp, caseexpr->args)
2914  {
2915  CaseWhen *when = lfirst_node(CaseWhen, temp);
2916 
2917  JumbleExpr(jstate, (Node *) when->expr);
2918  JumbleExpr(jstate, (Node *) when->result);
2919  }
2920  JumbleExpr(jstate, (Node *) caseexpr->defresult);
2921  }
2922  break;
2923  case T_CaseTestExpr:
2924  {
2925  CaseTestExpr *ct = (CaseTestExpr *) node;
2926 
2927  APP_JUMB(ct->typeId);
2928  }
2929  break;
2930  case T_ArrayExpr:
2931  JumbleExpr(jstate, (Node *) ((ArrayExpr *) node)->elements);
2932  break;
2933  case T_RowExpr:
2934  JumbleExpr(jstate, (Node *) ((RowExpr *) node)->args);
2935  break;
2936  case T_RowCompareExpr:
2937  {
2938  RowCompareExpr *rcexpr = (RowCompareExpr *) node;
2939 
2940  APP_JUMB(rcexpr->rctype);
2941  JumbleExpr(jstate, (Node *) rcexpr->largs);
2942  JumbleExpr(jstate, (Node *) rcexpr->rargs);
2943  }
2944  break;
2945  case T_CoalesceExpr:
2946  JumbleExpr(jstate, (Node *) ((CoalesceExpr *) node)->args);
2947  break;
2948  case T_MinMaxExpr:
2949  {
2950  MinMaxExpr *mmexpr = (MinMaxExpr *) node;
2951 
2952  APP_JUMB(mmexpr->op);
2953  JumbleExpr(jstate, (Node *) mmexpr->args);
2954  }
2955  break;
2956  case T_SQLValueFunction:
2957  {
2958  SQLValueFunction *svf = (SQLValueFunction *) node;
2959 
2960  APP_JUMB(svf->op);
2961  /* type is fully determined by op */
2962  APP_JUMB(svf->typmod);
2963  }
2964  break;
2965  case T_XmlExpr:
2966  {
2967  XmlExpr *xexpr = (XmlExpr *) node;
2968 
2969  APP_JUMB(xexpr->op);
2970  JumbleExpr(jstate, (Node *) xexpr->named_args);
2971  JumbleExpr(jstate, (Node *) xexpr->args);
2972  }
2973  break;
2974  case T_NullTest:
2975  {
2976  NullTest *nt = (NullTest *) node;
2977 
2978  APP_JUMB(nt->nulltesttype);
2979  JumbleExpr(jstate, (Node *) nt->arg);
2980  }
2981  break;
2982  case T_BooleanTest:
2983  {
2984  BooleanTest *bt = (BooleanTest *) node;
2985 
2986  APP_JUMB(bt->booltesttype);
2987  JumbleExpr(jstate, (Node *) bt->arg);
2988  }
2989  break;
2990  case T_CoerceToDomain:
2991  {
2992  CoerceToDomain *cd = (CoerceToDomain *) node;
2993 
2994  APP_JUMB(cd->resulttype);
2995  JumbleExpr(jstate, (Node *) cd->arg);
2996  }
2997  break;
2998  case T_CoerceToDomainValue:
2999  {
3000  CoerceToDomainValue *cdv = (CoerceToDomainValue *) node;
3001 
3002  APP_JUMB(cdv->typeId);
3003  }
3004  break;
3005  case T_SetToDefault:
3006  {
3007  SetToDefault *sd = (SetToDefault *) node;
3008 
3009  APP_JUMB(sd->typeId);
3010  }
3011  break;
3012  case T_CurrentOfExpr:
3013  {
3014  CurrentOfExpr *ce = (CurrentOfExpr *) node;
3015 
3016  APP_JUMB(ce->cvarno);
3017  if (ce->cursor_name)
3019  APP_JUMB(ce->cursor_param);
3020  }
3021  break;
3022  case T_NextValueExpr:
3023  {
3024  NextValueExpr *nve = (NextValueExpr *) node;
3025 
3026  APP_JUMB(nve->seqid);
3027  APP_JUMB(nve->typeId);
3028  }
3029  break;
3030  case T_InferenceElem:
3031  {
3032  InferenceElem *ie = (InferenceElem *) node;
3033 
3034  APP_JUMB(ie->infercollid);
3035  APP_JUMB(ie->inferopclass);
3036  JumbleExpr(jstate, ie->expr);
3037  }
3038  break;
3039  case T_TargetEntry:
3040  {
3041  TargetEntry *tle = (TargetEntry *) node;
3042 
3043  APP_JUMB(tle->resno);
3044  APP_JUMB(tle->ressortgroupref);
3045  JumbleExpr(jstate, (Node *) tle->expr);
3046  }
3047  break;
3048  case T_RangeTblRef:
3049  {
3050  RangeTblRef *rtr = (RangeTblRef *) node;
3051 
3052  APP_JUMB(rtr->rtindex);
3053  }
3054  break;
3055  case T_JoinExpr:
3056  {
3057  JoinExpr *join = (JoinExpr *) node;
3058 
3059  APP_JUMB(join->jointype);
3060  APP_JUMB(join->isNatural);
3061  APP_JUMB(join->rtindex);
3062  JumbleExpr(jstate, join->larg);
3063  JumbleExpr(jstate, join->rarg);
3064  JumbleExpr(jstate, join->quals);
3065  }
3066  break;
3067  case T_FromExpr:
3068  {
3069  FromExpr *from = (FromExpr *) node;
3070 
3071  JumbleExpr(jstate, (Node *) from->fromlist);
3072  JumbleExpr(jstate, from->quals);
3073  }
3074  break;
3075  case T_OnConflictExpr:
3076  {
3077  OnConflictExpr *conf = (OnConflictExpr *) node;
3078 
3079  APP_JUMB(conf->action);
3080  JumbleExpr(jstate, (Node *) conf->arbiterElems);
3081  JumbleExpr(jstate, conf->arbiterWhere);
3082  JumbleExpr(jstate, (Node *) conf->onConflictSet);
3083  JumbleExpr(jstate, conf->onConflictWhere);
3084  APP_JUMB(conf->constraint);
3085  APP_JUMB(conf->exclRelIndex);
3086  JumbleExpr(jstate, (Node *) conf->exclRelTlist);
3087  }
3088  break;
3089  case T_List:
3090  foreach(temp, (List *) node)
3091  {
3092  JumbleExpr(jstate, (Node *) lfirst(temp));
3093  }
3094  break;
3095  case T_IntList:
3096  foreach(temp, (List *) node)
3097  {
3098  APP_JUMB(lfirst_int(temp));
3099  }
3100  break;
3101  case T_SortGroupClause:
3102  {
3103  SortGroupClause *sgc = (SortGroupClause *) node;
3104 
3105  APP_JUMB(sgc->tleSortGroupRef);
3106  APP_JUMB(sgc->eqop);
3107  APP_JUMB(sgc->sortop);
3108  APP_JUMB(sgc->nulls_first);
3109  }
3110  break;
3111  case T_GroupingSet:
3112  {
3113  GroupingSet *gsnode = (GroupingSet *) node;
3114 
3115  JumbleExpr(jstate, (Node *) gsnode->content);
3116  }
3117  break;
3118  case T_WindowClause:
3119  {
3120  WindowClause *wc = (WindowClause *) node;
3121 
3122  APP_JUMB(wc->winref);
3123  APP_JUMB(wc->frameOptions);
3124  JumbleExpr(jstate, (Node *) wc->partitionClause);
3125  JumbleExpr(jstate, (Node *) wc->orderClause);
3126  JumbleExpr(jstate, wc->startOffset);
3127  JumbleExpr(jstate, wc->endOffset);
3128  }
3129  break;
3130  case T_CommonTableExpr:
3131  {
3132  CommonTableExpr *cte = (CommonTableExpr *) node;
3133 
3134  /* we store the string name because RTE_CTE RTEs need it */
3135  APP_JUMB_STRING(cte->ctename);
3136  APP_JUMB(cte->ctematerialized);
3137  JumbleQuery(jstate, castNode(Query, cte->ctequery));
3138  }
3139  break;
3140  case T_SetOperationStmt:
3141  {
3142  SetOperationStmt *setop = (SetOperationStmt *) node;
3143 
3144  APP_JUMB(setop->op);
3145  APP_JUMB(setop->all);
3146  JumbleExpr(jstate, setop->larg);
3147  JumbleExpr(jstate, setop->rarg);
3148  }
3149  break;
3150  case T_RangeTblFunction:
3151  {
3152  RangeTblFunction *rtfunc = (RangeTblFunction *) node;
3153 
3154  JumbleExpr(jstate, rtfunc->funcexpr);
3155  }
3156  break;
3157  case T_TableFunc:
3158  {
3159  TableFunc *tablefunc = (TableFunc *) node;
3160 
3161  JumbleExpr(jstate, tablefunc->docexpr);
3162  JumbleExpr(jstate, tablefunc->rowexpr);
3163  JumbleExpr(jstate, (Node *) tablefunc->colexprs);
3164  }
3165  break;
3166  case T_TableSampleClause:
3167  {
3168  TableSampleClause *tsc = (TableSampleClause *) node;
3169 
3170  APP_JUMB(tsc->tsmhandler);
3171  JumbleExpr(jstate, (Node *) tsc->args);
3172  JumbleExpr(jstate, (Node *) tsc->repeatable);
3173  }
3174  break;
3175  default:
3176  /* Only a warning, since we can stumble along anyway */
3177  elog(WARNING, "unrecognized node type: %d",
3178  (int) nodeTag(node));
3179  break;
3180  }
3181 }
3182 
3183 /*
3184  * Record location of constant within query string of query tree
3185  * that is currently being walked.
3186  */
3187 static void
3188 RecordConstLocation(pgssJumbleState *jstate, int location)
3189 {
3190  /* -1 indicates unknown or undefined location */
3191  if (location >= 0)
3192  {
3193  /* enlarge array if needed */
3194  if (jstate->clocations_count >= jstate->clocations_buf_size)
3195  {
3196  jstate->clocations_buf_size *= 2;
3197  jstate->clocations = (pgssLocationLen *)
3198  repalloc(jstate->clocations,
3199  jstate->clocations_buf_size *
3200  sizeof(pgssLocationLen));
3201  }
3202  jstate->clocations[jstate->clocations_count].location = location;
3203  /* initialize lengths to -1 to simplify fill_in_constant_lengths */
3204  jstate->clocations[jstate->clocations_count].length = -1;
3205  jstate->clocations_count++;
3206  }
3207 }
3208 
3209 /*
3210  * Generate a normalized version of the query string that will be used to
3211  * represent all similar queries.
3212  *
3213  * Note that the normalized representation may well vary depending on
3214  * just which "equivalent" query is used to create the hashtable entry.
3215  * We assume this is OK.
3216  *
3217  * If query_loc > 0, then "query" has been advanced by that much compared to
3218  * the original string start, so we need to translate the provided locations
3219  * to compensate. (This lets us avoid re-scanning statements before the one
3220  * of interest, so it's worth doing.)
3221  *
3222  * *query_len_p contains the input string length, and is updated with
3223  * the result string length on exit. The resulting string might be longer
3224  * or shorter depending on what happens with replacement of constants.
3225  *
3226  * Returns a palloc'd string.
3227  */
3228 static char *
3229 generate_normalized_query(pgssJumbleState *jstate, const char *query,
3230  int query_loc, int *query_len_p, int encoding)
3231 {
3232  char *norm_query;
3233  int query_len = *query_len_p;
3234  int i,
3235  norm_query_buflen, /* Space allowed for norm_query */
3236  len_to_wrt, /* Length (in bytes) to write */
3237  quer_loc = 0, /* Source query byte location */
3238  n_quer_loc = 0, /* Normalized query byte location */
3239  last_off = 0, /* Offset from start for previous tok */
3240  last_tok_len = 0; /* Length (in bytes) of that tok */
3241 
3242  /*
3243  * Get constants' lengths (core system only gives us locations). Note
3244  * this also ensures the items are sorted by location.
3245  */
3246  fill_in_constant_lengths(jstate, query, query_loc);
3247 
3248  /*
3249  * Allow for $n symbols to be longer than the constants they replace.
3250  * Constants must take at least one byte in text form, while a $n symbol
3251  * certainly isn't more than 11 bytes, even if n reaches INT_MAX. We
3252  * could refine that limit based on the max value of n for the current
3253  * query, but it hardly seems worth any extra effort to do so.
3254  */
3255  norm_query_buflen = query_len + jstate->clocations_count * 10;
3256 
3257  /* Allocate result buffer */
3258  norm_query = palloc(norm_query_buflen + 1);
3259 
3260  for (i = 0; i < jstate->clocations_count; i++)
3261  {
3262  int off, /* Offset from start for cur tok */
3263  tok_len; /* Length (in bytes) of that tok */
3264 
3265  off = jstate->clocations[i].location;
3266  /* Adjust recorded location if we're dealing with partial string */
3267  off -= query_loc;
3268 
3269  tok_len = jstate->clocations[i].length;
3270 
3271  if (tok_len < 0)
3272  continue; /* ignore any duplicates */
3273 
3274  /* Copy next chunk (what precedes the next constant) */
3275  len_to_wrt = off - last_off;
3276  len_to_wrt -= last_tok_len;
3277 
3278  Assert(len_to_wrt >= 0);
3279  memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
3280  n_quer_loc += len_to_wrt;
3281 
3282  /* And insert a param symbol in place of the constant token */
3283  n_quer_loc += sprintf(norm_query + n_quer_loc, "$%d",
3284  i + 1 + jstate->highest_extern_param_id);
3285 
3286  quer_loc = off + tok_len;
3287  last_off = off;
3288  last_tok_len = tok_len;
3289  }
3290 
3291  /*
3292  * We've copied up until the last ignorable constant. Copy over the
3293  * remaining bytes of the original query string.
3294  */
3295  len_to_wrt = query_len - quer_loc;
3296 
3297  Assert(len_to_wrt >= 0);
3298  memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
3299  n_quer_loc += len_to_wrt;
3300 
3301  Assert(n_quer_loc <= norm_query_buflen);
3302  norm_query[n_quer_loc] = '\0';
3303 
3304  *query_len_p = n_quer_loc;
3305  return norm_query;
3306 }
3307 
3308 /*
3309  * Given a valid SQL string and an array of constant-location records,
3310  * fill in the textual lengths of those constants.
3311  *
3312  * The constants may use any allowed constant syntax, such as float literals,
3313  * bit-strings, single-quoted strings and dollar-quoted strings. This is
3314  * accomplished by using the public API for the core scanner.
3315  *
3316  * It is the caller's job to ensure that the string is a valid SQL statement
3317  * with constants at the indicated locations. Since in practice the string
3318  * has already been parsed, and the locations that the caller provides will
3319  * have originated from within the authoritative parser, this should not be
3320  * a problem.
3321  *
3322  * Duplicate constant pointers are possible, and will have their lengths
3323  * marked as '-1', so that they are later ignored. (Actually, we assume the
3324  * lengths were initialized as -1 to start with, and don't change them here.)
3325  *
3326  * If query_loc > 0, then "query" has been advanced by that much compared to
3327  * the original string start, so we need to translate the provided locations
3328  * to compensate. (This lets us avoid re-scanning statements before the one
3329  * of interest, so it's worth doing.)
3330  *
3331  * N.B. There is an assumption that a '-' character at a Const location begins
3332  * a negative numeric constant. This precludes there ever being another
3333  * reason for a constant to start with a '-'.
3334  */
3335 static void
3336 fill_in_constant_lengths(pgssJumbleState *jstate, const char *query,
3337  int query_loc)
3338 {
3339  pgssLocationLen *locs;
3341  core_yy_extra_type yyextra;
3342  core_YYSTYPE yylval;
3343  YYLTYPE yylloc;
3344  int last_loc = -1;
3345  int i;
3346 
3347  /*
3348  * Sort the records by location so that we can process them in order while
3349  * scanning the query text.
3350  */
3351  if (jstate->clocations_count > 1)
3352  qsort(jstate->clocations, jstate->clocations_count,
3353  sizeof(pgssLocationLen), comp_location);
3354  locs = jstate->clocations;
3355 
3356  /* initialize the flex scanner --- should match raw_parser() */
3357  yyscanner = scanner_init(query,
3358  &yyextra,
3359  &ScanKeywords,
3361 
3362  /* we don't want to re-emit any escape string warnings */
3363  yyextra.escape_string_warning = false;
3364 
3365  /* Search for each constant, in sequence */
3366  for (i = 0; i < jstate->clocations_count; i++)
3367  {
3368  int loc = locs[i].location;
3369  int tok;
3370 
3371  /* Adjust recorded location if we're dealing with partial string */
3372  loc -= query_loc;
3373 
3374  Assert(loc >= 0);
3375 
3376  if (loc <= last_loc)
3377  continue; /* Duplicate constant, ignore */
3378 
3379  /* Lex tokens until we find the desired constant */
3380  for (;;)
3381  {
3382  tok = core_yylex(&yylval, &yylloc, yyscanner);
3383 
3384  /* We should not hit end-of-string, but if we do, behave sanely */
3385  if (tok == 0)
3386  break; /* out of inner for-loop */
3387 
3388  /*
3389  * We should find the token position exactly, but if we somehow
3390  * run past it, work with that.
3391  */
3392  if (yylloc >= loc)
3393  {
3394  if (query[loc] == '-')
3395  {
3396  /*
3397  * It's a negative value - this is the one and only case
3398  * where we replace more than a single token.
3399  *
3400  * Do not compensate for the core system's special-case
3401  * adjustment of location to that of the leading '-'
3402  * operator in the event of a negative constant. It is
3403  * also useful for our purposes to start from the minus
3404  * symbol. In this way, queries like "select * from foo
3405  * where bar = 1" and "select * from foo where bar = -2"
3406  * will have identical normalized query strings.
3407  */
3408  tok = core_yylex(&yylval, &yylloc, yyscanner);
3409  if (tok == 0)
3410  break; /* out of inner for-loop */
3411  }
3412 
3413  /*
3414  * We now rely on the assumption that flex has placed a zero
3415  * byte after the text of the current token in scanbuf.
3416  */
3417  locs[i].length = strlen(yyextra.scanbuf + loc);
3418  break; /* out of inner for-loop */
3419  }
3420  }
3421 
3422  /* If we hit end-of-string, give up, leaving remaining lengths -1 */
3423  if (tok == 0)
3424  break;
3425 
3426  last_loc = loc;
3427  }
3428 
3429  scanner_finish(yyscanner);
3430 }
3431 
3432 /*
3433  * comp_location: comparator for qsorting pgssLocationLen structs by location
3434  */
3435 static int
3436 comp_location(const void *a, const void *b)
3437 {
3438  int l = ((const pgssLocationLen *) a)->location;
3439  int r = ((const pgssLocationLen *) b)->location;
3440 
3441  if (l < r)
3442  return -1;
3443  else if (l > r)
3444  return +1;
3445  else
3446  return 0;
3447 }
void tuplestore_putvalues(Tuplestorestate *state, TupleDesc tdesc, Datum *values, bool *isnull)
Definition: tuplestore.c:750
List * aggdistinct
Definition: primnodes.h:321
Node * limitOffset
Definition: parsenodes.h:160
int slock_t
Definition: s_lock.h:934
void DefineCustomIntVariable(const char *name, const char *short_desc, const char *long_desc, int *valueAddr, int bootValue, int minValue, int maxValue, GucContext context, int flags, GucIntCheckHook check_hook, GucIntAssignHook assign_hook, GucShowHook show_hook)
Definition: guc.c:8818
void(* ExecutorRun_hook_type)(QueryDesc *queryDesc, ScanDirection direction, uint64 count, bool execute_once)
Definition: executor.h:69
long local_blks_hit
Definition: instrument.h:25
List * args
Definition: primnodes.h:1106
void standard_ExecutorRun(QueryDesc *queryDesc, ScanDirection direction, uint64 count, bool execute_once)
Definition: execMain.c:312
void _PG_init(void)
Definition: lwlock.h:32
static uint64 pgss_hash_string(const char *str, int len)
static void JumbleExpr(pgssJumbleState *jstate, Node *node)
#define IsA(nodeptr, _type_)
Definition: nodes.h:580
Node * docexpr
Definition: primnodes.h:87
void RequestAddinShmemSpace(Size size)
Definition: ipci.c:70
long local_blks_dirtied
Definition: instrument.h:27
#define PG_STAT_STATEMENTS_COLS_V1_3
#define DEBUG1
Definition: elog.h:25
TypeFuncClass get_call_result_type(FunctionCallInfo fcinfo, Oid *resultTypeId, TupleDesc *resultTupleDesc)
Definition: funcapi.c:205
Expr * arg
Definition: primnodes.h:791
Index varlevelsup
Definition: primnodes.h:191
void(* ProcessUtility_hook_type)(PlannedStmt *pstmt, const char *queryString, ProcessUtilityContext context, ParamListInfo params, QueryEnvironment *queryEnv, DestReceiver *dest, QueryCompletion *qc)
Definition: utility.h:71
double total_time[PGSS_NUMKIND]
long local_blks_read
Definition: instrument.h:26
WalUsage walusage
Definition: instrument.h:75
#define PG_STAT_STATEMENTS_COLS_V1_8
int stmt_location
Definition: parsenodes.h:181
List * sortClause
Definition: parsenodes.h:158
EState * estate
Definition: execdesc.h:48
List * content
Definition: parsenodes.h:1327
List * refs
Definition: primnodes.h:361
void * core_yyscan_t
Definition: scanner.h:121
List * args
Definition: primnodes.h:377
#define HASH_ELEM
Definition: hsearch.h:87
List * args
Definition: primnodes.h:477
static void JumbleRowMarks(pgssJumbleState *jstate, List *rowMarks)
static void error(void)
Definition: sql-dyntest.c:147
#define PG_STAT_STATEMENTS_COLS_V1_0
bool process_shared_preload_libraries_in_progress
Definition: miscinit.c:1568
#define USAGE_DEALLOC_PERCENT
int64 shared_blks_read
static const uint32 PGSS_PG_MAJOR_VERSION
FromExpr * jointree
Definition: parsenodes.h:138
int64 local_blks_written
Definition: guc.h:164
void RequestNamedLWLockTranche(const char *tranche_name, int num_lwlocks)
Definition: lwlock.c:705
Oid GetUserId(void)
Definition: miscinit.c:448
OnConflictExpr * onConflict
Definition: parsenodes.h:144
instr_time blk_read_time
Definition: instrument.h:31
#define castNode(_type_, nodeptr)
Definition: nodes.h:598
static ProcessUtility_hook_type prev_ProcessUtility
struct pgssSharedState pgssSharedState
static void AppendJumble(pgssJumbleState *jstate, const unsigned char *item, Size size)
WalUsage pgWalUsage
Definition: instrument.c:22
int64 shared_blks_dirtied
Datum pg_stat_statements_1_8(PG_FUNCTION_ARGS)
static void pgss_shmem_startup(void)
#define SpinLockInit(lock)
Definition: spin.h:60
#define INSTR_TIME_GET_MILLISEC(t)
Definition: instr_time.h:202
Oid resulttype
Definition: primnodes.h:835
PGDLLIMPORT const uint16 ScanKeywordTokens[]
struct timeval instr_time
Definition: instr_time.h:150
PlannedStmt *(* planner_hook_type)(Query *parse, const char *query_string, int cursorOptions, ParamListInfo boundParams)
Definition: planner.h:26
RowCompareType rctype
Definition: primnodes.h:1070
long shared_blks_read
Definition: instrument.h:22
void standard_ExecutorStart(QueryDesc *queryDesc, int eflags)
Definition: execMain.c:152
#define Min(x, y)
Definition: c.h:920
double max_time[PGSS_NUMKIND]
#define tuplestore_donestoring(state)
Definition: tuplestore.h:60
void WalUsageAccumDiff(WalUsage *dst, const WalUsage *add, const WalUsage *sub)
Definition: instrument.c:255
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
static ExecutorRun_hook_type prev_ExecutorRun
Index tleSortGroupRef
Definition: parsenodes.h:1257
Expr * arg
Definition: primnodes.h:814
List * groupingSets
Definition: parsenodes.h:150
static ExecutorEnd_hook_type prev_ExecutorEnd
Size entrysize
Definition: hsearch.h:73
#define PGSS_DUMP_FILE
ParamKind paramkind
Definition: primnodes.h:262
ProcessUtility_hook_type ProcessUtility_hook
Definition: utility.c:75
#define pgss_enabled(level)
Definition: nodes.h:529
static int pgss_track
static int entry_cmp(const void *lhs, const void *rhs)
int errcode(int sqlerrcode)
Definition: elog.c:610
void standard_ExecutorEnd(QueryDesc *queryDesc)
Definition: execMain.c:471
struct pgssEntry pgssEntry
#define PG_BINARY_W
Definition: c.h:1237
ProcessUtilityContext
Definition: utility.h:20
Instrumentation * InstrAlloc(int n, int instrument_options)
Definition: instrument.c:31
List * args
Definition: primnodes.h:319
static ExecutorStart_hook_type prev_ExecutorStart
void(* ExecutorFinish_hook_type)(QueryDesc *queryDesc)
Definition: executor.h:76
AttrNumber varattno
Definition: primnodes.h:186
Expr * arg
Definition: primnodes.h:762
long hash_get_num_entries(HTAB *hashp)
Definition: dynahash.c:1337
static void pgss_store(const char *query, uint64 queryId, int query_location, int query_len, pgssStoreKind kind, double total_time, uint64 rows, const BufferUsage *bufusage, const WalUsage *walusage, pgssJumbleState *jstate)
List * fromlist
Definition: primnodes.h:1510
unsigned char * jumble
#define PG_GETARG_BOOL(n)
Definition: fmgr.h:274
long temp_blks_written
Definition: instrument.h:30
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:908
Counters counters
static void pgss_ExecutorStart(QueryDesc *queryDesc, int eflags)
static int exec_nested_level
int64 temp_blks_read
#define JUMBLE_SIZE
#define PG_STAT_STATEMENTS_COLS_V1_1
#define LOG
Definition: elog.h:26
unsigned int Oid
Definition: postgres_ext.h:31
List * rowMarks
Definition: parsenodes.h:164
Index winref
Definition: primnodes.h:379
Node * utilityStmt
Definition: parsenodes.h:120
static post_parse_analyze_hook_type prev_post_parse_analyze_hook
Definition: primnodes.h:181
static Datum hash_any_extended(const unsigned char *k, int keylen, uint64 seed)
Definition: hashfn.h:37
List * refupperindexpr
Definition: primnodes.h:422
#define PG_BINARY_R
Definition: c.h:1236
static int fd(const char *x, int i)
Definition: preproc-init.c:105
#define PG_BINARY
Definition: c.h:1234
int stmt_len
Definition: plannodes.h:96
char * scanbuf
Definition: scanner.h:72
List * values_lists
Definition: parsenodes.h:1074
Node * quals
Definition: primnodes.h:1511
ssize_t pg_pwrite(int fd, const void *buf, size_t nbyte, off_t offset)
Definition: pwrite.c:27
void standard_ExecutorFinish(QueryDesc *queryDesc)
Definition: execMain.c:411
SQLValueFunctionOp op
Definition: primnodes.h:1143
int duration
Definition: pgbench.c:146
LockClauseStrength strength
Definition: parsenodes.h:1383
Datum pg_stat_statements_1_3(PG_FUNCTION_ARGS)
void DefineCustomEnumVariable(const char *name, const char *short_desc, const char *long_desc, int *valueAddr, int bootValue, const struct config_enum_entry *options, GucContext context, int flags, GucEnumCheckHook check_hook, GucEnumAssignHook assign_hook, GucShowHook show_hook)
Definition: guc.c:8903
#define MaxAllocHugeSize
Definition: memutils.h:44
signed int int32
Definition: c.h:355
#define APP_JUMB_STRING(str)
List * windowClause
Definition: parsenodes.h:154
void BufferUsageAccumDiff(BufferUsage *dst, const BufferUsage *add, const BufferUsage *sub)
Definition: instrument.c:225
List * targetList
Definition: parsenodes.h:140
static void RecordConstLocation(pgssJumbleState *jstate, int location)
List * arbiterElems
Definition: primnodes.h:1529
static void JumbleRangeTable(pgssJumbleState *jstate, List *rtable)
ExecutorStart_hook_type ExecutorStart_hook
Definition: execMain.c:70
static planner_hook_type prev_planner_hook
void InstrEndLoop(Instrumentation *instr)
Definition: instrument.c:121
Node * larg
Definition: primnodes.h:1490
PGDLLIMPORT const ScanKeywordList ScanKeywords
#define malloc(a)
Definition: header.h:50
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1812
#define ASSUMED_LENGTH_INIT
static bool pgss_track_planning
Oid consttype
Definition: primnodes.h:210
#define sprintf
Definition: port.h:195
long shared_blks_written
Definition: instrument.h:24
static pgssSharedState * pgss
#define SpinLockAcquire(lock)
Definition: spin.h:62
Definition: dynahash.c:210
static void pgss_ExecutorRun(QueryDesc *queryDesc, ScanDirection direction, uint64 count, bool execute_once)
void pfree(void *pointer)
Definition: mcxt.c:1056
MemoryContext es_query_cxt
Definition: execnodes.h:555
int64 shared_blks_hit
pgssLocationLen * clocations
static int pgss_max
static void entry_dealloc(void)
List * rtable
Definition: parsenodes.h:137
static HTAB * pgss_hash
List * distinctClause
Definition: parsenodes.h:156
Oid funcid
Definition: primnodes.h:469
#define ObjectIdGetDatum(X)
Definition: postgres.h:507
#define ERROR
Definition: elog.h:43
uint64 nprocessed
Definition: cmdtag.h:31
ExecutorRun_hook_type ExecutorRun_hook
Definition: execMain.c:71
ExecutorEnd_hook_type ExecutorEnd_hook
Definition: execMain.c:73
void standard_ProcessUtility(PlannedStmt *pstmt, const char *queryString, ProcessUtilityContext context, ParamListInfo params, QueryEnvironment *queryEnv, DestReceiver *dest, QueryCompletion *qc)
Definition: utility.c:541
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2370
static void pgss_post_parse_analyze(ParseState *pstate, Query *query)
struct pg_encoding enc
Definition: encode.c:556
fmNodePtr resultinfo
Definition: fmgr.h:89
static void pg_stat_statements_internal(FunctionCallInfo fcinfo, pgssVersion api_version, bool showtext)
TableFunc * tablefunc
Definition: parsenodes.h:1069
#define lfirst_int(lc)
Definition: pg_list.h:191
List * partitionClause
Definition: parsenodes.h:1353
#define INSTR_TIME_SUBTRACT(x, y)
Definition: instr_time.h:170
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:392
BoolExprType boolop
Definition: primnodes.h:582
Definition: guc.h:75
void EmitWarningsOnPlaceholders(const char *className)
Definition: guc.c:8931
Expr * arg
Definition: primnodes.h:1219
long wal_records
Definition: instrument.h:37
#define lfirst_node(type, lc)
Definition: pg_list.h:193
#define USAGE_EXEC(duration)
#define APP_JUMB(item)
Node * rowexpr
Definition: primnodes.h:88
Node * limitCount
Definition: parsenodes.h:161
char * c
void on_shmem_exit(pg_on_exit_callback function, Datum arg)
Definition: ipc.c:361
int stmt_location
Definition: plannodes.h:95
pgssStoreKind
NodeTag type
Definition: nodes.h:531
static char * buf
Definition: pg_test_fsync.c:67
static void JumbleQuery(pgssJumbleState *jstate, Query *query)
#define PG_GETARG_OID(n)
Definition: fmgr.h:275
List * exclRelTlist
Definition: primnodes.h:1538
void check_stack_depth(void)
Definition: postgres.c:3312
bool IsUnderPostmaster
Definition: globals.c:109
JoinType jointype
Definition: parsenodes.h:1049
void(* shmem_startup_hook_type)(void)
Definition: ipc.h:22
List * aggorder
Definition: primnodes.h:320
static pgssEntry * entry_alloc(pgssHashKey *key, Size query_offset, int query_len, int encoding, bool sticky)
int errdetail(const char *fmt,...)
Definition: elog.c:957
Expr * arg
Definition: primnodes.h:1242
int errcode_for_file_access(void)
Definition: elog.c:633
ScanDirection
Definition: sdir.h:22
AttrNumber resno
Definition: primnodes.h:1408
#define CStringGetDatum(X)
Definition: postgres.h:578
Node * utilityStmt
Definition: plannodes.h:92
static char * qtext_load_file(Size *buffer_size)
FILE * AllocateFile(const char *name, const char *mode)
Definition: fd.c:2320
static char * generate_normalized_query(pgssJumbleState *jstate, const char *query, int query_loc, int *query_len_p, int encoding)
long shared_blks_dirtied
Definition: instrument.h:23
char * cursor_name
Definition: primnodes.h:1317
#define YYLTYPE
Definition: scanner.h:44
unsigned int uint32
Definition: c.h:367
List * aggdirectargs
Definition: primnodes.h:318
#define IS_STICKY(c)
static ExecutorFinish_hook_type prev_ExecutorFinish
void _PG_fini(void)
Oid winfnoid
Definition: primnodes.h:373
planner_hook_type planner_hook
Definition: planner.c:74
Expr * arg
Definition: primnodes.h:834
Datum numeric_in(PG_FUNCTION_ARGS)
Definition: numeric.c:567
const char * p_sourcetext
Definition: parse_node.h:179
long temp_blks_read
Definition: instrument.h:29
Expr * elemexpr
Definition: primnodes.h:859
Definition: type.h:82
List * returningList
Definition: parsenodes.h:146
int64 temp_blks_written
Datum pg_stat_statements_reset(PG_FUNCTION_ARGS)
static bool qtext_store(const char *query, int query_len, Size *query_offset, int *gc_count)
int64 shared_blks_written
#define STICKY_DECREASE_FACTOR
char * enrname
Definition: parsenodes.h:1108
#define USAGE_DECREASE_FACTOR
int durable_rename(const char *oldfile, const char *newfile, int elevel)
Definition: fd.c:656
Definition: nodes.h:300
List * newvals
Definition: primnodes.h:792
#define DirectFunctionCall3(func, arg1, arg2, arg3)
Definition: fmgr.h:628
Definition: guc.h:72
Definition: nodes.h:154
OnConflictAction action
Definition: primnodes.h:1526
bool isNatural
Definition: primnodes.h:1489
core_yyscan_t scanner_init(const char *str, core_yy_extra_type *yyext, const ScanKeywordList *keywordlist, const uint16 *keyword_tokens)
static PlannedStmt * pgss_planner(Query *parse, const char *query_string, int cursorOptions, ParamListInfo boundParams)
static Size pgss_memsize(void)
uint64 queryId
Definition: parsenodes.h:116
int CloseTransientFile(int fd)
Definition: fd.c:2547
#define WARNING
Definition: elog.h:40
shmem_startup_hook_type shmem_startup_hook
Definition: ipci.c:52
Index varno
Definition: primnodes.h:184
Definition: nodes.h:153
#define stat(a, b)
Definition: win32_port.h:255
Size hash_estimate_size(long num_entries, Size entrysize)
Definition: dynahash.c:734
static void fill_in_constant_lengths(pgssJumbleState *jstate, const char *query, int query_loc)
XmlExprOp op
Definition: primnodes.h:1181
static int comp_location(const void *a, const void *b)
Node * startOffset
Definition: parsenodes.h:1356
#define SpinLockRelease(lock)
Definition: spin.h:64
Tuplestorestate * tuplestore_begin_heap(bool randomAccess, bool interXact, int maxKBytes)
Definition: tuplestore.c:318
#define HASH_BLOBS
Definition: hsearch.h:88
int64 local_blks_read
List * args
Definition: primnodes.h:933
static bool need_gc_qtexts(void)
#define PG_FINALLY()
Definition: elog.h:312
int location
Definition: primnodes.h:221
Node * quals
Definition: primnodes.h:1493
BoolTestType booltesttype
Definition: primnodes.h:1243
uintptr_t Datum
Definition: postgres.h:367
Datum pg_stat_statements_reset_1_7(PG_FUNCTION_ARGS)
int GetDatabaseEncoding(void)
Definition: mbutils.c:1151
static void pgss_shmem_shutdown(int code, Datum arg)
static void gc_qtexts(void)
static void pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString, ProcessUtilityContext context, ParamListInfo params, QueryEnvironment *queryEnv, DestReceiver *dest, QueryCompletion *qc)
Size add_size(Size s1, Size s2)
Definition: shmem.c:498
struct pgssLocationLen pgssLocationLen
double mean_time[PGSS_NUMKIND]
Oid MyDatabaseId
Definition: globals.c:85
PGSSTrackLevel
#define Int64GetDatumFast(X)
Definition: postgres.h:760
static void entry_reset(Oid userid, Oid dbid, uint64 queryid)
bool scanner_isspace(char ch)
Definition: scansup.c:220
Size keysize
Definition: hsearch.h:72
Oid resulttype
Definition: primnodes.h:815
int work_mem
Definition: globals.c:121
BufferUsage bufusage
Definition: instrument.h:74
#define USAGE_INIT
int core_yylex(core_YYSTYPE *lvalp, YYLTYPE *llocp, core_yyscan_t yyscanner)
NullTestType nulltesttype
Definition: primnodes.h:1220
static zic_t const max_time
Definition: zic.c:580
pgssHashKey key
Oid aggfnoid
Definition: primnodes.h:312
CommandTag commandTag
Definition: cmdtag.h:30
List * colexprs
Definition: primnodes.h:93
List * named_args
Definition: primnodes.h:1183
#define ereport(elevel,...)
Definition: elog.h:144
int allowedModes
Definition: execnodes.h:305
#define PG_STAT_STATEMENTS_COLS
#define free(a)
Definition: header.h:65
CmdType commandType
Definition: parsenodes.h:112
List * args
Definition: primnodes.h:1185
int64 calls[PGSS_NUMKIND]
#define PG_RETURN_VOID()
Definition: fmgr.h:348
bool is_member_of_role(Oid member, Oid role)
Definition: acl.c:4916
#define Float8GetDatumFast(X)
Definition: postgres.h:761
#define DatumGetUInt64(X)
Definition: postgres.h:634
SetFunctionReturnMode returnMode
Definition: execnodes.h:307
double min_time[PGSS_NUMKIND]
struct Instrumentation * totaltime
Definition: execdesc.h:55
#define Max(x, y)
Definition: c.h:914
void(* ExecutorEnd_hook_type)(QueryDesc *queryDesc)
Definition: executor.h:80
double blk_read_time
Node * rarg
Definition: primnodes.h:1491
#define PG_VALID_BE_ENCODING(_enc)
Definition: pg_wchar.h:295
Expr * arg
Definition: primnodes.h:498
JoinType jointype
Definition: primnodes.h:1488
static bool pgss_track_utility
#define Assert(condition)
Definition: c.h:738
#define lfirst(lc)
Definition: pg_list.h:190
LWLockPadded * GetNamedLWLockTranche(const char *tranche_name)
Definition: lwlock.c:599
CTEMaterialize ctematerialized
Definition: parsenodes.h:1452
List * functions
Definition: parsenodes.h:1063
Expr * aggfilter
Definition: primnodes.h:378
int64 local_blks_dirtied
Expr * expr
Definition: primnodes.h:1407
int paramid
Definition: primnodes.h:263
uint64 es_processed
Definition: execnodes.h:559
PG_MODULE_MAGIC
instr_time blk_write_time
Definition: instrument.h:32
int64 local_blks_hit
size_t Size
Definition: c.h:466
struct Counters Counters
Node * endOffset
Definition: parsenodes.h:1357
long wal_fpi
Definition: instrument.h:38
Expr * arg
Definition: primnodes.h:900
Expr * aggfilter
Definition: primnodes.h:322
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1208
SetOperation op
Definition: parsenodes.h:1658
double blk_write_time
Index ctelevelsup
Definition: parsenodes.h:1080
#define MAXALIGN(LEN)
Definition: c.h:691
#define record_gc_qtexts()
LockWaitPolicy waitPolicy
Definition: parsenodes.h:1384
struct pgssJumbleState pgssJumbleState
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1391
int32 encoding
Definition: pg_database.h:41
MemoryContext ecxt_per_query_memory
Definition: execnodes.h:233
void(* post_parse_analyze_hook_type)(ParseState *pstate, Query *query)
Definition: analyze.h:20
List * args
Definition: primnodes.h:583
static void header(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:208
void * repalloc(void *pointer, Size size)
Definition: mcxt.c:1069
double sum_var_time[PGSS_NUMKIND]
void hash_seq_init(HASH_SEQ_STATUS *status, HTAB *hashp)
Definition: dynahash.c:1381
#define INSTR_TIME_SET_CURRENT(t)
Definition: instr_time.h:156
static int plan_nested_level
#define nodeTag(nodeptr)
Definition: nodes.h:534
Tuplestorestate * setResult
Definition: execnodes.h:310
static shmem_startup_hook_type prev_shmem_startup_hook
const char * sourceText
Definition: execdesc.h:38
RTEKind rtekind
Definition: parsenodes.h:976
static core_yyscan_t yyscanner
Definition: pl_scanner.c:106
int FreeFile(FILE *file)
Definition: fd.c:2519
static Datum values[MAXATTR]
Definition: bootstrap.c:167
List * orderClause
Definition: parsenodes.h:1354
Node * arbiterWhere
Definition: primnodes.h:1531
Expr * refassgnexpr
Definition: primnodes.h:430
ExprContext * econtext
Definition: execnodes.h:303
#define Int32GetDatum(X)
Definition: postgres.h:479
char * ctename
Definition: parsenodes.h:1079
List * cteList
Definition: parsenodes.h:135
Node * setOperations
Definition: parsenodes.h:166
e
Definition: preproc-init.c:82
Query * subquery
Definition: parsenodes.h:1011
List * groupClause
Definition: parsenodes.h:148
TupleDesc setDesc
Definition: execnodes.h:311
static void pgss_ExecutorFinish(QueryDesc *queryDesc)
void * palloc(Size size)
Definition: mcxt.c:949
uint64 queryId
Definition: plannodes.h:48
int errmsg(const char *fmt,...)
Definition: elog.c:824
long shared_blks_hit
Definition: instrument.h:21
Datum pg_stat_statements(PG_FUNCTION_ARGS)
void(* ExecutorStart_hook_type)(QueryDesc *queryDesc, int eflags)
Definition: executor.h:65
long local_blks_written
Definition: instrument.h:28
List * reflowerindexpr
Definition: primnodes.h:424
#define elog(elevel,...)
Definition: elog.h:214
int i
PlannedStmt * standard_planner(Query *parse, const char *query_string, int cursorOptions, ParamListInfo boundParams)
Definition: planner.c:280
List * onConflictSet
Definition: primnodes.h:1535
bool escape_string_warning
Definition: scanner.h:88
static const uint32 PGSS_FILE_HEADER
Index ressortgroupref
Definition: primnodes.h:1410
#define CStringGetTextDatum(s)
Definition: builtins.h:87
void * arg
#define PG_FUNCTION_ARGS
Definition: fmgr.h:193
MinMaxOp op
Definition: primnodes.h:1105
void scanner_finish(core_yyscan_t yyscanner)
Expr * refexpr
Definition: primnodes.h:427
ExecutorFinish_hook_type ExecutorFinish_hook
Definition: execMain.c:72
Expr * arg
Definition: primnodes.h:932
Oid opno
Definition: primnodes.h:516
HTAB * ShmemInitHash(const char *name, long init_size, long max_size, HASHCTL *infoP, int hash_flags)
Definition: shmem.c:337
Expr * result
Definition: primnodes.h:945
struct pgssHashKey pgssHashKey
PlannedStmt * plannedstmt
Definition: execdesc.h:37
#define qsort(a, b, c, d)
Definition: port.h:479
List * args
Definition: primnodes.h:522
uint64 wal_bytes
Definition: instrument.h:39
Node * havingQual
Definition: parsenodes.h:152
Expr * defresult
Definition: primnodes.h:934
char * pg_any_to_server(const char *s, int len, int encoding)
Definition: mbutils.c:619
Expr * expr
Definition: primnodes.h:944
#define PG_TRY()
Definition: elog.h:295
static zic_t const min_time
Definition: zic.c:579
#define PG_GETARG_INT64(n)
Definition: fmgr.h:282
static void pgss_ExecutorEnd(QueryDesc *queryDesc)
Node * onConflictWhere
Definition: primnodes.h:1536
#define PGSS_TEXT_FILE
void DefineCustomBoolVariable(const char *name, const char *short_desc, const char *long_desc, bool *valueAddr, bool bootValue, GucContext context, int flags, GucBoolCheckHook check_hook, GucBoolAssignHook assign_hook, GucShowHook show_hook)
Definition: guc.c:8792
int rtindex
Definition: primnodes.h:1495
Definition: pg_list.h:50
#define snprintf
Definition: port.h:193
struct TableSampleClause * tablesample
Definition: parsenodes.h:1006
int stmt_len
Definition: parsenodes.h:182
#define UINT64_FORMAT
Definition: c.h:410
Oid paramtype
Definition: primnodes.h:264
post_parse_analyze_hook_type post_parse_analyze_hook
Definition: analyze.c:51
Datum pg_stat_statements_1_2(PG_FUNCTION_ARGS)
#define PG_END_TRY()
Definition: elog.h:320
#define ASSUMED_MEDIAN_INIT
#define read(a, b, c)
Definition: win32.h:13
void hash_seq_term(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1467
static const struct config_enum_entry track_options[]
BufferUsage pgBufferUsage
Definition: instrument.c:20
#define PG_STAT_STATEMENTS_COLS_V1_2
static bool pgss_save
static char * qtext_fetch(Size query_offset, int query_len, char *buffer, Size buffer_size)
PG_FUNCTION_INFO_V1(pg_stat_statements_reset)
static struct subre * parse(struct vars *, int, int, struct state *, struct state *)
Definition: regcomp.c:648
Definition: nodes.h:155
AttrNumber fieldnum
Definition: primnodes.h:763
#define ftruncate(a, b)
Definition: win32_port.h:59