Line data Source code
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 totaled 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 : * Starting in Postgres 9.2, this module normalized query entries. As of
12 : * Postgres 14, the normalization is done by the core if compute_query_id is
13 : * enabled, or optionally by third-party modules.
14 : *
15 : * To facilitate presenting entries to users, we create "representative" query
16 : * strings in which constants are replaced with parameter symbols ($n), to
17 : * make it clearer what a normalized entry can represent. To save on shared
18 : * memory, and to avoid having to truncate oversized query strings, we store
19 : * these strings in a temporary external query-texts file. Offsets into this
20 : * file are kept in shared memory.
21 : *
22 : * Note about locking issues: to create or delete an entry in the shared
23 : * hashtable, one must hold pgss->lock exclusively. Modifying any field
24 : * in an entry except the counters requires the same. To look up an entry,
25 : * one must hold the lock shared. To read or update the counters within
26 : * an entry, one must hold the lock shared or exclusive (so the entry doesn't
27 : * disappear!) and also take the entry's mutex spinlock.
28 : * The shared state variable pgss->extent (the next free spot in the external
29 : * query-text file) should be accessed only while holding either the
30 : * pgss->mutex spinlock, or exclusive lock on pgss->lock. We use the mutex to
31 : * allow reserving file space while holding only shared lock on pgss->lock.
32 : * Rewriting the entire external query-text file, eg for garbage collection,
33 : * requires holding pgss->lock exclusively; this allows individual entries
34 : * in the file to be read or written while holding only shared lock.
35 : *
36 : *
37 : * Copyright (c) 2008-2025, PostgreSQL Global Development Group
38 : *
39 : * IDENTIFICATION
40 : * contrib/pg_stat_statements/pg_stat_statements.c
41 : *
42 : *-------------------------------------------------------------------------
43 : */
44 : #include "postgres.h"
45 :
46 : #include <math.h>
47 : #include <sys/stat.h>
48 : #include <unistd.h>
49 :
50 : #include "access/parallel.h"
51 : #include "catalog/pg_authid.h"
52 : #include "common/int.h"
53 : #include "executor/instrument.h"
54 : #include "funcapi.h"
55 : #include "jit/jit.h"
56 : #include "mb/pg_wchar.h"
57 : #include "miscadmin.h"
58 : #include "nodes/queryjumble.h"
59 : #include "optimizer/planner.h"
60 : #include "parser/analyze.h"
61 : #include "parser/scanner.h"
62 : #include "pgstat.h"
63 : #include "storage/fd.h"
64 : #include "storage/ipc.h"
65 : #include "storage/lwlock.h"
66 : #include "storage/shmem.h"
67 : #include "storage/spin.h"
68 : #include "tcop/utility.h"
69 : #include "utils/acl.h"
70 : #include "utils/builtins.h"
71 : #include "utils/memutils.h"
72 : #include "utils/timestamp.h"
73 :
74 16 : PG_MODULE_MAGIC;
75 :
76 : /* Location of permanent stats file (valid when database is shut down) */
77 : #define PGSS_DUMP_FILE PGSTAT_STAT_PERMANENT_DIRECTORY "/pg_stat_statements.stat"
78 :
79 : /*
80 : * Location of external query text file.
81 : */
82 : #define PGSS_TEXT_FILE PG_STAT_TMP_DIR "/pgss_query_texts.stat"
83 :
84 : /* Magic number identifying the stats file format */
85 : static const uint32 PGSS_FILE_HEADER = 0x20220408;
86 :
87 : /* PostgreSQL major version number, changes in which invalidate all entries */
88 : static const uint32 PGSS_PG_MAJOR_VERSION = PG_VERSION_NUM / 100;
89 :
90 : /* XXX: Should USAGE_EXEC reflect execution time and/or buffer usage? */
91 : #define USAGE_EXEC(duration) (1.0)
92 : #define USAGE_INIT (1.0) /* including initial planning */
93 : #define ASSUMED_MEDIAN_INIT (10.0) /* initial assumed median usage */
94 : #define ASSUMED_LENGTH_INIT 1024 /* initial assumed mean query length */
95 : #define USAGE_DECREASE_FACTOR (0.99) /* decreased every entry_dealloc */
96 : #define STICKY_DECREASE_FACTOR (0.50) /* factor for sticky entries */
97 : #define USAGE_DEALLOC_PERCENT 5 /* free this % of entries at once */
98 : #define IS_STICKY(c) ((c.calls[PGSS_PLAN] + c.calls[PGSS_EXEC]) == 0)
99 :
100 : /*
101 : * Extension version number, for supporting older extension versions' objects
102 : */
103 : typedef enum pgssVersion
104 : {
105 : PGSS_V1_0 = 0,
106 : PGSS_V1_1,
107 : PGSS_V1_2,
108 : PGSS_V1_3,
109 : PGSS_V1_8,
110 : PGSS_V1_9,
111 : PGSS_V1_10,
112 : PGSS_V1_11,
113 : PGSS_V1_12,
114 : } pgssVersion;
115 :
116 : typedef enum pgssStoreKind
117 : {
118 : PGSS_INVALID = -1,
119 :
120 : /*
121 : * PGSS_PLAN and PGSS_EXEC must be respectively 0 and 1 as they're used to
122 : * reference the underlying values in the arrays in the Counters struct,
123 : * and this order is required in pg_stat_statements_internal().
124 : */
125 : PGSS_PLAN = 0,
126 : PGSS_EXEC,
127 : } pgssStoreKind;
128 :
129 : #define PGSS_NUMKIND (PGSS_EXEC + 1)
130 :
131 : /*
132 : * Hashtable key that defines the identity of a hashtable entry. We separate
133 : * queries by user and by database even if they are otherwise identical.
134 : *
135 : * If you add a new key to this struct, make sure to teach pgss_store() to
136 : * zero the padding bytes. Otherwise, things will break, because pgss_hash is
137 : * created using HASH_BLOBS, and thus tag_hash is used to hash this.
138 :
139 : */
140 : typedef struct pgssHashKey
141 : {
142 : Oid userid; /* user OID */
143 : Oid dbid; /* database OID */
144 : uint64 queryid; /* query identifier */
145 : bool toplevel; /* query executed at top level */
146 : } pgssHashKey;
147 :
148 : /*
149 : * The actual stats counters kept within pgssEntry.
150 : */
151 : typedef struct Counters
152 : {
153 : int64 calls[PGSS_NUMKIND]; /* # of times planned/executed */
154 : double total_time[PGSS_NUMKIND]; /* total planning/execution time,
155 : * in msec */
156 : double min_time[PGSS_NUMKIND]; /* minimum planning/execution time in
157 : * msec since min/max reset */
158 : double max_time[PGSS_NUMKIND]; /* maximum planning/execution time in
159 : * msec since min/max reset */
160 : double mean_time[PGSS_NUMKIND]; /* mean planning/execution time in
161 : * msec */
162 : double sum_var_time[PGSS_NUMKIND]; /* sum of variances in
163 : * planning/execution time in msec */
164 : int64 rows; /* total # of retrieved or affected rows */
165 : int64 shared_blks_hit; /* # of shared buffer hits */
166 : int64 shared_blks_read; /* # of shared disk blocks read */
167 : int64 shared_blks_dirtied; /* # of shared disk blocks dirtied */
168 : int64 shared_blks_written; /* # of shared disk blocks written */
169 : int64 local_blks_hit; /* # of local buffer hits */
170 : int64 local_blks_read; /* # of local disk blocks read */
171 : int64 local_blks_dirtied; /* # of local disk blocks dirtied */
172 : int64 local_blks_written; /* # of local disk blocks written */
173 : int64 temp_blks_read; /* # of temp blocks read */
174 : int64 temp_blks_written; /* # of temp blocks written */
175 : double shared_blk_read_time; /* time spent reading shared blocks,
176 : * in msec */
177 : double shared_blk_write_time; /* time spent writing shared blocks,
178 : * in msec */
179 : double local_blk_read_time; /* time spent reading local blocks, in
180 : * msec */
181 : double local_blk_write_time; /* time spent writing local blocks, in
182 : * msec */
183 : double temp_blk_read_time; /* time spent reading temp blocks, in msec */
184 : double temp_blk_write_time; /* time spent writing temp blocks, in
185 : * msec */
186 : double usage; /* usage factor */
187 : int64 wal_records; /* # of WAL records generated */
188 : int64 wal_fpi; /* # of WAL full page images generated */
189 : uint64 wal_bytes; /* total amount of WAL generated in bytes */
190 : int64 jit_functions; /* total number of JIT functions emitted */
191 : double jit_generation_time; /* total time to generate jit code */
192 : int64 jit_inlining_count; /* number of times inlining time has been
193 : * > 0 */
194 : double jit_deform_time; /* total time to deform tuples in jit code */
195 : int64 jit_deform_count; /* number of times deform time has been >
196 : * 0 */
197 :
198 : double jit_inlining_time; /* total time to inline jit code */
199 : int64 jit_optimization_count; /* number of times optimization time
200 : * has been > 0 */
201 : double jit_optimization_time; /* total time to optimize jit code */
202 : int64 jit_emission_count; /* number of times emission time has been
203 : * > 0 */
204 : double jit_emission_time; /* total time to emit jit code */
205 : int64 parallel_workers_to_launch; /* # of parallel workers planned
206 : * to be launched */
207 : int64 parallel_workers_launched; /* # of parallel workers actually
208 : * launched */
209 : } Counters;
210 :
211 : /*
212 : * Global statistics for pg_stat_statements
213 : */
214 : typedef struct pgssGlobalStats
215 : {
216 : int64 dealloc; /* # of times entries were deallocated */
217 : TimestampTz stats_reset; /* timestamp with all stats reset */
218 : } pgssGlobalStats;
219 :
220 : /*
221 : * Statistics per statement
222 : *
223 : * Note: in event of a failure in garbage collection of the query text file,
224 : * we reset query_offset to zero and query_len to -1. This will be seen as
225 : * an invalid state by qtext_fetch().
226 : */
227 : typedef struct pgssEntry
228 : {
229 : pgssHashKey key; /* hash key of entry - MUST BE FIRST */
230 : Counters counters; /* the statistics for this query */
231 : Size query_offset; /* query text offset in external file */
232 : int query_len; /* # of valid bytes in query string, or -1 */
233 : int encoding; /* query text encoding */
234 : TimestampTz stats_since; /* timestamp of entry allocation */
235 : TimestampTz minmax_stats_since; /* timestamp of last min/max values reset */
236 : slock_t mutex; /* protects the counters only */
237 : } pgssEntry;
238 :
239 : /*
240 : * Global shared state
241 : */
242 : typedef struct pgssSharedState
243 : {
244 : LWLock *lock; /* protects hashtable search/modification */
245 : double cur_median_usage; /* current median usage in hashtable */
246 : Size mean_query_len; /* current mean entry text length */
247 : slock_t mutex; /* protects following fields only: */
248 : Size extent; /* current extent of query file */
249 : int n_writers; /* number of active writers to query file */
250 : int gc_count; /* query file garbage collection cycle count */
251 : pgssGlobalStats stats; /* global statistics for pgss */
252 : } pgssSharedState;
253 :
254 : /*---- Local variables ----*/
255 :
256 : /* Current nesting depth of planner/ExecutorRun/ProcessUtility calls */
257 : static int nesting_level = 0;
258 :
259 : /* Saved hook values */
260 : static shmem_request_hook_type prev_shmem_request_hook = NULL;
261 : static shmem_startup_hook_type prev_shmem_startup_hook = NULL;
262 : static post_parse_analyze_hook_type prev_post_parse_analyze_hook = NULL;
263 : static planner_hook_type prev_planner_hook = NULL;
264 : static ExecutorStart_hook_type prev_ExecutorStart = NULL;
265 : static ExecutorRun_hook_type prev_ExecutorRun = NULL;
266 : static ExecutorFinish_hook_type prev_ExecutorFinish = NULL;
267 : static ExecutorEnd_hook_type prev_ExecutorEnd = NULL;
268 : static ProcessUtility_hook_type prev_ProcessUtility = NULL;
269 :
270 : /* Links to shared memory state */
271 : static pgssSharedState *pgss = NULL;
272 : static HTAB *pgss_hash = NULL;
273 :
274 : /*---- GUC variables ----*/
275 :
276 : typedef enum
277 : {
278 : PGSS_TRACK_NONE, /* track no statements */
279 : PGSS_TRACK_TOP, /* only top level statements */
280 : PGSS_TRACK_ALL, /* all statements, including nested ones */
281 : } PGSSTrackLevel;
282 :
283 : static const struct config_enum_entry track_options[] =
284 : {
285 : {"none", PGSS_TRACK_NONE, false},
286 : {"top", PGSS_TRACK_TOP, false},
287 : {"all", PGSS_TRACK_ALL, false},
288 : {NULL, 0, false}
289 : };
290 :
291 : static int pgss_max = 5000; /* max # statements to track */
292 : static int pgss_track = PGSS_TRACK_TOP; /* tracking level */
293 : static bool pgss_track_utility = true; /* whether to track utility commands */
294 : static bool pgss_track_planning = false; /* whether to track planning
295 : * duration */
296 : static bool pgss_save = true; /* whether to save stats across shutdown */
297 :
298 :
299 : #define pgss_enabled(level) \
300 : (!IsParallelWorker() && \
301 : (pgss_track == PGSS_TRACK_ALL || \
302 : (pgss_track == PGSS_TRACK_TOP && (level) == 0)))
303 :
304 : #define record_gc_qtexts() \
305 : do { \
306 : SpinLockAcquire(&pgss->mutex); \
307 : pgss->gc_count++; \
308 : SpinLockRelease(&pgss->mutex); \
309 : } while(0)
310 :
311 : /*---- Function declarations ----*/
312 :
313 12 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset);
314 12 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_7);
315 36 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_11);
316 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_2);
317 12 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_3);
318 12 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_8);
319 12 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_9);
320 12 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_10);
321 12 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_11);
322 44 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_12);
323 0 : PG_FUNCTION_INFO_V1(pg_stat_statements);
324 14 : PG_FUNCTION_INFO_V1(pg_stat_statements_info);
325 :
326 : static void pgss_shmem_request(void);
327 : static void pgss_shmem_startup(void);
328 : static void pgss_shmem_shutdown(int code, Datum arg);
329 : static void pgss_post_parse_analyze(ParseState *pstate, Query *query,
330 : JumbleState *jstate);
331 : static PlannedStmt *pgss_planner(Query *parse,
332 : const char *query_string,
333 : int cursorOptions,
334 : ParamListInfo boundParams);
335 : static void pgss_ExecutorStart(QueryDesc *queryDesc, int eflags);
336 : static void pgss_ExecutorRun(QueryDesc *queryDesc,
337 : ScanDirection direction,
338 : uint64 count);
339 : static void pgss_ExecutorFinish(QueryDesc *queryDesc);
340 : static void pgss_ExecutorEnd(QueryDesc *queryDesc);
341 : static void pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
342 : bool readOnlyTree,
343 : ProcessUtilityContext context, ParamListInfo params,
344 : QueryEnvironment *queryEnv,
345 : DestReceiver *dest, QueryCompletion *qc);
346 : static void pgss_store(const char *query, uint64 queryId,
347 : int query_location, int query_len,
348 : pgssStoreKind kind,
349 : double total_time, uint64 rows,
350 : const BufferUsage *bufusage,
351 : const WalUsage *walusage,
352 : const struct JitInstrumentation *jitusage,
353 : JumbleState *jstate,
354 : int parallel_workers_to_launch,
355 : int parallel_workers_launched);
356 : static void pg_stat_statements_internal(FunctionCallInfo fcinfo,
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 TimestampTz entry_reset(Oid userid, Oid dbid, uint64 queryid, bool minmax_only);
371 : static char *generate_normalized_query(JumbleState *jstate, const char *query,
372 : int query_loc, int *query_len_p);
373 : static void fill_in_constant_lengths(JumbleState *jstate, const char *query,
374 : int query_loc);
375 : static int comp_location(const void *a, const void *b);
376 :
377 :
378 : /*
379 : * Module load callback
380 : */
381 : void
382 16 : _PG_init(void)
383 : {
384 : /*
385 : * In order to create our shared memory area, we have to be loaded via
386 : * shared_preload_libraries. If not, fall out without hooking into any of
387 : * the main system. (We don't throw error here because it seems useful to
388 : * allow the pg_stat_statements functions to be created even when the
389 : * module isn't active. The functions must protect themselves against
390 : * being called then, however.)
391 : */
392 16 : if (!process_shared_preload_libraries_in_progress)
393 2 : return;
394 :
395 : /*
396 : * Inform the postmaster that we want to enable query_id calculation if
397 : * compute_query_id is set to auto.
398 : */
399 14 : EnableQueryId();
400 :
401 : /*
402 : * Define (or redefine) custom GUC variables.
403 : */
404 14 : DefineCustomIntVariable("pg_stat_statements.max",
405 : "Sets the maximum number of statements tracked by pg_stat_statements.",
406 : NULL,
407 : &pgss_max,
408 : 5000,
409 : 100,
410 : INT_MAX / 2,
411 : PGC_POSTMASTER,
412 : 0,
413 : NULL,
414 : NULL,
415 : NULL);
416 :
417 14 : DefineCustomEnumVariable("pg_stat_statements.track",
418 : "Selects which statements are tracked by pg_stat_statements.",
419 : NULL,
420 : &pgss_track,
421 : PGSS_TRACK_TOP,
422 : track_options,
423 : PGC_SUSET,
424 : 0,
425 : NULL,
426 : NULL,
427 : NULL);
428 :
429 14 : DefineCustomBoolVariable("pg_stat_statements.track_utility",
430 : "Selects whether utility commands are tracked by pg_stat_statements.",
431 : NULL,
432 : &pgss_track_utility,
433 : true,
434 : PGC_SUSET,
435 : 0,
436 : NULL,
437 : NULL,
438 : NULL);
439 :
440 14 : DefineCustomBoolVariable("pg_stat_statements.track_planning",
441 : "Selects whether planning duration is tracked by pg_stat_statements.",
442 : NULL,
443 : &pgss_track_planning,
444 : false,
445 : PGC_SUSET,
446 : 0,
447 : NULL,
448 : NULL,
449 : NULL);
450 :
451 14 : DefineCustomBoolVariable("pg_stat_statements.save",
452 : "Save pg_stat_statements statistics across server shutdowns.",
453 : NULL,
454 : &pgss_save,
455 : true,
456 : PGC_SIGHUP,
457 : 0,
458 : NULL,
459 : NULL,
460 : NULL);
461 :
462 14 : MarkGUCPrefixReserved("pg_stat_statements");
463 :
464 : /*
465 : * Install hooks.
466 : */
467 14 : prev_shmem_request_hook = shmem_request_hook;
468 14 : shmem_request_hook = pgss_shmem_request;
469 14 : prev_shmem_startup_hook = shmem_startup_hook;
470 14 : shmem_startup_hook = pgss_shmem_startup;
471 14 : prev_post_parse_analyze_hook = post_parse_analyze_hook;
472 14 : post_parse_analyze_hook = pgss_post_parse_analyze;
473 14 : prev_planner_hook = planner_hook;
474 14 : planner_hook = pgss_planner;
475 14 : prev_ExecutorStart = ExecutorStart_hook;
476 14 : ExecutorStart_hook = pgss_ExecutorStart;
477 14 : prev_ExecutorRun = ExecutorRun_hook;
478 14 : ExecutorRun_hook = pgss_ExecutorRun;
479 14 : prev_ExecutorFinish = ExecutorFinish_hook;
480 14 : ExecutorFinish_hook = pgss_ExecutorFinish;
481 14 : prev_ExecutorEnd = ExecutorEnd_hook;
482 14 : ExecutorEnd_hook = pgss_ExecutorEnd;
483 14 : prev_ProcessUtility = ProcessUtility_hook;
484 14 : ProcessUtility_hook = pgss_ProcessUtility;
485 : }
486 :
487 : /*
488 : * shmem_request hook: request additional shared resources. We'll allocate or
489 : * attach to the shared resources in pgss_shmem_startup().
490 : */
491 : static void
492 14 : pgss_shmem_request(void)
493 : {
494 14 : if (prev_shmem_request_hook)
495 0 : prev_shmem_request_hook();
496 :
497 14 : RequestAddinShmemSpace(pgss_memsize());
498 14 : RequestNamedLWLockTranche("pg_stat_statements", 1);
499 14 : }
500 :
501 : /*
502 : * shmem_startup hook: allocate or attach to shared memory,
503 : * then load any pre-existing statistics from file.
504 : * Also create and load the query-texts file, which is expected to exist
505 : * (even if empty) while the module is enabled.
506 : */
507 : static void
508 14 : pgss_shmem_startup(void)
509 : {
510 : bool found;
511 : HASHCTL info;
512 14 : FILE *file = NULL;
513 14 : FILE *qfile = NULL;
514 : uint32 header;
515 : int32 num;
516 : int32 pgver;
517 : int32 i;
518 : int buffer_size;
519 14 : char *buffer = NULL;
520 :
521 14 : if (prev_shmem_startup_hook)
522 0 : prev_shmem_startup_hook();
523 :
524 : /* reset in case this is a restart within the postmaster */
525 14 : pgss = NULL;
526 14 : pgss_hash = NULL;
527 :
528 : /*
529 : * Create or attach to the shared memory state, including hash table
530 : */
531 14 : LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
532 :
533 14 : pgss = ShmemInitStruct("pg_stat_statements",
534 : sizeof(pgssSharedState),
535 : &found);
536 :
537 14 : if (!found)
538 : {
539 : /* First time through ... */
540 14 : pgss->lock = &(GetNamedLWLockTranche("pg_stat_statements"))->lock;
541 14 : pgss->cur_median_usage = ASSUMED_MEDIAN_INIT;
542 14 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
543 14 : SpinLockInit(&pgss->mutex);
544 14 : pgss->extent = 0;
545 14 : pgss->n_writers = 0;
546 14 : pgss->gc_count = 0;
547 14 : pgss->stats.dealloc = 0;
548 14 : pgss->stats.stats_reset = GetCurrentTimestamp();
549 : }
550 :
551 14 : info.keysize = sizeof(pgssHashKey);
552 14 : info.entrysize = sizeof(pgssEntry);
553 14 : pgss_hash = ShmemInitHash("pg_stat_statements hash",
554 : pgss_max, pgss_max,
555 : &info,
556 : HASH_ELEM | HASH_BLOBS);
557 :
558 14 : LWLockRelease(AddinShmemInitLock);
559 :
560 : /*
561 : * If we're in the postmaster (or a standalone backend...), set up a shmem
562 : * exit hook to dump the statistics to disk.
563 : */
564 14 : if (!IsUnderPostmaster)
565 14 : on_shmem_exit(pgss_shmem_shutdown, (Datum) 0);
566 :
567 : /*
568 : * Done if some other process already completed our initialization.
569 : */
570 14 : if (found)
571 14 : return;
572 :
573 : /*
574 : * Note: we don't bother with locks here, because there should be no other
575 : * processes running when this code is reached.
576 : */
577 :
578 : /* Unlink query text file possibly left over from crash */
579 14 : unlink(PGSS_TEXT_FILE);
580 :
581 : /* Allocate new query text temp file */
582 14 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
583 14 : if (qfile == NULL)
584 0 : goto write_error;
585 :
586 : /*
587 : * If we were told not to load old statistics, we're done. (Note we do
588 : * not try to unlink any old dump file in this case. This seems a bit
589 : * questionable but it's the historical behavior.)
590 : */
591 14 : if (!pgss_save)
592 : {
593 2 : FreeFile(qfile);
594 2 : return;
595 : }
596 :
597 : /*
598 : * Attempt to load old statistics from the dump file.
599 : */
600 12 : file = AllocateFile(PGSS_DUMP_FILE, PG_BINARY_R);
601 12 : if (file == NULL)
602 : {
603 8 : if (errno != ENOENT)
604 0 : goto read_error;
605 : /* No existing persisted stats file, so we're done */
606 8 : FreeFile(qfile);
607 8 : return;
608 : }
609 :
610 4 : buffer_size = 2048;
611 4 : buffer = (char *) palloc(buffer_size);
612 :
613 8 : if (fread(&header, sizeof(uint32), 1, file) != 1 ||
614 8 : fread(&pgver, sizeof(uint32), 1, file) != 1 ||
615 4 : fread(&num, sizeof(int32), 1, file) != 1)
616 0 : goto read_error;
617 :
618 4 : if (header != PGSS_FILE_HEADER ||
619 4 : pgver != PGSS_PG_MAJOR_VERSION)
620 0 : goto data_error;
621 :
622 51664 : for (i = 0; i < num; i++)
623 : {
624 : pgssEntry temp;
625 : pgssEntry *entry;
626 : Size query_offset;
627 :
628 51660 : if (fread(&temp, sizeof(pgssEntry), 1, file) != 1)
629 0 : goto read_error;
630 :
631 : /* Encoding is the only field we can easily sanity-check */
632 51660 : if (!PG_VALID_BE_ENCODING(temp.encoding))
633 0 : goto data_error;
634 :
635 : /* Resize buffer as needed */
636 51660 : if (temp.query_len >= buffer_size)
637 : {
638 6 : buffer_size = Max(buffer_size * 2, temp.query_len + 1);
639 6 : buffer = repalloc(buffer, buffer_size);
640 : }
641 :
642 51660 : if (fread(buffer, 1, temp.query_len + 1, file) != temp.query_len + 1)
643 0 : goto read_error;
644 :
645 : /* Should have a trailing null, but let's make sure */
646 51660 : buffer[temp.query_len] = '\0';
647 :
648 : /* Skip loading "sticky" entries */
649 51660 : if (IS_STICKY(temp.counters))
650 1458 : continue;
651 :
652 : /* Store the query text */
653 50202 : query_offset = pgss->extent;
654 50202 : if (fwrite(buffer, 1, temp.query_len + 1, qfile) != temp.query_len + 1)
655 0 : goto write_error;
656 50202 : pgss->extent += temp.query_len + 1;
657 :
658 : /* make the hashtable entry (discards old entries if too many) */
659 50202 : entry = entry_alloc(&temp.key, query_offset, temp.query_len,
660 : temp.encoding,
661 : false);
662 :
663 : /* copy in the actual stats */
664 50202 : entry->counters = temp.counters;
665 50202 : entry->stats_since = temp.stats_since;
666 50202 : entry->minmax_stats_since = temp.minmax_stats_since;
667 : }
668 :
669 : /* Read global statistics for pg_stat_statements */
670 4 : if (fread(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
671 0 : goto read_error;
672 :
673 4 : pfree(buffer);
674 4 : FreeFile(file);
675 4 : FreeFile(qfile);
676 :
677 : /*
678 : * Remove the persisted stats file so it's not included in
679 : * backups/replication standbys, etc. A new file will be written on next
680 : * shutdown.
681 : *
682 : * Note: it's okay if the PGSS_TEXT_FILE is included in a basebackup,
683 : * because we remove that file on startup; it acts inversely to
684 : * PGSS_DUMP_FILE, in that it is only supposed to be around when the
685 : * server is running, whereas PGSS_DUMP_FILE is only supposed to be around
686 : * when the server is not running. Leaving the file creates no danger of
687 : * a newly restored database having a spurious record of execution costs,
688 : * which is what we're really concerned about here.
689 : */
690 4 : unlink(PGSS_DUMP_FILE);
691 :
692 4 : return;
693 :
694 0 : read_error:
695 0 : ereport(LOG,
696 : (errcode_for_file_access(),
697 : errmsg("could not read file \"%s\": %m",
698 : PGSS_DUMP_FILE)));
699 0 : goto fail;
700 0 : data_error:
701 0 : ereport(LOG,
702 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
703 : errmsg("ignoring invalid data in file \"%s\"",
704 : PGSS_DUMP_FILE)));
705 0 : goto fail;
706 0 : write_error:
707 0 : ereport(LOG,
708 : (errcode_for_file_access(),
709 : errmsg("could not write file \"%s\": %m",
710 : PGSS_TEXT_FILE)));
711 0 : fail:
712 0 : if (buffer)
713 0 : pfree(buffer);
714 0 : if (file)
715 0 : FreeFile(file);
716 0 : if (qfile)
717 0 : FreeFile(qfile);
718 : /* If possible, throw away the bogus file; ignore any error */
719 0 : unlink(PGSS_DUMP_FILE);
720 :
721 : /*
722 : * Don't unlink PGSS_TEXT_FILE here; it should always be around while the
723 : * server is running with pg_stat_statements enabled
724 : */
725 : }
726 :
727 : /*
728 : * shmem_shutdown hook: Dump statistics into file.
729 : *
730 : * Note: we don't bother with acquiring lock, because there should be no
731 : * other processes running when this is called.
732 : */
733 : static void
734 14 : pgss_shmem_shutdown(int code, Datum arg)
735 : {
736 : FILE *file;
737 14 : char *qbuffer = NULL;
738 14 : Size qbuffer_size = 0;
739 : HASH_SEQ_STATUS hash_seq;
740 : int32 num_entries;
741 : pgssEntry *entry;
742 :
743 : /* Don't try to dump during a crash. */
744 14 : if (code)
745 14 : return;
746 :
747 : /* Safety check ... shouldn't get here unless shmem is set up. */
748 14 : if (!pgss || !pgss_hash)
749 0 : return;
750 :
751 : /* Don't dump if told not to. */
752 14 : if (!pgss_save)
753 4 : return;
754 :
755 10 : file = AllocateFile(PGSS_DUMP_FILE ".tmp", PG_BINARY_W);
756 10 : if (file == NULL)
757 0 : goto error;
758 :
759 10 : if (fwrite(&PGSS_FILE_HEADER, sizeof(uint32), 1, file) != 1)
760 0 : goto error;
761 10 : if (fwrite(&PGSS_PG_MAJOR_VERSION, sizeof(uint32), 1, file) != 1)
762 0 : goto error;
763 10 : num_entries = hash_get_num_entries(pgss_hash);
764 10 : if (fwrite(&num_entries, sizeof(int32), 1, file) != 1)
765 0 : goto error;
766 :
767 10 : qbuffer = qtext_load_file(&qbuffer_size);
768 10 : if (qbuffer == NULL)
769 0 : goto error;
770 :
771 : /*
772 : * When serializing to disk, we store query texts immediately after their
773 : * entry data. Any orphaned query texts are thereby excluded.
774 : */
775 10 : hash_seq_init(&hash_seq, pgss_hash);
776 103744 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
777 : {
778 103734 : int len = entry->query_len;
779 103734 : char *qstr = qtext_fetch(entry->query_offset, len,
780 : qbuffer, qbuffer_size);
781 :
782 103734 : if (qstr == NULL)
783 0 : continue; /* Ignore any entries with bogus texts */
784 :
785 103734 : if (fwrite(entry, sizeof(pgssEntry), 1, file) != 1 ||
786 103734 : fwrite(qstr, 1, len + 1, file) != len + 1)
787 : {
788 : /* note: we assume hash_seq_term won't change errno */
789 0 : hash_seq_term(&hash_seq);
790 0 : goto error;
791 : }
792 : }
793 :
794 : /* Dump global statistics for pg_stat_statements */
795 10 : if (fwrite(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
796 0 : goto error;
797 :
798 10 : free(qbuffer);
799 10 : qbuffer = NULL;
800 :
801 10 : if (FreeFile(file))
802 : {
803 0 : file = NULL;
804 0 : goto error;
805 : }
806 :
807 : /*
808 : * Rename file into place, so we atomically replace any old one.
809 : */
810 10 : (void) durable_rename(PGSS_DUMP_FILE ".tmp", PGSS_DUMP_FILE, LOG);
811 :
812 : /* Unlink query-texts file; it's not needed while shutdown */
813 10 : unlink(PGSS_TEXT_FILE);
814 :
815 10 : return;
816 :
817 0 : error:
818 0 : ereport(LOG,
819 : (errcode_for_file_access(),
820 : errmsg("could not write file \"%s\": %m",
821 : PGSS_DUMP_FILE ".tmp")));
822 0 : free(qbuffer);
823 0 : if (file)
824 0 : FreeFile(file);
825 0 : unlink(PGSS_DUMP_FILE ".tmp");
826 0 : unlink(PGSS_TEXT_FILE);
827 : }
828 :
829 : /*
830 : * Post-parse-analysis hook: mark query with a queryId
831 : */
832 : static void
833 148770 : pgss_post_parse_analyze(ParseState *pstate, Query *query, JumbleState *jstate)
834 : {
835 148770 : if (prev_post_parse_analyze_hook)
836 0 : prev_post_parse_analyze_hook(pstate, query, jstate);
837 :
838 : /* Safety check... */
839 148770 : if (!pgss || !pgss_hash || !pgss_enabled(nesting_level))
840 25376 : return;
841 :
842 : /*
843 : * If it's EXECUTE, clear the queryId so that stats will accumulate for
844 : * the underlying PREPARE. But don't do this if we're not tracking
845 : * utility statements, to avoid messing up another extension that might be
846 : * tracking them.
847 : */
848 123394 : if (query->utilityStmt)
849 : {
850 55130 : if (pgss_track_utility && IsA(query->utilityStmt, ExecuteStmt))
851 : {
852 6348 : query->queryId = UINT64CONST(0);
853 6348 : return;
854 : }
855 : }
856 :
857 : /*
858 : * If query jumbling were able to identify any ignorable constants, we
859 : * immediately create a hash table entry for the query, so that we can
860 : * record the normalized form of the query string. If there were no such
861 : * constants, the normalized string would be the same as the query text
862 : * anyway, so there's no need for an early entry.
863 : */
864 117046 : if (jstate && jstate->clocations_count > 0)
865 67398 : pgss_store(pstate->p_sourcetext,
866 : query->queryId,
867 : query->stmt_location,
868 : query->stmt_len,
869 : PGSS_INVALID,
870 : 0,
871 : 0,
872 : NULL,
873 : NULL,
874 : NULL,
875 : jstate,
876 : 0,
877 : 0);
878 : }
879 :
880 : /*
881 : * Planner hook: forward to regular planner, but measure planning time
882 : * if needed.
883 : */
884 : static PlannedStmt *
885 91660 : pgss_planner(Query *parse,
886 : const char *query_string,
887 : int cursorOptions,
888 : ParamListInfo boundParams)
889 : {
890 : PlannedStmt *result;
891 :
892 : /*
893 : * We can't process the query if no query_string is provided, as
894 : * pgss_store needs it. We also ignore query without queryid, as it would
895 : * be treated as a utility statement, which may not be the case.
896 : */
897 91660 : if (pgss_enabled(nesting_level)
898 68572 : && pgss_track_planning && query_string
899 146 : && parse->queryId != UINT64CONST(0))
900 146 : {
901 : instr_time start;
902 : instr_time duration;
903 : BufferUsage bufusage_start,
904 : bufusage;
905 : WalUsage walusage_start,
906 : walusage;
907 :
908 : /* We need to track buffer usage as the planner can access them. */
909 146 : bufusage_start = pgBufferUsage;
910 :
911 : /*
912 : * Similarly the planner could write some WAL records in some cases
913 : * (e.g. setting a hint bit with those being WAL-logged)
914 : */
915 146 : walusage_start = pgWalUsage;
916 146 : INSTR_TIME_SET_CURRENT(start);
917 :
918 146 : nesting_level++;
919 146 : PG_TRY();
920 : {
921 146 : if (prev_planner_hook)
922 0 : result = prev_planner_hook(parse, query_string, cursorOptions,
923 : boundParams);
924 : else
925 146 : result = standard_planner(parse, query_string, cursorOptions,
926 : boundParams);
927 : }
928 0 : PG_FINALLY();
929 : {
930 146 : nesting_level--;
931 : }
932 146 : PG_END_TRY();
933 :
934 146 : INSTR_TIME_SET_CURRENT(duration);
935 146 : INSTR_TIME_SUBTRACT(duration, start);
936 :
937 : /* calc differences of buffer counters. */
938 146 : memset(&bufusage, 0, sizeof(BufferUsage));
939 146 : BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
940 :
941 : /* calc differences of WAL counters. */
942 146 : memset(&walusage, 0, sizeof(WalUsage));
943 146 : WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
944 :
945 146 : pgss_store(query_string,
946 : parse->queryId,
947 : parse->stmt_location,
948 : parse->stmt_len,
949 : PGSS_PLAN,
950 146 : INSTR_TIME_GET_MILLISEC(duration),
951 : 0,
952 : &bufusage,
953 : &walusage,
954 : NULL,
955 : NULL,
956 : 0,
957 : 0);
958 : }
959 : else
960 : {
961 : /*
962 : * Even though we're not tracking plan time for this statement, we
963 : * must still increment the nesting level, to ensure that functions
964 : * evaluated during planning are not seen as top-level calls.
965 : */
966 91514 : nesting_level++;
967 91514 : PG_TRY();
968 : {
969 91514 : if (prev_planner_hook)
970 0 : result = prev_planner_hook(parse, query_string, cursorOptions,
971 : boundParams);
972 : else
973 91514 : result = standard_planner(parse, query_string, cursorOptions,
974 : boundParams);
975 : }
976 1314 : PG_FINALLY();
977 : {
978 91514 : nesting_level--;
979 : }
980 91514 : PG_END_TRY();
981 : }
982 :
983 90346 : return result;
984 : }
985 :
986 : /*
987 : * ExecutorStart hook: start up tracking if needed
988 : */
989 : static void
990 109692 : pgss_ExecutorStart(QueryDesc *queryDesc, int eflags)
991 : {
992 109692 : if (prev_ExecutorStart)
993 0 : prev_ExecutorStart(queryDesc, eflags);
994 : else
995 109692 : standard_ExecutorStart(queryDesc, eflags);
996 :
997 : /*
998 : * If query has queryId zero, don't track it. This prevents double
999 : * counting of optimizable statements that are directly contained in
1000 : * utility statements.
1001 : */
1002 109068 : if (pgss_enabled(nesting_level) && queryDesc->plannedstmt->queryId != UINT64CONST(0))
1003 : {
1004 : /*
1005 : * Set up to track total elapsed time in ExecutorRun. Make sure the
1006 : * space is allocated in the per-query context so it will go away at
1007 : * ExecutorEnd.
1008 : */
1009 72432 : if (queryDesc->totaltime == NULL)
1010 : {
1011 : MemoryContext oldcxt;
1012 :
1013 72432 : oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
1014 72432 : queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL, false);
1015 72432 : MemoryContextSwitchTo(oldcxt);
1016 : }
1017 : }
1018 109068 : }
1019 :
1020 : /*
1021 : * ExecutorRun hook: all we need do is track nesting depth
1022 : */
1023 : static void
1024 106812 : pgss_ExecutorRun(QueryDesc *queryDesc, ScanDirection direction, uint64 count)
1025 : {
1026 106812 : nesting_level++;
1027 106812 : PG_TRY();
1028 : {
1029 106812 : if (prev_ExecutorRun)
1030 0 : prev_ExecutorRun(queryDesc, direction, count);
1031 : else
1032 106812 : standard_ExecutorRun(queryDesc, direction, count);
1033 : }
1034 6850 : PG_FINALLY();
1035 : {
1036 106812 : nesting_level--;
1037 : }
1038 106812 : PG_END_TRY();
1039 99962 : }
1040 :
1041 : /*
1042 : * ExecutorFinish hook: all we need do is track nesting depth
1043 : */
1044 : static void
1045 96706 : pgss_ExecutorFinish(QueryDesc *queryDesc)
1046 : {
1047 96706 : nesting_level++;
1048 96706 : PG_TRY();
1049 : {
1050 96706 : if (prev_ExecutorFinish)
1051 0 : prev_ExecutorFinish(queryDesc);
1052 : else
1053 96706 : standard_ExecutorFinish(queryDesc);
1054 : }
1055 330 : PG_FINALLY();
1056 : {
1057 96706 : nesting_level--;
1058 : }
1059 96706 : PG_END_TRY();
1060 96376 : }
1061 :
1062 : /*
1063 : * ExecutorEnd hook: store results if needed
1064 : */
1065 : static void
1066 101808 : pgss_ExecutorEnd(QueryDesc *queryDesc)
1067 : {
1068 101808 : uint64 queryId = queryDesc->plannedstmt->queryId;
1069 :
1070 101808 : if (queryId != UINT64CONST(0) && queryDesc->totaltime &&
1071 69450 : pgss_enabled(nesting_level))
1072 : {
1073 : /*
1074 : * Make sure stats accumulation is done. (Note: it's okay if several
1075 : * levels of hook all do this.)
1076 : */
1077 69450 : InstrEndLoop(queryDesc->totaltime);
1078 :
1079 69292 : pgss_store(queryDesc->sourceText,
1080 : queryId,
1081 69450 : queryDesc->plannedstmt->stmt_location,
1082 69450 : queryDesc->plannedstmt->stmt_len,
1083 : PGSS_EXEC,
1084 69450 : queryDesc->totaltime->total * 1000.0, /* convert to msec */
1085 69450 : queryDesc->estate->es_total_processed,
1086 69450 : &queryDesc->totaltime->bufusage,
1087 69450 : &queryDesc->totaltime->walusage,
1088 158 : queryDesc->estate->es_jit ? &queryDesc->estate->es_jit->instr : NULL,
1089 : NULL,
1090 69450 : queryDesc->estate->es_parallel_workers_to_launch,
1091 69450 : queryDesc->estate->es_parallel_workers_launched);
1092 : }
1093 :
1094 101808 : if (prev_ExecutorEnd)
1095 0 : prev_ExecutorEnd(queryDesc);
1096 : else
1097 101808 : standard_ExecutorEnd(queryDesc);
1098 101808 : }
1099 :
1100 : /*
1101 : * ProcessUtility hook
1102 : */
1103 : static void
1104 64402 : pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
1105 : bool readOnlyTree,
1106 : ProcessUtilityContext context,
1107 : ParamListInfo params, QueryEnvironment *queryEnv,
1108 : DestReceiver *dest, QueryCompletion *qc)
1109 : {
1110 64402 : Node *parsetree = pstmt->utilityStmt;
1111 64402 : uint64 saved_queryId = pstmt->queryId;
1112 64402 : int saved_stmt_location = pstmt->stmt_location;
1113 64402 : int saved_stmt_len = pstmt->stmt_len;
1114 64402 : bool enabled = pgss_track_utility && pgss_enabled(nesting_level);
1115 :
1116 : /*
1117 : * Force utility statements to get queryId zero. We do this even in cases
1118 : * where the statement contains an optimizable statement for which a
1119 : * queryId could be derived (such as EXPLAIN or DECLARE CURSOR). For such
1120 : * cases, runtime control will first go through ProcessUtility and then
1121 : * the executor, and we don't want the executor hooks to do anything,
1122 : * since we are already measuring the statement's costs at the utility
1123 : * level.
1124 : *
1125 : * Note that this is only done if pg_stat_statements is enabled and
1126 : * configured to track utility statements, in the unlikely possibility
1127 : * that user configured another extension to handle utility statements
1128 : * only.
1129 : */
1130 64402 : if (enabled)
1131 54978 : pstmt->queryId = UINT64CONST(0);
1132 :
1133 : /*
1134 : * If it's an EXECUTE statement, we don't track it and don't increment the
1135 : * nesting level. This allows the cycles to be charged to the underlying
1136 : * PREPARE instead (by the Executor hooks), which is much more useful.
1137 : *
1138 : * We also don't track execution of PREPARE. If we did, we would get one
1139 : * hash table entry for the PREPARE (with hash calculated from the query
1140 : * string), and then a different one with the same query string (but hash
1141 : * calculated from the query tree) would be used to accumulate costs of
1142 : * ensuing EXECUTEs. This would be confusing. Since PREPARE doesn't
1143 : * actually run the planner (only parse+rewrite), its costs are generally
1144 : * pretty negligible and it seems okay to just ignore it.
1145 : */
1146 64402 : if (enabled &&
1147 54978 : !IsA(parsetree, ExecuteStmt) &&
1148 48634 : !IsA(parsetree, PrepareStmt))
1149 43912 : {
1150 : instr_time start;
1151 : instr_time duration;
1152 : uint64 rows;
1153 : BufferUsage bufusage_start,
1154 : bufusage;
1155 : WalUsage walusage_start,
1156 : walusage;
1157 :
1158 48402 : bufusage_start = pgBufferUsage;
1159 48402 : walusage_start = pgWalUsage;
1160 48402 : INSTR_TIME_SET_CURRENT(start);
1161 :
1162 48402 : nesting_level++;
1163 48402 : PG_TRY();
1164 : {
1165 48402 : if (prev_ProcessUtility)
1166 0 : prev_ProcessUtility(pstmt, queryString, readOnlyTree,
1167 : context, params, queryEnv,
1168 : dest, qc);
1169 : else
1170 48402 : standard_ProcessUtility(pstmt, queryString, readOnlyTree,
1171 : context, params, queryEnv,
1172 : dest, qc);
1173 : }
1174 4490 : PG_FINALLY();
1175 : {
1176 48402 : nesting_level--;
1177 : }
1178 48402 : PG_END_TRY();
1179 :
1180 : /*
1181 : * CAUTION: do not access the *pstmt data structure again below here.
1182 : * If it was a ROLLBACK or similar, that data structure may have been
1183 : * freed. We must copy everything we still need into local variables,
1184 : * which we did above.
1185 : *
1186 : * For the same reason, we can't risk restoring pstmt->queryId to its
1187 : * former value, which'd otherwise be a good idea.
1188 : */
1189 :
1190 43912 : INSTR_TIME_SET_CURRENT(duration);
1191 43912 : INSTR_TIME_SUBTRACT(duration, start);
1192 :
1193 : /*
1194 : * Track the total number of rows retrieved or affected by the utility
1195 : * statements of COPY, FETCH, CREATE TABLE AS, CREATE MATERIALIZED
1196 : * VIEW, REFRESH MATERIALIZED VIEW and SELECT INTO.
1197 : */
1198 43906 : rows = (qc && (qc->commandTag == CMDTAG_COPY ||
1199 40902 : qc->commandTag == CMDTAG_FETCH ||
1200 40436 : qc->commandTag == CMDTAG_SELECT ||
1201 40068 : qc->commandTag == CMDTAG_REFRESH_MATERIALIZED_VIEW)) ?
1202 87818 : qc->nprocessed : 0;
1203 :
1204 : /* calc differences of buffer counters. */
1205 43912 : memset(&bufusage, 0, sizeof(BufferUsage));
1206 43912 : BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
1207 :
1208 : /* calc differences of WAL counters. */
1209 43912 : memset(&walusage, 0, sizeof(WalUsage));
1210 43912 : WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
1211 :
1212 43912 : pgss_store(queryString,
1213 : saved_queryId,
1214 : saved_stmt_location,
1215 : saved_stmt_len,
1216 : PGSS_EXEC,
1217 43912 : INSTR_TIME_GET_MILLISEC(duration),
1218 : rows,
1219 : &bufusage,
1220 : &walusage,
1221 : NULL,
1222 : NULL,
1223 : 0,
1224 : 0);
1225 : }
1226 : else
1227 : {
1228 : /*
1229 : * Even though we're not tracking execution time for this statement,
1230 : * we must still increment the nesting level, to ensure that functions
1231 : * evaluated within it are not seen as top-level calls. But don't do
1232 : * so for EXECUTE; that way, when control reaches pgss_planner or
1233 : * pgss_ExecutorStart, we will treat the costs as top-level if
1234 : * appropriate. Likewise, don't bump for PREPARE, so that parse
1235 : * analysis will treat the statement as top-level if appropriate.
1236 : *
1237 : * To be absolutely certain we don't mess up the nesting level,
1238 : * evaluate the bump_level condition just once.
1239 : */
1240 16000 : bool bump_level =
1241 25654 : !IsA(parsetree, ExecuteStmt) &&
1242 9654 : !IsA(parsetree, PrepareStmt);
1243 :
1244 16000 : if (bump_level)
1245 9420 : nesting_level++;
1246 16000 : PG_TRY();
1247 : {
1248 16000 : if (prev_ProcessUtility)
1249 0 : prev_ProcessUtility(pstmt, queryString, readOnlyTree,
1250 : context, params, queryEnv,
1251 : dest, qc);
1252 : else
1253 16000 : standard_ProcessUtility(pstmt, queryString, readOnlyTree,
1254 : context, params, queryEnv,
1255 : dest, qc);
1256 : }
1257 240 : PG_FINALLY();
1258 : {
1259 16000 : if (bump_level)
1260 9420 : nesting_level--;
1261 : }
1262 16000 : PG_END_TRY();
1263 : }
1264 59672 : }
1265 :
1266 : /*
1267 : * Store some statistics for a statement.
1268 : *
1269 : * If jstate is not NULL then we're trying to create an entry for which
1270 : * we have no statistics as yet; we just want to record the normalized
1271 : * query string. total_time, rows, bufusage and walusage are ignored in this
1272 : * case.
1273 : *
1274 : * If kind is PGSS_PLAN or PGSS_EXEC, its value is used as the array position
1275 : * for the arrays in the Counters field.
1276 : */
1277 : static void
1278 180906 : pgss_store(const char *query, uint64 queryId,
1279 : int query_location, int query_len,
1280 : pgssStoreKind kind,
1281 : double total_time, uint64 rows,
1282 : const BufferUsage *bufusage,
1283 : const WalUsage *walusage,
1284 : const struct JitInstrumentation *jitusage,
1285 : JumbleState *jstate,
1286 : int parallel_workers_to_launch,
1287 : int parallel_workers_launched)
1288 : {
1289 : pgssHashKey key;
1290 : pgssEntry *entry;
1291 180906 : char *norm_query = NULL;
1292 180906 : int encoding = GetDatabaseEncoding();
1293 :
1294 : Assert(query != NULL);
1295 :
1296 : /* Safety check... */
1297 180906 : if (!pgss || !pgss_hash)
1298 0 : return;
1299 :
1300 : /*
1301 : * Nothing to do if compute_query_id isn't enabled and no other module
1302 : * computed a query identifier.
1303 : */
1304 180906 : if (queryId == UINT64CONST(0))
1305 0 : return;
1306 :
1307 : /*
1308 : * Confine our attention to the relevant part of the string, if the query
1309 : * is a portion of a multi-statement source string, and update query
1310 : * location and length if needed.
1311 : */
1312 180906 : query = CleanQuerytext(query, &query_location, &query_len);
1313 :
1314 : /* Set up key for hashtable search */
1315 :
1316 : /* clear padding */
1317 180906 : memset(&key, 0, sizeof(pgssHashKey));
1318 :
1319 180906 : key.userid = GetUserId();
1320 180906 : key.dbid = MyDatabaseId;
1321 180906 : key.queryid = queryId;
1322 180906 : key.toplevel = (nesting_level == 0);
1323 :
1324 : /* Lookup the hash table entry with shared lock. */
1325 180906 : LWLockAcquire(pgss->lock, LW_SHARED);
1326 :
1327 180906 : entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
1328 :
1329 : /* Create new entry, if not present */
1330 180906 : if (!entry)
1331 : {
1332 : Size query_offset;
1333 : int gc_count;
1334 : bool stored;
1335 : bool do_gc;
1336 :
1337 : /*
1338 : * Create a new, normalized query string if caller asked. We don't
1339 : * need to hold the lock while doing this work. (Note: in any case,
1340 : * it's possible that someone else creates a duplicate hashtable entry
1341 : * in the interval where we don't hold the lock below. That case is
1342 : * handled by entry_alloc.)
1343 : */
1344 54870 : if (jstate)
1345 : {
1346 20748 : LWLockRelease(pgss->lock);
1347 20748 : norm_query = generate_normalized_query(jstate, query,
1348 : query_location,
1349 : &query_len);
1350 20748 : LWLockAcquire(pgss->lock, LW_SHARED);
1351 : }
1352 :
1353 : /* Append new query text to file with only shared lock held */
1354 54870 : stored = qtext_store(norm_query ? norm_query : query, query_len,
1355 : &query_offset, &gc_count);
1356 :
1357 : /*
1358 : * Determine whether we need to garbage collect external query texts
1359 : * while the shared lock is still held. This micro-optimization
1360 : * avoids taking the time to decide this while holding exclusive lock.
1361 : */
1362 54870 : do_gc = need_gc_qtexts();
1363 :
1364 : /* Need exclusive lock to make a new hashtable entry - promote */
1365 54870 : LWLockRelease(pgss->lock);
1366 54870 : LWLockAcquire(pgss->lock, LW_EXCLUSIVE);
1367 :
1368 : /*
1369 : * A garbage collection may have occurred while we weren't holding the
1370 : * lock. In the unlikely event that this happens, the query text we
1371 : * stored above will have been garbage collected, so write it again.
1372 : * This should be infrequent enough that doing it while holding
1373 : * exclusive lock isn't a performance problem.
1374 : */
1375 54870 : if (!stored || pgss->gc_count != gc_count)
1376 0 : stored = qtext_store(norm_query ? norm_query : query, query_len,
1377 : &query_offset, NULL);
1378 :
1379 : /* If we failed to write to the text file, give up */
1380 54870 : if (!stored)
1381 0 : goto done;
1382 :
1383 : /* OK to create a new hashtable entry */
1384 54870 : entry = entry_alloc(&key, query_offset, query_len, encoding,
1385 : jstate != NULL);
1386 :
1387 : /* If needed, perform garbage collection while exclusive lock held */
1388 54870 : if (do_gc)
1389 0 : gc_qtexts();
1390 : }
1391 :
1392 : /* Increment the counts, except when jstate is not NULL */
1393 180906 : if (!jstate)
1394 : {
1395 : Assert(kind == PGSS_PLAN || kind == PGSS_EXEC);
1396 :
1397 : /*
1398 : * Grab the spinlock while updating the counters (see comment about
1399 : * locking rules at the head of the file)
1400 : */
1401 113508 : SpinLockAcquire(&entry->mutex);
1402 :
1403 : /* "Unstick" entry if it was previously sticky */
1404 113508 : if (IS_STICKY(entry->counters))
1405 53364 : entry->counters.usage = USAGE_INIT;
1406 :
1407 113508 : entry->counters.calls[kind] += 1;
1408 113508 : entry->counters.total_time[kind] += total_time;
1409 :
1410 113508 : if (entry->counters.calls[kind] == 1)
1411 : {
1412 53472 : entry->counters.min_time[kind] = total_time;
1413 53472 : entry->counters.max_time[kind] = total_time;
1414 53472 : entry->counters.mean_time[kind] = total_time;
1415 : }
1416 : else
1417 : {
1418 : /*
1419 : * Welford's method for accurately computing variance. See
1420 : * <http://www.johndcook.com/blog/standard_deviation/>
1421 : */
1422 60036 : double old_mean = entry->counters.mean_time[kind];
1423 :
1424 60036 : entry->counters.mean_time[kind] +=
1425 60036 : (total_time - old_mean) / entry->counters.calls[kind];
1426 60036 : entry->counters.sum_var_time[kind] +=
1427 60036 : (total_time - old_mean) * (total_time - entry->counters.mean_time[kind]);
1428 :
1429 : /*
1430 : * Calculate min and max time. min = 0 and max = 0 means that the
1431 : * min/max statistics were reset
1432 : */
1433 60036 : if (entry->counters.min_time[kind] == 0
1434 4 : && entry->counters.max_time[kind] == 0)
1435 : {
1436 4 : entry->counters.min_time[kind] = total_time;
1437 4 : entry->counters.max_time[kind] = total_time;
1438 : }
1439 : else
1440 : {
1441 60032 : if (entry->counters.min_time[kind] > total_time)
1442 12030 : entry->counters.min_time[kind] = total_time;
1443 60032 : if (entry->counters.max_time[kind] < total_time)
1444 5754 : entry->counters.max_time[kind] = total_time;
1445 : }
1446 : }
1447 113508 : entry->counters.rows += rows;
1448 113508 : entry->counters.shared_blks_hit += bufusage->shared_blks_hit;
1449 113508 : entry->counters.shared_blks_read += bufusage->shared_blks_read;
1450 113508 : entry->counters.shared_blks_dirtied += bufusage->shared_blks_dirtied;
1451 113508 : entry->counters.shared_blks_written += bufusage->shared_blks_written;
1452 113508 : entry->counters.local_blks_hit += bufusage->local_blks_hit;
1453 113508 : entry->counters.local_blks_read += bufusage->local_blks_read;
1454 113508 : entry->counters.local_blks_dirtied += bufusage->local_blks_dirtied;
1455 113508 : entry->counters.local_blks_written += bufusage->local_blks_written;
1456 113508 : entry->counters.temp_blks_read += bufusage->temp_blks_read;
1457 113508 : entry->counters.temp_blks_written += bufusage->temp_blks_written;
1458 113508 : entry->counters.shared_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_read_time);
1459 113508 : entry->counters.shared_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_write_time);
1460 113508 : entry->counters.local_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_read_time);
1461 113508 : entry->counters.local_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_write_time);
1462 113508 : entry->counters.temp_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_read_time);
1463 113508 : entry->counters.temp_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_write_time);
1464 113508 : entry->counters.usage += USAGE_EXEC(total_time);
1465 113508 : entry->counters.wal_records += walusage->wal_records;
1466 113508 : entry->counters.wal_fpi += walusage->wal_fpi;
1467 113508 : entry->counters.wal_bytes += walusage->wal_bytes;
1468 113508 : if (jitusage)
1469 : {
1470 158 : entry->counters.jit_functions += jitusage->created_functions;
1471 158 : entry->counters.jit_generation_time += INSTR_TIME_GET_MILLISEC(jitusage->generation_counter);
1472 :
1473 158 : if (INSTR_TIME_GET_MILLISEC(jitusage->deform_counter))
1474 154 : entry->counters.jit_deform_count++;
1475 158 : entry->counters.jit_deform_time += INSTR_TIME_GET_MILLISEC(jitusage->deform_counter);
1476 :
1477 158 : if (INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter))
1478 74 : entry->counters.jit_inlining_count++;
1479 158 : entry->counters.jit_inlining_time += INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter);
1480 :
1481 158 : if (INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter))
1482 154 : entry->counters.jit_optimization_count++;
1483 158 : entry->counters.jit_optimization_time += INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter);
1484 :
1485 158 : if (INSTR_TIME_GET_MILLISEC(jitusage->emission_counter))
1486 154 : entry->counters.jit_emission_count++;
1487 158 : entry->counters.jit_emission_time += INSTR_TIME_GET_MILLISEC(jitusage->emission_counter);
1488 : }
1489 :
1490 : /* parallel worker counters */
1491 113508 : entry->counters.parallel_workers_to_launch += parallel_workers_to_launch;
1492 113508 : entry->counters.parallel_workers_launched += parallel_workers_launched;
1493 :
1494 113508 : SpinLockRelease(&entry->mutex);
1495 : }
1496 :
1497 67398 : done:
1498 180906 : LWLockRelease(pgss->lock);
1499 :
1500 : /* We postpone this clean-up until we're out of the lock */
1501 180906 : if (norm_query)
1502 20748 : pfree(norm_query);
1503 : }
1504 :
1505 : /*
1506 : * Reset statement statistics corresponding to userid, dbid, and queryid.
1507 : */
1508 : Datum
1509 2 : pg_stat_statements_reset_1_7(PG_FUNCTION_ARGS)
1510 : {
1511 : Oid userid;
1512 : Oid dbid;
1513 : uint64 queryid;
1514 :
1515 2 : userid = PG_GETARG_OID(0);
1516 2 : dbid = PG_GETARG_OID(1);
1517 2 : queryid = (uint64) PG_GETARG_INT64(2);
1518 :
1519 2 : entry_reset(userid, dbid, queryid, false);
1520 :
1521 2 : PG_RETURN_VOID();
1522 : }
1523 :
1524 : Datum
1525 156 : pg_stat_statements_reset_1_11(PG_FUNCTION_ARGS)
1526 : {
1527 : Oid userid;
1528 : Oid dbid;
1529 : uint64 queryid;
1530 : bool minmax_only;
1531 :
1532 156 : userid = PG_GETARG_OID(0);
1533 156 : dbid = PG_GETARG_OID(1);
1534 156 : queryid = (uint64) PG_GETARG_INT64(2);
1535 156 : minmax_only = PG_GETARG_BOOL(3);
1536 :
1537 156 : PG_RETURN_TIMESTAMPTZ(entry_reset(userid, dbid, queryid, minmax_only));
1538 : }
1539 :
1540 : /*
1541 : * Reset statement statistics.
1542 : */
1543 : Datum
1544 2 : pg_stat_statements_reset(PG_FUNCTION_ARGS)
1545 : {
1546 2 : entry_reset(0, 0, 0, false);
1547 :
1548 2 : PG_RETURN_VOID();
1549 : }
1550 :
1551 : /* Number of output arguments (columns) for various API versions */
1552 : #define PG_STAT_STATEMENTS_COLS_V1_0 14
1553 : #define PG_STAT_STATEMENTS_COLS_V1_1 18
1554 : #define PG_STAT_STATEMENTS_COLS_V1_2 19
1555 : #define PG_STAT_STATEMENTS_COLS_V1_3 23
1556 : #define PG_STAT_STATEMENTS_COLS_V1_8 32
1557 : #define PG_STAT_STATEMENTS_COLS_V1_9 33
1558 : #define PG_STAT_STATEMENTS_COLS_V1_10 43
1559 : #define PG_STAT_STATEMENTS_COLS_V1_11 49
1560 : #define PG_STAT_STATEMENTS_COLS_V1_12 51
1561 : #define PG_STAT_STATEMENTS_COLS 51 /* maximum of above */
1562 :
1563 : /*
1564 : * Retrieve statement statistics.
1565 : *
1566 : * The SQL API of this function has changed multiple times, and will likely
1567 : * do so again in future. To support the case where a newer version of this
1568 : * loadable module is being used with an old SQL declaration of the function,
1569 : * we continue to support the older API versions. For 1.2 and later, the
1570 : * expected API version is identified by embedding it in the C name of the
1571 : * function. Unfortunately we weren't bright enough to do that for 1.1.
1572 : */
1573 : Datum
1574 176 : pg_stat_statements_1_12(PG_FUNCTION_ARGS)
1575 : {
1576 176 : bool showtext = PG_GETARG_BOOL(0);
1577 :
1578 176 : pg_stat_statements_internal(fcinfo, PGSS_V1_12, showtext);
1579 :
1580 176 : return (Datum) 0;
1581 : }
1582 :
1583 : Datum
1584 2 : pg_stat_statements_1_11(PG_FUNCTION_ARGS)
1585 : {
1586 2 : bool showtext = PG_GETARG_BOOL(0);
1587 :
1588 2 : pg_stat_statements_internal(fcinfo, PGSS_V1_11, showtext);
1589 :
1590 2 : return (Datum) 0;
1591 : }
1592 :
1593 : Datum
1594 2 : pg_stat_statements_1_10(PG_FUNCTION_ARGS)
1595 : {
1596 2 : bool showtext = PG_GETARG_BOOL(0);
1597 :
1598 2 : pg_stat_statements_internal(fcinfo, PGSS_V1_10, showtext);
1599 :
1600 2 : return (Datum) 0;
1601 : }
1602 :
1603 : Datum
1604 2 : pg_stat_statements_1_9(PG_FUNCTION_ARGS)
1605 : {
1606 2 : bool showtext = PG_GETARG_BOOL(0);
1607 :
1608 2 : pg_stat_statements_internal(fcinfo, PGSS_V1_9, showtext);
1609 :
1610 2 : return (Datum) 0;
1611 : }
1612 :
1613 : Datum
1614 2 : pg_stat_statements_1_8(PG_FUNCTION_ARGS)
1615 : {
1616 2 : bool showtext = PG_GETARG_BOOL(0);
1617 :
1618 2 : pg_stat_statements_internal(fcinfo, PGSS_V1_8, showtext);
1619 :
1620 2 : return (Datum) 0;
1621 : }
1622 :
1623 : Datum
1624 2 : pg_stat_statements_1_3(PG_FUNCTION_ARGS)
1625 : {
1626 2 : bool showtext = PG_GETARG_BOOL(0);
1627 :
1628 2 : pg_stat_statements_internal(fcinfo, PGSS_V1_3, showtext);
1629 :
1630 2 : return (Datum) 0;
1631 : }
1632 :
1633 : Datum
1634 0 : pg_stat_statements_1_2(PG_FUNCTION_ARGS)
1635 : {
1636 0 : bool showtext = PG_GETARG_BOOL(0);
1637 :
1638 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_2, showtext);
1639 :
1640 0 : return (Datum) 0;
1641 : }
1642 :
1643 : /*
1644 : * Legacy entry point for pg_stat_statements() API versions 1.0 and 1.1.
1645 : * This can be removed someday, perhaps.
1646 : */
1647 : Datum
1648 0 : pg_stat_statements(PG_FUNCTION_ARGS)
1649 : {
1650 : /* If it's really API 1.1, we'll figure that out below */
1651 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_0, true);
1652 :
1653 0 : return (Datum) 0;
1654 : }
1655 :
1656 : /* Common code for all versions of pg_stat_statements() */
1657 : static void
1658 186 : pg_stat_statements_internal(FunctionCallInfo fcinfo,
1659 : pgssVersion api_version,
1660 : bool showtext)
1661 : {
1662 186 : ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1663 186 : Oid userid = GetUserId();
1664 186 : bool is_allowed_role = false;
1665 186 : char *qbuffer = NULL;
1666 186 : Size qbuffer_size = 0;
1667 186 : Size extent = 0;
1668 186 : int gc_count = 0;
1669 : HASH_SEQ_STATUS hash_seq;
1670 : pgssEntry *entry;
1671 :
1672 : /*
1673 : * Superusers or roles with the privileges of pg_read_all_stats members
1674 : * are allowed
1675 : */
1676 186 : is_allowed_role = has_privs_of_role(userid, ROLE_PG_READ_ALL_STATS);
1677 :
1678 : /* hash table must exist already */
1679 186 : if (!pgss || !pgss_hash)
1680 0 : ereport(ERROR,
1681 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1682 : errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
1683 :
1684 186 : InitMaterializedSRF(fcinfo, 0);
1685 :
1686 : /*
1687 : * Check we have the expected number of output arguments. Aside from
1688 : * being a good safety check, we need a kluge here to detect API version
1689 : * 1.1, which was wedged into the code in an ill-considered way.
1690 : */
1691 186 : switch (rsinfo->setDesc->natts)
1692 : {
1693 0 : case PG_STAT_STATEMENTS_COLS_V1_0:
1694 0 : if (api_version != PGSS_V1_0)
1695 0 : elog(ERROR, "incorrect number of output arguments");
1696 0 : break;
1697 0 : case PG_STAT_STATEMENTS_COLS_V1_1:
1698 : /* pg_stat_statements() should have told us 1.0 */
1699 0 : if (api_version != PGSS_V1_0)
1700 0 : elog(ERROR, "incorrect number of output arguments");
1701 0 : api_version = PGSS_V1_1;
1702 0 : break;
1703 0 : case PG_STAT_STATEMENTS_COLS_V1_2:
1704 0 : if (api_version != PGSS_V1_2)
1705 0 : elog(ERROR, "incorrect number of output arguments");
1706 0 : break;
1707 2 : case PG_STAT_STATEMENTS_COLS_V1_3:
1708 2 : if (api_version != PGSS_V1_3)
1709 0 : elog(ERROR, "incorrect number of output arguments");
1710 2 : break;
1711 2 : case PG_STAT_STATEMENTS_COLS_V1_8:
1712 2 : if (api_version != PGSS_V1_8)
1713 0 : elog(ERROR, "incorrect number of output arguments");
1714 2 : break;
1715 2 : case PG_STAT_STATEMENTS_COLS_V1_9:
1716 2 : if (api_version != PGSS_V1_9)
1717 0 : elog(ERROR, "incorrect number of output arguments");
1718 2 : break;
1719 2 : case PG_STAT_STATEMENTS_COLS_V1_10:
1720 2 : if (api_version != PGSS_V1_10)
1721 0 : elog(ERROR, "incorrect number of output arguments");
1722 2 : break;
1723 2 : case PG_STAT_STATEMENTS_COLS_V1_11:
1724 2 : if (api_version != PGSS_V1_11)
1725 0 : elog(ERROR, "incorrect number of output arguments");
1726 2 : break;
1727 176 : case PG_STAT_STATEMENTS_COLS_V1_12:
1728 176 : if (api_version != PGSS_V1_12)
1729 0 : elog(ERROR, "incorrect number of output arguments");
1730 176 : break;
1731 0 : default:
1732 0 : elog(ERROR, "incorrect number of output arguments");
1733 : }
1734 :
1735 : /*
1736 : * We'd like to load the query text file (if needed) while not holding any
1737 : * lock on pgss->lock. In the worst case we'll have to do this again
1738 : * after we have the lock, but it's unlikely enough to make this a win
1739 : * despite occasional duplicated work. We need to reload if anybody
1740 : * writes to the file (either a retail qtext_store(), or a garbage
1741 : * collection) between this point and where we've gotten shared lock. If
1742 : * a qtext_store is actually in progress when we look, we might as well
1743 : * skip the speculative load entirely.
1744 : */
1745 186 : if (showtext)
1746 : {
1747 : int n_writers;
1748 :
1749 : /* Take the mutex so we can examine variables */
1750 186 : SpinLockAcquire(&pgss->mutex);
1751 186 : extent = pgss->extent;
1752 186 : n_writers = pgss->n_writers;
1753 186 : gc_count = pgss->gc_count;
1754 186 : SpinLockRelease(&pgss->mutex);
1755 :
1756 : /* No point in loading file now if there are active writers */
1757 186 : if (n_writers == 0)
1758 186 : qbuffer = qtext_load_file(&qbuffer_size);
1759 : }
1760 :
1761 : /*
1762 : * Get shared lock, load or reload the query text file if we must, and
1763 : * iterate over the hashtable entries.
1764 : *
1765 : * With a large hash table, we might be holding the lock rather longer
1766 : * than one could wish. However, this only blocks creation of new hash
1767 : * table entries, and the larger the hash table the less likely that is to
1768 : * be needed. So we can hope this is okay. Perhaps someday we'll decide
1769 : * we need to partition the hash table to limit the time spent holding any
1770 : * one lock.
1771 : */
1772 186 : LWLockAcquire(pgss->lock, LW_SHARED);
1773 :
1774 186 : if (showtext)
1775 : {
1776 : /*
1777 : * Here it is safe to examine extent and gc_count without taking the
1778 : * mutex. Note that although other processes might change
1779 : * pgss->extent just after we look at it, the strings they then write
1780 : * into the file cannot yet be referenced in the hashtable, so we
1781 : * don't care whether we see them or not.
1782 : *
1783 : * If qtext_load_file fails, we just press on; we'll return NULL for
1784 : * every query text.
1785 : */
1786 186 : if (qbuffer == NULL ||
1787 186 : pgss->extent != extent ||
1788 186 : pgss->gc_count != gc_count)
1789 : {
1790 0 : free(qbuffer);
1791 0 : qbuffer = qtext_load_file(&qbuffer_size);
1792 : }
1793 : }
1794 :
1795 186 : hash_seq_init(&hash_seq, pgss_hash);
1796 52006 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
1797 : {
1798 : Datum values[PG_STAT_STATEMENTS_COLS];
1799 : bool nulls[PG_STAT_STATEMENTS_COLS];
1800 51820 : int i = 0;
1801 : Counters tmp;
1802 : double stddev;
1803 51820 : int64 queryid = entry->key.queryid;
1804 : TimestampTz stats_since;
1805 : TimestampTz minmax_stats_since;
1806 :
1807 51820 : memset(values, 0, sizeof(values));
1808 51820 : memset(nulls, 0, sizeof(nulls));
1809 :
1810 51820 : values[i++] = ObjectIdGetDatum(entry->key.userid);
1811 51820 : values[i++] = ObjectIdGetDatum(entry->key.dbid);
1812 51820 : if (api_version >= PGSS_V1_9)
1813 51794 : values[i++] = BoolGetDatum(entry->key.toplevel);
1814 :
1815 51820 : if (is_allowed_role || entry->key.userid == userid)
1816 : {
1817 51812 : if (api_version >= PGSS_V1_2)
1818 51812 : values[i++] = Int64GetDatumFast(queryid);
1819 :
1820 51812 : if (showtext)
1821 : {
1822 51812 : char *qstr = qtext_fetch(entry->query_offset,
1823 : entry->query_len,
1824 : qbuffer,
1825 : qbuffer_size);
1826 :
1827 51812 : if (qstr)
1828 : {
1829 : char *enc;
1830 :
1831 51812 : enc = pg_any_to_server(qstr,
1832 : entry->query_len,
1833 : entry->encoding);
1834 :
1835 51812 : values[i++] = CStringGetTextDatum(enc);
1836 :
1837 51812 : if (enc != qstr)
1838 0 : pfree(enc);
1839 : }
1840 : else
1841 : {
1842 : /* Just return a null if we fail to find the text */
1843 0 : nulls[i++] = true;
1844 : }
1845 : }
1846 : else
1847 : {
1848 : /* Query text not requested */
1849 0 : nulls[i++] = true;
1850 : }
1851 : }
1852 : else
1853 : {
1854 : /* Don't show queryid */
1855 8 : if (api_version >= PGSS_V1_2)
1856 8 : nulls[i++] = true;
1857 :
1858 : /*
1859 : * Don't show query text, but hint as to the reason for not doing
1860 : * so if it was requested
1861 : */
1862 8 : if (showtext)
1863 8 : values[i++] = CStringGetTextDatum("<insufficient privilege>");
1864 : else
1865 0 : nulls[i++] = true;
1866 : }
1867 :
1868 : /* copy counters to a local variable to keep locking time short */
1869 51820 : SpinLockAcquire(&entry->mutex);
1870 51820 : tmp = entry->counters;
1871 51820 : SpinLockRelease(&entry->mutex);
1872 :
1873 : /*
1874 : * The spinlock is not required when reading these two as they are
1875 : * always updated when holding pgss->lock exclusively.
1876 : */
1877 51820 : stats_since = entry->stats_since;
1878 51820 : minmax_stats_since = entry->minmax_stats_since;
1879 :
1880 : /* Skip entry if unexecuted (ie, it's a pending "sticky" entry) */
1881 51820 : if (IS_STICKY(tmp))
1882 78 : continue;
1883 :
1884 : /* Note that we rely on PGSS_PLAN being 0 and PGSS_EXEC being 1. */
1885 155226 : for (int kind = 0; kind < PGSS_NUMKIND; kind++)
1886 : {
1887 103484 : if (kind == PGSS_EXEC || api_version >= PGSS_V1_8)
1888 : {
1889 103476 : values[i++] = Int64GetDatumFast(tmp.calls[kind]);
1890 103476 : values[i++] = Float8GetDatumFast(tmp.total_time[kind]);
1891 : }
1892 :
1893 103484 : if ((kind == PGSS_EXEC && api_version >= PGSS_V1_3) ||
1894 : api_version >= PGSS_V1_8)
1895 : {
1896 103476 : values[i++] = Float8GetDatumFast(tmp.min_time[kind]);
1897 103476 : values[i++] = Float8GetDatumFast(tmp.max_time[kind]);
1898 103476 : values[i++] = Float8GetDatumFast(tmp.mean_time[kind]);
1899 :
1900 : /*
1901 : * Note we are calculating the population variance here, not
1902 : * the sample variance, as we have data for the whole
1903 : * population, so Bessel's correction is not used, and we
1904 : * don't divide by tmp.calls - 1.
1905 : */
1906 103476 : if (tmp.calls[kind] > 1)
1907 9156 : stddev = sqrt(tmp.sum_var_time[kind] / tmp.calls[kind]);
1908 : else
1909 94320 : stddev = 0.0;
1910 103476 : values[i++] = Float8GetDatumFast(stddev);
1911 : }
1912 : }
1913 51742 : values[i++] = Int64GetDatumFast(tmp.rows);
1914 51742 : values[i++] = Int64GetDatumFast(tmp.shared_blks_hit);
1915 51742 : values[i++] = Int64GetDatumFast(tmp.shared_blks_read);
1916 51742 : if (api_version >= PGSS_V1_1)
1917 51742 : values[i++] = Int64GetDatumFast(tmp.shared_blks_dirtied);
1918 51742 : values[i++] = Int64GetDatumFast(tmp.shared_blks_written);
1919 51742 : values[i++] = Int64GetDatumFast(tmp.local_blks_hit);
1920 51742 : values[i++] = Int64GetDatumFast(tmp.local_blks_read);
1921 51742 : if (api_version >= PGSS_V1_1)
1922 51742 : values[i++] = Int64GetDatumFast(tmp.local_blks_dirtied);
1923 51742 : values[i++] = Int64GetDatumFast(tmp.local_blks_written);
1924 51742 : values[i++] = Int64GetDatumFast(tmp.temp_blks_read);
1925 51742 : values[i++] = Int64GetDatumFast(tmp.temp_blks_written);
1926 51742 : if (api_version >= PGSS_V1_1)
1927 : {
1928 51742 : values[i++] = Float8GetDatumFast(tmp.shared_blk_read_time);
1929 51742 : values[i++] = Float8GetDatumFast(tmp.shared_blk_write_time);
1930 : }
1931 51742 : if (api_version >= PGSS_V1_11)
1932 : {
1933 51680 : values[i++] = Float8GetDatumFast(tmp.local_blk_read_time);
1934 51680 : values[i++] = Float8GetDatumFast(tmp.local_blk_write_time);
1935 : }
1936 51742 : if (api_version >= PGSS_V1_10)
1937 : {
1938 51702 : values[i++] = Float8GetDatumFast(tmp.temp_blk_read_time);
1939 51702 : values[i++] = Float8GetDatumFast(tmp.temp_blk_write_time);
1940 : }
1941 51742 : if (api_version >= PGSS_V1_8)
1942 : {
1943 : char buf[256];
1944 : Datum wal_bytes;
1945 :
1946 51734 : values[i++] = Int64GetDatumFast(tmp.wal_records);
1947 51734 : values[i++] = Int64GetDatumFast(tmp.wal_fpi);
1948 :
1949 51734 : snprintf(buf, sizeof buf, UINT64_FORMAT, tmp.wal_bytes);
1950 :
1951 : /* Convert to numeric. */
1952 51734 : wal_bytes = DirectFunctionCall3(numeric_in,
1953 : CStringGetDatum(buf),
1954 : ObjectIdGetDatum(0),
1955 : Int32GetDatum(-1));
1956 51734 : values[i++] = wal_bytes;
1957 : }
1958 51742 : if (api_version >= PGSS_V1_10)
1959 : {
1960 51702 : values[i++] = Int64GetDatumFast(tmp.jit_functions);
1961 51702 : values[i++] = Float8GetDatumFast(tmp.jit_generation_time);
1962 51702 : values[i++] = Int64GetDatumFast(tmp.jit_inlining_count);
1963 51702 : values[i++] = Float8GetDatumFast(tmp.jit_inlining_time);
1964 51702 : values[i++] = Int64GetDatumFast(tmp.jit_optimization_count);
1965 51702 : values[i++] = Float8GetDatumFast(tmp.jit_optimization_time);
1966 51702 : values[i++] = Int64GetDatumFast(tmp.jit_emission_count);
1967 51702 : values[i++] = Float8GetDatumFast(tmp.jit_emission_time);
1968 : }
1969 51742 : if (api_version >= PGSS_V1_11)
1970 : {
1971 51680 : values[i++] = Int64GetDatumFast(tmp.jit_deform_count);
1972 51680 : values[i++] = Float8GetDatumFast(tmp.jit_deform_time);
1973 : }
1974 51742 : if (api_version >= PGSS_V1_12)
1975 : {
1976 51654 : values[i++] = Int64GetDatumFast(tmp.parallel_workers_to_launch);
1977 51654 : values[i++] = Int64GetDatumFast(tmp.parallel_workers_launched);
1978 : }
1979 51742 : if (api_version >= PGSS_V1_11)
1980 : {
1981 51680 : values[i++] = TimestampTzGetDatum(stats_since);
1982 51680 : values[i++] = TimestampTzGetDatum(minmax_stats_since);
1983 : }
1984 :
1985 : Assert(i == (api_version == PGSS_V1_0 ? PG_STAT_STATEMENTS_COLS_V1_0 :
1986 : api_version == PGSS_V1_1 ? PG_STAT_STATEMENTS_COLS_V1_1 :
1987 : api_version == PGSS_V1_2 ? PG_STAT_STATEMENTS_COLS_V1_2 :
1988 : api_version == PGSS_V1_3 ? PG_STAT_STATEMENTS_COLS_V1_3 :
1989 : api_version == PGSS_V1_8 ? PG_STAT_STATEMENTS_COLS_V1_8 :
1990 : api_version == PGSS_V1_9 ? PG_STAT_STATEMENTS_COLS_V1_9 :
1991 : api_version == PGSS_V1_10 ? PG_STAT_STATEMENTS_COLS_V1_10 :
1992 : api_version == PGSS_V1_11 ? PG_STAT_STATEMENTS_COLS_V1_11 :
1993 : api_version == PGSS_V1_12 ? PG_STAT_STATEMENTS_COLS_V1_12 :
1994 : -1 /* fail if you forget to update this assert */ ));
1995 :
1996 51742 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
1997 : }
1998 :
1999 186 : LWLockRelease(pgss->lock);
2000 :
2001 186 : free(qbuffer);
2002 186 : }
2003 :
2004 : /* Number of output arguments (columns) for pg_stat_statements_info */
2005 : #define PG_STAT_STATEMENTS_INFO_COLS 2
2006 :
2007 : /*
2008 : * Return statistics of pg_stat_statements.
2009 : */
2010 : Datum
2011 4 : pg_stat_statements_info(PG_FUNCTION_ARGS)
2012 : {
2013 : pgssGlobalStats stats;
2014 : TupleDesc tupdesc;
2015 4 : Datum values[PG_STAT_STATEMENTS_INFO_COLS] = {0};
2016 4 : bool nulls[PG_STAT_STATEMENTS_INFO_COLS] = {0};
2017 :
2018 4 : if (!pgss || !pgss_hash)
2019 0 : ereport(ERROR,
2020 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2021 : errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
2022 :
2023 : /* Build a tuple descriptor for our result type */
2024 4 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
2025 0 : elog(ERROR, "return type must be a row type");
2026 :
2027 : /* Read global statistics for pg_stat_statements */
2028 4 : SpinLockAcquire(&pgss->mutex);
2029 4 : stats = pgss->stats;
2030 4 : SpinLockRelease(&pgss->mutex);
2031 :
2032 4 : values[0] = Int64GetDatum(stats.dealloc);
2033 4 : values[1] = TimestampTzGetDatum(stats.stats_reset);
2034 :
2035 4 : PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
2036 : }
2037 :
2038 : /*
2039 : * Estimate shared memory space needed.
2040 : */
2041 : static Size
2042 14 : pgss_memsize(void)
2043 : {
2044 : Size size;
2045 :
2046 14 : size = MAXALIGN(sizeof(pgssSharedState));
2047 14 : size = add_size(size, hash_estimate_size(pgss_max, sizeof(pgssEntry)));
2048 :
2049 14 : return size;
2050 : }
2051 :
2052 : /*
2053 : * Allocate a new hashtable entry.
2054 : * caller must hold an exclusive lock on pgss->lock
2055 : *
2056 : * "query" need not be null-terminated; we rely on query_len instead
2057 : *
2058 : * If "sticky" is true, make the new entry artificially sticky so that it will
2059 : * probably still be there when the query finishes execution. We do this by
2060 : * giving it a median usage value rather than the normal value. (Strictly
2061 : * speaking, query strings are normalized on a best effort basis, though it
2062 : * would be difficult to demonstrate this even under artificial conditions.)
2063 : *
2064 : * Note: despite needing exclusive lock, it's not an error for the target
2065 : * entry to already exist. This is because pgss_store releases and
2066 : * reacquires lock after failing to find a match; so someone else could
2067 : * have made the entry while we waited to get exclusive lock.
2068 : */
2069 : static pgssEntry *
2070 105072 : entry_alloc(pgssHashKey *key, Size query_offset, int query_len, int encoding,
2071 : bool sticky)
2072 : {
2073 : pgssEntry *entry;
2074 : bool found;
2075 :
2076 : /* Make space if needed */
2077 105072 : while (hash_get_num_entries(pgss_hash) >= pgss_max)
2078 0 : entry_dealloc();
2079 :
2080 : /* Find or create an entry with desired hash code */
2081 105072 : entry = (pgssEntry *) hash_search(pgss_hash, key, HASH_ENTER, &found);
2082 :
2083 105072 : if (!found)
2084 : {
2085 : /* New entry, initialize it */
2086 :
2087 : /* reset the statistics */
2088 105072 : memset(&entry->counters, 0, sizeof(Counters));
2089 : /* set the appropriate initial usage count */
2090 105072 : entry->counters.usage = sticky ? pgss->cur_median_usage : USAGE_INIT;
2091 : /* re-initialize the mutex each time ... we assume no one using it */
2092 105072 : SpinLockInit(&entry->mutex);
2093 : /* ... and don't forget the query text metadata */
2094 : Assert(query_len >= 0);
2095 105072 : entry->query_offset = query_offset;
2096 105072 : entry->query_len = query_len;
2097 105072 : entry->encoding = encoding;
2098 105072 : entry->stats_since = GetCurrentTimestamp();
2099 105072 : entry->minmax_stats_since = entry->stats_since;
2100 : }
2101 :
2102 105072 : return entry;
2103 : }
2104 :
2105 : /*
2106 : * qsort comparator for sorting into increasing usage order
2107 : */
2108 : static int
2109 0 : entry_cmp(const void *lhs, const void *rhs)
2110 : {
2111 0 : double l_usage = (*(pgssEntry *const *) lhs)->counters.usage;
2112 0 : double r_usage = (*(pgssEntry *const *) rhs)->counters.usage;
2113 :
2114 0 : if (l_usage < r_usage)
2115 0 : return -1;
2116 0 : else if (l_usage > r_usage)
2117 0 : return +1;
2118 : else
2119 0 : return 0;
2120 : }
2121 :
2122 : /*
2123 : * Deallocate least-used entries.
2124 : *
2125 : * Caller must hold an exclusive lock on pgss->lock.
2126 : */
2127 : static void
2128 0 : entry_dealloc(void)
2129 : {
2130 : HASH_SEQ_STATUS hash_seq;
2131 : pgssEntry **entries;
2132 : pgssEntry *entry;
2133 : int nvictims;
2134 : int i;
2135 : Size tottextlen;
2136 : int nvalidtexts;
2137 :
2138 : /*
2139 : * Sort entries by usage and deallocate USAGE_DEALLOC_PERCENT of them.
2140 : * While we're scanning the table, apply the decay factor to the usage
2141 : * values, and update the mean query length.
2142 : *
2143 : * Note that the mean query length is almost immediately obsolete, since
2144 : * we compute it before not after discarding the least-used entries.
2145 : * Hopefully, that doesn't affect the mean too much; it doesn't seem worth
2146 : * making two passes to get a more current result. Likewise, the new
2147 : * cur_median_usage includes the entries we're about to zap.
2148 : */
2149 :
2150 0 : entries = palloc(hash_get_num_entries(pgss_hash) * sizeof(pgssEntry *));
2151 :
2152 0 : i = 0;
2153 0 : tottextlen = 0;
2154 0 : nvalidtexts = 0;
2155 :
2156 0 : hash_seq_init(&hash_seq, pgss_hash);
2157 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2158 : {
2159 0 : entries[i++] = entry;
2160 : /* "Sticky" entries get a different usage decay rate. */
2161 0 : if (IS_STICKY(entry->counters))
2162 0 : entry->counters.usage *= STICKY_DECREASE_FACTOR;
2163 : else
2164 0 : entry->counters.usage *= USAGE_DECREASE_FACTOR;
2165 : /* In the mean length computation, ignore dropped texts. */
2166 0 : if (entry->query_len >= 0)
2167 : {
2168 0 : tottextlen += entry->query_len + 1;
2169 0 : nvalidtexts++;
2170 : }
2171 : }
2172 :
2173 : /* Sort into increasing order by usage */
2174 0 : qsort(entries, i, sizeof(pgssEntry *), entry_cmp);
2175 :
2176 : /* Record the (approximate) median usage */
2177 0 : if (i > 0)
2178 0 : pgss->cur_median_usage = entries[i / 2]->counters.usage;
2179 : /* Record the mean query length */
2180 0 : if (nvalidtexts > 0)
2181 0 : pgss->mean_query_len = tottextlen / nvalidtexts;
2182 : else
2183 0 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
2184 :
2185 : /* Now zap an appropriate fraction of lowest-usage entries */
2186 0 : nvictims = Max(10, i * USAGE_DEALLOC_PERCENT / 100);
2187 0 : nvictims = Min(nvictims, i);
2188 :
2189 0 : for (i = 0; i < nvictims; i++)
2190 : {
2191 0 : hash_search(pgss_hash, &entries[i]->key, HASH_REMOVE, NULL);
2192 : }
2193 :
2194 0 : pfree(entries);
2195 :
2196 : /* Increment the number of times entries are deallocated */
2197 0 : SpinLockAcquire(&pgss->mutex);
2198 0 : pgss->stats.dealloc += 1;
2199 0 : SpinLockRelease(&pgss->mutex);
2200 0 : }
2201 :
2202 : /*
2203 : * Given a query string (not necessarily null-terminated), allocate a new
2204 : * entry in the external query text file and store the string there.
2205 : *
2206 : * If successful, returns true, and stores the new entry's offset in the file
2207 : * into *query_offset. Also, if gc_count isn't NULL, *gc_count is set to the
2208 : * number of garbage collections that have occurred so far.
2209 : *
2210 : * On failure, returns false.
2211 : *
2212 : * At least a shared lock on pgss->lock must be held by the caller, so as
2213 : * to prevent a concurrent garbage collection. Share-lock-holding callers
2214 : * should pass a gc_count pointer to obtain the number of garbage collections,
2215 : * so that they can recheck the count after obtaining exclusive lock to
2216 : * detect whether a garbage collection occurred (and removed this entry).
2217 : */
2218 : static bool
2219 54870 : qtext_store(const char *query, int query_len,
2220 : Size *query_offset, int *gc_count)
2221 : {
2222 : Size off;
2223 : int fd;
2224 :
2225 : /*
2226 : * We use a spinlock to protect extent/n_writers/gc_count, so that
2227 : * multiple processes may execute this function concurrently.
2228 : */
2229 54870 : SpinLockAcquire(&pgss->mutex);
2230 54870 : off = pgss->extent;
2231 54870 : pgss->extent += query_len + 1;
2232 54870 : pgss->n_writers++;
2233 54870 : if (gc_count)
2234 54870 : *gc_count = pgss->gc_count;
2235 54870 : SpinLockRelease(&pgss->mutex);
2236 :
2237 54870 : *query_offset = off;
2238 :
2239 : /*
2240 : * Don't allow the file to grow larger than what qtext_load_file can
2241 : * (theoretically) handle. This has been seen to be reachable on 32-bit
2242 : * platforms.
2243 : */
2244 54870 : if (unlikely(query_len >= MaxAllocHugeSize - off))
2245 : {
2246 0 : errno = EFBIG; /* not quite right, but it'll do */
2247 0 : fd = -1;
2248 0 : goto error;
2249 : }
2250 :
2251 : /* Now write the data into the successfully-reserved part of the file */
2252 54870 : fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDWR | O_CREAT | PG_BINARY);
2253 54870 : if (fd < 0)
2254 0 : goto error;
2255 :
2256 54870 : if (pg_pwrite(fd, query, query_len, off) != query_len)
2257 0 : goto error;
2258 54870 : if (pg_pwrite(fd, "\0", 1, off + query_len) != 1)
2259 0 : goto error;
2260 :
2261 54870 : CloseTransientFile(fd);
2262 :
2263 : /* Mark our write complete */
2264 54870 : SpinLockAcquire(&pgss->mutex);
2265 54870 : pgss->n_writers--;
2266 54870 : SpinLockRelease(&pgss->mutex);
2267 :
2268 54870 : return true;
2269 :
2270 0 : error:
2271 0 : ereport(LOG,
2272 : (errcode_for_file_access(),
2273 : errmsg("could not write file \"%s\": %m",
2274 : PGSS_TEXT_FILE)));
2275 :
2276 0 : if (fd >= 0)
2277 0 : CloseTransientFile(fd);
2278 :
2279 : /* Mark our write complete */
2280 0 : SpinLockAcquire(&pgss->mutex);
2281 0 : pgss->n_writers--;
2282 0 : SpinLockRelease(&pgss->mutex);
2283 :
2284 0 : return false;
2285 : }
2286 :
2287 : /*
2288 : * Read the external query text file into a malloc'd buffer.
2289 : *
2290 : * Returns NULL (without throwing an error) if unable to read, eg
2291 : * file not there or insufficient memory.
2292 : *
2293 : * On success, the buffer size is also returned into *buffer_size.
2294 : *
2295 : * This can be called without any lock on pgss->lock, but in that case
2296 : * the caller is responsible for verifying that the result is sane.
2297 : */
2298 : static char *
2299 196 : qtext_load_file(Size *buffer_size)
2300 : {
2301 : char *buf;
2302 : int fd;
2303 : struct stat stat;
2304 : Size nread;
2305 :
2306 196 : fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDONLY | PG_BINARY);
2307 196 : if (fd < 0)
2308 : {
2309 0 : if (errno != ENOENT)
2310 0 : ereport(LOG,
2311 : (errcode_for_file_access(),
2312 : errmsg("could not read file \"%s\": %m",
2313 : PGSS_TEXT_FILE)));
2314 0 : return NULL;
2315 : }
2316 :
2317 : /* Get file length */
2318 196 : if (fstat(fd, &stat))
2319 : {
2320 0 : ereport(LOG,
2321 : (errcode_for_file_access(),
2322 : errmsg("could not stat file \"%s\": %m",
2323 : PGSS_TEXT_FILE)));
2324 0 : CloseTransientFile(fd);
2325 0 : return NULL;
2326 : }
2327 :
2328 : /* Allocate buffer; beware that off_t might be wider than size_t */
2329 196 : if (stat.st_size <= MaxAllocHugeSize)
2330 196 : buf = (char *) malloc(stat.st_size);
2331 : else
2332 0 : buf = NULL;
2333 196 : if (buf == NULL)
2334 : {
2335 0 : ereport(LOG,
2336 : (errcode(ERRCODE_OUT_OF_MEMORY),
2337 : errmsg("out of memory"),
2338 : errdetail("Could not allocate enough memory to read file \"%s\".",
2339 : PGSS_TEXT_FILE)));
2340 0 : CloseTransientFile(fd);
2341 0 : return NULL;
2342 : }
2343 :
2344 : /*
2345 : * OK, slurp in the file. Windows fails if we try to read more than
2346 : * INT_MAX bytes at once, and other platforms might not like that either,
2347 : * so read a very large file in 1GB segments.
2348 : */
2349 196 : nread = 0;
2350 390 : while (nread < stat.st_size)
2351 : {
2352 194 : int toread = Min(1024 * 1024 * 1024, stat.st_size - nread);
2353 :
2354 : /*
2355 : * If we get a short read and errno doesn't get set, the reason is
2356 : * probably that garbage collection truncated the file since we did
2357 : * the fstat(), so we don't log a complaint --- but we don't return
2358 : * the data, either, since it's most likely corrupt due to concurrent
2359 : * writes from garbage collection.
2360 : */
2361 194 : errno = 0;
2362 194 : if (read(fd, buf + nread, toread) != toread)
2363 : {
2364 0 : if (errno)
2365 0 : ereport(LOG,
2366 : (errcode_for_file_access(),
2367 : errmsg("could not read file \"%s\": %m",
2368 : PGSS_TEXT_FILE)));
2369 0 : free(buf);
2370 0 : CloseTransientFile(fd);
2371 0 : return NULL;
2372 : }
2373 194 : nread += toread;
2374 : }
2375 :
2376 196 : if (CloseTransientFile(fd) != 0)
2377 0 : ereport(LOG,
2378 : (errcode_for_file_access(),
2379 : errmsg("could not close file \"%s\": %m", PGSS_TEXT_FILE)));
2380 :
2381 196 : *buffer_size = nread;
2382 196 : return buf;
2383 : }
2384 :
2385 : /*
2386 : * Locate a query text in the file image previously read by qtext_load_file().
2387 : *
2388 : * We validate the given offset/length, and return NULL if bogus. Otherwise,
2389 : * the result points to a null-terminated string within the buffer.
2390 : */
2391 : static char *
2392 155546 : qtext_fetch(Size query_offset, int query_len,
2393 : char *buffer, Size buffer_size)
2394 : {
2395 : /* File read failed? */
2396 155546 : if (buffer == NULL)
2397 0 : return NULL;
2398 : /* Bogus offset/length? */
2399 155546 : if (query_len < 0 ||
2400 155546 : query_offset + query_len >= buffer_size)
2401 0 : return NULL;
2402 : /* As a further sanity check, make sure there's a trailing null */
2403 155546 : if (buffer[query_offset + query_len] != '\0')
2404 0 : return NULL;
2405 : /* Looks OK */
2406 155546 : return buffer + query_offset;
2407 : }
2408 :
2409 : /*
2410 : * Do we need to garbage-collect the external query text file?
2411 : *
2412 : * Caller should hold at least a shared lock on pgss->lock.
2413 : */
2414 : static bool
2415 54870 : need_gc_qtexts(void)
2416 : {
2417 : Size extent;
2418 :
2419 : /* Read shared extent pointer */
2420 54870 : SpinLockAcquire(&pgss->mutex);
2421 54870 : extent = pgss->extent;
2422 54870 : SpinLockRelease(&pgss->mutex);
2423 :
2424 : /*
2425 : * Don't proceed if file does not exceed 512 bytes per possible entry.
2426 : *
2427 : * Here and in the next test, 32-bit machines have overflow hazards if
2428 : * pgss_max and/or mean_query_len are large. Force the multiplications
2429 : * and comparisons to be done in uint64 arithmetic to forestall trouble.
2430 : */
2431 54870 : if ((uint64) extent < (uint64) 512 * pgss_max)
2432 54870 : return false;
2433 :
2434 : /*
2435 : * Don't proceed if file is less than about 50% bloat. Nothing can or
2436 : * should be done in the event of unusually large query texts accounting
2437 : * for file's large size. We go to the trouble of maintaining the mean
2438 : * query length in order to prevent garbage collection from thrashing
2439 : * uselessly.
2440 : */
2441 0 : if ((uint64) extent < (uint64) pgss->mean_query_len * pgss_max * 2)
2442 0 : return false;
2443 :
2444 0 : return true;
2445 : }
2446 :
2447 : /*
2448 : * Garbage-collect orphaned query texts in external file.
2449 : *
2450 : * This won't be called often in the typical case, since it's likely that
2451 : * there won't be too much churn, and besides, a similar compaction process
2452 : * occurs when serializing to disk at shutdown or as part of resetting.
2453 : * Despite this, it seems prudent to plan for the edge case where the file
2454 : * becomes unreasonably large, with no other method of compaction likely to
2455 : * occur in the foreseeable future.
2456 : *
2457 : * The caller must hold an exclusive lock on pgss->lock.
2458 : *
2459 : * At the first sign of trouble we unlink the query text file to get a clean
2460 : * slate (although existing statistics are retained), rather than risk
2461 : * thrashing by allowing the same problem case to recur indefinitely.
2462 : */
2463 : static void
2464 0 : gc_qtexts(void)
2465 : {
2466 : char *qbuffer;
2467 : Size qbuffer_size;
2468 0 : FILE *qfile = NULL;
2469 : HASH_SEQ_STATUS hash_seq;
2470 : pgssEntry *entry;
2471 : Size extent;
2472 : int nentries;
2473 :
2474 : /*
2475 : * When called from pgss_store, some other session might have proceeded
2476 : * with garbage collection in the no-lock-held interim of lock strength
2477 : * escalation. Check once more that this is actually necessary.
2478 : */
2479 0 : if (!need_gc_qtexts())
2480 0 : return;
2481 :
2482 : /*
2483 : * Load the old texts file. If we fail (out of memory, for instance),
2484 : * invalidate query texts. Hopefully this is rare. It might seem better
2485 : * to leave things alone on an OOM failure, but the problem is that the
2486 : * file is only going to get bigger; hoping for a future non-OOM result is
2487 : * risky and can easily lead to complete denial of service.
2488 : */
2489 0 : qbuffer = qtext_load_file(&qbuffer_size);
2490 0 : if (qbuffer == NULL)
2491 0 : goto gc_fail;
2492 :
2493 : /*
2494 : * We overwrite the query texts file in place, so as to reduce the risk of
2495 : * an out-of-disk-space failure. Since the file is guaranteed not to get
2496 : * larger, this should always work on traditional filesystems; though we
2497 : * could still lose on copy-on-write filesystems.
2498 : */
2499 0 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
2500 0 : if (qfile == NULL)
2501 : {
2502 0 : ereport(LOG,
2503 : (errcode_for_file_access(),
2504 : errmsg("could not write file \"%s\": %m",
2505 : PGSS_TEXT_FILE)));
2506 0 : goto gc_fail;
2507 : }
2508 :
2509 0 : extent = 0;
2510 0 : nentries = 0;
2511 :
2512 0 : hash_seq_init(&hash_seq, pgss_hash);
2513 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2514 : {
2515 0 : int query_len = entry->query_len;
2516 0 : char *qry = qtext_fetch(entry->query_offset,
2517 : query_len,
2518 : qbuffer,
2519 : qbuffer_size);
2520 :
2521 0 : if (qry == NULL)
2522 : {
2523 : /* Trouble ... drop the text */
2524 0 : entry->query_offset = 0;
2525 0 : entry->query_len = -1;
2526 : /* entry will not be counted in mean query length computation */
2527 0 : continue;
2528 : }
2529 :
2530 0 : if (fwrite(qry, 1, query_len + 1, qfile) != query_len + 1)
2531 : {
2532 0 : ereport(LOG,
2533 : (errcode_for_file_access(),
2534 : errmsg("could not write file \"%s\": %m",
2535 : PGSS_TEXT_FILE)));
2536 0 : hash_seq_term(&hash_seq);
2537 0 : goto gc_fail;
2538 : }
2539 :
2540 0 : entry->query_offset = extent;
2541 0 : extent += query_len + 1;
2542 0 : nentries++;
2543 : }
2544 :
2545 : /*
2546 : * Truncate away any now-unused space. If this fails for some odd reason,
2547 : * we log it, but there's no need to fail.
2548 : */
2549 0 : if (ftruncate(fileno(qfile), extent) != 0)
2550 0 : ereport(LOG,
2551 : (errcode_for_file_access(),
2552 : errmsg("could not truncate file \"%s\": %m",
2553 : PGSS_TEXT_FILE)));
2554 :
2555 0 : if (FreeFile(qfile))
2556 : {
2557 0 : ereport(LOG,
2558 : (errcode_for_file_access(),
2559 : errmsg("could not write file \"%s\": %m",
2560 : PGSS_TEXT_FILE)));
2561 0 : qfile = NULL;
2562 0 : goto gc_fail;
2563 : }
2564 :
2565 0 : elog(DEBUG1, "pgss gc of queries file shrunk size from %zu to %zu",
2566 : pgss->extent, extent);
2567 :
2568 : /* Reset the shared extent pointer */
2569 0 : pgss->extent = extent;
2570 :
2571 : /*
2572 : * Also update the mean query length, to be sure that need_gc_qtexts()
2573 : * won't still think we have a problem.
2574 : */
2575 0 : if (nentries > 0)
2576 0 : pgss->mean_query_len = extent / nentries;
2577 : else
2578 0 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
2579 :
2580 0 : free(qbuffer);
2581 :
2582 : /*
2583 : * OK, count a garbage collection cycle. (Note: even though we have
2584 : * exclusive lock on pgss->lock, we must take pgss->mutex for this, since
2585 : * other processes may examine gc_count while holding only the mutex.
2586 : * Also, we have to advance the count *after* we've rewritten the file,
2587 : * else other processes might not realize they read a stale file.)
2588 : */
2589 0 : record_gc_qtexts();
2590 :
2591 0 : return;
2592 :
2593 0 : gc_fail:
2594 : /* clean up resources */
2595 0 : if (qfile)
2596 0 : FreeFile(qfile);
2597 0 : free(qbuffer);
2598 :
2599 : /*
2600 : * Since the contents of the external file are now uncertain, mark all
2601 : * hashtable entries as having invalid texts.
2602 : */
2603 0 : hash_seq_init(&hash_seq, pgss_hash);
2604 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2605 : {
2606 0 : entry->query_offset = 0;
2607 0 : entry->query_len = -1;
2608 : }
2609 :
2610 : /*
2611 : * Destroy the query text file and create a new, empty one
2612 : */
2613 0 : (void) unlink(PGSS_TEXT_FILE);
2614 0 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
2615 0 : if (qfile == NULL)
2616 0 : ereport(LOG,
2617 : (errcode_for_file_access(),
2618 : errmsg("could not recreate file \"%s\": %m",
2619 : PGSS_TEXT_FILE)));
2620 : else
2621 0 : FreeFile(qfile);
2622 :
2623 : /* Reset the shared extent pointer */
2624 0 : pgss->extent = 0;
2625 :
2626 : /* Reset mean_query_len to match the new state */
2627 0 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
2628 :
2629 : /*
2630 : * Bump the GC count even though we failed.
2631 : *
2632 : * This is needed to make concurrent readers of file without any lock on
2633 : * pgss->lock notice existence of new version of file. Once readers
2634 : * subsequently observe a change in GC count with pgss->lock held, that
2635 : * forces a safe reopen of file. Writers also require that we bump here,
2636 : * of course. (As required by locking protocol, readers and writers don't
2637 : * trust earlier file contents until gc_count is found unchanged after
2638 : * pgss->lock acquired in shared or exclusive mode respectively.)
2639 : */
2640 0 : record_gc_qtexts();
2641 : }
2642 :
2643 : #define SINGLE_ENTRY_RESET(e) \
2644 : if (e) { \
2645 : if (minmax_only) { \
2646 : /* When requested reset only min/max statistics of an entry */ \
2647 : for (int kind = 0; kind < PGSS_NUMKIND; kind++) \
2648 : { \
2649 : e->counters.max_time[kind] = 0; \
2650 : e->counters.min_time[kind] = 0; \
2651 : } \
2652 : e->minmax_stats_since = stats_reset; \
2653 : } \
2654 : else \
2655 : { \
2656 : /* Remove the key otherwise */ \
2657 : hash_search(pgss_hash, &e->key, HASH_REMOVE, NULL); \
2658 : num_remove++; \
2659 : } \
2660 : }
2661 :
2662 : /*
2663 : * Reset entries corresponding to parameters passed.
2664 : */
2665 : static TimestampTz
2666 160 : entry_reset(Oid userid, Oid dbid, uint64 queryid, bool minmax_only)
2667 : {
2668 : HASH_SEQ_STATUS hash_seq;
2669 : pgssEntry *entry;
2670 : FILE *qfile;
2671 : long num_entries;
2672 160 : long num_remove = 0;
2673 : pgssHashKey key;
2674 : TimestampTz stats_reset;
2675 :
2676 160 : if (!pgss || !pgss_hash)
2677 0 : ereport(ERROR,
2678 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2679 : errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
2680 :
2681 160 : LWLockAcquire(pgss->lock, LW_EXCLUSIVE);
2682 160 : num_entries = hash_get_num_entries(pgss_hash);
2683 :
2684 160 : stats_reset = GetCurrentTimestamp();
2685 :
2686 160 : if (userid != 0 && dbid != 0 && queryid != UINT64CONST(0))
2687 : {
2688 : /* If all the parameters are available, use the fast path. */
2689 2 : memset(&key, 0, sizeof(pgssHashKey));
2690 2 : key.userid = userid;
2691 2 : key.dbid = dbid;
2692 2 : key.queryid = queryid;
2693 :
2694 : /*
2695 : * Reset the entry if it exists, starting with the non-top-level
2696 : * entry.
2697 : */
2698 2 : key.toplevel = false;
2699 2 : entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
2700 :
2701 2 : SINGLE_ENTRY_RESET(entry);
2702 :
2703 : /* Also reset the top-level entry if it exists. */
2704 2 : key.toplevel = true;
2705 2 : entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
2706 :
2707 2 : SINGLE_ENTRY_RESET(entry);
2708 : }
2709 158 : else if (userid != 0 || dbid != 0 || queryid != UINT64CONST(0))
2710 : {
2711 : /* Reset entries corresponding to valid parameters. */
2712 8 : hash_seq_init(&hash_seq, pgss_hash);
2713 102 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2714 : {
2715 94 : if ((!userid || entry->key.userid == userid) &&
2716 72 : (!dbid || entry->key.dbid == dbid) &&
2717 68 : (!queryid || entry->key.queryid == queryid))
2718 : {
2719 14 : SINGLE_ENTRY_RESET(entry);
2720 : }
2721 : }
2722 : }
2723 : else
2724 : {
2725 : /* Reset all entries. */
2726 150 : hash_seq_init(&hash_seq, pgss_hash);
2727 1490 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2728 : {
2729 1384 : SINGLE_ENTRY_RESET(entry);
2730 : }
2731 : }
2732 :
2733 : /* All entries are removed? */
2734 160 : if (num_entries != num_remove)
2735 12 : goto release_lock;
2736 :
2737 : /*
2738 : * Reset global statistics for pg_stat_statements since all entries are
2739 : * removed.
2740 : */
2741 148 : SpinLockAcquire(&pgss->mutex);
2742 148 : pgss->stats.dealloc = 0;
2743 148 : pgss->stats.stats_reset = stats_reset;
2744 148 : SpinLockRelease(&pgss->mutex);
2745 :
2746 : /*
2747 : * Write new empty query file, perhaps even creating a new one to recover
2748 : * if the file was missing.
2749 : */
2750 148 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
2751 148 : if (qfile == NULL)
2752 : {
2753 0 : ereport(LOG,
2754 : (errcode_for_file_access(),
2755 : errmsg("could not create file \"%s\": %m",
2756 : PGSS_TEXT_FILE)));
2757 0 : goto done;
2758 : }
2759 :
2760 : /* If ftruncate fails, log it, but it's not a fatal problem */
2761 148 : if (ftruncate(fileno(qfile), 0) != 0)
2762 0 : ereport(LOG,
2763 : (errcode_for_file_access(),
2764 : errmsg("could not truncate file \"%s\": %m",
2765 : PGSS_TEXT_FILE)));
2766 :
2767 148 : FreeFile(qfile);
2768 :
2769 148 : done:
2770 148 : pgss->extent = 0;
2771 : /* This counts as a query text garbage collection for our purposes */
2772 148 : record_gc_qtexts();
2773 :
2774 160 : release_lock:
2775 160 : LWLockRelease(pgss->lock);
2776 :
2777 160 : return stats_reset;
2778 : }
2779 :
2780 : /*
2781 : * Generate a normalized version of the query string that will be used to
2782 : * represent all similar queries.
2783 : *
2784 : * Note that the normalized representation may well vary depending on
2785 : * just which "equivalent" query is used to create the hashtable entry.
2786 : * We assume this is OK.
2787 : *
2788 : * If query_loc > 0, then "query" has been advanced by that much compared to
2789 : * the original string start, so we need to translate the provided locations
2790 : * to compensate. (This lets us avoid re-scanning statements before the one
2791 : * of interest, so it's worth doing.)
2792 : *
2793 : * *query_len_p contains the input string length, and is updated with
2794 : * the result string length on exit. The resulting string might be longer
2795 : * or shorter depending on what happens with replacement of constants.
2796 : *
2797 : * Returns a palloc'd string.
2798 : */
2799 : static char *
2800 20748 : generate_normalized_query(JumbleState *jstate, const char *query,
2801 : int query_loc, int *query_len_p)
2802 : {
2803 : char *norm_query;
2804 20748 : int query_len = *query_len_p;
2805 : int i,
2806 : norm_query_buflen, /* Space allowed for norm_query */
2807 : len_to_wrt, /* Length (in bytes) to write */
2808 20748 : quer_loc = 0, /* Source query byte location */
2809 20748 : n_quer_loc = 0, /* Normalized query byte location */
2810 20748 : last_off = 0, /* Offset from start for previous tok */
2811 20748 : last_tok_len = 0; /* Length (in bytes) of that tok */
2812 :
2813 : /*
2814 : * Get constants' lengths (core system only gives us locations). Note
2815 : * this also ensures the items are sorted by location.
2816 : */
2817 20748 : fill_in_constant_lengths(jstate, query, query_loc);
2818 :
2819 : /*
2820 : * Allow for $n symbols to be longer than the constants they replace.
2821 : * Constants must take at least one byte in text form, while a $n symbol
2822 : * certainly isn't more than 11 bytes, even if n reaches INT_MAX. We
2823 : * could refine that limit based on the max value of n for the current
2824 : * query, but it hardly seems worth any extra effort to do so.
2825 : */
2826 20748 : norm_query_buflen = query_len + jstate->clocations_count * 10;
2827 :
2828 : /* Allocate result buffer */
2829 20748 : norm_query = palloc(norm_query_buflen + 1);
2830 :
2831 85240 : for (i = 0; i < jstate->clocations_count; i++)
2832 : {
2833 : int off, /* Offset from start for cur tok */
2834 : tok_len; /* Length (in bytes) of that tok */
2835 :
2836 64492 : off = jstate->clocations[i].location;
2837 : /* Adjust recorded location if we're dealing with partial string */
2838 64492 : off -= query_loc;
2839 :
2840 64492 : tok_len = jstate->clocations[i].length;
2841 :
2842 64492 : if (tok_len < 0)
2843 466 : continue; /* ignore any duplicates */
2844 :
2845 : /* Copy next chunk (what precedes the next constant) */
2846 64026 : len_to_wrt = off - last_off;
2847 64026 : len_to_wrt -= last_tok_len;
2848 :
2849 : Assert(len_to_wrt >= 0);
2850 64026 : memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
2851 64026 : n_quer_loc += len_to_wrt;
2852 :
2853 : /* And insert a param symbol in place of the constant token */
2854 128052 : n_quer_loc += sprintf(norm_query + n_quer_loc, "$%d",
2855 64026 : i + 1 + jstate->highest_extern_param_id);
2856 :
2857 64026 : quer_loc = off + tok_len;
2858 64026 : last_off = off;
2859 64026 : last_tok_len = tok_len;
2860 : }
2861 :
2862 : /*
2863 : * We've copied up until the last ignorable constant. Copy over the
2864 : * remaining bytes of the original query string.
2865 : */
2866 20748 : len_to_wrt = query_len - quer_loc;
2867 :
2868 : Assert(len_to_wrt >= 0);
2869 20748 : memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
2870 20748 : n_quer_loc += len_to_wrt;
2871 :
2872 : Assert(n_quer_loc <= norm_query_buflen);
2873 20748 : norm_query[n_quer_loc] = '\0';
2874 :
2875 20748 : *query_len_p = n_quer_loc;
2876 20748 : return norm_query;
2877 : }
2878 :
2879 : /*
2880 : * Given a valid SQL string and an array of constant-location records,
2881 : * fill in the textual lengths of those constants.
2882 : *
2883 : * The constants may use any allowed constant syntax, such as float literals,
2884 : * bit-strings, single-quoted strings and dollar-quoted strings. This is
2885 : * accomplished by using the public API for the core scanner.
2886 : *
2887 : * It is the caller's job to ensure that the string is a valid SQL statement
2888 : * with constants at the indicated locations. Since in practice the string
2889 : * has already been parsed, and the locations that the caller provides will
2890 : * have originated from within the authoritative parser, this should not be
2891 : * a problem.
2892 : *
2893 : * Duplicate constant pointers are possible, and will have their lengths
2894 : * marked as '-1', so that they are later ignored. (Actually, we assume the
2895 : * lengths were initialized as -1 to start with, and don't change them here.)
2896 : *
2897 : * If query_loc > 0, then "query" has been advanced by that much compared to
2898 : * the original string start, so we need to translate the provided locations
2899 : * to compensate. (This lets us avoid re-scanning statements before the one
2900 : * of interest, so it's worth doing.)
2901 : *
2902 : * N.B. There is an assumption that a '-' character at a Const location begins
2903 : * a negative numeric constant. This precludes there ever being another
2904 : * reason for a constant to start with a '-'.
2905 : */
2906 : static void
2907 20748 : fill_in_constant_lengths(JumbleState *jstate, const char *query,
2908 : int query_loc)
2909 : {
2910 : LocationLen *locs;
2911 : core_yyscan_t yyscanner;
2912 : core_yy_extra_type yyextra;
2913 : core_YYSTYPE yylval;
2914 : YYLTYPE yylloc;
2915 20748 : int last_loc = -1;
2916 : int i;
2917 :
2918 : /*
2919 : * Sort the records by location so that we can process them in order while
2920 : * scanning the query text.
2921 : */
2922 20748 : if (jstate->clocations_count > 1)
2923 13500 : qsort(jstate->clocations, jstate->clocations_count,
2924 : sizeof(LocationLen), comp_location);
2925 20748 : locs = jstate->clocations;
2926 :
2927 : /* initialize the flex scanner --- should match raw_parser() */
2928 20748 : yyscanner = scanner_init(query,
2929 : &yyextra,
2930 : &ScanKeywords,
2931 : ScanKeywordTokens);
2932 :
2933 : /* we don't want to re-emit any escape string warnings */
2934 20748 : yyextra.escape_string_warning = false;
2935 :
2936 : /* Search for each constant, in sequence */
2937 85240 : for (i = 0; i < jstate->clocations_count; i++)
2938 : {
2939 64492 : int loc = locs[i].location;
2940 : int tok;
2941 :
2942 : /* Adjust recorded location if we're dealing with partial string */
2943 64492 : loc -= query_loc;
2944 :
2945 : Assert(loc >= 0);
2946 :
2947 64492 : if (loc <= last_loc)
2948 466 : continue; /* Duplicate constant, ignore */
2949 :
2950 : /* Lex tokens until we find the desired constant */
2951 : for (;;)
2952 : {
2953 474512 : tok = core_yylex(&yylval, &yylloc, yyscanner);
2954 :
2955 : /* We should not hit end-of-string, but if we do, behave sanely */
2956 474512 : if (tok == 0)
2957 0 : break; /* out of inner for-loop */
2958 :
2959 : /*
2960 : * We should find the token position exactly, but if we somehow
2961 : * run past it, work with that.
2962 : */
2963 474512 : if (yylloc >= loc)
2964 : {
2965 64026 : if (query[loc] == '-')
2966 : {
2967 : /*
2968 : * It's a negative value - this is the one and only case
2969 : * where we replace more than a single token.
2970 : *
2971 : * Do not compensate for the core system's special-case
2972 : * adjustment of location to that of the leading '-'
2973 : * operator in the event of a negative constant. It is
2974 : * also useful for our purposes to start from the minus
2975 : * symbol. In this way, queries like "select * from foo
2976 : * where bar = 1" and "select * from foo where bar = -2"
2977 : * will have identical normalized query strings.
2978 : */
2979 816 : tok = core_yylex(&yylval, &yylloc, yyscanner);
2980 816 : if (tok == 0)
2981 0 : break; /* out of inner for-loop */
2982 : }
2983 :
2984 : /*
2985 : * We now rely on the assumption that flex has placed a zero
2986 : * byte after the text of the current token in scanbuf.
2987 : */
2988 64026 : locs[i].length = strlen(yyextra.scanbuf + loc);
2989 64026 : break; /* out of inner for-loop */
2990 : }
2991 : }
2992 :
2993 : /* If we hit end-of-string, give up, leaving remaining lengths -1 */
2994 64026 : if (tok == 0)
2995 0 : break;
2996 :
2997 64026 : last_loc = loc;
2998 : }
2999 :
3000 20748 : scanner_finish(yyscanner);
3001 20748 : }
3002 :
3003 : /*
3004 : * comp_location: comparator for qsorting LocationLen structs by location
3005 : */
3006 : static int
3007 75416 : comp_location(const void *a, const void *b)
3008 : {
3009 75416 : int l = ((const LocationLen *) a)->location;
3010 75416 : int r = ((const LocationLen *) b)->location;
3011 :
3012 75416 : return pg_cmp_s32(l, r);
3013 : }
|
