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) 2008-2026, 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/htup_details.h"
51 : #include "access/parallel.h"
52 : #include "catalog/pg_authid.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 "pgstat.h"
62 : #include "storage/fd.h"
63 : #include "storage/ipc.h"
64 : #include "storage/lwlock.h"
65 : #include "storage/shmem.h"
66 : #include "storage/spin.h"
67 : #include "tcop/utility.h"
68 : #include "utils/acl.h"
69 : #include "utils/builtins.h"
70 : #include "utils/memutils.h"
71 : #include "utils/timestamp.h"
72 : #include "utils/tuplestore.h"
73 :
74 10 : PG_MODULE_MAGIC_EXT(
75 : .name = "pg_stat_statements",
76 : .version = PG_VERSION
77 : );
78 :
79 : /* Location of permanent stats file (valid when database is shut down) */
80 : #define PGSS_DUMP_FILE PGSTAT_STAT_PERMANENT_DIRECTORY "/pg_stat_statements.stat"
81 :
82 : /*
83 : * Location of external query text file.
84 : */
85 : #define PGSS_TEXT_FILE PG_STAT_TMP_DIR "/pgss_query_texts.stat"
86 :
87 : /* Magic number identifying the stats file format */
88 : static const uint32 PGSS_FILE_HEADER = 0x20250731;
89 :
90 : /* PostgreSQL major version number, changes in which invalidate all entries */
91 : static const uint32 PGSS_PG_MAJOR_VERSION = PG_VERSION_NUM / 100;
92 :
93 : /* XXX: Should USAGE_EXEC reflect execution time and/or buffer usage? */
94 : #define USAGE_EXEC(duration) (1.0)
95 : #define USAGE_INIT (1.0) /* including initial planning */
96 : #define ASSUMED_MEDIAN_INIT (10.0) /* initial assumed median usage */
97 : #define ASSUMED_LENGTH_INIT 1024 /* initial assumed mean query length */
98 : #define USAGE_DECREASE_FACTOR (0.99) /* decreased every entry_dealloc */
99 : #define STICKY_DECREASE_FACTOR (0.50) /* factor for sticky entries */
100 : #define USAGE_DEALLOC_PERCENT 5 /* free this % of entries at once */
101 : #define IS_STICKY(c) ((c.calls[PGSS_PLAN] + c.calls[PGSS_EXEC]) == 0)
102 :
103 : /*
104 : * Extension version number, for supporting older extension versions' objects
105 : */
106 : typedef enum pgssVersion
107 : {
108 : PGSS_V1_0 = 0,
109 : PGSS_V1_1,
110 : PGSS_V1_2,
111 : PGSS_V1_3,
112 : PGSS_V1_8,
113 : PGSS_V1_9,
114 : PGSS_V1_10,
115 : PGSS_V1_11,
116 : PGSS_V1_12,
117 : PGSS_V1_13,
118 : } pgssVersion;
119 :
120 : typedef enum pgssStoreKind
121 : {
122 : PGSS_INVALID = -1,
123 :
124 : /*
125 : * PGSS_PLAN and PGSS_EXEC must be respectively 0 and 1 as they're used to
126 : * reference the underlying values in the arrays in the Counters struct,
127 : * and this order is required in pg_stat_statements_internal().
128 : */
129 : PGSS_PLAN = 0,
130 : PGSS_EXEC,
131 : } pgssStoreKind;
132 :
133 : #define PGSS_NUMKIND (PGSS_EXEC + 1)
134 :
135 : /*
136 : * Hashtable key that defines the identity of a hashtable entry. We separate
137 : * queries by user and by database even if they are otherwise identical.
138 : *
139 : * If you add a new key to this struct, make sure to teach pgss_store() to
140 : * zero the padding bytes. Otherwise, things will break, because pgss_hash is
141 : * created using HASH_BLOBS, and thus tag_hash is used to hash this.
142 : */
143 : typedef struct pgssHashKey
144 : {
145 : Oid userid; /* user OID */
146 : Oid dbid; /* database OID */
147 : int64 queryid; /* query identifier */
148 : bool toplevel; /* query executed at top level */
149 : } pgssHashKey;
150 :
151 : /*
152 : * The actual stats counters kept within pgssEntry.
153 : */
154 : typedef struct Counters
155 : {
156 : int64 calls[PGSS_NUMKIND]; /* # of times planned/executed */
157 : double total_time[PGSS_NUMKIND]; /* total planning/execution time,
158 : * in msec */
159 : double min_time[PGSS_NUMKIND]; /* minimum planning/execution time in
160 : * msec since min/max reset */
161 : double max_time[PGSS_NUMKIND]; /* maximum planning/execution time in
162 : * msec since min/max reset */
163 : double mean_time[PGSS_NUMKIND]; /* mean planning/execution time in
164 : * msec */
165 : double sum_var_time[PGSS_NUMKIND]; /* sum of variances in
166 : * planning/execution time in msec */
167 : int64 rows; /* total # of retrieved or affected rows */
168 : int64 shared_blks_hit; /* # of shared buffer hits */
169 : int64 shared_blks_read; /* # of shared disk blocks read */
170 : int64 shared_blks_dirtied; /* # of shared disk blocks dirtied */
171 : int64 shared_blks_written; /* # of shared disk blocks written */
172 : int64 local_blks_hit; /* # of local buffer hits */
173 : int64 local_blks_read; /* # of local disk blocks read */
174 : int64 local_blks_dirtied; /* # of local disk blocks dirtied */
175 : int64 local_blks_written; /* # of local disk blocks written */
176 : int64 temp_blks_read; /* # of temp blocks read */
177 : int64 temp_blks_written; /* # of temp blocks written */
178 : double shared_blk_read_time; /* time spent reading shared blocks,
179 : * in msec */
180 : double shared_blk_write_time; /* time spent writing shared blocks,
181 : * in msec */
182 : double local_blk_read_time; /* time spent reading local blocks, in
183 : * msec */
184 : double local_blk_write_time; /* time spent writing local blocks, in
185 : * msec */
186 : double temp_blk_read_time; /* time spent reading temp blocks, in msec */
187 : double temp_blk_write_time; /* time spent writing temp blocks, in
188 : * msec */
189 : double usage; /* usage factor */
190 : int64 wal_records; /* # of WAL records generated */
191 : int64 wal_fpi; /* # of WAL full page images generated */
192 : uint64 wal_bytes; /* total amount of WAL generated in bytes */
193 : int64 wal_buffers_full; /* # of times the WAL buffers became full */
194 : int64 jit_functions; /* total number of JIT functions emitted */
195 : double jit_generation_time; /* total time to generate jit code */
196 : int64 jit_inlining_count; /* number of times inlining time has been
197 : * > 0 */
198 : double jit_deform_time; /* total time to deform tuples in jit code */
199 : int64 jit_deform_count; /* number of times deform time has been >
200 : * 0 */
201 :
202 : double jit_inlining_time; /* total time to inline jit code */
203 : int64 jit_optimization_count; /* number of times optimization time
204 : * has been > 0 */
205 : double jit_optimization_time; /* total time to optimize jit code */
206 : int64 jit_emission_count; /* number of times emission time has been
207 : * > 0 */
208 : double jit_emission_time; /* total time to emit jit code */
209 : int64 parallel_workers_to_launch; /* # of parallel workers planned
210 : * to be launched */
211 : int64 parallel_workers_launched; /* # of parallel workers actually
212 : * launched */
213 : int64 generic_plan_calls; /* number of calls using a generic plan */
214 : int64 custom_plan_calls; /* number of calls using a custom plan */
215 : } Counters;
216 :
217 : /*
218 : * Global statistics for pg_stat_statements
219 : */
220 : typedef struct pgssGlobalStats
221 : {
222 : int64 dealloc; /* # of times entries were deallocated */
223 : TimestampTz stats_reset; /* timestamp with all stats reset */
224 : } pgssGlobalStats;
225 :
226 : /*
227 : * Statistics per statement
228 : *
229 : * Note: in event of a failure in garbage collection of the query text file,
230 : * we reset query_offset to zero and query_len to -1. This will be seen as
231 : * an invalid state by qtext_fetch().
232 : */
233 : typedef struct pgssEntry
234 : {
235 : pgssHashKey key; /* hash key of entry - MUST BE FIRST */
236 : Counters counters; /* the statistics for this query */
237 : Size query_offset; /* query text offset in external file */
238 : int query_len; /* # of valid bytes in query string, or -1 */
239 : int encoding; /* query text encoding */
240 : TimestampTz stats_since; /* timestamp of entry allocation */
241 : TimestampTz minmax_stats_since; /* timestamp of last min/max values reset */
242 : slock_t mutex; /* protects the counters only */
243 : } pgssEntry;
244 :
245 : /*
246 : * Global shared state
247 : */
248 : typedef struct pgssSharedState
249 : {
250 : LWLockPadded lock; /* protects hashtable search/modification */
251 : double cur_median_usage; /* current median usage in hashtable */
252 : Size mean_query_len; /* current mean entry text length */
253 : slock_t mutex; /* protects following fields only: */
254 : Size extent; /* current extent of query file */
255 : int n_writers; /* number of active writers to query file */
256 : int gc_count; /* query file garbage collection cycle count */
257 : pgssGlobalStats stats; /* global statistics for pgss */
258 : } pgssSharedState;
259 :
260 : /* Links to shared memory state */
261 : static pgssSharedState *pgss;
262 : static HTAB *pgss_hash;
263 :
264 : static void pgss_shmem_request(void *arg);
265 : static void pgss_shmem_init(void *arg);
266 :
267 : static const ShmemCallbacks pgss_shmem_callbacks = {
268 : .request_fn = pgss_shmem_request,
269 : .init_fn = pgss_shmem_init,
270 : };
271 :
272 : /*---- Local variables ----*/
273 :
274 : /* Current nesting depth of planner/ExecutorRun/ProcessUtility calls */
275 : static int nesting_level = 0;
276 :
277 : /* Saved hook values */
278 : static post_parse_analyze_hook_type prev_post_parse_analyze_hook = NULL;
279 : static planner_hook_type prev_planner_hook = NULL;
280 : static ExecutorStart_hook_type prev_ExecutorStart = NULL;
281 : static ExecutorRun_hook_type prev_ExecutorRun = NULL;
282 : static ExecutorFinish_hook_type prev_ExecutorFinish = NULL;
283 : static ExecutorEnd_hook_type prev_ExecutorEnd = NULL;
284 : static ProcessUtility_hook_type prev_ProcessUtility = NULL;
285 :
286 : /*---- GUC variables ----*/
287 :
288 : typedef enum
289 : {
290 : PGSS_TRACK_NONE, /* track no statements */
291 : PGSS_TRACK_TOP, /* only top level statements */
292 : PGSS_TRACK_ALL, /* all statements, including nested ones */
293 : } PGSSTrackLevel;
294 :
295 : static const struct config_enum_entry track_options[] =
296 : {
297 : {"none", PGSS_TRACK_NONE, false},
298 : {"top", PGSS_TRACK_TOP, false},
299 : {"all", PGSS_TRACK_ALL, false},
300 : {NULL, 0, false}
301 : };
302 :
303 : static int pgss_max = 5000; /* max # statements to track */
304 : static int pgss_track = PGSS_TRACK_TOP; /* tracking level */
305 : static bool pgss_track_utility = true; /* whether to track utility commands */
306 : static bool pgss_track_planning = false; /* whether to track planning
307 : * duration */
308 : static bool pgss_save = true; /* whether to save stats across shutdown */
309 :
310 : #define pgss_enabled(level) \
311 : (!IsParallelWorker() && \
312 : (pgss_track == PGSS_TRACK_ALL || \
313 : (pgss_track == PGSS_TRACK_TOP && (level) == 0)))
314 :
315 : #define record_gc_qtexts() \
316 : do { \
317 : SpinLockAcquire(&pgss->mutex); \
318 : pgss->gc_count++; \
319 : SpinLockRelease(&pgss->mutex); \
320 : } while(0)
321 :
322 : /*---- Function declarations ----*/
323 :
324 6 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset);
325 6 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_7);
326 20 : PG_FUNCTION_INFO_V1(pg_stat_statements_reset_1_11);
327 0 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_2);
328 6 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_3);
329 6 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_8);
330 6 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_9);
331 6 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_10);
332 6 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_11);
333 6 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_12);
334 24 : PG_FUNCTION_INFO_V1(pg_stat_statements_1_13);
335 0 : PG_FUNCTION_INFO_V1(pg_stat_statements);
336 8 : PG_FUNCTION_INFO_V1(pg_stat_statements_info);
337 :
338 : static void pgss_shmem_shutdown(int code, Datum arg);
339 : static void pgss_post_parse_analyze(ParseState *pstate, Query *query,
340 : const JumbleState *jstate);
341 : static PlannedStmt *pgss_planner(Query *parse,
342 : const char *query_string,
343 : int cursorOptions,
344 : ParamListInfo boundParams,
345 : ExplainState *es);
346 : static void pgss_ExecutorStart(QueryDesc *queryDesc, int eflags);
347 : static void pgss_ExecutorRun(QueryDesc *queryDesc,
348 : ScanDirection direction,
349 : uint64 count);
350 : static void pgss_ExecutorFinish(QueryDesc *queryDesc);
351 : static void pgss_ExecutorEnd(QueryDesc *queryDesc);
352 : static void pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
353 : bool readOnlyTree,
354 : ProcessUtilityContext context, ParamListInfo params,
355 : QueryEnvironment *queryEnv,
356 : DestReceiver *dest, QueryCompletion *qc);
357 : static void pgss_store(const char *query, int64 queryId,
358 : int query_location, int query_len,
359 : pgssStoreKind kind,
360 : double total_time, uint64 rows,
361 : const BufferUsage *bufusage,
362 : const WalUsage *walusage,
363 : const struct JitInstrumentation *jitusage,
364 : const JumbleState *jstate,
365 : int parallel_workers_to_launch,
366 : int parallel_workers_launched,
367 : PlannedStmtOrigin planOrigin);
368 : static void pg_stat_statements_internal(FunctionCallInfo fcinfo,
369 : pgssVersion api_version,
370 : bool showtext);
371 : static pgssEntry *entry_alloc(pgssHashKey *key, Size query_offset, int query_len,
372 : int encoding, bool sticky);
373 : static void entry_dealloc(void);
374 : static bool qtext_store(const char *query, int query_len,
375 : Size *query_offset, int *gc_count);
376 : static char *qtext_load_file(Size *buffer_size);
377 : static char *qtext_fetch(Size query_offset, int query_len,
378 : char *buffer, Size buffer_size);
379 : static bool need_gc_qtexts(void);
380 : static void gc_qtexts(void);
381 : static TimestampTz entry_reset(Oid userid, Oid dbid, int64 queryid, bool minmax_only);
382 : static char *generate_normalized_query(const JumbleState *jstate,
383 : const char *query,
384 : int query_loc, int *query_len_p);
385 :
386 : /*
387 : * Module load callback
388 : */
389 : void
390 10 : _PG_init(void)
391 : {
392 : /*
393 : * In order to create our shared memory area, we have to be loaded via
394 : * shared_preload_libraries. If not, fall out without hooking into any of
395 : * the main system. (We don't throw error here because it seems useful to
396 : * allow the pg_stat_statements functions to be created even when the
397 : * module isn't active. The functions must protect themselves against
398 : * being called then, however.)
399 : */
400 10 : if (!process_shared_preload_libraries_in_progress)
401 1 : return;
402 :
403 : /*
404 : * Inform the postmaster that we want to enable query_id calculation if
405 : * compute_query_id is set to auto.
406 : */
407 9 : EnableQueryId();
408 :
409 : /*
410 : * Define (or redefine) custom GUC variables.
411 : */
412 9 : DefineCustomIntVariable("pg_stat_statements.max",
413 : "Sets the maximum number of statements tracked by pg_stat_statements.",
414 : NULL,
415 : &pgss_max,
416 : 5000,
417 : 100,
418 : INT_MAX / 2,
419 : PGC_POSTMASTER,
420 : 0,
421 : NULL,
422 : NULL,
423 : NULL);
424 :
425 9 : DefineCustomEnumVariable("pg_stat_statements.track",
426 : "Selects which statements are tracked by pg_stat_statements.",
427 : NULL,
428 : &pgss_track,
429 : PGSS_TRACK_TOP,
430 : track_options,
431 : PGC_SUSET,
432 : 0,
433 : NULL,
434 : NULL,
435 : NULL);
436 :
437 9 : DefineCustomBoolVariable("pg_stat_statements.track_utility",
438 : "Selects whether utility commands are tracked by pg_stat_statements.",
439 : NULL,
440 : &pgss_track_utility,
441 : true,
442 : PGC_SUSET,
443 : 0,
444 : NULL,
445 : NULL,
446 : NULL);
447 :
448 9 : DefineCustomBoolVariable("pg_stat_statements.track_planning",
449 : "Selects whether planning duration is tracked by pg_stat_statements.",
450 : NULL,
451 : &pgss_track_planning,
452 : false,
453 : PGC_SUSET,
454 : 0,
455 : NULL,
456 : NULL,
457 : NULL);
458 :
459 9 : DefineCustomBoolVariable("pg_stat_statements.save",
460 : "Save pg_stat_statements statistics across server shutdowns.",
461 : NULL,
462 : &pgss_save,
463 : true,
464 : PGC_SIGHUP,
465 : 0,
466 : NULL,
467 : NULL,
468 : NULL);
469 :
470 9 : MarkGUCPrefixReserved("pg_stat_statements");
471 :
472 : /*
473 : * Register our shared memory needs.
474 : */
475 9 : RegisterShmemCallbacks(&pgss_shmem_callbacks);
476 :
477 : /*
478 : * Install hooks.
479 : */
480 9 : prev_post_parse_analyze_hook = post_parse_analyze_hook;
481 9 : post_parse_analyze_hook = pgss_post_parse_analyze;
482 9 : prev_planner_hook = planner_hook;
483 9 : planner_hook = pgss_planner;
484 9 : prev_ExecutorStart = ExecutorStart_hook;
485 9 : ExecutorStart_hook = pgss_ExecutorStart;
486 9 : prev_ExecutorRun = ExecutorRun_hook;
487 9 : ExecutorRun_hook = pgss_ExecutorRun;
488 9 : prev_ExecutorFinish = ExecutorFinish_hook;
489 9 : ExecutorFinish_hook = pgss_ExecutorFinish;
490 9 : prev_ExecutorEnd = ExecutorEnd_hook;
491 9 : ExecutorEnd_hook = pgss_ExecutorEnd;
492 9 : prev_ProcessUtility = ProcessUtility_hook;
493 9 : ProcessUtility_hook = pgss_ProcessUtility;
494 : }
495 :
496 : /*
497 : * shmem request callback: Request shared memory resources.
498 : *
499 : * This is called at postmaster startup. Note that the shared memory isn't
500 : * allocated here yet, this merely register our needs.
501 : *
502 : * In EXEC_BACKEND mode, this is also called in each backend, to re-attach to
503 : * the shared memory area that was already initialized.
504 : */
505 : static void
506 11 : pgss_shmem_request(void *arg)
507 : {
508 11 : ShmemRequestHash(.name = "pg_stat_statements hash",
509 : .nelems = pgss_max,
510 : .hash_info.keysize = sizeof(pgssHashKey),
511 : .hash_info.entrysize = sizeof(pgssEntry),
512 : .hash_flags = HASH_ELEM | HASH_BLOBS,
513 : .ptr = &pgss_hash,
514 : );
515 11 : ShmemRequestStruct(.name = "pg_stat_statements",
516 : .size = sizeof(pgssSharedState),
517 : .ptr = (void **) &pgss,
518 : );
519 11 : }
520 :
521 : /*
522 : * shmem init callback: Initialize our shared memory data structures at
523 : * postmaster startup.
524 : *
525 : * Load any pre-existing statistics from file. Also create and load the
526 : * query-texts file, which is expected to exist (even if empty) while the
527 : * module is enabled.
528 : */
529 : static void
530 11 : pgss_shmem_init(void *arg)
531 : {
532 : int tranche_id;
533 11 : FILE *file = NULL;
534 11 : FILE *qfile = NULL;
535 : uint32 header;
536 : int32 num;
537 : int32 pgver;
538 : int32 i;
539 : int buffer_size;
540 11 : char *buffer = NULL;
541 :
542 : /*
543 : * We already checked that we're loaded from shared_preload_libraries in
544 : * _PG_init(), so we should not get here after postmaster startup.
545 : */
546 : Assert(!IsUnderPostmaster);
547 :
548 : /*
549 : * Initialize the shmem area with no statistics.
550 : */
551 11 : tranche_id = LWLockNewTrancheId("pg_stat_statements");
552 11 : LWLockInitialize(&pgss->lock.lock, tranche_id);
553 11 : pgss->cur_median_usage = ASSUMED_MEDIAN_INIT;
554 11 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
555 11 : SpinLockInit(&pgss->mutex);
556 11 : pgss->extent = 0;
557 11 : pgss->n_writers = 0;
558 11 : pgss->gc_count = 0;
559 11 : pgss->stats.dealloc = 0;
560 11 : pgss->stats.stats_reset = GetCurrentTimestamp();
561 :
562 : /* The hash table must've also been initialized by now */
563 : Assert(pgss_hash != NULL);
564 :
565 : /*
566 : * Set up a shmem exit hook to dump the statistics to disk on postmaster
567 : * (or standalone backend) exit.
568 : */
569 11 : on_shmem_exit(pgss_shmem_shutdown, (Datum) 0);
570 :
571 : /*
572 : * Load any pre-existing statistics from file.
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 11 : unlink(PGSS_TEXT_FILE);
580 :
581 : /* Allocate new query text temp file */
582 11 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
583 11 : 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 11 : if (!pgss_save)
592 : {
593 1 : FreeFile(qfile);
594 11 : return;
595 : }
596 :
597 : /*
598 : * Attempt to load old statistics from the dump file.
599 : */
600 10 : file = AllocateFile(PGSS_DUMP_FILE, PG_BINARY_R);
601 10 : if (file == NULL)
602 : {
603 7 : if (errno != ENOENT)
604 0 : goto read_error;
605 : /* No existing persisted stats file, so we're done */
606 7 : FreeFile(qfile);
607 7 : return;
608 : }
609 :
610 3 : buffer_size = 2048;
611 3 : buffer = (char *) palloc(buffer_size);
612 :
613 6 : if (fread(&header, sizeof(uint32), 1, file) != 1 ||
614 6 : fread(&pgver, sizeof(uint32), 1, file) != 1 ||
615 3 : fread(&num, sizeof(int32), 1, file) != 1)
616 0 : goto read_error;
617 :
618 3 : if (header != PGSS_FILE_HEADER ||
619 3 : pgver != PGSS_PG_MAJOR_VERSION)
620 0 : goto data_error;
621 :
622 28779 : for (i = 0; i < num; i++)
623 : {
624 : pgssEntry temp;
625 : pgssEntry *entry;
626 : Size query_offset;
627 :
628 28776 : 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 28776 : if (!PG_VALID_BE_ENCODING(temp.encoding))
633 0 : goto data_error;
634 :
635 : /* Resize buffer as needed */
636 28776 : if (temp.query_len >= buffer_size)
637 : {
638 4 : buffer_size = Max(buffer_size * 2, temp.query_len + 1);
639 4 : buffer = repalloc(buffer, buffer_size);
640 : }
641 :
642 28776 : 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 28776 : buffer[temp.query_len] = '\0';
647 :
648 : /* Skip loading "sticky" entries */
649 28776 : if (IS_STICKY(temp.counters))
650 784 : continue;
651 :
652 : /* Store the query text */
653 27992 : query_offset = pgss->extent;
654 27992 : if (fwrite(buffer, 1, temp.query_len + 1, qfile) != temp.query_len + 1)
655 0 : goto write_error;
656 27992 : pgss->extent += temp.query_len + 1;
657 :
658 : /* make the hashtable entry (discards old entries if too many) */
659 27992 : entry = entry_alloc(&temp.key, query_offset, temp.query_len,
660 : temp.encoding,
661 : false);
662 :
663 : /* copy in the actual stats */
664 27992 : entry->counters = temp.counters;
665 27992 : entry->stats_since = temp.stats_since;
666 27992 : entry->minmax_stats_since = temp.minmax_stats_since;
667 : }
668 :
669 : /* Read global statistics for pg_stat_statements */
670 3 : if (fread(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
671 0 : goto read_error;
672 :
673 3 : pfree(buffer);
674 3 : FreeFile(file);
675 3 : 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 3 : unlink(PGSS_DUMP_FILE);
691 :
692 3 : 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 11 : pgss_shmem_shutdown(int code, Datum arg)
735 : {
736 : FILE *file;
737 11 : char *qbuffer = NULL;
738 11 : 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 11 : if (code)
745 11 : return;
746 :
747 : /* Safety check ... shouldn't get here unless shmem is set up. */
748 9 : if (!pgss || !pgss_hash)
749 0 : return;
750 :
751 : /* Don't dump if told not to. */
752 9 : if (!pgss_save)
753 2 : return;
754 :
755 7 : file = AllocateFile(PGSS_DUMP_FILE ".tmp", PG_BINARY_W);
756 7 : if (file == NULL)
757 0 : goto error;
758 :
759 7 : if (fwrite(&PGSS_FILE_HEADER, sizeof(uint32), 1, file) != 1)
760 0 : goto error;
761 7 : if (fwrite(&PGSS_PG_MAJOR_VERSION, sizeof(uint32), 1, file) != 1)
762 0 : goto error;
763 7 : num_entries = hash_get_num_entries(pgss_hash);
764 7 : if (fwrite(&num_entries, sizeof(int32), 1, file) != 1)
765 0 : goto error;
766 :
767 7 : qbuffer = qtext_load_file(&qbuffer_size);
768 7 : 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 7 : hash_seq_init(&hash_seq, pgss_hash);
776 57852 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
777 : {
778 57845 : int len = entry->query_len;
779 57845 : char *qstr = qtext_fetch(entry->query_offset, len,
780 : qbuffer, qbuffer_size);
781 :
782 57845 : if (qstr == NULL)
783 0 : continue; /* Ignore any entries with bogus texts */
784 :
785 57845 : if (fwrite(entry, sizeof(pgssEntry), 1, file) != 1 ||
786 57845 : 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 7 : if (fwrite(&pgss->stats, sizeof(pgssGlobalStats), 1, file) != 1)
796 0 : goto error;
797 :
798 7 : pfree(qbuffer);
799 7 : qbuffer = NULL;
800 :
801 7 : 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 7 : (void) durable_rename(PGSS_DUMP_FILE ".tmp", PGSS_DUMP_FILE, LOG);
811 :
812 : /* Unlink query-texts file; it's not needed while shutdown */
813 7 : unlink(PGSS_TEXT_FILE);
814 :
815 7 : 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 : if (qbuffer)
823 0 : pfree(qbuffer);
824 0 : if (file)
825 0 : FreeFile(file);
826 0 : unlink(PGSS_DUMP_FILE ".tmp");
827 0 : unlink(PGSS_TEXT_FILE);
828 : }
829 :
830 : /*
831 : * Post-parse-analysis hook: mark query with a queryId
832 : */
833 : static void
834 83834 : pgss_post_parse_analyze(ParseState *pstate, Query *query, const JumbleState *jstate)
835 : {
836 83834 : if (prev_post_parse_analyze_hook)
837 0 : prev_post_parse_analyze_hook(pstate, query, jstate);
838 :
839 : /* Safety check... */
840 83834 : if (!pgss || !pgss_hash || !pgss_enabled(nesting_level))
841 12962 : return;
842 :
843 : /*
844 : * If it's EXECUTE, clear the queryId so that stats will accumulate for
845 : * the underlying PREPARE. But don't do this if we're not tracking
846 : * utility statements, to avoid messing up another extension that might be
847 : * tracking them.
848 : */
849 70872 : if (query->utilityStmt)
850 : {
851 31801 : if (pgss_track_utility && IsA(query->utilityStmt, ExecuteStmt))
852 : {
853 3374 : query->queryId = INT64CONST(0);
854 3374 : return;
855 : }
856 : }
857 :
858 : /*
859 : * If query jumbling were able to identify any ignorable constants, we
860 : * immediately create a hash table entry for the query, so that we can
861 : * record the normalized form of the query string. If there were no such
862 : * constants, the normalized string would be the same as the query text
863 : * anyway, so there's no need for an early entry.
864 : */
865 67498 : if (jstate && jstate->clocations_count > 0)
866 38915 : pgss_store(pstate->p_sourcetext,
867 : query->queryId,
868 : query->stmt_location,
869 : query->stmt_len,
870 : PGSS_INVALID,
871 : 0,
872 : 0,
873 : NULL,
874 : NULL,
875 : NULL,
876 : jstate,
877 : 0,
878 : 0,
879 : PLAN_STMT_UNKNOWN);
880 : }
881 :
882 : /*
883 : * Planner hook: forward to regular planner, but measure planning time
884 : * if needed.
885 : */
886 : static PlannedStmt *
887 50329 : pgss_planner(Query *parse,
888 : const char *query_string,
889 : int cursorOptions,
890 : ParamListInfo boundParams,
891 : ExplainState *es)
892 : {
893 : PlannedStmt *result;
894 :
895 : /*
896 : * We can't process the query if no query_string is provided, as
897 : * pgss_store needs it. We also ignore query without queryid, as it would
898 : * be treated as a utility statement, which may not be the case.
899 : */
900 50329 : if (pgss_enabled(nesting_level)
901 39223 : && pgss_track_planning && query_string
902 150 : && parse->queryId != INT64CONST(0))
903 150 : {
904 : instr_time start;
905 : instr_time duration;
906 : BufferUsage bufusage_start,
907 : bufusage;
908 : WalUsage walusage_start,
909 : walusage;
910 :
911 : /* We need to track buffer usage as the planner can access them. */
912 150 : bufusage_start = pgBufferUsage;
913 :
914 : /*
915 : * Similarly the planner could write some WAL records in some cases
916 : * (e.g. setting a hint bit with those being WAL-logged)
917 : */
918 150 : walusage_start = pgWalUsage;
919 150 : INSTR_TIME_SET_CURRENT(start);
920 :
921 150 : nesting_level++;
922 150 : PG_TRY();
923 : {
924 150 : if (prev_planner_hook)
925 0 : result = prev_planner_hook(parse, query_string, cursorOptions,
926 : boundParams, es);
927 : else
928 150 : result = standard_planner(parse, query_string, cursorOptions,
929 : boundParams, es);
930 : }
931 0 : PG_FINALLY();
932 : {
933 150 : nesting_level--;
934 : }
935 150 : PG_END_TRY();
936 :
937 150 : INSTR_TIME_SET_CURRENT(duration);
938 150 : INSTR_TIME_SUBTRACT(duration, start);
939 :
940 : /* calc differences of buffer counters. */
941 150 : memset(&bufusage, 0, sizeof(BufferUsage));
942 150 : BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
943 :
944 : /* calc differences of WAL counters. */
945 150 : memset(&walusage, 0, sizeof(WalUsage));
946 150 : WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
947 :
948 300 : pgss_store(query_string,
949 : parse->queryId,
950 : parse->stmt_location,
951 : parse->stmt_len,
952 : PGSS_PLAN,
953 150 : INSTR_TIME_GET_MILLISEC(duration),
954 : 0,
955 : &bufusage,
956 : &walusage,
957 : NULL,
958 : NULL,
959 : 0,
960 : 0,
961 : result->planOrigin);
962 : }
963 : else
964 : {
965 : /*
966 : * Even though we're not tracking plan time for this statement, we
967 : * must still increment the nesting level, to ensure that functions
968 : * evaluated during planning are not seen as top-level calls.
969 : */
970 50179 : nesting_level++;
971 50179 : PG_TRY();
972 : {
973 50179 : if (prev_planner_hook)
974 0 : result = prev_planner_hook(parse, query_string, cursorOptions,
975 : boundParams, es);
976 : else
977 50179 : result = standard_planner(parse, query_string, cursorOptions,
978 : boundParams, es);
979 : }
980 771 : PG_FINALLY();
981 : {
982 50179 : nesting_level--;
983 : }
984 50179 : PG_END_TRY();
985 : }
986 :
987 49558 : return result;
988 : }
989 :
990 : /*
991 : * ExecutorStart hook: start up tracking if needed
992 : */
993 : static void
994 60438 : pgss_ExecutorStart(QueryDesc *queryDesc, int eflags)
995 : {
996 : /*
997 : * If query has queryId zero, don't track it. This prevents double
998 : * counting of optimizable statements that are directly contained in
999 : * utility statements.
1000 : */
1001 60438 : if (pgss_enabled(nesting_level) && queryDesc->plannedstmt->queryId != INT64CONST(0))
1002 : {
1003 : /* Request all summary instrumentation, i.e. timing, buffers and WAL */
1004 41417 : queryDesc->query_instr_options |= INSTRUMENT_ALL;
1005 : }
1006 :
1007 60438 : if (prev_ExecutorStart)
1008 0 : prev_ExecutorStart(queryDesc, eflags);
1009 : else
1010 60438 : standard_ExecutorStart(queryDesc, eflags);
1011 60156 : }
1012 :
1013 : /*
1014 : * ExecutorRun hook: all we need do is track nesting depth
1015 : */
1016 : static void
1017 58821 : pgss_ExecutorRun(QueryDesc *queryDesc, ScanDirection direction, uint64 count)
1018 : {
1019 58821 : nesting_level++;
1020 58821 : PG_TRY();
1021 : {
1022 58821 : if (prev_ExecutorRun)
1023 0 : prev_ExecutorRun(queryDesc, direction, count);
1024 : else
1025 58821 : standard_ExecutorRun(queryDesc, direction, count);
1026 : }
1027 3414 : PG_FINALLY();
1028 : {
1029 58821 : nesting_level--;
1030 : }
1031 58821 : PG_END_TRY();
1032 55407 : }
1033 :
1034 : /*
1035 : * ExecutorFinish hook: all we need do is track nesting depth
1036 : */
1037 : static void
1038 53346 : pgss_ExecutorFinish(QueryDesc *queryDesc)
1039 : {
1040 53346 : nesting_level++;
1041 53346 : PG_TRY();
1042 : {
1043 53346 : if (prev_ExecutorFinish)
1044 0 : prev_ExecutorFinish(queryDesc);
1045 : else
1046 53346 : standard_ExecutorFinish(queryDesc);
1047 : }
1048 179 : PG_FINALLY();
1049 : {
1050 53346 : nesting_level--;
1051 : }
1052 53346 : PG_END_TRY();
1053 53167 : }
1054 :
1055 : /*
1056 : * ExecutorEnd hook: store results if needed
1057 : */
1058 : static void
1059 56237 : pgss_ExecutorEnd(QueryDesc *queryDesc)
1060 : {
1061 56237 : int64 queryId = queryDesc->plannedstmt->queryId;
1062 :
1063 56237 : if (queryId != INT64CONST(0) && queryDesc->query_instr &&
1064 39469 : pgss_enabled(nesting_level))
1065 : {
1066 39469 : pgss_store(queryDesc->sourceText,
1067 : queryId,
1068 39469 : queryDesc->plannedstmt->stmt_location,
1069 39469 : queryDesc->plannedstmt->stmt_len,
1070 : PGSS_EXEC,
1071 39469 : INSTR_TIME_GET_MILLISEC(queryDesc->query_instr->total),
1072 39469 : queryDesc->estate->es_total_processed,
1073 39469 : &queryDesc->query_instr->bufusage,
1074 39469 : &queryDesc->query_instr->walusage,
1075 0 : queryDesc->estate->es_jit ? &queryDesc->estate->es_jit->instr : NULL,
1076 : NULL,
1077 39469 : queryDesc->estate->es_parallel_workers_to_launch,
1078 39469 : queryDesc->estate->es_parallel_workers_launched,
1079 39469 : queryDesc->plannedstmt->planOrigin);
1080 : }
1081 :
1082 56237 : if (prev_ExecutorEnd)
1083 0 : prev_ExecutorEnd(queryDesc);
1084 : else
1085 56237 : standard_ExecutorEnd(queryDesc);
1086 56237 : }
1087 :
1088 : /*
1089 : * ProcessUtility hook
1090 : */
1091 : static void
1092 37433 : pgss_ProcessUtility(PlannedStmt *pstmt, const char *queryString,
1093 : bool readOnlyTree,
1094 : ProcessUtilityContext context,
1095 : ParamListInfo params, QueryEnvironment *queryEnv,
1096 : DestReceiver *dest, QueryCompletion *qc)
1097 : {
1098 37433 : Node *parsetree = pstmt->utilityStmt;
1099 37433 : int64 saved_queryId = pstmt->queryId;
1100 37433 : int saved_stmt_location = pstmt->stmt_location;
1101 37433 : int saved_stmt_len = pstmt->stmt_len;
1102 37433 : bool enabled = pgss_track_utility && pgss_enabled(nesting_level);
1103 :
1104 : /*
1105 : * Force utility statements to get queryId zero. We do this even in cases
1106 : * where the statement contains an optimizable statement for which a
1107 : * queryId could be derived (such as EXPLAIN or DECLARE CURSOR). For such
1108 : * cases, runtime control will first go through ProcessUtility and then
1109 : * the executor, and we don't want the executor hooks to do anything,
1110 : * since we are already measuring the statement's costs at the utility
1111 : * level.
1112 : *
1113 : * Note that this is only done if pg_stat_statements is enabled and
1114 : * configured to track utility statements, in the unlikely possibility
1115 : * that user configured another extension to handle utility statements
1116 : * only.
1117 : */
1118 37433 : if (enabled)
1119 31686 : pstmt->queryId = INT64CONST(0);
1120 :
1121 : /*
1122 : * If it's an EXECUTE statement, we don't track it and don't increment the
1123 : * nesting level. This allows the cycles to be charged to the underlying
1124 : * PREPARE instead (by the Executor hooks), which is much more useful.
1125 : *
1126 : * We also don't track execution of PREPARE. If we did, we would get one
1127 : * hash table entry for the PREPARE (with hash calculated from the query
1128 : * string), and then a different one with the same query string (but hash
1129 : * calculated from the query tree) would be used to accumulate costs of
1130 : * ensuing EXECUTEs. This would be confusing. Since PREPARE doesn't
1131 : * actually run the planner (only parse+rewrite), its costs are generally
1132 : * pretty negligible and it seems okay to just ignore it.
1133 : */
1134 37433 : if (enabled &&
1135 31686 : !IsA(parsetree, ExecuteStmt) &&
1136 28318 : !IsA(parsetree, PrepareStmt))
1137 25532 : {
1138 : instr_time start;
1139 : instr_time duration;
1140 : uint64 rows;
1141 : BufferUsage bufusage_start,
1142 : bufusage;
1143 : WalUsage walusage_start,
1144 : walusage;
1145 :
1146 28190 : bufusage_start = pgBufferUsage;
1147 28190 : walusage_start = pgWalUsage;
1148 28190 : INSTR_TIME_SET_CURRENT(start);
1149 :
1150 28190 : nesting_level++;
1151 28190 : PG_TRY();
1152 : {
1153 28190 : if (prev_ProcessUtility)
1154 0 : prev_ProcessUtility(pstmt, queryString, readOnlyTree,
1155 : context, params, queryEnv,
1156 : dest, qc);
1157 : else
1158 28190 : standard_ProcessUtility(pstmt, queryString, readOnlyTree,
1159 : context, params, queryEnv,
1160 : dest, qc);
1161 : }
1162 2658 : PG_FINALLY();
1163 : {
1164 28190 : nesting_level--;
1165 : }
1166 28190 : PG_END_TRY();
1167 :
1168 : /*
1169 : * CAUTION: do not access the *pstmt data structure again below here.
1170 : * If it was a ROLLBACK or similar, that data structure may have been
1171 : * freed. We must copy everything we still need into local variables,
1172 : * which we did above.
1173 : *
1174 : * For the same reason, we can't risk restoring pstmt->queryId to its
1175 : * former value, which'd otherwise be a good idea.
1176 : */
1177 :
1178 25532 : INSTR_TIME_SET_CURRENT(duration);
1179 25532 : INSTR_TIME_SUBTRACT(duration, start);
1180 :
1181 : /*
1182 : * Track the total number of rows retrieved or affected by the utility
1183 : * statements of COPY, FETCH, CREATE TABLE AS, CREATE MATERIALIZED
1184 : * VIEW, REFRESH MATERIALIZED VIEW and SELECT INTO.
1185 : */
1186 25529 : rows = (qc && (qc->commandTag == CMDTAG_COPY ||
1187 23726 : qc->commandTag == CMDTAG_FETCH ||
1188 23466 : qc->commandTag == CMDTAG_SELECT ||
1189 23273 : qc->commandTag == CMDTAG_REFRESH_MATERIALIZED_VIEW)) ?
1190 51061 : qc->nprocessed : 0;
1191 :
1192 : /* calc differences of buffer counters. */
1193 25532 : memset(&bufusage, 0, sizeof(BufferUsage));
1194 25532 : BufferUsageAccumDiff(&bufusage, &pgBufferUsage, &bufusage_start);
1195 :
1196 : /* calc differences of WAL counters. */
1197 25532 : memset(&walusage, 0, sizeof(WalUsage));
1198 25532 : WalUsageAccumDiff(&walusage, &pgWalUsage, &walusage_start);
1199 :
1200 51064 : pgss_store(queryString,
1201 : saved_queryId,
1202 : saved_stmt_location,
1203 : saved_stmt_len,
1204 : PGSS_EXEC,
1205 25532 : INSTR_TIME_GET_MILLISEC(duration),
1206 : rows,
1207 : &bufusage,
1208 : &walusage,
1209 : NULL,
1210 : NULL,
1211 : 0,
1212 : 0,
1213 : pstmt->planOrigin);
1214 : }
1215 : else
1216 : {
1217 : /*
1218 : * Even though we're not tracking execution time for this statement,
1219 : * we must still increment the nesting level, to ensure that functions
1220 : * evaluated within it are not seen as top-level calls. But don't do
1221 : * so for EXECUTE; that way, when control reaches pgss_planner or
1222 : * pgss_ExecutorStart, we will treat the costs as top-level if
1223 : * appropriate. Likewise, don't bump for PREPARE, so that parse
1224 : * analysis will treat the statement as top-level if appropriate.
1225 : *
1226 : * To be absolutely certain we don't mess up the nesting level,
1227 : * evaluate the bump_level condition just once.
1228 : */
1229 9243 : bool bump_level =
1230 15117 : !IsA(parsetree, ExecuteStmt) &&
1231 5874 : !IsA(parsetree, PrepareStmt);
1232 :
1233 9243 : if (bump_level)
1234 5745 : nesting_level++;
1235 9243 : PG_TRY();
1236 : {
1237 9243 : if (prev_ProcessUtility)
1238 0 : prev_ProcessUtility(pstmt, queryString, readOnlyTree,
1239 : context, params, queryEnv,
1240 : dest, qc);
1241 : else
1242 9243 : standard_ProcessUtility(pstmt, queryString, readOnlyTree,
1243 : context, params, queryEnv,
1244 : dest, qc);
1245 : }
1246 138 : PG_FINALLY();
1247 : {
1248 9243 : if (bump_level)
1249 5745 : nesting_level--;
1250 : }
1251 9243 : PG_END_TRY();
1252 : }
1253 34637 : }
1254 :
1255 : /*
1256 : * Store some statistics for a statement.
1257 : *
1258 : * If jstate is not NULL then we're trying to create an entry for which
1259 : * we have no statistics as yet; we just want to record the normalized
1260 : * query string. total_time, rows, bufusage and walusage are ignored in this
1261 : * case.
1262 : *
1263 : * If kind is PGSS_PLAN or PGSS_EXEC, its value is used as the array position
1264 : * for the arrays in the Counters field.
1265 : */
1266 : static void
1267 104066 : pgss_store(const char *query, int64 queryId,
1268 : int query_location, int query_len,
1269 : pgssStoreKind kind,
1270 : double total_time, uint64 rows,
1271 : const BufferUsage *bufusage,
1272 : const WalUsage *walusage,
1273 : const struct JitInstrumentation *jitusage,
1274 : const JumbleState *jstate,
1275 : int parallel_workers_to_launch,
1276 : int parallel_workers_launched,
1277 : PlannedStmtOrigin planOrigin)
1278 : {
1279 : pgssHashKey key;
1280 : pgssEntry *entry;
1281 104066 : char *norm_query = NULL;
1282 104066 : int encoding = GetDatabaseEncoding();
1283 :
1284 : Assert(query != NULL);
1285 :
1286 : /* Safety check... */
1287 104066 : if (!pgss || !pgss_hash)
1288 0 : return;
1289 :
1290 : /*
1291 : * Nothing to do if compute_query_id isn't enabled and no other module
1292 : * computed a query identifier.
1293 : */
1294 104066 : if (queryId == INT64CONST(0))
1295 0 : return;
1296 :
1297 : /*
1298 : * Confine our attention to the relevant part of the string, if the query
1299 : * is a portion of a multi-statement source string, and update query
1300 : * location and length if needed.
1301 : */
1302 104066 : query = CleanQuerytext(query, &query_location, &query_len);
1303 :
1304 : /* Set up key for hashtable search */
1305 :
1306 : /* clear padding */
1307 104066 : memset(&key, 0, sizeof(pgssHashKey));
1308 :
1309 104066 : key.userid = GetUserId();
1310 104066 : key.dbid = MyDatabaseId;
1311 104066 : key.queryid = queryId;
1312 104066 : key.toplevel = (nesting_level == 0);
1313 :
1314 : /* Lookup the hash table entry with shared lock. */
1315 104066 : LWLockAcquire(&pgss->lock.lock, LW_SHARED);
1316 :
1317 104066 : entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
1318 :
1319 : /* Create new entry, if not present */
1320 104066 : if (!entry)
1321 : {
1322 : Size query_offset;
1323 : int gc_count;
1324 : bool stored;
1325 : bool do_gc;
1326 :
1327 : /*
1328 : * Create a new, normalized query string if caller asked. We don't
1329 : * need to hold the lock while doing this work. (Note: in any case,
1330 : * it's possible that someone else creates a duplicate hashtable entry
1331 : * in the interval where we don't hold the lock below. That case is
1332 : * handled by entry_alloc.)
1333 : */
1334 30775 : if (jstate)
1335 : {
1336 11403 : LWLockRelease(&pgss->lock.lock);
1337 11403 : norm_query = generate_normalized_query(jstate, query,
1338 : query_location,
1339 : &query_len);
1340 11403 : LWLockAcquire(&pgss->lock.lock, LW_SHARED);
1341 : }
1342 :
1343 : /* Append new query text to file with only shared lock held */
1344 30775 : stored = qtext_store(norm_query ? norm_query : query, query_len,
1345 : &query_offset, &gc_count);
1346 :
1347 : /*
1348 : * Determine whether we need to garbage collect external query texts
1349 : * while the shared lock is still held. This micro-optimization
1350 : * avoids taking the time to decide this while holding exclusive lock.
1351 : */
1352 30775 : do_gc = need_gc_qtexts();
1353 :
1354 : /* Need exclusive lock to make a new hashtable entry - promote */
1355 30775 : LWLockRelease(&pgss->lock.lock);
1356 30775 : LWLockAcquire(&pgss->lock.lock, LW_EXCLUSIVE);
1357 :
1358 : /*
1359 : * A garbage collection may have occurred while we weren't holding the
1360 : * lock. In the unlikely event that this happens, the query text we
1361 : * stored above will have been garbage collected, so write it again.
1362 : * This should be infrequent enough that doing it while holding
1363 : * exclusive lock isn't a performance problem.
1364 : */
1365 30775 : if (!stored || pgss->gc_count != gc_count)
1366 0 : stored = qtext_store(norm_query ? norm_query : query, query_len,
1367 : &query_offset, NULL);
1368 :
1369 : /* If we failed to write to the text file, give up */
1370 30775 : if (!stored)
1371 0 : goto done;
1372 :
1373 : /* OK to create a new hashtable entry */
1374 30775 : entry = entry_alloc(&key, query_offset, query_len, encoding,
1375 : jstate != NULL);
1376 :
1377 : /* If needed, perform garbage collection while exclusive lock held */
1378 30775 : if (do_gc)
1379 0 : gc_qtexts();
1380 : }
1381 :
1382 : /* Increment the counts, except when jstate is not NULL */
1383 104066 : if (!jstate)
1384 : {
1385 : Assert(kind == PGSS_PLAN || kind == PGSS_EXEC);
1386 :
1387 : /*
1388 : * Grab the spinlock while updating the counters (see comment about
1389 : * locking rules at the head of the file)
1390 : */
1391 65151 : SpinLockAcquire(&entry->mutex);
1392 :
1393 : /* "Unstick" entry if it was previously sticky */
1394 65151 : if (IS_STICKY(entry->counters))
1395 29958 : entry->counters.usage = USAGE_INIT;
1396 :
1397 65151 : entry->counters.calls[kind] += 1;
1398 65151 : entry->counters.total_time[kind] += total_time;
1399 :
1400 65151 : if (entry->counters.calls[kind] == 1)
1401 : {
1402 30052 : entry->counters.min_time[kind] = total_time;
1403 30052 : entry->counters.max_time[kind] = total_time;
1404 30052 : entry->counters.mean_time[kind] = total_time;
1405 : }
1406 : else
1407 : {
1408 : /*
1409 : * Welford's method for accurately computing variance. See
1410 : * <http://www.johndcook.com/blog/standard_deviation/>
1411 : */
1412 35099 : double old_mean = entry->counters.mean_time[kind];
1413 :
1414 35099 : entry->counters.mean_time[kind] +=
1415 35099 : (total_time - old_mean) / entry->counters.calls[kind];
1416 35099 : entry->counters.sum_var_time[kind] +=
1417 35099 : (total_time - old_mean) * (total_time - entry->counters.mean_time[kind]);
1418 :
1419 : /*
1420 : * Calculate min and max time. min = 0 and max = 0 means that the
1421 : * min/max statistics were reset
1422 : */
1423 35099 : if (entry->counters.min_time[kind] == 0
1424 6 : && entry->counters.max_time[kind] == 0)
1425 : {
1426 3 : entry->counters.min_time[kind] = total_time;
1427 3 : entry->counters.max_time[kind] = total_time;
1428 : }
1429 : else
1430 : {
1431 35096 : if (entry->counters.min_time[kind] > total_time)
1432 6618 : entry->counters.min_time[kind] = total_time;
1433 35096 : if (entry->counters.max_time[kind] < total_time)
1434 3588 : entry->counters.max_time[kind] = total_time;
1435 : }
1436 : }
1437 65151 : entry->counters.rows += rows;
1438 65151 : entry->counters.shared_blks_hit += bufusage->shared_blks_hit;
1439 65151 : entry->counters.shared_blks_read += bufusage->shared_blks_read;
1440 65151 : entry->counters.shared_blks_dirtied += bufusage->shared_blks_dirtied;
1441 65151 : entry->counters.shared_blks_written += bufusage->shared_blks_written;
1442 65151 : entry->counters.local_blks_hit += bufusage->local_blks_hit;
1443 65151 : entry->counters.local_blks_read += bufusage->local_blks_read;
1444 65151 : entry->counters.local_blks_dirtied += bufusage->local_blks_dirtied;
1445 65151 : entry->counters.local_blks_written += bufusage->local_blks_written;
1446 65151 : entry->counters.temp_blks_read += bufusage->temp_blks_read;
1447 65151 : entry->counters.temp_blks_written += bufusage->temp_blks_written;
1448 65151 : entry->counters.shared_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_read_time);
1449 65151 : entry->counters.shared_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->shared_blk_write_time);
1450 65151 : entry->counters.local_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_read_time);
1451 65151 : entry->counters.local_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->local_blk_write_time);
1452 65151 : entry->counters.temp_blk_read_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_read_time);
1453 65151 : entry->counters.temp_blk_write_time += INSTR_TIME_GET_MILLISEC(bufusage->temp_blk_write_time);
1454 65151 : entry->counters.usage += USAGE_EXEC(total_time);
1455 65151 : entry->counters.wal_records += walusage->wal_records;
1456 65151 : entry->counters.wal_fpi += walusage->wal_fpi;
1457 65151 : entry->counters.wal_bytes += walusage->wal_bytes;
1458 65151 : entry->counters.wal_buffers_full += walusage->wal_buffers_full;
1459 65151 : if (jitusage)
1460 : {
1461 0 : entry->counters.jit_functions += jitusage->created_functions;
1462 0 : entry->counters.jit_generation_time += INSTR_TIME_GET_MILLISEC(jitusage->generation_counter);
1463 :
1464 0 : if (INSTR_TIME_GET_MILLISEC(jitusage->deform_counter))
1465 0 : entry->counters.jit_deform_count++;
1466 0 : entry->counters.jit_deform_time += INSTR_TIME_GET_MILLISEC(jitusage->deform_counter);
1467 :
1468 0 : if (INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter))
1469 0 : entry->counters.jit_inlining_count++;
1470 0 : entry->counters.jit_inlining_time += INSTR_TIME_GET_MILLISEC(jitusage->inlining_counter);
1471 :
1472 0 : if (INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter))
1473 0 : entry->counters.jit_optimization_count++;
1474 0 : entry->counters.jit_optimization_time += INSTR_TIME_GET_MILLISEC(jitusage->optimization_counter);
1475 :
1476 0 : if (INSTR_TIME_GET_MILLISEC(jitusage->emission_counter))
1477 0 : entry->counters.jit_emission_count++;
1478 0 : entry->counters.jit_emission_time += INSTR_TIME_GET_MILLISEC(jitusage->emission_counter);
1479 : }
1480 :
1481 : /* parallel worker counters */
1482 65151 : entry->counters.parallel_workers_to_launch += parallel_workers_to_launch;
1483 65151 : entry->counters.parallel_workers_launched += parallel_workers_launched;
1484 :
1485 : /* plan cache counters */
1486 65151 : if (planOrigin == PLAN_STMT_CACHE_GENERIC)
1487 3151 : entry->counters.generic_plan_calls++;
1488 62000 : else if (planOrigin == PLAN_STMT_CACHE_CUSTOM)
1489 375 : entry->counters.custom_plan_calls++;
1490 :
1491 65151 : SpinLockRelease(&entry->mutex);
1492 : }
1493 :
1494 38915 : done:
1495 104066 : LWLockRelease(&pgss->lock.lock);
1496 :
1497 : /* We postpone this clean-up until we're out of the lock */
1498 104066 : if (norm_query)
1499 11403 : pfree(norm_query);
1500 : }
1501 :
1502 : /*
1503 : * Reset statement statistics corresponding to userid, dbid, and queryid.
1504 : */
1505 : Datum
1506 1 : pg_stat_statements_reset_1_7(PG_FUNCTION_ARGS)
1507 : {
1508 : Oid userid;
1509 : Oid dbid;
1510 : int64 queryid;
1511 :
1512 1 : userid = PG_GETARG_OID(0);
1513 1 : dbid = PG_GETARG_OID(1);
1514 1 : queryid = PG_GETARG_INT64(2);
1515 :
1516 1 : entry_reset(userid, dbid, queryid, false);
1517 :
1518 1 : PG_RETURN_VOID();
1519 : }
1520 :
1521 : Datum
1522 119 : pg_stat_statements_reset_1_11(PG_FUNCTION_ARGS)
1523 : {
1524 : Oid userid;
1525 : Oid dbid;
1526 : int64 queryid;
1527 : bool minmax_only;
1528 :
1529 119 : userid = PG_GETARG_OID(0);
1530 119 : dbid = PG_GETARG_OID(1);
1531 119 : queryid = PG_GETARG_INT64(2);
1532 119 : minmax_only = PG_GETARG_BOOL(3);
1533 :
1534 119 : PG_RETURN_TIMESTAMPTZ(entry_reset(userid, dbid, queryid, minmax_only));
1535 : }
1536 :
1537 : /*
1538 : * Reset statement statistics.
1539 : */
1540 : Datum
1541 1 : pg_stat_statements_reset(PG_FUNCTION_ARGS)
1542 : {
1543 1 : entry_reset(0, 0, 0, false);
1544 :
1545 1 : PG_RETURN_VOID();
1546 : }
1547 :
1548 : /* Number of output arguments (columns) for various API versions */
1549 : #define PG_STAT_STATEMENTS_COLS_V1_0 14
1550 : #define PG_STAT_STATEMENTS_COLS_V1_1 18
1551 : #define PG_STAT_STATEMENTS_COLS_V1_2 19
1552 : #define PG_STAT_STATEMENTS_COLS_V1_3 23
1553 : #define PG_STAT_STATEMENTS_COLS_V1_8 32
1554 : #define PG_STAT_STATEMENTS_COLS_V1_9 33
1555 : #define PG_STAT_STATEMENTS_COLS_V1_10 43
1556 : #define PG_STAT_STATEMENTS_COLS_V1_11 49
1557 : #define PG_STAT_STATEMENTS_COLS_V1_12 52
1558 : #define PG_STAT_STATEMENTS_COLS_V1_13 54
1559 : #define PG_STAT_STATEMENTS_COLS 54 /* maximum of above */
1560 :
1561 : /*
1562 : * Retrieve statement statistics.
1563 : *
1564 : * The SQL API of this function has changed multiple times, and will likely
1565 : * do so again in future. To support the case where a newer version of this
1566 : * loadable module is being used with an old SQL declaration of the function,
1567 : * we continue to support the older API versions. For 1.2 and later, the
1568 : * expected API version is identified by embedding it in the C name of the
1569 : * function. Unfortunately we weren't bright enough to do that for 1.1.
1570 : */
1571 : Datum
1572 129 : pg_stat_statements_1_13(PG_FUNCTION_ARGS)
1573 : {
1574 129 : bool showtext = PG_GETARG_BOOL(0);
1575 :
1576 129 : pg_stat_statements_internal(fcinfo, PGSS_V1_13, showtext);
1577 :
1578 129 : return (Datum) 0;
1579 : }
1580 :
1581 : Datum
1582 1 : pg_stat_statements_1_12(PG_FUNCTION_ARGS)
1583 : {
1584 1 : bool showtext = PG_GETARG_BOOL(0);
1585 :
1586 1 : pg_stat_statements_internal(fcinfo, PGSS_V1_12, showtext);
1587 :
1588 1 : return (Datum) 0;
1589 : }
1590 :
1591 : Datum
1592 1 : pg_stat_statements_1_11(PG_FUNCTION_ARGS)
1593 : {
1594 1 : bool showtext = PG_GETARG_BOOL(0);
1595 :
1596 1 : pg_stat_statements_internal(fcinfo, PGSS_V1_11, showtext);
1597 :
1598 1 : return (Datum) 0;
1599 : }
1600 :
1601 : Datum
1602 1 : pg_stat_statements_1_10(PG_FUNCTION_ARGS)
1603 : {
1604 1 : bool showtext = PG_GETARG_BOOL(0);
1605 :
1606 1 : pg_stat_statements_internal(fcinfo, PGSS_V1_10, showtext);
1607 :
1608 1 : return (Datum) 0;
1609 : }
1610 :
1611 : Datum
1612 1 : pg_stat_statements_1_9(PG_FUNCTION_ARGS)
1613 : {
1614 1 : bool showtext = PG_GETARG_BOOL(0);
1615 :
1616 1 : pg_stat_statements_internal(fcinfo, PGSS_V1_9, showtext);
1617 :
1618 1 : return (Datum) 0;
1619 : }
1620 :
1621 : Datum
1622 1 : pg_stat_statements_1_8(PG_FUNCTION_ARGS)
1623 : {
1624 1 : bool showtext = PG_GETARG_BOOL(0);
1625 :
1626 1 : pg_stat_statements_internal(fcinfo, PGSS_V1_8, showtext);
1627 :
1628 1 : return (Datum) 0;
1629 : }
1630 :
1631 : Datum
1632 1 : pg_stat_statements_1_3(PG_FUNCTION_ARGS)
1633 : {
1634 1 : bool showtext = PG_GETARG_BOOL(0);
1635 :
1636 1 : pg_stat_statements_internal(fcinfo, PGSS_V1_3, showtext);
1637 :
1638 1 : return (Datum) 0;
1639 : }
1640 :
1641 : Datum
1642 0 : pg_stat_statements_1_2(PG_FUNCTION_ARGS)
1643 : {
1644 0 : bool showtext = PG_GETARG_BOOL(0);
1645 :
1646 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_2, showtext);
1647 :
1648 0 : return (Datum) 0;
1649 : }
1650 :
1651 : /*
1652 : * Legacy entry point for pg_stat_statements() API versions 1.0 and 1.1.
1653 : * This can be removed someday, perhaps.
1654 : */
1655 : Datum
1656 0 : pg_stat_statements(PG_FUNCTION_ARGS)
1657 : {
1658 : /* If it's really API 1.1, we'll figure that out below */
1659 0 : pg_stat_statements_internal(fcinfo, PGSS_V1_0, true);
1660 :
1661 0 : return (Datum) 0;
1662 : }
1663 :
1664 : /* Common code for all versions of pg_stat_statements() */
1665 : static void
1666 135 : pg_stat_statements_internal(FunctionCallInfo fcinfo,
1667 : pgssVersion api_version,
1668 : bool showtext)
1669 : {
1670 135 : ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1671 135 : Oid userid = GetUserId();
1672 135 : bool is_allowed_role = false;
1673 135 : char *qbuffer = NULL;
1674 135 : Size qbuffer_size = 0;
1675 135 : Size extent = 0;
1676 135 : int gc_count = 0;
1677 : HASH_SEQ_STATUS hash_seq;
1678 : pgssEntry *entry;
1679 :
1680 : /*
1681 : * Superusers or roles with the privileges of pg_read_all_stats members
1682 : * are allowed
1683 : */
1684 135 : is_allowed_role = has_privs_of_role(userid, ROLE_PG_READ_ALL_STATS);
1685 :
1686 : /* hash table must exist already */
1687 135 : if (!pgss || !pgss_hash)
1688 0 : ereport(ERROR,
1689 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1690 : errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
1691 :
1692 135 : InitMaterializedSRF(fcinfo, 0);
1693 :
1694 : /*
1695 : * Check we have the expected number of output arguments. Aside from
1696 : * being a good safety check, we need a kluge here to detect API version
1697 : * 1.1, which was wedged into the code in an ill-considered way.
1698 : */
1699 135 : switch (rsinfo->setDesc->natts)
1700 : {
1701 0 : case PG_STAT_STATEMENTS_COLS_V1_0:
1702 0 : if (api_version != PGSS_V1_0)
1703 0 : elog(ERROR, "incorrect number of output arguments");
1704 0 : break;
1705 0 : case PG_STAT_STATEMENTS_COLS_V1_1:
1706 : /* pg_stat_statements() should have told us 1.0 */
1707 0 : if (api_version != PGSS_V1_0)
1708 0 : elog(ERROR, "incorrect number of output arguments");
1709 0 : api_version = PGSS_V1_1;
1710 0 : break;
1711 0 : case PG_STAT_STATEMENTS_COLS_V1_2:
1712 0 : if (api_version != PGSS_V1_2)
1713 0 : elog(ERROR, "incorrect number of output arguments");
1714 0 : break;
1715 1 : case PG_STAT_STATEMENTS_COLS_V1_3:
1716 1 : if (api_version != PGSS_V1_3)
1717 0 : elog(ERROR, "incorrect number of output arguments");
1718 1 : break;
1719 1 : case PG_STAT_STATEMENTS_COLS_V1_8:
1720 1 : if (api_version != PGSS_V1_8)
1721 0 : elog(ERROR, "incorrect number of output arguments");
1722 1 : break;
1723 1 : case PG_STAT_STATEMENTS_COLS_V1_9:
1724 1 : if (api_version != PGSS_V1_9)
1725 0 : elog(ERROR, "incorrect number of output arguments");
1726 1 : break;
1727 1 : case PG_STAT_STATEMENTS_COLS_V1_10:
1728 1 : if (api_version != PGSS_V1_10)
1729 0 : elog(ERROR, "incorrect number of output arguments");
1730 1 : break;
1731 1 : case PG_STAT_STATEMENTS_COLS_V1_11:
1732 1 : if (api_version != PGSS_V1_11)
1733 0 : elog(ERROR, "incorrect number of output arguments");
1734 1 : break;
1735 1 : case PG_STAT_STATEMENTS_COLS_V1_12:
1736 1 : if (api_version != PGSS_V1_12)
1737 0 : elog(ERROR, "incorrect number of output arguments");
1738 1 : break;
1739 129 : case PG_STAT_STATEMENTS_COLS_V1_13:
1740 129 : if (api_version != PGSS_V1_13)
1741 0 : elog(ERROR, "incorrect number of output arguments");
1742 129 : break;
1743 0 : default:
1744 0 : elog(ERROR, "incorrect number of output arguments");
1745 : }
1746 :
1747 : /*
1748 : * We'd like to load the query text file (if needed) while not holding any
1749 : * lock on pgss->lock. In the worst case we'll have to do this again
1750 : * after we have the lock, but it's unlikely enough to make this a win
1751 : * despite occasional duplicated work. We need to reload if anybody
1752 : * writes to the file (either a retail qtext_store(), or a garbage
1753 : * collection) between this point and where we've gotten shared lock. If
1754 : * a qtext_store is actually in progress when we look, we might as well
1755 : * skip the speculative load entirely.
1756 : */
1757 135 : if (showtext)
1758 : {
1759 : int n_writers;
1760 :
1761 : /* Take the mutex so we can examine variables */
1762 135 : SpinLockAcquire(&pgss->mutex);
1763 135 : extent = pgss->extent;
1764 135 : n_writers = pgss->n_writers;
1765 135 : gc_count = pgss->gc_count;
1766 135 : SpinLockRelease(&pgss->mutex);
1767 :
1768 : /* No point in loading file now if there are active writers */
1769 135 : if (n_writers == 0)
1770 135 : qbuffer = qtext_load_file(&qbuffer_size);
1771 : }
1772 :
1773 : /*
1774 : * Get shared lock, load or reload the query text file if we must, and
1775 : * iterate over the hashtable entries.
1776 : *
1777 : * With a large hash table, we might be holding the lock rather longer
1778 : * than one could wish. However, this only blocks creation of new hash
1779 : * table entries, and the larger the hash table the less likely that is to
1780 : * be needed. So we can hope this is okay. Perhaps someday we'll decide
1781 : * we need to partition the hash table to limit the time spent holding any
1782 : * one lock.
1783 : */
1784 135 : LWLockAcquire(&pgss->lock.lock, LW_SHARED);
1785 :
1786 135 : if (showtext)
1787 : {
1788 : /*
1789 : * Here it is safe to examine extent and gc_count without taking the
1790 : * mutex. Note that although other processes might change
1791 : * pgss->extent just after we look at it, the strings they then write
1792 : * into the file cannot yet be referenced in the hashtable, so we
1793 : * don't care whether we see them or not.
1794 : *
1795 : * If qtext_load_file fails, we just press on; we'll return NULL for
1796 : * every query text.
1797 : */
1798 135 : if (qbuffer == NULL ||
1799 135 : pgss->extent != extent ||
1800 135 : pgss->gc_count != gc_count)
1801 : {
1802 0 : if (qbuffer)
1803 0 : pfree(qbuffer);
1804 0 : qbuffer = qtext_load_file(&qbuffer_size);
1805 : }
1806 : }
1807 :
1808 135 : hash_seq_init(&hash_seq, pgss_hash);
1809 29146 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
1810 : {
1811 : Datum values[PG_STAT_STATEMENTS_COLS];
1812 : bool nulls[PG_STAT_STATEMENTS_COLS];
1813 29011 : int i = 0;
1814 : Counters tmp;
1815 : double stddev;
1816 29011 : int64 queryid = entry->key.queryid;
1817 : TimestampTz stats_since;
1818 : TimestampTz minmax_stats_since;
1819 :
1820 29011 : memset(values, 0, sizeof(values));
1821 29011 : memset(nulls, 0, sizeof(nulls));
1822 :
1823 29011 : values[i++] = ObjectIdGetDatum(entry->key.userid);
1824 29011 : values[i++] = ObjectIdGetDatum(entry->key.dbid);
1825 29011 : if (api_version >= PGSS_V1_9)
1826 28999 : values[i++] = BoolGetDatum(entry->key.toplevel);
1827 :
1828 29011 : if (is_allowed_role || entry->key.userid == userid)
1829 : {
1830 29007 : if (api_version >= PGSS_V1_2)
1831 29007 : values[i++] = Int64GetDatumFast(queryid);
1832 :
1833 29007 : if (showtext)
1834 : {
1835 29007 : char *qstr = qtext_fetch(entry->query_offset,
1836 : entry->query_len,
1837 : qbuffer,
1838 : qbuffer_size);
1839 :
1840 29007 : if (qstr)
1841 : {
1842 : char *enc;
1843 :
1844 29007 : enc = pg_any_to_server(qstr,
1845 : entry->query_len,
1846 : entry->encoding);
1847 :
1848 29007 : values[i++] = CStringGetTextDatum(enc);
1849 :
1850 29007 : if (enc != qstr)
1851 0 : pfree(enc);
1852 : }
1853 : else
1854 : {
1855 : /* Just return a null if we fail to find the text */
1856 0 : nulls[i++] = true;
1857 : }
1858 : }
1859 : else
1860 : {
1861 : /* Query text not requested */
1862 0 : nulls[i++] = true;
1863 : }
1864 : }
1865 : else
1866 : {
1867 : /* Don't show queryid */
1868 4 : if (api_version >= PGSS_V1_2)
1869 4 : nulls[i++] = true;
1870 :
1871 : /*
1872 : * Don't show query text, but hint as to the reason for not doing
1873 : * so if it was requested
1874 : */
1875 4 : if (showtext)
1876 4 : values[i++] = CStringGetTextDatum("<insufficient privilege>");
1877 : else
1878 0 : nulls[i++] = true;
1879 : }
1880 :
1881 : /* copy counters to a local variable to keep locking time short */
1882 29011 : SpinLockAcquire(&entry->mutex);
1883 29011 : tmp = entry->counters;
1884 29011 : SpinLockRelease(&entry->mutex);
1885 :
1886 : /*
1887 : * The spinlock is not required when reading these two as they are
1888 : * always updated when holding pgss->lock exclusively.
1889 : */
1890 29011 : stats_since = entry->stats_since;
1891 29011 : minmax_stats_since = entry->minmax_stats_since;
1892 :
1893 : /* Skip entry if unexecuted (ie, it's a pending "sticky" entry) */
1894 29011 : if (IS_STICKY(tmp))
1895 45 : continue;
1896 :
1897 : /* Note that we rely on PGSS_PLAN being 0 and PGSS_EXEC being 1. */
1898 86898 : for (int kind = 0; kind < PGSS_NUMKIND; kind++)
1899 : {
1900 57932 : if (kind == PGSS_EXEC || api_version >= PGSS_V1_8)
1901 : {
1902 57928 : values[i++] = Int64GetDatumFast(tmp.calls[kind]);
1903 57928 : values[i++] = Float8GetDatumFast(tmp.total_time[kind]);
1904 : }
1905 :
1906 57932 : if ((kind == PGSS_EXEC && api_version >= PGSS_V1_3) ||
1907 : api_version >= PGSS_V1_8)
1908 : {
1909 57928 : values[i++] = Float8GetDatumFast(tmp.min_time[kind]);
1910 57928 : values[i++] = Float8GetDatumFast(tmp.max_time[kind]);
1911 57928 : values[i++] = Float8GetDatumFast(tmp.mean_time[kind]);
1912 :
1913 : /*
1914 : * Note we are calculating the population variance here, not
1915 : * the sample variance, as we have data for the whole
1916 : * population, so Bessel's correction is not used, and we
1917 : * don't divide by tmp.calls - 1.
1918 : */
1919 57928 : if (tmp.calls[kind] > 1)
1920 5394 : stddev = sqrt(tmp.sum_var_time[kind] / tmp.calls[kind]);
1921 : else
1922 52534 : stddev = 0.0;
1923 57928 : values[i++] = Float8GetDatumFast(stddev);
1924 : }
1925 : }
1926 28966 : values[i++] = Int64GetDatumFast(tmp.rows);
1927 28966 : values[i++] = Int64GetDatumFast(tmp.shared_blks_hit);
1928 28966 : values[i++] = Int64GetDatumFast(tmp.shared_blks_read);
1929 28966 : if (api_version >= PGSS_V1_1)
1930 28966 : values[i++] = Int64GetDatumFast(tmp.shared_blks_dirtied);
1931 28966 : values[i++] = Int64GetDatumFast(tmp.shared_blks_written);
1932 28966 : values[i++] = Int64GetDatumFast(tmp.local_blks_hit);
1933 28966 : values[i++] = Int64GetDatumFast(tmp.local_blks_read);
1934 28966 : if (api_version >= PGSS_V1_1)
1935 28966 : values[i++] = Int64GetDatumFast(tmp.local_blks_dirtied);
1936 28966 : values[i++] = Int64GetDatumFast(tmp.local_blks_written);
1937 28966 : values[i++] = Int64GetDatumFast(tmp.temp_blks_read);
1938 28966 : values[i++] = Int64GetDatumFast(tmp.temp_blks_written);
1939 28966 : if (api_version >= PGSS_V1_1)
1940 : {
1941 28966 : values[i++] = Float8GetDatumFast(tmp.shared_blk_read_time);
1942 28966 : values[i++] = Float8GetDatumFast(tmp.shared_blk_write_time);
1943 : }
1944 28966 : if (api_version >= PGSS_V1_11)
1945 : {
1946 28938 : values[i++] = Float8GetDatumFast(tmp.local_blk_read_time);
1947 28938 : values[i++] = Float8GetDatumFast(tmp.local_blk_write_time);
1948 : }
1949 28966 : if (api_version >= PGSS_V1_10)
1950 : {
1951 28947 : values[i++] = Float8GetDatumFast(tmp.temp_blk_read_time);
1952 28947 : values[i++] = Float8GetDatumFast(tmp.temp_blk_write_time);
1953 : }
1954 28966 : if (api_version >= PGSS_V1_8)
1955 : {
1956 : char buf[256];
1957 : Datum wal_bytes;
1958 :
1959 28962 : values[i++] = Int64GetDatumFast(tmp.wal_records);
1960 28962 : values[i++] = Int64GetDatumFast(tmp.wal_fpi);
1961 :
1962 28962 : snprintf(buf, sizeof buf, UINT64_FORMAT, tmp.wal_bytes);
1963 :
1964 : /* Convert to numeric. */
1965 28962 : wal_bytes = DirectFunctionCall3(numeric_in,
1966 : CStringGetDatum(buf),
1967 : ObjectIdGetDatum(0),
1968 : Int32GetDatum(-1));
1969 28962 : values[i++] = wal_bytes;
1970 : }
1971 28966 : if (api_version >= PGSS_V1_12)
1972 : {
1973 28928 : values[i++] = Int64GetDatumFast(tmp.wal_buffers_full);
1974 : }
1975 28966 : if (api_version >= PGSS_V1_10)
1976 : {
1977 28947 : values[i++] = Int64GetDatumFast(tmp.jit_functions);
1978 28947 : values[i++] = Float8GetDatumFast(tmp.jit_generation_time);
1979 28947 : values[i++] = Int64GetDatumFast(tmp.jit_inlining_count);
1980 28947 : values[i++] = Float8GetDatumFast(tmp.jit_inlining_time);
1981 28947 : values[i++] = Int64GetDatumFast(tmp.jit_optimization_count);
1982 28947 : values[i++] = Float8GetDatumFast(tmp.jit_optimization_time);
1983 28947 : values[i++] = Int64GetDatumFast(tmp.jit_emission_count);
1984 28947 : values[i++] = Float8GetDatumFast(tmp.jit_emission_time);
1985 : }
1986 28966 : if (api_version >= PGSS_V1_11)
1987 : {
1988 28938 : values[i++] = Int64GetDatumFast(tmp.jit_deform_count);
1989 28938 : values[i++] = Float8GetDatumFast(tmp.jit_deform_time);
1990 : }
1991 28966 : if (api_version >= PGSS_V1_12)
1992 : {
1993 28928 : values[i++] = Int64GetDatumFast(tmp.parallel_workers_to_launch);
1994 28928 : values[i++] = Int64GetDatumFast(tmp.parallel_workers_launched);
1995 : }
1996 28966 : if (api_version >= PGSS_V1_13)
1997 : {
1998 28923 : values[i++] = Int64GetDatumFast(tmp.generic_plan_calls);
1999 28923 : values[i++] = Int64GetDatumFast(tmp.custom_plan_calls);
2000 : }
2001 28966 : if (api_version >= PGSS_V1_11)
2002 : {
2003 28938 : values[i++] = TimestampTzGetDatum(stats_since);
2004 28938 : values[i++] = TimestampTzGetDatum(minmax_stats_since);
2005 : }
2006 :
2007 : Assert(i == (api_version == PGSS_V1_0 ? PG_STAT_STATEMENTS_COLS_V1_0 :
2008 : api_version == PGSS_V1_1 ? PG_STAT_STATEMENTS_COLS_V1_1 :
2009 : api_version == PGSS_V1_2 ? PG_STAT_STATEMENTS_COLS_V1_2 :
2010 : api_version == PGSS_V1_3 ? PG_STAT_STATEMENTS_COLS_V1_3 :
2011 : api_version == PGSS_V1_8 ? PG_STAT_STATEMENTS_COLS_V1_8 :
2012 : api_version == PGSS_V1_9 ? PG_STAT_STATEMENTS_COLS_V1_9 :
2013 : api_version == PGSS_V1_10 ? PG_STAT_STATEMENTS_COLS_V1_10 :
2014 : api_version == PGSS_V1_11 ? PG_STAT_STATEMENTS_COLS_V1_11 :
2015 : api_version == PGSS_V1_12 ? PG_STAT_STATEMENTS_COLS_V1_12 :
2016 : api_version == PGSS_V1_13 ? PG_STAT_STATEMENTS_COLS_V1_13 :
2017 : -1 /* fail if you forget to update this assert */ ));
2018 :
2019 28966 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
2020 : }
2021 :
2022 135 : LWLockRelease(&pgss->lock.lock);
2023 :
2024 135 : if (qbuffer)
2025 135 : pfree(qbuffer);
2026 135 : }
2027 :
2028 : /* Number of output arguments (columns) for pg_stat_statements_info */
2029 : #define PG_STAT_STATEMENTS_INFO_COLS 2
2030 :
2031 : /*
2032 : * Return statistics of pg_stat_statements.
2033 : */
2034 : Datum
2035 3 : pg_stat_statements_info(PG_FUNCTION_ARGS)
2036 : {
2037 : pgssGlobalStats stats;
2038 : TupleDesc tupdesc;
2039 3 : Datum values[PG_STAT_STATEMENTS_INFO_COLS] = {0};
2040 3 : bool nulls[PG_STAT_STATEMENTS_INFO_COLS] = {0};
2041 :
2042 3 : if (!pgss || !pgss_hash)
2043 0 : ereport(ERROR,
2044 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2045 : errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
2046 :
2047 : /* Build a tuple descriptor for our result type */
2048 3 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
2049 0 : elog(ERROR, "return type must be a row type");
2050 :
2051 : /* Read global statistics for pg_stat_statements */
2052 3 : SpinLockAcquire(&pgss->mutex);
2053 3 : stats = pgss->stats;
2054 3 : SpinLockRelease(&pgss->mutex);
2055 :
2056 3 : values[0] = Int64GetDatum(stats.dealloc);
2057 3 : values[1] = TimestampTzGetDatum(stats.stats_reset);
2058 :
2059 3 : PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
2060 : }
2061 :
2062 : /*
2063 : * Allocate a new hashtable entry.
2064 : * caller must hold an exclusive lock on pgss->lock
2065 : *
2066 : * "query" need not be null-terminated; we rely on query_len instead
2067 : *
2068 : * If "sticky" is true, make the new entry artificially sticky so that it will
2069 : * probably still be there when the query finishes execution. We do this by
2070 : * giving it a median usage value rather than the normal value. (Strictly
2071 : * speaking, query strings are normalized on a best effort basis, though it
2072 : * would be difficult to demonstrate this even under artificial conditions.)
2073 : *
2074 : * Note: despite needing exclusive lock, it's not an error for the target
2075 : * entry to already exist. This is because pgss_store releases and
2076 : * reacquires lock after failing to find a match; so someone else could
2077 : * have made the entry while we waited to get exclusive lock.
2078 : */
2079 : static pgssEntry *
2080 58767 : entry_alloc(pgssHashKey *key, Size query_offset, int query_len, int encoding,
2081 : bool sticky)
2082 : {
2083 : pgssEntry *entry;
2084 : bool found;
2085 :
2086 : /* Make space if needed */
2087 58767 : while (hash_get_num_entries(pgss_hash) >= pgss_max)
2088 0 : entry_dealloc();
2089 :
2090 : /* Find or create an entry with desired hash code */
2091 58767 : entry = (pgssEntry *) hash_search(pgss_hash, key, HASH_ENTER, &found);
2092 :
2093 58767 : if (!found)
2094 : {
2095 : /* New entry, initialize it */
2096 :
2097 : /* reset the statistics */
2098 58767 : memset(&entry->counters, 0, sizeof(Counters));
2099 : /* set the appropriate initial usage count */
2100 58767 : entry->counters.usage = sticky ? pgss->cur_median_usage : USAGE_INIT;
2101 : /* re-initialize the mutex each time ... we assume no one using it */
2102 58767 : SpinLockInit(&entry->mutex);
2103 : /* ... and don't forget the query text metadata */
2104 : Assert(query_len >= 0);
2105 58767 : entry->query_offset = query_offset;
2106 58767 : entry->query_len = query_len;
2107 58767 : entry->encoding = encoding;
2108 58767 : entry->stats_since = GetCurrentTimestamp();
2109 58767 : entry->minmax_stats_since = entry->stats_since;
2110 : }
2111 :
2112 58767 : return entry;
2113 : }
2114 :
2115 : /*
2116 : * qsort comparator for sorting into increasing usage order
2117 : */
2118 : static int
2119 0 : entry_cmp(const void *lhs, const void *rhs)
2120 : {
2121 0 : double l_usage = (*(pgssEntry *const *) lhs)->counters.usage;
2122 0 : double r_usage = (*(pgssEntry *const *) rhs)->counters.usage;
2123 :
2124 0 : if (l_usage < r_usage)
2125 0 : return -1;
2126 0 : else if (l_usage > r_usage)
2127 0 : return +1;
2128 : else
2129 0 : return 0;
2130 : }
2131 :
2132 : /*
2133 : * Deallocate least-used entries.
2134 : *
2135 : * Caller must hold an exclusive lock on pgss->lock.
2136 : */
2137 : static void
2138 0 : entry_dealloc(void)
2139 : {
2140 : HASH_SEQ_STATUS hash_seq;
2141 : pgssEntry **entries;
2142 : pgssEntry *entry;
2143 : int nvictims;
2144 : int i;
2145 : Size tottextlen;
2146 : int nvalidtexts;
2147 :
2148 : /*
2149 : * Sort entries by usage and deallocate USAGE_DEALLOC_PERCENT of them.
2150 : * While we're scanning the table, apply the decay factor to the usage
2151 : * values, and update the mean query length.
2152 : *
2153 : * Note that the mean query length is almost immediately obsolete, since
2154 : * we compute it before not after discarding the least-used entries.
2155 : * Hopefully, that doesn't affect the mean too much; it doesn't seem worth
2156 : * making two passes to get a more current result. Likewise, the new
2157 : * cur_median_usage includes the entries we're about to zap.
2158 : */
2159 :
2160 0 : entries = palloc(hash_get_num_entries(pgss_hash) * sizeof(pgssEntry *));
2161 :
2162 0 : i = 0;
2163 0 : tottextlen = 0;
2164 0 : nvalidtexts = 0;
2165 :
2166 0 : hash_seq_init(&hash_seq, pgss_hash);
2167 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2168 : {
2169 0 : entries[i++] = entry;
2170 : /* "Sticky" entries get a different usage decay rate. */
2171 0 : if (IS_STICKY(entry->counters))
2172 0 : entry->counters.usage *= STICKY_DECREASE_FACTOR;
2173 : else
2174 0 : entry->counters.usage *= USAGE_DECREASE_FACTOR;
2175 : /* In the mean length computation, ignore dropped texts. */
2176 0 : if (entry->query_len >= 0)
2177 : {
2178 0 : tottextlen += entry->query_len + 1;
2179 0 : nvalidtexts++;
2180 : }
2181 : }
2182 :
2183 : /* Sort into increasing order by usage */
2184 0 : qsort(entries, i, sizeof(pgssEntry *), entry_cmp);
2185 :
2186 : /* Record the (approximate) median usage */
2187 0 : if (i > 0)
2188 0 : pgss->cur_median_usage = entries[i / 2]->counters.usage;
2189 : /* Record the mean query length */
2190 0 : if (nvalidtexts > 0)
2191 0 : pgss->mean_query_len = tottextlen / nvalidtexts;
2192 : else
2193 0 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
2194 :
2195 : /* Now zap an appropriate fraction of lowest-usage entries */
2196 0 : nvictims = Max(10, i * USAGE_DEALLOC_PERCENT / 100);
2197 0 : nvictims = Min(nvictims, i);
2198 :
2199 0 : for (i = 0; i < nvictims; i++)
2200 : {
2201 0 : hash_search(pgss_hash, &entries[i]->key, HASH_REMOVE, NULL);
2202 : }
2203 :
2204 0 : pfree(entries);
2205 :
2206 : /* Increment the number of times entries are deallocated */
2207 0 : SpinLockAcquire(&pgss->mutex);
2208 0 : pgss->stats.dealloc += 1;
2209 0 : SpinLockRelease(&pgss->mutex);
2210 0 : }
2211 :
2212 : /*
2213 : * Given a query string (not necessarily null-terminated), allocate a new
2214 : * entry in the external query text file and store the string there.
2215 : *
2216 : * If successful, returns true, and stores the new entry's offset in the file
2217 : * into *query_offset. Also, if gc_count isn't NULL, *gc_count is set to the
2218 : * number of garbage collections that have occurred so far.
2219 : *
2220 : * On failure, returns false.
2221 : *
2222 : * At least a shared lock on pgss->lock must be held by the caller, so as
2223 : * to prevent a concurrent garbage collection. Share-lock-holding callers
2224 : * should pass a gc_count pointer to obtain the number of garbage collections,
2225 : * so that they can recheck the count after obtaining exclusive lock to
2226 : * detect whether a garbage collection occurred (and removed this entry).
2227 : */
2228 : static bool
2229 30775 : qtext_store(const char *query, int query_len,
2230 : Size *query_offset, int *gc_count)
2231 : {
2232 : Size off;
2233 : int fd;
2234 :
2235 : /*
2236 : * We use a spinlock to protect extent/n_writers/gc_count, so that
2237 : * multiple processes may execute this function concurrently.
2238 : */
2239 30775 : SpinLockAcquire(&pgss->mutex);
2240 30775 : off = pgss->extent;
2241 30775 : pgss->extent += query_len + 1;
2242 30775 : pgss->n_writers++;
2243 30775 : if (gc_count)
2244 30775 : *gc_count = pgss->gc_count;
2245 30775 : SpinLockRelease(&pgss->mutex);
2246 :
2247 30775 : *query_offset = off;
2248 :
2249 : /*
2250 : * Don't allow the file to grow larger than what qtext_load_file can
2251 : * (theoretically) handle. This has been seen to be reachable on 32-bit
2252 : * platforms.
2253 : */
2254 30775 : if (unlikely(query_len >= MaxAllocHugeSize - off))
2255 : {
2256 0 : errno = EFBIG; /* not quite right, but it'll do */
2257 0 : fd = -1;
2258 0 : goto error;
2259 : }
2260 :
2261 : /* Now write the data into the successfully-reserved part of the file */
2262 30775 : fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDWR | O_CREAT | PG_BINARY);
2263 30775 : if (fd < 0)
2264 0 : goto error;
2265 :
2266 30775 : if (pg_pwrite(fd, query, query_len, off) != query_len)
2267 0 : goto error;
2268 30775 : if (pg_pwrite(fd, "\0", 1, off + query_len) != 1)
2269 0 : goto error;
2270 :
2271 30775 : CloseTransientFile(fd);
2272 :
2273 : /* Mark our write complete */
2274 30775 : SpinLockAcquire(&pgss->mutex);
2275 30775 : pgss->n_writers--;
2276 30775 : SpinLockRelease(&pgss->mutex);
2277 :
2278 30775 : return true;
2279 :
2280 0 : error:
2281 0 : ereport(LOG,
2282 : (errcode_for_file_access(),
2283 : errmsg("could not write file \"%s\": %m",
2284 : PGSS_TEXT_FILE)));
2285 :
2286 0 : if (fd >= 0)
2287 0 : CloseTransientFile(fd);
2288 :
2289 : /* Mark our write complete */
2290 0 : SpinLockAcquire(&pgss->mutex);
2291 0 : pgss->n_writers--;
2292 0 : SpinLockRelease(&pgss->mutex);
2293 :
2294 0 : return false;
2295 : }
2296 :
2297 : /*
2298 : * Read the external query text file into a palloc'd buffer.
2299 : *
2300 : * Returns NULL (without throwing an error) if unable to read, eg
2301 : * file not there or insufficient memory.
2302 : *
2303 : * On success, the buffer size is also returned into *buffer_size.
2304 : *
2305 : * This can be called without any lock on pgss->lock, but in that case
2306 : * the caller is responsible for verifying that the result is sane.
2307 : */
2308 : static char *
2309 142 : qtext_load_file(Size *buffer_size)
2310 : {
2311 : char *buf;
2312 : int fd;
2313 : struct stat stat;
2314 : Size nread;
2315 :
2316 142 : fd = OpenTransientFile(PGSS_TEXT_FILE, O_RDONLY | PG_BINARY);
2317 142 : if (fd < 0)
2318 : {
2319 0 : if (errno != ENOENT)
2320 0 : ereport(LOG,
2321 : (errcode_for_file_access(),
2322 : errmsg("could not read file \"%s\": %m",
2323 : PGSS_TEXT_FILE)));
2324 0 : return NULL;
2325 : }
2326 :
2327 : /* Get file length */
2328 142 : if (fstat(fd, &stat))
2329 : {
2330 0 : ereport(LOG,
2331 : (errcode_for_file_access(),
2332 : errmsg("could not stat file \"%s\": %m",
2333 : PGSS_TEXT_FILE)));
2334 0 : CloseTransientFile(fd);
2335 0 : return NULL;
2336 : }
2337 :
2338 : /* Allocate buffer; beware that off_t might be wider than size_t */
2339 142 : if (stat.st_size <= MaxAllocHugeSize)
2340 142 : buf = (char *) palloc_extended(stat.st_size, MCXT_ALLOC_HUGE | MCXT_ALLOC_NO_OOM);
2341 : else
2342 0 : buf = NULL;
2343 142 : if (buf == NULL)
2344 : {
2345 0 : ereport(LOG,
2346 : (errcode(ERRCODE_OUT_OF_MEMORY),
2347 : errmsg("out of memory"),
2348 : errdetail("Could not allocate enough memory to read file \"%s\".",
2349 : PGSS_TEXT_FILE)));
2350 0 : CloseTransientFile(fd);
2351 0 : return NULL;
2352 : }
2353 :
2354 : /*
2355 : * OK, slurp in the file. Windows fails if we try to read more than
2356 : * INT_MAX bytes at once, and other platforms might not like that either,
2357 : * so read a very large file in 1GB segments.
2358 : */
2359 142 : nread = 0;
2360 283 : while (nread < stat.st_size)
2361 : {
2362 141 : int toread = Min(1024 * 1024 * 1024, stat.st_size - nread);
2363 :
2364 : /*
2365 : * If we get a short read and errno doesn't get set, the reason is
2366 : * probably that garbage collection truncated the file since we did
2367 : * the fstat(), so we don't log a complaint --- but we don't return
2368 : * the data, either, since it's most likely corrupt due to concurrent
2369 : * writes from garbage collection.
2370 : */
2371 141 : errno = 0;
2372 141 : if (read(fd, buf + nread, toread) != toread)
2373 : {
2374 0 : if (errno)
2375 0 : ereport(LOG,
2376 : (errcode_for_file_access(),
2377 : errmsg("could not read file \"%s\": %m",
2378 : PGSS_TEXT_FILE)));
2379 0 : pfree(buf);
2380 0 : CloseTransientFile(fd);
2381 0 : return NULL;
2382 : }
2383 141 : nread += toread;
2384 : }
2385 :
2386 142 : if (CloseTransientFile(fd) != 0)
2387 0 : ereport(LOG,
2388 : (errcode_for_file_access(),
2389 : errmsg("could not close file \"%s\": %m", PGSS_TEXT_FILE)));
2390 :
2391 142 : *buffer_size = nread;
2392 142 : return buf;
2393 : }
2394 :
2395 : /*
2396 : * Locate a query text in the file image previously read by qtext_load_file().
2397 : *
2398 : * We validate the given offset/length, and return NULL if bogus. Otherwise,
2399 : * the result points to a null-terminated string within the buffer.
2400 : */
2401 : static char *
2402 86852 : qtext_fetch(Size query_offset, int query_len,
2403 : char *buffer, Size buffer_size)
2404 : {
2405 : /* File read failed? */
2406 86852 : if (buffer == NULL)
2407 0 : return NULL;
2408 : /* Bogus offset/length? */
2409 86852 : if (query_len < 0 ||
2410 86852 : query_offset + query_len >= buffer_size)
2411 0 : return NULL;
2412 : /* As a further sanity check, make sure there's a trailing null */
2413 86852 : if (buffer[query_offset + query_len] != '\0')
2414 0 : return NULL;
2415 : /* Looks OK */
2416 86852 : return buffer + query_offset;
2417 : }
2418 :
2419 : /*
2420 : * Do we need to garbage-collect the external query text file?
2421 : *
2422 : * Caller should hold at least a shared lock on pgss->lock.
2423 : */
2424 : static bool
2425 30775 : need_gc_qtexts(void)
2426 : {
2427 : Size extent;
2428 :
2429 : /* Read shared extent pointer */
2430 30775 : SpinLockAcquire(&pgss->mutex);
2431 30775 : extent = pgss->extent;
2432 30775 : SpinLockRelease(&pgss->mutex);
2433 :
2434 : /*
2435 : * Don't proceed if file does not exceed 512 bytes per possible entry.
2436 : *
2437 : * Here and in the next test, 32-bit machines have overflow hazards if
2438 : * pgss_max and/or mean_query_len are large. Force the multiplications
2439 : * and comparisons to be done in uint64 arithmetic to forestall trouble.
2440 : */
2441 30775 : if ((uint64) extent < (uint64) 512 * pgss_max)
2442 30775 : return false;
2443 :
2444 : /*
2445 : * Don't proceed if file is less than about 50% bloat. Nothing can or
2446 : * should be done in the event of unusually large query texts accounting
2447 : * for file's large size. We go to the trouble of maintaining the mean
2448 : * query length in order to prevent garbage collection from thrashing
2449 : * uselessly.
2450 : */
2451 0 : if ((uint64) extent < (uint64) pgss->mean_query_len * pgss_max * 2)
2452 0 : return false;
2453 :
2454 0 : return true;
2455 : }
2456 :
2457 : /*
2458 : * Garbage-collect orphaned query texts in external file.
2459 : *
2460 : * This won't be called often in the typical case, since it's likely that
2461 : * there won't be too much churn, and besides, a similar compaction process
2462 : * occurs when serializing to disk at shutdown or as part of resetting.
2463 : * Despite this, it seems prudent to plan for the edge case where the file
2464 : * becomes unreasonably large, with no other method of compaction likely to
2465 : * occur in the foreseeable future.
2466 : *
2467 : * The caller must hold an exclusive lock on pgss->lock.
2468 : *
2469 : * At the first sign of trouble we unlink the query text file to get a clean
2470 : * slate (although existing statistics are retained), rather than risk
2471 : * thrashing by allowing the same problem case to recur indefinitely.
2472 : */
2473 : static void
2474 0 : gc_qtexts(void)
2475 : {
2476 : char *qbuffer;
2477 : Size qbuffer_size;
2478 0 : FILE *qfile = NULL;
2479 : HASH_SEQ_STATUS hash_seq;
2480 : pgssEntry *entry;
2481 : Size extent;
2482 : int nentries;
2483 :
2484 : /*
2485 : * When called from pgss_store, some other session might have proceeded
2486 : * with garbage collection in the no-lock-held interim of lock strength
2487 : * escalation. Check once more that this is actually necessary.
2488 : */
2489 0 : if (!need_gc_qtexts())
2490 0 : return;
2491 :
2492 : /*
2493 : * Load the old texts file. If we fail (out of memory, for instance),
2494 : * invalidate query texts. Hopefully this is rare. It might seem better
2495 : * to leave things alone on an OOM failure, but the problem is that the
2496 : * file is only going to get bigger; hoping for a future non-OOM result is
2497 : * risky and can easily lead to complete denial of service.
2498 : */
2499 0 : qbuffer = qtext_load_file(&qbuffer_size);
2500 0 : if (qbuffer == NULL)
2501 0 : goto gc_fail;
2502 :
2503 : /*
2504 : * We overwrite the query texts file in place, so as to reduce the risk of
2505 : * an out-of-disk-space failure. Since the file is guaranteed not to get
2506 : * larger, this should always work on traditional filesystems; though we
2507 : * could still lose on copy-on-write filesystems.
2508 : */
2509 0 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
2510 0 : if (qfile == NULL)
2511 : {
2512 0 : ereport(LOG,
2513 : (errcode_for_file_access(),
2514 : errmsg("could not write file \"%s\": %m",
2515 : PGSS_TEXT_FILE)));
2516 0 : goto gc_fail;
2517 : }
2518 :
2519 0 : extent = 0;
2520 0 : nentries = 0;
2521 :
2522 0 : hash_seq_init(&hash_seq, pgss_hash);
2523 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2524 : {
2525 0 : int query_len = entry->query_len;
2526 0 : char *qry = qtext_fetch(entry->query_offset,
2527 : query_len,
2528 : qbuffer,
2529 : qbuffer_size);
2530 :
2531 0 : if (qry == NULL)
2532 : {
2533 : /* Trouble ... drop the text */
2534 0 : entry->query_offset = 0;
2535 0 : entry->query_len = -1;
2536 : /* entry will not be counted in mean query length computation */
2537 0 : continue;
2538 : }
2539 :
2540 0 : if (fwrite(qry, 1, query_len + 1, qfile) != query_len + 1)
2541 : {
2542 0 : ereport(LOG,
2543 : (errcode_for_file_access(),
2544 : errmsg("could not write file \"%s\": %m",
2545 : PGSS_TEXT_FILE)));
2546 0 : hash_seq_term(&hash_seq);
2547 0 : goto gc_fail;
2548 : }
2549 :
2550 0 : entry->query_offset = extent;
2551 0 : extent += query_len + 1;
2552 0 : nentries++;
2553 : }
2554 :
2555 : /*
2556 : * Truncate away any now-unused space. If this fails for some odd reason,
2557 : * we log it, but there's no need to fail.
2558 : */
2559 0 : if (ftruncate(fileno(qfile), extent) != 0)
2560 0 : ereport(LOG,
2561 : (errcode_for_file_access(),
2562 : errmsg("could not truncate file \"%s\": %m",
2563 : PGSS_TEXT_FILE)));
2564 :
2565 0 : if (FreeFile(qfile))
2566 : {
2567 0 : ereport(LOG,
2568 : (errcode_for_file_access(),
2569 : errmsg("could not write file \"%s\": %m",
2570 : PGSS_TEXT_FILE)));
2571 0 : qfile = NULL;
2572 0 : goto gc_fail;
2573 : }
2574 :
2575 0 : elog(DEBUG1, "pgss gc of queries file shrunk size from %zu to %zu",
2576 : pgss->extent, extent);
2577 :
2578 : /* Reset the shared extent pointer */
2579 0 : pgss->extent = extent;
2580 :
2581 : /*
2582 : * Also update the mean query length, to be sure that need_gc_qtexts()
2583 : * won't still think we have a problem.
2584 : */
2585 0 : if (nentries > 0)
2586 0 : pgss->mean_query_len = extent / nentries;
2587 : else
2588 0 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
2589 :
2590 0 : pfree(qbuffer);
2591 :
2592 : /*
2593 : * OK, count a garbage collection cycle. (Note: even though we have
2594 : * exclusive lock on pgss->lock, we must take pgss->mutex for this, since
2595 : * other processes may examine gc_count while holding only the mutex.
2596 : * Also, we have to advance the count *after* we've rewritten the file,
2597 : * else other processes might not realize they read a stale file.)
2598 : */
2599 0 : record_gc_qtexts();
2600 :
2601 0 : return;
2602 :
2603 0 : gc_fail:
2604 : /* clean up resources */
2605 0 : if (qfile)
2606 0 : FreeFile(qfile);
2607 0 : if (qbuffer)
2608 0 : pfree(qbuffer);
2609 :
2610 : /*
2611 : * Since the contents of the external file are now uncertain, mark all
2612 : * hashtable entries as having invalid texts.
2613 : */
2614 0 : hash_seq_init(&hash_seq, pgss_hash);
2615 0 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2616 : {
2617 0 : entry->query_offset = 0;
2618 0 : entry->query_len = -1;
2619 : }
2620 :
2621 : /*
2622 : * Destroy the query text file and create a new, empty one
2623 : */
2624 0 : (void) unlink(PGSS_TEXT_FILE);
2625 0 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
2626 0 : if (qfile == NULL)
2627 0 : ereport(LOG,
2628 : (errcode_for_file_access(),
2629 : errmsg("could not recreate file \"%s\": %m",
2630 : PGSS_TEXT_FILE)));
2631 : else
2632 0 : FreeFile(qfile);
2633 :
2634 : /* Reset the shared extent pointer */
2635 0 : pgss->extent = 0;
2636 :
2637 : /* Reset mean_query_len to match the new state */
2638 0 : pgss->mean_query_len = ASSUMED_LENGTH_INIT;
2639 :
2640 : /*
2641 : * Bump the GC count even though we failed.
2642 : *
2643 : * This is needed to make concurrent readers of file without any lock on
2644 : * pgss->lock notice existence of new version of file. Once readers
2645 : * subsequently observe a change in GC count with pgss->lock held, that
2646 : * forces a safe reopen of file. Writers also require that we bump here,
2647 : * of course. (As required by locking protocol, readers and writers don't
2648 : * trust earlier file contents until gc_count is found unchanged after
2649 : * pgss->lock acquired in shared or exclusive mode respectively.)
2650 : */
2651 0 : record_gc_qtexts();
2652 : }
2653 :
2654 : #define SINGLE_ENTRY_RESET(e) \
2655 : if (e) { \
2656 : if (minmax_only) { \
2657 : /* When requested reset only min/max statistics of an entry */ \
2658 : for (int kind = 0; kind < PGSS_NUMKIND; kind++) \
2659 : { \
2660 : e->counters.max_time[kind] = 0; \
2661 : e->counters.min_time[kind] = 0; \
2662 : } \
2663 : e->minmax_stats_since = stats_reset; \
2664 : } \
2665 : else \
2666 : { \
2667 : /* Remove the key otherwise */ \
2668 : hash_search(pgss_hash, &e->key, HASH_REMOVE, NULL); \
2669 : num_remove++; \
2670 : } \
2671 : }
2672 :
2673 : /*
2674 : * Reset entries corresponding to parameters passed.
2675 : */
2676 : static TimestampTz
2677 121 : entry_reset(Oid userid, Oid dbid, int64 queryid, bool minmax_only)
2678 : {
2679 : HASH_SEQ_STATUS hash_seq;
2680 : pgssEntry *entry;
2681 : FILE *qfile;
2682 : int64 num_entries;
2683 121 : int64 num_remove = 0;
2684 : pgssHashKey key;
2685 : TimestampTz stats_reset;
2686 :
2687 121 : if (!pgss || !pgss_hash)
2688 0 : ereport(ERROR,
2689 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2690 : errmsg("pg_stat_statements must be loaded via \"shared_preload_libraries\"")));
2691 :
2692 121 : LWLockAcquire(&pgss->lock.lock, LW_EXCLUSIVE);
2693 121 : num_entries = hash_get_num_entries(pgss_hash);
2694 :
2695 121 : stats_reset = GetCurrentTimestamp();
2696 :
2697 121 : if (userid != 0 && dbid != 0 && queryid != INT64CONST(0))
2698 : {
2699 : /* If all the parameters are available, use the fast path. */
2700 1 : memset(&key, 0, sizeof(pgssHashKey));
2701 1 : key.userid = userid;
2702 1 : key.dbid = dbid;
2703 1 : key.queryid = queryid;
2704 :
2705 : /*
2706 : * Reset the entry if it exists, starting with the non-top-level
2707 : * entry.
2708 : */
2709 1 : key.toplevel = false;
2710 1 : entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
2711 :
2712 1 : SINGLE_ENTRY_RESET(entry);
2713 :
2714 : /* Also reset the top-level entry if it exists. */
2715 1 : key.toplevel = true;
2716 1 : entry = (pgssEntry *) hash_search(pgss_hash, &key, HASH_FIND, NULL);
2717 :
2718 1 : SINGLE_ENTRY_RESET(entry);
2719 : }
2720 120 : else if (userid != 0 || dbid != 0 || queryid != INT64CONST(0))
2721 : {
2722 : /* Reset entries corresponding to valid parameters. */
2723 4 : hash_seq_init(&hash_seq, pgss_hash);
2724 51 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2725 : {
2726 47 : if ((!userid || entry->key.userid == userid) &&
2727 36 : (!dbid || entry->key.dbid == dbid) &&
2728 34 : (!queryid || entry->key.queryid == queryid))
2729 : {
2730 7 : SINGLE_ENTRY_RESET(entry);
2731 : }
2732 : }
2733 : }
2734 : else
2735 : {
2736 : /* Reset all entries. */
2737 116 : hash_seq_init(&hash_seq, pgss_hash);
2738 1151 : while ((entry = hash_seq_search(&hash_seq)) != NULL)
2739 : {
2740 941 : SINGLE_ENTRY_RESET(entry);
2741 : }
2742 : }
2743 :
2744 : /* All entries are removed? */
2745 121 : if (num_entries != num_remove)
2746 6 : goto release_lock;
2747 :
2748 : /*
2749 : * Reset global statistics for pg_stat_statements since all entries are
2750 : * removed.
2751 : */
2752 115 : SpinLockAcquire(&pgss->mutex);
2753 115 : pgss->stats.dealloc = 0;
2754 115 : pgss->stats.stats_reset = stats_reset;
2755 115 : SpinLockRelease(&pgss->mutex);
2756 :
2757 : /*
2758 : * Write new empty query file, perhaps even creating a new one to recover
2759 : * if the file was missing.
2760 : */
2761 115 : qfile = AllocateFile(PGSS_TEXT_FILE, PG_BINARY_W);
2762 115 : if (qfile == NULL)
2763 : {
2764 0 : ereport(LOG,
2765 : (errcode_for_file_access(),
2766 : errmsg("could not create file \"%s\": %m",
2767 : PGSS_TEXT_FILE)));
2768 0 : goto done;
2769 : }
2770 :
2771 : /* If ftruncate fails, log it, but it's not a fatal problem */
2772 115 : if (ftruncate(fileno(qfile), 0) != 0)
2773 0 : ereport(LOG,
2774 : (errcode_for_file_access(),
2775 : errmsg("could not truncate file \"%s\": %m",
2776 : PGSS_TEXT_FILE)));
2777 :
2778 115 : FreeFile(qfile);
2779 :
2780 115 : done:
2781 115 : pgss->extent = 0;
2782 : /* This counts as a query text garbage collection for our purposes */
2783 115 : record_gc_qtexts();
2784 :
2785 121 : release_lock:
2786 121 : LWLockRelease(&pgss->lock.lock);
2787 :
2788 121 : return stats_reset;
2789 : }
2790 :
2791 : /*
2792 : * Generate a normalized version of the query string that will be used to
2793 : * represent all similar queries.
2794 : *
2795 : * Note that the normalized representation may well vary depending on
2796 : * just which "equivalent" query is used to create the hashtable entry.
2797 : * We assume this is OK.
2798 : *
2799 : * If query_loc > 0, then "query" has been advanced by that much compared to
2800 : * the original string start, so we need to translate the provided locations
2801 : * to compensate. (This lets us avoid re-scanning statements before the one
2802 : * of interest, so it's worth doing.)
2803 : *
2804 : * *query_len_p contains the input string length, and is updated with
2805 : * the result string length on exit. The resulting string might be longer
2806 : * or shorter depending on what happens with replacement of constants.
2807 : *
2808 : * Returns a palloc'd string.
2809 : */
2810 : static char *
2811 11403 : generate_normalized_query(const JumbleState *jstate, const char *query,
2812 : int query_loc, int *query_len_p)
2813 : {
2814 : char *norm_query;
2815 11403 : int query_len = *query_len_p;
2816 : int norm_query_buflen, /* Space allowed for norm_query */
2817 : len_to_wrt, /* Length (in bytes) to write */
2818 11403 : quer_loc = 0, /* Source query byte location */
2819 11403 : n_quer_loc = 0, /* Normalized query byte location */
2820 11403 : last_off = 0, /* Offset from start for previous tok */
2821 11403 : last_tok_len = 0; /* Length (in bytes) of that tok */
2822 11403 : int num_constants_replaced = 0;
2823 11403 : LocationLen *locs = NULL;
2824 :
2825 : /*
2826 : * Determine constants' lengths (core system only gives us locations), and
2827 : * return a sorted copy of jstate's LocationLen data with lengths filled
2828 : * in.
2829 : */
2830 11403 : locs = ComputeConstantLengths(jstate, query, query_loc);
2831 :
2832 : /*
2833 : * Allow for $n symbols to be longer than the constants they replace.
2834 : * Constants must take at least one byte in text form, while a $n symbol
2835 : * certainly isn't more than 11 bytes, even if n reaches INT_MAX. We
2836 : * could refine that limit based on the max value of n for the current
2837 : * query, but it hardly seems worth any extra effort to do so.
2838 : */
2839 11403 : norm_query_buflen = query_len + jstate->clocations_count * 10;
2840 :
2841 : /* Allocate result buffer */
2842 11403 : norm_query = palloc(norm_query_buflen + 1);
2843 :
2844 45757 : for (int i = 0; i < jstate->clocations_count; i++)
2845 : {
2846 : int off, /* Offset from start for cur tok */
2847 : tok_len; /* Length (in bytes) of that tok */
2848 :
2849 : /*
2850 : * If we have an external param at this location, but no lists are
2851 : * being squashed across the query, then we skip here; this will make
2852 : * us print the characters found in the original query that represent
2853 : * the parameter in the next iteration (or after the loop is done),
2854 : * which is a bit odd but seems to work okay in most cases.
2855 : */
2856 34354 : if (locs[i].extern_param && !jstate->has_squashed_lists)
2857 162 : continue;
2858 :
2859 34192 : off = locs[i].location;
2860 :
2861 : /* Adjust recorded location if we're dealing with partial string */
2862 34192 : off -= query_loc;
2863 :
2864 34192 : tok_len = locs[i].length;
2865 :
2866 34192 : if (tok_len < 0)
2867 506 : continue; /* ignore any duplicates */
2868 :
2869 : /* Copy next chunk (what precedes the next constant) */
2870 33686 : len_to_wrt = off - last_off;
2871 33686 : len_to_wrt -= last_tok_len;
2872 : Assert(len_to_wrt >= 0);
2873 33686 : memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
2874 33686 : n_quer_loc += len_to_wrt;
2875 :
2876 : /*
2877 : * And insert a param symbol in place of the constant token; and, if
2878 : * we have a squashable list, insert a placeholder comment starting
2879 : * from the list's second value.
2880 : */
2881 33686 : n_quer_loc += sprintf(norm_query + n_quer_loc, "$%d%s",
2882 33686 : num_constants_replaced + 1 + jstate->highest_extern_param_id,
2883 33686 : locs[i].squashed ? " /*, ... */" : "");
2884 33686 : num_constants_replaced++;
2885 :
2886 : /* move forward */
2887 33686 : quer_loc = off + tok_len;
2888 33686 : last_off = off;
2889 33686 : last_tok_len = tok_len;
2890 : }
2891 :
2892 : /* Clean up, if needed */
2893 11403 : if (locs)
2894 11403 : pfree(locs);
2895 :
2896 : /*
2897 : * We've copied up until the last ignorable constant. Copy over the
2898 : * remaining bytes of the original query string.
2899 : */
2900 11403 : len_to_wrt = query_len - quer_loc;
2901 :
2902 : Assert(len_to_wrt >= 0);
2903 11403 : memcpy(norm_query + n_quer_loc, query + quer_loc, len_to_wrt);
2904 11403 : n_quer_loc += len_to_wrt;
2905 :
2906 : Assert(n_quer_loc <= norm_query_buflen);
2907 11403 : norm_query[n_quer_loc] = '\0';
2908 :
2909 11403 : *query_len_p = n_quer_loc;
2910 11403 : return norm_query;
2911 : }
|
