1PGBENCH(1) PostgreSQL 13.3 Documentation PGBENCH(1)
2
3
4
6 pgbench - run a benchmark test on PostgreSQL
7
9 pgbench -i [option...] [dbname]
10
11 pgbench [option...] [dbname]
12
14 pgbench is a simple program for running benchmark tests on PostgreSQL.
15 It runs the same sequence of SQL commands over and over, possibly in
16 multiple concurrent database sessions, and then calculates the average
17 transaction rate (transactions per second). By default, pgbench tests a
18 scenario that is loosely based on TPC-B, involving five SELECT, UPDATE,
19 and INSERT commands per transaction. However, it is easy to test other
20 cases by writing your own transaction script files.
21
22 Typical output from pgbench looks like:
23
24 transaction type: <builtin: TPC-B (sort of)>
25 scaling factor: 10
26 query mode: simple
27 number of clients: 10
28 number of threads: 1
29 number of transactions per client: 1000
30 number of transactions actually processed: 10000/10000
31 tps = 85.184871 (including connections establishing)
32 tps = 85.296346 (excluding connections establishing)
33
34 The first six lines report some of the most important parameter
35 settings. The next line reports the number of transactions completed
36 and intended (the latter being just the product of number of clients
37 and number of transactions per client); these will be equal unless the
38 run failed before completion. (In -T mode, only the actual number of
39 transactions is printed.) The last two lines report the number of
40 transactions per second, figured with and without counting the time to
41 start database sessions.
42
43 The default TPC-B-like transaction test requires specific tables to be
44 set up beforehand. pgbench should be invoked with the -i (initialize)
45 option to create and populate these tables. (When you are testing a
46 custom script, you don't need this step, but will instead need to do
47 whatever setup your test needs.) Initialization looks like:
48
49 pgbench -i [ other-options ] dbname
50
51 where dbname is the name of the already-created database to test in.
52 (You may also need -h, -p, and/or -U options to specify how to connect
53 to the database server.)
54
55 Caution
56 pgbench -i creates four tables pgbench_accounts, pgbench_branches,
57 pgbench_history, and pgbench_tellers, destroying any existing
58 tables of these names. Be very careful to use another database if
59 you have tables having these names!
60
61 At the default “scale factor” of 1, the tables initially contain this
62 many rows:
63
64 table # of rows
65 ---------------------------------
66 pgbench_branches 1
67 pgbench_tellers 10
68 pgbench_accounts 100000
69 pgbench_history 0
70
71 You can (and, for most purposes, probably should) increase the number
72 of rows by using the -s (scale factor) option. The -F (fillfactor)
73 option might also be used at this point.
74
75 Once you have done the necessary setup, you can run your benchmark with
76 a command that doesn't include -i, that is
77
78 pgbench [ options ] dbname
79
80 In nearly all cases, you'll need some options to make a useful test.
81 The most important options are -c (number of clients), -t (number of
82 transactions), -T (time limit), and -f (specify a custom script file).
83 See below for a full list.
84
86 The following is divided into three subsections. Different options are
87 used during database initialization and while running benchmarks, but
88 some options are useful in both cases.
89
90 Initialization Options
91 pgbench accepts the following command-line initialization arguments:
92
93 dbname
94 Specifies the name of the database to test in. If this is not
95 specified, the environment variable PGDATABASE is used. If that is
96 not set, the user name specified for the connection is used.
97
98 -i
99 --initialize
100 Required to invoke initialization mode.
101
102 -I init_steps
103 --init-steps=init_steps
104 Perform just a selected set of the normal initialization steps.
105 init_steps specifies the initialization steps to be performed,
106 using one character per step. Each step is invoked in the specified
107 order. The default is dtgvp. The available steps are:
108
109 d (Drop)
110 Drop any existing pgbench tables.
111
112 t (create Tables)
113 Create the tables used by the standard pgbench scenario, namely
114 pgbench_accounts, pgbench_branches, pgbench_history, and
115 pgbench_tellers.
116
117 g or G (Generate data, client-side or server-side)
118 Generate data and load it into the standard tables, replacing
119 any data already present.
120
121 With g (client-side data generation), data is generated in
122 pgbench client and then sent to the server. This uses the
123 client/server bandwidth extensively through a COPY. Using g
124 causes logging to print one message every 100,000 rows while
125 generating data for the pgbench_accounts table.
126
127 With G (server-side data generation), only small queries are
128 sent from the pgbench client and then data is actually
129 generated in the server. No significant bandwidth is required
130 for this variant, but the server will do more work. Using G
131 causes logging not to print any progress message while
132 generating data.
133
134 The default initialization behavior uses client-side data
135 generation (equivalent to g).
136
137 v (Vacuum)
138 Invoke VACUUM on the standard tables.
139
140 p (create Primary keys)
141 Create primary key indexes on the standard tables.
142
143 f (create Foreign keys)
144 Create foreign key constraints between the standard tables.
145 (Note that this step is not performed by default.)
146
147 -F fillfactor
148 --fillfactor=fillfactor
149 Create the pgbench_accounts, pgbench_tellers and pgbench_branches
150 tables with the given fillfactor. Default is 100.
151
152 -n
153 --no-vacuum
154 Perform no vacuuming during initialization. (This option suppresses
155 the v initialization step, even if it was specified in -I.)
156
157 -q
158 --quiet
159 Switch logging to quiet mode, producing only one progress message
160 per 5 seconds. The default logging prints one message each 100,000
161 rows, which often outputs many lines per second (especially on good
162 hardware).
163
164 This setting has no effect if G is specified in -I.
165
166 -s scale_factor
167 --scale=scale_factor
168 Multiply the number of rows generated by the scale factor. For
169 example, -s 100 will create 10,000,000 rows in the pgbench_accounts
170 table. Default is 1. When the scale is 20,000 or larger, the
171 columns used to hold account identifiers (aid columns) will switch
172 to using larger integers (bigint), in order to be big enough to
173 hold the range of account identifiers.
174
175 --foreign-keys
176 Create foreign key constraints between the standard tables. (This
177 option adds the f step to the initialization step sequence, if it
178 is not already present.)
179
180 --index-tablespace=index_tablespace
181 Create indexes in the specified tablespace, rather than the default
182 tablespace.
183
184 --partition-method=NAME
185 Create a partitioned pgbench_accounts table with NAME method.
186 Expected values are range or hash. This option requires that
187 --partitions is set to non-zero. If unspecified, default is range.
188
189 --partitions=NUM
190 Create a partitioned pgbench_accounts table with NUM partitions of
191 nearly equal size for the scaled number of accounts. Default is 0,
192 meaning no partitioning.
193
194 --tablespace=tablespace
195 Create tables in the specified tablespace, rather than the default
196 tablespace.
197
198 --unlogged-tables
199 Create all tables as unlogged tables, rather than permanent tables.
200
201 Benchmarking Options
202 pgbench accepts the following command-line benchmarking arguments:
203
204 -b scriptname[@weight]
205 --builtin=scriptname[@weight]
206 Add the specified built-in script to the list of scripts to be
207 executed. Available built-in scripts are: tpcb-like, simple-update
208 and select-only. Unambiguous prefixes of built-in names are
209 accepted. With the special name list, show the list of built-in
210 scripts and exit immediately.
211
212 Optionally, write an integer weight after @ to adjust the
213 probability of selecting this script versus other ones. The default
214 weight is 1. See below for details.
215
216 -c clients
217 --client=clients
218 Number of clients simulated, that is, number of concurrent database
219 sessions. Default is 1.
220
221 -C
222 --connect
223 Establish a new connection for each transaction, rather than doing
224 it just once per client session. This is useful to measure the
225 connection overhead.
226
227 -d
228 --debug
229 Print debugging output.
230
231 -D varname=value
232 --define=varname=value
233 Define a variable for use by a custom script (see below). Multiple
234 -D options are allowed.
235
236 -f filename[@weight]
237 --file=filename[@weight]
238 Add a transaction script read from filename to the list of scripts
239 to be executed.
240
241 Optionally, write an integer weight after @ to adjust the
242 probability of selecting this script versus other ones. The default
243 weight is 1. (To use a script file name that includes an @
244 character, append a weight so that there is no ambiguity, for
245 example filen@me@1.) See below for details.
246
247 -j threads
248 --jobs=threads
249 Number of worker threads within pgbench. Using more than one thread
250 can be helpful on multi-CPU machines. Clients are distributed as
251 evenly as possible among available threads. Default is 1.
252
253 -l
254 --log
255 Write information about each transaction to a log file. See below
256 for details.
257
258 -L limit
259 --latency-limit=limit
260 Transactions that last more than limit milliseconds are counted and
261 reported separately, as late.
262
263 When throttling is used (--rate=...), transactions that lag behind
264 schedule by more than limit ms, and thus have no hope of meeting
265 the latency limit, are not sent to the server at all. They are
266 counted and reported separately as skipped.
267
268 -M querymode
269 --protocol=querymode
270 Protocol to use for submitting queries to the server:
271
272 • simple: use simple query protocol.
273
274 • extended: use extended query protocol.
275
276 • prepared: use extended query protocol with prepared statements.
277
278 In the prepared mode, pgbench reuses the parse analysis result
279 starting from the second query iteration, so pgbench runs faster
280 than in other modes.
281
282 The default is simple query protocol. (See Chapter 52 for more
283 information.)
284
285 -n
286 --no-vacuum
287 Perform no vacuuming before running the test. This option is
288 necessary if you are running a custom test scenario that does not
289 include the standard tables pgbench_accounts, pgbench_branches,
290 pgbench_history, and pgbench_tellers.
291
292 -N
293 --skip-some-updates
294 Run built-in simple-update script. Shorthand for -b simple-update.
295
296 -P sec
297 --progress=sec
298 Show progress report every sec seconds. The report includes the
299 time since the beginning of the run, the TPS since the last report,
300 and the transaction latency average and standard deviation since
301 the last report. Under throttling (-R), the latency is computed
302 with respect to the transaction scheduled start time, not the
303 actual transaction beginning time, thus it also includes the
304 average schedule lag time.
305
306 -r
307 --report-latencies
308 Report the average per-statement latency (execution time from the
309 perspective of the client) of each command after the benchmark
310 finishes. See below for details.
311
312 -R rate
313 --rate=rate
314 Execute transactions targeting the specified rate instead of
315 running as fast as possible (the default). The rate is given in
316 transactions per second. If the targeted rate is above the maximum
317 possible rate, the rate limit won't impact the results.
318
319 The rate is targeted by starting transactions along a
320 Poisson-distributed schedule time line. The expected start time
321 schedule moves forward based on when the client first started, not
322 when the previous transaction ended. That approach means that when
323 transactions go past their original scheduled end time, it is
324 possible for later ones to catch up again.
325
326 When throttling is active, the transaction latency reported at the
327 end of the run is calculated from the scheduled start times, so it
328 includes the time each transaction had to wait for the previous
329 transaction to finish. The wait time is called the schedule lag
330 time, and its average and maximum are also reported separately. The
331 transaction latency with respect to the actual transaction start
332 time, i.e., the time spent executing the transaction in the
333 database, can be computed by subtracting the schedule lag time from
334 the reported latency.
335
336 If --latency-limit is used together with --rate, a transaction can
337 lag behind so much that it is already over the latency limit when
338 the previous transaction ends, because the latency is calculated
339 from the scheduled start time. Such transactions are not sent to
340 the server, but are skipped altogether and counted separately.
341
342 A high schedule lag time is an indication that the system cannot
343 process transactions at the specified rate, with the chosen number
344 of clients and threads. When the average transaction execution time
345 is longer than the scheduled interval between each transaction,
346 each successive transaction will fall further behind, and the
347 schedule lag time will keep increasing the longer the test run is.
348 When that happens, you will have to reduce the specified
349 transaction rate.
350
351 -s scale_factor
352 --scale=scale_factor
353 Report the specified scale factor in pgbench's output. With the
354 built-in tests, this is not necessary; the correct scale factor
355 will be detected by counting the number of rows in the
356 pgbench_branches table. However, when testing only custom
357 benchmarks (-f option), the scale factor will be reported as 1
358 unless this option is used.
359
360 -S
361 --select-only
362 Run built-in select-only script. Shorthand for -b select-only.
363
364 -t transactions
365 --transactions=transactions
366 Number of transactions each client runs. Default is 10.
367
368 -T seconds
369 --time=seconds
370 Run the test for this many seconds, rather than a fixed number of
371 transactions per client. -t and -T are mutually exclusive.
372
373 -v
374 --vacuum-all
375 Vacuum all four standard tables before running the test. With
376 neither -n nor -v, pgbench will vacuum the pgbench_tellers and
377 pgbench_branches tables, and will truncate pgbench_history.
378
379 --aggregate-interval=seconds
380 Length of aggregation interval (in seconds). May be used only with
381 -l option. With this option, the log contains per-interval summary
382 data, as described below.
383
384 --log-prefix=prefix
385 Set the filename prefix for the log files created by --log. The
386 default is pgbench_log.
387
388 --progress-timestamp
389 When showing progress (option -P), use a timestamp (Unix epoch)
390 instead of the number of seconds since the beginning of the run.
391 The unit is in seconds, with millisecond precision after the dot.
392 This helps compare logs generated by various tools.
393
394 --random-seed=seed
395 Set random generator seed. Seeds the system random number
396 generator, which then produces a sequence of initial generator
397 states, one for each thread. Values for seed may be: time (the
398 default, the seed is based on the current time), rand (use a strong
399 random source, failing if none is available), or an unsigned
400 decimal integer value. The random generator is invoked explicitly
401 from a pgbench script (random... functions) or implicitly (for
402 instance option --rate uses it to schedule transactions). When
403 explicitly set, the value used for seeding is shown on the
404 terminal. Any value allowed for seed may also be provided through
405 the environment variable PGBENCH_RANDOM_SEED. To ensure that the
406 provided seed impacts all possible uses, put this option first or
407 use the environment variable.
408
409 Setting the seed explicitly allows to reproduce a pgbench run
410 exactly, as far as random numbers are concerned. As the random
411 state is managed per thread, this means the exact same pgbench run
412 for an identical invocation if there is one client per thread and
413 there are no external or data dependencies. From a statistical
414 viewpoint reproducing runs exactly is a bad idea because it can
415 hide the performance variability or improve performance unduly,
416 e.g., by hitting the same pages as a previous run. However, it may
417 also be of great help for debugging, for instance re-running a
418 tricky case which leads to an error. Use wisely.
419
420 --sampling-rate=rate
421 Sampling rate, used when writing data into the log, to reduce the
422 amount of log generated. If this option is given, only the
423 specified fraction of transactions are logged. 1.0 means all
424 transactions will be logged, 0.05 means only 5% of the transactions
425 will be logged.
426
427 Remember to take the sampling rate into account when processing the
428 log file. For example, when computing TPS values, you need to
429 multiply the numbers accordingly (e.g., with 0.01 sample rate,
430 you'll only get 1/100 of the actual TPS).
431
432 --show-script=scriptname
433 Show the actual code of builtin script scriptname on stderr, and
434 exit immediately.
435
436 Common Options
437 pgbench also accepts the following common command-line arguments for
438 connection parameters:
439
440 -h hostname
441 --host=hostname
442 The database server's host name
443
444 -p port
445 --port=port
446 The database server's port number
447
448 -U login
449 --username=login
450 The user name to connect as
451
452 -V
453 --version
454 Print the pgbench version and exit.
455
456 -?
457 --help
458 Show help about pgbench command line arguments, and exit.
459
461 A successful run will exit with status 0. Exit status 1 indicates
462 static problems such as invalid command-line options. Errors during the
463 run such as database errors or problems in the script will result in
464 exit status 2. In the latter case, pgbench will print partial results.
465
467 PGDATABASE
468 PGHOST
469 PGPORT
470 PGUSER
471 Default connection parameters.
472
473 This utility, like most other PostgreSQL utilities, uses the
474 environment variables supported by libpq (see Section 33.14).
475
476 The environment variable PG_COLOR specifies whether to use color in
477 diagnostic messages. Possible values are always, auto and never.
478
480 What Is the “Transaction” Actually Performed in pgbench?
481 pgbench executes test scripts chosen randomly from a specified list.
482 The scripts may include built-in scripts specified with -b and
483 user-provided scripts specified with -f. Each script may be given a
484 relative weight specified after an @ so as to change its selection
485 probability. The default weight is 1. Scripts with a weight of 0 are
486 ignored.
487
488 The default built-in transaction script (also invoked with -b
489 tpcb-like) issues seven commands per transaction over randomly chosen
490 aid, tid, bid and delta. The scenario is inspired by the TPC-B
491 benchmark, but is not actually TPC-B, hence the name.
492
493 1. BEGIN;
494
495 2. UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid
496 = :aid;
497
498 3. SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
499
500 4. UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid =
501 :tid;
502
503 5. UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid
504 = :bid;
505
506 6. INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES
507 (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
508
509 7. END;
510
511 If you select the simple-update built-in (also -N), steps 4 and 5
512 aren't included in the transaction. This will avoid update contention
513 on these tables, but it makes the test case even less like TPC-B.
514
515 If you select the select-only built-in (also -S), only the SELECT is
516 issued.
517
518 Custom Scripts
519 pgbench has support for running custom benchmark scenarios by replacing
520 the default transaction script (described above) with a transaction
521 script read from a file (-f option). In this case a “transaction”
522 counts as one execution of a script file.
523
524 A script file contains one or more SQL commands terminated by
525 semicolons. Empty lines and lines beginning with -- are ignored. Script
526 files can also contain “meta commands”, which are interpreted by
527 pgbench itself, as described below.
528
529 Note
530 Before PostgreSQL 9.6, SQL commands in script files were terminated
531 by newlines, and so they could not be continued across lines. Now a
532 semicolon is required to separate consecutive SQL commands (though
533 a SQL command does not need one if it is followed by a meta
534 command). If you need to create a script file that works with both
535 old and new versions of pgbench, be sure to write each SQL command
536 on a single line ending with a semicolon.
537
538 There is a simple variable-substitution facility for script files.
539 Variable names must consist of letters (including non-Latin letters),
540 digits, and underscores, with the first character not being a digit.
541 Variables can be set by the command-line -D option, explained above, or
542 by the meta commands explained below. In addition to any variables
543 preset by -D command-line options, there are a few variables that are
544 preset automatically, listed in Table 273. A value specified for these
545 variables using -D takes precedence over the automatic presets. Once
546 set, a variable's value can be inserted into a SQL command by writing
547 :variablename. When running more than one client session, each session
548 has its own set of variables. pgbench supports up to 255 variable uses
549 in one statement.
550
551 Table 273. pgbench Automatic Variables
552 ┌─────────────┬────────────────────────────┐
553 │Variable │ Description │
554 ├─────────────┼────────────────────────────┤
555 │client_id │ unique number identifying │
556 │ │ the client session (starts │
557 │ │ from zero) │
558 ├─────────────┼────────────────────────────┤
559 │default_seed │ seed used in hash │
560 │ │ functions by default │
561 ├─────────────┼────────────────────────────┤
562 │random_seed │ random generator seed │
563 │ │ (unless overwritten with │
564 │ │ -D) │
565 ├─────────────┼────────────────────────────┤
566 │scale │ current scale factor │
567 └─────────────┴────────────────────────────┘
568
569 Script file meta commands begin with a backslash (\) and normally
570 extend to the end of the line, although they can be continued to
571 additional lines by writing backslash-return. Arguments to a meta
572 command are separated by white space. These meta commands are
573 supported:
574
575 \gset [prefix] \aset [prefix]
576 These commands may be used to end SQL queries, taking the place of
577 the terminating semicolon (;).
578
579 When the \gset command is used, the preceding SQL query is expected
580 to return one row, the columns of which are stored into variables
581 named after column names, and prefixed with prefix if provided.
582
583 When the \aset command is used, all combined SQL queries (separated
584 by \;) have their columns stored into variables named after column
585 names, and prefixed with prefix if provided. If a query returns no
586 row, no assignment is made and the variable can be tested for
587 existence to detect this. If a query returns more than one row, the
588 last value is kept.
589
590 The following example puts the final account balance from the first
591 query into variable abalance, and fills variables p_two and p_three
592 with integers from the third query. The result of the second query
593 is discarded. The result of the two last combined queries are
594 stored in variables four and five.
595
596 UPDATE pgbench_accounts
597 SET abalance = abalance + :delta
598 WHERE aid = :aid
599 RETURNING abalance \gset
600 -- compound of two queries
601 SELECT 1 \;
602 SELECT 2 AS two, 3 AS three \gset p_
603 SELECT 4 AS four \; SELECT 5 AS five \aset
604
605 \if expression
606 \elif expression
607 \else
608 \endif
609 This group of commands implements nestable conditional blocks,
610 similarly to psql's \if expression. Conditional expressions are
611 identical to those with \set, with non-zero values interpreted as
612 true.
613
614 \set varname expression
615 Sets variable varname to a value calculated from expression. The
616 expression may contain the NULL constant, Boolean constants TRUE
617 and FALSE, integer constants such as 5432, double constants such as
618 3.14159, references to variables :variablename, operators with
619 their usual SQL precedence and associativity, function calls, SQL
620 CASE generic conditional expressions and parentheses.
621
622 Functions and most operators return NULL on NULL input.
623
624 For conditional purposes, non zero numerical values are TRUE, zero
625 numerical values and NULL are FALSE.
626
627 Too large or small integer and double constants, as well as integer
628 arithmetic operators (+, -, * and /) raise errors on overflows.
629
630 When no final ELSE clause is provided to a CASE, the default value
631 is NULL.
632
633 Examples:
634
635 \set ntellers 10 * :scale
636 \set aid (1021 * random(1, 100000 * :scale)) % \
637 (100000 * :scale) + 1
638 \set divx CASE WHEN :x <> 0 THEN :y/:x ELSE NULL END
639
640 \sleep number [ us | ms | s ]
641 Causes script execution to sleep for the specified duration in
642 microseconds (us), milliseconds (ms) or seconds (s). If the unit is
643 omitted then seconds are the default. number can be either an
644 integer constant or a :variablename reference to a variable having
645 an integer value.
646
647 Example:
648
649 \sleep 10 ms
650
651 \setshell varname command [ argument ... ]
652 Sets variable varname to the result of the shell command command
653 with the given argument(s). The command must return an integer
654 value through its standard output.
655
656 command and each argument can be either a text constant or a
657 :variablename reference to a variable. If you want to use an
658 argument starting with a colon, write an additional colon at the
659 beginning of argument.
660
661 Example:
662
663 \setshell variable_to_be_assigned command literal_argument :variable ::literal_starting_with_colon
664
665 \shell command [ argument ... ]
666 Same as \setshell, but the result of the command is discarded.
667
668 Example:
669
670 \shell command literal_argument :variable ::literal_starting_with_colon
671
672 Built-in Operators
673 The arithmetic, bitwise, comparison and logical operators listed in
674 Table 274 are built into pgbench and may be used in expressions
675 appearing in \set. The operators are listed in increasing precedence
676 order. Except as noted, operators taking two numeric inputs will
677 produce a double value if either input is double, otherwise they
678 produce an integer result.
679
680 Table 274. pgbench Operators
681 ┌────────────────────────────────────────┐
682 │ │
683 │ Operator │
684 │ │
685 │ .PP Description │
686 │ │
687 │ .PP Example(s) │
688 ├────────────────────────────────────────┤
689 │ │
690 │ boolean OR boolean → boolean │
691 │ │
692 │ .PP Logical OR │
693 │ │
694 │ .PP 5 or 0 → TRUE │
695 ├────────────────────────────────────────┤
696 │ │
697 │ boolean AND boolean → boolean │
698 │ │
699 │ .PP Logical AND │
700 │ │
701 │ .PP 3 and 0 → FALSE │
702 ├────────────────────────────────────────┤
703 │ │
704 │ NOT boolean → boolean │
705 │ │
706 │ .PP Logical NOT │
707 │ │
708 │ .PP not false → TRUE │
709 ├────────────────────────────────────────┤
710 │ │
711 │ boolean IS [NOT] │
712 │ (NULL|TRUE|FALSE) → boolean │
713 │ │
714 │ .PP Boolean value tests │
715 │ │
716 │ .PP 1 is null → FALSE │
717 ├────────────────────────────────────────┤
718 │ │
719 │ value ISNULL|NOTNULL → boolean │
720 │ │
721 │ .PP Nullness tests │
722 │ │
723 │ .PP 1 notnull → TRUE │
724 ├────────────────────────────────────────┤
725 │ │
726 │ number = number → boolean │
727 │ │
728 │ .PP Equal │
729 │ │
730 │ .PP 5 = 4 → FALSE │
731 ├────────────────────────────────────────┤
732 │ │
733 │ number <> number → boolean │
734 │ │
735 │ .PP Not equal │
736 │ │
737 │ .PP 5 <> 4 → TRUE │
738 ├────────────────────────────────────────┤
739 │ │
740 │ number != number → boolean │
741 │ │
742 │ .PP Not equal │
743 │ │
744 │ .PP 5 != 5 → FALSE │
745 ├────────────────────────────────────────┤
746 │ │
747 │ number < number → boolean │
748 │ │
749 │ .PP Less than │
750 │ │
751 │ .PP 5 < 4 → FALSE │
752 ├────────────────────────────────────────┤
753 │ │
754 │ number <= number → boolean │
755 │ │
756 │ .PP Less than or equal to │
757 │ │
758 │ .PP 5 <= 4 → FALSE │
759 ├────────────────────────────────────────┤
760 │ │
761 │ number > number → boolean │
762 │ │
763 │ .PP Greater than │
764 │ │
765 │ .PP 5 > 4 → TRUE │
766 ├────────────────────────────────────────┤
767 │ │
768 │ number >= number → boolean │
769 │ │
770 │ .PP Greater than or equal │
771 │ to │
772 │ │
773 │ .PP 5 >= 4 → TRUE │
774 ├────────────────────────────────────────┤
775 │ │
776 │ integer | integer → integer │
777 │ │
778 │ .PP Bitwise OR │
779 │ │
780 │ .PP 1 | 2 → 3 │
781 ├────────────────────────────────────────┤
782 │ │
783 │ integer # integer → integer │
784 │ │
785 │ .PP Bitwise XOR │
786 │ │
787 │ .PP 1 # 3 → 2 │
788 ├────────────────────────────────────────┤
789 │ │
790 │ integer & integer → integer │
791 │ │
792 │ .PP Bitwise AND │
793 │ │
794 │ .PP 1 & 3 → 1 │
795 ├────────────────────────────────────────┤
796 │ │
797 │ ~ integer → integer │
798 │ │
799 │ .PP Bitwise NOT │
800 │ │
801 │ .PP ~ 1 → -2 │
802 ├────────────────────────────────────────┤
803 │ │
804 │ integer << integer → integer │
805 │ │
806 │ .PP Bitwise shift left │
807 │ │
808 │ .PP 1 << 2 → 4 │
809 ├────────────────────────────────────────┤
810 │ │
811 │ integer >> integer → integer │
812 │ │
813 │ .PP Bitwise shift right │
814 │ │
815 │ .PP 8 >> 2 → 2 │
816 ├────────────────────────────────────────┤
817 │ │
818 │ number + number → number │
819 │ │
820 │ .PP Addition │
821 │ │
822 │ .PP 5 + 4 → 9 │
823 ├────────────────────────────────────────┤
824 │ │
825 │ number - number → number │
826 │ │
827 │ .PP Subtraction │
828 │ │
829 │ .PP 3 - 2.0 → 1.0 │
830 ├────────────────────────────────────────┤
831 │ │
832 │ number * number → number │
833 │ │
834 │ .PP Multiplication │
835 │ │
836 │ .PP 5 * 4 → 20 │
837 ├────────────────────────────────────────┤
838 │ │
839 │ number / number → number │
840 │ │
841 │ .PP Division (truncates │
842 │ the result towards zero if both │
843 │ inputs are integers) │
844 │ │
845 │ .PP 5 / 3 → 1 │
846 ├────────────────────────────────────────┤
847 │ │
848 │ integer % integer → integer │
849 │ │
850 │ .PP Modulo (remainder) │
851 │ │
852 │ .PP 3 % 2 → 1 │
853 ├────────────────────────────────────────┤
854 │ │
855 │ - number → number │
856 │ │
857 │ .PP Negation │
858 │ │
859 │ .PP - 2.0 → -2.0 │
860 └────────────────────────────────────────┘
861
862 Built-In Functions
863 The functions listed in Table 275 are built into pgbench and may be
864 used in expressions appearing in \set.
865
866 Table 275. pgbench Functions
867 ┌────────────────────────────────────────┐
868 │ │
869 │ Function │
870 │ │
871 │ .PP Description │
872 │ │
873 │ .PP Example(s) │
874 ├────────────────────────────────────────┤
875 │ │
876 │ abs ( number ) → same type as │
877 │ input │
878 │ │
879 │ .PP Absolute value │
880 │ │
881 │ .PP abs(-17) → 17 │
882 ├────────────────────────────────────────┤
883 │ │
884 │ debug ( number ) → same type as │
885 │ input │
886 │ │
887 │ .PP Prints the argument │
888 │ to stderr, and returns the │
889 │ argument. │
890 │ │
891 │ .PP debug(5432.1) → │
892 │ 5432.1 │
893 ├────────────────────────────────────────┤
894 │ │
895 │ double ( number ) → double │
896 │ │
897 │ .PP Casts to double. │
898 │ │
899 │ .PP double(5432) → 5432.0 │
900 ├────────────────────────────────────────┤
901 │ │
902 │ exp ( number ) → double │
903 │ │
904 │ .PP Exponential (e raised │
905 │ to the given power) │
906 │ │
907 │ .PP exp(1.0) → │
908 │ 2.718281828459045 │
909 ├────────────────────────────────────────┤
910 │ │
911 │ greatest ( number [, ... ] ) → │
912 │ double if any argument is │
913 │ double, else integer │
914 │ │
915 │ .PP Selects the largest │
916 │ value among the arguments. │
917 │ │
918 │ .PP greatest(5, 4, 3, 2) │
919 │ → 5 │
920 ├────────────────────────────────────────┤
921 │ │
922 │ hash ( value [, seed ] ) → │
923 │ integer │
924 │ │
925 │ .PP This is an alias for │
926 │ hash_murmur2. │
927 │ │
928 │ .PP hash(10, 5432) → │
929 │ -5817877081768721676 │
930 ├────────────────────────────────────────┤
931 │ │
932 │ hash_fnv1a ( value [, seed ] ) → │
933 │ integer │
934 │ │
935 │ .PP Computes FNV-1a hash. │
936 │ │
937 │ .PP hash_fnv1a(10, 5432) │
938 │ → -7793829335365542153 │
939 ├────────────────────────────────────────┤
940 │ │
941 │ hash_murmur2 ( value [, seed ] ) │
942 │ → integer │
943 │ │
944 │ .PP Computes MurmurHash2 │
945 │ hash. │
946 │ │
947 │ .PP hash_murmur2(10, │
948 │ 5432) → -5817877081768721676 │
949 ├────────────────────────────────────────┤
950 │ │
951 │ int ( number ) → integer │
952 │ │
953 │ .PP Casts to integer. │
954 │ │
955 │ .PP int(5.4 + 3.8) → 9 │
956 ├────────────────────────────────────────┤
957 │ │
958 │ least ( number [, ... ] ) → │
959 │ double if any argument is │
960 │ double, else integer │
961 │ │
962 │ .PP Selects the smallest │
963 │ value among the arguments. │
964 │ │
965 │ .PP least(5, 4, 3, 2.1) → │
966 │ 2.1 │
967 ├────────────────────────────────────────┤
968 │ │
969 │ ln ( number ) → double │
970 │ │
971 │ .PP Natural logarithm │
972 │ │
973 │ .PP ln(2.718281828459045) │
974 │ → 1.0 │
975 ├────────────────────────────────────────┤
976 │ │
977 │ mod ( integer, integer ) → │
978 │ integer │
979 │ │
980 │ .PP Modulo (remainder) │
981 │ │
982 │ .PP mod(54, 32) → 22 │
983 ├────────────────────────────────────────┤
984 │ │
985 │ pi () → double │
986 │ │
987 │ .PP Approximate value of │
988 │ π │
989 │ │
990 │ .PP pi() → │
991 │ 3.14159265358979323846 │
992 ├────────────────────────────────────────┤
993 │ │
994 │ pow ( x, y ) → double │
995 │ │
996 │ .PP power ( x, y ) → │
997 │ double │
998 │ │
999 │ .PP x raised to the power │
1000 │ of y │
1001 │ │
1002 │ .PP pow(2.0, 10) → 1024.0 │
1003 ├────────────────────────────────────────┤
1004 │ │
1005 │ random ( lb, ub ) → integer │
1006 │ │
1007 │ .PP Computes a │
1008 │ uniformly-distributed random │
1009 │ integer in [lb, ub]. │
1010 │ │
1011 │ .PP random(1, 10) → an │
1012 │ integer between 1 and 10 │
1013 ├────────────────────────────────────────┤
1014 │ │
1015 │ random_exponential ( lb, ub, │
1016 │ parameter ) → integer │
1017 │ │
1018 │ .PP Computes an │
1019 │ exponentially-distributed random │
1020 │ integer in [lb, ub], see below. │
1021 │ │
1022 │ .PP random_exponential(1, │
1023 │ 10, 3.0) → an integer between 1 │
1024 │ and 10 │
1025 ├────────────────────────────────────────┤
1026 │ │
1027 │ random_gaussian ( lb, ub, │
1028 │ parameter ) → integer │
1029 │ │
1030 │ .PP Computes a │
1031 │ Gaussian-distributed random │
1032 │ integer in [lb, ub], see below. │
1033 │ │
1034 │ .PP random_gaussian(1, │
1035 │ 10, 2.5) → an integer between 1 │
1036 │ and 10 │
1037 ├────────────────────────────────────────┤
1038 │ │
1039 │ random_zipfian ( lb, ub, │
1040 │ parameter ) → integer │
1041 │ │
1042 │ .PP Computes a │
1043 │ Zipfian-distributed random │
1044 │ integer in [lb, ub], see below. │
1045 │ │
1046 │ .PP random_zipfian(1, 10, │
1047 │ 1.5) → an integer between 1 and │
1048 │ 10 │
1049 ├────────────────────────────────────────┤
1050 │ │
1051 │ sqrt ( number ) → double │
1052 │ │
1053 │ .PP Square root │
1054 │ │
1055 │ .PP sqrt(2.0) → │
1056 │ 1.414213562 │
1057 └────────────────────────────────────────┘
1058
1059 The random function generates values using a uniform distribution, that
1060 is all the values are drawn within the specified range with equal
1061 probability. The random_exponential, random_gaussian and random_zipfian
1062 functions require an additional double parameter which determines the
1063 precise shape of the distribution.
1064
1065 • For an exponential distribution, parameter controls the
1066 distribution by truncating a quickly-decreasing exponential
1067 distribution at parameter, and then projecting onto integers
1068 between the bounds. To be precise, with
1069
1070 f(x) = exp(-parameter * (x - min) / (max - min + 1)) / (1 - exp(-parameter))
1071
1072 Then value i between min and max inclusive is drawn with
1073 probability: f(i) - f(i + 1).
1074
1075 Intuitively, the larger the parameter, the more frequently values
1076 close to min are accessed, and the less frequently values close to
1077 max are accessed. The closer to 0 parameter is, the flatter (more
1078 uniform) the access distribution. A crude approximation of the
1079 distribution is that the most frequent 1% values in the range,
1080 close to min, are drawn parameter% of the time. The parameter value
1081 must be strictly positive.
1082
1083 • For a Gaussian distribution, the interval is mapped onto a standard
1084 normal distribution (the classical bell-shaped Gaussian curve)
1085 truncated at -parameter on the left and +parameter on the right.
1086 Values in the middle of the interval are more likely to be drawn.
1087 To be precise, if PHI(x) is the cumulative distribution function of
1088 the standard normal distribution, with mean mu defined as (max +
1089 min) / 2.0, with
1090
1091 f(x) = PHI(2.0 * parameter * (x - mu) / (max - min + 1)) /
1092 (2.0 * PHI(parameter) - 1)
1093
1094 then value i between min and max inclusive is drawn with
1095 probability: f(i + 0.5) - f(i - 0.5). Intuitively, the larger the
1096 parameter, the more frequently values close to the middle of the
1097 interval are drawn, and the less frequently values close to the min
1098 and max bounds. About 67% of values are drawn from the middle 1.0 /
1099 parameter, that is a relative 0.5 / parameter around the mean, and
1100 95% in the middle 2.0 / parameter, that is a relative 1.0 /
1101 parameter around the mean; for instance, if parameter is 4.0, 67%
1102 of values are drawn from the middle quarter (1.0 / 4.0) of the
1103 interval (i.e., from 3.0 / 8.0 to 5.0 / 8.0) and 95% from the
1104 middle half (2.0 / 4.0) of the interval (second and third
1105 quartiles). The minimum allowed parameter value is 2.0.
1106
1107 • random_zipfian generates a bounded Zipfian distribution. parameter
1108 defines how skewed the distribution is. The larger the parameter,
1109 the more frequently values closer to the beginning of the interval
1110 are drawn. The distribution is such that, assuming the range starts
1111 from 1, the ratio of the probability of drawing k versus drawing
1112 k+1 is ((k+1)/k)**parameter. For example, random_zipfian(1, ...,
1113 2.5) produces the value 1 about (2/1)**2.5 = 5.66 times more
1114 frequently than 2, which itself is produced (3/2)**2.5 = 2.76 times
1115 more frequently than 3, and so on.
1116
1117 pgbench's implementation is based on "Non-Uniform Random Variate
1118 Generation", Luc Devroye, p. 550-551, Springer 1986. Due to
1119 limitations of that algorithm, the parameter value is restricted to
1120 the range [1.001, 1000].
1121
1122 Hash functions hash, hash_murmur2 and hash_fnv1a accept an input value
1123 and an optional seed parameter. In case the seed isn't provided the
1124 value of :default_seed is used, which is initialized randomly unless
1125 set by the command-line -D option. Hash functions can be used to
1126 scatter the distribution of random functions such as random_zipfian or
1127 random_exponential. For instance, the following pgbench script
1128 simulates possible real world workload typical for social media and
1129 blogging platforms where few accounts generate excessive load:
1130
1131 \set r random_zipfian(0, 100000000, 1.07)
1132 \set k abs(hash(:r)) % 1000000
1133
1134 In some cases several distinct distributions are needed which don't
1135 correlate with each other and this is when implicit seed parameter
1136 comes in handy:
1137
1138 \set k1 abs(hash(:r, :default_seed + 123)) % 1000000
1139 \set k2 abs(hash(:r, :default_seed + 321)) % 1000000
1140
1141 As an example, the full definition of the built-in TPC-B-like
1142 transaction is:
1143
1144 \set aid random(1, 100000 * :scale)
1145 \set bid random(1, 1 * :scale)
1146 \set tid random(1, 10 * :scale)
1147 \set delta random(-5000, 5000)
1148 BEGIN;
1149 UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
1150 SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
1151 UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;
1152 UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;
1153 INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
1154 END;
1155
1156 This script allows each iteration of the transaction to reference
1157 different, randomly-chosen rows. (This example also shows why it's
1158 important for each client session to have its own variables — otherwise
1159 they'd not be independently touching different rows.)
1160
1161 Per-Transaction Logging
1162 With the -l option (but without the --aggregate-interval option),
1163 pgbench writes information about each transaction to a log file. The
1164 log file will be named prefix.nnn, where prefix defaults to
1165 pgbench_log, and nnn is the PID of the pgbench process. The prefix can
1166 be changed by using the --log-prefix option. If the -j option is 2 or
1167 higher, so that there are multiple worker threads, each will have its
1168 own log file. The first worker will use the same name for its log file
1169 as in the standard single worker case. The additional log files for the
1170 other workers will be named prefix.nnn.mmm, where mmm is a sequential
1171 number for each worker starting with 1.
1172
1173 The format of the log is:
1174
1175 client_id transaction_no time script_no time_epoch time_us [ schedule_lag ]
1176
1177 where client_id indicates which client session ran the transaction,
1178 transaction_no counts how many transactions have been run by that
1179 session, time is the total elapsed transaction time in microseconds,
1180 script_no identifies which script file was used (useful when multiple
1181 scripts were specified with -f or -b), and time_epoch/time_us are a
1182 Unix-epoch time stamp and an offset in microseconds (suitable for
1183 creating an ISO 8601 time stamp with fractional seconds) showing when
1184 the transaction completed. The schedule_lag field is the difference
1185 between the transaction's scheduled start time, and the time it
1186 actually started, in microseconds. It is only present when the --rate
1187 option is used. When both --rate and --latency-limit are used, the time
1188 for a skipped transaction will be reported as skipped.
1189
1190 Here is a snippet of a log file generated in a single-client run:
1191
1192 0 199 2241 0 1175850568 995598
1193 0 200 2465 0 1175850568 998079
1194 0 201 2513 0 1175850569 608
1195 0 202 2038 0 1175850569 2663
1196
1197 Another example with --rate=100 and --latency-limit=5 (note the
1198 additional schedule_lag column):
1199
1200 0 81 4621 0 1412881037 912698 3005
1201 0 82 6173 0 1412881037 914578 4304
1202 0 83 skipped 0 1412881037 914578 5217
1203 0 83 skipped 0 1412881037 914578 5099
1204 0 83 4722 0 1412881037 916203 3108
1205 0 84 4142 0 1412881037 918023 2333
1206 0 85 2465 0 1412881037 919759 740
1207
1208 In this example, transaction 82 was late, because its latency (6.173
1209 ms) was over the 5 ms limit. The next two transactions were skipped,
1210 because they were already late before they were even started.
1211
1212 When running a long test on hardware that can handle a lot of
1213 transactions, the log files can become very large. The --sampling-rate
1214 option can be used to log only a random sample of transactions.
1215
1216 Aggregated Logging
1217 With the --aggregate-interval option, a different format is used for
1218 the log files:
1219
1220 interval_start num_transactions sum_latency sum_latency_2 min_latency max_latency [ sum_lag sum_lag_2 min_lag max_lag [ skipped ] ]
1221
1222 where interval_start is the start of the interval (as a Unix epoch time
1223 stamp), num_transactions is the number of transactions within the
1224 interval, sum_latency is the sum of the transaction latencies within
1225 the interval, sum_latency_2 is the sum of squares of the transaction
1226 latencies within the interval, min_latency is the minimum latency
1227 within the interval, and max_latency is the maximum latency within the
1228 interval. The next fields, sum_lag, sum_lag_2, min_lag, and max_lag,
1229 are only present if the --rate option is used. They provide statistics
1230 about the time each transaction had to wait for the previous one to
1231 finish, i.e., the difference between each transaction's scheduled start
1232 time and the time it actually started. The very last field, skipped, is
1233 only present if the --latency-limit option is used, too. It counts the
1234 number of transactions skipped because they would have started too
1235 late. Each transaction is counted in the interval when it was
1236 committed.
1237
1238 Here is some example output:
1239
1240 1345828501 5601 1542744 483552416 61 2573
1241 1345828503 7884 1979812 565806736 60 1479
1242 1345828505 7208 1979422 567277552 59 1391
1243 1345828507 7685 1980268 569784714 60 1398
1244 1345828509 7073 1979779 573489941 236 1411
1245
1246 Notice that while the plain (unaggregated) log file shows which script
1247 was used for each transaction, the aggregated log does not. Therefore
1248 if you need per-script data, you need to aggregate the data on your
1249 own.
1250
1251 Per-Statement Latencies
1252 With the -r option, pgbench collects the elapsed transaction time of
1253 each statement executed by every client. It then reports an average of
1254 those values, referred to as the latency for each statement, after the
1255 benchmark has finished.
1256
1257 For the default script, the output will look similar to this:
1258
1259 starting vacuum...end.
1260 transaction type: <builtin: TPC-B (sort of)>
1261 scaling factor: 1
1262 query mode: simple
1263 number of clients: 10
1264 number of threads: 1
1265 number of transactions per client: 1000
1266 number of transactions actually processed: 10000/10000
1267 latency average = 15.844 ms
1268 latency stddev = 2.715 ms
1269 tps = 618.764555 (including connections establishing)
1270 tps = 622.977698 (excluding connections establishing)
1271 statement latencies in milliseconds:
1272 0.002 \set aid random(1, 100000 * :scale)
1273 0.005 \set bid random(1, 1 * :scale)
1274 0.002 \set tid random(1, 10 * :scale)
1275 0.001 \set delta random(-5000, 5000)
1276 0.326 BEGIN;
1277 0.603 UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
1278 0.454 SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
1279 5.528 UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;
1280 7.335 UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;
1281 0.371 INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
1282 1.212 END;
1283
1284 If multiple script files are specified, the averages are reported
1285 separately for each script file.
1286
1287 Note that collecting the additional timing information needed for
1288 per-statement latency computation adds some overhead. This will slow
1289 average execution speed and lower the computed TPS. The amount of
1290 slowdown varies significantly depending on platform and hardware.
1291 Comparing average TPS values with and without latency reporting enabled
1292 is a good way to measure if the timing overhead is significant.
1293
1294 Good Practices
1295 It is very easy to use pgbench to produce completely meaningless
1296 numbers. Here are some guidelines to help you get useful results.
1297
1298 In the first place, never believe any test that runs for only a few
1299 seconds. Use the -t or -T option to make the run last at least a few
1300 minutes, so as to average out noise. In some cases you could need hours
1301 to get numbers that are reproducible. It's a good idea to try the test
1302 run a few times, to find out if your numbers are reproducible or not.
1303
1304 For the default TPC-B-like test scenario, the initialization scale
1305 factor (-s) should be at least as large as the largest number of
1306 clients you intend to test (-c); else you'll mostly be measuring update
1307 contention. There are only -s rows in the pgbench_branches table, and
1308 every transaction wants to update one of them, so -c values in excess
1309 of -s will undoubtedly result in lots of transactions blocked waiting
1310 for other transactions.
1311
1312 The default test scenario is also quite sensitive to how long it's been
1313 since the tables were initialized: accumulation of dead rows and dead
1314 space in the tables changes the results. To understand the results you
1315 must keep track of the total number of updates and when vacuuming
1316 happens. If autovacuum is enabled it can result in unpredictable
1317 changes in measured performance.
1318
1319 A limitation of pgbench is that it can itself become the bottleneck
1320 when trying to test a large number of client sessions. This can be
1321 alleviated by running pgbench on a different machine from the database
1322 server, although low network latency will be essential. It might even
1323 be useful to run several pgbench instances concurrently, on several
1324 client machines, against the same database server.
1325
1326 Security
1327 If untrusted users have access to a database that has not adopted a
1328 secure schema usage pattern, do not run pgbench in that database.
1329 pgbench uses unqualified names and does not manipulate the search path.
1330
1331
1332
1333PostgreSQL 13.3 2021 PGBENCH(1)