1PGBENCH(1) PostgreSQL 15.4 Documentation PGBENCH(1)
2
6 pgbench - run a benchmark test on PostgreSQL
7
9 pgbench -i [option...] [dbname]
10 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 Typical output from pgbench looks like:
23 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 maximum number of tries: 1
30 number of transactions per client: 1000
31 number of transactions actually processed: 10000/10000
32 number of failed transactions: 0 (0.000%)
33 latency average = 11.013 ms
34 latency stddev = 7.351 ms
35 initial connection time = 45.758 ms
36 tps = 896.967014 (without initial connection time)
37 The first seven lines report some of the most important parameter
39 settings. The sixth line reports the maximum number of tries for
40 transactions with serialization or deadlock errors (see Failures and
41 Serialization/Deadlock Retries for more information). The eighth line
42 reports the number of transactions completed and intended (the latter
43 being just the product of number of clients and number of transactions
44 per client); these will be equal unless the run failed before
45 completion or some SQL command(s) failed. (In -T mode, only the actual
46 number of transactions is printed.) The next line reports the number of
47 failed transactions due to serialization or deadlock errors (see
48 Failures and Serialization/Deadlock Retries for more information). The
49 last line reports the number of transactions per second.
50 The default TPC-B-like transaction test requires specific tables to be
52 set up beforehand. pgbench should be invoked with the -i (initialize)
53 option to create and populate these tables. (When you are testing a
54 custom script, you don't need this step, but will instead need to do
55 whatever setup your test needs.) Initialization looks like:
56 pgbench -i [ other-options ] dbname
58 where dbname is the name of the already-created database to test in.
60 (You may also need -h, -p, and/or -U options to specify how to connect
61 to the database server.)
62 Caution
64 pgbench -i creates four tables pgbench_accounts, pgbench_branches,
65 pgbench_history, and pgbench_tellers, destroying any existing
66 tables of these names. Be very careful to use another database if
67 you have tables having these names!
68 At the default “scale factor” of 1, the tables initially contain this
70 many rows:
71 table # of rows
73 ---------------------------------
74 pgbench_branches 1
75 pgbench_tellers 10
76 pgbench_accounts 100000
77 pgbench_history 0
78 You can (and, for most purposes, probably should) increase the number
80 of rows by using the -s (scale factor) option. The -F (fillfactor)
81 option might also be used at this point.
82 Once you have done the necessary setup, you can run your benchmark with
84 a command that doesn't include -i, that is
85 pgbench [ options ] dbname
87 In nearly all cases, you'll need some options to make a useful test.
89 The most important options are -c (number of clients), -t (number of
90 transactions), -T (time limit), and -f (specify a custom script file).
91 See below for a full list.
92
94 The following is divided into three subsections. Different options are
95 used during database initialization and while running benchmarks, but
96 some options are useful in both cases.
97 Initialization Options
99 pgbench accepts the following command-line initialization arguments:
100 dbname
102 Specifies the name of the database to test in. If this is not
103 specified, the environment variable PGDATABASE is used. If that is
104 not set, the user name specified for the connection is used.
105 -i
107 --initialize
108 Required to invoke initialization mode.
109 -I init_steps
111 --init-steps=init_steps
112 Perform just a selected set of the normal initialization steps.
113 init_steps specifies the initialization steps to be performed,
114 using one character per step. Each step is invoked in the specified
115 order. The default is dtgvp. The available steps are:
116 d (Drop)
118 Drop any existing pgbench tables.
119 t (create Tables)
121 Create the tables used by the standard pgbench scenario, namely
122 pgbench_accounts, pgbench_branches, pgbench_history, and
123 pgbench_tellers.
124 g or G (Generate data, client-side or server-side)
126 Generate data and load it into the standard tables, replacing
127 any data already present.
128 With g (client-side data generation), data is generated in
130 pgbench client and then sent to the server. This uses the
131 client/server bandwidth extensively through a COPY. pgbench
132 uses the FREEZE option with version 14 or later of PostgreSQL
133 to speed up subsequent VACUUM, unless partitions are enabled.
134 Using g causes logging to print one message every 100,000 rows
135 while generating data for the pgbench_accounts table.
136 With G (server-side data generation), only small queries are
138 sent from the pgbench client and then data is actually
139 generated in the server. No significant bandwidth is required
140 for this variant, but the server will do more work. Using G
141 causes logging not to print any progress message while
142 generating data.
143 The default initialization behavior uses client-side data
145 generation (equivalent to g).
146 v (Vacuum)
148 Invoke VACUUM on the standard tables.
149 p (create Primary keys)
151 Create primary key indexes on the standard tables.
152 f (create Foreign keys)
154 Create foreign key constraints between the standard tables.
155 (Note that this step is not performed by default.)
156 -F fillfactor
158 --fillfactor=fillfactor
159 Create the pgbench_accounts, pgbench_tellers and pgbench_branches
160 tables with the given fillfactor. Default is 100.
161 -n
163 --no-vacuum
164 Perform no vacuuming during initialization. (This option suppresses
165 the v initialization step, even if it was specified in -I.)
166 -q
168 --quiet
169 Switch logging to quiet mode, producing only one progress message
170 per 5 seconds. The default logging prints one message each 100,000
171 rows, which often outputs many lines per second (especially on good
172 hardware).
173 This setting has no effect if G is specified in -I.
175 -s scale_factor
177 --scale=scale_factor
178 Multiply the number of rows generated by the scale factor. For
179 example, -s 100 will create 10,000,000 rows in the pgbench_accounts
180 table. Default is 1. When the scale is 20,000 or larger, the
181 columns used to hold account identifiers (aid columns) will switch
182 to using larger integers (bigint), in order to be big enough to
183 hold the range of account identifiers.
184 --foreign-keys
186 Create foreign key constraints between the standard tables. (This
187 option adds the f step to the initialization step sequence, if it
188 is not already present.)
189 --index-tablespace=index_tablespace
191 Create indexes in the specified tablespace, rather than the default
192 tablespace.
193 --partition-method=NAME
195 Create a partitioned pgbench_accounts table with NAME method.
196 Expected values are range or hash. This option requires that
197 --partitions is set to non-zero. If unspecified, default is range.
198 --partitions=NUM
200 Create a partitioned pgbench_accounts table with NUM partitions of
201 nearly equal size for the scaled number of accounts. Default is 0,
202 meaning no partitioning.
203 --tablespace=tablespace
205 Create tables in the specified tablespace, rather than the default
206 tablespace.
207 --unlogged-tables
209 Create all tables as unlogged tables, rather than permanent tables.
210 Benchmarking Options
212 pgbench accepts the following command-line benchmarking arguments:
213 -b scriptname[@weight]
215 --builtin=scriptname[@weight]
216 Add the specified built-in script to the list of scripts to be
217 executed. Available built-in scripts are: tpcb-like, simple-update
218 and select-only. Unambiguous prefixes of built-in names are
219 accepted. With the special name list, show the list of built-in
220 scripts and exit immediately.
221 Optionally, write an integer weight after @ to adjust the
223 probability of selecting this script versus other ones. The default
224 weight is 1. See below for details.
225 -c clients
227 --client=clients
228 Number of clients simulated, that is, number of concurrent database
229 sessions. Default is 1.
230 -C
232 --connect
233 Establish a new connection for each transaction, rather than doing
234 it just once per client session. This is useful to measure the
235 connection overhead.
236 -d
238 --debug
239 Print debugging output.
240 -D varname=value
242 --define=varname=value
243 Define a variable for use by a custom script (see below). Multiple
244 -D options are allowed.
245 -f filename[@weight]
247 --file=filename[@weight]
248 Add a transaction script read from filename to the list of scripts
249 to be executed.
250 Optionally, write an integer weight after @ to adjust the
252 probability of selecting this script versus other ones. The default
253 weight is 1. (To use a script file name that includes an @
254 character, append a weight so that there is no ambiguity, for
255 example filen@me@1.) See below for details.
256 -j threads
258 --jobs=threads
259 Number of worker threads within pgbench. Using more than one thread
260 can be helpful on multi-CPU machines. Clients are distributed as
261 evenly as possible among available threads. Default is 1.
262 -l
264 --log
265 Write information about each transaction to a log file. See below
266 for details.
267 -L limit
269 --latency-limit=limit
270 Transactions that last more than limit milliseconds are counted and
271 reported separately, as late.
272 When throttling is used (--rate=...), transactions that lag behind
274 schedule by more than limit ms, and thus have no hope of meeting
275 the latency limit, are not sent to the server at all. They are
276 counted and reported separately as skipped.
277 When the --max-tries option is used, a transaction which fails due
279 to a serialization anomaly or from a deadlock will not be retried
280 if the total time of all its tries is greater than limit ms. To
281 limit only the time of tries and not their number, use
282 --max-tries=0. By default, the option --max-tries is set to 1 and
283 transactions with serialization/deadlock errors are not retried.
284 See Failures and Serialization/Deadlock Retries for more
285 information about retrying such transactions.
286 -M querymode
288 --protocol=querymode
289 Protocol to use for submitting queries to the server:
290 • simple: use simple query protocol.
292 • extended: use extended query protocol.
294 • prepared: use extended query protocol with prepared statements.
296 In the prepared mode, pgbench reuses the parse analysis result
298 starting from the second query iteration, so pgbench runs faster
299 than in other modes.
300 The default is simple query protocol. (See Chapter 55 for more
302 information.)
303 -n
305 --no-vacuum
306 Perform no vacuuming before running the test. This option is
307 necessary if you are running a custom test scenario that does not
308 include the standard tables pgbench_accounts, pgbench_branches,
309 pgbench_history, and pgbench_tellers.
310 -N
312 --skip-some-updates
313 Run built-in simple-update script. Shorthand for -b simple-update.
314 -P sec
316 --progress=sec
317 Show progress report every sec seconds. The report includes the
318 time since the beginning of the run, the TPS since the last report,
319 and the transaction latency average, standard deviation, and the
320 number of failed transactions since the last report. Under
321 throttling (-R), the latency is computed with respect to the
322 transaction scheduled start time, not the actual transaction
323 beginning time, thus it also includes the average schedule lag
324 time. When --max-tries is used to enable transaction retries after
325 serialization/deadlock errors, the report includes the number of
326 retried transactions and the sum of all retries.
327 -r
329 --report-per-command
330 Report the following statistics for each command after the
331 benchmark finishes: the average per-statement latency (execution
332 time from the perspective of the client), the number of failures,
333 and the number of retries after serialization or deadlock errors in
334 this command. The report displays retry statistics only if the
335 --max-tries option is not equal to 1.
336 -R rate
338 --rate=rate
339 Execute transactions targeting the specified rate instead of
340 running as fast as possible (the default). The rate is given in
341 transactions per second. If the targeted rate is above the maximum
342 possible rate, the rate limit won't impact the results.
343 The rate is targeted by starting transactions along a
345 Poisson-distributed schedule time line. The expected start time
346 schedule moves forward based on when the client first started, not
347 when the previous transaction ended. That approach means that when
348 transactions go past their original scheduled end time, it is
349 possible for later ones to catch up again.
350 When throttling is active, the transaction latency reported at the
352 end of the run is calculated from the scheduled start times, so it
353 includes the time each transaction had to wait for the previous
354 transaction to finish. The wait time is called the schedule lag
355 time, and its average and maximum are also reported separately. The
356 transaction latency with respect to the actual transaction start
357 time, i.e., the time spent executing the transaction in the
358 database, can be computed by subtracting the schedule lag time from
359 the reported latency.
360 If --latency-limit is used together with --rate, a transaction can
362 lag behind so much that it is already over the latency limit when
363 the previous transaction ends, because the latency is calculated
364 from the scheduled start time. Such transactions are not sent to
365 the server, but are skipped altogether and counted separately.
366 A high schedule lag time is an indication that the system cannot
368 process transactions at the specified rate, with the chosen number
369 of clients and threads. When the average transaction execution time
370 is longer than the scheduled interval between each transaction,
371 each successive transaction will fall further behind, and the
372 schedule lag time will keep increasing the longer the test run is.
373 When that happens, you will have to reduce the specified
374 transaction rate.
375 -s scale_factor
377 --scale=scale_factor
378 Report the specified scale factor in pgbench's output. With the
379 built-in tests, this is not necessary; the correct scale factor
380 will be detected by counting the number of rows in the
381 pgbench_branches table. However, when testing only custom
382 benchmarks (-f option), the scale factor will be reported as 1
383 unless this option is used.
384 -S
386 --select-only
387 Run built-in select-only script. Shorthand for -b select-only.
388 -t transactions
390 --transactions=transactions
391 Number of transactions each client runs. Default is 10.
392 -T seconds
394 --time=seconds
395 Run the test for this many seconds, rather than a fixed number of
396 transactions per client. -t and -T are mutually exclusive.
397 -v
399 --vacuum-all
400 Vacuum all four standard tables before running the test. With
401 neither -n nor -v, pgbench will vacuum the pgbench_tellers and
402 pgbench_branches tables, and will truncate pgbench_history.
403 --aggregate-interval=seconds
405 Length of aggregation interval (in seconds). May be used only with
406 -l option. With this option, the log contains per-interval summary
407 data, as described below.
408 --failures-detailed
410 Report failures in per-transaction and aggregation logs, as well as
411 in the main and per-script reports, grouped by the following types:
412 • serialization failures;
414 • deadlock failures;
416 See Failures and Serialization/Deadlock Retries for more
418 information.
419 --log-prefix=prefix
421 Set the filename prefix for the log files created by --log. The
422 default is pgbench_log.
423 --max-tries=number_of_tries
425 Enable retries for transactions with serialization/deadlock errors
426 and set the maximum number of these tries. This option can be
427 combined with the --latency-limit option which limits the total
428 time of all transaction tries; moreover, you cannot use an
429 unlimited number of tries (--max-tries=0) without --latency-limit
430 or --time. The default value is 1 and transactions with
431 serialization/deadlock errors are not retried. See Failures and
432 Serialization/Deadlock Retries for more information about retrying
433 such transactions.
434 --progress-timestamp
436 When showing progress (option -P), use a timestamp (Unix epoch)
437 instead of the number of seconds since the beginning of the run.
438 The unit is in seconds, with millisecond precision after the dot.
439 This helps compare logs generated by various tools.
440 --random-seed=seed
442 Set random generator seed. Seeds the system random number
443 generator, which then produces a sequence of initial generator
444 states, one for each thread. Values for seed may be: time (the
445 default, the seed is based on the current time), rand (use a strong
446 random source, failing if none is available), or an unsigned
447 decimal integer value. The random generator is invoked explicitly
448 from a pgbench script (random... functions) or implicitly (for
449 instance option --rate uses it to schedule transactions). When
450 explicitly set, the value used for seeding is shown on the
451 terminal. Any value allowed for seed may also be provided through
452 the environment variable PGBENCH_RANDOM_SEED. To ensure that the
453 provided seed impacts all possible uses, put this option first or
454 use the environment variable.
455 Setting the seed explicitly allows to reproduce a pgbench run
457 exactly, as far as random numbers are concerned. As the random
458 state is managed per thread, this means the exact same pgbench run
459 for an identical invocation if there is one client per thread and
460 there are no external or data dependencies. From a statistical
461 viewpoint reproducing runs exactly is a bad idea because it can
462 hide the performance variability or improve performance unduly,
463 e.g., by hitting the same pages as a previous run. However, it may
464 also be of great help for debugging, for instance re-running a
465 tricky case which leads to an error. Use wisely.
466 --sampling-rate=rate
468 Sampling rate, used when writing data into the log, to reduce the
469 amount of log generated. If this option is given, only the
470 specified fraction of transactions are logged. 1.0 means all
471 transactions will be logged, 0.05 means only 5% of the transactions
472 will be logged.
473 Remember to take the sampling rate into account when processing the
475 log file. For example, when computing TPS values, you need to
476 multiply the numbers accordingly (e.g., with 0.01 sample rate,
477 you'll only get 1/100 of the actual TPS).
478 --show-script=scriptname
480 Show the actual code of builtin script scriptname on stderr, and
481 exit immediately.
482 --verbose-errors
484 Print messages about all errors and failures (errors without
485 retrying) including which limit for retries was exceeded and how
486 far it was exceeded for the serialization/deadlock failures. (Note
487 that in this case the output can be significantly increased.). See
488 Failures and Serialization/Deadlock Retries for more information.
489 Common Options
491 pgbench also accepts the following common command-line arguments for
492 connection parameters:
493 -h hostname
495 --host=hostname
496 The database server's host name
497 -p port
499 --port=port
500 The database server's port number
501 -U login
503 --username=login
504 The user name to connect as
505 -V
507 --version
508 Print the pgbench version and exit.
509 -?
511 --help
512 Show help about pgbench command line arguments, and exit.
513
515 A successful run will exit with status 0. Exit status 1 indicates
516 static problems such as invalid command-line options or internal errors
517 which are supposed to never occur. Early errors that occur when
518 starting benchmark such as initial connection failures also exit with
519 status 1. Errors during the run such as database errors or problems in
520 the script will result in exit status 2. In the latter case, pgbench
521 will print partial results.
522
524 PGDATABASE
525 PGHOST
526 PGPORT
527 PGUSER
528 Default connection parameters.
529 This utility, like most other PostgreSQL utilities, uses the
531 environment variables supported by libpq (see Section 34.15).
532 The environment variable PG_COLOR specifies whether to use color in
534 diagnostic messages. Possible values are always, auto and never.
535
537 What Is the “Transaction” Actually Performed in pgbench?
538 pgbench executes test scripts chosen randomly from a specified list.
539 The scripts may include built-in scripts specified with -b and
540 user-provided scripts specified with -f. Each script may be given a
541 relative weight specified after an @ so as to change its selection
542 probability. The default weight is 1. Scripts with a weight of 0 are
543 ignored.
544 The default built-in transaction script (also invoked with -b
546 tpcb-like) issues seven commands per transaction over randomly chosen
547 aid, tid, bid and delta. The scenario is inspired by the TPC-B
548 benchmark, but is not actually TPC-B, hence the name.
549 1. BEGIN;
551 2. UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid
553 = :aid;
554 3. SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
556 4. UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid =
558 :tid;
559 5. UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid
561 = :bid;
562 6. INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES
564 (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
565 7. END;
567 If you select the simple-update built-in (also -N), steps 4 and 5
569 aren't included in the transaction. This will avoid update contention
570 on these tables, but it makes the test case even less like TPC-B.
571 If you select the select-only built-in (also -S), only the SELECT is
573 issued.
574 Custom Scripts
576 pgbench has support for running custom benchmark scenarios by replacing
577 the default transaction script (described above) with a transaction
578 script read from a file (-f option). In this case a “transaction”
579 counts as one execution of a script file.
580 A script file contains one or more SQL commands terminated by
582 semicolons. Empty lines and lines beginning with -- are ignored. Script
583 files can also contain “meta commands”, which are interpreted by
584 pgbench itself, as described below.
585 Note
587 Before PostgreSQL 9.6, SQL commands in script files were terminated
588 by newlines, and so they could not be continued across lines. Now a
589 semicolon is required to separate consecutive SQL commands (though
590 an SQL command does not need one if it is followed by a meta
591 command). If you need to create a script file that works with both
592 old and new versions of pgbench, be sure to write each SQL command
593 on a single line ending with a semicolon.
594 It is assumed that pgbench scripts do not contain incomplete blocks
596 of SQL transactions. If at runtime the client reaches the end of
597 the script without completing the last transaction block, it will
598 be aborted.
599 There is a simple variable-substitution facility for script files.
601 Variable names must consist of letters (including non-Latin letters),
602 digits, and underscores, with the first character not being a digit.
603 Variables can be set by the command-line -D option, explained above, or
604 by the meta commands explained below. In addition to any variables
605 preset by -D command-line options, there are a few variables that are
606 preset automatically, listed in Table 288. A value specified for these
607 variables using -D takes precedence over the automatic presets. Once
608 set, a variable's value can be inserted into an SQL command by writing
609 :variablename. When running more than one client session, each session
610 has its own set of variables. pgbench supports up to 255 variable uses
611 in one statement.
612 Table 288. pgbench Automatic Variables
614 ┌─────────────┬────────────────────────────┐
615 │Variable │ Description │
616 ├─────────────┼────────────────────────────┤
617 │client_id │ unique number identifying │
618 │ │ the client session (starts │
619 │ │ from zero) │
620 ├─────────────┼────────────────────────────┤
621 │default_seed │ seed used in hash and │
622 │ │ pseudorandom permutation │
623 │ │ functions by default │
624 ├─────────────┼────────────────────────────┤
625 │random_seed │ random generator seed │
626 │ │ (unless overwritten with │
627 │ │ -D) │
628 ├─────────────┼────────────────────────────┤
629 │scale │ current scale factor │
630 └─────────────┴────────────────────────────┘
631 Script file meta commands begin with a backslash (\) and normally
633 extend to the end of the line, although they can be continued to
634 additional lines by writing backslash-return. Arguments to a meta
635 command are separated by white space. These meta commands are
636 supported:
637 \gset [prefix] \aset [prefix]
639 These commands may be used to end SQL queries, taking the place of
640 the terminating semicolon (;).
641 When the \gset command is used, the preceding SQL query is expected
643 to return one row, the columns of which are stored into variables
644 named after column names, and prefixed with prefix if provided.
645 When the \aset command is used, all combined SQL queries (separated
647 by \;) have their columns stored into variables named after column
648 names, and prefixed with prefix if provided. If a query returns no
649 row, no assignment is made and the variable can be tested for
650 existence to detect this. If a query returns more than one row, the
651 last value is kept.
652 \gset and \aset cannot be used in pipeline mode, since the query
654 results are not yet available by the time the commands would need
655 them.
656 The following example puts the final account balance from the first
658 query into variable abalance, and fills variables p_two and p_three
659 with integers from the third query. The result of the second query
660 is discarded. The result of the two last combined queries are
661 stored in variables four and five.
662 UPDATE pgbench_accounts
664 SET abalance = abalance + :delta
665 WHERE aid = :aid
666 RETURNING abalance \gset
667 -- compound of two queries
668 SELECT 1 \;
669 SELECT 2 AS two, 3 AS three \gset p_
670 SELECT 4 AS four \; SELECT 5 AS five \aset
671 \if expression
673 \elif expression
674 \else
675 \endif
676 This group of commands implements nestable conditional blocks,
677 similarly to psql's \if expression. Conditional expressions are
678 identical to those with \set, with non-zero values interpreted as
679 true.
680 \set varname expression
682 Sets variable varname to a value calculated from expression. The
683 expression may contain the NULL constant, Boolean constants TRUE
684 and FALSE, integer constants such as 5432, double constants such as
685 3.14159, references to variables :variablename, operators with
686 their usual SQL precedence and associativity, function calls, SQL
687 CASE generic conditional expressions and parentheses.
688 Functions and most operators return NULL on NULL input.
690 For conditional purposes, non zero numerical values are TRUE, zero
692 numerical values and NULL are FALSE.
693 Too large or small integer and double constants, as well as integer
695 arithmetic operators (+, -, * and /) raise errors on overflows.
696 When no final ELSE clause is provided to a CASE, the default value
698 is NULL.
699 Examples:
701 \set ntellers 10 * :scale
703 \set aid (1021 * random(1, 100000 * :scale)) % \
704 (100000 * :scale) + 1
705 \set divx CASE WHEN :x <> 0 THEN :y/:x ELSE NULL END
706 \sleep number [ us | ms | s ]
708 Causes script execution to sleep for the specified duration in
709 microseconds (us), milliseconds (ms) or seconds (s). If the unit is
710 omitted then seconds are the default. number can be either an
711 integer constant or a :variablename reference to a variable having
712 an integer value.
713 Example:
715 \sleep 10 ms
717 \setshell varname command [ argument ... ]
719 Sets variable varname to the result of the shell command command
720 with the given argument(s). The command must return an integer
721 value through its standard output.
722 command and each argument can be either a text constant or a
724 :variablename reference to a variable. If you want to use an
725 argument starting with a colon, write an additional colon at the
726 beginning of argument.
727 Example:
729 \setshell variable_to_be_assigned command literal_argument :variable ::literal_starting_with_colon
731 \shell command [ argument ... ]
733 Same as \setshell, but the result of the command is discarded.
734 Example:
736 \shell command literal_argument :variable ::literal_starting_with_colon
738 \startpipeline
740 \endpipeline
741 These commands delimit the start and end of a pipeline of SQL
742 statements. In pipeline mode, statements are sent to the server
743 without waiting for the results of previous statements. See
744 Section 34.5 for more details. Pipeline mode requires the use of
745 extended query protocol.
746 Built-in Operators
748 The arithmetic, bitwise, comparison and logical operators listed in
749 Table 289 are built into pgbench and may be used in expressions
750 appearing in \set. The operators are listed in increasing precedence
751 order. Except as noted, operators taking two numeric inputs will
752 produce a double value if either input is double, otherwise they
753 produce an integer result.
754 Table 289. pgbench Operators
756 ┌────────────────────────────────────────┐
757 │ │
758 │ Operator │
759 │ │
760 │ .PP Description │
761 │ │
762 │ .PP Example(s) │
763 ├────────────────────────────────────────┤
764 │ │
765 │ boolean OR boolean → boolean │
766 │ │
767 │ .PP Logical OR │
768 │ │
769 │ .PP 5 or 0 → TRUE │
770 ├────────────────────────────────────────┤
771 │ │
772 │ boolean AND boolean → boolean │
773 │ │
774 │ .PP Logical AND │
775 │ │
776 │ .PP 3 and 0 → FALSE │
777 ├────────────────────────────────────────┤
778 │ │
779 │ NOT boolean → boolean │
780 │ │
781 │ .PP Logical NOT │
782 │ │
783 │ .PP not false → TRUE │
784 ├────────────────────────────────────────┤
785 │ │
786 │ boolean IS [NOT] │
787 │ (NULL|TRUE|FALSE) → boolean │
788 │ │
789 │ .PP Boolean value tests │
790 │ │
791 │ .PP 1 is null → FALSE │
792 ├────────────────────────────────────────┤
793 │ │
794 │ value ISNULL|NOTNULL → boolean │
795 │ │
796 │ .PP Nullness tests │
797 │ │
798 │ .PP 1 notnull → TRUE │
799 ├────────────────────────────────────────┤
800 │ │
801 │ number = number → boolean │
802 │ │
803 │ .PP Equal │
804 │ │
805 │ .PP 5 = 4 → FALSE │
806 ├────────────────────────────────────────┤
807 │ │
808 │ number <> number → boolean │
809 │ │
810 │ .PP Not equal │
811 │ │
812 │ .PP 5 <> 4 → TRUE │
813 ├────────────────────────────────────────┤
814 │ │
815 │ number != number → boolean │
816 │ │
817 │ .PP Not equal │
818 │ │
819 │ .PP 5 != 5 → FALSE │
820 ├────────────────────────────────────────┤
821 │ │
822 │ number < number → boolean │
823 │ │
824 │ .PP Less than │
825 │ │
826 │ .PP 5 < 4 → FALSE │
827 ├────────────────────────────────────────┤
828 │ │
829 │ number <= number → boolean │
830 │ │
831 │ .PP Less than or equal to │
832 │ │
833 │ .PP 5 <= 4 → FALSE │
834 ├────────────────────────────────────────┤
835 │ │
836 │ number > number → boolean │
837 │ │
838 │ .PP Greater than │
839 │ │
840 │ .PP 5 > 4 → TRUE │
841 ├────────────────────────────────────────┤
842 │ │
843 │ number >= number → boolean │
844 │ │
845 │ .PP Greater than or equal │
846 │ to │
847 │ │
848 │ .PP 5 >= 4 → TRUE │
849 ├────────────────────────────────────────┤
850 │ │
851 │ integer | integer → integer │
852 │ │
853 │ .PP Bitwise OR │
854 │ │
855 │ .PP 1 | 2 → 3 │
856 ├────────────────────────────────────────┤
857 │ │
858 │ integer # integer → integer │
859 │ │
860 │ .PP Bitwise XOR │
861 │ │
862 │ .PP 1 # 3 → 2 │
863 ├────────────────────────────────────────┤
864 │ │
865 │ integer & integer → integer │
866 │ │
867 │ .PP Bitwise AND │
868 │ │
869 │ .PP 1 & 3 → 1 │
870 ├────────────────────────────────────────┤
871 │ │
872 │ ~ integer → integer │
873 │ │
874 │ .PP Bitwise NOT │
875 │ │
876 │ .PP ~ 1 → -2 │
877 ├────────────────────────────────────────┤
878 │ │
879 │ integer << integer → integer │
880 │ │
881 │ .PP Bitwise shift left │
882 │ │
883 │ .PP 1 << 2 → 4 │
884 ├────────────────────────────────────────┤
885 │ │
886 │ integer >> integer → integer │
887 │ │
888 │ .PP Bitwise shift right │
889 │ │
890 │ .PP 8 >> 2 → 2 │
891 ├────────────────────────────────────────┤
892 │ │
893 │ number + number → number │
894 │ │
895 │ .PP Addition │
896 │ │
897 │ .PP 5 + 4 → 9 │
898 ├────────────────────────────────────────┤
899 │ │
900 │ number - number → number │
901 │ │
902 │ .PP Subtraction │
903 │ │
904 │ .PP 3 - 2.0 → 1.0 │
905 ├────────────────────────────────────────┤
906 │ │
907 │ number * number → number │
908 │ │
909 │ .PP Multiplication │
910 │ │
911 │ .PP 5 * 4 → 20 │
912 ├────────────────────────────────────────┤
913 │ │
914 │ number / number → number │
915 │ │
916 │ .PP Division (truncates │
917 │ the result towards zero if both │
918 │ inputs are integers) │
919 │ │
920 │ .PP 5 / 3 → 1 │
921 ├────────────────────────────────────────┤
922 │ │
923 │ integer % integer → integer │
924 │ │
925 │ .PP Modulo (remainder) │
926 │ │
927 │ .PP 3 % 2 → 1 │
928 ├────────────────────────────────────────┤
929 │ │
930 │ - number → number │
931 │ │
932 │ .PP Negation │
933 │ │
934 │ .PP - 2.0 → -2.0 │
935 └────────────────────────────────────────┘
936 Built-In Functions
938 The functions listed in Table 290 are built into pgbench and may be
939 used in expressions appearing in \set.
940 Table 290. pgbench Functions
942 ┌────────────────────────────────────────┐
943 │ │
944 │ Function │
945 │ │
946 │ .PP Description │
947 │ │
948 │ .PP Example(s) │
949 ├────────────────────────────────────────┤
950 │ │
951 │ abs ( number ) → same type as │
952 │ input │
953 │ │
954 │ .PP Absolute value │
955 │ │
956 │ .PP abs(-17) → 17 │
957 ├────────────────────────────────────────┤
958 │ │
959 │ debug ( number ) → same type as │
960 │ input │
961 │ │
962 │ .PP Prints the argument │
963 │ to stderr, and returns the │
964 │ argument. │
965 │ │
966 │ .PP debug(5432.1) → │
967 │ 5432.1 │
968 ├────────────────────────────────────────┤
969 │ │
970 │ double ( number ) → double │
971 │ │
972 │ .PP Casts to double. │
973 │ │
974 │ .PP double(5432) → 5432.0 │
975 ├────────────────────────────────────────┤
976 │ │
977 │ exp ( number ) → double │
978 │ │
979 │ .PP Exponential (e raised │
980 │ to the given power) │
981 │ │
982 │ .PP exp(1.0) → │
983 │ 2.718281828459045 │
984 ├────────────────────────────────────────┤
985 │ │
986 │ greatest ( number [, ... ] ) → │
987 │ double if any argument is │
988 │ double, else integer │
989 │ │
990 │ .PP Selects the largest │
991 │ value among the arguments. │
992 │ │
993 │ .PP greatest(5, 4, 3, 2) │
994 │ → 5 │
995 ├────────────────────────────────────────┤
996 │ │
997 │ hash ( value [, seed ] ) → │
998 │ integer │
999 │ │
1000 │ .PP This is an alias for │
1001 │ hash_murmur2. │
1002 │ │
1003 │ .PP hash(10, 5432) → │
1004 │ -5817877081768721676 │
1005 ├────────────────────────────────────────┤
1006 │ │
1007 │ hash_fnv1a ( value [, seed ] ) → │
1008 │ integer │
1009 │ │
1010 │ .PP Computes FNV-1a hash. │
1011 │ │
1012 │ .PP hash_fnv1a(10, 5432) │
1013 │ → -7793829335365542153 │
1014 ├────────────────────────────────────────┤
1015 │ │
1016 │ hash_murmur2 ( value [, seed ] ) │
1017 │ → integer │
1018 │ │
1019 │ .PP Computes MurmurHash2 │
1020 │ hash. │
1021 │ │
1022 │ .PP hash_murmur2(10, │
1023 │ 5432) → -5817877081768721676 │
1024 ├────────────────────────────────────────┤
1025 │ │
1026 │ int ( number ) → integer │
1027 │ │
1028 │ .PP Casts to integer. │
1029 │ │
1030 │ .PP int(5.4 + 3.8) → 9 │
1031 ├────────────────────────────────────────┤
1032 │ │
1033 │ least ( number [, ... ] ) → │
1034 │ double if any argument is │
1035 │ double, else integer │
1036 │ │
1037 │ .PP Selects the smallest │
1038 │ value among the arguments. │
1039 │ │
1040 │ .PP least(5, 4, 3, 2.1) → │
1041 │ 2.1 │
1042 ├────────────────────────────────────────┤
1043 │ │
1044 │ ln ( number ) → double │
1045 │ │
1046 │ .PP Natural logarithm │
1047 │ │
1048 │ .PP ln(2.718281828459045) │
1049 │ → 1.0 │
1050 ├────────────────────────────────────────┤
1051 │ │
1052 │ mod ( integer, integer ) → │
1053 │ integer │
1054 │ │
1055 │ .PP Modulo (remainder) │
1056 │ │
1057 │ .PP mod(54, 32) → 22 │
1058 ├────────────────────────────────────────┤
1059 │ │
1060 │ permute ( i, size [, seed ] ) → │
1061 │ integer │
1062 │ │
1063 │ .PP Permuted value of i, │
1064 │ in the range [0, size). This is │
1065 │ the new position of i (modulo │
1066 │ size) in a pseudorandom │
1067 │ permutation of the integers │
1068 │ 0...size-1, parameterized by │
1069 │ seed, see below. │
1070 │ │
1071 │ .PP permute(0, 4) → an │
1072 │ integer between 0 and 3 │
1073 ├────────────────────────────────────────┤
1074 │ │
1075 │ pi () → double │
1076 │ │
1077 │ .PP Approximate value of │
1078 │ π │
1079 │ │
1080 │ .PP pi() → │
1081 │ 3.14159265358979323846 │
1082 ├────────────────────────────────────────┤
1083 │ │
1084 │ pow ( x, y ) → double │
1085 │ │
1086 │ .PP power ( x, y ) → │
1087 │ double │
1088 │ │
1089 │ .PP x raised to the power │
1090 │ of y │
1091 │ │
1092 │ .PP pow(2.0, 10) → 1024.0 │
1093 ├────────────────────────────────────────┤
1094 │ │
1095 │ random ( lb, ub ) → integer │
1096 │ │
1097 │ .PP Computes a │
1098 │ uniformly-distributed random │
1099 │ integer in [lb, ub]. │
1100 │ │
1101 │ .PP random(1, 10) → an │
1102 │ integer between 1 and 10 │
1103 ├────────────────────────────────────────┤
1104 │ │
1105 │ random_exponential ( lb, ub, │
1106 │ parameter ) → integer │
1107 │ │
1108 │ .PP Computes an │
1109 │ exponentially-distributed random │
1110 │ integer in [lb, ub], see below. │
1111 │ │
1112 │ .PP random_exponential(1, │
1113 │ 10, 3.0) → an integer between 1 │
1114 │ and 10 │
1115 ├────────────────────────────────────────┤
1116 │ │
1117 │ random_gaussian ( lb, ub, │
1118 │ parameter ) → integer │
1119 │ │
1120 │ .PP Computes a │
1121 │ Gaussian-distributed random │
1122 │ integer in [lb, ub], see below. │
1123 │ │
1124 │ .PP random_gaussian(1, │
1125 │ 10, 2.5) → an integer between 1 │
1126 │ and 10 │
1127 ├────────────────────────────────────────┤
1128 │ │
1129 │ random_zipfian ( lb, ub, │
1130 │ parameter ) → integer │
1131 │ │
1132 │ .PP Computes a │
1133 │ Zipfian-distributed random │
1134 │ integer in [lb, ub], see below. │
1135 │ │
1136 │ .PP random_zipfian(1, 10, │
1137 │ 1.5) → an integer between 1 and │
1138 │ 10 │
1139 ├────────────────────────────────────────┤
1140 │ │
1141 │ sqrt ( number ) → double │
1142 │ │
1143 │ .PP Square root │
1144 │ │
1145 │ .PP sqrt(2.0) → │
1146 │ 1.414213562 │
1147 └────────────────────────────────────────┘
1148 The random function generates values using a uniform distribution, that
1150 is all the values are drawn within the specified range with equal
1151 probability. The random_exponential, random_gaussian and random_zipfian
1152 functions require an additional double parameter which determines the
1153 precise shape of the distribution.
1154 • For an exponential distribution, parameter controls the
1156 distribution by truncating a quickly-decreasing exponential
1157 distribution at parameter, and then projecting onto integers
1158 between the bounds. To be precise, with
1159 f(x) = exp(-parameter * (x - min) / (max - min + 1)) / (1 - exp(-parameter))
1161 Then value i between min and max inclusive is drawn with
1163 probability: f(i) - f(i + 1).
1164 Intuitively, the larger the parameter, the more frequently values
1166 close to min are accessed, and the less frequently values close to
1167 max are accessed. The closer to 0 parameter is, the flatter (more
1168 uniform) the access distribution. A crude approximation of the
1169 distribution is that the most frequent 1% values in the range,
1170 close to min, are drawn parameter% of the time. The parameter value
1171 must be strictly positive.
1172 • For a Gaussian distribution, the interval is mapped onto a standard
1174 normal distribution (the classical bell-shaped Gaussian curve)
1175 truncated at -parameter on the left and +parameter on the right.
1176 Values in the middle of the interval are more likely to be drawn.
1177 To be precise, if PHI(x) is the cumulative distribution function of
1178 the standard normal distribution, with mean mu defined as (max +
1179 min) / 2.0, with
1180 f(x) = PHI(2.0 * parameter * (x - mu) / (max - min + 1)) /
1182 (2.0 * PHI(parameter) - 1)
1183 then value i between min and max inclusive is drawn with
1185 probability: f(i + 0.5) - f(i - 0.5). Intuitively, the larger the
1186 parameter, the more frequently values close to the middle of the
1187 interval are drawn, and the less frequently values close to the min
1188 and max bounds. About 67% of values are drawn from the middle 1.0 /
1189 parameter, that is a relative 0.5 / parameter around the mean, and
1190 95% in the middle 2.0 / parameter, that is a relative 1.0 /
1191 parameter around the mean; for instance, if parameter is 4.0, 67%
1192 of values are drawn from the middle quarter (1.0 / 4.0) of the
1193 interval (i.e., from 3.0 / 8.0 to 5.0 / 8.0) and 95% from the
1194 middle half (2.0 / 4.0) of the interval (second and third
1195 quartiles). The minimum allowed parameter value is 2.0.
1196 • random_zipfian generates a bounded Zipfian distribution. parameter
1198 defines how skewed the distribution is. The larger the parameter,
1199 the more frequently values closer to the beginning of the interval
1200 are drawn. The distribution is such that, assuming the range starts
1201 from 1, the ratio of the probability of drawing k versus drawing
1202 k+1 is ((k+1)/k)**parameter. For example, random_zipfian(1, ...,
1203 2.5) produces the value 1 about (2/1)**2.5 = 5.66 times more
1204 frequently than 2, which itself is produced (3/2)**2.5 = 2.76 times
1205 more frequently than 3, and so on.
1206 pgbench's implementation is based on "Non-Uniform Random Variate
1208 Generation", Luc Devroye, p. 550-551, Springer 1986. Due to
1209 limitations of that algorithm, the parameter value is restricted to
1210 the range [1.001, 1000].
1211 Note
1213 When designing a benchmark which selects rows non-uniformly, be
1214 aware that the rows chosen may be correlated with other data such
1215 as IDs from a sequence or the physical row ordering, which may skew
1216 performance measurements.
1217 To avoid this, you may wish to use the permute function, or some
1219 other additional step with similar effect, to shuffle the selected
1220 rows and remove such correlations.
1221 Hash functions hash, hash_murmur2 and hash_fnv1a accept an input value
1223 and an optional seed parameter. In case the seed isn't provided the
1224 value of :default_seed is used, which is initialized randomly unless
1225 set by the command-line -D option.
1226 permute accepts an input value, a size, and an optional seed parameter.
1228 It generates a pseudorandom permutation of integers in the range [0,
1229 size), and returns the index of the input value in the permuted values.
1230 The permutation chosen is parameterized by the seed, which defaults to
1231 :default_seed, if not specified. Unlike the hash functions, permute
1232 ensures that there are no collisions or holes in the output values.
1233 Input values outside the interval are interpreted modulo the size. The
1234 function raises an error if the size is not positive. permute can be
1235 used to scatter the distribution of non-uniform random functions such
1236 as random_zipfian or random_exponential so that values drawn more often
1237 are not trivially correlated. For instance, the following pgbench
1238 script simulates a possible real world workload typical for social
1239 media and blogging platforms where a few accounts generate excessive
1240 load:
1241 \set size 1000000
1243 \set r random_zipfian(1, :size, 1.07)
1244 \set k 1 + permute(:r, :size)
1245 In some cases several distinct distributions are needed which don't
1247 correlate with each other and this is when the optional seed parameter
1248 comes in handy:
1249 \set k1 1 + permute(:r, :size, :default_seed + 123)
1251 \set k2 1 + permute(:r, :size, :default_seed + 321)
1252 A similar behavior can also be approximated with hash:
1254 \set size 1000000
1256 \set r random_zipfian(1, 100 * :size, 1.07)
1257 \set k 1 + abs(hash(:r)) % :size
1258 However, since hash generates collisions, some values will not be
1260 reachable and others will be more frequent than expected from the
1261 original distribution.
1262 As an example, the full definition of the built-in TPC-B-like
1264 transaction is:
1265 \set aid random(1, 100000 * :scale)
1267 \set bid random(1, 1 * :scale)
1268 \set tid random(1, 10 * :scale)
1269 \set delta random(-5000, 5000)
1270 BEGIN;
1271 UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
1272 SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
1273 UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;
1274 UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;
1275 INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
1276 END;
1277 This script allows each iteration of the transaction to reference
1279 different, randomly-chosen rows. (This example also shows why it's
1280 important for each client session to have its own variables — otherwise
1281 they'd not be independently touching different rows.)
1282 Per-Transaction Logging
1284 With the -l option (but without the --aggregate-interval option),
1285 pgbench writes information about each transaction to a log file. The
1286 log file will be named prefix.nnn, where prefix defaults to
1287 pgbench_log, and nnn is the PID of the pgbench process. The prefix can
1288 be changed by using the --log-prefix option. If the -j option is 2 or
1289 higher, so that there are multiple worker threads, each will have its
1290 own log file. The first worker will use the same name for its log file
1291 as in the standard single worker case. The additional log files for the
1292 other workers will be named prefix.nnn.mmm, where mmm is a sequential
1293 number for each worker starting with 1.
1294 Each line in a log file describes one transaction. It contains the
1296 following space-separated fields:
1297 client_id
1299 identifies the client session that ran the transaction
1300 transaction_no
1302 counts how many transactions have been run by that session
1303 time
1305 transaction's elapsed time, in microseconds
1306 script_no
1308 identifies the script file that was used for the transaction
1309 (useful when multiple scripts are specified with -f or -b)
1310 time_epoch
1312 transaction's completion time, as a Unix-epoch time stamp
1313 time_us
1315 fractional-second part of transaction's completion time, in
1316 microseconds
1317 schedule_lag
1319 transaction start delay, that is the difference between the
1320 transaction's scheduled start time and the time it actually
1321 started, in microseconds (present only if --rate is specified)
1322 retries
1324 count of retries after serialization or deadlock errors during the
1325 transaction (present only if --max-tries is not equal to one)
1326 When both --rate and --latency-limit are used, the time for a skipped
1328 transaction will be reported as skipped. If the transaction ends with a
1329 failure, its time will be reported as failed. If you use the
1330 --failures-detailed option, the time of the failed transaction will be
1331 reported as serialization or deadlock depending on the type of failure
1332 (see Failures and Serialization/Deadlock Retries for more information).
1333 Here is a snippet of a log file generated in a single-client run:
1335 0 199 2241 0 1175850568 995598
1337 0 200 2465 0 1175850568 998079
1338 0 201 2513 0 1175850569 608
1339 0 202 2038 0 1175850569 2663
1340 Another example with --rate=100 and --latency-limit=5 (note the
1342 additional schedule_lag column):
1343 0 81 4621 0 1412881037 912698 3005
1345 0 82 6173 0 1412881037 914578 4304
1346 0 83 skipped 0 1412881037 914578 5217
1347 0 83 skipped 0 1412881037 914578 5099
1348 0 83 4722 0 1412881037 916203 3108
1349 0 84 4142 0 1412881037 918023 2333
1350 0 85 2465 0 1412881037 919759 740
1351 In this example, transaction 82 was late, because its latency (6.173
1353 ms) was over the 5 ms limit. The next two transactions were skipped,
1354 because they were already late before they were even started.
1355 The following example shows a snippet of a log file with failures and
1357 retries, with the maximum number of tries set to 10 (note the
1358 additional retries column):
1359 3 0 47423 0 1499414498 34501 3
1361 3 1 8333 0 1499414498 42848 0
1362 3 2 8358 0 1499414498 51219 0
1363 4 0 72345 0 1499414498 59433 6
1364 1 3 41718 0 1499414498 67879 4
1365 1 4 8416 0 1499414498 76311 0
1366 3 3 33235 0 1499414498 84469 3
1367 0 0 failed 0 1499414498 84905 9
1368 2 0 failed 0 1499414498 86248 9
1369 3 4 8307 0 1499414498 92788 0
1370 If the --failures-detailed option is used, the type of failure is
1372 reported in the time like this:
1373 3 0 47423 0 1499414498 34501 3
1375 3 1 8333 0 1499414498 42848 0
1376 3 2 8358 0 1499414498 51219 0
1377 4 0 72345 0 1499414498 59433 6
1378 1 3 41718 0 1499414498 67879 4
1379 1 4 8416 0 1499414498 76311 0
1380 3 3 33235 0 1499414498 84469 3
1381 0 0 serialization 0 1499414498 84905 9
1382 2 0 serialization 0 1499414498 86248 9
1383 3 4 8307 0 1499414498 92788 0
1384 When running a long test on hardware that can handle a lot of
1386 transactions, the log files can become very large. The --sampling-rate
1387 option can be used to log only a random sample of transactions.
1388 Aggregated Logging
1390 With the --aggregate-interval option, a different format is used for
1391 the log files. Each log line describes one aggregation interval. It
1392 contains the following space-separated fields:
1393 interval_start
1395 start time of the interval, as a Unix-epoch time stamp
1396 num_transactions
1398 number of transactions within the interval
1399 sum_latency
1401 sum of transaction latencies
1402 sum_latency_2
1404 sum of squares of transaction latencies
1405 min_latency
1407 minimum transaction latency
1408 max_latency
1410 maximum transaction latency
1411 sum_lag
1413 sum of transaction start delays (zero unless --rate is specified)
1414 sum_lag_2
1416 sum of squares of transaction start delays (zero unless --rate is
1417 specified)
1418 min_lag
1420 minimum transaction start delay (zero unless --rate is specified)
1421 max_lag
1423 maximum transaction start delay (zero unless --rate is specified)
1424 skipped
1426 number of transactions skipped because they would have started too
1427 late (zero unless --rate and --latency-limit are specified)
1428 retried
1430 number of retried transactions (zero unless --max-tries is not
1431 equal to one)
1432 retries
1434 number of retries after serialization or deadlock errors (zero
1435 unless --max-tries is not equal to one)
1436 serialization_failures
1438 number of transactions that got a serialization error and were not
1439 retried afterwards (zero unless --failures-detailed is specified)
1440 deadlock_failures
1442 number of transactions that got a deadlock error and were not
1443 retried afterwards (zero unless --failures-detailed is specified)
1444 Here is some example output generated with these options:
1446 pgbench --aggregate-interval=10 --time=20 --client=10 --log --rate=1000 --latency-limit=10 --failures-detailed --max-tries=10 test
1448 1650260552 5178 26171317 177284491527 1136 44462 2647617 7321113867 0 9866 64 7564 28340 4148 0
1450 1650260562 4808 25573984 220121792172 1171 62083 3037380 9666800914 0 9998 598 7392 26621 4527 0
1451 Notice that while the plain (unaggregated) log format shows which
1453 script was used for each transaction, the aggregated format does not.
1454 Therefore if you need per-script data, you need to aggregate the data
1455 on your own.
1456 Per-Statement Report
1458 With the -r option, pgbench collects the following statistics for each
1459 statement:
1460 • latency — elapsed transaction time for each statement. pgbench
1462 reports an average value of all successful runs of the statement.
1463 • The number of failures in this statement. See Failures and
1465 Serialization/Deadlock Retries for more information.
1466 • The number of retries after a serialization or a deadlock error in
1468 this statement. See Failures and Serialization/Deadlock Retries for
1469 more information.
1470 The report displays retry statistics only if the --max-tries option is
1472 not equal to 1.
1473 All values are computed for each statement executed by every client and
1475 are reported after the benchmark has finished.
1476 For the default script, the output will look similar to this:
1478 starting vacuum...end.
1480 transaction type: <builtin: TPC-B (sort of)>
1481 scaling factor: 1
1482 query mode: simple
1483 number of clients: 10
1484 number of threads: 1
1485 maximum number of tries: 1
1486 number of transactions per client: 1000
1487 number of transactions actually processed: 10000/10000
1488 number of failed transactions: 0 (0.000%)
1489 number of transactions above the 50.0 ms latency limit: 1311/10000 (13.110 %)
1490 latency average = 28.488 ms
1491 latency stddev = 21.009 ms
1492 initial connection time = 69.068 ms
1493 tps = 346.224794 (without initial connection time)
1494 statement latencies in milliseconds and failures:
1495 0.012 0 \set aid random(1, 100000 * :scale)
1496 0.002 0 \set bid random(1, 1 * :scale)
1497 0.002 0 \set tid random(1, 10 * :scale)
1498 0.002 0 \set delta random(-5000, 5000)
1499 0.319 0 BEGIN;
1500 0.834 0 UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
1501 0.641 0 SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
1502 11.126 0 UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;
1503 12.961 0 UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;
1504 0.634 0 INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
1505 1.957 0 END;
1506 Another example of output for the default script using serializable
1508 default transaction isolation level (PGOPTIONS='-c
1509 default_transaction_isolation=serializable' pgbench ...):
1510 starting vacuum...end.
1512 transaction type: <builtin: TPC-B (sort of)>
1513 scaling factor: 1
1514 query mode: simple
1515 number of clients: 10
1516 number of threads: 1
1517 maximum number of tries: 10
1518 number of transactions per client: 1000
1519 number of transactions actually processed: 6317/10000
1520 number of failed transactions: 3683 (36.830%)
1521 number of transactions retried: 7667 (76.670%)
1522 total number of retries: 45339
1523 number of transactions above the 50.0 ms latency limit: 106/6317 (1.678 %)
1524 latency average = 17.016 ms
1525 latency stddev = 13.283 ms
1526 initial connection time = 45.017 ms
1527 tps = 186.792667 (without initial connection time)
1528 statement latencies in milliseconds, failures and retries:
1529 0.006 0 0 \set aid random(1, 100000 * :scale)
1530 0.001 0 0 \set bid random(1, 1 * :scale)
1531 0.001 0 0 \set tid random(1, 10 * :scale)
1532 0.001 0 0 \set delta random(-5000, 5000)
1533 0.385 0 0 BEGIN;
1534 0.773 0 1 UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
1535 0.624 0 0 SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
1536 1.098 320 3762 UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;
1537 0.582 3363 41576 UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;
1538 0.465 0 0 INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
1539 1.933 0 0 END;
1540 If multiple script files are specified, all statistics are reported
1542 separately for each script file.
1543 Note that collecting the additional timing information needed for
1545 per-statement latency computation adds some overhead. This will slow
1546 average execution speed and lower the computed TPS. The amount of
1547 slowdown varies significantly depending on platform and hardware.
1548 Comparing average TPS values with and without latency reporting enabled
1549 is a good way to measure if the timing overhead is significant.
1550 Failures and Serialization/Deadlock Retries
1552 When executing pgbench, there are three main types of errors:
1553 • Errors of the main program. They are the most serious and always
1555 result in an immediate exit from pgbench with the corresponding
1556 error message. They include:
1557 • errors at the beginning of pgbench (e.g. an invalid option
1559 value);
1560 • errors in the initialization mode (e.g. the query to create
1562 tables for built-in scripts fails);
1563 • errors before starting threads (e.g. could not connect to the
1565 database server, syntax error in the meta command, thread
1566 creation failure);
1567 • internal pgbench errors (which are supposed to never occur...).
1569 • Errors when the thread manages its clients (e.g. the client could
1571 not start a connection to the database server / the socket for
1572 connecting the client to the database server has become invalid).
1573 In such cases all clients of this thread stop while other threads
1574 continue to work.
1575 • Direct client errors. They lead to immediate exit from pgbench with
1577 the corresponding error message only in the case of an internal
1578 pgbench error (which are supposed to never occur...). Otherwise in
1579 the worst case they only lead to the abortion of the failed client
1580 while other clients continue their run (but some client errors are
1581 handled without an abortion of the client and reported separately,
1582 see below). Later in this section it is assumed that the discussed
1583 errors are only the direct client errors and they are not internal
1584 pgbench errors.
1585 A client's run is aborted in case of a serious error; for example, the
1587 connection with the database server was lost or the end of script was
1588 reached without completing the last transaction. In addition, if
1589 execution of an SQL or meta command fails for reasons other than
1590 serialization or deadlock errors, the client is aborted. Otherwise, if
1591 an SQL command fails with serialization or deadlock errors, the client
1592 is not aborted. In such cases, the current transaction is rolled back,
1593 which also includes setting the client variables as they were before
1594 the run of this transaction (it is assumed that one transaction script
1595 contains only one transaction; see What Is the "Transaction" Actually
1596 Performed in pgbench? for more information). Transactions with
1597 serialization or deadlock errors are repeated after rollbacks until
1598 they complete successfully or reach the maximum number of tries
1599 (specified by the --max-tries option) / the maximum time of retries
1600 (specified by the --latency-limit option) / the end of benchmark
1601 (specified by the --time option). If the last trial run fails, this
1602 transaction will be reported as failed but the client is not aborted
1603 and continues to work.
1604 Note
1606 Without specifying the --max-tries option, a transaction will never
1607 be retried after a serialization or deadlock error because its
1608 default value is 1. Use an unlimited number of tries
1609 (--max-tries=0) and the --latency-limit option to limit only the
1610 maximum time of tries. You can also use the --time option to limit
1611 the benchmark duration under an unlimited number of tries.
1612 Be careful when repeating scripts that contain multiple
1614 transactions: the script is always retried completely, so
1615 successful transactions can be performed several times.
1616 Be careful when repeating transactions with shell commands. Unlike
1618 the results of SQL commands, the results of shell commands are not
1619 rolled back, except for the variable value of the \setshell
1620 command.
1621 The latency of a successful transaction includes the entire time of
1623 transaction execution with rollbacks and retries. The latency is
1624 measured only for successful transactions and commands but not for
1625 failed transactions or commands.
1626 The main report contains the number of failed transactions. If the
1628 --max-tries option is not equal to 1, the main report also contains
1629 statistics related to retries: the total number of retried transactions
1630 and total number of retries. The per-script report inherits all these
1631 fields from the main report. The per-statement report displays retry
1632 statistics only if the --max-tries option is not equal to 1.
1633 If you want to group failures by basic types in per-transaction and
1635 aggregation logs, as well as in the main and per-script reports, use
1636 the --failures-detailed option. If you also want to distinguish all
1637 errors and failures (errors without retrying) by type including which
1638 limit for retries was exceeded and how much it was exceeded by for the
1639 serialization/deadlock failures, use the --verbose-errors option.
1640 Good Practices
1642 It is very easy to use pgbench to produce completely meaningless
1643 numbers. Here are some guidelines to help you get useful results.
1644 In the first place, never believe any test that runs for only a few
1646 seconds. Use the -t or -T option to make the run last at least a few
1647 minutes, so as to average out noise. In some cases you could need hours
1648 to get numbers that are reproducible. It's a good idea to try the test
1649 run a few times, to find out if your numbers are reproducible or not.
1650 For the default TPC-B-like test scenario, the initialization scale
1652 factor (-s) should be at least as large as the largest number of
1653 clients you intend to test (-c); else you'll mostly be measuring update
1654 contention. There are only -s rows in the pgbench_branches table, and
1655 every transaction wants to update one of them, so -c values in excess
1656 of -s will undoubtedly result in lots of transactions blocked waiting
1657 for other transactions.
1658 The default test scenario is also quite sensitive to how long it's been
1660 since the tables were initialized: accumulation of dead rows and dead
1661 space in the tables changes the results. To understand the results you
1662 must keep track of the total number of updates and when vacuuming
1663 happens. If autovacuum is enabled it can result in unpredictable
1664 changes in measured performance.
1665 A limitation of pgbench is that it can itself become the bottleneck
1667 when trying to test a large number of client sessions. This can be
1668 alleviated by running pgbench on a different machine from the database
1669 server, although low network latency will be essential. It might even
1670 be useful to run several pgbench instances concurrently, on several
1671 client machines, against the same database server.
1672 Security
1674 If untrusted users have access to a database that has not adopted a
1675 secure schema usage pattern, do not run pgbench in that database.
1676 pgbench uses unqualified names and does not manipulate the search path.
1677PostgreSQL 15.4 2023 PGBENCH(1)