1PGBENCH(1) PostgreSQL 10.7 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 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 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 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 pgbench -i [ other-options ] dbname
50 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 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 At the default “scale factor” of 1, the tables initially contain this
62 many rows:
63 table # of rows
65 ---------------------------------
66 pgbench_branches 1
67 pgbench_tellers 10
68 pgbench_accounts 100000
69 pgbench_history 0
70 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 Once you have done the necessary setup, you can run your benchmark with
76 a command that doesn't include -i, that is
77 pgbench [ options ] dbname
79 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, some
88 options are useful in both cases.
89 Initialization Options
91 pgbench accepts the following command-line initialization arguments:
92 -i
94 --initialize
95 Required to invoke initialization mode.
96 -F fillfactor
98 --fillfactor=fillfactor
99 Create the pgbench_accounts, pgbench_tellers and pgbench_branches
100 tables with the given fillfactor. Default is 100.
101 -n
103 --no-vacuum
104 Perform no vacuuming after initialization.
105 -q
107 --quiet
108 Switch logging to quiet mode, producing only one progress message
109 per 5 seconds. The default logging prints one message each 100000
110 rows, which often outputs many lines per second (especially on good
111 hardware).
112 -s scale_factor
114 --scale=scale_factor
115 Multiply the number of rows generated by the scale factor. For
116 example, -s 100 will create 10,000,000 rows in the pgbench_accounts
117 table. Default is 1. When the scale is 20,000 or larger, the
118 columns used to hold account identifiers (aid columns) will switch
119 to using larger integers (bigint), in order to be big enough to
120 hold the range of account identifiers.
121 --foreign-keys
123 Create foreign key constraints between the standard tables.
124 --index-tablespace=index_tablespace
126 Create indexes in the specified tablespace, rather than the default
127 tablespace.
128 --tablespace=tablespace
130 Create tables in the specified tablespace, rather than the default
131 tablespace.
132 --unlogged-tables
134 Create all tables as unlogged tables, rather than permanent tables.
135 Benchmarking Options
137 pgbench accepts the following command-line benchmarking arguments:
138 -b scriptname[@weight]
140 --builtin=scriptname[@weight]
141 Add the specified built-in script to the list of executed scripts.
142 An optional integer weight after @ allows to adjust the probability
143 of drawing the script. If not specified, it is set to 1. Available
144 built-in scripts are: tpcb-like, simple-update and select-only.
145 Unambiguous prefixes of built-in names are accepted. With special
146 name list, show the list of built-in scripts and exit immediately.
147 -c clients
149 --client=clients
150 Number of clients simulated, that is, number of concurrent database
151 sessions. Default is 1.
152 -C
154 --connect
155 Establish a new connection for each transaction, rather than doing
156 it just once per client session. This is useful to measure the
157 connection overhead.
158 -d
160 --debug
161 Print debugging output.
162 -D varname=value
164 --define=varname=value
165 Define a variable for use by a custom script (see below). Multiple
166 -D options are allowed.
167 -f filename[@weight]
169 --file=filename[@weight]
170 Add a transaction script read from filename to the list of executed
171 scripts. An optional integer weight after @ allows to adjust the
172 probability of drawing the test. See below for details.
173 -j threads
175 --jobs=threads
176 Number of worker threads within pgbench. Using more than one thread
177 can be helpful on multi-CPU machines. Clients are distributed as
178 evenly as possible among available threads. Default is 1.
179 -l
181 --log
182 Write information about each transaction to a log file. See below
183 for details.
184 -L limit
186 --latency-limit=limit
187 Transaction which last more than limit milliseconds are counted and
188 reported separately, as late.
189 When throttling is used (--rate=...), transactions that lag behind
191 schedule by more than limit ms, and thus have no hope of meeting
192 the latency limit, are not sent to the server at all. They are
193 counted and reported separately as skipped.
194 -M querymode
196 --protocol=querymode
197 Protocol to use for submitting queries to the server:
198 · simple: use simple query protocol.
200 · extended: use extended query protocol.
202 · prepared: use extended query protocol with prepared statements.
204 The default is simple query protocol. (See Chapter 52 for more
206 information.)
207 -n
209 --no-vacuum
210 Perform no vacuuming before running the test. This option is
211 necessary if you are running a custom test scenario that does not
212 include the standard tables pgbench_accounts, pgbench_branches,
213 pgbench_history, and pgbench_tellers.
214 -N
216 --skip-some-updates
217 Run built-in simple-update script. Shorthand for -b simple-update.
218 -P sec
220 --progress=sec
221 Show progress report every sec seconds. The report includes the
222 time since the beginning of the run, the tps since the last report,
223 and the transaction latency average and standard deviation since
224 the last report. Under throttling (-R), the latency is computed
225 with respect to the transaction scheduled start time, not the
226 actual transaction beginning time, thus it also includes the
227 average schedule lag time.
228 -r
230 --report-latencies
231 Report the average per-statement latency (execution time from the
232 perspective of the client) of each command after the benchmark
233 finishes. See below for details.
234 -R rate
236 --rate=rate
237 Execute transactions targeting the specified rate instead of
238 running as fast as possible (the default). The rate is given in
239 transactions per second. If the targeted rate is above the maximum
240 possible rate, the rate limit won't impact the results.
241 The rate is targeted by starting transactions along a
243 Poisson-distributed schedule time line. The expected start time
244 schedule moves forward based on when the client first started, not
245 when the previous transaction ended. That approach means that when
246 transactions go past their original scheduled end time, it is
247 possible for later ones to catch up again.
248 When throttling is active, the transaction latency reported at the
250 end of the run is calculated from the scheduled start times, so it
251 includes the time each transaction had to wait for the previous
252 transaction to finish. The wait time is called the schedule lag
253 time, and its average and maximum are also reported separately. The
254 transaction latency with respect to the actual transaction start
255 time, i.e. the time spent executing the transaction in the
256 database, can be computed by subtracting the schedule lag time from
257 the reported latency.
258 If --latency-limit is used together with --rate, a transaction can
260 lag behind so much that it is already over the latency limit when
261 the previous transaction ends, because the latency is calculated
262 from the scheduled start time. Such transactions are not sent to
263 the server, but are skipped altogether and counted separately.
264 A high schedule lag time is an indication that the system cannot
266 process transactions at the specified rate, with the chosen number
267 of clients and threads. When the average transaction execution time
268 is longer than the scheduled interval between each transaction,
269 each successive transaction will fall further behind, and the
270 schedule lag time will keep increasing the longer the test run is.
271 When that happens, you will have to reduce the specified
272 transaction rate.
273 -s scale_factor
275 --scale=scale_factor
276 Report the specified scale factor in pgbench's output. With the
277 built-in tests, this is not necessary; the correct scale factor
278 will be detected by counting the number of rows in the
279 pgbench_branches table. However, when testing only custom
280 benchmarks (-f option), the scale factor will be reported as 1
281 unless this option is used.
282 -S
284 --select-only
285 Run built-in select-only script. Shorthand for -b select-only.
286 -t transactions
288 --transactions=transactions
289 Number of transactions each client runs. Default is 10.
290 -T seconds
292 --time=seconds
293 Run the test for this many seconds, rather than a fixed number of
294 transactions per client. -t and -T are mutually exclusive.
295 -v
297 --vacuum-all
298 Vacuum all four standard tables before running the test. With
299 neither -n nor -v, pgbench will vacuum the pgbench_tellers and
300 pgbench_branches tables, and will truncate pgbench_history.
301 --aggregate-interval=seconds
303 Length of aggregation interval (in seconds). May be used only with
304 -l option. With this option, the log contains per-interval summary
305 data, as described below.
306 --log-prefix=prefix
308 Set the filename prefix for the log files created by --log. The
309 default is pgbench_log.
310 --progress-timestamp
312 When showing progress (option -P), use a timestamp (Unix epoch)
313 instead of the number of seconds since the beginning of the run.
314 The unit is in seconds, with millisecond precision after the dot.
315 This helps compare logs generated by various tools.
316 --sampling-rate=rate
318 Sampling rate, used when writing data into the log, to reduce the
319 amount of log generated. If this option is given, only the
320 specified fraction of transactions are logged. 1.0 means all
321 transactions will be logged, 0.05 means only 5% of the transactions
322 will be logged.
323 Remember to take the sampling rate into account when processing the
325 log file. For example, when computing tps values, you need to
326 multiply the numbers accordingly (e.g. with 0.01 sample rate,
327 you'll only get 1/100 of the actual tps).
328 Common Options
330 pgbench accepts the following command-line common arguments:
331 -h hostname
333 --host=hostname
334 The database server's host name
335 -p port
337 --port=port
338 The database server's port number
339 -U login
341 --username=login
342 The user name to connect as
343 -V
345 --version
346 Print the pgbench version and exit.
347 -?
349 --help
350 Show help about pgbench command line arguments, and exit.
351
353 What is the “Transaction” Actually Performed in pgbench?
354 pgbench executes test scripts chosen randomly from a specified list.
355 They include built-in scripts with -b and user-provided custom scripts
356 with -f. Each script may be given a relative weight specified after a @
357 so as to change its drawing probability. The default weight is 1.
358 Scripts with a weight of 0 are ignored.
359 The default built-in transaction script (also invoked with -b
361 tpcb-like) issues seven commands per transaction over randomly chosen
362 aid, tid, bid and balance. The scenario is inspired by the TPC-B
363 benchmark, but is not actually TPC-B, hence the name.
364 1. BEGIN;
366 2. UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid
368 = :aid;
369 3. SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
371 4. UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid =
373 :tid;
374 5. UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid
376 = :bid;
377 6. INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES
379 (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
380 7. END;
382 If you select the simple-update built-in (also -N), steps 4 and 5
384 aren't included in the transaction. This will avoid update contention
385 on these tables, but it makes the test case even less like TPC-B.
386 If you select the select-only built-in (also -S), only the SELECT is
388 issued.
389 Custom Scripts
391 pgbench has support for running custom benchmark scenarios by replacing
392 the default transaction script (described above) with a transaction
393 script read from a file (-f option). In this case a “transaction”
394 counts as one execution of a script file.
395 A script file contains one or more SQL commands terminated by
397 semicolons. Empty lines and lines beginning with -- are ignored. Script
398 files can also contain “meta commands”, which are interpreted by
399 pgbench itself, as described below.
400 Note
402 Before PostgreSQL 9.6, SQL commands in script files were terminated
403 by newlines, and so they could not be continued across lines. Now a
404 semicolon is required to separate consecutive SQL commands (though
405 a SQL command does not need one if it is followed by a meta
406 command). If you need to create a script file that works with both
407 old and new versions of pgbench, be sure to write each SQL command
408 on a single line ending with a semicolon.
409 There is a simple variable-substitution facility for script files.
411 Variables can be set by the command-line -D option, explained above, or
412 by the meta commands explained below. In addition to any variables
413 preset by -D command-line options, there are a few variables that are
414 preset automatically, listed in Table 241. A value specified for these
415 variables using -D takes precedence over the automatic presets. Once
416 set, a variable's value can be inserted into a SQL command by writing
417 :variablename. When running more than one client session, each session
418 has its own set of variables.
419 Table 241. Automatic Variables
421 ┌──────────┬────────────────────────────┐
422 │Variable │ Description │
423 ├──────────┼────────────────────────────┤
424 │scale │ current scale factor │
425 ├──────────┼────────────────────────────┤
426 │client_id │ unique number identifying │
427 │ │ the client session (starts │
428 │ │ from zero) │
429 └──────────┴────────────────────────────┘
430 Script file meta commands begin with a backslash (\) and normally
432 extend to the end of the line, although they can be continued to
433 additional lines by writing backslash-return. Arguments to a meta
434 command are separated by white space. These meta commands are
435 supported:
436 \set varname expression
438 Sets variable varname to a value calculated from expression. The
439 expression may contain integer constants such as 5432, double
440 constants such as 3.14159, references to variables :variablename,
441 unary operators (+, -) and binary operators (+, -, *, /, %) with
442 their usual precedence and associativity, function calls, and
443 parentheses.
444 Examples:
446 \set ntellers 10 * :scale
448 \set aid (1021 * random(1, 100000 * :scale)) % \
449 (100000 * :scale) + 1
450 \sleep number [ us | ms | s ]
452 Causes script execution to sleep for the specified duration in
453 microseconds (us), milliseconds (ms) or seconds (s). If the unit is
454 omitted then seconds are the default. number can be either an
455 integer constant or a :variablename reference to a variable having
456 an integer value.
457 Example:
459 \sleep 10 ms
461 \setshell varname command [ argument ... ]
463 Sets variable varname to the result of the shell command command
464 with the given argument(s). The command must return an integer
465 value through its standard output.
466 command and each argument can be either a text constant or a
468 :variablename reference to a variable. If you want to use an
469 argument starting with a colon, write an additional colon at the
470 beginning of argument.
471 Example:
473 \setshell variable_to_be_assigned command literal_argument :variable ::literal_starting_with_colon
475 \shell command [ argument ... ]
477 Same as \setshell, but the result of the command is discarded.
478 Example:
480 \shell command literal_argument :variable ::literal_starting_with_colon
482 Built-In Functions
484 The functions listed in Table 242 are built into pgbench and may be
485 used in expressions appearing in \set.
486 Table 242. pgbench Functions
488 ┌───────────────────────┬───────────────┬───────────────────────────┬───────────────────────┬────────────────────────┐
489 │Function │ Return Type │ Description │ Example │ Result │
490 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
491 │abs(a) │ same as a │ absolute │ abs(-17) │ 17 │
492 │ │ │ value │ │ │
493 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
494 │debug(a) │ same as a │ print a to │ debug(5432.1) │ 5432.1 │
495 │ │ │ stderr, │ │ │
496 │ │ │ and │ │ │
497 │ │ │ return a │ │ │
498 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
499 │double(i) │ double │ cast to │ double(5432) │ 5432.0 │
500 │ │ │ double │ │ │
501 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
502 │greatest(a [, │ double if any │ largest value │ greatest(5, │ 5 │
503 │... ] ) │ a is double, │ among │ 4, 3, 2) │ │
504 │ │ else integer │ arguments │ │ │
505 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
506 │int(x) │ integer │ cast to int │ int(5.4 + │ 9 │
507 │ │ │ │ 3.8) │ │
508 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
509 │least(a [, │ double if any │ smallest │ least(5, 4, │ 2.1 │
510 │... ] ) │ a is double, │ value among │ 3, 2.1) │ │
511 │ │ else integer │ arguments │ │ │
512 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
513 │pi() │ double │ value of the │ pi() │ 3.14159265358979323846 │
514 │ │ │ constant PI │ │ │
515 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
516 │random(lb, │ integer │ uniformly-distributed │ random(1, 10) │ an integer between 1 │
517 │ub) │ │ random │ │ and 10 │
518 │ │ │ integer in │ │ │
519 │ │ │ [lb, ub] │ │ │
520 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
521 │random_exponential(lb, │ integer │ exponentially-distributed │ random_exponential(1, │ an integer between 1 │
522 │ub, │ │ random integer in │ 10, 3.0) │ and 10 │
523 │parameter) │ │ [lb, ub], │ │ │
524 │ │ │ see │ │ │
525 │ │ │ below │ │ │
526 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
527 │random_gaussian(lb, │ integer │ Gaussian-distributed │ random_gaussian(1, │ an integer between 1 │
528 │ub, parameter) │ │ random integer in [lb, │ 10, 2.5) │ and 10 │
529 │ │ │ ub], │ │ │
530 │ │ │ see below │ │ │
531 ├───────────────────────┼───────────────┼───────────────────────────┼───────────────────────┼────────────────────────┤
532 │sqrt(x) │ double │ square root │ sqrt(2.0) │ 1.414213562 │
533 └───────────────────────┴───────────────┴───────────────────────────┴───────────────────────┴────────────────────────┘
534 The random function generates values using a uniform distribution, that
536 is all the values are drawn within the specified range with equal
537 probability. The random_exponential and random_gaussian functions
538 require an additional double parameter which determines the precise
539 shape of the distribution.
540 · For an exponential distribution, parameter controls the
542 distribution by truncating a quickly-decreasing exponential
543 distribution at parameter, and then projecting onto integers
544 between the bounds. To be precise, with
545 f(x) = exp(-parameter * (x - min) / (max - min + 1)) / (1 - exp(-parameter))
547 Then value i between min and max inclusive is drawn with
549 probability: f(i) - f(i + 1).
550 Intuitively, the larger the parameter, the more frequently values
552 close to min are accessed, and the less frequently values close to
553 max are accessed. The closer to 0 parameter is, the flatter (more
554 uniform) the access distribution. A crude approximation of the
555 distribution is that the most frequent 1% values in the range,
556 close to min, are drawn parameter% of the time. The parameter value
557 must be strictly positive.
558 · For a Gaussian distribution, the interval is mapped onto a standard
560 normal distribution (the classical bell-shaped Gaussian curve)
561 truncated at -parameter on the left and +parameter on the right.
562 Values in the middle of the interval are more likely to be drawn.
563 To be precise, if PHI(x) is the cumulative distribution function of
564 the standard normal distribution, with mean mu defined as (max +
565 min) / 2.0, with
566 f(x) = PHI(2.0 * parameter * (x - mu) / (max - min + 1)) /
568 (2.0 * PHI(parameter) - 1)
569 then value i between min and max inclusive is drawn with
571 probability: f(i + 0.5) - f(i - 0.5). Intuitively, the larger the
572 parameter, the more frequently values close to the middle of the
573 interval are drawn, and the less frequently values close to the min
574 and max bounds. About 67% of values are drawn from the middle 1.0 /
575 parameter, that is a relative 0.5 / parameter around the mean, and
576 95% in the middle 2.0 / parameter, that is a relative 1.0 /
577 parameter around the mean; for instance, if parameter is 4.0, 67%
578 of values are drawn from the middle quarter (1.0 / 4.0) of the
579 interval (i.e. from 3.0 / 8.0 to 5.0 / 8.0) and 95% from the middle
580 half (2.0 / 4.0) of the interval (second and third quartiles). The
581 minimum parameter is 2.0 for performance of the Box-Muller
582 transform.
583 As an example, the full definition of the built-in TPC-B-like
585 transaction is:
586 \set aid random(1, 100000 * :scale)
588 \set bid random(1, 1 * :scale)
589 \set tid random(1, 10 * :scale)
590 \set delta random(-5000, 5000)
591 BEGIN;
592 UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
593 SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
594 UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;
595 UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;
596 INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
597 END;
598 This script allows each iteration of the transaction to reference
600 different, randomly-chosen rows. (This example also shows why it's
601 important for each client session to have its own variables — otherwise
602 they'd not be independently touching different rows.)
603 Per-Transaction Logging
605 With the -l option (but without the --aggregate-interval option),
606 pgbench writes information about each transaction to a log file. The
607 log file will be named prefix.nnn, where prefix defaults to
608 pgbench_log, and nnn is the PID of the pgbench process. The prefix can
609 be changed by using the --log-prefix option. If the -j option is 2 or
610 higher, so that there are multiple worker threads, each will have its
611 own log file. The first worker will use the same name for its log file
612 as in the standard single worker case. The additional log files for the
613 other workers will be named prefix.nnn.mmm, where mmm is a sequential
614 number for each worker starting with 1.
615 The format of the log is:
617 client_id transaction_no time script_no time_epoch time_us [ schedule_lag ]
619 where client_id indicates which client session ran the transaction,
621 transaction_no counts how many transactions have been run by that
622 session, time is the total elapsed transaction time in microseconds,
623 script_no identifies which script file was used (useful when multiple
624 scripts were specified with -f or -b), and time_epoch/time_us are a
625 Unix-epoch time stamp and an offset in microseconds (suitable for
626 creating an ISO 8601 time stamp with fractional seconds) showing when
627 the transaction completed. The schedule_lag field is the difference
628 between the transaction's scheduled start time, and the time it
629 actually started, in microseconds. It is only present when the --rate
630 option is used. When both --rate and --latency-limit are used, the time
631 for a skipped transaction will be reported as skipped.
632 Here is a snippet of a log file generated in a single-client run:
634 0 199 2241 0 1175850568 995598
636 0 200 2465 0 1175850568 998079
637 0 201 2513 0 1175850569 608
638 0 202 2038 0 1175850569 2663
639 Another example with --rate=100 and --latency-limit=5 (note the
641 additional schedule_lag column):
642 0 81 4621 0 1412881037 912698 3005
644 0 82 6173 0 1412881037 914578 4304
645 0 83 skipped 0 1412881037 914578 5217
646 0 83 skipped 0 1412881037 914578 5099
647 0 83 4722 0 1412881037 916203 3108
648 0 84 4142 0 1412881037 918023 2333
649 0 85 2465 0 1412881037 919759 740
650 In this example, transaction 82 was late, because its latency (6.173
652 ms) was over the 5 ms limit. The next two transactions were skipped,
653 because they were already late before they were even started.
654 When running a long test on hardware that can handle a lot of
656 transactions, the log files can become very large. The --sampling-rate
657 option can be used to log only a random sample of transactions.
658 Aggregated Logging
660 With the --aggregate-interval option, a different format is used for
661 the log files:
662 interval_start num_transactions sum_latency sum_latency_2 min_latency max_latency [ sum_lag sum_lag_2 min_lag max_lag [ skipped ] ]
664 where interval_start is the start of the interval (as a Unix epoch time
666 stamp), num_transactions is the number of transactions within the
667 interval, sum_latency is the sum of the transaction latencies within
668 the interval, sum_latency_2 is the sum of squares of the transaction
669 latencies within the interval, min_latency is the minimum latency
670 within the interval, and max_latency is the maximum latency within the
671 interval. The next fields, sum_lag, sum_lag_2, min_lag, and max_lag,
672 are only present if the --rate option is used. They provide statistics
673 about the time each transaction had to wait for the previous one to
674 finish, i.e. the difference between each transaction's scheduled start
675 time and the time it actually started. The very last field, skipped, is
676 only present if the --latency-limit option is used, too. It counts the
677 number of transactions skipped because they would have started too
678 late. Each transaction is counted in the interval when it was
679 committed.
680 Here is some example output:
682 1345828501 5601 1542744 483552416 61 2573
684 1345828503 7884 1979812 565806736 60 1479
685 1345828505 7208 1979422 567277552 59 1391
686 1345828507 7685 1980268 569784714 60 1398
687 1345828509 7073 1979779 573489941 236 1411
688 Notice that while the plain (unaggregated) log file shows which script
690 was used for each transaction, the aggregated log does not. Therefore
691 if you need per-script data, you need to aggregate the data on your
692 own.
693 Per-Statement Latencies
695 With the -r option, pgbench collects the elapsed transaction time of
696 each statement executed by every client. It then reports an average of
697 those values, referred to as the latency for each statement, after the
698 benchmark has finished.
699 For the default script, the output will look similar to this:
701 starting vacuum...end.
703 transaction type: <builtin: TPC-B (sort of)>
704 scaling factor: 1
705 query mode: simple
706 number of clients: 10
707 number of threads: 1
708 number of transactions per client: 1000
709 number of transactions actually processed: 10000/10000
710 latency average = 15.844 ms
711 latency stddev = 2.715 ms
712 tps = 618.764555 (including connections establishing)
713 tps = 622.977698 (excluding connections establishing)
714 script statistics:
715 - statement latencies in milliseconds:
716 0.002 \set aid random(1, 100000 * :scale)
717 0.005 \set bid random(1, 1 * :scale)
718 0.002 \set tid random(1, 10 * :scale)
719 0.001 \set delta random(-5000, 5000)
720 0.326 BEGIN;
721 0.603 UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
722 0.454 SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
723 5.528 UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;
724 7.335 UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;
725 0.371 INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
726 1.212 END;
727 If multiple script files are specified, the averages are reported
729 separately for each script file.
730 Note that collecting the additional timing information needed for
732 per-statement latency computation adds some overhead. This will slow
733 average execution speed and lower the computed TPS. The amount of
734 slowdown varies significantly depending on platform and hardware.
735 Comparing average TPS values with and without latency reporting enabled
736 is a good way to measure if the timing overhead is significant.
737 Good Practices
739 It is very easy to use pgbench to produce completely meaningless
740 numbers. Here are some guidelines to help you get useful results.
741 In the first place, never believe any test that runs for only a few
743 seconds. Use the -t or -T option to make the run last at least a few
744 minutes, so as to average out noise. In some cases you could need hours
745 to get numbers that are reproducible. It's a good idea to try the test
746 run a few times, to find out if your numbers are reproducible or not.
747 For the default TPC-B-like test scenario, the initialization scale
749 factor (-s) should be at least as large as the largest number of
750 clients you intend to test (-c); else you'll mostly be measuring update
751 contention. There are only -s rows in the pgbench_branches table, and
752 every transaction wants to update one of them, so -c values in excess
753 of -s will undoubtedly result in lots of transactions blocked waiting
754 for other transactions.
755 The default test scenario is also quite sensitive to how long it's been
757 since the tables were initialized: accumulation of dead rows and dead
758 space in the tables changes the results. To understand the results you
759 must keep track of the total number of updates and when vacuuming
760 happens. If autovacuum is enabled it can result in unpredictable
761 changes in measured performance.
762 A limitation of pgbench is that it can itself become the bottleneck
764 when trying to test a large number of client sessions. This can be
765 alleviated by running pgbench on a different machine from the database
766 server, although low network latency will be essential. It might even
767 be useful to run several pgbench instances concurrently, on several
768 client machines, against the same database server.
769 Security
771 If untrusted users have access to a database that has not adopted a
772 secure schema usage pattern, do not run pgbench in that database.
773 pgbench uses unqualified names and does not manipulate the search path.
774PostgreSQL 10.7 2019 PGBENCH(1)