| blob | 46240e3f72599e93e6267c7015bfda80f6a22272 |
1 <!--
2 doc/src/sgml/ref/pgbench.sgml
3 PostgreSQL documentation
4 -->
9 </indexterm>
11 <refmeta>
15 </refmeta>
17 <refnamediv>
20 </refnamediv>
22 <refsynopsisdiv>
23 <cmdsynopsis>
28 </cmdsynopsis>
29 <cmdsynopsis>
33 </cmdsynopsis>
34 </refsynopsisdiv>
36 <refsect1>
38 <para>
41 commands over and over, possibly in multiple concurrent database sessions,
42 and then calculates the average transaction rate (transactions per second).
46 However, it is easy to test other cases by writing your own transaction
47 script files.
48 </para>
50 <para>
53 <screen>
55 scaling factor: 10
56 query mode: simple
57 number of clients: 10
58 number of threads: 1
59 maximum number of tries: 1
60 number of transactions per client: 1000
63 latency average = 11.013 ms
64 latency stddev = 7.351 ms
65 initial connection time = 45.758 ms
66 tps = 896.967014 (without initial connection time)
67 </screen>
69 The first seven lines report some of the most important parameter
70 settings.
71 The sixth line reports the maximum number of tries for transactions with
73 for more information).
74 The eighth line reports the number of transactions completed
75 and intended (the latter being just the product of number of clients
76 and number of transactions per client); these will be equal unless the run
77 failed before completion or some SQL command(s) failed. (In
79 The next line reports the number of failed transactions due to
81 for more information).
82 The last line reports the number of transactions per second.
83 </para>
85 <para>
86 The default TPC-B-like transaction test requires specific tables to be
89 tables. (When you are testing a custom script, you don't need this
90 step, but will instead need to do whatever setup your test needs.)
91 Initialization looks like:
93 <programlisting>
94 pgbench -i <optional> <replaceable>other-options</replaceable> </optional> <replaceable>dbname</replaceable>
95 </programlisting>
100 connect to the database server.)
101 </para>
103 <caution>
104 <para>
108 destroying any existing tables of these names.
109 Be very careful to use another database if you have tables having these
110 names!
111 </para>
112 </caution>
114 <para>
116 contain this many rows:
117 <screen>
118 table # of rows
119 ---------------------------------
120 pgbench_branches 1
121 pgbench_tellers 10
122 pgbench_accounts 100000
123 pgbench_history 0
124 </screen>
125 You can (and, for most purposes, probably should) increase the number
128 </para>
130 <para>
131 Once you have done the necessary setup, you can run your benchmark
134 <programlisting>
135 pgbench <optional> <replaceable>options</replaceable> </optional> <replaceable>dbname</replaceable>
136 </programlisting>
138 In nearly all cases, you'll need some options to make a useful test.
142 See below for a full list.
143 </para>
144 </refsect1>
146 <refsect1>
149 <para>
150 The following is divided into three subsections. Different options are
151 used during database initialization and while running benchmarks, but some
152 options are useful in both cases.
153 </para>
158 <para>
160 initialization arguments:
162 <variablelist>
165 <term><option><optional>-d</optional> <replaceable class="parameter">dbname</replaceable></option></term>
166 <term><option><optional>--dbname=</optional><replaceable class="parameter">dbname</replaceable></option></term>
167 <listitem>
168 <para>
169 Specifies the name of the database to test in. If this is
170 not specified, the environment variable
172 user name specified for the connection is used.
173 </para>
174 </listitem>
175 </varlistentry>
180 <listitem>
181 <para>
182 Required to invoke initialization mode.
183 </para>
184 </listitem>
185 </varlistentry>
190 <listitem>
191 <para>
192 Perform just a selected set of the normal initialization steps.
194 initialization steps to be performed, using one character per step.
195 Each step is invoked in the specified order.
197 The available steps are:
199 <variablelist>
202 <listitem>
203 <para>
205 </para>
206 </listitem>
207 </varlistentry>
210 <listitem>
211 <para>
212 Create the tables used by the
218 </para>
219 </listitem>
220 </varlistentry>
222 <term><literal>g</literal> or <literal>G</literal> (Generate data, client-side or server-side)</term>
223 <listitem>
224 <para>
225 Generate data and load it into the standard tables,
226 replacing any data already present.
227 </para>
228 <para>
231 sent to the server. This uses the client/server bandwidth
234 with version 14 or later
240 tables.
241 </para>
242 <para>
245 client and then data is actually generated in the server.
246 No significant bandwidth is required for this variant, but
247 the server will do more work.
249 message while generating data.
250 </para>
251 <para>
252 The default initialization behavior uses client-side data
254 </para>
255 </listitem>
256 </varlistentry>
259 <listitem>
260 <para>
262 </para>
263 </listitem>
264 </varlistentry>
267 <listitem>
268 <para>
269 Create primary key indexes on the standard tables.
270 </para>
271 </listitem>
272 </varlistentry>
275 <listitem>
276 <para>
277 Create foreign key constraints between the standard tables.
278 (Note that this step is not performed by default.)
279 </para>
280 </listitem>
281 </varlistentry>
282 </variablelist></para>
283 </listitem>
284 </varlistentry>
289 <listitem>
290 <para>
294 Default is 100.
295 </para>
296 </listitem>
297 </varlistentry>
302 <listitem>
303 <para>
304 Perform no vacuuming during initialization.
307 </para>
308 </listitem>
309 </varlistentry>
314 <listitem>
315 <para>
316 Switch logging to quiet mode, producing only one progress message per 5
318 often outputs many lines per second (especially on good hardware).
319 </para>
320 <para>
323 </para>
324 </listitem>
325 </varlistentry>
330 <listitem>
331 <para>
332 Multiply the number of rows generated by the scale factor.
338 in order to be big enough to hold the range of account
339 identifiers.
340 </para>
341 </listitem>
342 </varlistentry>
346 <listitem>
347 <para>
348 Create foreign key constraints between the standard tables.
350 step sequence, if it is not already present.)
351 </para>
352 </listitem>
353 </varlistentry>
357 <listitem>
358 <para>
359 Create indexes in the specified tablespace, rather than the default
360 tablespace.
361 </para>
362 </listitem>
363 </varlistentry>
367 <listitem>
368 <para>
374 </para>
375 </listitem>
376 </varlistentry>
380 <listitem>
381 <para>
384 the scaled number of accounts.
386 </para>
387 </listitem>
388 </varlistentry>
392 <listitem>
393 <para>
394 Create tables in the specified tablespace, rather than the default
395 tablespace.
396 </para>
397 </listitem>
398 </varlistentry>
402 <listitem>
403 <para>
404 Create all tables as unlogged tables, rather than permanent tables.
405 </para>
406 </listitem>
407 </varlistentry>
409 </variablelist>
410 </para>
412 </refsect2>
417 <para>
419 benchmarking arguments:
421 <variablelist>
425 <listitem>
426 <para>
427 Add the specified built-in script to the list of scripts to be executed.
430 Unambiguous prefixes of built-in names are accepted.
432 and exit immediately.
433 </para>
434 <para>
436 adjust the probability of selecting this script versus other ones.
437 The default weight is 1.
438 See below for details.
439 </para>
440 </listitem>
441 </varlistentry>
446 <listitem>
447 <para>
448 Number of clients simulated, that is, number of concurrent database
449 sessions. Default is 1.
450 </para>
451 </listitem>
452 </varlistentry>
457 <listitem>
458 <para>
459 Establish a new connection for each transaction, rather than
460 doing it just once per client session.
461 This is useful to measure the connection overhead.
462 </para>
463 </listitem>
464 </varlistentry>
467 <term><option>-D</option> <replaceable>varname</replaceable><literal>=</literal><replaceable>value</replaceable></term>
468 <term><option>--define=</option><replaceable>varname</replaceable><literal>=</literal><replaceable>value</replaceable></term>
469 <listitem>
470 <para>
471 Define a variable for use by a custom script (see below).
473 </para>
474 </listitem>
475 </varlistentry>
480 <listitem>
481 <para>
483 to the list of scripts to be executed.
484 </para>
485 <para>
487 adjust the probability of selecting this script versus other ones.
488 The default weight is 1.
490 character, append a weight so that there is no ambiguity, for
492 See below for details.
493 </para>
494 </listitem>
495 </varlistentry>
500 <listitem>
501 <para>
503 Using more than one thread can be helpful on multi-CPU machines.
504 Clients are distributed as evenly as possible among available threads.
505 Default is 1.
506 </para>
507 </listitem>
508 </varlistentry>
513 <listitem>
514 <para>
515 Write information about each transaction to a log file.
516 See below for details.
517 </para>
518 </listitem>
519 </varlistentry>
524 <listitem>
525 <para>
528 </para>
529 <para>
532 have no hope of meeting the latency limit, are not sent to the server
533 at all. They are counted and reported separately as
535 </para>
536 <para>
538 which fails due to a serialization anomaly or from a deadlock will not
539 be retried if the total time of all its tries is greater than
543 transactions with serialization/deadlock errors are not retried. See
545 retrying such transactions.
546 </para>
547 </listitem>
548 </varlistentry>
553 <listitem>
554 <para>
555 Protocol to use for submitting queries to the server:
556 <itemizedlist>
557 <listitem>
559 </listitem>
560 <listitem>
562 </listitem>
563 <listitem>
565 </listitem>
566 </itemizedlist>
569 reuses the parse analysis result starting from the second query
571 than in other modes.
572 </para>
573 <para>
575 for more information.)
576 </para>
577 </listitem>
578 </varlistentry>
583 <listitem>
584 <para>
585 Perform no vacuuming before running the test.
587 if you are running a custom test scenario that does not include
591 </para>
592 </listitem>
593 </varlistentry>
598 <listitem>
599 <para>
600 Run built-in simple-update script.
602 </para>
603 </listitem>
604 </varlistentry>
609 <listitem>
610 <para>
612 includes the time since the beginning of the run, the TPS since the
613 last report, and the transaction latency average, standard deviation,
614 and the number of failed transactions since the last report. Under
616 to the transaction scheduled start time, not the actual transaction
617 beginning time, thus it also includes the average schedule lag time.
619 after serialization/deadlock errors, the report includes the number of
620 retried transactions and the sum of all retries.
621 </para>
622 </listitem>
623 </varlistentry>
628 <listitem>
629 <para>
630 Report the following statistics for each command after the benchmark
631 finishes: the average per-statement latency (execution time from the
632 perspective of the client), the number of failures, and the number of
633 retries after serialization or deadlock errors in this command. The
634 report displays retry statistics only if the
636 </para>
637 </listitem>
638 </varlistentry>
643 <listitem>
644 <para>
645 Execute transactions targeting the specified rate instead of running
646 as fast as possible (the default). The rate is given in transactions
647 per second. If the targeted rate is above the maximum possible rate,
648 the rate limit won't impact the results.
649 </para>
650 <para>
651 The rate is targeted by starting transactions along a
652 Poisson-distributed schedule time line. The expected start time
653 schedule moves forward based on when the client first started, not
654 when the previous transaction ended. That approach means that when
655 transactions go past their original scheduled end time, it is
656 possible for later ones to catch up again.
657 </para>
658 <para>
659 When throttling is active, the transaction latency reported at the
660 end of the run is calculated from the scheduled start times, so it
661 includes the time each transaction had to wait for the previous
662 transaction to finish. The wait time is called the schedule lag time,
663 and its average and maximum are also reported separately. The
664 transaction latency with respect to the actual transaction start time,
665 i.e., the time spent executing the transaction in the database, can be
666 computed by subtracting the schedule lag time from the reported
667 latency.
668 </para>
670 <para>
672 a transaction can lag behind so much that it is already over the
673 latency limit when the previous transaction ends, because the latency
674 is calculated from the scheduled start time. Such transactions are
675 not sent to the server, but are skipped altogether and counted
676 separately.
677 </para>
679 <para>
680 A high schedule lag time is an indication that the system cannot
681 process transactions at the specified rate, with the chosen number of
682 clients and threads. When the average transaction execution time is
683 longer than the scheduled interval between each transaction, each
684 successive transaction will fall further behind, and the schedule lag
685 time will keep increasing the longer the test run is. When that
686 happens, you will have to reduce the specified transaction rate.
687 </para>
688 </listitem>
689 </varlistentry>
694 <listitem>
695 <para>
697 output. With the built-in tests, this is not necessary; the
698 correct scale factor will be detected by counting the number of
701 the scale factor will be reported as 1 unless this option is used.
702 </para>
703 </listitem>
704 </varlistentry>
709 <listitem>
710 <para>
711 Run built-in select-only script.
713 </para>
714 </listitem>
715 </varlistentry>
720 <listitem>
721 <para>
722 Number of transactions each client runs. Default is 10.
723 </para>
724 </listitem>
725 </varlistentry>
730 <listitem>
731 <para>
732 Run the test for this many seconds, rather than a fixed number of
735 </para>
736 </listitem>
737 </varlistentry>
742 <listitem>
743 <para>
744 Vacuum all four standard tables before running the test.
745 With neither <option>-n</option> nor <option>-v</option>, <application>pgbench</application> will vacuum the
748 </para>
749 </listitem>
750 </varlistentry>
754 <listitem>
755 <para>
756 Length of aggregation interval (in seconds). May be used only
758 per-interval summary data, as described below.
759 </para>
760 </listitem>
761 </varlistentry>
765 <listitem>
766 <para>
767 Exit immediately when any client is aborted due to some error. Without
768 this option, even when a client is aborted, other clients could
771 will print an incomplete results in this case.
772 </para>
773 <para>
774 Note that serialization failures or deadlock failures do not abort the
775 client, so they are not affected by this option.
777 </para>
778 </listitem>
779 </varlistentry>
783 <listitem>
784 <para>
785 Report failures in per-transaction and aggregation logs, as well as in
786 the main and per-script reports, grouped by the following types:
787 <itemizedlist>
788 <listitem>
790 </listitem>
791 <listitem>
793 </listitem>
794 </itemizedlist>
796 </para>
797 </listitem>
798 </varlistentry>
802 <listitem>
803 <para>
804 Set the filename prefix for the log files created by
806 </para>
807 </listitem>
808 </varlistentry>
812 <listitem>
813 <para>
814 Enable retries for transactions with serialization/deadlock errors and
815 set the maximum number of these tries. This option can be combined with
817 of all transaction tries; moreover, you cannot use an unlimited number
820 The default value is 1 and transactions with serialization/deadlock
822 for more information about retrying such transactions.
823 </para>
824 </listitem>
825 </varlistentry>
829 <listitem>
830 <para>
832 (Unix epoch) instead of the number of seconds since the
833 beginning of the run. The unit is in seconds, with millisecond
834 precision after the dot.
835 This helps compare logs generated by various tools.
836 </para>
837 </listitem>
838 </varlistentry>
842 <listitem>
843 <para>
844 Set random generator seed. Seeds the system random number generator,
845 which then produces a sequence of initial generator states, one for
846 each thread.
850 is available), or an unsigned decimal integer value.
851 The random generator is invoked explicitly from a pgbench script
854 When explicitly set, the value used for seeding is shown on the terminal.
856 provided through the environment variable
858 To ensure that the provided seed impacts all possible uses, put this option
859 first or use the environment variable.
860 </para>
861 <para>
863 run exactly, as far as random numbers are concerned.
864 As the random state is managed per thread, this means the exact same
866 client per thread and there are no external or data dependencies.
867 From a statistical viewpoint reproducing runs exactly is a bad idea because
868 it can hide the performance variability or improve performance unduly,
869 e.g., by hitting the same pages as a previous run.
870 However, it may also be of great help for debugging, for instance
871 re-running a tricky case which leads to an error.
872 Use wisely.
873 </para>
874 </listitem>
875 </varlistentry>
879 <listitem>
880 <para>
881 Sampling rate, used when writing data into the log, to reduce the
882 amount of log generated. If this option is given, only the specified
883 fraction of transactions are logged. 1.0 means all transactions will
885 </para>
886 <para>
887 Remember to take the sampling rate into account when processing the
888 log file. For example, when computing TPS values, you need to multiply
889 the numbers accordingly (e.g., with 0.01 sample rate, you'll only get
891 </para>
892 </listitem>
893 </varlistentry>
897 <listitem>
898 <para>
900 on stderr, and exit immediately.
901 </para>
902 </listitem>
903 </varlistentry>
907 <listitem>
908 <para>
909 Print messages about all errors and failures (errors without retrying)
910 including which limit for retries was exceeded and how far it was
911 exceeded for the serialization/deadlock failures. (Note that in this
912 case the output can be significantly increased.).
914 </para>
915 </listitem>
916 </varlistentry>
918 </variablelist>
919 </para>
921 </refsect2>
926 <para>
928 arguments for connection parameters and other common settings:
930 <variablelist>
933 <listitem>
934 <para>
935 Print debugging output.
936 </para>
937 </listitem>
938 </varlistentry>
943 <listitem>
944 <para>
945 The database server's host name
946 </para>
947 </listitem>
948 </varlistentry>
953 <listitem>
954 <para>
955 The database server's port number
956 </para>
957 </listitem>
958 </varlistentry>
963 <listitem>
964 <para>
965 The user name to connect as
966 </para>
967 </listitem>
968 </varlistentry>
973 <listitem>
974 <para>
976 </para>
977 </listitem>
978 </varlistentry>
983 <listitem>
984 <para>
986 arguments, and exit.
987 </para>
988 </listitem>
989 </varlistentry>
990 </variablelist>
991 </para>
993 </refsect2>
994 </refsect1>
996 <refsect1>
999 <para>
1001 problems such as invalid command-line options or internal errors which
1002 are supposed to never occur. Early errors that occur when starting
1003 benchmark such as initial connection failures also exit with status 1.
1004 Errors during the run such as database errors or problems in the script
1005 will result in exit status 2. In the latter case,
1008 </para>
1009 </refsect1>
1011 <refsect1>
1014 <variablelist>
1021 <listitem>
1022 <para>
1023 Default connection parameters.
1024 </para>
1025 </listitem>
1026 </varlistentry>
1027 </variablelist>
1029 <para>
1033 </para>
1035 <para>
1037 color in diagnostic messages. Possible values are
1040 </para>
1041 </refsect1>
1043 <refsect1>
1046 <refsect2 id="transactions-and-scripts" xreflabel="What Is the "Transaction" Actually Performed in pgbench?">
1047 <title>What Is the <quote>Transaction</quote> Actually Performed in <application>pgbench</application>?</title>
1049 <para>
1051 from a specified list.
1054 Each script may be given a relative weight specified after an
1058 </para>
1060 <para>
1064 The scenario is inspired by the TPC-B benchmark, but is not actually TPC-B,
1065 hence the name.
1066 </para>
1068 <orderedlist>
1070 <listitem><para><literal>UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;</literal></para></listitem>
1071 <listitem><para><literal>SELECT abalance FROM pgbench_accounts WHERE aid = :aid;</literal></para></listitem>
1072 <listitem><para><literal>UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;</literal></para></listitem>
1073 <listitem><para><literal>UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;</literal></para></listitem>
1074 <listitem><para><literal>INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);</literal></para></listitem>
1076 </orderedlist>
1078 <para>
1081 This will avoid update contention on these tables, but
1082 it makes the test case even less like TPC-B.
1083 </para>
1085 <para>
1088 </para>
1089 </refsect2>
1091 <refsect2>
1094 <para>
1096 benchmark scenarios by replacing the default transaction script
1097 (described above) with a transaction script read from a file
1099 counts as one execution of a script file.
1100 </para>
1102 <para>
1103 A script file contains one or more SQL commands terminated by
1104 semicolons. Empty lines and lines beginning with
1107 itself, as described below.
1108 </para>
1110 <note>
1111 <para>
1113 were terminated by newlines, and so they could not be continued across
1115 SQL commands (though an SQL command does not need one if it is followed
1116 by a meta command). If you need to create a script file that works with
1118 each SQL command on a single line ending with a semicolon.
1119 </para>
1120 <para>
1122 incomplete blocks of SQL
1123 transactions. If at runtime the client reaches the end of the script without
1124 completing the last transaction block, it will be aborted.
1125 </para>
1126 </note>
1128 <para>
1129 There is a simple variable-substitution facility for script files.
1130 Variable names must consist of letters (including non-Latin letters),
1131 digits, and underscores, with the first character not being a digit.
1133 explained above, or by the meta commands explained below.
1135 there are a few variables that are preset automatically, listed in
1138 Once set, a variable's
1139 value can be inserted into an SQL command by writing
1141 one client session, each session has its own set of variables.
1143 statement.
1144 </para>
1151 <thead>
1152 <row>
1155 </row>
1156 </thead>
1158 <tbody>
1159 <row>
1162 </row>
1164 <row>
1167 </row>
1169 <row>
1172 </row>
1174 <row>
1177 </row>
1178 </tbody>
1179 </tgroup>
1180 </table>
1182 <para>
1184 normally extend to the end of the line, although they can be continued
1185 to additional lines by writing backslash-return.
1186 Arguments to a meta command are separated by white space.
1187 These meta commands are supported:
1188 </para>
1190 <variablelist>
1192 <term>
1195 </term>
1197 <listitem>
1198 <para>
1199 These commands may be used to end SQL queries, taking the place of the
1201 </para>
1203 <para>
1205 expected to return one row, the columns of which are stored into variables
1207 if provided.
1208 </para>
1210 <para>
1214 if provided. If a query returns no row, no assignment is made and the variable
1215 can be tested for existence to detect this. If a query returns more than one
1216 row, the last value is kept.
1217 </para>
1219 <para>
1221 pipeline mode, since the query results are not yet available by the time
1222 the commands would need them.
1223 </para>
1225 <para>
1226 The following example puts the final account balance from the first query
1229 with integers from the third query.
1230 The result of the second query is discarded.
1231 The result of the two last combined queries are stored in variables
1233 <programlisting>
1234 UPDATE pgbench_accounts
1235 SET abalance = abalance + :delta
1236 WHERE aid = :aid
1237 RETURNING abalance \gset
1238 -- compound of two queries
1239 SELECT 1 \;
1242 </programlisting></para>
1243 </listitem>
1244 </varlistentry>
1251 <listitem>
1252 <para>
1253 This group of commands implements nestable conditional blocks,
1256 with non-zero values interpreted as true.
1257 </para>
1258 </listitem>
1259 </varlistentry>
1262 <term>
1263 <literal>\set <replaceable>varname</replaceable> <replaceable>expression</replaceable></literal>
1264 </term>
1266 <listitem>
1267 <para>
1276 with their usual SQL precedence and associativity,
1279 expressions</link> and parentheses.
1280 </para>
1282 <para>
1285 </para>
1287 <para>
1288 For conditional purposes, non zero numerical values are
1291 </para>
1293 <para>
1294 Too large or small integer and double constants, as well as
1297 raise errors on overflows.
1298 </para>
1300 <para>
1303 </para>
1305 <para>
1306 Examples:
1307 <programlisting>
1308 \set ntellers 10 * :scale
1312 </programlisting></para>
1313 </listitem>
1314 </varlistentry>
1317 <term>
1319 </term>
1321 <listitem>
1322 <para>
1323 Causes script execution to sleep for the specified duration in
1328 having an integer value.
1329 </para>
1331 <para>
1332 Example:
1333 <programlisting>
1334 \sleep 10 ms
1335 </programlisting></para>
1336 </listitem>
1337 </varlistentry>
1340 <term>
1341 <literal>\setshell <replaceable>varname</replaceable> <replaceable>command</replaceable> [ <replaceable>argument</replaceable> ... ]</literal>
1342 </term>
1344 <listitem>
1345 <para>
1348 The command must return an integer value through its standard output.
1349 </para>
1351 <para>
1355 with a colon, write an additional colon at the beginning of
1357 </para>
1359 <para>
1360 Example:
1361 <programlisting>
1362 \setshell variable_to_be_assigned command literal_argument :variable ::literal_starting_with_colon
1363 </programlisting></para>
1364 </listitem>
1365 </varlistentry>
1368 <term>
1369 <literal>\shell <replaceable>command</replaceable> [ <replaceable>argument</replaceable> ... ]</literal>
1370 </term>
1372 <listitem>
1373 <para>
1375 is discarded.
1376 </para>
1378 <para>
1379 Example:
1380 <programlisting>
1381 \shell command literal_argument :variable ::literal_starting_with_colon
1382 </programlisting></para>
1383 </listitem>
1384 </varlistentry>
1391 <listitem>
1392 <para>
1393 This group of commands implements pipelining of SQL statements.
1398 without ending the ongoing pipeline and flushing the send buffer.
1399 In pipeline mode, statements are sent to the server without waiting
1400 for the results of previous statements. See
1402 Pipeline mode requires the use of extended query protocol.
1403 </para>
1404 </listitem>
1405 </varlistentry>
1407 </variablelist>
1408 </refsect2>
1413 <para>
1414 The arithmetic, bitwise, comparison and logical operators listed in
1416 and may be used in expressions appearing in
1418 The operators are listed in increasing precedence order.
1419 Except as noted, operators taking two numeric inputs will produce
1420 a double value if either input is double, otherwise they produce
1421 an integer result.
1422 </para>
1427 <thead>
1428 <row>
1430 Operator
1431 </para>
1432 <para>
1433 Description
1434 </para>
1435 <para>
1436 Example(s)
1437 </para></entry>
1438 </row>
1439 </thead>
1441 <tbody>
1442 <row>
1446 </para>
1447 <para>
1448 Logical OR
1449 </para>
1450 <para>
1453 </para></entry>
1454 </row>
1456 <row>
1460 </para>
1461 <para>
1462 Logical AND
1463 </para>
1464 <para>
1467 </para></entry>
1468 </row>
1470 <row>
1474 </para>
1475 <para>
1476 Logical NOT
1477 </para>
1478 <para>
1481 </para></entry>
1482 </row>
1484 <row>
1488 </para>
1489 <para>
1490 Boolean value tests
1491 </para>
1492 <para>
1495 </para></entry>
1496 </row>
1498 <row>
1502 </para>
1503 <para>
1504 Nullness tests
1505 </para>
1506 <para>
1509 </para></entry>
1510 </row>
1512 <row>
1516 </para>
1517 <para>
1518 Equal
1519 </para>
1520 <para>
1523 </para></entry>
1524 </row>
1526 <row>
1528 <replaceable>number</replaceable> <literal><></literal> <replaceable>number</replaceable>
1530 </para>
1531 <para>
1532 Not equal
1533 </para>
1534 <para>
1537 </para></entry>
1538 </row>
1540 <row>
1544 </para>
1545 <para>
1546 Not equal
1547 </para>
1548 <para>
1551 </para></entry>
1552 </row>
1554 <row>
1558 </para>
1559 <para>
1560 Less than
1561 </para>
1562 <para>
1565 </para></entry>
1566 </row>
1568 <row>
1572 </para>
1573 <para>
1574 Less than or equal to
1575 </para>
1576 <para>
1579 </para></entry>
1580 </row>
1582 <row>
1586 </para>
1587 <para>
1588 Greater than
1589 </para>
1590 <para>
1593 </para></entry>
1594 </row>
1596 <row>
1600 </para>
1601 <para>
1602 Greater than or equal to
1603 </para>
1604 <para>
1607 </para></entry>
1608 </row>
1610 <row>
1614 </para>
1615 <para>
1616 Bitwise OR
1617 </para>
1618 <para>
1621 </para></entry>
1622 </row>
1624 <row>
1628 </para>
1629 <para>
1630 Bitwise XOR
1631 </para>
1632 <para>
1635 </para></entry>
1636 </row>
1638 <row>
1642 </para>
1643 <para>
1644 Bitwise AND
1645 </para>
1646 <para>
1649 </para></entry>
1650 </row>
1652 <row>
1656 </para>
1657 <para>
1658 Bitwise NOT
1659 </para>
1660 <para>
1663 </para></entry>
1664 </row>
1666 <row>
1668 <replaceable>integer</replaceable> <literal><<</literal> <replaceable>integer</replaceable>
1670 </para>
1671 <para>
1672 Bitwise shift left
1673 </para>
1674 <para>
1677 </para></entry>
1678 </row>
1680 <row>
1682 <replaceable>integer</replaceable> <literal>>></literal> <replaceable>integer</replaceable>
1684 </para>
1685 <para>
1686 Bitwise shift right
1687 </para>
1688 <para>
1691 </para></entry>
1692 </row>
1694 <row>
1698 </para>
1699 <para>
1700 Addition
1701 </para>
1702 <para>
1705 </para></entry>
1706 </row>
1708 <row>
1712 </para>
1713 <para>
1714 Subtraction
1715 </para>
1716 <para>
1719 </para></entry>
1720 </row>
1722 <row>
1726 </para>
1727 <para>
1728 Multiplication
1729 </para>
1730 <para>
1733 </para></entry>
1734 </row>
1736 <row>
1740 </para>
1741 <para>
1742 Division (truncates the result towards zero if both inputs are integers)
1743 </para>
1744 <para>
1747 </para></entry>
1748 </row>
1750 <row>
1754 </para>
1755 <para>
1756 Modulo (remainder)
1757 </para>
1758 <para>
1761 </para></entry>
1762 </row>
1764 <row>
1768 </para>
1769 <para>
1770 Negation
1771 </para>
1772 <para>
1775 </para></entry>
1776 </row>
1777 </tbody>
1778 </tgroup>
1779 </table>
1780 </refsect2>
1785 <para>
1789 </para>
1791 <!-- list pgbench functions in alphabetical order -->
1795 <thead>
1796 <row>
1798 Function
1799 </para>
1800 <para>
1801 Description
1802 </para>
1803 <para>
1804 Example(s)
1805 </para></entry>
1806 </row>
1807 </thead>
1809 <tbody>
1810 <row>
1813 <returnvalue></returnvalue> same type as input
1814 </para>
1815 <para>
1816 Absolute value
1817 </para>
1818 <para>
1821 </para></entry>
1822 </row>
1824 <row>
1827 <returnvalue></returnvalue> same type as input
1828 </para>
1829 <para>
1831 and returns the argument.
1832 </para>
1833 <para>
1836 </para></entry>
1837 </row>
1839 <row>
1843 </para>
1844 <para>
1845 Casts to double.
1846 </para>
1847 <para>
1850 </para></entry>
1851 </row>
1853 <row>
1857 </para>
1858 <para>
1860 </para>
1861 <para>
1864 </para></entry>
1865 </row>
1867 <row>
1869 <function>greatest</function> ( <replaceable>number</replaceable> <optional>, <literal>...</literal> </optional> )
1870 <returnvalue></returnvalue> <type>double</type> if any argument is double, else <type>integer</type>
1871 </para>
1872 <para>
1873 Selects the largest value among the arguments.
1874 </para>
1875 <para>
1878 </para></entry>
1879 </row>
1881 <row>
1883 <function>hash</function> ( <parameter>value</parameter> <optional>, <parameter>seed</parameter> </optional> )
1885 </para>
1886 <para>
1888 </para>
1889 <para>
1892 </para></entry>
1893 </row>
1895 <row>
1897 <function>hash_fnv1a</function> ( <parameter>value</parameter> <optional>, <parameter>seed</parameter> </optional> )
1899 </para>
1900 <para>
1901 Computes <ulink url="https://en.wikipedia.org/wiki/Fowler%E2%80%93Noll%E2%80%93Vo_hash_function">FNV-1a hash</ulink>.
1902 </para>
1903 <para>
1906 </para></entry>
1907 </row>
1909 <row>
1911 <function>hash_murmur2</function> ( <parameter>value</parameter> <optional>, <parameter>seed</parameter> </optional> )
1913 </para>
1914 <para>
1916 </para>
1917 <para>
1920 </para></entry>
1921 </row>
1923 <row>
1927 </para>
1928 <para>
1929 Casts to integer.
1930 </para>
1931 <para>
1934 </para></entry>
1935 </row>
1937 <row>
1939 <function>least</function> ( <replaceable>number</replaceable> <optional>, <literal>...</literal> </optional> )
1940 <returnvalue></returnvalue> <type>double</type> if any argument is double, else <type>integer</type>
1941 </para>
1942 <para>
1943 Selects the smallest value among the arguments.
1944 </para>
1945 <para>
1948 </para></entry>
1949 </row>
1951 <row>
1955 </para>
1956 <para>
1957 Natural logarithm
1958 </para>
1959 <para>
1962 </para></entry>
1963 </row>
1965 <row>
1967 <function>mod</function> ( <replaceable>integer</replaceable>, <replaceable>integer</replaceable> )
1969 </para>
1970 <para>
1971 Modulo (remainder)
1972 </para>
1973 <para>
1976 </para></entry>
1977 </row>
1979 <row>
1981 <function>permute</function> ( <parameter>i</parameter>, <parameter>size</parameter> [, <parameter>seed</parameter> ] )
1983 </para>
1984 <para>
1990 </para>
1991 <para>
1994 </para></entry>
1995 </row>
1997 <row>
2001 </para>
2002 <para>
2004 </para>
2005 <para>
2008 </para></entry>
2009 </row>
2011 <row>
2015 </para>
2019 </para>
2020 <para>
2022 </para>
2023 <para>
2026 </para></entry>
2027 </row>
2029 <row>
2033 </para>
2034 <para>
2035 Computes a uniformly-distributed random integer in <literal>[lb,
2036 ub]</literal>.
2037 </para>
2038 <para>
2041 </para></entry>
2042 </row>
2044 <row>
2046 <function>random_exponential</function> ( <parameter>lb</parameter>, <parameter>ub</parameter>, <parameter>parameter</parameter> )
2048 </para>
2049 <para>
2050 Computes an exponentially-distributed random integer in <literal>[lb,
2051 ub]</literal>, see below.
2052 </para>
2053 <para>
2056 </para></entry>
2057 </row>
2059 <row>
2061 <function>random_gaussian</function> ( <parameter>lb</parameter>, <parameter>ub</parameter>, <parameter>parameter</parameter> )
2063 </para>
2064 <para>
2065 Computes a Gaussian-distributed random integer in <literal>[lb,
2066 ub]</literal>, see below.
2067 </para>
2068 <para>
2071 </para></entry>
2072 </row>
2074 <row>
2076 <function>random_zipfian</function> ( <parameter>lb</parameter>, <parameter>ub</parameter>, <parameter>parameter</parameter> )
2078 </para>
2079 <para>
2080 Computes a Zipfian-distributed random integer in <literal>[lb,
2081 ub]</literal>, see below.
2082 </para>
2083 <para>
2086 </para></entry>
2087 </row>
2089 <row>
2093 </para>
2094 <para>
2095 Square root
2096 </para>
2097 <para>
2100 </para></entry>
2101 </row>
2102 </tbody>
2103 </tgroup>
2104 </table>
2106 <para>
2108 distribution, that is all the values are drawn within the specified
2111 functions require an additional double parameter which determines the precise
2112 shape of the distribution.
2113 </para>
2115 <itemizedlist>
2116 <listitem>
2117 <para>
2119 controls the distribution by truncating a quickly-decreasing
2121 projecting onto integers between the bounds.
2122 To be precise, with
2123 <literallayout>
2125 </literallayout>
2129 </para>
2131 <para>
2136 uniform) the access distribution.
2137 A crude approximation of the distribution is that the most frequent 1%
2141 </para>
2142 </listitem>
2144 <listitem>
2145 <para>
2146 For a Gaussian distribution, the interval is mapped onto a standard
2147 normal distribution (the classical bell-shaped Gaussian curve) truncated
2149 on the right.
2150 Values in the middle of the interval are more likely to be drawn.
2154 <literallayout>
2157 </literallayout>
2162 frequently values close to the middle of the interval are drawn, and the
2174 value is 2.0.
2175 </para>
2176 </listitem>
2177 <listitem>
2178 <para>
2180 distribution.
2183 frequently values closer to the beginning of the interval are drawn.
2184 The distribution is such that, assuming the range starts from 1,
2187 <literal>((<replaceable>k</replaceable>+1)/<replaceable>k</replaceable>)**<replaceable>parameter</replaceable></literal>.
2193 </para>
2194 <para>
2197 Springer 1986. Due to limitations of that algorithm,
2200 </para>
2201 </listitem>
2202 </itemizedlist>
2204 <note>
2205 <para>
2206 When designing a benchmark which selects rows non-uniformly, be aware
2207 that the rows chosen may be correlated with other data such as IDs from
2208 a sequence or the physical row ordering, which may skew performance
2209 measurements.
2210 </para>
2211 <para>
2213 function, or some other additional step with similar effect, to shuffle
2214 the selected rows and remove such correlations.
2215 </para>
2216 </note>
2218 <para>
2222 is used, which is initialized randomly unless set by the command-line
2224 </para>
2226 <para>
2228 seed parameter. It generates a pseudorandom permutation of integers in
2230 value in the permuted values. The permutation chosen is parameterized by
2233 that there are no collisions or holes in the output values. Input values
2234 outside the interval are interpreted modulo the size. The function raises
2236 used to scatter the distribution of non-uniform random functions such as
2238 so that values drawn more often are not trivially correlated. For
2240 simulates a possible real world workload typical for social media and
2241 blogging platforms where a few accounts generate excessive load:
2243 <programlisting>
2244 \set size 1000000
2246 \set k 1 + permute(:r, :size)
2247 </programlisting>
2249 In some cases several distinct distributions are needed which don't correlate
2250 with each other and this is when the optional seed parameter comes in handy:
2252 <programlisting>
2255 </programlisting>
2259 <programlisting>
2260 \set size 1000000
2262 \set k 1 + abs(hash(:r)) % :size
2263 </programlisting>
2266 will not be reachable and others will be more frequent than expected from
2267 the original distribution.
2268 </para>
2270 <para>
2271 As an example, the full definition of the built-in TPC-B-like
2272 transaction is:
2274 <programlisting>
2279 BEGIN;
2280 UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
2281 SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
2282 UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;
2283 UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;
2284 INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
2285 END;
2286 </programlisting>
2288 This script allows each iteration of the transaction to reference
2289 different, randomly-chosen rows. (This example also shows why it's
2290 important for each client session to have its own variables —
2291 otherwise they'd not be independently touching different rows.)
2292 </para>
2294 </refsect2>
2296 <refsect2>
2299 <para>
2303 to a log file. The log file will be named
2310 worker threads, each will have its own log file. The first worker will
2311 use the same name for its log file as in the standard single worker case.
2312 The additional log files for the other workers will be named
2313 <filename><replaceable>prefix</replaceable>.<replaceable>nnn</replaceable>.<replaceable>mmm</replaceable></filename>,
2315 with 1.
2316 </para>
2318 <para>
2319 Each line in a log file describes one transaction.
2320 It contains the following space-separated fields:
2322 <variablelist>
2323 <varlistentry>
2325 <listitem>
2326 <para>
2327 identifies the client session that ran the transaction
2328 </para>
2329 </listitem>
2330 </varlistentry>
2332 <varlistentry>
2334 <listitem>
2335 <para>
2336 counts how many transactions have been run by that session
2337 </para>
2338 </listitem>
2339 </varlistentry>
2341 <varlistentry>
2343 <listitem>
2344 <para>
2345 transaction's elapsed time, in microseconds
2346 </para>
2347 </listitem>
2348 </varlistentry>
2350 <varlistentry>
2352 <listitem>
2353 <para>
2354 identifies the script file that was used for the transaction
2355 (useful when multiple scripts are specified
2357 </para>
2358 </listitem>
2359 </varlistentry>
2361 <varlistentry>
2363 <listitem>
2364 <para>
2365 transaction's completion time, as a Unix-epoch time stamp
2366 </para>
2367 </listitem>
2368 </varlistentry>
2370 <varlistentry>
2372 <listitem>
2373 <para>
2374 fractional-second part of transaction's completion time, in
2375 microseconds
2376 </para>
2377 </listitem>
2378 </varlistentry>
2380 <varlistentry>
2382 <listitem>
2383 <para>
2384 transaction start delay, that is the difference between the
2385 transaction's scheduled start time and the time it actually
2386 started, in microseconds
2388 </para>
2389 </listitem>
2390 </varlistentry>
2392 <varlistentry>
2394 <listitem>
2395 <para>
2396 count of retries after serialization or deadlock errors during the
2397 transaction
2399 </para>
2400 </listitem>
2401 </varlistentry>
2402 </variablelist>
2403 </para>
2405 <para>
2416 </para>
2418 <para>
2419 Here is a snippet of a log file generated in a single-client run:
2420 <screen>
2421 0 199 2241 0 1175850568 995598
2422 0 200 2465 0 1175850568 998079
2423 0 201 2513 0 1175850569 608
2424 0 202 2038 0 1175850569 2663
2425 </screen>
2430 <screen>
2431 0 81 4621 0 1412881037 912698 3005
2432 0 82 6173 0 1412881037 914578 4304
2435 0 83 4722 0 1412881037 916203 3108
2436 0 84 4142 0 1412881037 918023 2333
2437 0 85 2465 0 1412881037 919759 740
2438 </screen>
2440 over the 5 ms limit. The next two transactions were skipped, because they
2441 were already late before they were even started.
2442 </para>
2444 <para>
2445 The following example shows a snippet of a log file with failures and
2446 retries, with the maximum number of tries set to 10 (note the additional
2448 <screen>
2449 3 0 47423 0 1499414498 34501 3
2450 3 1 8333 0 1499414498 42848 0
2451 3 2 8358 0 1499414498 51219 0
2452 4 0 72345 0 1499414498 59433 6
2453 1 3 41718 0 1499414498 67879 4
2454 1 4 8416 0 1499414498 76311 0
2455 3 3 33235 0 1499414498 84469 3
2458 3 4 8307 0 1499414498 92788 0
2459 </screen>
2460 </para>
2462 <para>
2465 <screen>
2466 3 0 47423 0 1499414498 34501 3
2467 3 1 8333 0 1499414498 42848 0
2468 3 2 8358 0 1499414498 51219 0
2469 4 0 72345 0 1499414498 59433 6
2470 1 3 41718 0 1499414498 67879 4
2471 1 4 8416 0 1499414498 76311 0
2472 3 3 33235 0 1499414498 84469 3
2475 3 4 8307 0 1499414498 92788 0
2476 </screen>
2477 </para>
2479 <para>
2480 When running a long test on hardware that can handle a lot of transactions,
2482 can be used to log only a random sample of transactions.
2483 </para>
2484 </refsect2>
2486 <refsect2>
2489 <para>
2491 format is used for the log files. Each log line describes one
2492 aggregation interval. It contains the following space-separated
2493 fields:
2495 <variablelist>
2496 <varlistentry>
2498 <listitem>
2499 <para>
2500 start time of the interval, as a Unix-epoch time stamp
2501 </para>
2502 </listitem>
2503 </varlistentry>
2505 <varlistentry>
2507 <listitem>
2508 <para>
2509 number of transactions within the interval
2510 </para>
2511 </listitem>
2512 </varlistentry>
2514 <varlistentry>
2516 <listitem>
2517 <para>
2518 sum of transaction latencies
2519 </para>
2520 </listitem>
2521 </varlistentry>
2523 <varlistentry>
2525 <listitem>
2526 <para>
2527 sum of squares of transaction latencies
2528 </para>
2529 </listitem>
2530 </varlistentry>
2532 <varlistentry>
2534 <listitem>
2535 <para>
2536 minimum transaction latency
2537 </para>
2538 </listitem>
2539 </varlistentry>
2541 <varlistentry>
2543 <listitem>
2544 <para>
2545 maximum transaction latency
2546 </para>
2547 </listitem>
2548 </varlistentry>
2550 <varlistentry>
2552 <listitem>
2553 <para>
2554 sum of transaction start delays
2556 </para>
2557 </listitem>
2558 </varlistentry>
2560 <varlistentry>
2562 <listitem>
2563 <para>
2564 sum of squares of transaction start delays
2566 </para>
2567 </listitem>
2568 </varlistentry>
2570 <varlistentry>
2572 <listitem>
2573 <para>
2574 minimum transaction start delay
2576 </para>
2577 </listitem>
2578 </varlistentry>
2580 <varlistentry>
2582 <listitem>
2583 <para>
2584 maximum transaction start delay
2586 </para>
2587 </listitem>
2588 </varlistentry>
2590 <varlistentry>
2592 <listitem>
2593 <para>
2594 number of transactions skipped because they would have started too late
2597 </para>
2598 </listitem>
2599 </varlistentry>
2601 <varlistentry>
2603 <listitem>
2604 <para>
2605 number of retried transactions
2607 </para>
2608 </listitem>
2609 </varlistentry>
2611 <varlistentry>
2613 <listitem>
2614 <para>
2615 number of retries after serialization or deadlock errors
2617 </para>
2618 </listitem>
2619 </varlistentry>
2621 <varlistentry>
2623 <listitem>
2624 <para>
2625 number of transactions that got a serialization error and were not
2626 retried afterwards
2628 </para>
2629 </listitem>
2630 </varlistentry>
2632 <varlistentry>
2634 <listitem>
2635 <para>
2636 number of transactions that got a deadlock error and were not
2637 retried afterwards
2639 </para>
2640 </listitem>
2641 </varlistentry>
2642 </variablelist>
2643 </para>
2645 <para>
2646 Here is some example output generated with this option:
2647 <screen>
2648 <userinput>pgbench --aggregate-interval=10 --time=20 --client=10 --log --rate=1000 --latency-limit=10 --failures-detailed --max-tries=10 test</userinput>
2650 1650260552 5178 26171317 177284491527 1136 44462 2647617 7321113867 0 9866 64 7564 28340 4148 0
2651 1650260562 4808 25573984 220121792172 1171 62083 3037380 9666800914 0 9998 598 7392 26621 4527 0
2652 </screen>
2653 </para>
2655 <para>
2656 Notice that while the plain (unaggregated) log format shows which script
2657 was used for each transaction, the aggregated format does not. Therefore if
2658 you need per-script data, you need to aggregate the data on your own.
2659 </para>
2661 </refsect2>
2663 <refsect2>
2666 <para>
2668 collects the following statistics for each statement:
2669 <itemizedlist>
2670 <listitem>
2671 <para>
2674 of all successful runs of the statement.
2675 </para>
2676 </listitem>
2677 <listitem>
2678 <para>
2679 The number of failures in this statement. See
2681 </para>
2682 </listitem>
2683 <listitem>
2684 <para>
2685 The number of retries after a serialization or a deadlock error in this
2687 </para>
2688 </listitem>
2689 </itemizedlist>
2690 </para>
2692 <para>
2694 option is not equal to 1.
2695 </para>
2697 <para>
2698 All values are computed for each statement executed by every client and are
2699 reported after the benchmark has finished.
2700 </para>
2702 <para>
2703 For the default script, the output will look similar to this:
2704 <screen>
2705 starting vacuum...end.
2707 scaling factor: 1
2708 query mode: simple
2709 number of clients: 10
2710 number of threads: 1
2711 maximum number of tries: 1
2712 number of transactions per client: 1000
2716 latency average = 28.488 ms
2717 latency stddev = 21.009 ms
2718 initial connection time = 69.068 ms
2719 tps = 346.224794 (without initial connection time)
2720 statement latencies in milliseconds and failures:
2725 0.319 0 BEGIN;
2726 0.834 0 UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
2727 0.641 0 SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
2728 11.126 0 UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;
2729 12.961 0 UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;
2730 0.634 0 INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
2731 1.957 0 END;
2732 </screen>
2734 Another example of output for the default script using serializable default
2735 transaction isolation level (<command>PGOPTIONS='-c
2736 default_transaction_isolation=serializable' pgbench ...</command>):
2737 <screen>
2738 starting vacuum...end.
2740 scaling factor: 1
2741 query mode: simple
2742 number of clients: 10
2743 number of threads: 1
2744 maximum number of tries: 10
2745 number of transactions per client: 1000
2749 total number of retries: 45339
2751 latency average = 17.016 ms
2752 latency stddev = 13.283 ms
2753 initial connection time = 45.017 ms
2754 tps = 186.792667 (without initial connection time)
2755 statement latencies in milliseconds, failures and retries:
2760 0.385 0 0 BEGIN;
2761 0.773 0 1 UPDATE pgbench_accounts SET abalance = abalance + :delta WHERE aid = :aid;
2762 0.624 0 0 SELECT abalance FROM pgbench_accounts WHERE aid = :aid;
2763 1.098 320 3762 UPDATE pgbench_tellers SET tbalance = tbalance + :delta WHERE tid = :tid;
2764 0.582 3363 41576 UPDATE pgbench_branches SET bbalance = bbalance + :delta WHERE bid = :bid;
2765 0.465 0 0 INSERT INTO pgbench_history (tid, bid, aid, delta, mtime) VALUES (:tid, :bid, :aid, :delta, CURRENT_TIMESTAMP);
2766 1.933 0 0 END;
2767 </screen></para>
2769 <para>
2770 If multiple script files are specified, all statistics are reported
2771 separately for each script file.
2772 </para>
2774 <para>
2775 Note that collecting the additional timing information needed for
2776 per-statement latency computation adds some overhead. This will slow
2777 average execution speed and lower the computed TPS. The amount
2778 of slowdown varies significantly depending on platform and hardware.
2779 Comparing average TPS values with and without latency reporting enabled
2780 is a good way to measure if the timing overhead is significant.
2781 </para>
2782 </refsect2>
2787 <para>
2789 of errors:
2790 <itemizedlist>
2791 <listitem>
2792 <para>
2793 Errors of the main program. They are the most serious and always result
2795 corresponding error message. They include:
2796 <itemizedlist>
2797 <listitem>
2798 <para>
2800 (e.g. an invalid option value);
2801 </para>
2802 </listitem>
2803 <listitem>
2804 <para>
2805 errors in the initialization mode (e.g. the query to create
2806 tables for built-in scripts fails);
2807 </para>
2808 </listitem>
2809 <listitem>
2810 <para>
2811 errors before starting threads (e.g. could not connect to the
2812 database server, syntax error in the meta command, thread
2813 creation failure);
2814 </para>
2815 </listitem>
2816 <listitem>
2817 <para>
2819 supposed to never occur...).
2820 </para>
2821 </listitem>
2822 </itemizedlist></para>
2823 </listitem>
2824 <listitem>
2825 <para>
2826 Errors when the thread manages its clients (e.g. the client could not
2827 start a connection to the database server / the socket for connecting
2828 the client to the database server has become invalid). In such cases
2829 all clients of this thread stop while other threads continue to work.
2831 threads stop immediately in this case.
2832 </para>
2833 </listitem>
2834 <listitem>
2835 <para>
2836 Direct client errors. They lead to immediate exit from
2839 error (which are supposed to never occur...) or when
2841 case they only lead to the abortion of the failed client while other
2842 clients continue their run (but some client errors are handled without
2843 an abortion of the client and reported separately, see below). Later in
2844 this section it is assumed that the discussed errors are only the
2845 direct client errors and they are not internal
2847 </para>
2848 </listitem>
2849 </itemizedlist>
2850 </para>
2852 <para>
2853 A client's run is aborted in case of a serious error; for example, the
2854 connection with the database server was lost or the end of script was reached
2855 without completing the last transaction. In addition, if execution of an SQL
2856 or meta command fails for reasons other than serialization or deadlock errors,
2857 the client is aborted. Otherwise, if an SQL command fails with serialization or
2858 deadlock errors, the client is not aborted. In such cases, the current
2859 transaction is rolled back, which also includes setting the client variables
2860 as they were before the run of this transaction (it is assumed that one
2861 transaction script contains only one transaction; see
2863 Transactions with serialization or deadlock errors are repeated after
2864 rollbacks until they complete successfully or reach the maximum
2868 the last trial run fails, this transaction will be reported as failed but
2869 the client is not aborted and continues to work.
2870 </para>
2872 <note>
2873 <para>
2875 never be retried after a serialization or deadlock error because its default
2879 benchmark duration under an unlimited number of tries.
2880 </para>
2881 <para>
2882 Be careful when repeating scripts that contain multiple transactions: the
2883 script is always retried completely, so successful transactions can be
2884 performed several times.
2885 </para>
2886 <para>
2887 Be careful when repeating transactions with shell commands. Unlike the
2888 results of SQL commands, the results of shell commands are not rolled back,
2890 </para>
2891 </note>
2893 <para>
2894 The latency of a successful transaction includes the entire time of
2895 transaction execution with rollbacks and retries. The latency is measured
2896 only for successful transactions and commands but not for failed transactions
2897 or commands.
2898 </para>
2900 <para>
2901 The main report contains the number of failed transactions. If the
2903 contains statistics related to retries: the total number of retried
2904 transactions and total number of retries. The per-script report inherits all
2905 these fields from the main report. The per-statement report displays retry
2907 </para>
2909 <para>
2910 If you want to group failures by basic types in per-transaction and
2911 aggregation logs, as well as in the main and per-script reports, use the
2913 all errors and failures (errors without retrying) by type including which
2914 limit for retries was exceeded and how much it was exceeded by for the
2916 option.
2917 </para>
2918 </refsect2>
2920 <refsect2>
2923 <para>
2926 specifies database configuration options that are passed to PostgreSQL via
2928 For example, a hypothetical default Table Access Method for the tables that
2930 <programlisting>
2931 PGOPTIONS='-c default_table_access_method=wuzza'
2932 </programlisting>
2933 </para>
2934 </refsect2>
2936 <refsect2>
2939 <para>
2941 meaningless numbers. Here are some guidelines to help you get useful
2942 results.
2943 </para>
2945 <para>
2948 to make the run last at least a few minutes, so as to average out noise.
2949 In some cases you could need hours to get numbers that are reproducible.
2950 It's a good idea to try the test run a few times, to find out if your
2951 numbers are reproducible or not.
2952 </para>
2954 <para>
2955 For the default TPC-B-like test scenario, the initialization scale factor
2961 will undoubtedly result in lots of transactions blocked waiting for
2962 other transactions.
2963 </para>
2965 <para>
2966 The default test scenario is also quite sensitive to how long it's been
2967 since the tables were initialized: accumulation of dead rows and dead space
2968 in the tables changes the results. To understand the results you must keep
2969 track of the total number of updates and when vacuuming happens. If
2970 autovacuum is enabled it can result in unpredictable changes in measured
2971 performance.
2972 </para>
2974 <para>
2976 the bottleneck when trying to test a large number of client sessions.
2978 machine from the database server, although low network latency will be
2980 instances concurrently, on several client machines, against the same
2981 database server.
2982 </para>
2983 </refsect2>
2984 <refsect2>
2987 <para>
2988 If untrusted users have access to a database that has not adopted a
2992 does not manipulate the search path.
2993 </para>
2994 </refsect2>
2995 </refsect1>
2996 </refentry>