1PGBOUNCER.INI(5) Databases PGBOUNCER.INI(5)
2
6 pgbouncer.ini - configuration file for pgbouncer
7
9 The configuration file is in “ini” format. Section names are between
10 “[" and "]”. Lines starting with “;” or “#” are taken as comments and
11 ignored. The characters “;” and “#” are not recognized as special when
12 they appear later in the line.
13
15 logfile
16 Specifies the log file. For daemonization (-d), either this or syslog
17 need to be set.
18 The log file is kept open, so after rotation, kill -HUP or on console
20 RELOAD; should be done. On Windows, the service must be stopped and
21 started.
22 Note that setting logfile does not by itself turn off logging to
24 stderr. Use the command-line option -q or -d for that.
25 Default: not set
27 pidfile
29 Specifies the PID file. Without pidfile set, daemonization (-d) is not
30 allowed.
31 Default: not set
33 listen_addr
35 Specifies a list (comma-separated) of addresses where to listen for TCP
36 connections. You may also use * meaning “listen on all addresses”.
37 When not set, only Unix socket connections are accepted.
38 Addresses can be specified numerically (IPv4/IPv6) or by name.
40 Default: not set
42 listen_port
44 Which port to listen on. Applies to both TCP and Unix sockets.
45 Default: 6432
47 unix_socket_dir
49 Specifies the location for Unix sockets. Applies to both the listening
50 socket and to server connections. If set to an empty string, Unix
51 sockets are disabled. A value that starts with @ specifies that a Unix
52 socket in the abstract namespace should be created (currently supported
53 on Linux and Windows).
54 For online reboot (-R) to work, a Unix socket needs to be configured,
56 and it needs to be in the file-system namespace.
57 Default: /tmp (empty on Windows)
59 unix_socket_mode
61 File system mode for Unix socket. Ignored for sockets in the abstract
62 namespace. Not supported on Windows.
63 Default: 0777
65 unix_socket_group
67 Group name to use for Unix socket. Ignored for sockets in the abstract
68 namespace. Not supported on Windows.
69 Default: not set
71 user
73 If set, specifies the Unix user to change to after startup. Works only
74 if PgBouncer is started as root or if it’s already running as the given
75 user. Not supported on Windows.
76 Default: not set
78 pool_mode
80 Specifies when a server connection can be reused by other clients.
81 session
83 Server is released back to pool after client disconnects. De‐
84 fault.
85 transaction
87 Server is released back to pool after transaction finishes.
88 statement
90 Server is released back to pool after query finishes. Transac‐
91 tions spanning multiple statements are disallowed in this mode.
92 max_client_conn
94 Maximum number of client connections allowed.
95 When this setting is increased, then the file descriptor limits in the
97 operating system might also have to be increased. Note that the number
98 of file descriptors potentially used is more than max_client_conn. If
99 each user connects under its own user name to the server, the theoreti‐
100 cal maximum used is:
101 max_client_conn + (max pool_size * total databases * total users)
103 If a database user is specified in the connection string (all users
105 connect under the same user name), the theoretical maximum is:
106 max_client_conn + (max pool_size * total databases)
108 The theoretical maximum should never be reached, unless somebody delib‐
110 erately crafts a special load for it. Still, it means you should set
111 the number of file descriptors to a safely high number.
112 Search for ulimit in your favorite shell man page. Note: ulimit does
114 not apply in a Windows environment.
115 Default: 100
117 default_pool_size
119 How many server connections to allow per user/database pair. Can be
120 overridden in the per-database configuration.
121 Default: 20
123 min_pool_size
125 Add more server connections to pool if below this number. Improves be‐
126 havior when the normal load suddenly comes back after a period of total
127 inactivity. The value is effectively capped at the pool size.
128 Only enforced for pools where at least one of the following is true:
130 • the entry in the [database] section for the pool has a value set for
132 the user key (aka forced user)
133 • there is at least one client connected to the pool
135 Default: 0 (disabled)
137 reserve_pool_size
139 How many additional connections to allow to a pool (see re‐
140 serve_pool_timeout). 0 disables.
141 Default: 0 (disabled)
143 reserve_pool_timeout
145 If a client has not been serviced in this time, use additional connec‐
146 tions from the reserve pool. 0 disables. [seconds]
147 Default: 5.0
149 max_db_connections
151 Do not allow more than this many server connections per database (re‐
152 gardless of user). This considers the PgBouncer database that the
153 client has connected to, not the PostgreSQL database of the outgoing
154 connection.
155 This can also be set per database in the [databases] section.
157 Note that when you hit the limit, closing a client connection to one
159 pool will not immediately allow a server connection to be established
160 for another pool, because the server connection for the first pool is
161 still open. Once the server connection closes (due to idle timeout), a
162 new server connection will immediately be opened for the waiting pool.
163 Default: 0 (unlimited)
165 max_user_connections
167 Do not allow more than this many server connections per user (regard‐
168 less of database). This considers the PgBouncer user that is associat‐
169 ed with a pool, which is either the user specified for the server con‐
170 nection or in absence of that the user the client has connected as.
171 This can also be set per user in the [users] section.
173 Note that when you hit the limit, closing a client connection to one
175 pool will not immediately allow a server connection to be established
176 for another pool, because the server connection for the first pool is
177 still open. Once the server connection closes (due to idle timeout), a
178 new server connection will immediately be opened for the waiting pool.
179 Default: 0 (unlimited)
181 server_round_robin
183 By default, PgBouncer reuses server connections in LIFO (last-in,
184 first-out) manner, so that few connections get the most load. This
185 gives best performance if you have a single server serving a database.
186 But if there is a round-robin system behind a database address (TCP,
187 DNS, or host list), then it is better if PgBouncer also uses connec‐
188 tions in that manner, thus achieving uniform load.
189 Default: 0
191 track_extra_parameters
193 By default, PgBouncer tracks client_encoding, datestyle, timezone,
194 standard_conforming_strings and application_name parameters per client.
195 To allow other parameters to be tracked, they can be specified here, so
196 that PgBouncer knows that they should be maintained in the client vari‐
197 able cache and restored in the server whenever the client becomes ac‐
198 tive.
199 If you need to specify multiple values, use a comma-separated list
201 (e.g. default_transaction_readonly, IntervalStyle)
202 Note: Most parameters cannot be tracked this way. The only parameters
204 that can be tracked are ones that Postgres reports to the client.
205 Postgres has an official list of parameters that it reports to the
206 client (https://www.postgresql.org/docs/15/protocol-flow.html#PROTOCOL-
207 ASYNC). Postgres extensions can change this list though, they can add
208 parameters themselves that they also report, and they can start report‐
209 ing already existing parameters that Postgres does not report. Notably
210 Citus 12.0+ causes Postgres to also report search_path.
211 The Postgres protocol allows specifying parameters settings, both di‐
213 rectly as a parameter in the startup packet, or inside the options
214 startup packet (https://www.postgresql.org/docs/current/libpq-con‐
215 nect.html#LIBPQ-CONNECT-OPTIONS). Parameters specified using both of
216 these methods are supported by track_extra_parameters. However, it’s
217 not possible to include options itself in track_extra_parameters, only
218 the parameters contained in options.
219 Default: IntervalStyle
221 ignore_startup_parameters
223 By default, PgBouncer allows only parameters it can keep track of in
224 startup packets: client_encoding, datestyle, timezone and standard_con‐
225 forming_strings. All others parameters will raise an error. To allow
226 others parameters, they can be specified here, so that PgBouncer knows
227 that they are handled by the admin and it can ignore them.
228 If you need to specify multiple values, use a comma-separated list
230 (e.g. options,extra_float_digits)
231 The Postgres protocol allows specifying parameters settings, both di‐
233 rectly as a parameter in the startup packet, or inside the options
234 startup packet (https://www.postgresql.org/docs/current/libpq-con‐
235 nect.html#LIBPQ-CONNECT-OPTIONS). Parameters specified using both of
236 these methods are supported by ignore_startup_parameters. It’s even
237 possible to include options itself in track_extra_parameters, which re‐
238 sults in any unknown parameters contained inside options to be ignored.
239 Default: empty
241 peer_id
243 The peer id used to identify this PgBouncer process in a group of Pg‐
244 Bouncer processes that are peered together. The peer_id value should
245 be unique within a group of peered PgBouncer processes. When set to 0
246 pgbouncer peering is disabled. See the docs for the [peers] section
247 for more information. The maximum value that can be used for the
248 peer_id is 16383.
249 Default: 0
251 disable_pqexec
253 Disable the Simple Query protocol (PQexec). Unlike the Extended Query
254 protocol, Simple Query allows multiple queries in one packet, which al‐
255 lows some classes of SQL-injection attacks. Disabling it can improve
256 security. Obviously, this means only clients that exclusively use the
257 Extended Query protocol will stay working.
258 Default: 0
260 application_name_add_host
262 Add the client host address and port to the application name setting
263 set on connection start. This helps in identifying the source of bad
264 queries etc. This logic applies only at the start of a connection. If
265 application_name is later changed with SET, PgBouncer does not change
266 it again.
267 Default: 0
269 conffile
271 Show location of current config file. Changing it will make PgBouncer
272 use another config file for next RELOAD / SIGHUP.
273 Default: file from command line
275 service_name
277 Used on win32 service registration.
278 Default: pgbouncer
280 job_name
282 Alias for service_name.
283 stats_period
285 Sets how often the averages shown in various SHOW commands are updated
286 and how often aggregated statistics are written to the log (but see
287 log_stats). [seconds]
288 Default: 60
290 max_prepared_statements
292 When this is set to a non-zero value PgBouncer tracks protocol-level
293 named prepared statements related commands sent by the client in trans‐
294 action and statement pooling mode. PgBouncer makes sure that any
295 statement prepared by a client is available on the backing server con‐
296 nection. Even when the statement was originally prepared on another
297 server connection.
298 PgBouncer internally examines all the queries that are sent as a pre‐
300 pared statement by clients and gives each unique query string an inter‐
301 nal name with the format PGBOUNCER_{unique_id}. Prepared statements
302 are only prepared using this name on the corresponding PostgreSQL serv‐
303 er. PgBouncer keeps track of the name that the client gave to each
304 prepared statement. It rewrites each command that uses a prepared
305 statement to use the matching internal name (e.g. PGBOUNCER_123) be‐
306 fore forwarding that command to the server. More importantly, if the
307 prepared statement that the client wants to use is not prepared on the
308 server yet, it automatically prepares that statement before forwarding
309 the command that the client sent.
310 Note: This tracking and rewriting of prepared statement commands does
312 not work for SQL-level prepared statement commands such as PREPARE, EX‐
313 ECUTE, DEALLOCATE, DEALLOCATE ALL and DISCARD ALL. Running DEALLOCATE
314 ALL and DISCARD ALL is especially problematic, since those commands ap‐
315 pear to run successfully, but they mess up with the state of the server
316 connection significantly without PgBouncer noticing. Which in turn
317 will very likely break the execution of any further prepared statements
318 on that server connection.
319 The actual value of this setting controls the number of prepared state‐
321 ments kept active on a single server connection. When the setting is
322 set to 0 prepared statement support for transaction and statement pool‐
323 ing is disabled. To get the best performance you should try to make
324 sure that this setting is larger than the amount of commonly used pre‐
325 pared statements in your application. Keep in mind that the higher
326 this value, the larger the memory footprint of each PgBouncer connec‐
327 tion will be on your PostgreSQL server, because it will keep more
328 queries prepared on those connections. It also increases the memory
329 footprint of PgBouncer itself, because it now needs to keep track of
330 query strings.
331 The impact on PgBouncer memory usage is not that big though: - Each
333 unique query is stored once in a global query cache. - Each client
334 connection keeps a buffer that it uses to rewrite packets. This is at
335 most 4 times the size of pkt_buf. This limit is often not reached
336 though, it only happens when the queries in your prepared statements
337 are between 2 and 4 times the size of pkt_buf.
338 So if you consider the following as an example scenario: - There are
340 1000 active clients - The clients prepare 200 unique queries - The av‐
341 erage size of a query is 5kB - pkt_buf parameter is set to the default
342 of 4096 (4kB)
343 Then PgBouncer needs at most the following amount of memory to handle
345 these prepared statements:
346 200 x 5kB + 1000 x 4 x 4kB = ~17MB of memory.
348 Tracking prepared statements does not only come with a memory cost, but
350 also with increased CPU usage, because PgBouncer needs to inspect and
351 rewrite the queries. Multiple PgBouncer instances can listen on the
352 same port to use more than one core for processing, see the documenta‐
353 tion for the so_reuseport option for details.
354 But of course there are also performance benefits to prepared state‐
356 ments. Just as when connecting to PostgreSQL directly, by preparing a
357 query that is executed many times, it reduces the total amount of pars‐
358 ing and planning that needs to be done. The way that PgBouncer tracks
359 prepared statements is especially beneficial to performance when multi‐
360 ple clients prepare the same queries. Because client connections auto‐
361 matically reuse a prepared statement on a server connection even if it
362 was prepared by another client. As an example if you have a pool_size
363 of 20 and you have 100 clients that all prepare the exact same query,
364 then the query is prepared (and thus parsed) only 20 times on the Post‐
365 greSQL server.
366 The reuse of prepared statements has one downside. If the return or
368 argument types of a prepared statement changes across executions then
369 PostgreSQL currently throws an error such as:
370 ERROR: cached plan must not change result type
372 You can avoid such errors by not having multiple clients that use the
374 exact same query string in a prepared statement, but expecting differ‐
375 ent argument or result types. One of the most common ways of running
376 into this issue is during a DDL migration where you add a new column or
377 change a column type on an existing table. In those cases you can run
378 RECONNECT on the PgBouncer admin console after doing the migration to
379 force a re-prepare of the query and make the error goes away.
380 Default: 0
382
384 PgBouncer handles its own client authentication and has its own data‐
385 base of users. These settings control this.
386 auth_type
388 How to authenticate users.
389 cert Client must connect over TLS connection with a valid client cer‐
391 tificate. The user name is then taken from the CommonName field
392 from the certificate.
393 md5 Use MD5-based password check. This is the default authentica‐
395 tion method. auth_file may contain both MD5-encrypted and
396 plain-text passwords. If md5 is configured and a user has a
397 SCRAM secret, then SCRAM authentication is used automatically
398 instead.
399 scram-sha-256
401 Use password check with SCRAM-SHA-256. auth_file has to contain
402 SCRAM secrets or plain-text passwords.
403 plain The clear-text password is sent over the wire. Deprecated.
405 trust No authentication is done. The user name must still exist in
407 auth_file.
408 any Like the trust method, but the user name given is ignored. Re‐
410 quires that all databases are configured to log in as a specific
411 user. Additionally, the console database allows any user to log
412 in as admin.
413 hba The actual authentication type is loaded from auth_hba_file.
415 This allows different authentication methods for different ac‐
416 cess paths, for example: connections over Unix socket use the
417 peer auth method, connections over TCP must use TLS.
418 pam PAM is used to authenticate users, auth_file is ignored. This
420 method is not compatible with databases using the auth_user op‐
421 tion. The service name reported to PAM is “pgbouncer”. pam is
422 not supported in the HBA configuration file.
423 auth_hba_file
425 HBA configuration file to use when auth_type is hba.
426 Default: not set
428 auth_file
430 The name of the file to load user names and passwords from. See sec‐
431 tion Authentication file format below about details.
432 Most authentication types (see above) require that either auth_file or
434 auth_user be set; otherwise there would be no users defined.
435 Default: not set
437 auth_user
439 If auth_user is set, then any user not specified in auth_file will be
440 queried through the auth_query query from pg_shadow in the database,
441 using auth_user. The password of auth_user will be taken from
442 auth_file. (If the auth_user does not require a password then it does
443 not need to be defined in auth_file.)
444 Direct access to pg_shadow requires admin rights. It’s preferable to
446 use a non-superuser that calls a SECURITY DEFINER function instead.
447 Default: not set
449 auth_query
451 Query to load user’s password from database.
452 Direct access to pg_shadow requires admin rights. It’s preferable to
454 use a non-superuser that calls a SECURITY DEFINER function instead.
455 Note that the query is run inside the target database. So if a func‐
457 tion is used, it needs to be installed into each database.
458 Default: SELECT usename, passwd FROM pg_shadow WHERE usename=$1
460 auth_dbname
462 Database name in the [database] section to be used for authentication
463 purposes. This option can be either global or overridden in the con‐
464 nection string if this parameter is specified.
465
467 syslog
468 Toggles syslog on/off. On Windows, the event log is used instead.
469 Default: 0
471 syslog_ident
473 Under what name to send logs to syslog.
474 Default: pgbouncer (program name)
476 syslog_facility
478 Under what facility to send logs to syslog. Possibilities: auth, auth‐
479 priv, daemon, user, local0-7.
480 Default: daemon
482 log_connections
484 Log successful logins.
485 Default: 1
487 log_disconnections
489 Log disconnections with reasons.
490 Default: 1
492 log_pooler_errors
494 Log error messages the pooler sends to clients.
495 Default: 1
497 log_stats
499 Write aggregated statistics into the log, every stats_period. This can
500 be disabled if external monitoring tools are used to grab the same data
501 from SHOW commands.
502 Default: 1
504 verbose
506 Increase verbosity. Mirrors the “-v” switch on the command line. For
507 example, using “-v -v” on the command line is the same as verbose=2.
508 Default: 0
510
512 admin_users
513 Comma-separated list of database users that are allowed to connect and
514 run all commands on the console. Ignored when auth_type is any, in
515 which case any user name is allowed in as admin.
516 Default: empty
518 stats_users
520 Comma-separated list of database users that are allowed to connect and
521 run read-only queries on the console. That means all SHOW commands ex‐
522 cept SHOW FDS.
523 Default: empty
525
527 server_reset_query
528 Query sent to server on connection release, before making it available
529 to other clients. At that moment no transaction is in progress, so the
530 value should not include ABORT or ROLLBACK.
531 The query is supposed to clean any changes made to the database session
533 so that the next client gets the connection in a well-defined state.
534 The default is DISCARD ALL, which cleans everything, but that leaves
535 the next client no pre-cached state. It can be made lighter, e.g. DE‐
536 ALLOCATE ALL to just drop prepared statements, if the application does
537 not break when some state is kept around.
538 When transaction pooling is used, the server_reset_query is not used,
540 because in that mode, clients must not use any session-based features,
541 since each transaction ends up in a different connection and thus gets
542 a different session state.
543 Default: DISCARD ALL
545 server_reset_query_always
547 Whether server_reset_query should be run in all pooling modes. When
548 this setting is off (default), the server_reset_query will be run only
549 in pools that are in sessions-pooling mode. Connections in transac‐
550 tion-pooling mode should not have any need for a reset query.
551 This setting is for working around broken setups that run applications
553 that use session features over a transaction-pooled PgBouncer. It
554 changes non-deterministic breakage to deterministic breakage: Clients
555 always lose their state after each transaction.
556 Default: 0
558 server_check_delay
560 How long to keep released connections available for immediate re-use,
561 without running server_check_query on it. If 0 then the check is al‐
562 ways run.
563 Default: 30.0
565 server_check_query
567 Simple do-nothing query to check if the server connection is alive.
568 If an empty string, then sanity checking is disabled.
570 Default: select 1
572 server_fast_close
574 Disconnect a server in session pooling mode immediately or after the
575 end of the current transaction if it is in “close_needed” mode (set by
576 RECONNECT, RELOAD that changes connection settings, or DNS change),
577 rather than waiting for the session end. In statement or transaction
578 pooling mode, this has no effect since that is the default behavior
579 there.
580 If because of this setting a server connection is closed before the end
582 of the client session, the client connection is also closed. This en‐
583 sures that the client notices that the session has been interrupted.
584 This setting makes connection configuration changes take effect sooner
586 if session pooling and long-running sessions are used. The downside is
587 that client sessions are liable to be interrupted by a configuration
588 change, so client applications will need logic to reconnect and
589 reestablish session state. But note that no transactions will be lost,
590 because running transactions are not interrupted, only idle sessions.
591 Default: 0
593 server_lifetime
595 The pooler will close an unused (not currently linked to any client
596 connection) server connection that has been connected longer than this.
597 Setting it to 0 means the connection is to be used only once, then
598 closed. [seconds]
599 Default: 3600.0
601 server_idle_timeout
603 If a server connection has been idle more than this many seconds it
604 will be closed. If 0 then this timeout is disabled. [seconds]
605 Default: 600.0
607 server_connect_timeout
609 If connection and login don’t finish in this amount of time, the con‐
610 nection will be closed. [seconds]
611 Default: 15.0
613 server_login_retry
615 If login to the server failed, because of failure to connect or from
616 authentication, the pooler waits this much before retrying to connect.
617 During the waiting interval, new clients trying to connect to the fail‐
618 ing server will get an error immediately without another connection at‐
619 tempt. [seconds]
620 The purpose of this behavior is that clients don’t unnecessarily queue
622 up waiting for a server connection to become available if the server is
623 not working. However, it also means that if a server is momentarily
624 failing, for example during a restart or if the configuration was erro‐
625 neous, then it will take at least this long until the pooler will con‐
626 sider connecting to it again. Planned events such as restarts should
627 normally be managed using the PAUSE command to avoid this.
628 Default: 15.0
630 client_login_timeout
632 If a client connects but does not manage to log in in this amount of
633 time, it will be disconnected. Mainly needed to avoid dead connections
634 stalling SUSPEND and thus online restart. [seconds]
635 Default: 60.0
637 autodb_idle_timeout
639 If the automatically created (via "*") database pools have been unused
640 this many seconds, they are freed. The negative aspect of that is that
641 their statistics are also forgotten. [seconds]
642 Default: 3600.0
644 dns_max_ttl
646 How long DNS lookups can be cached. The actual DNS TTL is ignored.
647 [seconds]
648 Default: 15.0
650 dns_nxdomain_ttl
652 How long DNS errors and NXDOMAIN DNS lookups can be cached. [seconds]
653 Default: 15.0
655 dns_zone_check_period
657 Period to check if a zone serial has changed.
658 PgBouncer can collect DNS zones from host names (everything after first
660 dot) and then periodically check if the zone serial changes. If it no‐
661 tices changes, all host names under that zone are looked up again. If
662 any host IP changes, its connections are invalidated.
663 Works only with c-ares backend (configure option --with-cares).
665 Default: 0.0 (disabled)
667 resolv_conf
669 The location of a custom resolv.conf file. This is to allow specifying
670 custom DNS servers and perhaps other name resolution options, indepen‐
671 dent of the global operating system configuration.
672 Requires evdns (>= 2.0.3) or c-ares (>= 1.15.0) backend.
674 The parsing of the file is done by the DNS backend library, not Pg‐
676 Bouncer, so see the library’s documentation for details on allowed syn‐
677 tax and directives.
678 Default: empty (use operating system defaults)
680
682 client_tls_sslmode
683 TLS mode to use for connections from clients. TLS connections are dis‐
684 abled by default. When enabled, client_tls_key_file and
685 client_tls_cert_file must be also configured to set up the key and cer‐
686 tificate PgBouncer uses to accept client connections.
687 disable
689 Plain TCP. If client requests TLS, it’s ignored. Default.
690 allow If client requests TLS, it is used. If not, plain TCP is used.
692 If the client presents a client certificate, it is not validat‐
693 ed.
694 prefer Same as allow.
696 require
698 Client must use TLS. If not, the client connection is rejected.
699 If the client presents a client certificate, it is not validat‐
700 ed.
701 verify-ca
703 Client must use TLS with valid client certificate.
704 verify-full
706 Same as verify-ca.
707 client_tls_key_file
709 Private key for PgBouncer to accept client connections.
710 Default: not set
712 client_tls_cert_file
714 Certificate for private key. Clients can validate it.
715 Default: not set
717 client_tls_ca_file
719 Root certificate file to validate client certificates.
720 Default: not set
722 client_tls_protocols
724 Which TLS protocol versions are allowed. Allowed values: tlsv1.0,
725 tlsv1.1, tlsv1.2, tlsv1.3. Shortcuts: all
726 (tlsv1.0,tlsv1.1,tlsv1.2,tlsv1.3), secure (tlsv1.2,tlsv1.3), legacy
727 (all).
728 Default: secure
730 client_tls_ciphers
732 Allowed TLS ciphers, in OpenSSL syntax. Shortcuts:
733 • default/secure/fast/normal (these all use system wide OpenSSL de‐
735 faults)
736 • all (enables all ciphers, not recommended)
738 Only connections using TLS version 1.2 and lower are affected. There
740 is currently no setting that controls the cipher choices used by TLS
741 version 1.3 connections.
742 Default: default
744 client_tls_ecdhcurve
746 Elliptic Curve name to use for ECDH key exchanges.
747 Allowed values: none (DH is disabled), auto (256-bit ECDH), curve name
749 Default: auto
751 client_tls_dheparams
753 DHE key exchange type.
754 Allowed values: none (DH is disabled), auto (2048-bit DH), legacy
756 (1024-bit DH)
757 Default: auto
759 server_tls_sslmode
761 TLS mode to use for connections to PostgreSQL servers. The default
762 mode is prefer.
763 disable
765 Plain TCP. TLS is not even requested from the server.
766 allow FIXME: if server rejects plain, try TLS?
768 prefer TLS connection is always requested first from PostgreSQL. If
770 refused, the connection will be established over plain TCP.
771 Server certificate is not validated. Default
772 require
774 Connection must go over TLS. If server rejects it, plain TCP is
775 not attempted. Server certificate is not validated.
776 verify-ca
778 Connection must go over TLS and server certificate must be valid
779 according to server_tls_ca_file. Server host name is not
780 checked against certificate.
781 verify-full
783 Connection must go over TLS and server certificate must be valid
784 according to server_tls_ca_file. Server host name must match
785 certificate information.
786 server_tls_ca_file
788 Root certificate file to validate PostgreSQL server certificates.
789 Default: not set
791 server_tls_key_file
793 Private key for PgBouncer to authenticate against PostgreSQL server.
794 Default: not set
796 server_tls_cert_file
798 Certificate for private key. PostgreSQL server can validate it.
799 Default: not set
801 server_tls_protocols
803 Which TLS protocol versions are allowed. Allowed values: tlsv1.0,
804 tlsv1.1, tlsv1.2, tlsv1.3. Shortcuts: all
805 (tlsv1.0,tlsv1.1,tlsv1.2,tlsv1.3), secure (tlsv1.2,tlsv1.3), legacy
806 (all).
807 Default: secure
809 server_tls_ciphers
811 Allowed TLS ciphers, in OpenSSL syntax. Shortcuts:
812 • default/secure/fast/normal (these all use system wide OpenSSL de‐
814 faults)
815 • all (enables all ciphers, not recommended)
817 Only connections using TLS version 1.2 and lower are affected. There
819 is currently no setting that controls the cipher choices used by TLS
820 version 1.3 connections.
821 Default: default
823
825 Setting the following timeouts can cause unexpected errors.
826 query_timeout
828 Queries running longer than that are canceled. This should be used on‐
829 ly with a slightly smaller server-side statement_timeout, to apply only
830 for network problems. [seconds]
831 Default: 0.0 (disabled)
833 query_wait_timeout
835 Maximum time queries are allowed to spend waiting for execution. If
836 the query is not assigned to a server during that time, the client is
837 disconnected. 0 disables. If this is disabled, clients will be queued
838 indefinitely. [seconds]
839 This setting is used to prevent unresponsive servers from grabbing up
841 connections. It also helps when the server is down or rejects connec‐
842 tions for any reason.
843 Default: 120.0
845 cancel_wait_timeout
847 Maximum time cancellation requests are allowed to spend waiting for ex‐
848 ecution. If the cancel request is not assigned to a server during that
849 time, the client is disconnected. 0 disables. If this is disabled,
850 cancel requests will be queued indefinitely. [seconds]
851 This setting is used to prevent a client locking up when a cancel can‐
853 not be forwarded due to the server being down.
854 Default: 10.0
856 client_idle_timeout
858 Client connections idling longer than this many seconds are closed.
859 This should be larger than the client-side connection lifetime set‐
860 tings, and only used for network problems. [seconds]
861 Default: 0.0 (disabled)
863 idle_transaction_timeout
865 If a client has been in “idle in transaction” state longer, it will be
866 disconnected. [seconds]
867 Default: 0.0 (disabled)
869 suspend_timeout
871 How long to wait for buffer flush during SUSPEND or reboot (-R). A
872 connection is dropped if the flush does not succeed. [seconds]
873 Default: 10
875
877 pkt_buf
878 Internal buffer size for packets. Affects size of TCP packets sent and
879 general memory usage. Actual libpq packets can be larger than this, so
880 no need to set it large.
881 Default: 4096
883 max_packet_size
885 Maximum size for PostgreSQL packets that PgBouncer allows through. One
886 packet is either one query or one result set row. The full result set
887 can be larger.
888 Default: 2147483647
890 listen_backlog
892 Backlog argument for listen(2). Determines how many new unanswered
893 connection attempts are kept in the queue. When the queue is full,
894 further new connections are dropped.
895 Default: 128
897 sbuf_loopcnt
899 How many times to process data on one connection, before proceeding.
900 Without this limit, one connection with a big result set can stall Pg‐
901 Bouncer for a long time. One loop processes one pkt_buf amount of da‐
902 ta. 0 means no limit.
903 Default: 5
905 so_reuseport
907 Specifies whether to set the socket option SO_REUSEPORT on TCP listen‐
908 ing sockets. On some operating systems, this allows running multiple
909 PgBouncer instances on the same host listening on the same port and
910 having the kernel distribute the connections automatically. This op‐
911 tion is a way to get PgBouncer to use more CPU cores. (PgBouncer is
912 single-threaded and uses one CPU core per instance.)
913 The behavior in detail depends on the operating system kernel. As of
915 this writing, this setting has the desired effect on (sufficiently re‐
916 cent versions of) Linux, DragonFlyBSD, and FreeBSD. (On FreeBSD, it
917 applies the socket option SO_REUSEPORT_LB instead.) Some other operat‐
918 ing systems support the socket option but it won’t have the desired ef‐
919 fect: It will allow multiple processes to bind to the same port but on‐
920 ly one of them will get the connections. See your operating system’s
921 setsockopt() documentation for details.
922 On systems that don’t support the socket option at all, turning this
924 setting on will result in an error.
925 Each PgBouncer instance on the same host needs different settings for
927 at least unix_socket_dir and pidfile, as well as logfile if that is
928 used. Also note that if you make use of this option, you can no longer
929 connect to a specific PgBouncer instance via TCP/IP, which might have
930 implications for monitoring and metrics collection.
931 To make sure query cancellations keep working, you should set up Pg‐
933 Bouncer peering between the different PgBouncer processes. For details
934 look at docs for the peer_id configuration option and the peers config‐
935 uration section. There’s also an example that uses peering and
936 so_reuseport in the example section of these docs.
937 Default: 0
939 tcp_defer_accept
941 Sets the TCP_DEFER_ACCEPT socket option; see man 7 tcp for details.
942 (This is a Boolean option: 1 means enabled. The actual value set if
943 enabled is currently hardcoded to 45 seconds.)
944 This is currently only supported on Linux.
946 Default: 1 on Linux, otherwise 0
948 tcp_socket_buffer
950 Default: not set
951 tcp_keepalive
953 Turns on basic keepalive with OS defaults.
954 On Linux, the system defaults are tcp_keepidle=7200, tcp_keepintvl=75,
956 tcp_keepcnt=9. They are probably similar on other operating systems.
957 Default: 1
959 tcp_keepcnt
961 Default: not set
962 tcp_keepidle
964 Default: not set
965 tcp_keepintvl
967 Default: not set
968 tcp_user_timeout
970 Sets the TCP_USER_TIMEOUT socket option. This specifies the maximum
971 amount of time in milliseconds that transmitted data may remain unac‐
972 knowledged before the TCP connection is forcibly closed. If set to 0,
973 then operating system’s default is used.
974 This is currently only supported on Linux.
976 Default: 0
978
980 The section [databases] defines the names of the databases that clients
981 of PgBouncer can connect to and specifies where those connections will
982 be routed. The section contains key=value lines like
983 dbname = connection string
985 where the key will be taken as a database name and the value as a con‐
987 nection string, consisting of key=value pairs of connection parameters,
988 described below (similar to libpq, but the actual libpq is not used and
989 the set of available features is different). Example:
990 foodb = host=host1.example.com port=5432
992 bardb = host=localhost dbname=bazdb
993 The database name can contain characters _0-9A-Za-z without quoting.
995 Names that contain other characters need to be quoted with standard SQL
996 identifier quoting: double quotes, with "" for a single instance of a
997 double quote.
998 The database name “pgbouncer” is reserved for the admin console and
1000 cannot be used as a key here.
1001 "*" acts as a fallback database: If the exact name does not exist, its
1003 value is taken as connection string for the requested database. For
1004 example, if there is an entry (and no other overriding entries)
1005 * = host=foo
1007 then a connection to PgBouncer specifying a database “bar” will effec‐
1009 tively behave as if an entry
1010 bar = host=foo dbname=bar
1012 exists (taking advantage of the default for dbname being the client-
1014 side database name; see below).
1015 Such automatically created database entries are cleaned up if they stay
1017 idle longer than the time specified by the autodb_idle_timeout parame‐
1018 ter.
1019 dbname
1021 Destination database name.
1022 Default: same as client-side database name
1024 host
1026 Host name or IP address to connect to. Host names are resolved at con‐
1027 nection time, the result is cached per dns_max_ttl parameter. When a
1028 host name’s resolution changes, existing server connections are auto‐
1029 matically closed when they are released (according to the pooling
1030 mode), and new server connections immediately use the new resolution.
1031 If DNS returns several results, they are used in a round-robin manner.
1032 If the value begins with /, then a Unix socket in the file-system name‐
1034 space is used. If the value begins with @, then a Unix socket in the
1035 abstract namespace is used.
1036 A comma-separated list of host names or addresses can be specified. In
1038 that case, connections are made in a round-robin manner. (If a host
1039 list contains host names that in turn resolve via DNS to multiple ad‐
1040 dresses, the round-robin systems operate independently. This is an im‐
1041 plementation dependency that is subject to change.) Note that in a
1042 list, all hosts must be available at all times: There are no mechanisms
1043 to skip unreachable hosts or to select only available hosts from a list
1044 or similar. (This is different from what a host list in libpq means.)
1045 Also note that this only affects how the destinations of new connec‐
1046 tions are chosen. See also the setting server_round_robin for how
1047 clients are assigned to already established server connections.
1048 Examples:
1050 host=localhost
1052 host=127.0.0.1
1053 host=2001:0db8:85a3:0000:0000:8a2e:0370:7334
1054 host=/var/run/postgresql
1055 host=192.168.0.1,192.168.0.2,192.168.0.3
1056 Default: not set, meaning to use a Unix socket
1058 port
1060 Default: 5432
1061 user
1063 If user= is set, all connections to the destination database will be
1064 done with the specified user, meaning that there will be only one pool
1065 for this database.
1066 Otherwise, PgBouncer logs into the destination database with the client
1068 user name, meaning that there will be one pool per user.
1069 password
1071 If no password is specified here, the password from the auth_file or
1072 auth_query will be used.
1073 auth_user
1075 Override of the global auth_user setting, if specified.
1076 pool_size
1078 Set the maximum size of pools for this database. If not set, the de‐
1079 fault_pool_size is used.
1080 min_pool_size
1082 Set the minimum pool size for this database. If not set, the global
1083 min_pool_size is used.
1084 Only enforced if at least one of the following is true:
1086 • this entry in the [database] section has a value set for the user key
1088 (aka forced user)
1089 • there is at least one client connected to the pool
1091 reserve_pool
1093 Set additional connections for this database. If not set, re‐
1094 serve_pool_size is used.
1095 connect_query
1097 Query to be executed after a connection is established, but before al‐
1098 lowing the connection to be used by any clients. If the query raises
1099 errors, they are logged but ignored otherwise.
1100 pool_mode
1102 Set the pool mode specific to this database. If not set, the default
1103 pool_mode is used.
1104 max_db_connections
1106 Configure a database-wide maximum (i.e. all pools within the database
1107 will not have more than this many server connections).
1108 client_encoding
1110 Ask specific client_encoding from server.
1111 datestyle
1113 Ask specific datestyle from server.
1114 timezone
1116 Ask specific timezone from server.
1117
1119 This section contains key=value lines like
1120 user1 = settings
1122 where the key will be taken as a user name and the value as a list of
1124 key=value pairs of configuration settings specific for this user. Ex‐
1125 ample:
1126 user1 = pool_mode=session
1128 Only a few settings are available here.
1130 pool_mode
1132 Set the pool mode to be used for all connections from this user. If
1133 not set, the database or default pool_mode is used.
1134 max_user_connections
1136 Configure a maximum for the user (i.e. all pools with the user will not
1137 have more than this many server connections).
1138
1140 The section [peers] defines the peers that PgBouncer can forward can‐
1141 cellation requests to and where those cancellation requests will be
1142 routed.
1143 PgBouncer processes can be peered together in a group by defining a
1145 peer_id value and a [peers] section in the configs of all the PgBouncer
1146 processes. These PgBouncer processes can then forward cancellations
1147 requests to the process that it originated from. This is needed to
1148 make cancellations work when multiple PgBouncer processes (possibly on
1149 different servers) are behind the same TCP load balancer. Cancellation
1150 requests are sent over different TCP connections than the query they
1151 are cancelling, so a TCP load balancer might send the cancellation re‐
1152 quest connection to a different process than the one that it was meant
1153 for. By peering them these cancellation requests eventually end up at
1154 the right process. A more in-depth explanation is provided in this
1155 recording of a conference talk (https://www.youtube.com/watch?v=M585Ff‐
1156 bboNA).
1157 The section contains key=value lines like
1159 peer_id = connection string
1161 Where the key will be taken as a peer_id and the value as a connection
1163 string, consisting of key=value pairs of connection parameters, de‐
1164 scribed below (similar to libpq, but the actual libpq is not used and
1165 the set of available features is different). Example:
1166 1 = host=host1.example.com
1168 2 = host=/tmp/pgbouncer-2 port=5555
1169 Note 1: For peering to work, the peer_id of each PgBouncer process in
1171 the group must be unique within the peered group. And the [peers] sec‐
1172 tion should contain entries for each of those peer ids. An example can
1173 be found in the examples section of these docs. It is allowed, but not
1174 necessary, for the [peers] section to contain the peer_id of the Pg‐
1175 Bouncer that the config is for. Such an entry will be ignored, but it
1176 is allowed to config management easy. Because it allows using the ex‐
1177 act same [peers] section for multiple configs.
1178 Note 2: Cross-version peering is supported as long as all peers are on
1180 the same side of the v1.21.0 version boundary. In v1.21.0 some break‐
1181 ing changes were made in how we encode the cancellation tokens that
1182 made them incompatible with the ones created by earlier versions.
1183 host
1185 Host name or IP address to connect to. Host names are resolved at con‐
1186 nection time, the result is cached per dns_max_ttl parameter. If DNS
1187 returns several results, they are used in a round-robin manner. But in
1188 general it’s not recommended to use a hostname that resolves to multi‐
1189 ple IPs, because then the cancel request might still be forwarded to
1190 the wrong node and it would need to be forwarded again (which is only
1191 allowed up to three times).
1192 If the value begins with /, then a Unix socket in the file-system name‐
1194 space is used. If the value begins with @, then a Unix socket in the
1195 abstract namespace is used.
1196 Examples:
1198 host=localhost
1200 host=127.0.0.1
1201 host=2001:0db8:85a3:0000:0000:8a2e:0370:7334
1202 host=/var/run/pgbouncer-1
1203 port
1205 Default: 6432
1206 pool_size
1208 Set the maximum number of cancel requests that can be in flight to the
1209 peer at the same time. It’s quite normal for cancel requests to arrive
1210 in bursts, e.g. when the backing Postgres server slow or down. So
1211 it’s important for pool_size to not be so low that it cannot handle
1212 these bursts.
1213 If not set, the default_pool_size is used.
1215
1217 The PgBouncer configuration file can contain include directives, which
1218 specify another configuration file to read and process. This allows
1219 splitting the configuration file into physically separate parts. The
1220 include directives look like this:
1221 %include filename
1223 If the file name is not an absolute path, it is taken as relative to
1225 the current working directory.
1226
1228 This section describes the format of the file specified by the
1229 auth_file setting. It is a text file in the following format:
1230 "username1" "password" ...
1232 "username2" "md5abcdef012342345" ...
1233 "username2" "SCRAM-SHA-256$<iterations>:<salt>$<storedkey>:<serverkey>"
1234 There should be at least 2 fields, surrounded by double quotes. The
1236 first field is the user name and the second is either a plain-text, a
1237 MD5-hashed password, or a SCRAM secret. PgBouncer ignores the rest of
1238 the line. Double quotes in a field value can be escaped by writing two
1239 double quotes.
1240 PostgreSQL MD5-hashed password format:
1242 "md5" + md5(password + username)
1244 So user admin with password 1234 will have MD5-hashed password
1246 md545f2603610af569b6155c45067268c6b.
1247 PostgreSQL SCRAM secret format:
1249 SCRAM-SHA-256$<iterations>:<salt>$<storedkey>:<serverkey>
1251 See the PostgreSQL documentation and RFC 5803 for details on this.
1253 The passwords or secrets stored in the authentication file serve two
1255 purposes. First, they are used to verify the passwords of incoming
1256 client connections, if a password-based authentication method is con‐
1257 figured. Second, they are used as the passwords for outgoing connec‐
1258 tions to the backend server, if the backend server requires password-
1259 based authentication (unless the password is specified directly in the
1260 database’s connection string). The latter works if the password is
1261 stored in plain text or MD5-hashed. SCRAM secrets can only be used for
1262 logging into a server if the client authentication also uses SCRAM, the
1263 PgBouncer database definition does not specify a user name, and the
1264 SCRAM secrets are identical in PgBouncer and the PostgreSQL server
1265 (same salt and iterations, not merely the same password). This is due
1266 to an inherent security property of SCRAM: The stored SCRAM secret can‐
1267 not by itself be used for deriving login credentials.
1268 The authentication file can be written by hand, but it’s also useful to
1270 generate it from some other list of users and passwords. See
1271 ./etc/mkauth.py for a sample script to generate the authentication file
1272 from the pg_shadow system table. Alternatively, use auth_query instead
1273 of auth_file to avoid having to maintain a separate authentication
1274 file.
1275
1277 The location of the HBA file is specified by the setting auth_hba_file.
1278 It is only used if auth_type is set to hba.
1279 The file follows the format of the PostgreSQL pg_hba.conf file (see
1281 <https://www.postgresql.org/docs/current/auth-pg-hba-conf.html>).
1282 • Supported record types: local, host, hostssl, hostnossl.
1284 • Database field: Supports all, sameuser, @file, multiple names. Not
1286 supported: replication, samerole, samegroup.
1287 • User name field: Supports all, @file, multiple names. Not supported:
1289 +groupname.
1290 • Address field: Supports IPv4, IPv6. Not supported: DNS names, domain
1292 prefixes.
1293 • Auth-method field: Only methods supported by PgBouncer’s auth_type
1295 are supported, plus peer and reject, but except any and pam, which
1296 only work globally. User name map (map=) parameter is not supported.
1297
1299 Small example configuration:
1300 [databases]
1302 template1 = host=localhost dbname=template1 auth_user=someuser
1303 [pgbouncer]
1305 pool_mode = session
1306 listen_port = 6432
1307 listen_addr = localhost
1308 auth_type = md5
1309 auth_file = users.txt
1310 logfile = pgbouncer.log
1311 pidfile = pgbouncer.pid
1312 admin_users = someuser
1313 stats_users = stat_collector
1314 Database examples:
1316 [databases]
1318 ; foodb over Unix socket
1320 foodb =
1321 ; redirect bardb to bazdb on localhost
1323 bardb = host=localhost dbname=bazdb
1324 ; access to destination database will go with single user
1326 forcedb = host=localhost port=300 user=baz password=foo client_encoding=UNICODE datestyle=ISO
1327 Example of a secure function for auth_query:
1329 CREATE OR REPLACE FUNCTION pgbouncer.user_lookup(in i_username text, out uname text, out phash text)
1331 RETURNS record AS $$
1332 BEGIN
1333 SELECT usename, passwd FROM pg_catalog.pg_shadow
1334 WHERE usename = i_username INTO uname, phash;
1335 RETURN;
1336 END;
1337 $$ LANGUAGE plpgsql SECURITY DEFINER;
1338 REVOKE ALL ON FUNCTION pgbouncer.user_lookup(text) FROM public, pgbouncer;
1339 GRANT EXECUTE ON FUNCTION pgbouncer.user_lookup(text) TO pgbouncer;
1340 Example configs for 2 peered PgBouncer processes to create a multi-core
1342 PgBouncer setup using so_reuseport. The config for the first process:
1343 [databases]
1345 postgres = host=localhost dbname=postgres
1346 [peers]
1348 1 = host=/tmp/pgbouncer1
1349 2 = host=/tmp/pgbouncer2
1350 [pgbouncer]
1352 listen_addr=127.0.0.1
1353 auth_file=auth_file.conf
1354 so_reuseport=1
1355 unix_socket_dir=/tmp/pgbouncer1
1356 peer_id=1
1357 The config for the second process:
1359 [databases]
1361 postgres = host=localhost dbname=postgres
1362 [peers]
1364 1 = host=/tmp/pgbouncer1
1365 2 = host=/tmp/pgbouncer2
1366 [pgbouncer]
1368 listen_addr=127.0.0.1
1369 auth_file=auth_file.conf
1370 so_reuseport=1
1371 ; only unix_socket_dir and peer_id are different
1372 unix_socket_dir=/tmp/pgbouncer2
1373 peer_id=2
1374
1376 pgbouncer(1) - man page for general usage, console commands
1377 <https://www.pgbouncer.org/>
13791.21.0 PGBOUNCER.INI(5)