1 <!-- $PostgreSQL: pgsql/doc/src/sgml/protocol.sgml,v 1.83 2010/02/22 18:12:04 momjian Exp $ -->
3 <chapter id="protocol">
4 <title>Frontend/Backend Protocol</title>
6 <indexterm zone="protocol">
7 <primary>protocol</primary>
8 <secondary>frontend-backend</secondary>
12 <productname>PostgreSQL</productname> uses a message-based protocol
13 for communication between frontends and backends (clients and servers).
14 The protocol is supported over <acronym>TCP/IP</acronym> and also over
15 Unix-domain sockets. Port number 5432 has been registered with IANA as
16 the customary TCP port number for servers supporting this protocol, but
17 in practice any non-privileged port number can be used.
21 This document describes version 3.0 of the protocol, implemented in
22 <productname>PostgreSQL</productname> 7.4 and later. For descriptions
23 of the earlier protocol versions, see previous releases of the
24 <productname>PostgreSQL</productname> documentation. A single server
25 can support multiple protocol versions. The initial
26 startup-request message tells the server which protocol version the
27 client is attempting to use, and then the server follows that protocol
32 In order to serve multiple clients efficiently, the server launches
33 a new <quote>backend</> process for each client.
34 In the current implementation, a new child
35 process is created immediately after an incoming connection is detected.
36 This is transparent to the protocol, however. For purposes of the
37 protocol, the terms <quote>backend</> and <quote>server</> are
38 interchangeable; likewise <quote>frontend</> and <quote>client</>
42 <sect1 id="protocol-overview">
43 <title>Overview</title>
46 The protocol has separate phases for startup and normal operation.
47 In the startup phase, the frontend opens a connection to the server
48 and authenticates itself to the satisfaction of the server. (This might
49 involve a single message, or multiple messages depending on the
50 authentication method being used.) If all goes well, the server then sends
51 status information to the frontend, and finally enters normal operation.
52 Except for the initial startup-request message, this part of the
53 protocol is driven by the server.
57 During normal operation, the frontend sends queries and
58 other commands to the backend, and the backend sends back query results
59 and other responses. There are a few cases (such as <command>NOTIFY</>)
61 backend will send unsolicited messages, but for the most part this portion
62 of a session is driven by frontend requests.
66 Termination of the session is normally by frontend choice, but can be
67 forced by the backend in certain cases. In any case, when the backend
68 closes the connection, it will roll back any open (incomplete) transaction
73 Within normal operation, SQL commands can be executed through either of
74 two sub-protocols. In the <quote>simple query</> protocol, the frontend
75 just sends a textual query string, which is parsed and immediately
76 executed by the backend. In the <quote>extended query</> protocol,
77 processing of queries is separated into multiple steps: parsing,
78 binding of parameter values, and execution. This offers flexibility
79 and performance benefits, at the cost of extra complexity.
83 Normal operation has additional sub-protocols for special operations
84 such as <command>COPY</>.
87 <sect2 id="protocol-message-concepts">
88 <title>Messaging Overview</title>
91 All communication is through a stream of messages. The first byte of a
92 message identifies the message type, and the next four bytes specify the
93 length of the rest of the message (this length count includes itself, but
94 not the message-type byte). The remaining contents of the message are
95 determined by the message type. For historical reasons, the very first
96 message sent by the client (the startup message) has no initial
101 To avoid losing synchronization with the message stream, both servers and
102 clients typically read an entire message into a buffer (using the byte
103 count) before attempting to process its contents. This allows easy
104 recovery if an error is detected while processing the contents. In
105 extreme situations (such as not having enough memory to buffer the
106 message), the receiver can use the byte count to determine how much
107 input to skip before it resumes reading messages.
111 Conversely, both servers and clients must take care never to send an
112 incomplete message. This is commonly done by marshaling the entire message
113 in a buffer before beginning to send it. If a communications failure
114 occurs partway through sending or receiving a message, the only sensible
115 response is to abandon the connection, since there is little hope of
116 recovering message-boundary synchronization.
120 <sect2 id="protocol-query-concepts">
121 <title>Extended Query Overview</title>
124 In the extended-query protocol, execution of SQL commands is divided
125 into multiple steps. The state retained between steps is represented
126 by two types of objects: <firstterm>prepared statements</> and
127 <firstterm>portals</>. A prepared statement represents the result of
128 parsing, semantic analysis, and (optionally) planning of a textual query
130 A prepared statement is not necessarily ready to execute, because it might
131 lack specific values for <firstterm>parameters</>. A portal represents
132 a ready-to-execute or already-partially-executed statement, with any
133 missing parameter values filled in. (For <command>SELECT</> statements,
134 a portal is equivalent to an open cursor, but we choose to use a different
135 term since cursors don't handle non-<command>SELECT</> statements.)
139 The overall execution cycle consists of a <firstterm>parse</> step,
140 which creates a prepared statement from a textual query string; a
141 <firstterm>bind</> step, which creates a portal given a prepared
142 statement and values for any needed parameters; and an
143 <firstterm>execute</> step that runs a portal's query. In the case of
144 a query that returns rows (<command>SELECT</>, <command>SHOW</>, etc),
145 the execute step can be told to fetch only
146 a limited number of rows, so that multiple execute steps might be needed
147 to complete the operation.
151 The backend can keep track of multiple prepared statements and portals
152 (but note that these exist only within a session, and are never shared
153 across sessions). Existing prepared statements and portals are
154 referenced by names assigned when they were created. In addition,
155 an <quote>unnamed</> prepared statement and portal exist. Although these
156 behave largely the same as named objects, operations on them are optimized
157 for the case of executing a query only once and then discarding it,
158 whereas operations on named objects are optimized on the expectation
163 <sect2 id="protocol-format-codes">
164 <title>Formats and Format Codes</title>
167 Data of a particular data type might be transmitted in any of several
168 different <firstterm>formats</>. As of <productname>PostgreSQL</> 7.4
169 the only supported formats are <quote>text</> and <quote>binary</>,
170 but the protocol makes provision for future extensions. The desired
171 format for any value is specified by a <firstterm>format code</>.
172 Clients can specify a format code for each transmitted parameter value
173 and for each column of a query result. Text has format code zero,
174 binary has format code one, and all other format codes are reserved
175 for future definition.
179 The text representation of values is whatever strings are produced
180 and accepted by the input/output conversion functions for the
181 particular data type. In the transmitted representation, there is
182 no trailing null character; the frontend must add one to received
183 values if it wants to process them as C strings.
184 (The text format does not allow embedded nulls, by the way.)
188 Binary representations for integers use network byte order (most
189 significant byte first). For other data types consult the documentation
190 or source code to learn about the binary representation. Keep in mind
191 that binary representations for complex data types might change across
192 server versions; the text format is usually the more portable choice.
197 <sect1 id="protocol-flow">
198 <title>Message Flow</title>
201 This section describes the message flow and the semantics of each
202 message type. (Details of the exact representation of each message
203 appear in <xref linkend="protocol-message-formats">.) There are
204 several different sub-protocols depending on the state of the
205 connection: start-up, query, function call,
206 <command>COPY</command>, and termination. There are also special
207 provisions for asynchronous operations (including notification
208 responses and command cancellation), which can occur at any time
209 after the start-up phase.
213 <title>Start-Up</title>
216 To begin a session, a frontend opens a connection to the server and sends
217 a startup message. This message includes the names of the user and of the
218 database the user wants to connect to; it also identifies the particular
219 protocol version to be used. (Optionally, the startup message can include
220 additional settings for run-time parameters.)
221 The server then uses this information and
222 the contents of its configuration files (such as
223 <filename>pg_hba.conf</filename>) to determine
224 whether the connection is provisionally acceptable, and what additional
225 authentication is required (if any).
229 The server then sends an appropriate authentication request message,
230 to which the frontend must reply with an appropriate authentication
231 response message (such as a password).
232 For all authentication methods except GSSAPI and SSPI, there is at most
233 one request and one response. In some methods, no response
234 at all is needed from the frontend, and so no authentication request
235 occurs. For GSSAPI and SSPI, multiple exchanges of packets may be needed
236 to complete the authentication.
240 The authentication cycle ends with the server either rejecting the
241 connection attempt (ErrorResponse), or sending AuthenticationOk.
245 The possible messages from the server in this phase are:
249 <term>ErrorResponse</term>
252 The connection attempt has been rejected.
253 The server then immediately closes the connection.
259 <term>AuthenticationOk</term>
262 The authentication exchange is successfully completed.
268 <term>AuthenticationKerberosV5</term>
271 The frontend must now take part in a Kerberos V5
272 authentication dialog (not described here, part of the
273 Kerberos specification) with the server. If this is
274 successful, the server responds with an AuthenticationOk,
275 otherwise it responds with an ErrorResponse.
281 <term>AuthenticationCleartextPassword</term>
284 The frontend must now send a PasswordMessage containing the
285 password in clear-text form. If
286 this is the correct password, the server responds with an
287 AuthenticationOk, otherwise it responds with an ErrorResponse.
293 <term>AuthenticationMD5Password</term>
296 The frontend must now send a PasswordMessage containing the
297 password encrypted via MD5, using the 4-character salt
298 specified in the AuthenticationMD5Password message. If
299 this is the correct password, the server responds with an
300 AuthenticationOk, otherwise it responds with an ErrorResponse.
306 <term>AuthenticationSCMCredential</term>
309 This response is only possible for local Unix-domain connections
310 on platforms that support SCM credential messages. The frontend
311 must issue an SCM credential message and then send a single data
312 byte. (The contents of the data byte are uninteresting; it's
313 only used to ensure that the server waits long enough to receive
314 the credential message.) If the credential is acceptable,
315 the server responds with an
316 AuthenticationOk, otherwise it responds with an ErrorResponse.
322 <term>AuthenticationGSS</term>
325 The frontend must now initiate a GSSAPI negotiation. The frontend
326 will send a PasswordMessage with the first part of the GSSAPI
327 data stream in response to this. If further messages are needed,
328 the server will respond with AuthenticationGSSContinue.
334 <term>AuthenticationSSPI</term>
337 The frontend must now initiate a SSPI negotiation. The frontend
338 will send a PasswordMessage with the first part of the SSPI
339 data stream in response to this. If further messages are needed,
340 the server will respond with AuthenticationGSSContinue.
346 <term>AuthenticationGSSContinue</term>
349 This message contains the response data from the previous step
350 of GSSAPI or SSPI negotiation (AuthenticationGSS, AuthenticationSSPI
351 or a previous AuthenticationGSSContinue). If the GSSAPI
352 or SSPI data in this message
353 indicates more data is needed to complete the authentication,
354 the frontend must send that data as another PasswordMessage. If
355 GSSAPI or SSPI authentication is completed by this message, the server
356 will next send AuthenticationOk to indicate successful authentication
357 or ErrorResponse to indicate failure.
366 If the frontend does not support the authentication method
367 requested by the server, then it should immediately close the
372 After having received AuthenticationOk, the frontend must wait
373 for further messages from the server. In this phase a backend process
374 is being started, and the frontend is just an interested bystander.
375 It is still possible for the startup attempt
376 to fail (ErrorResponse), but in the normal case the backend will send
377 some ParameterStatus messages, BackendKeyData, and finally ReadyForQuery.
381 During this phase the backend will attempt to apply any additional
382 run-time parameter settings that were given in the startup message.
383 If successful, these values become session defaults. An error causes
384 ErrorResponse and exit.
388 The possible messages from the backend in this phase are:
392 <term>BackendKeyData</term>
395 This message provides secret-key data that the frontend must
396 save if it wants to be able to issue cancel requests later.
397 The frontend should not respond to this message, but should
398 continue listening for a ReadyForQuery message.
404 <term>ParameterStatus</term>
407 This message informs the frontend about the current (initial)
408 setting of backend parameters, such as <xref
409 linkend="guc-client-encoding"> or <xref linkend="guc-datestyle">.
410 The frontend can ignore this message, or record the settings
411 for its future use; see <xref linkend="protocol-async"> for
412 more details. The frontend should not respond to this
413 message, but should continue listening for a ReadyForQuery
420 <term>ReadyForQuery</term>
423 Start-up is completed. The frontend can now issue commands.
429 <term>ErrorResponse</term>
432 Start-up failed. The connection is closed after sending this
439 <term>NoticeResponse</term>
442 A warning message has been issued. The frontend should
443 display the message but continue listening for ReadyForQuery
452 The ReadyForQuery message is the same one that the backend will
453 issue after each command cycle. Depending on the coding needs of
454 the frontend, it is reasonable to consider ReadyForQuery as
455 starting a command cycle, or to consider ReadyForQuery as ending the
456 start-up phase and each subsequent command cycle.
461 <title>Simple Query</title>
464 A simple query cycle is initiated by the frontend sending a Query message
465 to the backend. The message includes an SQL command (or commands)
466 expressed as a text string.
467 The backend then sends one or more response
468 messages depending on the contents of the query command string,
469 and finally a ReadyForQuery response message. ReadyForQuery
470 informs the frontend that it can safely send a new command.
471 (It is not actually necessary for the frontend to wait for
472 ReadyForQuery before issuing another command, but the frontend must
473 then take responsibility for figuring out what happens if the earlier
474 command fails and already-issued later commands succeed.)
478 The possible response messages from the backend are:
482 <term>CommandComplete</term>
485 An SQL command completed normally.
491 <term>CopyInResponse</term>
494 The backend is ready to copy data from the frontend to a
495 table; see <xref linkend="protocol-copy">.
501 <term>CopyOutResponse</term>
504 The backend is ready to copy data from a table to the
505 frontend; see <xref linkend="protocol-copy">.
511 <term>RowDescription</term>
514 Indicates that rows are about to be returned in response to
515 a <command>SELECT</command>, <command>FETCH</command>, etc query.
516 The contents of this message describe the column layout of the rows.
517 This will be followed by a DataRow message for each row being returned
527 One of the set of rows returned by
528 a <command>SELECT</command>, <command>FETCH</command>, etc query.
534 <term>EmptyQueryResponse</term>
537 An empty query string was recognized.
543 <term>ErrorResponse</term>
546 An error has occurred.
552 <term>ReadyForQuery</term>
555 Processing of the query string is complete. A separate
556 message is sent to indicate this because the query string might
557 contain multiple SQL commands. (CommandComplete marks the
558 end of processing one SQL command, not the whole string.)
559 ReadyForQuery will always be sent, whether processing
560 terminates successfully or with an error.
566 <term>NoticeResponse</term>
569 A warning message has been issued in relation to the query.
570 Notices are in addition to other responses, i.e., the backend
571 will continue processing the command.
580 The response to a <command>SELECT</> query (or other queries that
581 return row sets, such as <command>EXPLAIN</> or <command>SHOW</>)
582 normally consists of RowDescription, zero or more
583 DataRow messages, and then CommandComplete.
584 <command>COPY</> to or from the frontend invokes special protocol
585 as described in <xref linkend="protocol-copy">.
586 All other query types normally produce only
587 a CommandComplete message.
591 Since a query string could contain several queries (separated by
592 semicolons), there might be several such response sequences before the
593 backend finishes processing the query string. ReadyForQuery is issued
594 when the entire string has been processed and the backend is ready to
595 accept a new query string.
599 If a completely empty (no contents other than whitespace) query string
600 is received, the response is EmptyQueryResponse followed by ReadyForQuery.
604 In the event of an error, ErrorResponse is issued followed by
605 ReadyForQuery. All further processing of the query string is aborted by
606 ErrorResponse (even if more queries remained in it). Note that this
607 might occur partway through the sequence of messages generated by an
612 In simple Query mode, the format of retrieved values is always text,
613 except when the given command is a <command>FETCH</> from a cursor
614 declared with the <literal>BINARY</> option. In that case, the
615 retrieved values are in binary format. The format codes given in
616 the RowDescription message tell which format is being used.
620 A frontend must be prepared to accept ErrorResponse and
621 NoticeResponse messages whenever it is expecting any other type of
622 message. See also <xref linkend="protocol-async"> concerning messages
623 that the backend might generate due to outside events.
627 Recommended practice is to code frontends in a state-machine style
628 that will accept any message type at any time that it could make sense,
629 rather than wiring in assumptions about the exact sequence of messages.
633 <sect2 id="protocol-flow-ext-query">
634 <title>Extended Query</title>
637 The extended query protocol breaks down the above-described simple
638 query protocol into multiple steps. The results of preparatory
639 steps can be re-used multiple times for improved efficiency.
640 Furthermore, additional features are available, such as the possibility
641 of supplying data values as separate parameters instead of having to
642 insert them directly into a query string.
646 In the extended protocol, the frontend first sends a Parse message,
647 which contains a textual query string, optionally some information
648 about data types of parameter placeholders, and the
649 name of a destination prepared-statement object (an empty string
650 selects the unnamed prepared statement). The response is
651 either ParseComplete or ErrorResponse. Parameter data types can be
652 specified by OID; if not given, the parser attempts to infer the
653 data types in the same way as it would do for untyped literal string
659 A parameter data type can be left unspecified by setting it to zero,
660 or by making the array of parameter type OIDs shorter than the
661 number of parameter symbols (<literal>$</><replaceable>n</>)
662 used in the query string. Another special case is that a parameter's
663 type can be specified as <type>void</> (that is, the OID of the
664 <type>void</> pseudotype). This is meant to allow parameter symbols
665 to be used for function parameters that are actually OUT parameters.
666 Ordinarily there is no context in which a <type>void</> parameter
667 could be used, but if such a parameter symbol appears in a function's
668 parameter list, it is effectively ignored. For example, a function
669 call such as <literal>foo($1,$2,$3,$4)</> could match a function with
670 two IN and two OUT arguments, if <literal>$3</> and <literal>$4</>
671 are specified as having type <type>void</>.
677 The query string contained in a Parse message cannot include more
678 than one SQL statement; else a syntax error is reported. This
679 restriction does not exist in the simple-query protocol, but it
680 does exist in the extended protocol, because allowing prepared
681 statements or portals to contain multiple commands would complicate
687 If successfully created, a named prepared-statement object lasts till
688 the end of the current session, unless explicitly destroyed. An unnamed
689 prepared statement lasts only until the next Parse statement specifying
690 the unnamed statement as destination is issued. (Note that a simple
691 Query message also destroys the unnamed statement.) Named prepared
692 statements must be explicitly closed before they can be redefined by
693 a Parse message, but this is not required for the unnamed statement.
694 Named prepared statements can also be created and accessed at the SQL
695 command level, using <command>PREPARE</> and <command>EXECUTE</>.
699 Once a prepared statement exists, it can be readied for execution using a
700 Bind message. The Bind message gives the name of the source prepared
701 statement (empty string denotes the unnamed prepared statement), the name
702 of the destination portal (empty string denotes the unnamed portal), and
703 the values to use for any parameter placeholders present in the prepared
705 supplied parameter set must match those needed by the prepared statement.
706 (If you declared any <type>void</> parameters in the Parse message,
707 pass NULL values for them in the Bind message.)
708 Bind also specifies the format to use for any data returned
709 by the query; the format can be specified overall, or per-column.
710 The response is either BindComplete or ErrorResponse.
715 The choice between text and binary output is determined by the format
716 codes given in Bind, regardless of the SQL command involved. The
717 <literal>BINARY</> attribute in cursor declarations is irrelevant when
718 using extended query protocol.
723 Query planning for named prepared-statement objects occurs when the Parse
724 message is processed. If a query will be repeatedly executed with
725 different parameters, it might be beneficial to send a single Parse message
726 containing a parameterized query, followed by multiple Bind
727 and Execute messages. This will avoid replanning the query on each
732 The unnamed prepared statement is likewise planned during Parse processing
733 if the Parse message defines no parameters. But if there are parameters,
734 query planning occurs every time Bind parameters are supplied. This allows the
735 planner to make use of the actual values of the parameters provided by
736 each Bind message, rather than use generic estimates.
741 Query plans generated from a parameterized query might be less
742 efficient than query plans generated from an equivalent query with actual
743 parameter values substituted. The query planner cannot make decisions
744 based on actual parameter values (for example, index selectivity) when
745 planning a parameterized query assigned to a named prepared-statement
746 object. This possible penalty is avoided when using the unnamed
747 statement, since it is not planned until actual parameter values are
748 available. The cost is that planning must occur afresh for each Bind,
749 even if the query stays the same.
754 If successfully created, a named portal object lasts till the end of the
755 current transaction, unless explicitly destroyed. An unnamed portal is
756 destroyed at the end of the transaction, or as soon as the next Bind
757 statement specifying the unnamed portal as destination is issued. (Note
758 that a simple Query message also destroys the unnamed portal.) Named
759 portals must be explicitly closed before they can be redefined by a Bind
760 message, but this is not required for the unnamed portal.
761 Named portals can also be created and accessed at the SQL
762 command level, using <command>DECLARE CURSOR</> and <command>FETCH</>.
766 Once a portal exists, it can be executed using an Execute message.
767 The Execute message specifies the portal name (empty string denotes the
769 a maximum result-row count (zero meaning <quote>fetch all rows</>).
770 The result-row count is only meaningful for portals
771 containing commands that return row sets; in other cases the command is
772 always executed to completion, and the row count is ignored.
774 responses to Execute are the same as those described above for queries
775 issued via simple query protocol, except that Execute doesn't cause
776 ReadyForQuery or RowDescription to be issued.
780 If Execute terminates before completing the execution of a portal
781 (due to reaching a nonzero result-row count), it will send a
782 PortalSuspended message; the appearance of this message tells the frontend
783 that another Execute should be issued against the same portal to
784 complete the operation. The CommandComplete message indicating
785 completion of the source SQL command is not sent until
786 the portal's execution is completed. Therefore, an Execute phase is
787 always terminated by the appearance of exactly one of these messages:
788 CommandComplete, EmptyQueryResponse (if the portal was created from
789 an empty query string), ErrorResponse, or PortalSuspended.
793 At completion of each series of extended-query messages, the frontend
794 should issue a Sync message. This parameterless message causes the
795 backend to close the current transaction if it's not inside a
796 <command>BEGIN</>/<command>COMMIT</> transaction block (<quote>close</>
797 meaning to commit if no error, or roll back if error). Then a
798 ReadyForQuery response is issued. The purpose of Sync is to provide
799 a resynchronization point for error recovery. When an error is detected
800 while processing any extended-query message, the backend issues
801 ErrorResponse, then reads and discards messages until a Sync is reached,
802 then issues ReadyForQuery and returns to normal message processing.
803 (But note that no skipping occurs if an error is detected
804 <emphasis>while</> processing Sync — this ensures that there is one
805 and only one ReadyForQuery sent for each Sync.)
810 Sync does not cause a transaction block opened with <command>BEGIN</>
811 to be closed. It is possible to detect this situation since the
812 ReadyForQuery message includes transaction status information.
817 In addition to these fundamental, required operations, there are several
818 optional operations that can be used with extended-query protocol.
822 The Describe message (portal variant) specifies the name of an existing
823 portal (or an empty string for the unnamed portal). The response is a
824 RowDescription message describing the rows that will be returned by
825 executing the portal; or a NoData message if the portal does not contain a
826 query that will return rows; or ErrorResponse if there is no such portal.
830 The Describe message (statement variant) specifies the name of an existing
831 prepared statement (or an empty string for the unnamed prepared
832 statement). The response is a ParameterDescription message describing the
833 parameters needed by the statement, followed by a RowDescription message
834 describing the rows that will be returned when the statement is eventually
835 executed (or a NoData message if the statement will not return rows).
836 ErrorResponse is issued if there is no such prepared statement. Note that
837 since Bind has not yet been issued, the formats to be used for returned
838 columns are not yet known to the backend; the format code fields in the
839 RowDescription message will be zeroes in this case.
844 In most scenarios the frontend should issue one or the other variant
845 of Describe before issuing Execute, to ensure that it knows how to
846 interpret the results it will get back.
851 The Close message closes an existing prepared statement or portal
852 and releases resources. It is not an error to issue Close against
853 a nonexistent statement or portal name. The response is normally
854 CloseComplete, but could be ErrorResponse if some difficulty is
855 encountered while releasing resources. Note that closing a prepared
856 statement implicitly closes any open portals that were constructed
861 The Flush message does not cause any specific output to be generated,
862 but forces the backend to deliver any data pending in its output
863 buffers. A Flush must be sent after any extended-query command except
864 Sync, if the frontend wishes to examine the results of that command before
865 issuing more commands. Without Flush, messages returned by the backend
866 will be combined into the minimum possible number of packets to minimize
872 The simple Query message is approximately equivalent to the series Parse,
873 Bind, portal Describe, Execute, Close, Sync, using the unnamed prepared
874 statement and portal objects and no parameters. One difference is that
875 it will accept multiple SQL statements in the query string, automatically
876 performing the bind/describe/execute sequence for each one in succession.
877 Another difference is that it will not return ParseComplete, BindComplete,
878 CloseComplete, or NoData messages.
884 <title>Function Call</title>
887 The Function Call sub-protocol allows the client to request a direct
888 call of any function that exists in the database's
889 <structname>pg_proc</structname> system catalog. The client must have
890 execute permission for the function.
895 The Function Call sub-protocol is a legacy feature that is probably best
896 avoided in new code. Similar results can be accomplished by setting up
897 a prepared statement that does <literal>SELECT function($1, ...)</>.
898 The Function Call cycle can then be replaced with Bind/Execute.
903 A Function Call cycle is initiated by the frontend sending a
904 FunctionCall message to the backend. The backend then sends one
905 or more response messages depending on the results of the function
906 call, and finally a ReadyForQuery response message. ReadyForQuery
907 informs the frontend that it can safely send a new query or
912 The possible response messages from the backend are:
916 <term>ErrorResponse</term>
919 An error has occurred.
925 <term>FunctionCallResponse</term>
928 The function call was completed and returned the result given
930 (Note that the Function Call protocol can only handle a single
931 scalar result, not a row type or set of results.)
937 <term>ReadyForQuery</term>
940 Processing of the function call is complete. ReadyForQuery
941 will always be sent, whether processing terminates
942 successfully or with an error.
948 <term>NoticeResponse</term>
951 A warning message has been issued in relation to the function
952 call. Notices are in addition to other responses, i.e., the
953 backend will continue processing the command.
961 <sect2 id="protocol-copy">
962 <title>COPY Operations</title>
965 The <command>COPY</> command allows high-speed bulk data transfer
966 to or from the server. Copy-in and copy-out operations each switch
967 the connection into a distinct sub-protocol, which lasts until the
968 operation is completed.
972 Copy-in mode (data transfer to the server) is initiated when the
973 backend executes a <command>COPY FROM STDIN</> SQL statement. The backend
974 sends a CopyInResponse message to the frontend. The frontend should
975 then send zero or more CopyData messages, forming a stream of input
976 data. (The message boundaries are not required to have anything to do
977 with row boundaries, although that is often a reasonable choice.)
978 The frontend can terminate the copy-in mode by sending either a CopyDone
979 message (allowing successful termination) or a CopyFail message (which
980 will cause the <command>COPY</> SQL statement to fail with an
981 error). The backend then reverts to the command-processing mode it was
982 in before the <command>COPY</> started, which will be either simple or
983 extended query protocol. It will next send either CommandComplete
984 (if successful) or ErrorResponse (if not).
988 In the event of a backend-detected error during copy-in mode (including
989 receipt of a CopyFail message), the backend will issue an ErrorResponse
990 message. If the <command>COPY</> command was issued via an extended-query
991 message, the backend will now discard frontend messages until a Sync
992 message is received, then it will issue ReadyForQuery and return to normal
993 processing. If the <command>COPY</> command was issued in a simple
994 Query message, the rest of that message is discarded and ReadyForQuery
995 is issued. In either case, any subsequent CopyData, CopyDone, or CopyFail
996 messages issued by the frontend will simply be dropped.
1000 The backend will ignore Flush and Sync messages received during copy-in
1001 mode. Receipt of any other non-copy message type constitutes an error
1002 that will abort the copy-in state as described above. (The exception for
1003 Flush and Sync is for the convenience of client libraries that always
1004 send Flush or Sync after an Execute message, without checking whether
1005 the command to be executed is a <command>COPY FROM STDIN</>.)
1009 Copy-out mode (data transfer from the server) is initiated when the
1010 backend executes a <command>COPY TO STDOUT</> SQL statement. The backend
1011 sends a CopyOutResponse message to the frontend, followed by
1012 zero or more CopyData messages (always one per row), followed by CopyDone.
1013 The backend then reverts to the command-processing mode it was
1014 in before the <command>COPY</> started, and sends CommandComplete.
1015 The frontend cannot abort the transfer (except by closing the connection
1016 or issuing a Cancel request),
1017 but it can discard unwanted CopyData and CopyDone messages.
1021 In the event of a backend-detected error during copy-out mode,
1022 the backend will issue an ErrorResponse message and revert to normal
1023 processing. The frontend should treat receipt of ErrorResponse as
1024 terminating the copy-out mode.
1028 It is possible for NoticeResponse and ParameterStatus messages to be
1029 interspersed between CopyData messages; frontends must handle these cases,
1030 and should be prepared for other asynchronous message types as well (see
1031 <xref linkend="protocol-async">). Otherwise, any message type other than
1032 CopyData or CopyDone may be treated as terminating copy-out mode.
1036 The CopyInResponse and CopyOutResponse messages include fields that
1037 inform the frontend of the number of columns per row and the format
1038 codes being used for each column. (As of the present implementation,
1039 all columns in a given <command>COPY</> operation will use the same
1040 format, but the message design does not assume this.)
1044 <sect2 id="protocol-async">
1045 <title>Asynchronous Operations</title>
1048 There are several cases in which the backend will send messages that
1049 are not specifically prompted by the frontend's command stream.
1050 Frontends must be prepared to deal with these messages at any time,
1051 even when not engaged in a query.
1052 At minimum, one should check for these cases before beginning to
1053 read a query response.
1057 It is possible for NoticeResponse messages to be generated due to
1058 outside activity; for example, if the database administrator commands
1059 a <quote>fast</> database shutdown, the backend will send a NoticeResponse
1060 indicating this fact before closing the connection. Accordingly,
1061 frontends should always be prepared to accept and display NoticeResponse
1062 messages, even when the connection is nominally idle.
1066 ParameterStatus messages will be generated whenever the active
1067 value changes for any of the parameters the backend believes the
1068 frontend should know about. Most commonly this occurs in response
1069 to a <command>SET</> SQL command executed by the frontend, and
1070 this case is effectively synchronous — but it is also possible
1071 for parameter status changes to occur because the administrator
1072 changed a configuration file and then sent the
1073 <systemitem>SIGHUP</systemitem> signal to the server. Also,
1074 if a <command>SET</command> command is rolled back, an appropriate
1075 ParameterStatus message will be generated to report the current
1080 At present there is a hard-wired set of parameters for which
1081 ParameterStatus will be generated: they are
1082 <literal>server_version</>,
1083 <literal>server_encoding</>,
1084 <literal>client_encoding</>,
1085 <literal>application_name</>,
1086 <literal>is_superuser</>,
1087 <literal>session_authorization</>,
1088 <literal>DateStyle</>,
1089 <literal>IntervalStyle</>,
1090 <literal>TimeZone</>,
1091 <literal>integer_datetimes</>, and
1092 <literal>standard_conforming_strings</>.
1093 (<literal>server_encoding</>, <literal>TimeZone</>, and
1094 <literal>integer_datetimes</> were not reported by releases before 8.0;
1095 <literal>standard_conforming_strings</> was not reported by releases
1097 <literal>IntervalStyle</> was not reported by releases before 8.4;
1098 <literal>application_name</> was not reported by releases before 9.0.)
1100 <literal>server_version</>,
1101 <literal>server_encoding</> and
1102 <literal>integer_datetimes</>
1103 are pseudo-parameters that cannot change after startup.
1104 This set might change in the future, or even become configurable.
1105 Accordingly, a frontend should simply ignore ParameterStatus for
1106 parameters that it does not understand or care about.
1110 If a frontend issues a <command>LISTEN</command> command, then the
1111 backend will send a NotificationResponse message (not to be
1112 confused with NoticeResponse!) whenever a
1113 <command>NOTIFY</command> command is executed for the same
1119 At present, NotificationResponse can only be sent outside a
1120 transaction, and thus it will not occur in the middle of a
1121 command-response series, though it might occur just before ReadyForQuery.
1122 It is unwise to design frontend logic that assumes that, however.
1123 Good practice is to be able to accept NotificationResponse at any
1124 point in the protocol.
1130 <title>Cancelling Requests in Progress</title>
1133 During the processing of a query, the frontend might request
1134 cancellation of the query. The cancel request is not sent
1135 directly on the open connection to the backend for reasons of
1136 implementation efficiency: we don't want to have the backend
1137 constantly checking for new input from the frontend during query
1138 processing. Cancel requests should be relatively infrequent, so
1139 we make them slightly cumbersome in order to avoid a penalty in
1144 To issue a cancel request, the frontend opens a new connection to
1145 the server and sends a CancelRequest message, rather than the
1146 StartupMessage message that would ordinarily be sent across a new
1147 connection. The server will process this request and then close
1148 the connection. For security reasons, no direct reply is made to
1149 the cancel request message.
1153 A CancelRequest message will be ignored unless it contains the
1154 same key data (PID and secret key) passed to the frontend during
1155 connection start-up. If the request matches the PID and secret
1156 key for a currently executing backend, the processing of the
1157 current query is aborted. (In the existing implementation, this is
1158 done by sending a special signal to the backend process that is
1159 processing the query.)
1163 The cancellation signal might or might not have any effect — for
1164 example, if it arrives after the backend has finished processing
1165 the query, then it will have no effect. If the cancellation is
1166 effective, it results in the current command being terminated
1167 early with an error message.
1171 The upshot of all this is that for reasons of both security and
1172 efficiency, the frontend has no direct way to tell whether a
1173 cancel request has succeeded. It must continue to wait for the
1174 backend to respond to the query. Issuing a cancel simply improves
1175 the odds that the current query will finish soon, and improves the
1176 odds that it will fail with an error message instead of
1181 Since the cancel request is sent across a new connection to the
1182 server and not across the regular frontend/backend communication
1183 link, it is possible for the cancel request to be issued by any
1184 process, not just the frontend whose query is to be canceled.
1185 This might provide additional flexibility when building
1186 multiple-process applications. It also introduces a security
1187 risk, in that unauthorized persons might try to cancel queries.
1188 The security risk is addressed by requiring a dynamically
1189 generated secret key to be supplied in cancel requests.
1194 <title>Termination</title>
1197 The normal, graceful termination procedure is that the frontend
1198 sends a Terminate message and immediately closes the connection.
1199 On receipt of this message, the backend closes the connection and
1204 In rare cases (such as an administrator-commanded database shutdown)
1205 the backend might disconnect without any frontend request to do so.
1206 In such cases the backend will attempt to send an error or notice message
1207 giving the reason for the disconnection before it closes the connection.
1211 Other termination scenarios arise from various failure cases, such as core
1212 dump at one end or the other, loss of the communications link, loss of
1213 message-boundary synchronization, etc. If either frontend or backend sees
1214 an unexpected closure of the connection, it should clean
1215 up and terminate. The frontend has the option of launching a new backend
1216 by recontacting the server if it doesn't want to terminate itself.
1217 Closing the connection is also advisable if an unrecognizable message type
1218 is received, since this probably indicates loss of message-boundary sync.
1222 For either normal or abnormal termination, any open transaction is
1223 rolled back, not committed. One should note however that if a
1224 frontend disconnects while a non-<command>SELECT</command> query
1225 is being processed, the backend will probably finish the query
1226 before noticing the disconnection. If the query is outside any
1227 transaction block (<command>BEGIN</> ... <command>COMMIT</>
1228 sequence) then its results might be committed before the
1229 disconnection is recognized.
1234 <title><acronym>SSL</acronym> Session Encryption</title>
1237 If <productname>PostgreSQL</> was built with
1238 <acronym>SSL</acronym> support, frontend/backend communications
1239 can be encrypted using <acronym>SSL</acronym>. This provides
1240 communication security in environments where attackers might be
1241 able to capture the session traffic. For more information on
1242 encrypting <productname>PostgreSQL</productname> sessions with
1243 <acronym>SSL</acronym>, see <xref linkend="ssl-tcp">.
1247 To initiate an <acronym>SSL</acronym>-encrypted connection, the
1248 frontend initially sends an SSLRequest message rather than a
1249 StartupMessage. The server then responds with a single byte
1250 containing <literal>S</> or <literal>N</>, indicating that it is
1251 willing or unwilling to perform <acronym>SSL</acronym>,
1252 respectively. The frontend might close the connection at this point
1253 if it is dissatisfied with the response. To continue after
1254 <literal>S</>, perform an <acronym>SSL</acronym> startup handshake
1255 (not described here, part of the <acronym>SSL</acronym>
1256 specification) with the server. If this is successful, continue
1257 with sending the usual StartupMessage. In this case the
1258 StartupMessage and all subsequent data will be
1259 <acronym>SSL</acronym>-encrypted. To continue after
1260 <literal>N</>, send the usual StartupMessage and proceed without
1265 The frontend should also be prepared to handle an ErrorMessage
1266 response to SSLRequest from the server. This would only occur if
1267 the server predates the addition of <acronym>SSL</acronym> support
1268 to <productname>PostgreSQL</>. In this case the connection must
1269 be closed, but the frontend might choose to open a fresh connection
1270 and proceed without requesting <acronym>SSL</acronym>.
1274 An initial SSLRequest can also be used in a connection that is being
1275 opened to send a CancelRequest message.
1279 While the protocol itself does not provide a way for the server to
1280 force <acronym>SSL</acronym> encryption, the administrator can
1281 configure the server to reject unencrypted sessions as a byproduct
1282 of authentication checking.
1287 <sect1 id="protocol-message-types">
1288 <title>Message Data Types</title>
1291 This section describes the base data types used in messages.
1297 Int<replaceable>n</replaceable>(<replaceable>i</replaceable>)
1301 An <replaceable>n</replaceable>-bit integer in network byte
1302 order (most significant byte first).
1303 If <replaceable>i</replaceable> is specified it
1304 is the exact value that will appear, otherwise the value
1305 is variable. Eg. Int16, Int32(42).
1312 Int<replaceable>n</replaceable>[<replaceable>k</replaceable>]
1316 An array of <replaceable>k</replaceable>
1317 <replaceable>n</replaceable>-bit integers, each in network
1318 byte order. The array length <replaceable>k</replaceable>
1319 is always determined by an earlier field in the message.
1327 String(<replaceable>s</replaceable>)
1331 A null-terminated string (C-style string). There is no
1332 specific length limitation on strings.
1333 If <replaceable>s</replaceable> is specified it is the exact
1334 value that will appear, otherwise the value is variable.
1335 Eg. String, String("user").
1340 <emphasis>There is no predefined limit</emphasis> on the length of a string
1341 that can be returned by the backend. Good coding strategy for a frontend
1342 is to use an expandable buffer so that anything that fits in memory can be
1343 accepted. If that's not feasible, read the full string and discard trailing
1344 characters that don't fit into your fixed-size buffer.
1352 Byte<replaceable>n</replaceable>(<replaceable>c</replaceable>)
1356 Exactly <replaceable>n</replaceable> bytes. If the field
1357 width <replaceable>n</replaceable> is not a constant, it is
1358 always determinable from an earlier field in the message.
1359 If <replaceable>c</replaceable> is specified it is the exact
1360 value. Eg. Byte2, Byte1('\n').
1369 <sect1 id="protocol-message-formats">
1370 <title>Message Formats</title>
1373 This section describes the detailed format of each message. Each is marked to
1374 indicate that it can be sent by a frontend (F), a backend (B), or both
1376 Notice that although each message includes a byte count at the beginning,
1377 the message format is defined so that the message end can be found without
1378 reference to the byte count. This aids validity checking. (The CopyData
1379 message is an exception, because it forms part of a data stream; the contents
1380 of any individual CopyData message cannot be interpretable on their own.)
1388 AuthenticationOk (B)
1400 Identifies the message as an authentication request.
1410 Length of message contents in bytes, including self.
1420 Specifies that the authentication was successful.
1433 AuthenticationKerberosV5 (B)
1445 Identifies the message as an authentication request.
1455 Length of message contents in bytes, including self.
1465 Specifies that Kerberos V5 authentication is required.
1477 AuthenticationCleartextPassword (B)
1489 Identifies the message as an authentication request.
1499 Length of message contents in bytes, including self.
1509 Specifies that a clear-text password is required.
1521 AuthenticationMD5Password (B)
1533 Identifies the message as an authentication request.
1543 Length of message contents in bytes, including self.
1553 Specifies that an MD5-encrypted password is required.
1563 The salt to use when encrypting the password.
1576 AuthenticationSCMCredential (B)
1588 Identifies the message as an authentication request.
1598 Length of message contents in bytes, including self.
1608 Specifies that an SCM credentials message is required.
1621 AuthenticationGSS (B)
1633 Identifies the message as an authentication request.
1643 Length of message contents in bytes, including self.
1653 Specifies that GSSAPI authentication is required.
1666 AuthenticationSSPI (B)
1678 Identifies the message as an authentication request.
1688 Length of message contents in bytes, including self.
1698 Specifies that SSPI authentication is required.
1709 AuthenticationGSSContinue (B)
1721 Identifies the message as an authentication request.
1731 Length of message contents in bytes, including self.
1741 Specifies that this message contains GSSAPI or SSPI data.
1747 Byte<replaceable>n</replaceable>
1751 GSSAPI or SSPI authentication data.
1776 Identifies the message as cancellation key data.
1777 The frontend must save these values if it wishes to be
1778 able to issue CancelRequest messages later.
1788 Length of message contents in bytes, including self.
1798 The process ID of this backend.
1808 The secret key of this backend.
1833 Identifies the message as a Bind command.
1843 Length of message contents in bytes, including self.
1853 The name of the destination portal
1854 (an empty string selects the unnamed portal).
1864 The name of the source prepared statement
1865 (an empty string selects the unnamed prepared statement).
1875 The number of parameter format codes that follow
1876 (denoted <replaceable>C</> below).
1877 This can be zero to indicate that there are no parameters
1878 or that the parameters all use the default format (text);
1879 or one, in which case the specified format code is applied
1880 to all parameters; or it can equal the actual number of
1887 Int16[<replaceable>C</>]
1891 The parameter format codes. Each must presently be
1892 zero (text) or one (binary).
1902 The number of parameter values that follow (possibly zero).
1903 This must match the number of parameters needed by the query.
1908 Next, the following pair of fields appear for each parameter:
1916 The length of the parameter value, in bytes (this count
1917 does not include itself). Can be zero.
1918 As a special case, -1 indicates a NULL parameter value.
1919 No value bytes follow in the NULL case.
1925 Byte<replaceable>n</replaceable>
1929 The value of the parameter, in the format indicated by the
1930 associated format code.
1931 <replaceable>n</replaceable> is the above length.
1936 After the last parameter, the following fields appear:
1944 The number of result-column format codes that follow
1945 (denoted <replaceable>R</> below).
1946 This can be zero to indicate that there are no result columns
1947 or that the result columns should all use the default format
1949 or one, in which case the specified format code is applied
1950 to all result columns (if any); or it can equal the actual
1951 number of result columns of the query.
1957 Int16[<replaceable>R</>]
1961 The result-column format codes. Each must presently be
1962 zero (text) or one (binary).
1986 Identifies the message as a Bind-complete indicator.
1996 Length of message contents in bytes, including self.
2021 Length of message contents in bytes, including self.
2031 The cancel request code. The value is chosen to contain
2032 <literal>1234</> in the most significant 16 bits, and <literal>5678</> in the
2033 least 16 significant bits. (To avoid confusion, this code
2034 must not be the same as any protocol version number.)
2044 The process ID of the target backend.
2054 The secret key for the target backend.
2079 Identifies the message as a Close command.
2089 Length of message contents in bytes, including self.
2099 '<literal>S</>' to close a prepared statement; or
2100 '<literal>P</>' to close a portal.
2110 The name of the prepared statement or portal to close
2111 (an empty string selects the unnamed prepared statement
2136 Identifies the message as a Close-complete indicator.
2146 Length of message contents in bytes, including self.
2171 Identifies the message as a command-completed response.
2181 Length of message contents in bytes, including self.
2191 The command tag. This is usually a single
2192 word that identifies which SQL command was completed.
2196 For an <command>INSERT</command> command, the tag is
2197 <literal>INSERT <replaceable>oid</replaceable>
2198 <replaceable>rows</replaceable></literal>, where
2199 <replaceable>rows</replaceable> is the number of rows
2200 inserted. <replaceable>oid</replaceable> is the object ID
2201 of the inserted row if <replaceable>rows</replaceable> is 1
2202 and the target table has OIDs;
2203 otherwise <replaceable>oid</replaceable> is 0.
2207 For a <command>DELETE</command> command, the tag is
2208 <literal>DELETE <replaceable>rows</replaceable></literal> where
2209 <replaceable>rows</replaceable> is the number of rows deleted.
2213 For an <command>UPDATE</command> command, the tag is
2214 <literal>UPDATE <replaceable>rows</replaceable></literal> where
2215 <replaceable>rows</replaceable> is the number of rows updated.
2219 For a <command>SELECT</command> or <command>CREATE TABLE AS</command>
2220 command, the tag is <literal>SELECT <replaceable>rows</replaceable></literal>
2221 where <replaceable>rows</replaceable> is the number of rows retrieved.
2225 For a <command>MOVE</command> command, the tag is
2226 <literal>MOVE <replaceable>rows</replaceable></literal> where
2227 <replaceable>rows</replaceable> is the number of rows the
2228 cursor's position has been changed by.
2232 For a <command>FETCH</command> command, the tag is
2233 <literal>FETCH <replaceable>rows</replaceable></literal> where
2234 <replaceable>rows</replaceable> is the number of rows that
2235 have been retrieved from the cursor.
2239 For a <command>COPY</command> command, the tag is
2240 <literal>COPY <replaceable>rows</replaceable></literal> where
2241 <replaceable>rows</replaceable> is the number of rows copied.
2242 (Note: the row count appears only in
2243 <productname>PostgreSQL</productname> 8.2 and later.)
2257 CopyData (F & B)
2268 Identifies the message as <command>COPY</command> data.
2278 Length of message contents in bytes, including self.
2284 Byte<replaceable>n</replaceable>
2288 Data that forms part of a <command>COPY</command> data stream. Messages sent
2289 from the backend will always correspond to single data rows,
2290 but messages sent by frontends might divide the data stream
2303 CopyDone (F & B)
2315 Identifies the message as a <command>COPY</command>-complete indicator.
2325 Length of message contents in bytes, including self.
2350 Identifies the message as a <command>COPY</command>-failure indicator.
2360 Length of message contents in bytes, including self.
2370 An error message to report as the cause of failure.
2395 Identifies the message as a Start Copy In response.
2396 The frontend must now send copy-in data (if not
2397 prepared to do so, send a CopyFail message).
2407 Length of message contents in bytes, including self.
2417 0 indicates the overall <command>COPY</command> format is textual (rows
2418 separated by newlines, columns separated by separator
2420 1 indicates the overall copy format is binary (similar
2422 See <xref linkend="sql-copy" endterm="sql-copy-title">
2423 for more information.
2433 The number of columns in the data to be copied
2434 (denoted <replaceable>N</> below).
2440 Int16[<replaceable>N</>]
2444 The format codes to be used for each column.
2445 Each must presently be zero (text) or one (binary).
2446 All must be zero if the overall copy format is textual.
2471 Identifies the message as a Start Copy Out response.
2472 This message will be followed by copy-out data.
2482 Length of message contents in bytes, including self.
2492 0 indicates the overall <command>COPY</command> format
2493 is textual (rows separated by newlines, columns
2494 separated by separator characters, etc). 1 indicates
2495 the overall copy format is binary (similar to DataRow
2496 format). See <xref linkend="sql-copy"
2497 endterm="sql-copy-title"> for more information.
2507 The number of columns in the data to be copied
2508 (denoted <replaceable>N</> below).
2514 Int16[<replaceable>N</>]
2518 The format codes to be used for each column.
2519 Each must presently be zero (text) or one (binary).
2520 All must be zero if the overall copy format is textual.
2544 Identifies the message as a data row.
2554 Length of message contents in bytes, including self.
2564 The number of column values that follow (possibly zero).
2569 Next, the following pair of fields appear for each column:
2577 The length of the column value, in bytes (this count
2578 does not include itself). Can be zero.
2579 As a special case, -1 indicates a NULL column value.
2580 No value bytes follow in the NULL case.
2586 Byte<replaceable>n</replaceable>
2590 The value of the column, in the format indicated by the
2591 associated format code.
2592 <replaceable>n</replaceable> is the above length.
2617 Identifies the message as a Describe command.
2627 Length of message contents in bytes, including self.
2637 '<literal>S</>' to describe a prepared statement; or
2638 '<literal>P</>' to describe a portal.
2648 The name of the prepared statement or portal to describe
2649 (an empty string selects the unnamed prepared statement
2662 EmptyQueryResponse (B)
2674 Identifies the message as a response to an empty query string.
2675 (This substitutes for CommandComplete.)
2685 Length of message contents in bytes, including self.
2710 Identifies the message as an error.
2720 Length of message contents in bytes, including self.
2725 The message body consists of one or more identified fields,
2726 followed by a zero byte as a terminator. Fields can appear in
2727 any order. For each field there is the following:
2735 A code identifying the field type; if zero, this is
2736 the message terminator and no string follows.
2737 The presently defined field types are listed in
2738 <xref linkend="protocol-error-fields">.
2739 Since more field types might be added in future,
2740 frontends should silently ignore fields of unrecognized
2776 Identifies the message as an Execute command.
2786 Length of message contents in bytes, including self.
2796 The name of the portal to execute
2797 (an empty string selects the unnamed portal).
2807 Maximum number of rows to return, if portal contains
2808 a query that returns rows (ignored otherwise). Zero
2809 denotes <quote>no limit</>.
2833 Identifies the message as a Flush command.
2843 Length of message contents in bytes, including self.
2868 Identifies the message as a function call.
2878 Length of message contents in bytes, including self.
2888 Specifies the object ID of the function to call.
2898 The number of argument format codes that follow
2899 (denoted <replaceable>C</> below).
2900 This can be zero to indicate that there are no arguments
2901 or that the arguments all use the default format (text);
2902 or one, in which case the specified format code is applied
2903 to all arguments; or it can equal the actual number of
2910 Int16[<replaceable>C</>]
2914 The argument format codes. Each must presently be
2915 zero (text) or one (binary).
2925 Specifies the number of arguments being supplied to the
2931 Next, the following pair of fields appear for each argument:
2939 The length of the argument value, in bytes (this count
2940 does not include itself). Can be zero.
2941 As a special case, -1 indicates a NULL argument value.
2942 No value bytes follow in the NULL case.
2948 Byte<replaceable>n</replaceable>
2952 The value of the argument, in the format indicated by the
2953 associated format code.
2954 <replaceable>n</replaceable> is the above length.
2959 After the last argument, the following field appears:
2967 The format code for the function result. Must presently be
2968 zero (text) or one (binary).
2981 FunctionCallResponse (B)
2993 Identifies the message as a function call result.
3003 Length of message contents in bytes, including self.
3013 The length of the function result value, in bytes (this count
3014 does not include itself). Can be zero.
3015 As a special case, -1 indicates a NULL function result.
3016 No value bytes follow in the NULL case.
3022 Byte<replaceable>n</replaceable>
3026 The value of the function result, in the format indicated by
3027 the associated format code.
3028 <replaceable>n</replaceable> is the above length.
3053 Identifies the message as a no-data indicator.
3063 Length of message contents in bytes, including self.
3088 Identifies the message as a notice.
3098 Length of message contents in bytes, including self.
3103 The message body consists of one or more identified fields,
3104 followed by a zero byte as a terminator. Fields can appear in
3105 any order. For each field there is the following:
3113 A code identifying the field type; if zero, this is
3114 the message terminator and no string follows.
3115 The presently defined field types are listed in
3116 <xref linkend="protocol-error-fields">.
3117 Since more field types might be added in future,
3118 frontends should silently ignore fields of unrecognized
3142 NotificationResponse (B)
3154 Identifies the message as a notification response.
3164 Length of message contents in bytes, including self.
3174 The process ID of the notifying backend process.
3184 The name of the channel that the notify has been raised on.
3194 The <quote>payload</> string passed from the notifying process.
3207 ParameterDescription (B)
3219 Identifies the message as a parameter description.
3229 Length of message contents in bytes, including self.
3239 The number of parameters used by the statement
3245 Then, for each parameter, there is the following:
3253 Specifies the object ID of the parameter data type.
3277 Identifies the message as a run-time parameter status report.
3287 Length of message contents in bytes, including self.
3297 The name of the run-time parameter being reported.
3307 The current value of the parameter.
3331 Identifies the message as a Parse command.
3341 Length of message contents in bytes, including self.
3351 The name of the destination prepared statement
3352 (an empty string selects the unnamed prepared statement).
3362 The query string to be parsed.
3372 The number of parameter data types specified
3373 (can be zero). Note that this is not an indication of
3374 the number of parameters that might appear in the
3375 query string, only the number that the frontend wants to
3376 prespecify types for.
3381 Then, for each parameter, there is the following:
3389 Specifies the object ID of the parameter data type.
3390 Placing a zero here is equivalent to leaving the type
3415 Identifies the message as a Parse-complete indicator.
3425 Length of message contents in bytes, including self.
3450 Identifies the message as a password response. Note that
3451 this is also used for GSSAPI and SSPI response messages
3452 (which is really a design error, since the contained data
3453 is not a null-terminated string in that case, but can be
3454 arbitrary binary data).
3464 Length of message contents in bytes, including self.
3474 The password (encrypted, if requested).
3498 Identifies the message as a portal-suspended indicator.
3499 Note this only appears if an Execute message's row-count limit
3510 Length of message contents in bytes, including self.
3535 Identifies the message as a simple query.
3545 Length of message contents in bytes, including self.
3555 The query string itself.
3580 Identifies the message type. ReadyForQuery is sent
3581 whenever the backend is ready for a new query cycle.
3591 Length of message contents in bytes, including self.
3601 Current backend transaction status indicator.
3602 Possible values are '<literal>I</>' if idle (not in
3603 a transaction block); '<literal>T</>' if in a transaction
3604 block; or '<literal>E</>' if in a failed transaction
3605 block (queries will be rejected until block is ended).
3630 Identifies the message as a row description.
3640 Length of message contents in bytes, including self.
3650 Specifies the number of fields in a row (can be zero).
3655 Then, for each field, there is the following:
3673 If the field can be identified as a column of a specific
3674 table, the object ID of the table; otherwise zero.
3684 If the field can be identified as a column of a specific
3685 table, the attribute number of the column; otherwise zero.
3695 The object ID of the field's data type.
3705 The data type size (see <varname>pg_type.typlen</>).
3706 Note that negative values denote variable-width types.
3716 The type modifier (see <varname>pg_attribute.atttypmod</>).
3717 The meaning of the modifier is type-specific.
3727 The format code being used for the field. Currently will
3728 be zero (text) or one (binary). In a RowDescription
3729 returned from the statement variant of Describe, the
3730 format code is not yet known and will always be zero.
3755 Length of message contents in bytes, including self.
3765 The <acronym>SSL</acronym> request code. The value is chosen to contain
3766 <literal>1234</> in the most significant 16 bits, and <literal>5679</> in the
3767 least 16 significant bits. (To avoid confusion, this code
3768 must not be the same as any protocol version number.)
3793 Length of message contents in bytes, including self.
3803 The protocol version number. The most significant 16 bits are
3804 the major version number (3 for the protocol described here).
3805 The least significant 16 bits are the minor version number
3806 (0 for the protocol described here).
3811 The protocol version number is followed by one or more pairs of
3812 parameter name and value strings. A zero byte is required as a
3813 terminator after the last name/value pair.
3814 Parameters can appear in any
3815 order. <literal>user</> is required, others are optional.
3816 Each parameter is specified as:
3824 The parameter name. Currently recognized names are:
3833 The database user name to connect as. Required;
3834 there is no default.
3840 <literal>database</>
3844 The database to connect to. Defaults to the user name.
3854 Command-line arguments for the backend. (This is
3855 deprecated in favor of setting individual run-time
3862 In addition to the above, any run-time parameter that can be
3863 set at backend start time might be listed. Such settings
3864 will be applied during backend start (after parsing the
3865 command-line options if any). The values will act as
3876 The parameter value.
3901 Identifies the message as a Sync command.
3911 Length of message contents in bytes, including self.
3936 Identifies the message as a termination.
3946 Length of message contents in bytes, including self.
3962 <sect1 id="protocol-error-fields">
3963 <title>Error and Notice Message Fields</title>
3966 This section describes the fields that can appear in ErrorResponse and
3967 NoticeResponse messages. Each field type has a single-byte identification
3968 token. Note that any given field type should appear at most once per
3980 Severity: the field contents are
3981 <literal>ERROR</>, <literal>FATAL</>, or
3982 <literal>PANIC</> (in an error message), or
3983 <literal>WARNING</>, <literal>NOTICE</>, <literal>DEBUG</>,
3984 <literal>INFO</>, or <literal>LOG</> (in a notice message),
3985 or a localized translation of one of these. Always present.
3996 Code: the SQLSTATE code for the error (see <xref
3997 linkend="errcodes-appendix">). Not localizable. Always present.
4008 Message: the primary human-readable error message.
4009 This should be accurate but terse (typically one line).
4021 Detail: an optional secondary error message carrying more
4022 detail about the problem. Might run to multiple lines.
4033 Hint: an optional suggestion what to do about the problem.
4034 This is intended to differ from Detail in that it offers advice
4035 (potentially inappropriate) rather than hard facts.
4036 Might run to multiple lines.
4047 Position: the field value is a decimal ASCII integer, indicating
4048 an error cursor position as an index into the original query string.
4049 The first character has index 1, and positions are measured in
4050 characters not bytes.
4061 Internal position: this is defined the same as the <literal>P</>
4062 field, but it is used when the cursor position refers to an internally
4063 generated command rather than the one submitted by the client.
4064 The <literal>q</> field will always appear when this field appears.
4075 Internal query: the text of a failed internally-generated command.
4076 This could be, for example, a SQL query issued by a PL/pgSQL function.
4087 Where: an indication of the context in which the error occurred.
4088 Presently this includes a call stack traceback of active
4089 procedural language functions and internally-generated queries.
4090 The trace is one entry per line, most recent first.
4101 File: the file name of the source-code location where the error
4113 Line: the line number of the source-code location where the error
4125 Routine: the name of the source-code routine reporting the error.
4133 The client is responsible for formatting displayed information to meet its
4134 needs; in particular it should break long lines as needed. Newline characters
4135 appearing in the error message fields should be treated as paragraph breaks,
4141 <sect1 id="protocol-replication">
4142 <title>Streaming Replication Protocol</title>
4145 To initiate streaming replication, the frontend sends the "replication"
4146 parameter in the startup message. This tells the backend to go into
4147 walsender mode, where a small set of replication commands can be issued
4148 instead of SQL statements. Only the simple query protocol can be used in
4151 The commands accepted in walsender mode are:
4155 <term>IDENTIFY_SYSTEM</term>
4158 Requests the server to idenfity itself. Server replies with a result
4159 set of a single row, and two fields:
4161 systemid: The unique system identifier identifying the cluster. This
4162 can be used to check that the base backup used to initialize the
4163 slave came from the same cluster.
4165 timeline: Current TimelineID. Also used to check that the slave is
4166 consistent with the master.
4172 <term>START_REPLICATION XXX/XXX</term>
4175 Instructs backend to start streaming WAL, starting at point XXX/XXX.
4176 Server can reply with an error e.g if the requested piece of WAL has
4177 already been recycled. On success, server responds with a
4178 CopyOutResponse message, and backend starts to stream WAL as CopyData
4180 The payload in CopyData message consists of the following format.
4198 Identifies the message as WAL data.
4208 The log file number of the LSN, indicating the starting point of
4209 the WAL in the message.
4219 The byte offset of the LSN, indicating the starting point of
4220 the WAL in the message.
4226 Byte<replaceable>n</replaceable>
4230 Data that forms part of WAL data stream.
4241 A single WAL record is never split across two CopyData messages. When
4242 a WAL record crosses a WAL page boundary, however, and is therefore
4243 already split using continuation records, it can be split at the page
4244 boundary. In other words, the first main WAL record and its
4245 continuation records can be split across different CopyData messages.
4255 <sect1 id="protocol-changes">
4256 <title>Summary of Changes since Protocol 2.0</title>
4259 This section provides a quick checklist of changes, for the benefit of
4260 developers trying to update existing client libraries to protocol 3.0.
4264 The initial startup packet uses a flexible list-of-strings format
4265 instead of a fixed format. Notice that session default values for run-time
4266 parameters can now be specified directly in the startup packet. (Actually,
4267 you could do that before using the <literal>options</> field, but given the
4268 limited width of <literal>options</> and the lack of any way to quote
4269 whitespace in the values, it wasn't a very safe technique.)
4273 All messages now have a length count immediately following the message type
4274 byte (except for startup packets, which have no type byte). Also note that
4275 PasswordMessage now has a type byte.
4279 ErrorResponse and NoticeResponse ('<literal>E</>' and '<literal>N</>')
4280 messages now contain multiple fields, from which the client code can
4281 assemble an error message of the desired level of verbosity. Note that
4282 individual fields will typically not end with a newline, whereas the single
4283 string sent in the older protocol always did.
4287 The ReadyForQuery ('<literal>Z</>') message includes a transaction status
4292 The distinction between BinaryRow and DataRow message types is gone; the
4293 single DataRow message type serves for returning data in all formats.
4294 Note that the layout of DataRow has changed to make it easier to parse.
4295 Also, the representation of binary values has changed: it is no longer
4296 directly tied to the server's internal representation.
4300 There is a new <quote>extended query</> sub-protocol, which adds the frontend
4301 message types Parse, Bind, Execute, Describe, Close, Flush, and Sync, and the
4302 backend message types ParseComplete, BindComplete, PortalSuspended,
4303 ParameterDescription, NoData, and CloseComplete. Existing clients do not
4304 have to concern themselves with this sub-protocol, but making use of it
4305 might allow improvements in performance or functionality.
4309 <command>COPY</command> data is now encapsulated into CopyData and CopyDone messages. There
4310 is a well-defined way to recover from errors during <command>COPY</command>. The special
4311 <quote><literal>\.</></quote> last line is not needed anymore, and is not sent
4312 during <command>COPY OUT</command>.
4313 (It is still recognized as a terminator during <command>COPY IN</command>, but its use is
4314 deprecated and will eventually be removed.) Binary <command>COPY</command> is supported.
4315 The CopyInResponse and CopyOutResponse messages include fields indicating
4316 the number of columns and the format of each column.
4320 The layout of FunctionCall and FunctionCallResponse messages has changed.
4321 FunctionCall can now support passing NULL arguments to functions. It also
4322 can handle passing parameters and retrieving results in either text or
4323 binary format. There is no longer any reason to consider FunctionCall a
4324 potential security hole, since it does not offer direct access to internal
4325 server data representations.
4329 The backend sends ParameterStatus ('<literal>S</>') messages during connection
4330 startup for all parameters it considers interesting to the client library.
4331 Subsequently, a ParameterStatus message is sent whenever the active value
4332 changes for any of these parameters.
4336 The RowDescription ('<literal>T</>') message carries new table OID and column
4337 number fields for each column of the described row. It also shows the format
4338 code for each column.
4342 The CursorResponse ('<literal>P</>') message is no longer generated by
4347 The NotificationResponse ('<literal>A</>') message has an additional string
4348 field, which can carry a <quote>payload</> string passed
4349 from the <command>NOTIFY</command> event sender.
4353 The EmptyQueryResponse ('<literal>I</>') message used to include an empty
4354 string parameter; this has been removed.