4 32.5.1. Using Pipeline Mode
5 32.5.2. Functions Associated with Pipeline Mode
6 32.5.3. When to Use Pipeline Mode
8 libpq pipeline mode allows applications to send a query without having
9 to read the result of the previously sent query. Taking advantage of
10 the pipeline mode, a client will wait less for the server, since
11 multiple queries/results can be sent/received in a single network
14 While pipeline mode provides a significant performance boost, writing
15 clients using the pipeline mode is more complex because it involves
16 managing a queue of pending queries and finding which result
17 corresponds to which query in the queue.
19 Pipeline mode also generally consumes more memory on both the client
20 and server, though careful and aggressive management of the
21 send/receive queue can mitigate this. This applies whether or not the
22 connection is in blocking or non-blocking mode.
24 While libpq's pipeline API was introduced in PostgreSQL 14, it is a
25 client-side feature which doesn't require special server support and
26 works on any server that supports the v3 extended query protocol. For
27 more information see Section 54.2.4.
29 32.5.1. Using Pipeline Mode #
31 To issue pipelines, the application must switch the connection into
32 pipeline mode, which is done with PQenterPipelineMode. PQpipelineStatus
33 can be used to test whether pipeline mode is active. In pipeline mode,
34 only asynchronous operations that utilize the extended query protocol
35 are permitted, command strings containing multiple SQL commands are
36 disallowed, and so is COPY. Using synchronous command execution
37 functions such as PQfn, PQexec, PQexecParams, PQprepare,
38 PQexecPrepared, PQdescribePrepared, PQdescribePortal, PQclosePrepared,
39 PQclosePortal, is an error condition. PQsendQuery is also disallowed,
40 because it uses the simple query protocol. Once all dispatched commands
41 have had their results processed, and the end pipeline result has been
42 consumed, the application may return to non-pipelined mode with
47 It is best to use pipeline mode with libpq in non-blocking mode. If
48 used in blocking mode it is possible for a client/server deadlock to
51 32.5.1.1. Issuing Queries #
53 After entering pipeline mode, the application dispatches requests using
54 PQsendQueryParams or its prepared-query sibling PQsendQueryPrepared.
55 These requests are queued on the client-side until flushed to the
56 server; this occurs when PQpipelineSync is used to establish a
57 synchronization point in the pipeline, or when PQflush is called. The
58 functions PQsendPrepare, PQsendDescribePrepared, PQsendDescribePortal,
59 PQsendClosePrepared, and PQsendClosePortal also work in pipeline mode.
60 Result processing is described below.
62 The server executes statements, and returns results, in the order the
63 client sends them. The server will begin executing the commands in the
64 pipeline immediately, not waiting for the end of the pipeline. Note
65 that results are buffered on the server side; the server flushes that
66 buffer when a synchronization point is established with either
67 PQpipelineSync or PQsendPipelineSync, or when PQsendFlushRequest is
68 called. If any statement encounters an error, the server aborts the
69 current transaction and does not execute any subsequent command in the
70 queue until the next synchronization point; a PGRES_PIPELINE_ABORTED
71 result is produced for each such command. (This remains true even if
72 the commands in the pipeline would rollback the transaction.) Query
73 processing resumes after the synchronization point.
75 It's fine for one operation to depend on the results of a prior one;
76 for example, one query may define a table that the next query in the
77 same pipeline uses. Similarly, an application may create a named
78 prepared statement and execute it with later statements in the same
81 32.5.1.2. Processing Results #
83 To process the result of one query in a pipeline, the application calls
84 PQgetResult repeatedly and handles each result until PQgetResult
85 returns null. The result from the next query in the pipeline may then
86 be retrieved using PQgetResult again and the cycle repeated. The
87 application handles individual statement results as normal. When the
88 results of all the queries in the pipeline have been returned,
89 PQgetResult returns a result containing the status value
92 The client may choose to defer result processing until the complete
93 pipeline has been sent, or interleave that with sending further queries
94 in the pipeline; see Section 32.5.1.4.
96 PQgetResult behaves the same as for normal asynchronous processing
97 except that it may contain the new PGresult types PGRES_PIPELINE_SYNC
98 and PGRES_PIPELINE_ABORTED. PGRES_PIPELINE_SYNC is reported exactly
99 once for each PQpipelineSync or PQsendPipelineSync at the corresponding
100 point in the pipeline. PGRES_PIPELINE_ABORTED is emitted in place of a
101 normal query result for the first error and all subsequent results
102 until the next PGRES_PIPELINE_SYNC; see Section 32.5.1.3.
104 PQisBusy, PQconsumeInput, etc operate as normal when processing
105 pipeline results. In particular, a call to PQisBusy in the middle of a
106 pipeline returns 0 if the results for all the queries issued so far
109 libpq does not provide any information to the application about the
110 query currently being processed (except that PQgetResult returns null
111 to indicate that we start returning the results of next query). The
112 application must keep track of the order in which it sent queries, to
113 associate them with their corresponding results. Applications will
114 typically use a state machine or a FIFO queue for this.
116 32.5.1.3. Error Handling #
118 From the client's perspective, after PQresultStatus returns
119 PGRES_FATAL_ERROR, the pipeline is flagged as aborted. PQresultStatus
120 will report a PGRES_PIPELINE_ABORTED result for each remaining queued
121 operation in an aborted pipeline. The result for PQpipelineSync or
122 PQsendPipelineSync is reported as PGRES_PIPELINE_SYNC to signal the end
123 of the aborted pipeline and resumption of normal result processing.
125 The client must process results with PQgetResult during error recovery.
127 If the pipeline used an implicit transaction, then operations that have
128 already executed are rolled back and operations that were queued to
129 follow the failed operation are skipped entirely. The same behavior
130 holds if the pipeline starts and commits a single explicit transaction
131 (i.e. the first statement is BEGIN and the last is COMMIT) except that
132 the session remains in an aborted transaction state at the end of the
133 pipeline. If a pipeline contains multiple explicit transactions, all
134 transactions that committed prior to the error remain committed, the
135 currently in-progress transaction is aborted, and all subsequent
136 operations are skipped completely, including subsequent transactions.
137 If a pipeline synchronization point occurs with an explicit transaction
138 block in aborted state, the next pipeline will become aborted
139 immediately unless the next command puts the transaction in normal mode
144 The client must not assume that work is committed when it sends a
145 COMMIT — only when the corresponding result is received to confirm the
146 commit is complete. Because errors arrive asynchronously, the
147 application needs to be able to restart from the last received
148 committed change and resend work done after that point if something
151 32.5.1.4. Interleaving Result Processing and Query Dispatch #
153 To avoid deadlocks on large pipelines the client should be structured
154 around a non-blocking event loop using operating system facilities such
155 as select, poll, WaitForMultipleObjectEx, etc.
157 The client application should generally maintain a queue of work
158 remaining to be dispatched and a queue of work that has been dispatched
159 but not yet had its results processed. When the socket is writable it
160 should dispatch more work. When the socket is readable it should read
161 results and process them, matching them up to the next entry in its
162 corresponding results queue. Based on available memory, results from
163 the socket should be read frequently: there's no need to wait until the
164 pipeline end to read the results. Pipelines should be scoped to logical
165 units of work, usually (but not necessarily) one transaction per
166 pipeline. There's no need to exit pipeline mode and re-enter it between
167 pipelines, or to wait for one pipeline to finish before sending the
170 An example using select() and a simple state machine to track sent and
171 received work is in src/test/modules/libpq_pipeline/libpq_pipeline.c in
172 the PostgreSQL source distribution.
174 32.5.2. Functions Associated with Pipeline Mode #
177 Returns the current pipeline mode status of the libpq
180 PGpipelineStatus PQpipelineStatus(const PGconn *conn);
182 PQpipelineStatus can return one of the following values:
185 The libpq connection is in pipeline mode.
188 The libpq connection is not in pipeline mode.
191 The libpq connection is in pipeline mode and an error
192 occurred while processing the current pipeline. The
193 aborted flag is cleared when PQgetResult returns a result
194 of type PGRES_PIPELINE_SYNC.
196 PQenterPipelineMode #
197 Causes a connection to enter pipeline mode if it is currently
198 idle or already in pipeline mode.
200 int PQenterPipelineMode(PGconn *conn);
202 Returns 1 for success. Returns 0 and has no effect if the
203 connection is not currently idle, i.e., it has a result ready,
204 or it is waiting for more input from the server, etc. This
205 function does not actually send anything to the server, it just
206 changes the libpq connection state.
209 Causes a connection to exit pipeline mode if it is currently in
210 pipeline mode with an empty queue and no pending results.
212 int PQexitPipelineMode(PGconn *conn);
214 Returns 1 for success. Returns 1 and takes no action if not in
215 pipeline mode. If the current statement isn't finished
216 processing, or PQgetResult has not been called to collect
217 results from all previously sent query, returns 0 (in which
218 case, use PQerrorMessage to get more information about the
222 Marks a synchronization point in a pipeline by sending a sync
223 message and flushing the send buffer. This serves as the
224 delimiter of an implicit transaction and an error recovery
225 point; see Section 32.5.1.3.
227 int PQpipelineSync(PGconn *conn);
229 Returns 1 for success. Returns 0 if the connection is not in
230 pipeline mode or sending a sync message failed.
233 Marks a synchronization point in a pipeline by sending a sync
234 message without flushing the send buffer. This serves as the
235 delimiter of an implicit transaction and an error recovery
236 point; see Section 32.5.1.3.
238 int PQsendPipelineSync(PGconn *conn);
240 Returns 1 for success. Returns 0 if the connection is not in
241 pipeline mode or sending a sync message failed. Note that the
242 message is not itself flushed to the server automatically; use
243 PQflush if necessary.
246 Sends a request for the server to flush its output buffer.
248 int PQsendFlushRequest(PGconn *conn);
250 Returns 1 for success. Returns 0 on any failure.
252 The server flushes its output buffer automatically as a result
253 of PQpipelineSync being called, or on any request when not in
254 pipeline mode; this function is useful to cause the server to
255 flush its output buffer in pipeline mode without establishing a
256 synchronization point. Note that the request is not itself
257 flushed to the server automatically; use PQflush if necessary.
259 32.5.3. When to Use Pipeline Mode #
261 Much like asynchronous query mode, there is no meaningful performance
262 overhead when using pipeline mode. It increases client application
263 complexity, and extra caution is required to prevent client/server
264 deadlocks, but pipeline mode can offer considerable performance
265 improvements, in exchange for increased memory usage from leaving state
268 Pipeline mode is most useful when the server is distant, i.e., network
269 latency (“ping time”) is high, and also when many small operations are
270 being performed in rapid succession. There is usually less benefit in
271 using pipelined commands when each query takes many multiples of the
272 client/server round-trip time to execute. A 100-statement operation run
273 on a server 300 ms round-trip-time away would take 30 seconds in
274 network latency alone without pipelining; with pipelining it may spend
275 as little as 0.3 s waiting for results from the server.
277 Use pipelined commands when your application does lots of small INSERT,
278 UPDATE and DELETE operations that can't easily be transformed into
279 operations on sets, or into a COPY operation.
281 Pipeline mode is not useful when information from one operation is
282 required by the client to produce the next operation. In such cases,
283 the client would have to introduce a synchronization point and wait for
284 a full client/server round-trip to get the results it needs. However,
285 it's often possible to adjust the client design to exchange the
286 required information server-side. Read-modify-write cycles are
287 especially good candidates; for example:
289 SELECT x FROM mytable WHERE id = 42 FOR UPDATE;
291 -- client adds 1 to x:
292 UPDATE mytable SET x = 3 WHERE id = 42;
295 could be much more efficiently done with:
296 UPDATE mytable SET x = x + 1 WHERE id = 42;
298 Pipelining is less useful, and more complex, when a single pipeline
299 contains multiple transactions (see Section 32.5.1.3).
301 ^[15] The client will block trying to send queries to the server, but
302 the server will block trying to send results to the client from queries
303 it has already processed. This only occurs when the client sends enough
304 queries to fill both its output buffer and the server's receive buffer
305 before it switches to processing input from the server, but it's hard
306 to predict exactly when that will happen.