1CREATE AGGREGATE(7) PostgreSQL 12.6 Documentation CREATE AGGREGATE(7)
2
3
4
6 CREATE_AGGREGATE - define a new aggregate function
7
9 CREATE [ OR REPLACE ] AGGREGATE name ( [ argmode ] [ argname ] arg_data_type [ , ... ] ) (
10 SFUNC = sfunc,
11 STYPE = state_data_type
12 [ , SSPACE = state_data_size ]
13 [ , FINALFUNC = ffunc ]
14 [ , FINALFUNC_EXTRA ]
15 [ , FINALFUNC_MODIFY = { READ_ONLY | SHAREABLE | READ_WRITE } ]
16 [ , COMBINEFUNC = combinefunc ]
17 [ , SERIALFUNC = serialfunc ]
18 [ , DESERIALFUNC = deserialfunc ]
19 [ , INITCOND = initial_condition ]
20 [ , MSFUNC = msfunc ]
21 [ , MINVFUNC = minvfunc ]
22 [ , MSTYPE = mstate_data_type ]
23 [ , MSSPACE = mstate_data_size ]
24 [ , MFINALFUNC = mffunc ]
25 [ , MFINALFUNC_EXTRA ]
26 [ , MFINALFUNC_MODIFY = { READ_ONLY | SHAREABLE | READ_WRITE } ]
27 [ , MINITCOND = minitial_condition ]
28 [ , SORTOP = sort_operator ]
29 [ , PARALLEL = { SAFE | RESTRICTED | UNSAFE } ]
30 )
31
32 CREATE [ OR REPLACE ] AGGREGATE name ( [ [ argmode ] [ argname ] arg_data_type [ , ... ] ]
33 ORDER BY [ argmode ] [ argname ] arg_data_type [ , ... ] ) (
34 SFUNC = sfunc,
35 STYPE = state_data_type
36 [ , SSPACE = state_data_size ]
37 [ , FINALFUNC = ffunc ]
38 [ , FINALFUNC_EXTRA ]
39 [ , FINALFUNC_MODIFY = { READ_ONLY | SHAREABLE | READ_WRITE } ]
40 [ , INITCOND = initial_condition ]
41 [ , PARALLEL = { SAFE | RESTRICTED | UNSAFE } ]
42 [ , HYPOTHETICAL ]
43 )
44
45 or the old syntax
46
47 CREATE [ OR REPLACE ] AGGREGATE name (
48 BASETYPE = base_type,
49 SFUNC = sfunc,
50 STYPE = state_data_type
51 [ , SSPACE = state_data_size ]
52 [ , FINALFUNC = ffunc ]
53 [ , FINALFUNC_EXTRA ]
54 [ , FINALFUNC_MODIFY = { READ_ONLY | SHAREABLE | READ_WRITE } ]
55 [ , COMBINEFUNC = combinefunc ]
56 [ , SERIALFUNC = serialfunc ]
57 [ , DESERIALFUNC = deserialfunc ]
58 [ , INITCOND = initial_condition ]
59 [ , MSFUNC = msfunc ]
60 [ , MINVFUNC = minvfunc ]
61 [ , MSTYPE = mstate_data_type ]
62 [ , MSSPACE = mstate_data_size ]
63 [ , MFINALFUNC = mffunc ]
64 [ , MFINALFUNC_EXTRA ]
65 [ , MFINALFUNC_MODIFY = { READ_ONLY | SHAREABLE | READ_WRITE } ]
66 [ , MINITCOND = minitial_condition ]
67 [ , SORTOP = sort_operator ]
68 )
69
71 CREATE AGGREGATE defines a new aggregate function. CREATE OR REPLACE
72 AGGREGATE will either define a new aggregate function or replace an
73 existing definition. Some basic and commonly-used aggregate functions
74 are included with the distribution; they are documented in
75 Section 9.20. If one defines new types or needs an aggregate function
76 not already provided, then CREATE AGGREGATE can be used to provide the
77 desired features.
78
79 When replacing an existing definition, the argument types, result type,
80 and number of direct arguments may not be changed. Also, the new
81 definition must be of the same kind (ordinary aggregate, ordered-set
82 aggregate, or hypothetical-set aggregate) as the old one.
83
84 If a schema name is given (for example, CREATE AGGREGATE myschema.myagg
85 ...) then the aggregate function is created in the specified schema.
86 Otherwise it is created in the current schema.
87
88 An aggregate function is identified by its name and input data type(s).
89 Two aggregates in the same schema can have the same name if they
90 operate on different input types. The name and input data type(s) of an
91 aggregate must also be distinct from the name and input data type(s) of
92 every ordinary function in the same schema. This behavior is identical
93 to overloading of ordinary function names (see CREATE FUNCTION
94 (CREATE_FUNCTION(7))).
95
96 A simple aggregate function is made from one or two ordinary functions:
97 a state transition function sfunc, and an optional final calculation
98 function ffunc. These are used as follows:
99
100 sfunc( internal-state, next-data-values ) ---> next-internal-state
101 ffunc( internal-state ) ---> aggregate-value
102
103 PostgreSQL creates a temporary variable of data type stype to hold the
104 current internal state of the aggregate. At each input row, the
105 aggregate argument value(s) are calculated and the state transition
106 function is invoked with the current state value and the new argument
107 value(s) to calculate a new internal state value. After all the rows
108 have been processed, the final function is invoked once to calculate
109 the aggregate's return value. If there is no final function then the
110 ending state value is returned as-is.
111
112 An aggregate function can provide an initial condition, that is, an
113 initial value for the internal state value. This is specified and
114 stored in the database as a value of type text, but it must be a valid
115 external representation of a constant of the state value data type. If
116 it is not supplied then the state value starts out null.
117
118 If the state transition function is declared “strict”, then it cannot
119 be called with null inputs. With such a transition function, aggregate
120 execution behaves as follows. Rows with any null input values are
121 ignored (the function is not called and the previous state value is
122 retained). If the initial state value is null, then at the first row
123 with all-nonnull input values, the first argument value replaces the
124 state value, and the transition function is invoked at each subsequent
125 row with all-nonnull input values. This is handy for implementing
126 aggregates like max. Note that this behavior is only available when
127 state_data_type is the same as the first arg_data_type. When these
128 types are different, you must supply a nonnull initial condition or use
129 a nonstrict transition function.
130
131 If the state transition function is not strict, then it will be called
132 unconditionally at each input row, and must deal with null inputs and
133 null state values for itself. This allows the aggregate author to have
134 full control over the aggregate's handling of null values.
135
136 If the final function is declared “strict”, then it will not be called
137 when the ending state value is null; instead a null result will be
138 returned automatically. (Of course this is just the normal behavior of
139 strict functions.) In any case the final function has the option of
140 returning a null value. For example, the final function for avg returns
141 null when it sees there were zero input rows.
142
143 Sometimes it is useful to declare the final function as taking not just
144 the state value, but extra parameters corresponding to the aggregate's
145 input values. The main reason for doing this is if the final function
146 is polymorphic and the state value's data type would be inadequate to
147 pin down the result type. These extra parameters are always passed as
148 NULL (and so the final function must not be strict when the
149 FINALFUNC_EXTRA option is used), but nonetheless they are valid
150 parameters. The final function could for example make use of
151 get_fn_expr_argtype to identify the actual argument type in the current
152 call.
153
154 An aggregate can optionally support moving-aggregate mode, as described
155 in Section 37.12.1. This requires specifying the MSFUNC, MINVFUNC, and
156 MSTYPE parameters, and optionally the MSSPACE, MFINALFUNC,
157 MFINALFUNC_EXTRA, MFINALFUNC_MODIFY, and MINITCOND parameters. Except
158 for MINVFUNC, these parameters work like the corresponding
159 simple-aggregate parameters without M; they define a separate
160 implementation of the aggregate that includes an inverse transition
161 function.
162
163 The syntax with ORDER BY in the parameter list creates a special type
164 of aggregate called an ordered-set aggregate; or if HYPOTHETICAL is
165 specified, then a hypothetical-set aggregate is created. These
166 aggregates operate over groups of sorted values in order-dependent
167 ways, so that specification of an input sort order is an essential part
168 of a call. Also, they can have direct arguments, which are arguments
169 that are evaluated only once per aggregation rather than once per input
170 row. Hypothetical-set aggregates are a subclass of ordered-set
171 aggregates in which some of the direct arguments are required to match,
172 in number and data types, the aggregated argument columns. This allows
173 the values of those direct arguments to be added to the collection of
174 aggregate-input rows as an additional “hypothetical” row.
175
176 An aggregate can optionally support partial aggregation, as described
177 in Section 37.12.4. This requires specifying the COMBINEFUNC parameter.
178 If the state_data_type is internal, it's usually also appropriate to
179 provide the SERIALFUNC and DESERIALFUNC parameters so that parallel
180 aggregation is possible. Note that the aggregate must also be marked
181 PARALLEL SAFE to enable parallel aggregation.
182
183 Aggregates that behave like MIN or MAX can sometimes be optimized by
184 looking into an index instead of scanning every input row. If this
185 aggregate can be so optimized, indicate it by specifying a sort
186 operator. The basic requirement is that the aggregate must yield the
187 first element in the sort ordering induced by the operator; in other
188 words:
189
190 SELECT agg(col) FROM tab;
191
192 must be equivalent to:
193
194 SELECT col FROM tab ORDER BY col USING sortop LIMIT 1;
195
196 Further assumptions are that the aggregate ignores null inputs, and
197 that it delivers a null result if and only if there were no non-null
198 inputs. Ordinarily, a data type's < operator is the proper sort
199 operator for MIN, and > is the proper sort operator for MAX. Note that
200 the optimization will never actually take effect unless the specified
201 operator is the “less than” or “greater than” strategy member of a
202 B-tree index operator class.
203
204 To be able to create an aggregate function, you must have USAGE
205 privilege on the argument types, the state type(s), and the return
206 type, as well as EXECUTE privilege on the supporting functions.
207
209 name
210 The name (optionally schema-qualified) of the aggregate function to
211 create.
212
213 argmode
214 The mode of an argument: IN or VARIADIC. (Aggregate functions do
215 not support OUT arguments.) If omitted, the default is IN. Only the
216 last argument can be marked VARIADIC.
217
218 argname
219 The name of an argument. This is currently only useful for
220 documentation purposes. If omitted, the argument has no name.
221
222 arg_data_type
223 An input data type on which this aggregate function operates. To
224 create a zero-argument aggregate function, write * in place of the
225 list of argument specifications. (An example of such an aggregate
226 is count(*).)
227
228 base_type
229 In the old syntax for CREATE AGGREGATE, the input data type is
230 specified by a basetype parameter rather than being written next to
231 the aggregate name. Note that this syntax allows only one input
232 parameter. To define a zero-argument aggregate function with this
233 syntax, specify the basetype as "ANY" (not *). Ordered-set
234 aggregates cannot be defined with the old syntax.
235
236 sfunc
237 The name of the state transition function to be called for each
238 input row. For a normal N-argument aggregate function, the sfunc
239 must take N+1 arguments, the first being of type state_data_type
240 and the rest matching the declared input data type(s) of the
241 aggregate. The function must return a value of type
242 state_data_type. This function takes the current state value and
243 the current input data value(s), and returns the next state value.
244
245 For ordered-set (including hypothetical-set) aggregates, the state
246 transition function receives only the current state value and the
247 aggregated arguments, not the direct arguments. Otherwise it is the
248 same.
249
250 state_data_type
251 The data type for the aggregate's state value.
252
253 state_data_size
254 The approximate average size (in bytes) of the aggregate's state
255 value. If this parameter is omitted or is zero, a default estimate
256 is used based on the state_data_type. The planner uses this value
257 to estimate the memory required for a grouped aggregate query. The
258 planner will consider using hash aggregation for such a query only
259 if the hash table is estimated to fit in work_mem; therefore, large
260 values of this parameter discourage use of hash aggregation.
261
262 ffunc
263 The name of the final function called to compute the aggregate's
264 result after all input rows have been traversed. For a normal
265 aggregate, this function must take a single argument of type
266 state_data_type. The return data type of the aggregate is defined
267 as the return type of this function. If ffunc is not specified,
268 then the ending state value is used as the aggregate's result, and
269 the return type is state_data_type.
270
271 For ordered-set (including hypothetical-set) aggregates, the final
272 function receives not only the final state value, but also the
273 values of all the direct arguments.
274
275 If FINALFUNC_EXTRA is specified, then in addition to the final
276 state value and any direct arguments, the final function receives
277 extra NULL values corresponding to the aggregate's regular
278 (aggregated) arguments. This is mainly useful to allow correct
279 resolution of the aggregate result type when a polymorphic
280 aggregate is being defined.
281
282 FINALFUNC_MODIFY = { READ_ONLY | SHAREABLE | READ_WRITE }
283 This option specifies whether the final function is a pure function
284 that does not modify its arguments. READ_ONLY indicates it does
285 not; the other two values indicate that it may change the
286 transition state value. See NOTES below for more detail. The
287 default is READ_ONLY, except for ordered-set aggregates, for which
288 the default is READ_WRITE.
289
290 combinefunc
291 The combinefunc function may optionally be specified to allow the
292 aggregate function to support partial aggregation. If provided, the
293 combinefunc must combine two state_data_type values, each
294 containing the result of aggregation over some subset of the input
295 values, to produce a new state_data_type that represents the result
296 of aggregating over both sets of inputs. This function can be
297 thought of as an sfunc, where instead of acting upon an individual
298 input row and adding it to the running aggregate state, it adds
299 another aggregate state to the running state.
300
301 The combinefunc must be declared as taking two arguments of the
302 state_data_type and returning a value of the state_data_type.
303 Optionally this function may be “strict”. In this case the function
304 will not be called when either of the input states are null; the
305 other state will be taken as the correct result.
306
307 For aggregate functions whose state_data_type is internal, the
308 combinefunc must not be strict. In this case the combinefunc must
309 ensure that null states are handled correctly and that the state
310 being returned is properly stored in the aggregate memory context.
311
312 serialfunc
313 An aggregate function whose state_data_type is internal can
314 participate in parallel aggregation only if it has a serialfunc
315 function, which must serialize the aggregate state into a bytea
316 value for transmission to another process. This function must take
317 a single argument of type internal and return type bytea. A
318 corresponding deserialfunc is also required.
319
320 deserialfunc
321 Deserialize a previously serialized aggregate state back into
322 state_data_type. This function must take two arguments of types
323 bytea and internal, and produce a result of type internal. (Note:
324 the second, internal argument is unused, but is required for type
325 safety reasons.)
326
327 initial_condition
328 The initial setting for the state value. This must be a string
329 constant in the form accepted for the data type state_data_type. If
330 not specified, the state value starts out null.
331
332 msfunc
333 The name of the forward state transition function to be called for
334 each input row in moving-aggregate mode. This is exactly like the
335 regular transition function, except that its first argument and
336 result are of type mstate_data_type, which might be different from
337 state_data_type.
338
339 minvfunc
340 The name of the inverse state transition function to be used in
341 moving-aggregate mode. This function has the same argument and
342 result types as msfunc, but it is used to remove a value from the
343 current aggregate state, rather than add a value to it. The inverse
344 transition function must have the same strictness attribute as the
345 forward state transition function.
346
347 mstate_data_type
348 The data type for the aggregate's state value, when using
349 moving-aggregate mode.
350
351 mstate_data_size
352 The approximate average size (in bytes) of the aggregate's state
353 value, when using moving-aggregate mode. This works the same as
354 state_data_size.
355
356 mffunc
357 The name of the final function called to compute the aggregate's
358 result after all input rows have been traversed, when using
359 moving-aggregate mode. This works the same as ffunc, except that
360 its first argument's type is mstate_data_type and extra dummy
361 arguments are specified by writing MFINALFUNC_EXTRA. The aggregate
362 result type determined by mffunc or mstate_data_type must match
363 that determined by the aggregate's regular implementation.
364
365 MFINALFUNC_MODIFY = { READ_ONLY | SHAREABLE | READ_WRITE }
366 This option is like FINALFUNC_MODIFY, but it describes the behavior
367 of the moving-aggregate final function.
368
369 minitial_condition
370 The initial setting for the state value, when using
371 moving-aggregate mode. This works the same as initial_condition.
372
373 sort_operator
374 The associated sort operator for a MIN- or MAX-like aggregate. This
375 is just an operator name (possibly schema-qualified). The operator
376 is assumed to have the same input data types as the aggregate
377 (which must be a single-argument normal aggregate).
378
379 PARALLEL = { SAFE | RESTRICTED | UNSAFE }
380 The meanings of PARALLEL SAFE, PARALLEL RESTRICTED, and PARALLEL
381 UNSAFE are the same as in CREATE FUNCTION (CREATE_FUNCTION(7)). An
382 aggregate will not be considered for parallelization if it is
383 marked PARALLEL UNSAFE (which is the default!) or PARALLEL
384 RESTRICTED. Note that the parallel-safety markings of the
385 aggregate's support functions are not consulted by the planner,
386 only the marking of the aggregate itself.
387
388 HYPOTHETICAL
389 For ordered-set aggregates only, this flag specifies that the
390 aggregate arguments are to be processed according to the
391 requirements for hypothetical-set aggregates: that is, the last few
392 direct arguments must match the data types of the aggregated
393 (WITHIN GROUP) arguments. The HYPOTHETICAL flag has no effect on
394 run-time behavior, only on parse-time resolution of the data types
395 and collations of the aggregate's arguments.
396
397 The parameters of CREATE AGGREGATE can be written in any order, not
398 just the order illustrated above.
399
401 In parameters that specify support function names, you can write a
402 schema name if needed, for example SFUNC = public.sum. Do not write
403 argument types there, however — the argument types of the support
404 functions are determined from other parameters.
405
406 Ordinarily, PostgreSQL functions are expected to be true functions that
407 do not modify their input values. However, an aggregate transition
408 function, when used in the context of an aggregate, is allowed to cheat
409 and modify its transition-state argument in place. This can provide
410 substantial performance benefits compared to making a fresh copy of the
411 transition state each time.
412
413 Likewise, while an aggregate final function is normally expected not to
414 modify its input values, sometimes it is impractical to avoid modifying
415 the transition-state argument. Such behavior must be declared using the
416 FINALFUNC_MODIFY parameter. The READ_WRITE value indicates that the
417 final function modifies the transition state in unspecified ways. This
418 value prevents use of the aggregate as a window function, and it also
419 prevents merging of transition states for aggregate calls that share
420 the same input values and transition functions. The SHAREABLE value
421 indicates that the transition function cannot be applied after the
422 final function, but multiple final-function calls can be performed on
423 the ending transition state value. This value prevents use of the
424 aggregate as a window function, but it allows merging of transition
425 states. (That is, the optimization of interest here is not applying the
426 same final function repeatedly, but applying different final functions
427 to the same ending transition state value. This is allowed as long as
428 none of the final functions are marked READ_WRITE.)
429
430 If an aggregate supports moving-aggregate mode, it will improve
431 calculation efficiency when the aggregate is used as a window function
432 for a window with moving frame start (that is, a frame start mode other
433 than UNBOUNDED PRECEDING). Conceptually, the forward transition
434 function adds input values to the aggregate's state when they enter the
435 window frame from the bottom, and the inverse transition function
436 removes them again when they leave the frame at the top. So, when
437 values are removed, they are always removed in the same order they were
438 added. Whenever the inverse transition function is invoked, it will
439 thus receive the earliest added but not yet removed argument value(s).
440 The inverse transition function can assume that at least one row will
441 remain in the current state after it removes the oldest row. (When this
442 would not be the case, the window function mechanism simply starts a
443 fresh aggregation, rather than using the inverse transition function.)
444
445 The forward transition function for moving-aggregate mode is not
446 allowed to return NULL as the new state value. If the inverse
447 transition function returns NULL, this is taken as an indication that
448 the inverse function cannot reverse the state calculation for this
449 particular input, and so the aggregate calculation will be redone from
450 scratch for the current frame starting position. This convention allows
451 moving-aggregate mode to be used in situations where there are some
452 infrequent cases that are impractical to reverse out of the running
453 state value.
454
455 If no moving-aggregate implementation is supplied, the aggregate can
456 still be used with moving frames, but PostgreSQL will recompute the
457 whole aggregation whenever the start of the frame moves. Note that
458 whether or not the aggregate supports moving-aggregate mode, PostgreSQL
459 can handle a moving frame end without recalculation; this is done by
460 continuing to add new values to the aggregate's state. This is why use
461 of an aggregate as a window function requires that the final function
462 be read-only: it must not damage the aggregate's state value, so that
463 the aggregation can be continued even after an aggregate result value
464 has been obtained for one set of frame boundaries.
465
466 The syntax for ordered-set aggregates allows VARIADIC to be specified
467 for both the last direct parameter and the last aggregated (WITHIN
468 GROUP) parameter. However, the current implementation restricts use of
469 VARIADIC in two ways. First, ordered-set aggregates can only use
470 VARIADIC "any", not other variadic array types. Second, if the last
471 direct parameter is VARIADIC "any", then there can be only one
472 aggregated parameter and it must also be VARIADIC "any". (In the
473 representation used in the system catalogs, these two parameters are
474 merged into a single VARIADIC "any" item, since pg_proc cannot
475 represent functions with more than one VARIADIC parameter.) If the
476 aggregate is a hypothetical-set aggregate, the direct arguments that
477 match the VARIADIC "any" parameter are the hypothetical ones; any
478 preceding parameters represent additional direct arguments that are not
479 constrained to match the aggregated arguments.
480
481 Currently, ordered-set aggregates do not need to support
482 moving-aggregate mode, since they cannot be used as window functions.
483
484 Partial (including parallel) aggregation is currently not supported for
485 ordered-set aggregates. Also, it will never be used for aggregate calls
486 that include DISTINCT or ORDER BY clauses, since those semantics cannot
487 be supported during partial aggregation.
488
490 See Section 37.12.
491
493 CREATE AGGREGATE is a PostgreSQL language extension. The SQL standard
494 does not provide for user-defined aggregate functions.
495
497 ALTER AGGREGATE (ALTER_AGGREGATE(7)), DROP AGGREGATE
498 (DROP_AGGREGATE(7))
499
500
501
502PostgreSQL 12.6 2021 CREATE AGGREGATE(7)