1CREATE AGGREGATE(7) PostgreSQL 11.3 Documentation CREATE AGGREGATE(7)
2
6 CREATE_AGGREGATE - define a new aggregate function
7
9 CREATE 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 CREATE 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 or the old syntax
46 CREATE 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. Some basic and
72 commonly-used aggregate functions are included with the distribution;
73 they are documented in Section 9.20. If one defines new types or needs
74 an aggregate function not already provided, then CREATE AGGREGATE can
75 be used to provide the desired features.
76 If a schema name is given (for example, CREATE AGGREGATE myschema.myagg
78 ...) then the aggregate function is created in the specified schema.
79 Otherwise it is created in the current schema.
80 An aggregate function is identified by its name and input data type(s).
82 Two aggregates in the same schema can have the same name if they
83 operate on different input types. The name and input data type(s) of an
84 aggregate must also be distinct from the name and input data type(s) of
85 every ordinary function in the same schema. This behavior is identical
86 to overloading of ordinary function names (see CREATE FUNCTION
87 (CREATE_FUNCTION(7))).
88 A simple aggregate function is made from one or two ordinary functions:
90 a state transition function sfunc, and an optional final calculation
91 function ffunc. These are used as follows:
92 sfunc( internal-state, next-data-values ) ---> next-internal-state
94 ffunc( internal-state ) ---> aggregate-value
95 PostgreSQL creates a temporary variable of data type stype to hold the
97 current internal state of the aggregate. At each input row, the
98 aggregate argument value(s) are calculated and the state transition
99 function is invoked with the current state value and the new argument
100 value(s) to calculate a new internal state value. After all the rows
101 have been processed, the final function is invoked once to calculate
102 the aggregate's return value. If there is no final function then the
103 ending state value is returned as-is.
104 An aggregate function can provide an initial condition, that is, an
106 initial value for the internal state value. This is specified and
107 stored in the database as a value of type text, but it must be a valid
108 external representation of a constant of the state value data type. If
109 it is not supplied then the state value starts out null.
110 If the state transition function is declared “strict”, then it cannot
112 be called with null inputs. With such a transition function, aggregate
113 execution behaves as follows. Rows with any null input values are
114 ignored (the function is not called and the previous state value is
115 retained). If the initial state value is null, then at the first row
116 with all-nonnull input values, the first argument value replaces the
117 state value, and the transition function is invoked at each subsequent
118 row with all-nonnull input values. This is handy for implementing
119 aggregates like max. Note that this behavior is only available when
120 state_data_type is the same as the first arg_data_type. When these
121 types are different, you must supply a nonnull initial condition or use
122 a nonstrict transition function.
123 If the state transition function is not strict, then it will be called
125 unconditionally at each input row, and must deal with null inputs and
126 null state values for itself. This allows the aggregate author to have
127 full control over the aggregate's handling of null values.
128 If the final function is declared “strict”, then it will not be called
130 when the ending state value is null; instead a null result will be
131 returned automatically. (Of course this is just the normal behavior of
132 strict functions.) In any case the final function has the option of
133 returning a null value. For example, the final function for avg returns
134 null when it sees there were zero input rows.
135 Sometimes it is useful to declare the final function as taking not just
137 the state value, but extra parameters corresponding to the aggregate's
138 input values. The main reason for doing this is if the final function
139 is polymorphic and the state value's data type would be inadequate to
140 pin down the result type. These extra parameters are always passed as
141 NULL (and so the final function must not be strict when the
142 FINALFUNC_EXTRA option is used), but nonetheless they are valid
143 parameters. The final function could for example make use of
144 get_fn_expr_argtype to identify the actual argument type in the current
145 call.
146 An aggregate can optionally support moving-aggregate mode, as described
148 in Section 38.11.1. This requires specifying the MSFUNC, MINVFUNC, and
149 MSTYPE parameters, and optionally the MSPACE, MFINALFUNC,
150 MFINALFUNC_EXTRA, MFINALFUNC_MODIFY, and MINITCOND parameters. Except
151 for MINVFUNC, these parameters work like the corresponding
152 simple-aggregate parameters without M; they define a separate
153 implementation of the aggregate that includes an inverse transition
154 function.
155 The syntax with ORDER BY in the parameter list creates a special type
157 of aggregate called an ordered-set aggregate; or if HYPOTHETICAL is
158 specified, then a hypothetical-set aggregate is created. These
159 aggregates operate over groups of sorted values in order-dependent
160 ways, so that specification of an input sort order is an essential part
161 of a call. Also, they can have direct arguments, which are arguments
162 that are evaluated only once per aggregation rather than once per input
163 row. Hypothetical-set aggregates are a subclass of ordered-set
164 aggregates in which some of the direct arguments are required to match,
165 in number and data types, the aggregated argument columns. This allows
166 the values of those direct arguments to be added to the collection of
167 aggregate-input rows as an additional “hypothetical” row.
168 An aggregate can optionally support partial aggregation, as described
170 in Section 38.11.4. This requires specifying the COMBINEFUNC parameter.
171 If the state_data_type is internal, it's usually also appropriate to
172 provide the SERIALFUNC and DESERIALFUNC parameters so that parallel
173 aggregation is possible. Note that the aggregate must also be marked
174 PARALLEL SAFE to enable parallel aggregation.
175 Aggregates that behave like MIN or MAX can sometimes be optimized by
177 looking into an index instead of scanning every input row. If this
178 aggregate can be so optimized, indicate it by specifying a sort
179 operator. The basic requirement is that the aggregate must yield the
180 first element in the sort ordering induced by the operator; in other
181 words:
182 SELECT agg(col) FROM tab;
184 must be equivalent to:
186 SELECT col FROM tab ORDER BY col USING sortop LIMIT 1;
188 Further assumptions are that the aggregate ignores null inputs, and
190 that it delivers a null result if and only if there were no non-null
191 inputs. Ordinarily, a data type's < operator is the proper sort
192 operator for MIN, and > is the proper sort operator for MAX. Note that
193 the optimization will never actually take effect unless the specified
194 operator is the “less than” or “greater than” strategy member of a
195 B-tree index operator class.
196 To be able to create an aggregate function, you must have USAGE
198 privilege on the argument types, the state type(s), and the return
199 type, as well as EXECUTE privilege on the supporting functions.
200
202 name
203 The name (optionally schema-qualified) of the aggregate function to
204 create.
205 argmode
207 The mode of an argument: IN or VARIADIC. (Aggregate functions do
208 not support OUT arguments.) If omitted, the default is IN. Only the
209 last argument can be marked VARIADIC.
210 argname
212 The name of an argument. This is currently only useful for
213 documentation purposes. If omitted, the argument has no name.
214 arg_data_type
216 An input data type on which this aggregate function operates. To
217 create a zero-argument aggregate function, write * in place of the
218 list of argument specifications. (An example of such an aggregate
219 is count(*).)
220 base_type
222 In the old syntax for CREATE AGGREGATE, the input data type is
223 specified by a basetype parameter rather than being written next to
224 the aggregate name. Note that this syntax allows only one input
225 parameter. To define a zero-argument aggregate function with this
226 syntax, specify the basetype as "ANY" (not *). Ordered-set
227 aggregates cannot be defined with the old syntax.
228 sfunc
230 The name of the state transition function to be called for each
231 input row. For a normal N-argument aggregate function, the sfunc
232 must take N+1 arguments, the first being of type state_data_type
233 and the rest matching the declared input data type(s) of the
234 aggregate. The function must return a value of type
235 state_data_type. This function takes the current state value and
236 the current input data value(s), and returns the next state value.
237 For ordered-set (including hypothetical-set) aggregates, the state
239 transition function receives only the current state value and the
240 aggregated arguments, not the direct arguments. Otherwise it is the
241 same.
242 state_data_type
244 The data type for the aggregate's state value.
245 state_data_size
247 The approximate average size (in bytes) of the aggregate's state
248 value. If this parameter is omitted or is zero, a default estimate
249 is used based on the state_data_type. The planner uses this value
250 to estimate the memory required for a grouped aggregate query. The
251 planner will consider using hash aggregation for such a query only
252 if the hash table is estimated to fit in work_mem; therefore, large
253 values of this parameter discourage use of hash aggregation.
254 ffunc
256 The name of the final function called to compute the aggregate's
257 result after all input rows have been traversed. For a normal
258 aggregate, this function must take a single argument of type
259 state_data_type. The return data type of the aggregate is defined
260 as the return type of this function. If ffunc is not specified,
261 then the ending state value is used as the aggregate's result, and
262 the return type is state_data_type.
263 For ordered-set (including hypothetical-set) aggregates, the final
265 function receives not only the final state value, but also the
266 values of all the direct arguments.
267 If FINALFUNC_EXTRA is specified, then in addition to the final
269 state value and any direct arguments, the final function receives
270 extra NULL values corresponding to the aggregate's regular
271 (aggregated) arguments. This is mainly useful to allow correct
272 resolution of the aggregate result type when a polymorphic
273 aggregate is being defined.
274 FINALFUNC_MODIFY = { READ_ONLY | SHAREABLE | READ_WRITE }
276 This option specifies whether the final function is a pure function
277 that does not modify its arguments. READ_ONLY indicates it does
278 not; the other two values indicate that it may change the
279 transition state value. See NOTES below for more detail. The
280 default is READ_ONLY, except for ordered-set aggregates, for which
281 the default is READ_WRITE.
282 combinefunc
284 The combinefunc function may optionally be specified to allow the
285 aggregate function to support partial aggregation. If provided, the
286 combinefunc must combine two state_data_type values, each
287 containing the result of aggregation over some subset of the input
288 values, to produce a new state_data_type that represents the result
289 of aggregating over both sets of inputs. This function can be
290 thought of as an sfunc, where instead of acting upon an individual
291 input row and adding it to the running aggregate state, it adds
292 another aggregate state to the running state.
293 The combinefunc must be declared as taking two arguments of the
295 state_data_type and returning a value of the state_data_type.
296 Optionally this function may be “strict”. In this case the function
297 will not be called when either of the input states are null; the
298 other state will be taken as the correct result.
299 For aggregate functions whose state_data_type is internal, the
301 combinefunc must not be strict. In this case the combinefunc must
302 ensure that null states are handled correctly and that the state
303 being returned is properly stored in the aggregate memory context.
304 serialfunc
306 An aggregate function whose state_data_type is internal can
307 participate in parallel aggregation only if it has a serialfunc
308 function, which must serialize the aggregate state into a bytea
309 value for transmission to another process. This function must take
310 a single argument of type internal and return type bytea. A
311 corresponding deserialfunc is also required.
312 deserialfunc
314 Deserialize a previously serialized aggregate state back into
315 state_data_type. This function must take two arguments of types
316 bytea and internal, and produce a result of type internal. (Note:
317 the second, internal argument is unused, but is required for type
318 safety reasons.)
319 initial_condition
321 The initial setting for the state value. This must be a string
322 constant in the form accepted for the data type state_data_type. If
323 not specified, the state value starts out null.
324 msfunc
326 The name of the forward state transition function to be called for
327 each input row in moving-aggregate mode. This is exactly like the
328 regular transition function, except that its first argument and
329 result are of type mstate_data_type, which might be different from
330 state_data_type.
331 minvfunc
333 The name of the inverse state transition function to be used in
334 moving-aggregate mode. This function has the same argument and
335 result types as msfunc, but it is used to remove a value from the
336 current aggregate state, rather than add a value to it. The inverse
337 transition function must have the same strictness attribute as the
338 forward state transition function.
339 mstate_data_type
341 The data type for the aggregate's state value, when using
342 moving-aggregate mode.
343 mstate_data_size
345 The approximate average size (in bytes) of the aggregate's state
346 value, when using moving-aggregate mode. This works the same as
347 state_data_size.
348 mffunc
350 The name of the final function called to compute the aggregate's
351 result after all input rows have been traversed, when using
352 moving-aggregate mode. This works the same as ffunc, except that
353 its first argument's type is mstate_data_type and extra dummy
354 arguments are specified by writing MFINALFUNC_EXTRA. The aggregate
355 result type determined by mffunc or mstate_data_type must match
356 that determined by the aggregate's regular implementation.
357 MFINALFUNC_MODIFY = { READ_ONLY | SHAREABLE | READ_WRITE }
359 This option is like FINALFUNC_MODIFY, but it describes the behavior
360 of the moving-aggregate final function.
361 minitial_condition
363 The initial setting for the state value, when using
364 moving-aggregate mode. This works the same as initial_condition.
365 sort_operator
367 The associated sort operator for a MIN- or MAX-like aggregate. This
368 is just an operator name (possibly schema-qualified). The operator
369 is assumed to have the same input data types as the aggregate
370 (which must be a single-argument normal aggregate).
371 PARALLEL = { SAFE | RESTRICTED | UNSAFE }
373 The meanings of PARALLEL SAFE, PARALLEL RESTRICTED, and PARALLEL
374 UNSAFE are the same as in CREATE FUNCTION (CREATE_FUNCTION(7)). An
375 aggregate will not be considered for parallelization if it is
376 marked PARALLEL UNSAFE (which is the default!) or PARALLEL
377 RESTRICTED. Note that the parallel-safety markings of the
378 aggregate's support functions are not consulted by the planner,
379 only the marking of the aggregate itself.
380 HYPOTHETICAL
382 For ordered-set aggregates only, this flag specifies that the
383 aggregate arguments are to be processed according to the
384 requirements for hypothetical-set aggregates: that is, the last few
385 direct arguments must match the data types of the aggregated
386 (WITHIN GROUP) arguments. The HYPOTHETICAL flag has no effect on
387 run-time behavior, only on parse-time resolution of the data types
388 and collations of the aggregate's arguments.
389 The parameters of CREATE AGGREGATE can be written in any order, not
391 just the order illustrated above.
392
394 In parameters that specify support function names, you can write a
395 schema name if needed, for example SFUNC = public.sum. Do not write
396 argument types there, however — the argument types of the support
397 functions are determined from other parameters.
398 Ordinarily, PostgreSQL functions are expected to be true functions that
400 do not modify their input values. However, an aggregate transition
401 function, when used in the context of an aggregate, is allowed to cheat
402 and modify its transition-state argument in place. This can provide
403 substantial performance benefits compared to making a fresh copy of the
404 transition state each time.
405 Likewise, while an aggregate final function is normally expected not to
407 modify its input values, sometimes it is impractical to avoid modifying
408 the transition-state argument. Such behavior must be declared using the
409 FINALFUNC_MODIFY parameter. The READ_WRITE value indicates that the
410 final function modifies the transition state in unspecified ways. This
411 value prevents use of the aggregate as a window function, and it also
412 prevents merging of transition states for aggregate calls that share
413 the same input values and transition functions. The SHAREABLE value
414 indicates that the transition function cannot be applied after the
415 final function, but multiple final-function calls can be performed on
416 the ending transition state value. This value prevents use of the
417 aggregate as a window function, but it allows merging of transition
418 states. (That is, the optimization of interest here is not applying the
419 same final function repeatedly, but applying different final functions
420 to the same ending transition state value. This is allowed as long as
421 none of the final functions are marked READ_WRITE.)
422 If an aggregate supports moving-aggregate mode, it will improve
424 calculation efficiency when the aggregate is used as a window function
425 for a window with moving frame start (that is, a frame start mode other
426 than UNBOUNDED PRECEDING). Conceptually, the forward transition
427 function adds input values to the aggregate's state when they enter the
428 window frame from the bottom, and the inverse transition function
429 removes them again when they leave the frame at the top. So, when
430 values are removed, they are always removed in the same order they were
431 added. Whenever the inverse transition function is invoked, it will
432 thus receive the earliest added but not yet removed argument value(s).
433 The inverse transition function can assume that at least one row will
434 remain in the current state after it removes the oldest row. (When this
435 would not be the case, the window function mechanism simply starts a
436 fresh aggregation, rather than using the inverse transition function.)
437 The forward transition function for moving-aggregate mode is not
439 allowed to return NULL as the new state value. If the inverse
440 transition function returns NULL, this is taken as an indication that
441 the inverse function cannot reverse the state calculation for this
442 particular input, and so the aggregate calculation will be redone from
443 scratch for the current frame starting position. This convention allows
444 moving-aggregate mode to be used in situations where there are some
445 infrequent cases that are impractical to reverse out of the running
446 state value.
447 If no moving-aggregate implementation is supplied, the aggregate can
449 still be used with moving frames, but PostgreSQL will recompute the
450 whole aggregation whenever the start of the frame moves. Note that
451 whether or not the aggregate supports moving-aggregate mode, PostgreSQL
452 can handle a moving frame end without recalculation; this is done by
453 continuing to add new values to the aggregate's state. This is why use
454 of an aggregate as a window function requires that the final function
455 be read-only: it must not damage the aggregate's state value, so that
456 the aggregation can be continued even after an aggregate result value
457 has been obtained for one set of frame boundaries.
458 The syntax for ordered-set aggregates allows VARIADIC to be specified
460 for both the last direct parameter and the last aggregated (WITHIN
461 GROUP) parameter. However, the current implementation restricts use of
462 VARIADIC in two ways. First, ordered-set aggregates can only use
463 VARIADIC "any", not other variadic array types. Second, if the last
464 direct parameter is VARIADIC "any", then there can be only one
465 aggregated parameter and it must also be VARIADIC "any". (In the
466 representation used in the system catalogs, these two parameters are
467 merged into a single VARIADIC "any" item, since pg_proc cannot
468 represent functions with more than one VARIADIC parameter.) If the
469 aggregate is a hypothetical-set aggregate, the direct arguments that
470 match the VARIADIC "any" parameter are the hypothetical ones; any
471 preceding parameters represent additional direct arguments that are not
472 constrained to match the aggregated arguments.
473 Currently, ordered-set aggregates do not need to support
475 moving-aggregate mode, since they cannot be used as window functions.
476 Partial (including parallel) aggregation is currently not supported for
478 ordered-set aggregates. Also, it will never be used for aggregate calls
479 that include DISTINCT or ORDER BY clauses, since those semantics cannot
480 be supported during partial aggregation.
481
483 See Section 38.11.
484
486 CREATE AGGREGATE is a PostgreSQL language extension. The SQL standard
487 does not provide for user-defined aggregate functions.
488
490 ALTER AGGREGATE (ALTER_AGGREGATE(7)), DROP AGGREGATE
491 (DROP_AGGREGATE(7))
492PostgreSQL 11.3 2019 CREATE AGGREGATE(7)