| 1 |
<!-- doc/src/sgml/xaggr.sgml --> |
| 2 |
|
| 3 |
<sect1 id="xaggr"> |
| 4 |
<title>User-Defined Aggregates</title> |
| 5 |
|
| 6 |
<indexterm zone="xaggr"> |
| 7 |
<primary>aggregate function</primary> |
| 8 |
<secondary>user-defined</secondary> |
| 9 |
</indexterm> |
| 10 |
|
| 11 |
<para> |
| 12 |
Aggregate functions in <productname>PostgreSQL</productname> |
| 13 |
are defined in terms of <firstterm>state values</firstterm> |
| 14 |
and <firstterm>state transition functions</firstterm>. |
| 15 |
That is, an aggregate operates using a state value that is updated |
| 16 |
as each successive input row is processed. |
| 17 |
To define a new aggregate |
| 18 |
function, one selects a data type for the state value, |
| 19 |
an initial value for the state, and a state transition |
| 20 |
function. The state transition function takes the previous state |
| 21 |
value and the aggregate's input value(s) for the current row, and |
| 22 |
returns a new state value. |
| 23 |
A <firstterm>final function</firstterm> |
| 24 |
can also be specified, in case the desired result of the aggregate |
| 25 |
is different from the data that needs to be kept in the running |
| 26 |
state value. The final function takes the ending state value |
| 27 |
and returns whatever is wanted as the aggregate result. |
| 28 |
In principle, the transition and final functions are just ordinary |
| 29 |
functions that could also be used outside the context of the |
| 30 |
aggregate. (In practice, it's often helpful for performance reasons |
| 31 |
to create specialized transition functions that can only work when |
| 32 |
called as part of an aggregate.) |
| 33 |
</para> |
| 34 |
|
| 35 |
<para> |
| 36 |
Thus, in addition to the argument and result data types seen by a user |
| 37 |
of the aggregate, there is an internal state-value data type that |
| 38 |
might be different from both the argument and result types. |
| 39 |
</para> |
| 40 |
|
| 41 |
<para> |
| 42 |
If we define an aggregate that does not use a final function, |
| 43 |
we have an aggregate that computes a running function of |
| 44 |
the column values from each row. <function>sum</function> is an |
| 45 |
example of this kind of aggregate. <function>sum</function> starts at |
| 46 |
zero and always adds the current row's value to |
| 47 |
its running total. For example, if we want to make a <function>sum</function> |
| 48 |
aggregate to work on a data type for complex numbers, |
| 49 |
we only need the addition function for that data type. |
| 50 |
The aggregate definition would be: |
| 51 |
|
| 52 |
<programlisting> |
| 53 |
CREATE AGGREGATE sum (complex) |
| 54 |
( |
| 55 |
sfunc = complex_add, |
| 56 |
stype = complex, |
| 57 |
initcond = '(0,0)' |
| 58 |
); |
| 59 |
</programlisting> |
| 60 |
|
| 61 |
which we might use like this: |
| 62 |
|
| 63 |
<programlisting> |
| 64 |
SELECT sum(a) FROM test_complex; |
| 65 |
|
| 66 |
sum |
| 67 |
----------- |
| 68 |
(34,53.9) |
| 69 |
</programlisting> |
| 70 |
|
| 71 |
(Notice that we are relying on function overloading: there is more than |
| 72 |
one aggregate named <function>sum</function>, but |
| 73 |
<productname>PostgreSQL</productname> can figure out which kind |
| 74 |
of sum applies to a column of type <type>complex</type>.) |
| 75 |
</para> |
| 76 |
|
| 77 |
<para> |
| 78 |
The above definition of <function>sum</function> will return zero |
| 79 |
(the initial state value) if there are no nonnull input values. |
| 80 |
Perhaps we want to return null in that case instead — the SQL standard |
| 81 |
expects <function>sum</function> to behave that way. We can do this simply by |
| 82 |
omitting the <literal>initcond</literal> phrase, so that the initial state |
| 83 |
value is null. Ordinarily this would mean that the <literal>sfunc</literal> |
| 84 |
would need to check for a null state-value input. But for |
| 85 |
<function>sum</function> and some other simple aggregates like |
| 86 |
<function>max</function> and <function>min</function>, |
| 87 |
it is sufficient to insert the first nonnull input value into |
| 88 |
the state variable and then start applying the transition function |
| 89 |
at the second nonnull input value. <productname>PostgreSQL</productname> |
| 90 |
will do that automatically if the initial state value is null and |
| 91 |
the transition function is marked <quote>strict</quote> (i.e., not to be called |
| 92 |
for null inputs). |
| 93 |
</para> |
| 94 |
|
| 95 |
<para> |
| 96 |
Another bit of default behavior for a <quote>strict</quote> transition function |
| 97 |
is that the previous state value is retained unchanged whenever a |
| 98 |
null input value is encountered. Thus, null values are ignored. If you |
| 99 |
need some other behavior for null inputs, do not declare your |
| 100 |
transition function as strict; instead code it to test for null inputs and |
| 101 |
do whatever is needed. |
| 102 |
</para> |
| 103 |
|
| 104 |
<para> |
| 105 |
<function>avg</function> (average) is a more complex example of an aggregate. |
| 106 |
It requires |
| 107 |
two pieces of running state: the sum of the inputs and the count |
| 108 |
of the number of inputs. The final result is obtained by dividing |
| 109 |
these quantities. Average is typically implemented by using an |
| 110 |
array as the state value. For example, |
| 111 |
the built-in implementation of <function>avg(float8)</function> |
| 112 |
looks like: |
| 113 |
|
| 114 |
<programlisting> |
| 115 |
CREATE AGGREGATE avg (float8) |
| 116 |
( |
| 117 |
sfunc = float8_accum, |
| 118 |
stype = float8[], |
| 119 |
finalfunc = float8_avg, |
| 120 |
initcond = '{0,0,0}' |
| 121 |
); |
| 122 |
</programlisting> |
| 123 |
</para> |
| 124 |
|
| 125 |
<note> |
| 126 |
<para> |
| 127 |
<function>float8_accum</function> requires a three-element array, not just |
| 128 |
two elements, because it accumulates the sum of squares as well as |
| 129 |
the sum and count of the inputs. This is so that it can be used for |
| 130 |
some other aggregates as well as <function>avg</function>. |
| 131 |
</para> |
| 132 |
</note> |
| 133 |
|
| 134 |
<para> |
| 135 |
Aggregate function calls in SQL allow <literal>DISTINCT</literal> |
| 136 |
and <literal>ORDER BY</literal> options that control which rows are fed |
| 137 |
to the aggregate's transition function and in what order. These |
| 138 |
options are implemented behind the scenes and are not the concern |
| 139 |
of the aggregate's support functions. |
| 140 |
</para> |
| 141 |
|
| 142 |
<para> |
| 143 |
For further details see the |
| 144 |
<xref linkend="sql-createaggregate"/> |
| 145 |
command. |
| 146 |
</para> |
| 147 |
|
| 148 |
<sect2 id="xaggr-moving-aggregates"> |
| 149 |
<title>Moving-Aggregate Mode</title> |
| 150 |
|
| 151 |
<indexterm> |
| 152 |
<primary>moving-aggregate mode</primary> |
| 153 |
</indexterm> |
| 154 |
|
| 155 |
<indexterm> |
| 156 |
<primary>aggregate function</primary> |
| 157 |
<secondary>moving aggregate</secondary> |
| 158 |
</indexterm> |
| 159 |
|
| 160 |
<para> |
| 161 |
Aggregate functions can optionally support <firstterm>moving-aggregate |
| 162 |
mode</firstterm>, which allows substantially faster execution of aggregate |
| 163 |
functions within windows with moving frame starting points. |
| 164 |
(See <xref linkend="tutorial-window"/> |
| 165 |
and <xref linkend="syntax-window-functions"/> for information about use of |
| 166 |
aggregate functions as window functions.) |
| 167 |
The basic idea is that in addition to a normal <quote>forward</quote> |
| 168 |
transition function, the aggregate provides an <firstterm>inverse |
| 169 |
transition function</firstterm>, which allows rows to be removed from the |
| 170 |
aggregate's running state value when they exit the window frame. |
| 171 |
For example a <function>sum</function> aggregate, which uses addition as the |
| 172 |
forward transition function, would use subtraction as the inverse |
| 173 |
transition function. Without an inverse transition function, the window |
| 174 |
function mechanism must recalculate the aggregate from scratch each time |
| 175 |
the frame starting point moves, resulting in run time proportional to the |
| 176 |
number of input rows times the average frame length. With an inverse |
| 177 |
transition function, the run time is only proportional to the number of |
| 178 |
input rows. |
| 179 |
</para> |
| 180 |
|
| 181 |
<para> |
| 182 |
The inverse transition function is passed the current state value and the |
| 183 |
aggregate input value(s) for the earliest row included in the current |
| 184 |
state. It must reconstruct what the state value would have been if the |
| 185 |
given input row had never been aggregated, but only the rows following |
| 186 |
it. This sometimes requires that the forward transition function keep |
| 187 |
more state than is needed for plain aggregation mode. Therefore, the |
| 188 |
moving-aggregate mode uses a completely separate implementation from the |
| 189 |
plain mode: it has its own state data type, its own forward transition |
| 190 |
function, and its own final function if needed. These can be the same as |
| 191 |
the plain mode's data type and functions, if there is no need for extra |
| 192 |
state. |
| 193 |
</para> |
| 194 |
|
| 195 |
<para> |
| 196 |
As an example, we could extend the <function>sum</function> aggregate given above |
| 197 |
to support moving-aggregate mode like this: |
| 198 |
|
| 199 |
<programlisting> |
| 200 |
CREATE AGGREGATE sum (complex) |
| 201 |
( |
| 202 |
sfunc = complex_add, |
| 203 |
stype = complex, |
| 204 |
initcond = '(0,0)', |
| 205 |
msfunc = complex_add, |
| 206 |
minvfunc = complex_sub, |
| 207 |
mstype = complex, |
| 208 |
minitcond = '(0,0)' |
| 209 |
); |
| 210 |
</programlisting> |
| 211 |
|
| 212 |
The parameters whose names begin with <literal>m</literal> define the |
| 213 |
moving-aggregate implementation. Except for the inverse transition |
| 214 |
function <literal>minvfunc</literal>, they correspond to the plain-aggregate |
| 215 |
parameters without <literal>m</literal>. |
| 216 |
</para> |
| 217 |
|
| 218 |
<para> |
| 219 |
The forward transition function for moving-aggregate mode is not allowed |
| 220 |
to return null as the new state value. If the inverse transition |
| 221 |
function returns null, this is taken as an indication that the inverse |
| 222 |
function cannot reverse the state calculation for this particular input, |
| 223 |
and so the aggregate calculation will be redone from scratch for the |
| 224 |
current frame starting position. This convention allows moving-aggregate |
| 225 |
mode to be used in situations where there are some infrequent cases that |
| 226 |
are impractical to reverse out of the running state value. The inverse |
| 227 |
transition function can <quote>punt</quote> on these cases, and yet still come |
| 228 |
out ahead so long as it can work for most cases. As an example, an |
| 229 |
aggregate working with floating-point numbers might choose to punt when |
| 230 |
a <literal>NaN</literal> (not a number) input has to be removed from the running |
| 231 |
state value. |
| 232 |
</para> |
| 233 |
|
| 234 |
<para> |
| 235 |
When writing moving-aggregate support functions, it is important to be |
| 236 |
sure that the inverse transition function can reconstruct the correct |
| 237 |
state value exactly. Otherwise there might be user-visible differences |
| 238 |
in results depending on whether the moving-aggregate mode is used. |
| 239 |
An example of an aggregate for which adding an inverse transition |
| 240 |
function seems easy at first, yet where this requirement cannot be met |
| 241 |
is <function>sum</function> over <type>float4</type> or <type>float8</type> inputs. A |
| 242 |
naive declaration of <function>sum(<type>float8</type>)</function> could be |
| 243 |
|
| 244 |
<programlisting> |
| 245 |
CREATE AGGREGATE unsafe_sum (float8) |
| 246 |
( |
| 247 |
stype = float8, |
| 248 |
sfunc = float8pl, |
| 249 |
mstype = float8, |
| 250 |
msfunc = float8pl, |
| 251 |
minvfunc = float8mi |
| 252 |
); |
| 253 |
</programlisting> |
| 254 |
|
| 255 |
This aggregate, however, can give wildly different results than it would |
| 256 |
have without the inverse transition function. For example, consider |
| 257 |
|
| 258 |
<programlisting> |
| 259 |
SELECT |
| 260 |
unsafe_sum(x) OVER (ORDER BY n ROWS BETWEEN CURRENT ROW AND 1 FOLLOWING) |
| 261 |
FROM (VALUES (1, 1.0e20::float8), |
| 262 |
(2, 1.0::float8)) AS v (n,x); |
| 263 |
</programlisting> |
| 264 |
|
| 265 |
This query returns <literal>0</literal> as its second result, rather than the |
| 266 |
expected answer of <literal>1</literal>. The cause is the limited precision of |
| 267 |
floating-point values: adding <literal>1</literal> to <literal>1e20</literal> results |
| 268 |
in <literal>1e20</literal> again, and so subtracting <literal>1e20</literal> from that |
| 269 |
yields <literal>0</literal>, not <literal>1</literal>. Note that this is a limitation |
| 270 |
of floating-point arithmetic in general, not a limitation |
| 271 |
of <productname>PostgreSQL</productname>. |
| 272 |
</para> |
| 273 |
|
| 274 |
</sect2> |
| 275 |
|
| 276 |
<sect2 id="xaggr-polymorphic-aggregates"> |
| 277 |
<title>Polymorphic and Variadic Aggregates</title> |
| 278 |
|
| 279 |
<indexterm> |
| 280 |
<primary>aggregate function</primary> |
| 281 |
<secondary>polymorphic</secondary> |
| 282 |
</indexterm> |
| 283 |
|
| 284 |
<indexterm> |
| 285 |
<primary>aggregate function</primary> |
| 286 |
<secondary>variadic</secondary> |
| 287 |
</indexterm> |
| 288 |
|
| 289 |
<para> |
| 290 |
Aggregate functions can use polymorphic |
| 291 |
state transition functions or final functions, so that the same functions |
| 292 |
can be used to implement multiple aggregates. |
| 293 |
See <xref linkend="extend-types-polymorphic"/> |
| 294 |
for an explanation of polymorphic functions. |
| 295 |
Going a step further, the aggregate function itself can be specified |
| 296 |
with polymorphic input type(s) and state type, allowing a single |
| 297 |
aggregate definition to serve for multiple input data types. |
| 298 |
Here is an example of a polymorphic aggregate: |
| 299 |
|
| 300 |
<programlisting> |
| 301 |
CREATE AGGREGATE array_accum (anycompatible) |
| 302 |
( |
| 303 |
sfunc = array_append, |
| 304 |
stype = anycompatiblearray, |
| 305 |
initcond = '{}' |
| 306 |
); |
| 307 |
</programlisting> |
| 308 |
|
| 309 |
Here, the actual state type for any given aggregate call is the array type |
| 310 |
having the actual input type as elements. The behavior of the aggregate |
| 311 |
is to concatenate all the inputs into an array of that type. |
| 312 |
(Note: the built-in aggregate <function>array_agg</function> provides similar |
| 313 |
functionality, with better performance than this definition would have.) |
| 314 |
</para> |
| 315 |
|
| 316 |
<para> |
| 317 |
Here's the output using two different actual data types as arguments: |
| 318 |
|
| 319 |
<programlisting> |
| 320 |
SELECT attrelid::regclass, array_accum(attname) |
| 321 |
FROM pg_attribute |
| 322 |
WHERE attnum > 0 AND attrelid = 'pg_tablespace'::regclass |
| 323 |
GROUP BY attrelid; |
| 324 |
|
| 325 |
attrelid | array_accum |
| 326 |
---------------+--------------------------------------- |
| 327 |
pg_tablespace | {spcname,spcowner,spcacl,spcoptions} |
| 328 |
(1 row) |
| 329 |
|
| 330 |
SELECT attrelid::regclass, array_accum(atttypid::regtype) |
| 331 |
FROM pg_attribute |
| 332 |
WHERE attnum > 0 AND attrelid = 'pg_tablespace'::regclass |
| 333 |
GROUP BY attrelid; |
| 334 |
|
| 335 |
attrelid | array_accum |
| 336 |
---------------+--------------------------- |
| 337 |
pg_tablespace | {name,oid,aclitem[],text[]} |
| 338 |
(1 row) |
| 339 |
</programlisting> |
| 340 |
</para> |
| 341 |
|
| 342 |
<para> |
| 343 |
Ordinarily, an aggregate function with a polymorphic result type has a |
| 344 |
polymorphic state type, as in the above example. This is necessary |
| 345 |
because otherwise the final function cannot be declared sensibly: it |
| 346 |
would need to have a polymorphic result type but no polymorphic argument |
| 347 |
type, which <command>CREATE FUNCTION</command> will reject on the grounds that |
| 348 |
the result type cannot be deduced from a call. But sometimes it is |
| 349 |
inconvenient to use a polymorphic state type. The most common case is |
| 350 |
where the aggregate support functions are to be written in C and the |
| 351 |
state type should be declared as <type>internal</type> because there is |
| 352 |
no SQL-level equivalent for it. To address this case, it is possible to |
| 353 |
declare the final function as taking extra <quote>dummy</quote> arguments |
| 354 |
that match the input arguments of the aggregate. Such dummy arguments |
| 355 |
are always passed as null values since no specific value is available when the |
| 356 |
final function is called. Their only use is to allow a polymorphic |
| 357 |
final function's result type to be connected to the aggregate's input |
| 358 |
type(s). For example, the definition of the built-in |
| 359 |
aggregate <function>array_agg</function> is equivalent to |
| 360 |
|
| 361 |
<programlisting> |
| 362 |
CREATE FUNCTION array_agg_transfn(internal, anynonarray) |
| 363 |
RETURNS internal ...; |
| 364 |
CREATE FUNCTION array_agg_finalfn(internal, anynonarray) |
| 365 |
RETURNS anyarray ...; |
| 366 |
|
| 367 |
CREATE AGGREGATE array_agg (anynonarray) |
| 368 |
( |
| 369 |
sfunc = array_agg_transfn, |
| 370 |
stype = internal, |
| 371 |
finalfunc = array_agg_finalfn, |
| 372 |
finalfunc_extra |
| 373 |
); |
| 374 |
</programlisting> |
| 375 |
|
| 376 |
Here, the <literal>finalfunc_extra</literal> option specifies that the final |
| 377 |
function receives, in addition to the state value, extra dummy |
| 378 |
argument(s) corresponding to the aggregate's input argument(s). |
| 379 |
The extra <type>anynonarray</type> argument allows the declaration |
| 380 |
of <function>array_agg_finalfn</function> to be valid. |
| 381 |
</para> |
| 382 |
|
| 383 |
<para> |
| 384 |
An aggregate function can be made to accept a varying number of arguments |
| 385 |
by declaring its last argument as a <literal>VARIADIC</literal> array, in much |
| 386 |
the same fashion as for regular functions; see |
| 387 |
<xref linkend="xfunc-sql-variadic-functions"/>. The aggregate's transition |
| 388 |
function(s) must have the same array type as their last argument. The |
| 389 |
transition function(s) typically would also be marked <literal>VARIADIC</literal>, |
| 390 |
but this is not strictly required. |
| 391 |
</para> |
| 392 |
|
| 393 |
<note> |
| 394 |
<para> |
| 395 |
Variadic aggregates are easily misused in connection with |
| 396 |
the <literal>ORDER BY</literal> option (see <xref linkend="syntax-aggregates"/>), |
| 397 |
since the parser cannot tell whether the wrong number of actual arguments |
| 398 |
have been given in such a combination. Keep in mind that everything to |
| 399 |
the right of <literal>ORDER BY</literal> is a sort key, not an argument to the |
| 400 |
aggregate. For example, in |
| 401 |
<programlisting> |
| 402 |
SELECT myaggregate(a ORDER BY a, b, c) FROM ... |
| 403 |
</programlisting> |
| 404 |
the parser will see this as a single aggregate function argument and |
| 405 |
three sort keys. However, the user might have intended |
| 406 |
<programlisting> |
| 407 |
SELECT myaggregate(a, b, c ORDER BY a) FROM ... |
| 408 |
</programlisting> |
| 409 |
If <literal>myaggregate</literal> is variadic, both these calls could be |
| 410 |
perfectly valid. |
| 411 |
</para> |
| 412 |
|
| 413 |
<para> |
| 414 |
For the same reason, it's wise to think twice before creating aggregate |
| 415 |
functions with the same names and different numbers of regular arguments. |
| 416 |
</para> |
| 417 |
</note> |
| 418 |
|
| 419 |
</sect2> |
| 420 |
|
| 421 |
<sect2 id="xaggr-ordered-set-aggregates"> |
| 422 |
<title>Ordered-Set Aggregates</title> |
| 423 |
|
| 424 |
<indexterm> |
| 425 |
<primary>aggregate function</primary> |
| 426 |
<secondary>ordered set</secondary> |
| 427 |
</indexterm> |
| 428 |
|
| 429 |
<para> |
| 430 |
The aggregates we have been describing so far are <quote>normal</quote> |
| 431 |
aggregates. <productname>PostgreSQL</productname> also |
| 432 |
supports <firstterm>ordered-set aggregates</firstterm>, which differ from |
| 433 |
normal aggregates in two key ways. First, in addition to ordinary |
| 434 |
aggregated arguments that are evaluated once per input row, an |
| 435 |
ordered-set aggregate can have <quote>direct</quote> arguments that are |
| 436 |
evaluated only once per aggregation operation. Second, the syntax |
| 437 |
for the ordinary aggregated arguments specifies a sort ordering |
| 438 |
for them explicitly. An ordered-set aggregate is usually |
| 439 |
used to implement a computation that depends on a specific row |
| 440 |
ordering, for instance rank or percentile, so that the sort ordering |
| 441 |
is a required aspect of any call. For example, the built-in |
| 442 |
definition of <function>percentile_disc</function> is equivalent to: |
| 443 |
|
| 444 |
<programlisting> |
| 445 |
CREATE FUNCTION ordered_set_transition(internal, anyelement) |
| 446 |
RETURNS internal ...; |
| 447 |
CREATE FUNCTION percentile_disc_final(internal, float8, anyelement) |
| 448 |
RETURNS anyelement ...; |
| 449 |
|
| 450 |
CREATE AGGREGATE percentile_disc (float8 ORDER BY anyelement) |
| 451 |
( |
| 452 |
sfunc = ordered_set_transition, |
| 453 |
stype = internal, |
| 454 |
finalfunc = percentile_disc_final, |
| 455 |
finalfunc_extra |
| 456 |
); |
| 457 |
</programlisting> |
| 458 |
|
| 459 |
This aggregate takes a <type>float8</type> direct argument (the percentile |
| 460 |
fraction) and an aggregated input that can be of any sortable data type. |
| 461 |
It could be used to obtain a median household income like this: |
| 462 |
|
| 463 |
<programlisting> |
| 464 |
SELECT percentile_disc(0.5) WITHIN GROUP (ORDER BY income) FROM households; |
| 465 |
percentile_disc |
| 466 |
----------------- |
| 467 |
50489 |
| 468 |
</programlisting> |
| 469 |
|
| 470 |
Here, <literal>0.5</literal> is a direct argument; it would make no sense |
| 471 |
for the percentile fraction to be a value varying across rows. |
| 472 |
</para> |
| 473 |
|
| 474 |
<para> |
| 475 |
Unlike the case for normal aggregates, the sorting of input rows for |
| 476 |
an ordered-set aggregate is <emphasis>not</emphasis> done behind the scenes, |
| 477 |
but is the responsibility of the aggregate's support functions. |
| 478 |
The typical implementation approach is to keep a reference to |
| 479 |
a <quote>tuplesort</quote> object in the aggregate's state value, feed the |
| 480 |
incoming rows into that object, and then complete the sorting and |
| 481 |
read out the data in the final function. This design allows the |
| 482 |
final function to perform special operations such as injecting |
| 483 |
additional <quote>hypothetical</quote> rows into the data to be sorted. |
| 484 |
While normal aggregates can often be implemented with support |
| 485 |
functions written in <application>PL/pgSQL</application> or another |
| 486 |
PL language, ordered-set aggregates generally have to be written in |
| 487 |
C, since their state values aren't definable as any SQL data type. |
| 488 |
(In the above example, notice that the state value is declared as |
| 489 |
type <type>internal</type> — this is typical.) |
| 490 |
Also, because the final function performs the sort, it is not possible |
| 491 |
to continue adding input rows by executing the transition function again |
| 492 |
later. This means the final function is not <literal>READ_ONLY</literal>; |
| 493 |
it must be declared in <link linkend="sql-createaggregate"><command>CREATE AGGREGATE</command></link> |
| 494 |
as <literal>READ_WRITE</literal>, or as <literal>SHAREABLE</literal> if |
| 495 |
it's possible for additional final-function calls to make use of the |
| 496 |
already-sorted state. |
| 497 |
</para> |
| 498 |
|
| 499 |
<para> |
| 500 |
The state transition function for an ordered-set aggregate receives |
| 501 |
the current state value plus the aggregated input values for |
| 502 |
each row, and returns the updated state value. This is the |
| 503 |
same definition as for normal aggregates, but note that the direct |
| 504 |
arguments (if any) are not provided. The final function receives |
| 505 |
the last state value, the values of the direct arguments if any, |
| 506 |
and (if <literal>finalfunc_extra</literal> is specified) null values |
| 507 |
corresponding to the aggregated input(s). As with normal |
| 508 |
aggregates, <literal>finalfunc_extra</literal> is only really useful if the |
| 509 |
aggregate is polymorphic; then the extra dummy argument(s) are needed |
| 510 |
to connect the final function's result type to the aggregate's input |
| 511 |
type(s). |
| 512 |
</para> |
| 513 |
|
| 514 |
<para> |
| 515 |
Currently, ordered-set aggregates cannot be used as window functions, |
| 516 |
and therefore there is no need for them to support moving-aggregate mode. |
| 517 |
</para> |
| 518 |
|
| 519 |
</sect2> |
| 520 |
|
| 521 |
<sect2 id="xaggr-partial-aggregates"> |
| 522 |
<title>Partial Aggregation</title> |
| 523 |
|
| 524 |
<indexterm> |
| 525 |
<primary>aggregate function</primary> |
| 526 |
<secondary>partial aggregation</secondary> |
| 527 |
</indexterm> |
| 528 |
|
| 529 |
<para> |
| 530 |
Optionally, an aggregate function can support <firstterm>partial |
| 531 |
aggregation</firstterm>. The idea of partial aggregation is to run the aggregate's |
| 532 |
state transition function over different subsets of the input data |
| 533 |
independently, and then to combine the state values resulting from those |
| 534 |
subsets to produce the same state value that would have resulted from |
| 535 |
scanning all the input in a single operation. This mode can be used for |
| 536 |
parallel aggregation by having different worker processes scan different |
| 537 |
portions of a table. Each worker produces a partial state value, and at |
| 538 |
the end those state values are combined to produce a final state value. |
| 539 |
(In the future this mode might also be used for purposes such as combining |
| 540 |
aggregations over local and remote tables; but that is not implemented |
| 541 |
yet.) |
| 542 |
</para> |
| 543 |
|
| 544 |
<para> |
| 545 |
To support partial aggregation, the aggregate definition must provide |
| 546 |
a <firstterm>combine function</firstterm>, which takes two values of the |
| 547 |
aggregate's state type (representing the results of aggregating over two |
| 548 |
subsets of the input rows) and produces a new value of the state type, |
| 549 |
representing what the state would have been after aggregating over the |
| 550 |
combination of those sets of rows. It is unspecified what the relative |
| 551 |
order of the input rows from the two sets would have been. This means |
| 552 |
that it's usually impossible to define a useful combine function for |
| 553 |
aggregates that are sensitive to input row order. |
| 554 |
</para> |
| 555 |
|
| 556 |
<para> |
| 557 |
As simple examples, <literal>MAX</literal> and <literal>MIN</literal> aggregates can be |
| 558 |
made to support partial aggregation by specifying the combine function as |
| 559 |
the same greater-of-two or lesser-of-two comparison function that is used |
| 560 |
as their transition function. <literal>SUM</literal> aggregates just need an |
| 561 |
addition function as combine function. (Again, this is the same as their |
| 562 |
transition function, unless the state value is wider than the input data |
| 563 |
type.) |
| 564 |
</para> |
| 565 |
|
| 566 |
<para> |
| 567 |
The combine function is treated much like a transition function that |
| 568 |
happens to take a value of the state type, not of the underlying input |
| 569 |
type, as its second argument. In particular, the rules for dealing |
| 570 |
with null values and strict functions are similar. Also, if the aggregate |
| 571 |
definition specifies a non-null <literal>initcond</literal>, keep in mind that |
| 572 |
that will be used not only as the initial state for each partial |
| 573 |
aggregation run, but also as the initial state for the combine function, |
| 574 |
which will be called to combine each partial result into that state. |
| 575 |
</para> |
| 576 |
|
| 577 |
<para> |
| 578 |
If the aggregate's state type is declared as <type>internal</type>, it is |
| 579 |
the combine function's responsibility that its result is allocated in |
| 580 |
the correct memory context for aggregate state values. This means in |
| 581 |
particular that when the first input is <literal>NULL</literal> it's invalid |
| 582 |
to simply return the second input, as that value will be in the wrong |
| 583 |
context and will not have sufficient lifespan. |
| 584 |
</para> |
| 585 |
|
| 586 |
<para> |
| 587 |
When the aggregate's state type is declared as <type>internal</type>, it is |
| 588 |
usually also appropriate for the aggregate definition to provide a |
| 589 |
<firstterm>serialization function</firstterm> and a <firstterm>deserialization |
| 590 |
function</firstterm>, which allow such a state value to be copied from one process |
| 591 |
to another. Without these functions, parallel aggregation cannot be |
| 592 |
performed, and future applications such as local/remote aggregation will |
| 593 |
probably not work either. |
| 594 |
</para> |
| 595 |
|
| 596 |
<para> |
| 597 |
A serialization function must take a single argument of |
| 598 |
type <type>internal</type> and return a result of type <type>bytea</type>, which |
| 599 |
represents the state value packaged up into a flat blob of bytes. |
| 600 |
Conversely, a deserialization function reverses that conversion. It must |
| 601 |
take two arguments of types <type>bytea</type> and <type>internal</type>, and |
| 602 |
return a result of type <type>internal</type>. (The second argument is unused |
| 603 |
and is always zero, but it is required for type-safety reasons.) The |
| 604 |
result of the deserialization function should simply be allocated in the |
| 605 |
current memory context, as unlike the combine function's result, it is not |
| 606 |
long-lived. |
| 607 |
</para> |
| 608 |
|
| 609 |
<para> |
| 610 |
Worth noting also is that for an aggregate to be executed in parallel, |
| 611 |
the aggregate itself must be marked <literal>PARALLEL SAFE</literal>. The |
| 612 |
parallel-safety markings on its support functions are not consulted. |
| 613 |
</para> |
| 614 |
|
| 615 |
</sect2> |
| 616 |
|
| 617 |
<sect2 id="xaggr-support-functions"> |
| 618 |
<title>Support Functions for Aggregates</title> |
| 619 |
|
| 620 |
<indexterm> |
| 621 |
<primary>aggregate function</primary> |
| 622 |
<secondary>support functions for</secondary> |
| 623 |
</indexterm> |
| 624 |
|
| 625 |
<para> |
| 626 |
A function written in C can detect that it is being called as an |
| 627 |
aggregate support function by calling |
| 628 |
<function>AggCheckCallContext</function>, for example: |
| 629 |
<programlisting> |
| 630 |
if (AggCheckCallContext(fcinfo, NULL)) |
| 631 |
</programlisting> |
| 632 |
One reason for checking this is that when it is true, the first input |
| 633 |
must be a temporary state value and can therefore safely be modified |
| 634 |
in-place rather than allocating a new copy. |
| 635 |
See <function>int8inc()</function> for an example. |
| 636 |
(While aggregate transition functions are always allowed to modify |
| 637 |
the transition value in-place, aggregate final functions are generally |
| 638 |
discouraged from doing so; if they do so, the behavior must be declared |
| 639 |
when creating the aggregate. See <xref linkend="sql-createaggregate"/> |
| 640 |
for more detail.) |
| 641 |
</para> |
| 642 |
|
| 643 |
<para> |
| 644 |
The second argument of <function>AggCheckCallContext</function> can be used to |
| 645 |
retrieve the memory context in which aggregate state values are being kept. |
| 646 |
This is useful for transition functions that wish to use <quote>expanded</quote> |
| 647 |
objects (see <xref linkend="xtypes-toast"/>) as their state values. |
| 648 |
On first call, the transition function should return an expanded object |
| 649 |
whose memory context is a child of the aggregate state context, and then |
| 650 |
keep returning the same expanded object on subsequent calls. See |
| 651 |
<function>array_append()</function> for an example. (<function>array_append()</function> |
| 652 |
is not the transition function of any built-in aggregate, but it is written |
| 653 |
to behave efficiently when used as transition function of a custom |
| 654 |
aggregate.) |
| 655 |
</para> |
| 656 |
|
| 657 |
<para> |
| 658 |
Another support routine available to aggregate functions written in C |
| 659 |
is <function>AggGetAggref</function>, which returns the <literal>Aggref</literal> |
| 660 |
parse node that defines the aggregate call. This is mainly useful |
| 661 |
for ordered-set aggregates, which can inspect the substructure of |
| 662 |
the <literal>Aggref</literal> node to find out what sort ordering they are |
| 663 |
supposed to implement. Examples can be found |
| 664 |
in <filename>orderedsetaggs.c</filename> in the <productname>PostgreSQL</productname> |
| 665 |
source code. |
| 666 |
</para> |
| 667 |
|
| 668 |
</sect2> |
| 669 |
|
| 670 |
</sect1> |