Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * nodeAgg.c
4 : : * Routines to handle aggregate nodes.
5 : : *
6 : : * ExecAgg normally evaluates each aggregate in the following steps:
7 : : *
8 : : * transvalue = initcond
9 : : * foreach input_tuple do
10 : : * transvalue = transfunc(transvalue, input_value(s))
11 : : * result = finalfunc(transvalue, direct_argument(s))
12 : : *
13 : : * If a finalfunc is not supplied then the result is just the ending
14 : : * value of transvalue.
15 : : *
16 : : * Other behaviors can be selected by the "aggsplit" mode, which exists
17 : : * to support partial aggregation. It is possible to:
18 : : * * Skip running the finalfunc, so that the output is always the
19 : : * final transvalue state.
20 : : * * Substitute the combinefunc for the transfunc, so that transvalue
21 : : * states (propagated up from a child partial-aggregation step) are merged
22 : : * rather than processing raw input rows. (The statements below about
23 : : * the transfunc apply equally to the combinefunc, when it's selected.)
24 : : * * Apply the serializefunc to the output values (this only makes sense
25 : : * when skipping the finalfunc, since the serializefunc works on the
26 : : * transvalue data type).
27 : : * * Apply the deserializefunc to the input values (this only makes sense
28 : : * when using the combinefunc, for similar reasons).
29 : : * It is the planner's responsibility to connect up Agg nodes using these
30 : : * alternate behaviors in a way that makes sense, with partial aggregation
31 : : * results being fed to nodes that expect them.
32 : : *
33 : : * If a normal aggregate call specifies DISTINCT or ORDER BY, we sort the
34 : : * input tuples and eliminate duplicates (if required) before performing
35 : : * the above-depicted process. (However, we don't do that for ordered-set
36 : : * aggregates; their "ORDER BY" inputs are ordinary aggregate arguments
37 : : * so far as this module is concerned.) Note that partial aggregation
38 : : * is not supported in these cases, since we couldn't ensure global
39 : : * ordering or distinctness of the inputs.
40 : : *
41 : : * If transfunc is marked "strict" in pg_proc and initcond is NULL,
42 : : * then the first non-NULL input_value is assigned directly to transvalue,
43 : : * and transfunc isn't applied until the second non-NULL input_value.
44 : : * The agg's first input type and transtype must be the same in this case!
45 : : *
46 : : * If transfunc is marked "strict" then NULL input_values are skipped,
47 : : * keeping the previous transvalue. If transfunc is not strict then it
48 : : * is called for every input tuple and must deal with NULL initcond
49 : : * or NULL input_values for itself.
50 : : *
51 : : * If finalfunc is marked "strict" then it is not called when the
52 : : * ending transvalue is NULL, instead a NULL result is created
53 : : * automatically (this is just the usual handling of strict functions,
54 : : * of course). A non-strict finalfunc can make its own choice of
55 : : * what to return for a NULL ending transvalue.
56 : : *
57 : : * Ordered-set aggregates are treated specially in one other way: we
58 : : * evaluate any "direct" arguments and pass them to the finalfunc along
59 : : * with the transition value.
60 : : *
61 : : * A finalfunc can have additional arguments beyond the transvalue and
62 : : * any "direct" arguments, corresponding to the input arguments of the
63 : : * aggregate. These are always just passed as NULL. Such arguments may be
64 : : * needed to allow resolution of a polymorphic aggregate's result type.
65 : : *
66 : : * We compute aggregate input expressions and run the transition functions
67 : : * in a temporary econtext (aggstate->tmpcontext). This is reset at least
68 : : * once per input tuple, so when the transvalue datatype is
69 : : * pass-by-reference, we have to be careful to copy it into a longer-lived
70 : : * memory context, and free the prior value to avoid memory leakage. We
71 : : * store transvalues in another set of econtexts, aggstate->aggcontexts
72 : : * (one per grouping set, see below), which are also used for the hashtable
73 : : * structures in AGG_HASHED mode. These econtexts are rescanned, not just
74 : : * reset, at group boundaries so that aggregate transition functions can
75 : : * register shutdown callbacks via AggRegisterCallback.
76 : : *
77 : : * The node's regular econtext (aggstate->ss.ps.ps_ExprContext) is used to
78 : : * run finalize functions and compute the output tuple; this context can be
79 : : * reset once per output tuple.
80 : : *
81 : : * The executor's AggState node is passed as the fmgr "context" value in
82 : : * all transfunc and finalfunc calls. It is not recommended that the
83 : : * transition functions look at the AggState node directly, but they can
84 : : * use AggCheckCallContext() to verify that they are being called by
85 : : * nodeAgg.c (and not as ordinary SQL functions). The main reason a
86 : : * transition function might want to know this is so that it can avoid
87 : : * palloc'ing a fixed-size pass-by-ref transition value on every call:
88 : : * it can instead just scribble on and return its left input. Ordinarily
89 : : * it is completely forbidden for functions to modify pass-by-ref inputs,
90 : : * but in the aggregate case we know the left input is either the initial
91 : : * transition value or a previous function result, and in either case its
92 : : * value need not be preserved. See int8inc() for an example. Notice that
93 : : * the EEOP_AGG_PLAIN_TRANS step is coded to avoid a data copy step when
94 : : * the previous transition value pointer is returned. It is also possible
95 : : * to avoid repeated data copying when the transition value is an expanded
96 : : * object: to do that, the transition function must take care to return
97 : : * an expanded object that is in a child context of the memory context
98 : : * returned by AggCheckCallContext(). Also, some transition functions want
99 : : * to store working state in addition to the nominal transition value; they
100 : : * can use the memory context returned by AggCheckCallContext() to do that.
101 : : *
102 : : * Note: AggCheckCallContext() is available as of PostgreSQL 9.0. The
103 : : * AggState is available as context in earlier releases (back to 8.1),
104 : : * but direct examination of the node is needed to use it before 9.0.
105 : : *
106 : : * As of 9.4, aggregate transition functions can also use AggGetAggref()
107 : : * to get hold of the Aggref expression node for their aggregate call.
108 : : * This is mainly intended for ordered-set aggregates, which are not
109 : : * supported as window functions. (A regular aggregate function would
110 : : * need some fallback logic to use this, since there's no Aggref node
111 : : * for a window function.)
112 : : *
113 : : * Grouping sets:
114 : : *
115 : : * A list of grouping sets which is structurally equivalent to a ROLLUP
116 : : * clause (e.g. (a,b,c), (a,b), (a)) can be processed in a single pass over
117 : : * ordered data. We do this by keeping a separate set of transition values
118 : : * for each grouping set being concurrently processed; for each input tuple
119 : : * we update them all, and on group boundaries we reset those states
120 : : * (starting at the front of the list) whose grouping values have changed
121 : : * (the list of grouping sets is ordered from most specific to least
122 : : * specific).
123 : : *
124 : : * Where more complex grouping sets are used, we break them down into
125 : : * "phases", where each phase has a different sort order (except phase 0
126 : : * which is reserved for hashing). During each phase but the last, the
127 : : * input tuples are additionally stored in a tuplesort which is keyed to the
128 : : * next phase's sort order; during each phase but the first, the input
129 : : * tuples are drawn from the previously sorted data. (The sorting of the
130 : : * data for the first phase is handled by the planner, as it might be
131 : : * satisfied by underlying nodes.)
132 : : *
133 : : * Hashing can be mixed with sorted grouping. To do this, we have an
134 : : * AGG_MIXED strategy that populates the hashtables during the first sorted
135 : : * phase, and switches to reading them out after completing all sort phases.
136 : : * We can also support AGG_HASHED with multiple hash tables and no sorting
137 : : * at all.
138 : : *
139 : : * From the perspective of aggregate transition and final functions, the
140 : : * only issue regarding grouping sets is this: a single call site (flinfo)
141 : : * of an aggregate function may be used for updating several different
142 : : * transition values in turn. So the function must not cache in the flinfo
143 : : * anything which logically belongs as part of the transition value (most
144 : : * importantly, the memory context in which the transition value exists).
145 : : * The support API functions (AggCheckCallContext, AggRegisterCallback) are
146 : : * sensitive to the grouping set for which the aggregate function is
147 : : * currently being called.
148 : : *
149 : : * Plan structure:
150 : : *
151 : : * What we get from the planner is actually one "real" Agg node which is
152 : : * part of the plan tree proper, but which optionally has an additional list
153 : : * of Agg nodes hung off the side via the "chain" field. This is because an
154 : : * Agg node happens to be a convenient representation of all the data we
155 : : * need for grouping sets.
156 : : *
157 : : * For many purposes, we treat the "real" node as if it were just the first
158 : : * node in the chain. The chain must be ordered such that hashed entries
159 : : * come before sorted/plain entries; the real node is marked AGG_MIXED if
160 : : * there are both types present (in which case the real node describes one
161 : : * of the hashed groupings, other AGG_HASHED nodes may optionally follow in
162 : : * the chain, followed in turn by AGG_SORTED or (one) AGG_PLAIN node). If
163 : : * the real node is marked AGG_HASHED or AGG_SORTED, then all the chained
164 : : * nodes must be of the same type; if it is AGG_PLAIN, there can be no
165 : : * chained nodes.
166 : : *
167 : : * We collect all hashed nodes into a single "phase", numbered 0, and create
168 : : * a sorted phase (numbered 1..n) for each AGG_SORTED or AGG_PLAIN node.
169 : : * Phase 0 is allocated even if there are no hashes, but remains unused in
170 : : * that case.
171 : : *
172 : : * AGG_HASHED nodes actually refer to only a single grouping set each,
173 : : * because for each hashed grouping we need a separate grpColIdx and
174 : : * numGroups estimate. AGG_SORTED nodes represent a "rollup", a list of
175 : : * grouping sets that share a sort order. Each AGG_SORTED node other than
176 : : * the first one has an associated Sort node which describes the sort order
177 : : * to be used; the first sorted node takes its input from the outer subtree,
178 : : * which the planner has already arranged to provide ordered data.
179 : : *
180 : : * Memory and ExprContext usage:
181 : : *
182 : : * Because we're accumulating aggregate values across input rows, we need to
183 : : * use more memory contexts than just simple input/output tuple contexts.
184 : : * In fact, for a rollup, we need a separate context for each grouping set
185 : : * so that we can reset the inner (finer-grained) aggregates on their group
186 : : * boundaries while continuing to accumulate values for outer
187 : : * (coarser-grained) groupings. On top of this, we might be simultaneously
188 : : * populating hashtables; however, we only need one context for all the
189 : : * hashtables.
190 : : *
191 : : * So we create an array, aggcontexts, with an ExprContext for each grouping
192 : : * set in the largest rollup that we're going to process, and use the
193 : : * per-tuple memory context of those ExprContexts to store the aggregate
194 : : * transition values. hashcontext is the single context created to support
195 : : * all hash tables.
196 : : *
197 : : * Spilling To Disk
198 : : *
199 : : * When performing hash aggregation, if the hash table memory exceeds the
200 : : * limit (see hash_agg_check_limits()), we enter "spill mode". In spill
201 : : * mode, we advance the transition states only for groups already in the
202 : : * hash table. For tuples that would need to create a new hash table
203 : : * entries (and initialize new transition states), we instead spill them to
204 : : * disk to be processed later. The tuples are spilled in a partitioned
205 : : * manner, so that subsequent batches are smaller and less likely to exceed
206 : : * hash_mem (if a batch does exceed hash_mem, it must be spilled
207 : : * recursively).
208 : : *
209 : : * Spilled data is written to logical tapes. These provide better control
210 : : * over memory usage, disk space, and the number of files than if we were
211 : : * to use a BufFile for each spill. We don't know the number of tapes needed
212 : : * at the start of the algorithm (because it can recurse), so a tape set is
213 : : * allocated at the beginning, and individual tapes are created as needed.
214 : : * As a particular tape is read, logtape.c recycles its disk space. When a
215 : : * tape is read to completion, it is destroyed entirely.
216 : : *
217 : : * Tapes' buffers can take up substantial memory when many tapes are open at
218 : : * once. We only need one tape open at a time in read mode (using a buffer
219 : : * that's a multiple of BLCKSZ); but we need one tape open in write mode (each
220 : : * requiring a buffer of size BLCKSZ) for each partition.
221 : : *
222 : : * Note that it's possible for transition states to start small but then
223 : : * grow very large; for instance in the case of ARRAY_AGG. In such cases,
224 : : * it's still possible to significantly exceed hash_mem. We try to avoid
225 : : * this situation by estimating what will fit in the available memory, and
226 : : * imposing a limit on the number of groups separately from the amount of
227 : : * memory consumed.
228 : : *
229 : : * Transition / Combine function invocation:
230 : : *
231 : : * For performance reasons transition functions, including combine
232 : : * functions, aren't invoked one-by-one from nodeAgg.c after computing
233 : : * arguments using the expression evaluation engine. Instead
234 : : * ExecBuildAggTrans() builds one large expression that does both argument
235 : : * evaluation and transition function invocation. That avoids performance
236 : : * issues due to repeated uses of expression evaluation, complications due
237 : : * to filter expressions having to be evaluated early, and allows to JIT
238 : : * the entire expression into one native function.
239 : : *
240 : : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
241 : : * Portions Copyright (c) 1994, Regents of the University of California
242 : : *
243 : : * IDENTIFICATION
244 : : * src/backend/executor/nodeAgg.c
245 : : *
246 : : *-------------------------------------------------------------------------
247 : : */
248 : :
249 : : #include "postgres.h"
250 : :
251 : : #include "access/htup_details.h"
252 : : #include "access/parallel.h"
253 : : #include "catalog/objectaccess.h"
254 : : #include "catalog/pg_aggregate.h"
255 : : #include "catalog/pg_proc.h"
256 : : #include "catalog/pg_type.h"
257 : : #include "common/hashfn.h"
258 : : #include "executor/execExpr.h"
259 : : #include "executor/executor.h"
260 : : #include "executor/nodeAgg.h"
261 : : #include "lib/hyperloglog.h"
262 : : #include "miscadmin.h"
263 : : #include "nodes/nodeFuncs.h"
264 : : #include "optimizer/optimizer.h"
265 : : #include "parser/parse_agg.h"
266 : : #include "parser/parse_coerce.h"
267 : : #include "utils/acl.h"
268 : : #include "utils/builtins.h"
269 : : #include "utils/datum.h"
270 : : #include "utils/dynahash.h"
271 : : #include "utils/expandeddatum.h"
272 : : #include "utils/injection_point.h"
273 : : #include "utils/logtape.h"
274 : : #include "utils/lsyscache.h"
275 : : #include "utils/memutils.h"
276 : : #include "utils/memutils_memorychunk.h"
277 : : #include "utils/syscache.h"
278 : : #include "utils/tuplesort.h"
279 : :
280 : : /*
281 : : * Control how many partitions are created when spilling HashAgg to
282 : : * disk.
283 : : *
284 : : * HASHAGG_PARTITION_FACTOR is multiplied by the estimated number of
285 : : * partitions needed such that each partition will fit in memory. The factor
286 : : * is set higher than one because there's not a high cost to having a few too
287 : : * many partitions, and it makes it less likely that a partition will need to
288 : : * be spilled recursively. Another benefit of having more, smaller partitions
289 : : * is that small hash tables may perform better than large ones due to memory
290 : : * caching effects.
291 : : *
292 : : * We also specify a min and max number of partitions per spill. Too few might
293 : : * mean a lot of wasted I/O from repeated spilling of the same tuples. Too
294 : : * many will result in lots of memory wasted buffering the spill files (which
295 : : * could instead be spent on a larger hash table).
296 : : */
297 : : #define HASHAGG_PARTITION_FACTOR 1.50
298 : : #define HASHAGG_MIN_PARTITIONS 4
299 : : #define HASHAGG_MAX_PARTITIONS 1024
300 : :
301 : : /*
302 : : * For reading from tapes, the buffer size must be a multiple of
303 : : * BLCKSZ. Larger values help when reading from multiple tapes concurrently,
304 : : * but that doesn't happen in HashAgg, so we simply use BLCKSZ. Writing to a
305 : : * tape always uses a buffer of size BLCKSZ.
306 : : */
307 : : #define HASHAGG_READ_BUFFER_SIZE BLCKSZ
308 : : #define HASHAGG_WRITE_BUFFER_SIZE BLCKSZ
309 : :
310 : : /*
311 : : * HyperLogLog is used for estimating the cardinality of the spilled tuples in
312 : : * a given partition. 5 bits corresponds to a size of about 32 bytes and a
313 : : * worst-case error of around 18%. That's effective enough to choose a
314 : : * reasonable number of partitions when recursing.
315 : : */
316 : : #define HASHAGG_HLL_BIT_WIDTH 5
317 : :
318 : : /*
319 : : * Assume the palloc overhead always uses sizeof(MemoryChunk) bytes.
320 : : */
321 : : #define CHUNKHDRSZ sizeof(MemoryChunk)
322 : :
323 : : /*
324 : : * Represents partitioned spill data for a single hashtable. Contains the
325 : : * necessary information to route tuples to the correct partition, and to
326 : : * transform the spilled data into new batches.
327 : : *
328 : : * The high bits are used for partition selection (when recursing, we ignore
329 : : * the bits that have already been used for partition selection at an earlier
330 : : * level).
331 : : */
332 : : typedef struct HashAggSpill
333 : : {
334 : : int npartitions; /* number of partitions */
335 : : LogicalTape **partitions; /* spill partition tapes */
336 : : int64 *ntuples; /* number of tuples in each partition */
337 : : uint32 mask; /* mask to find partition from hash value */
338 : : int shift; /* after masking, shift by this amount */
339 : : hyperLogLogState *hll_card; /* cardinality estimate for contents */
340 : : } HashAggSpill;
341 : :
342 : : /*
343 : : * Represents work to be done for one pass of hash aggregation (with only one
344 : : * grouping set).
345 : : *
346 : : * Also tracks the bits of the hash already used for partition selection by
347 : : * earlier iterations, so that this batch can use new bits. If all bits have
348 : : * already been used, no partitioning will be done (any spilled data will go
349 : : * to a single output tape).
350 : : */
351 : : typedef struct HashAggBatch
352 : : {
353 : : int setno; /* grouping set */
354 : : int used_bits; /* number of bits of hash already used */
355 : : LogicalTape *input_tape; /* input partition tape */
356 : : int64 input_tuples; /* number of tuples in this batch */
357 : : double input_card; /* estimated group cardinality */
358 : : } HashAggBatch;
359 : :
360 : : /* used to find referenced colnos */
361 : : typedef struct FindColsContext
362 : : {
363 : : bool is_aggref; /* is under an aggref */
364 : : Bitmapset *aggregated; /* column references under an aggref */
365 : : Bitmapset *unaggregated; /* other column references */
366 : : } FindColsContext;
367 : :
368 : : static void select_current_set(AggState *aggstate, int setno, bool is_hash);
369 : : static void initialize_phase(AggState *aggstate, int newphase);
370 : : static TupleTableSlot *fetch_input_tuple(AggState *aggstate);
371 : : static void initialize_aggregates(AggState *aggstate,
372 : : AggStatePerGroup *pergroups,
373 : : int numReset);
374 : : static void advance_transition_function(AggState *aggstate,
375 : : AggStatePerTrans pertrans,
376 : : AggStatePerGroup pergroupstate);
377 : : static void advance_aggregates(AggState *aggstate);
378 : : static void process_ordered_aggregate_single(AggState *aggstate,
379 : : AggStatePerTrans pertrans,
380 : : AggStatePerGroup pergroupstate);
381 : : static void process_ordered_aggregate_multi(AggState *aggstate,
382 : : AggStatePerTrans pertrans,
383 : : AggStatePerGroup pergroupstate);
384 : : static void finalize_aggregate(AggState *aggstate,
385 : : AggStatePerAgg peragg,
386 : : AggStatePerGroup pergroupstate,
387 : : Datum *resultVal, bool *resultIsNull);
388 : : static void finalize_partialaggregate(AggState *aggstate,
389 : : AggStatePerAgg peragg,
390 : : AggStatePerGroup pergroupstate,
391 : : Datum *resultVal, bool *resultIsNull);
392 : : static inline void prepare_hash_slot(AggStatePerHash perhash,
393 : : TupleTableSlot *inputslot,
394 : : TupleTableSlot *hashslot);
395 : : static void prepare_projection_slot(AggState *aggstate,
396 : : TupleTableSlot *slot,
397 : : int currentSet);
398 : : static void finalize_aggregates(AggState *aggstate,
399 : : AggStatePerAgg peraggs,
400 : : AggStatePerGroup pergroup);
401 : : static TupleTableSlot *project_aggregates(AggState *aggstate);
402 : : static void find_cols(AggState *aggstate, Bitmapset **aggregated,
403 : : Bitmapset **unaggregated);
404 : : static bool find_cols_walker(Node *node, FindColsContext *context);
405 : : static void build_hash_tables(AggState *aggstate);
406 : : static void build_hash_table(AggState *aggstate, int setno, long nbuckets);
407 : : static void hashagg_recompile_expressions(AggState *aggstate, bool minslot,
408 : : bool nullcheck);
409 : : static void hash_create_memory(AggState *aggstate);
410 : : static long hash_choose_num_buckets(double hashentrysize,
411 : : long ngroups, Size memory);
412 : : static int hash_choose_num_partitions(double input_groups,
413 : : double hashentrysize,
414 : : int used_bits,
415 : : int *log2_npartitions);
416 : : static void initialize_hash_entry(AggState *aggstate,
417 : : TupleHashTable hashtable,
418 : : TupleHashEntry entry);
419 : : static void lookup_hash_entries(AggState *aggstate);
420 : : static TupleTableSlot *agg_retrieve_direct(AggState *aggstate);
421 : : static void agg_fill_hash_table(AggState *aggstate);
422 : : static bool agg_refill_hash_table(AggState *aggstate);
423 : : static TupleTableSlot *agg_retrieve_hash_table(AggState *aggstate);
424 : : static TupleTableSlot *agg_retrieve_hash_table_in_memory(AggState *aggstate);
425 : : static void hash_agg_check_limits(AggState *aggstate);
426 : : static void hash_agg_enter_spill_mode(AggState *aggstate);
427 : : static void hash_agg_update_metrics(AggState *aggstate, bool from_tape,
428 : : int npartitions);
429 : : static void hashagg_finish_initial_spills(AggState *aggstate);
430 : : static void hashagg_reset_spill_state(AggState *aggstate);
431 : : static HashAggBatch *hashagg_batch_new(LogicalTape *input_tape, int setno,
432 : : int64 input_tuples, double input_card,
433 : : int used_bits);
434 : : static MinimalTuple hashagg_batch_read(HashAggBatch *batch, uint32 *hashp);
435 : : static void hashagg_spill_init(HashAggSpill *spill, LogicalTapeSet *tapeset,
436 : : int used_bits, double input_groups,
437 : : double hashentrysize);
438 : : static Size hashagg_spill_tuple(AggState *aggstate, HashAggSpill *spill,
439 : : TupleTableSlot *inputslot, uint32 hash);
440 : : static void hashagg_spill_finish(AggState *aggstate, HashAggSpill *spill,
441 : : int setno);
442 : : static Datum GetAggInitVal(Datum textInitVal, Oid transtype);
443 : : static void build_pertrans_for_aggref(AggStatePerTrans pertrans,
444 : : AggState *aggstate, EState *estate,
445 : : Aggref *aggref, Oid transfn_oid,
446 : : Oid aggtranstype, Oid aggserialfn,
447 : : Oid aggdeserialfn, Datum initValue,
448 : : bool initValueIsNull, Oid *inputTypes,
449 : : int numArguments);
450 : :
451 : :
452 : : /*
453 : : * Select the current grouping set; affects current_set and
454 : : * curaggcontext.
455 : : */
456 : : static void
3085 rhodiumtoad@postgres 457 :CBC 3440309 : select_current_set(AggState *aggstate, int setno, bool is_hash)
458 : : {
459 : : /*
460 : : * When changing this, also adapt ExecAggPlainTransByVal() and
461 : : * ExecAggPlainTransByRef().
462 : : */
463 [ + + ]: 3440309 : if (is_hash)
464 : 3107469 : aggstate->curaggcontext = aggstate->hashcontext;
465 : : else
466 : 332840 : aggstate->curaggcontext = aggstate->aggcontexts[setno];
467 : :
468 : 3440309 : aggstate->current_set = setno;
469 : 3440309 : }
470 : :
471 : : /*
472 : : * Switch to phase "newphase", which must either be 0 or 1 (to reset) or
473 : : * current_phase + 1. Juggle the tuplesorts accordingly.
474 : : *
475 : : * Phase 0 is for hashing, which we currently handle last in the AGG_MIXED
476 : : * case, so when entering phase 0, all we need to do is drop open sorts.
477 : : */
478 : : static void
3766 andres@anarazel.de 479 : 42785 : initialize_phase(AggState *aggstate, int newphase)
480 : : {
3085 rhodiumtoad@postgres 481 [ + + - + ]: 42785 : Assert(newphase <= 1 || newphase == aggstate->current_phase + 1);
482 : :
483 : : /*
484 : : * Whatever the previous state, we're now done with whatever input
485 : : * tuplesort was in use.
486 : : */
3766 andres@anarazel.de 487 [ + + ]: 42785 : if (aggstate->sort_in)
488 : : {
489 : 21 : tuplesort_end(aggstate->sort_in);
490 : 21 : aggstate->sort_in = NULL;
491 : : }
492 : :
3085 rhodiumtoad@postgres 493 [ + + ]: 42785 : if (newphase <= 1)
494 : : {
495 : : /*
496 : : * Discard any existing output tuplesort.
497 : : */
3766 andres@anarazel.de 498 [ + + ]: 42686 : if (aggstate->sort_out)
499 : : {
500 : 3 : tuplesort_end(aggstate->sort_out);
501 : 3 : aggstate->sort_out = NULL;
502 : : }
503 : : }
504 : : else
505 : : {
506 : : /*
507 : : * The old output tuplesort becomes the new input one, and this is the
508 : : * right time to actually sort it.
509 : : */
510 : 99 : aggstate->sort_in = aggstate->sort_out;
511 : 99 : aggstate->sort_out = NULL;
512 [ - + ]: 99 : Assert(aggstate->sort_in);
513 : 99 : tuplesort_performsort(aggstate->sort_in);
514 : : }
515 : :
516 : : /*
517 : : * If this isn't the last phase, we need to sort appropriately for the
518 : : * next phase in sequence.
519 : : */
3085 rhodiumtoad@postgres 520 [ + + + + ]: 42785 : if (newphase > 0 && newphase < aggstate->numphases - 1)
521 : : {
3759 bruce@momjian.us 522 : 123 : Sort *sortnode = aggstate->phases[newphase + 1].sortnode;
3766 andres@anarazel.de 523 : 123 : PlanState *outerNode = outerPlanState(aggstate);
524 : 123 : TupleDesc tupDesc = ExecGetResultType(outerNode);
525 : :
526 : 123 : aggstate->sort_out = tuplesort_begin_heap(tupDesc,
527 : : sortnode->numCols,
528 : : sortnode->sortColIdx,
529 : : sortnode->sortOperators,
530 : : sortnode->collations,
531 : : sortnode->nullsFirst,
532 : : work_mem,
533 : : NULL, TUPLESORT_NONE);
534 : : }
535 : :
536 : 42785 : aggstate->current_phase = newphase;
537 : 42785 : aggstate->phase = &aggstate->phases[newphase];
538 : 42785 : }
539 : :
540 : : /*
541 : : * Fetch a tuple from either the outer plan (for phase 1) or from the sorter
542 : : * populated by the previous phase. Copy it to the sorter for the next phase
543 : : * if any.
544 : : *
545 : : * Callers cannot rely on memory for tuple in returned slot remaining valid
546 : : * past any subsequently fetched tuple.
547 : : */
548 : : static TupleTableSlot *
549 : 13657713 : fetch_input_tuple(AggState *aggstate)
550 : : {
551 : : TupleTableSlot *slot;
552 : :
553 [ + + ]: 13657713 : if (aggstate->sort_in)
554 : : {
555 : : /* make sure we check for interrupts in either path through here */
2965 556 [ - + ]: 147447 : CHECK_FOR_INTERRUPTS();
3075 557 [ + + ]: 147447 : if (!tuplesort_gettupleslot(aggstate->sort_in, true, false,
558 : : aggstate->sort_slot, NULL))
3766 559 : 99 : return NULL;
560 : 147348 : slot = aggstate->sort_slot;
561 : : }
562 : : else
563 : 13510266 : slot = ExecProcNode(outerPlanState(aggstate));
564 : :
565 [ + + + + : 13657580 : if (!TupIsNull(slot) && aggstate->sort_out)
+ + ]
566 : 147348 : tuplesort_puttupleslot(aggstate->sort_out, slot);
567 : :
568 : 13657580 : return slot;
569 : : }
570 : :
571 : : /*
572 : : * (Re)Initialize an individual aggregate.
573 : : *
574 : : * This function handles only one grouping set, already set in
575 : : * aggstate->current_set.
576 : : *
577 : : * When called, CurrentMemoryContext should be the per-query context.
578 : : */
579 : : static void
3686 heikki.linnakangas@i 580 : 562948 : initialize_aggregate(AggState *aggstate, AggStatePerTrans pertrans,
581 : : AggStatePerGroup pergroupstate)
582 : : {
583 : : /*
584 : : * Start a fresh sort operation for each DISTINCT/ORDER BY aggregate.
585 : : */
1131 drowley@postgresql.o 586 [ + + ]: 562948 : if (pertrans->aggsortrequired)
587 : : {
588 : : /*
589 : : * In case of rescan, maybe there could be an uncompleted sort
590 : : * operation? Clean it up if so.
591 : : */
3686 heikki.linnakangas@i 592 [ - + ]: 26921 : if (pertrans->sortstates[aggstate->current_set])
3686 heikki.linnakangas@i 593 :UBC 0 : tuplesort_end(pertrans->sortstates[aggstate->current_set]);
594 : :
595 : :
596 : : /*
597 : : * We use a plain Datum sorter when there's a single input column;
598 : : * otherwise sort the full tuple. (See comments for
599 : : * process_ordered_aggregate_single.)
600 : : */
3686 heikki.linnakangas@i 601 [ + + ]:CBC 26921 : if (pertrans->numInputs == 1)
602 : : {
2939 andres@anarazel.de 603 : 26879 : Form_pg_attribute attr = TupleDescAttr(pertrans->sortdesc, 0);
604 : :
3686 heikki.linnakangas@i 605 : 26879 : pertrans->sortstates[aggstate->current_set] =
2939 andres@anarazel.de 606 : 26879 : tuplesort_begin_datum(attr->atttypid,
3686 heikki.linnakangas@i 607 : 26879 : pertrans->sortOperators[0],
608 : 26879 : pertrans->sortCollations[0],
609 : 26879 : pertrans->sortNullsFirst[0],
610 : : work_mem, NULL, TUPLESORT_NONE);
611 : : }
612 : : else
613 : 42 : pertrans->sortstates[aggstate->current_set] =
3202 andres@anarazel.de 614 : 42 : tuplesort_begin_heap(pertrans->sortdesc,
615 : : pertrans->numSortCols,
616 : : pertrans->sortColIdx,
617 : : pertrans->sortOperators,
618 : : pertrans->sortCollations,
619 : : pertrans->sortNullsFirst,
620 : : work_mem, NULL, TUPLESORT_NONE);
621 : : }
622 : :
623 : : /*
624 : : * (Re)set transValue to the initial value.
625 : : *
626 : : * Note that when the initial value is pass-by-ref, we must copy it (into
627 : : * the aggcontext) since we will pfree the transValue later.
628 : : */
3686 heikki.linnakangas@i 629 [ + + ]: 562948 : if (pertrans->initValueIsNull)
630 : 296618 : pergroupstate->transValue = pertrans->initValue;
631 : : else
632 : : {
633 : : MemoryContext oldContext;
634 : :
2046 alvherre@alvh.no-ip. 635 : 266330 : oldContext = MemoryContextSwitchTo(aggstate->curaggcontext->ecxt_per_tuple_memory);
3686 heikki.linnakangas@i 636 : 532660 : pergroupstate->transValue = datumCopy(pertrans->initValue,
637 : 266330 : pertrans->transtypeByVal,
638 : 266330 : pertrans->transtypeLen);
3766 andres@anarazel.de 639 : 266330 : MemoryContextSwitchTo(oldContext);
640 : : }
3686 heikki.linnakangas@i 641 : 562948 : pergroupstate->transValueIsNull = pertrans->initValueIsNull;
642 : :
643 : : /*
644 : : * If the initial value for the transition state doesn't exist in the
645 : : * pg_aggregate table then we will let the first non-NULL value returned
646 : : * from the outer procNode become the initial value. (This is useful for
647 : : * aggregates like max() and min().) The noTransValue flag signals that we
648 : : * still need to do this.
649 : : */
650 : 562948 : pergroupstate->noTransValue = pertrans->initValueIsNull;
3766 andres@anarazel.de 651 : 562948 : }
652 : :
653 : : /*
654 : : * Initialize all aggregate transition states for a new group of input values.
655 : : *
656 : : * If there are multiple grouping sets, we initialize only the first numReset
657 : : * of them (the grouping sets are ordered so that the most specific one, which
658 : : * is reset most often, is first). As a convenience, if numReset is 0, we
659 : : * reinitialize all sets.
660 : : *
661 : : * NB: This cannot be used for hash aggregates, as for those the grouping set
662 : : * number has to be specified from further up.
663 : : *
664 : : * When called, CurrentMemoryContext should be the per-query context.
665 : : */
666 : : static void
667 : 149703 : initialize_aggregates(AggState *aggstate,
668 : : AggStatePerGroup *pergroups,
669 : : int numReset)
670 : : {
671 : : int transno;
3759 bruce@momjian.us 672 : 149703 : int numGroupingSets = Max(aggstate->phase->numsets, 1);
673 : 149703 : int setno = 0;
3085 rhodiumtoad@postgres 674 : 149703 : int numTrans = aggstate->numtrans;
3686 heikki.linnakangas@i 675 : 149703 : AggStatePerTrans transstates = aggstate->pertrans;
676 : :
3085 rhodiumtoad@postgres 677 [ - + ]: 149703 : if (numReset == 0)
3766 andres@anarazel.de 678 :UBC 0 : numReset = numGroupingSets;
679 : :
2804 andres@anarazel.de 680 [ + + ]:CBC 306487 : for (setno = 0; setno < numReset; setno++)
681 : : {
682 : 156784 : AggStatePerGroup pergroup = pergroups[setno];
683 : :
684 : 156784 : select_current_set(aggstate, setno, false);
685 : :
686 [ + + ]: 489595 : for (transno = 0; transno < numTrans; transno++)
687 : : {
688 : 332811 : AggStatePerTrans pertrans = &transstates[transno];
689 : 332811 : AggStatePerGroup pergroupstate = &pergroup[transno];
690 : :
691 : 332811 : initialize_aggregate(aggstate, pertrans, pergroupstate);
692 : : }
693 : : }
9399 tgl@sss.pgh.pa.us 694 : 149703 : }
695 : :
696 : : /*
697 : : * Given new input value(s), advance the transition function of one aggregate
698 : : * state within one grouping set only (already set in aggstate->current_set)
699 : : *
700 : : * The new values (and null flags) have been preloaded into argument positions
701 : : * 1 and up in pertrans->transfn_fcinfo, so that we needn't copy them again to
702 : : * pass to the transition function. We also expect that the static fields of
703 : : * the fcinfo are already initialized; that was done by ExecInitAgg().
704 : : *
705 : : * It doesn't matter which memory context this is called in.
706 : : */
707 : : static void
8340 708 : 362149 : advance_transition_function(AggState *aggstate,
709 : : AggStatePerTrans pertrans,
710 : : AggStatePerGroup pergroupstate)
711 : : {
2415 andres@anarazel.de 712 : 362149 : FunctionCallInfo fcinfo = pertrans->transfn_fcinfo;
713 : : MemoryContext oldContext;
714 : : Datum newVal;
715 : :
3686 heikki.linnakangas@i 716 [ + + ]: 362149 : if (pertrans->transfn.fn_strict)
717 : : {
718 : : /*
719 : : * For a strict transfn, nothing happens when there's a NULL input; we
720 : : * just keep the prior transValue.
721 : : */
722 : 112500 : int numTransInputs = pertrans->numTransInputs;
723 : : int i;
724 : :
4275 tgl@sss.pgh.pa.us 725 [ + + ]: 225000 : for (i = 1; i <= numTransInputs; i++)
726 : : {
2415 andres@anarazel.de 727 [ - + ]: 112500 : if (fcinfo->args[i].isnull)
6981 tgl@sss.pgh.pa.us 728 :UBC 0 : return;
729 : : }
8340 tgl@sss.pgh.pa.us 730 [ - + ]:CBC 112500 : if (pergroupstate->noTransValue)
731 : : {
732 : : /*
733 : : * transValue has not been initialized. This is the first non-NULL
734 : : * input value. We use it as the initial value for transValue. (We
735 : : * already checked that the agg's input type is binary-compatible
736 : : * with its transtype, so straight copy here is OK.)
737 : : *
738 : : * We must copy the datum into aggcontext if it is pass-by-ref. We
739 : : * do not need to pfree the old transValue, since it's NULL.
740 : : */
2046 alvherre@alvh.no-ip. 741 :UBC 0 : oldContext = MemoryContextSwitchTo(aggstate->curaggcontext->ecxt_per_tuple_memory);
2415 andres@anarazel.de 742 : 0 : pergroupstate->transValue = datumCopy(fcinfo->args[1].value,
3686 heikki.linnakangas@i 743 : 0 : pertrans->transtypeByVal,
744 : 0 : pertrans->transtypeLen);
8340 tgl@sss.pgh.pa.us 745 : 0 : pergroupstate->transValueIsNull = false;
746 : 0 : pergroupstate->noTransValue = false;
747 : 0 : MemoryContextSwitchTo(oldContext);
9182 748 : 0 : return;
749 : : }
8340 tgl@sss.pgh.pa.us 750 [ - + ]:CBC 112500 : if (pergroupstate->transValueIsNull)
751 : : {
752 : : /*
753 : : * Don't call a strict function with NULL inputs. Note it is
754 : : * possible to get here despite the above tests, if the transfn is
755 : : * strict *and* returned a NULL on a prior cycle. If that happens
756 : : * we will propagate the NULL all the way to the end.
757 : : */
9182 tgl@sss.pgh.pa.us 758 :UBC 0 : return;
759 : : }
760 : : }
761 : :
762 : : /* We run the transition functions in per-input-tuple memory context */
8340 tgl@sss.pgh.pa.us 763 :CBC 362149 : oldContext = MemoryContextSwitchTo(aggstate->tmpcontext->ecxt_per_tuple_memory);
764 : :
765 : : /* set up aggstate->curpertrans for AggGetAggref() */
3686 heikki.linnakangas@i 766 : 362149 : aggstate->curpertrans = pertrans;
767 : :
768 : : /*
769 : : * OK to call the transition function
770 : : */
2415 andres@anarazel.de 771 : 362149 : fcinfo->args[0].value = pergroupstate->transValue;
772 : 362149 : fcinfo->args[0].isnull = pergroupstate->transValueIsNull;
4259 tgl@sss.pgh.pa.us 773 : 362149 : fcinfo->isnull = false; /* just in case transfn doesn't set it */
774 : :
6981 775 : 362149 : newVal = FunctionCallInvoke(fcinfo);
776 : :
3686 heikki.linnakangas@i 777 : 362149 : aggstate->curpertrans = NULL;
778 : :
779 : : /*
780 : : * If pass-by-ref datatype, must copy the new value into aggcontext and
781 : : * free the prior transValue. But if transfn returned a pointer to its
782 : : * first input, we don't need to do anything.
783 : : *
784 : : * It's safe to compare newVal with pergroup->transValue without regard
785 : : * for either being NULL, because ExecAggCopyTransValue takes care to set
786 : : * transValue to 0 when NULL. Otherwise we could end up accidentally not
787 : : * reparenting, when the transValue has the same numerical value as
788 : : * newValue, despite being NULL. This is a somewhat hot path, making it
789 : : * undesirable to instead solve this with another branch for the common
790 : : * case of the transition function returning its (modified) input
791 : : * argument.
792 : : */
793 [ - + - - ]: 362149 : if (!pertrans->transtypeByVal &&
7266 bruce@momjian.us 794 :UBC 0 : DatumGetPointer(newVal) != DatumGetPointer(pergroupstate->transValue))
866 tgl@sss.pgh.pa.us 795 : 0 : newVal = ExecAggCopyTransValue(aggstate, pertrans,
796 : 0 : newVal, fcinfo->isnull,
797 : : pergroupstate->transValue,
798 : 0 : pergroupstate->transValueIsNull);
799 : :
8340 tgl@sss.pgh.pa.us 800 :CBC 362149 : pergroupstate->transValue = newVal;
6981 801 : 362149 : pergroupstate->transValueIsNull = fcinfo->isnull;
802 : :
8340 803 : 362149 : MemoryContextSwitchTo(oldContext);
804 : : }
805 : :
806 : : /*
807 : : * Advance each aggregate transition state for one input tuple. The input
808 : : * tuple has been stored in tmpcontext->ecxt_outertuple, so that it is
809 : : * accessible to ExecEvalExpr.
810 : : *
811 : : * We have two sets of transition states to handle: one for sorted aggregation
812 : : * and one for hashed; we do them both here, to avoid multiple evaluation of
813 : : * the inputs.
814 : : *
815 : : * When called, CurrentMemoryContext should be the per-query context.
816 : : */
817 : : static void
2797 andres@anarazel.de 818 : 13996417 : advance_aggregates(AggState *aggstate)
819 : : {
179 dgustafsson@postgres 820 : 13996417 : ExecEvalExprNoReturnSwitchContext(aggstate->phase->evaltrans,
821 : : aggstate->tmpcontext);
8340 tgl@sss.pgh.pa.us 822 : 13996378 : }
823 : :
824 : : /*
825 : : * Run the transition function for a DISTINCT or ORDER BY aggregate
826 : : * with only one input. This is called after we have completed
827 : : * entering all the input values into the sort object. We complete the
828 : : * sort, read out the values in sorted order, and run the transition
829 : : * function on each value (applying DISTINCT if appropriate).
830 : : *
831 : : * Note that the strictness of the transition function was checked when
832 : : * entering the values into the sort, so we don't check it again here;
833 : : * we just apply standard SQL DISTINCT logic.
834 : : *
835 : : * The one-input case is handled separately from the multi-input case
836 : : * for performance reasons: for single by-value inputs, such as the
837 : : * common case of count(distinct id), the tuplesort_getdatum code path
838 : : * is around 300% faster. (The speedup for by-reference types is less
839 : : * but still noticeable.)
840 : : *
841 : : * This function handles only one grouping set (already set in
842 : : * aggstate->current_set).
843 : : *
844 : : * When called, CurrentMemoryContext should be the per-query context.
845 : : */
846 : : static void
5744 847 : 26879 : process_ordered_aggregate_single(AggState *aggstate,
848 : : AggStatePerTrans pertrans,
849 : : AggStatePerGroup pergroupstate)
850 : : {
9187 851 : 26879 : Datum oldVal = (Datum) 0;
5671 bruce@momjian.us 852 : 26879 : bool oldIsNull = true;
9187 tgl@sss.pgh.pa.us 853 : 26879 : bool haveOldVal = false;
8340 854 : 26879 : MemoryContext workcontext = aggstate->tmpcontext->ecxt_per_tuple_memory;
855 : : MemoryContext oldContext;
3686 heikki.linnakangas@i 856 : 26879 : bool isDistinct = (pertrans->numDistinctCols > 0);
3489 rhaas@postgresql.org 857 : 26879 : Datum newAbbrevVal = (Datum) 0;
858 : 26879 : Datum oldAbbrevVal = (Datum) 0;
2415 andres@anarazel.de 859 : 26879 : FunctionCallInfo fcinfo = pertrans->transfn_fcinfo;
860 : : Datum *newVal;
861 : : bool *isNull;
862 : :
3686 heikki.linnakangas@i 863 [ - + ]: 26879 : Assert(pertrans->numDistinctCols < 2);
864 : :
865 : 26879 : tuplesort_performsort(pertrans->sortstates[aggstate->current_set]);
866 : :
867 : : /* Load the column into argument 1 (arg 0 will be transition value) */
2415 andres@anarazel.de 868 : 26879 : newVal = &fcinfo->args[1].value;
869 : 26879 : isNull = &fcinfo->args[1].isnull;
870 : :
871 : : /*
872 : : * Note: if input type is pass-by-ref, the datums returned by the sort are
873 : : * freshly palloc'd in the per-query context, so we must be careful to
874 : : * pfree them when they are no longer needed.
875 : : */
876 : :
3686 heikki.linnakangas@i 877 [ + + ]: 449138 : while (tuplesort_getdatum(pertrans->sortstates[aggstate->current_set],
878 : : true, false, newVal, isNull, &newAbbrevVal))
879 : : {
880 : : /*
881 : : * Clear and select the working context for evaluation of the equality
882 : : * function and transition function.
883 : : */
8340 tgl@sss.pgh.pa.us 884 : 422259 : MemoryContextReset(workcontext);
885 : 422259 : oldContext = MemoryContextSwitchTo(workcontext);
886 : :
887 : : /*
888 : : * If DISTINCT mode, and not distinct from prior, skip it.
889 : : */
5744 890 [ + + + + ]: 422259 : if (isDistinct &&
891 [ - + ]: 155227 : haveOldVal &&
5744 tgl@sss.pgh.pa.us 892 [ # # ]:UBC 0 : ((oldIsNull && *isNull) ||
5744 tgl@sss.pgh.pa.us 893 [ + - + - ]:CBC 155227 : (!oldIsNull && !*isNull &&
3489 rhaas@postgresql.org 894 [ + + + + ]: 302974 : oldAbbrevVal == newAbbrevVal &&
2360 peter@eisentraut.org 895 : 147747 : DatumGetBool(FunctionCall2Coll(&pertrans->equalfnOne,
896 : : pertrans->aggCollation,
897 : : oldVal, *newVal)))))
898 : : {
1044 drowley@postgresql.o 899 : 60218 : MemoryContextSwitchTo(oldContext);
900 : 60218 : continue;
901 : : }
902 : : else
903 : : {
3686 heikki.linnakangas@i 904 : 362041 : advance_transition_function(aggstate, pertrans, pergroupstate);
905 : :
1044 drowley@postgresql.o 906 : 362041 : MemoryContextSwitchTo(oldContext);
907 : :
908 : : /*
909 : : * Forget the old value, if any, and remember the new one for
910 : : * subsequent equality checks.
911 : : */
912 [ + + ]: 362041 : if (!pertrans->inputtypeByVal)
913 : : {
914 [ + + ]: 262644 : if (!oldIsNull)
915 : 262554 : pfree(DatumGetPointer(oldVal));
916 [ + + ]: 262644 : if (!*isNull)
917 : 262614 : oldVal = datumCopy(*newVal, pertrans->inputtypeByVal,
918 : 262614 : pertrans->inputtypeLen);
919 : : }
920 : : else
921 : 99397 : oldVal = *newVal;
3489 rhaas@postgresql.org 922 : 362041 : oldAbbrevVal = newAbbrevVal;
5744 tgl@sss.pgh.pa.us 923 : 362041 : oldIsNull = *isNull;
9399 924 : 362041 : haveOldVal = true;
925 : : }
926 : : }
927 : :
3686 heikki.linnakangas@i 928 [ + + + + ]: 26879 : if (!oldIsNull && !pertrans->inputtypeByVal)
9187 tgl@sss.pgh.pa.us 929 : 60 : pfree(DatumGetPointer(oldVal));
930 : :
3686 heikki.linnakangas@i 931 : 26879 : tuplesort_end(pertrans->sortstates[aggstate->current_set]);
932 : 26879 : pertrans->sortstates[aggstate->current_set] = NULL;
9187 tgl@sss.pgh.pa.us 933 : 26879 : }
934 : :
935 : : /*
936 : : * Run the transition function for a DISTINCT or ORDER BY aggregate
937 : : * with more than one input. This is called after we have completed
938 : : * entering all the input values into the sort object. We complete the
939 : : * sort, read out the values in sorted order, and run the transition
940 : : * function on each value (applying DISTINCT if appropriate).
941 : : *
942 : : * This function handles only one grouping set (already set in
943 : : * aggstate->current_set).
944 : : *
945 : : * When called, CurrentMemoryContext should be the per-query context.
946 : : */
947 : : static void
5744 948 : 42 : process_ordered_aggregate_multi(AggState *aggstate,
949 : : AggStatePerTrans pertrans,
950 : : AggStatePerGroup pergroupstate)
951 : : {
2760 andres@anarazel.de 952 : 42 : ExprContext *tmpcontext = aggstate->tmpcontext;
2415 953 : 42 : FunctionCallInfo fcinfo = pertrans->transfn_fcinfo;
3202 954 : 42 : TupleTableSlot *slot1 = pertrans->sortslot;
3686 heikki.linnakangas@i 955 : 42 : TupleTableSlot *slot2 = pertrans->uniqslot;
956 : 42 : int numTransInputs = pertrans->numTransInputs;
957 : 42 : int numDistinctCols = pertrans->numDistinctCols;
3489 rhaas@postgresql.org 958 : 42 : Datum newAbbrevVal = (Datum) 0;
959 : 42 : Datum oldAbbrevVal = (Datum) 0;
5671 bruce@momjian.us 960 : 42 : bool haveOldValue = false;
2760 andres@anarazel.de 961 : 42 : TupleTableSlot *save = aggstate->tmpcontext->ecxt_outertuple;
962 : : int i;
963 : :
3686 heikki.linnakangas@i 964 : 42 : tuplesort_performsort(pertrans->sortstates[aggstate->current_set]);
965 : :
5744 tgl@sss.pgh.pa.us 966 : 42 : ExecClearTuple(slot1);
967 [ - + ]: 42 : if (slot2)
5744 tgl@sss.pgh.pa.us 968 :UBC 0 : ExecClearTuple(slot2);
969 : :
3686 heikki.linnakangas@i 970 [ + + ]:CBC 150 : while (tuplesort_gettupleslot(pertrans->sortstates[aggstate->current_set],
971 : : true, true, slot1, &newAbbrevVal))
972 : : {
2965 andres@anarazel.de 973 [ - + ]: 108 : CHECK_FOR_INTERRUPTS();
974 : :
2760 975 : 108 : tmpcontext->ecxt_outertuple = slot1;
976 : 108 : tmpcontext->ecxt_innertuple = slot2;
977 : :
5744 tgl@sss.pgh.pa.us 978 [ - + ]: 108 : if (numDistinctCols == 0 ||
5744 tgl@sss.pgh.pa.us 979 [ # # ]:UBC 0 : !haveOldValue ||
3489 rhaas@postgresql.org 980 [ # # ]: 0 : newAbbrevVal != oldAbbrevVal ||
2760 andres@anarazel.de 981 [ # # ]: 0 : !ExecQual(pertrans->equalfnMulti, tmpcontext))
982 : : {
983 : : /*
984 : : * Extract the first numTransInputs columns as datums to pass to
985 : : * the transfn.
986 : : */
2760 andres@anarazel.de 987 :CBC 108 : slot_getsomeattrs(slot1, numTransInputs);
988 : :
989 : : /* Load values into fcinfo */
990 : : /* Start from 1, since the 0th arg will be the transition value */
4275 tgl@sss.pgh.pa.us 991 [ + + ]: 306 : for (i = 0; i < numTransInputs; i++)
992 : : {
2415 andres@anarazel.de 993 : 198 : fcinfo->args[i + 1].value = slot1->tts_values[i];
994 : 198 : fcinfo->args[i + 1].isnull = slot1->tts_isnull[i];
995 : : }
996 : :
3686 heikki.linnakangas@i 997 : 108 : advance_transition_function(aggstate, pertrans, pergroupstate);
998 : :
5744 tgl@sss.pgh.pa.us 999 [ - + ]: 108 : if (numDistinctCols > 0)
1000 : : {
1001 : : /* swap the slot pointers to retain the current tuple */
5744 tgl@sss.pgh.pa.us 1002 :UBC 0 : TupleTableSlot *tmpslot = slot2;
1003 : :
1004 : 0 : slot2 = slot1;
1005 : 0 : slot1 = tmpslot;
1006 : : /* avoid ExecQual() calls by reusing abbreviated keys */
3489 rhaas@postgresql.org 1007 : 0 : oldAbbrevVal = newAbbrevVal;
5744 tgl@sss.pgh.pa.us 1008 : 0 : haveOldValue = true;
1009 : : }
1010 : : }
1011 : :
1012 : : /* Reset context each time */
2760 andres@anarazel.de 1013 :CBC 108 : ResetExprContext(tmpcontext);
1014 : :
5744 tgl@sss.pgh.pa.us 1015 : 108 : ExecClearTuple(slot1);
1016 : : }
1017 : :
1018 [ - + ]: 42 : if (slot2)
5744 tgl@sss.pgh.pa.us 1019 :UBC 0 : ExecClearTuple(slot2);
1020 : :
3686 heikki.linnakangas@i 1021 :CBC 42 : tuplesort_end(pertrans->sortstates[aggstate->current_set]);
1022 : 42 : pertrans->sortstates[aggstate->current_set] = NULL;
1023 : :
1024 : : /* restore previous slot, potentially in use for grouping sets */
2760 andres@anarazel.de 1025 : 42 : tmpcontext->ecxt_outertuple = save;
5744 tgl@sss.pgh.pa.us 1026 : 42 : }
1027 : :
1028 : : /*
1029 : : * Compute the final value of one aggregate.
1030 : : *
1031 : : * This function handles only one grouping set (already set in
1032 : : * aggstate->current_set).
1033 : : *
1034 : : * The finalfn will be run, and the result delivered, in the
1035 : : * output-tuple context; caller's CurrentMemoryContext does not matter.
1036 : : * (But note that in some cases, such as when there is no finalfn, the
1037 : : * result might be a pointer to or into the agg's transition value.)
1038 : : *
1039 : : * The finalfn uses the state as set in the transno. This also might be
1040 : : * being used by another aggregate function, so it's important that we do
1041 : : * nothing destructive here. Moreover, the aggregate's final value might
1042 : : * get used in multiple places, so we mustn't return a R/W expanded datum.
1043 : : */
1044 : : static void
8340 1045 : 556900 : finalize_aggregate(AggState *aggstate,
1046 : : AggStatePerAgg peragg,
1047 : : AggStatePerGroup pergroupstate,
1048 : : Datum *resultVal, bool *resultIsNull)
1049 : : {
2415 andres@anarazel.de 1050 : 556900 : LOCAL_FCINFO(fcinfo, FUNC_MAX_ARGS);
4275 tgl@sss.pgh.pa.us 1051 : 556900 : bool anynull = false;
1052 : : MemoryContext oldContext;
1053 : : int i;
1054 : : ListCell *lc;
3686 heikki.linnakangas@i 1055 : 556900 : AggStatePerTrans pertrans = &aggstate->pertrans[peragg->transno];
1056 : :
8311 tgl@sss.pgh.pa.us 1057 : 556900 : oldContext = MemoryContextSwitchTo(aggstate->ss.ps.ps_ExprContext->ecxt_per_tuple_memory);
1058 : :
1059 : : /*
1060 : : * Evaluate any direct arguments. We do this even if there's no finalfn
1061 : : * (which is unlikely anyway), so that side-effects happen as expected.
1062 : : * The direct arguments go into arg positions 1 and up, leaving position 0
1063 : : * for the transition state value.
1064 : : */
4275 1065 : 556900 : i = 1;
2882 1066 [ + + + + : 557387 : foreach(lc, peragg->aggdirectargs)
+ + ]
1067 : : {
4275 1068 : 487 : ExprState *expr = (ExprState *) lfirst(lc);
1069 : :
2415 andres@anarazel.de 1070 : 487 : fcinfo->args[i].value = ExecEvalExpr(expr,
1071 : : aggstate->ss.ps.ps_ExprContext,
1072 : : &fcinfo->args[i].isnull);
1073 : 487 : anynull |= fcinfo->args[i].isnull;
4275 tgl@sss.pgh.pa.us 1074 : 487 : i++;
1075 : : }
1076 : :
1077 : : /*
1078 : : * Apply the agg's finalfn if one is provided, else return transValue.
1079 : : */
3686 heikki.linnakangas@i 1080 [ + + ]: 556900 : if (OidIsValid(peragg->finalfn_oid))
1081 : : {
1082 : 168453 : int numFinalArgs = peragg->numFinalArgs;
1083 : :
1084 : : /* set up aggstate->curperagg for AggGetAggref() */
2886 tgl@sss.pgh.pa.us 1085 : 168453 : aggstate->curperagg = peragg;
1086 : :
2415 andres@anarazel.de 1087 : 168453 : InitFunctionCallInfoData(*fcinfo, &peragg->finalfn,
1088 : : numFinalArgs,
1089 : : pertrans->aggCollation,
1090 : : (Node *) aggstate, NULL);
1091 : :
1092 : : /* Fill in the transition state value */
1093 : 168453 : fcinfo->args[0].value =
1094 [ + + + + ]: 168453 : MakeExpandedObjectReadOnly(pergroupstate->transValue,
1095 : : pergroupstate->transValueIsNull,
1096 : : pertrans->transtypeLen);
1097 : 168453 : fcinfo->args[0].isnull = pergroupstate->transValueIsNull;
4275 tgl@sss.pgh.pa.us 1098 : 168453 : anynull |= pergroupstate->transValueIsNull;
1099 : :
1100 : : /* Fill any remaining argument positions with nulls */
4154 1101 [ + + ]: 244066 : for (; i < numFinalArgs; i++)
1102 : : {
2415 andres@anarazel.de 1103 : 75613 : fcinfo->args[i].value = (Datum) 0;
1104 : 75613 : fcinfo->args[i].isnull = true;
4275 tgl@sss.pgh.pa.us 1105 : 75613 : anynull = true;
1106 : : }
1107 : :
2415 andres@anarazel.de 1108 [ + + - + ]: 168453 : if (fcinfo->flinfo->fn_strict && anynull)
1109 : : {
1110 : : /* don't call a strict function with NULL inputs */
9232 tgl@sss.pgh.pa.us 1111 :UBC 0 : *resultVal = (Datum) 0;
1112 : 0 : *resultIsNull = true;
1113 : : }
1114 : : else
1115 : : {
1116 : : Datum result;
1117 : :
874 tgl@sss.pgh.pa.us 1118 :CBC 168453 : result = FunctionCallInvoke(fcinfo);
2415 andres@anarazel.de 1119 : 168447 : *resultIsNull = fcinfo->isnull;
874 tgl@sss.pgh.pa.us 1120 [ + + + + ]: 168447 : *resultVal = MakeExpandedObjectReadOnly(result,
1121 : : fcinfo->isnull,
1122 : : peragg->resulttypeLen);
1123 : : }
2886 1124 : 168447 : aggstate->curperagg = NULL;
1125 : : }
1126 : : else
1127 : : {
1066 1128 : 388447 : *resultVal =
1129 [ + + + + ]: 388447 : MakeExpandedObjectReadOnly(pergroupstate->transValue,
1130 : : pergroupstate->transValueIsNull,
1131 : : pertrans->transtypeLen);
8340 1132 : 388447 : *resultIsNull = pergroupstate->transValueIsNull;
1133 : : }
1134 : :
1135 : 556894 : MemoryContextSwitchTo(oldContext);
9399 1136 : 556894 : }
1137 : :
1138 : : /*
1139 : : * Compute the output value of one partial aggregate.
1140 : : *
1141 : : * The serialization function will be run, and the result delivered, in the
1142 : : * output-tuple context; caller's CurrentMemoryContext does not matter.
1143 : : */
1144 : : static void
3448 rhaas@postgresql.org 1145 : 7818 : finalize_partialaggregate(AggState *aggstate,
1146 : : AggStatePerAgg peragg,
1147 : : AggStatePerGroup pergroupstate,
1148 : : Datum *resultVal, bool *resultIsNull)
1149 : : {
3376 1150 : 7818 : AggStatePerTrans pertrans = &aggstate->pertrans[peragg->transno];
1151 : : MemoryContext oldContext;
1152 : :
3448 1153 : 7818 : oldContext = MemoryContextSwitchTo(aggstate->ss.ps.ps_ExprContext->ecxt_per_tuple_memory);
1154 : :
1155 : : /*
1156 : : * serialfn_oid will be set if we must serialize the transvalue before
1157 : : * returning it
1158 : : */
1159 [ + + ]: 7818 : if (OidIsValid(pertrans->serialfn_oid))
1160 : : {
1161 : : /* Don't call a strict serialization function with NULL input. */
1162 [ + - + + ]: 333 : if (pertrans->serialfn.fn_strict && pergroupstate->transValueIsNull)
1163 : : {
1164 : 36 : *resultVal = (Datum) 0;
1165 : 36 : *resultIsNull = true;
1166 : : }
1167 : : else
1168 : : {
2415 andres@anarazel.de 1169 : 297 : FunctionCallInfo fcinfo = pertrans->serialfn_fcinfo;
1170 : : Datum result;
1171 : :
1172 : 297 : fcinfo->args[0].value =
1173 [ + - + - ]: 297 : MakeExpandedObjectReadOnly(pergroupstate->transValue,
1174 : : pergroupstate->transValueIsNull,
1175 : : pertrans->transtypeLen);
1176 : 297 : fcinfo->args[0].isnull = pergroupstate->transValueIsNull;
1964 tgl@sss.pgh.pa.us 1177 : 297 : fcinfo->isnull = false;
1178 : :
874 1179 : 297 : result = FunctionCallInvoke(fcinfo);
3448 rhaas@postgresql.org 1180 : 297 : *resultIsNull = fcinfo->isnull;
874 tgl@sss.pgh.pa.us 1181 [ + - + - ]: 297 : *resultVal = MakeExpandedObjectReadOnly(result,
1182 : : fcinfo->isnull,
1183 : : peragg->resulttypeLen);
1184 : : }
1185 : : }
1186 : : else
1187 : : {
1066 1188 : 7485 : *resultVal =
1189 [ + + + + ]: 7485 : MakeExpandedObjectReadOnly(pergroupstate->transValue,
1190 : : pergroupstate->transValueIsNull,
1191 : : pertrans->transtypeLen);
3448 rhaas@postgresql.org 1192 : 7485 : *resultIsNull = pergroupstate->transValueIsNull;
1193 : : }
1194 : :
1195 : 7818 : MemoryContextSwitchTo(oldContext);
1196 : 7818 : }
1197 : :
1198 : : /*
1199 : : * Extract the attributes that make up the grouping key into the
1200 : : * hashslot. This is necessary to compute the hash or perform a lookup.
1201 : : */
1202 : : static inline void
1868 jdavis@postgresql.or 1203 : 3611834 : prepare_hash_slot(AggStatePerHash perhash,
1204 : : TupleTableSlot *inputslot,
1205 : : TupleTableSlot *hashslot)
1206 : : {
1207 : : int i;
1208 : :
1209 : : /* transfer just the needed columns into hashslot */
2026 1210 : 3611834 : slot_getsomeattrs(inputslot, perhash->largestGrpColIdx);
1211 : 3611834 : ExecClearTuple(hashslot);
1212 : :
1213 [ + + ]: 8697963 : for (i = 0; i < perhash->numhashGrpCols; i++)
1214 : : {
1215 : 5086129 : int varNumber = perhash->hashGrpColIdxInput[i] - 1;
1216 : :
1217 : 5086129 : hashslot->tts_values[i] = inputslot->tts_values[varNumber];
1218 : 5086129 : hashslot->tts_isnull[i] = inputslot->tts_isnull[varNumber];
1219 : : }
1220 : 3611834 : ExecStoreVirtualTuple(hashslot);
1221 : 3611834 : }
1222 : :
1223 : : /*
1224 : : * Prepare to finalize and project based on the specified representative tuple
1225 : : * slot and grouping set.
1226 : : *
1227 : : * In the specified tuple slot, force to null all attributes that should be
1228 : : * read as null in the context of the current grouping set. Also stash the
1229 : : * current group bitmap where GroupingExpr can get at it.
1230 : : *
1231 : : * This relies on three conditions:
1232 : : *
1233 : : * 1) Nothing is ever going to try and extract the whole tuple from this slot,
1234 : : * only reference it in evaluations, which will only access individual
1235 : : * attributes.
1236 : : *
1237 : : * 2) No system columns are going to need to be nulled. (If a system column is
1238 : : * referenced in a group clause, it is actually projected in the outer plan
1239 : : * tlist.)
1240 : : *
1241 : : * 3) Within a given phase, we never need to recover the value of an attribute
1242 : : * once it has been set to null.
1243 : : *
1244 : : * Poking into the slot this way is a bit ugly, but the consensus is that the
1245 : : * alternative was worse.
1246 : : */
1247 : : static void
3766 andres@anarazel.de 1248 : 414465 : prepare_projection_slot(AggState *aggstate, TupleTableSlot *slot, int currentSet)
1249 : : {
1250 [ + + ]: 414465 : if (aggstate->phase->grouped_cols)
1251 : : {
3759 bruce@momjian.us 1252 : 272448 : Bitmapset *grouped_cols = aggstate->phase->grouped_cols[currentSet];
1253 : :
3766 andres@anarazel.de 1254 : 272448 : aggstate->grouped_cols = grouped_cols;
1255 : :
2518 1256 [ + + ]: 272448 : if (TTS_EMPTY(slot))
1257 : : {
1258 : : /*
1259 : : * Force all values to be NULL if working on an empty input tuple
1260 : : * (i.e. an empty grouping set for which no input rows were
1261 : : * supplied).
1262 : : */
3766 1263 : 24 : ExecStoreAllNullTuple(slot);
1264 : : }
1265 [ + + ]: 272424 : else if (aggstate->all_grouped_cols)
1266 : : {
1267 : : ListCell *lc;
1268 : :
1269 : : /* all_grouped_cols is arranged in desc order */
1270 : 272400 : slot_getsomeattrs(slot, linitial_int(aggstate->all_grouped_cols));
1271 : :
1272 [ + - + + : 748365 : foreach(lc, aggstate->all_grouped_cols)
+ + ]
1273 : : {
3759 bruce@momjian.us 1274 : 475965 : int attnum = lfirst_int(lc);
1275 : :
3766 andres@anarazel.de 1276 [ + + ]: 475965 : if (!bms_is_member(attnum, grouped_cols))
1277 : 28913 : slot->tts_isnull[attnum - 1] = true;
1278 : : }
1279 : : }
1280 : : }
1281 : 414465 : }
1282 : :
1283 : : /*
1284 : : * Compute the final value of all aggregates for one group.
1285 : : *
1286 : : * This function handles only one grouping set at a time, which the caller must
1287 : : * have selected. It's also the caller's responsibility to adjust the supplied
1288 : : * pergroup parameter to point to the current set's transvalues.
1289 : : *
1290 : : * Results are stored in the output econtext aggvalues/aggnulls.
1291 : : */
1292 : : static void
1293 : 414465 : finalize_aggregates(AggState *aggstate,
1294 : : AggStatePerAgg peraggs,
1295 : : AggStatePerGroup pergroup)
1296 : : {
1297 : 414465 : ExprContext *econtext = aggstate->ss.ps.ps_ExprContext;
1298 : 414465 : Datum *aggvalues = econtext->ecxt_aggvalues;
1299 : 414465 : bool *aggnulls = econtext->ecxt_aggnulls;
1300 : : int aggno;
1301 : :
1302 : : /*
1303 : : * If there were any DISTINCT and/or ORDER BY aggregates, sort their
1304 : : * inputs and run the transition functions.
1305 : : */
1109 drowley@postgresql.o 1306 [ + + ]: 979054 : for (int transno = 0; transno < aggstate->numtrans; transno++)
1307 : : {
3686 heikki.linnakangas@i 1308 : 564589 : AggStatePerTrans pertrans = &aggstate->pertrans[transno];
1309 : : AggStatePerGroup pergroupstate;
1310 : :
3085 rhodiumtoad@postgres 1311 : 564589 : pergroupstate = &pergroup[transno];
1312 : :
1131 drowley@postgresql.o 1313 [ + + ]: 564589 : if (pertrans->aggsortrequired)
1314 : : {
3085 rhodiumtoad@postgres 1315 [ + - - + ]: 26921 : Assert(aggstate->aggstrategy != AGG_HASHED &&
1316 : : aggstate->aggstrategy != AGG_MIXED);
1317 : :
3686 heikki.linnakangas@i 1318 [ + + ]: 26921 : if (pertrans->numInputs == 1)
3766 andres@anarazel.de 1319 : 26879 : process_ordered_aggregate_single(aggstate,
1320 : : pertrans,
1321 : : pergroupstate);
1322 : : else
1323 : 42 : process_ordered_aggregate_multi(aggstate,
1324 : : pertrans,
1325 : : pergroupstate);
1326 : : }
1131 drowley@postgresql.o 1327 [ + + + + ]: 537668 : else if (pertrans->numDistinctCols > 0 && pertrans->haslast)
1328 : : {
1329 : 9179 : pertrans->haslast = false;
1330 : :
1331 [ + + ]: 9179 : if (pertrans->numDistinctCols == 1)
1332 : : {
1333 [ + + + + ]: 9131 : if (!pertrans->inputtypeByVal && !pertrans->lastisnull)
1334 : 131 : pfree(DatumGetPointer(pertrans->lastdatum));
1335 : :
1336 : 9131 : pertrans->lastisnull = false;
1337 : 9131 : pertrans->lastdatum = (Datum) 0;
1338 : : }
1339 : : else
1340 : 48 : ExecClearTuple(pertrans->uniqslot);
1341 : : }
1342 : : }
1343 : :
1344 : : /*
1345 : : * Run the final functions.
1346 : : */
3182 heikki.linnakangas@i 1347 [ + + ]: 979177 : for (aggno = 0; aggno < aggstate->numaggs; aggno++)
1348 : : {
1349 : 564718 : AggStatePerAgg peragg = &peraggs[aggno];
1350 : 564718 : int transno = peragg->transno;
1351 : : AggStatePerGroup pergroupstate;
1352 : :
3085 rhodiumtoad@postgres 1353 : 564718 : pergroupstate = &pergroup[transno];
1354 : :
3359 tgl@sss.pgh.pa.us 1355 [ + + ]: 564718 : if (DO_AGGSPLIT_SKIPFINAL(aggstate->aggsplit))
3448 rhaas@postgresql.org 1356 : 7818 : finalize_partialaggregate(aggstate, peragg, pergroupstate,
1357 : 7818 : &aggvalues[aggno], &aggnulls[aggno]);
1358 : : else
3359 tgl@sss.pgh.pa.us 1359 : 556900 : finalize_aggregate(aggstate, peragg, pergroupstate,
1360 : 556900 : &aggvalues[aggno], &aggnulls[aggno]);
1361 : : }
3766 andres@anarazel.de 1362 : 414459 : }
1363 : :
1364 : : /*
1365 : : * Project the result of a group (whose aggs have already been calculated by
1366 : : * finalize_aggregates). Returns the result slot, or NULL if no row is
1367 : : * projected (suppressed by qual).
1368 : : */
1369 : : static TupleTableSlot *
1370 : 414459 : project_aggregates(AggState *aggstate)
1371 : : {
1372 : 414459 : ExprContext *econtext = aggstate->ss.ps.ps_ExprContext;
1373 : :
1374 : : /*
1375 : : * Check the qual (HAVING clause); if the group does not match, ignore it.
1376 : : */
3098 1377 [ + + ]: 414459 : if (ExecQual(aggstate->ss.ps.qual, econtext))
1378 : : {
1379 : : /*
1380 : : * Form and return projection tuple using the aggregate results and
1381 : : * the representative input tuple.
1382 : : */
3152 1383 : 361229 : return ExecProject(aggstate->ss.ps.ps_ProjInfo);
1384 : : }
1385 : : else
3766 1386 [ - + ]: 53230 : InstrCountFiltered1(aggstate, 1);
1387 : :
1388 : 53230 : return NULL;
1389 : : }
1390 : :
1391 : : /*
1392 : : * Find input-tuple columns that are needed, dividing them into
1393 : : * aggregated and unaggregated sets.
1394 : : */
1395 : : static void
1882 jdavis@postgresql.or 1396 : 2832 : find_cols(AggState *aggstate, Bitmapset **aggregated, Bitmapset **unaggregated)
1397 : : {
1578 tgl@sss.pgh.pa.us 1398 : 2832 : Agg *agg = (Agg *) aggstate->ss.ps.plan;
1399 : : FindColsContext context;
1400 : :
1882 jdavis@postgresql.or 1401 : 2832 : context.is_aggref = false;
1402 : 2832 : context.aggregated = NULL;
1403 : 2832 : context.unaggregated = NULL;
1404 : :
1405 : : /* Examine tlist and quals */
1406 : 2832 : (void) find_cols_walker((Node *) agg->plan.targetlist, &context);
1407 : 2832 : (void) find_cols_walker((Node *) agg->plan.qual, &context);
1408 : :
1409 : : /* In some cases, grouping columns will not appear in the tlist */
1675 tgl@sss.pgh.pa.us 1410 [ + + ]: 7193 : for (int i = 0; i < agg->numCols; i++)
1411 : 4361 : context.unaggregated = bms_add_member(context.unaggregated,
1412 : 4361 : agg->grpColIdx[i]);
1413 : :
1882 jdavis@postgresql.or 1414 : 2832 : *aggregated = context.aggregated;
1415 : 2832 : *unaggregated = context.unaggregated;
7010 tgl@sss.pgh.pa.us 1416 : 2832 : }
1417 : :
1418 : : static bool
1882 jdavis@postgresql.or 1419 : 33465 : find_cols_walker(Node *node, FindColsContext *context)
1420 : : {
7010 tgl@sss.pgh.pa.us 1421 [ + + ]: 33465 : if (node == NULL)
1422 : 5973 : return false;
1423 [ + + ]: 27492 : if (IsA(node, Var))
1424 : : {
1425 : 7286 : Var *var = (Var *) node;
1426 : :
1427 : : /* setrefs.c should have set the varno to OUTER_VAR */
5079 1428 [ - + ]: 7286 : Assert(var->varno == OUTER_VAR);
7010 1429 [ - + ]: 7286 : Assert(var->varlevelsup == 0);
1882 jdavis@postgresql.or 1430 [ + + ]: 7286 : if (context->is_aggref)
1431 : 2290 : context->aggregated = bms_add_member(context->aggregated,
1432 : 2290 : var->varattno);
1433 : : else
1434 : 4996 : context->unaggregated = bms_add_member(context->unaggregated,
1435 : 4996 : var->varattno);
7010 tgl@sss.pgh.pa.us 1436 : 7286 : return false;
1437 : : }
1882 jdavis@postgresql.or 1438 [ + + ]: 20206 : if (IsA(node, Aggref))
1439 : : {
1440 [ - + ]: 3385 : Assert(!context->is_aggref);
1441 : 3385 : context->is_aggref = true;
282 peter@eisentraut.org 1442 : 3385 : expression_tree_walker(node, find_cols_walker, context);
1882 jdavis@postgresql.or 1443 : 3385 : context->is_aggref = false;
7010 tgl@sss.pgh.pa.us 1444 : 3385 : return false;
1445 : : }
282 peter@eisentraut.org 1446 : 16821 : return expression_tree_walker(node, find_cols_walker, context);
1447 : : }
1448 : :
1449 : : /*
1450 : : * (Re-)initialize the hash table(s) to empty.
1451 : : *
1452 : : * To implement hashed aggregation, we need a hashtable that stores a
1453 : : * representative tuple and an array of AggStatePerGroup structs for each
1454 : : * distinct set of GROUP BY column values. We compute the hash key from the
1455 : : * GROUP BY columns. The per-group data is allocated in initialize_hash_entry(),
1456 : : * for each entry.
1457 : : *
1458 : : * We have a separate hashtable and associated perhash data structure for each
1459 : : * grouping set for which we're doing hashing.
1460 : : *
1461 : : * The contents of the hash tables always live in the hashcontext's per-tuple
1462 : : * memory context (there is only one of these for all tables together, since
1463 : : * they are all reset at the same time).
1464 : : */
1465 : : static void
2026 jdavis@postgresql.or 1466 : 8323 : build_hash_tables(AggState *aggstate)
1467 : : {
1468 : : int setno;
1469 : :
1470 [ + + ]: 16818 : for (setno = 0; setno < aggstate->num_hashes; ++setno)
1471 : : {
1472 : 8495 : AggStatePerHash perhash = &aggstate->perhash[setno];
1473 : : long nbuckets;
1474 : : Size memory;
1475 : :
1998 1476 [ + + ]: 8495 : if (perhash->hashtable != NULL)
1477 : : {
1478 : 6183 : ResetTupleHashTable(perhash->hashtable);
1479 : 6183 : continue;
1480 : : }
1481 : :
3085 rhodiumtoad@postgres 1482 [ - + ]: 2312 : Assert(perhash->aggnode->numGroups > 0);
1483 : :
1998 jdavis@postgresql.or 1484 : 2312 : memory = aggstate->hash_mem_limit / aggstate->num_hashes;
1485 : :
1486 : : /* choose reasonable number of buckets per hashtable */
1941 tgl@sss.pgh.pa.us 1487 : 2312 : nbuckets = hash_choose_num_buckets(aggstate->hashentrysize,
1488 : 2312 : perhash->aggnode->numGroups,
1489 : : memory);
1490 : :
1491 : : #ifdef USE_INJECTION_POINTS
1492 : : if (IS_INJECTION_POINT_ATTACHED("hash-aggregate-oversize-table"))
1493 : : {
1494 : : nbuckets = memory / TupleHashEntrySize();
1495 : : INJECTION_POINT_CACHED("hash-aggregate-oversize-table", NULL);
1496 : : }
1497 : : #endif
1498 : :
1998 jdavis@postgresql.or 1499 : 2312 : build_hash_table(aggstate, setno, nbuckets);
1500 : : }
1501 : :
1502 : 8323 : aggstate->hash_ngroups_current = 0;
6169 neilc@samurai.com 1503 : 8323 : }
1504 : :
1505 : : /*
1506 : : * Build a single hashtable for this grouping set.
1507 : : */
1508 : : static void
2026 jdavis@postgresql.or 1509 : 2312 : build_hash_table(AggState *aggstate, int setno, long nbuckets)
1510 : : {
1511 : 2312 : AggStatePerHash perhash = &aggstate->perhash[setno];
1941 tgl@sss.pgh.pa.us 1512 : 2312 : MemoryContext metacxt = aggstate->hash_metacxt;
166 jdavis@postgresql.or 1513 : 2312 : MemoryContext tablecxt = aggstate->hash_tablecxt;
1941 tgl@sss.pgh.pa.us 1514 : 2312 : MemoryContext tmpcxt = aggstate->tmpcontext->ecxt_per_tuple_memory;
1515 : : Size additionalsize;
1516 : :
2026 jdavis@postgresql.or 1517 [ + + - + ]: 2312 : Assert(aggstate->aggstrategy == AGG_HASHED ||
1518 : : aggstate->aggstrategy == AGG_MIXED);
1519 : :
1520 : : /*
1521 : : * Used to make sure initial hash table allocation does not exceed
1522 : : * hash_mem. Note that the estimate does not include space for
1523 : : * pass-by-reference transition data values, nor for the representative
1524 : : * tuple of each group.
1525 : : */
1526 : 2312 : additionalsize = aggstate->numtrans * sizeof(AggStatePerGroupData);
1527 : :
261 tgl@sss.pgh.pa.us 1528 : 4624 : perhash->hashtable = BuildTupleHashTable(&aggstate->ss.ps,
1529 : 2312 : perhash->hashslot->tts_tupleDescriptor,
1530 : 2312 : perhash->hashslot->tts_ops,
1531 : : perhash->numCols,
1532 : : perhash->hashGrpColIdxHash,
1533 : 2312 : perhash->eqfuncoids,
1534 : : perhash->hashfunctions,
1535 : 2312 : perhash->aggnode->grpCollations,
1536 : : nbuckets,
1537 : : additionalsize,
1538 : : metacxt,
1539 : : tablecxt,
1540 : : tmpcxt,
1541 : 2312 : DO_AGGSPLIT_SKIPFINAL(aggstate->aggsplit));
2026 jdavis@postgresql.or 1542 : 2312 : }
1543 : :
1544 : : /*
1545 : : * Compute columns that actually need to be stored in hashtable entries. The
1546 : : * incoming tuples from the child plan node will contain grouping columns,
1547 : : * other columns referenced in our targetlist and qual, columns used to
1548 : : * compute the aggregate functions, and perhaps just junk columns we don't use
1549 : : * at all. Only columns of the first two types need to be stored in the
1550 : : * hashtable, and getting rid of the others can make the table entries
1551 : : * significantly smaller. The hashtable only contains the relevant columns,
1552 : : * and is packed/unpacked in lookup_hash_entries() / agg_retrieve_hash_table()
1553 : : * into the format of the normal input descriptor.
1554 : : *
1555 : : * Additional columns, in addition to the columns grouped by, come from two
1556 : : * sources: Firstly functionally dependent columns that we don't need to group
1557 : : * by themselves, and secondly ctids for row-marks.
1558 : : *
1559 : : * To eliminate duplicates, we build a bitmapset of the needed columns, and
1560 : : * then build an array of the columns included in the hashtable. We might
1561 : : * still have duplicates if the passed-in grpColIdx has them, which can happen
1562 : : * in edge cases from semijoins/distinct; these can't always be removed,
1563 : : * because it's not certain that the duplicate cols will be using the same
1564 : : * hash function.
1565 : : *
1566 : : * Note that the array is preserved over ExecReScanAgg, so we allocate it in
1567 : : * the per-query context (unlike the hash table itself).
1568 : : */
1569 : : static void
6169 neilc@samurai.com 1570 : 2832 : find_hash_columns(AggState *aggstate)
1571 : : {
1572 : : Bitmapset *base_colnos;
1573 : : Bitmapset *aggregated_colnos;
1882 jdavis@postgresql.or 1574 : 2832 : TupleDesc scanDesc = aggstate->ss.ss_ScanTupleSlot->tts_tupleDescriptor;
3202 andres@anarazel.de 1575 : 2832 : List *outerTlist = outerPlanState(aggstate)->plan->targetlist;
3085 rhodiumtoad@postgres 1576 : 2832 : int numHashes = aggstate->num_hashes;
2760 andres@anarazel.de 1577 : 2832 : EState *estate = aggstate->ss.ps.state;
1578 : : int j;
1579 : :
1580 : : /* Find Vars that will be needed in tlist and qual */
1882 jdavis@postgresql.or 1581 : 2832 : find_cols(aggstate, &aggregated_colnos, &base_colnos);
1582 : 2832 : aggstate->colnos_needed = bms_union(base_colnos, aggregated_colnos);
1583 : 2832 : aggstate->max_colno_needed = 0;
1584 : 2832 : aggstate->all_cols_needed = true;
1585 : :
1586 [ + + ]: 11848 : for (int i = 0; i < scanDesc->natts; i++)
1587 : : {
1578 tgl@sss.pgh.pa.us 1588 : 9016 : int colno = i + 1;
1589 : :
1882 jdavis@postgresql.or 1590 [ + + ]: 9016 : if (bms_is_member(colno, aggstate->colnos_needed))
1591 : 6032 : aggstate->max_colno_needed = colno;
1592 : : else
1593 : 2984 : aggstate->all_cols_needed = false;
1594 : : }
1595 : :
3085 rhodiumtoad@postgres 1596 [ + + ]: 5921 : for (j = 0; j < numHashes; ++j)
1597 : : {
1598 : 3089 : AggStatePerHash perhash = &aggstate->perhash[j];
1599 : 3089 : Bitmapset *colnos = bms_copy(base_colnos);
1600 : 3089 : AttrNumber *grpColIdx = perhash->aggnode->grpColIdx;
1601 : 3089 : List *hashTlist = NIL;
1602 : : TupleDesc hashDesc;
1603 : : int maxCols;
1604 : : int i;
1605 : :
1606 : 3089 : perhash->largestGrpColIdx = 0;
1607 : :
1608 : : /*
1609 : : * If we're doing grouping sets, then some Vars might be referenced in
1610 : : * tlist/qual for the benefit of other grouping sets, but not needed
1611 : : * when hashing; i.e. prepare_projection_slot will null them out, so
1612 : : * there'd be no point storing them. Use prepare_projection_slot's
1613 : : * logic to determine which.
1614 : : */
1615 [ + - ]: 3089 : if (aggstate->phases[0].grouped_cols)
1616 : : {
1617 : 3089 : Bitmapset *grouped_cols = aggstate->phases[0].grouped_cols[j];
1618 : : ListCell *lc;
1619 : :
1620 [ + - + + : 8433 : foreach(lc, aggstate->all_grouped_cols)
+ + ]
1621 : : {
1622 : 5344 : int attnum = lfirst_int(lc);
1623 : :
1624 [ + + ]: 5344 : if (!bms_is_member(attnum, grouped_cols))
1625 : 672 : colnos = bms_del_member(colnos, attnum);
1626 : : }
1627 : : }
1628 : :
1629 : : /*
1630 : : * Compute maximum number of input columns accounting for possible
1631 : : * duplications in the grpColIdx array, which can happen in some edge
1632 : : * cases where HashAggregate was generated as part of a semijoin or a
1633 : : * DISTINCT.
1634 : : */
2298 1635 : 3089 : maxCols = bms_num_members(colnos) + perhash->numCols;
1636 : :
3085 1637 : 3089 : perhash->hashGrpColIdxInput =
2298 1638 : 3089 : palloc(maxCols * sizeof(AttrNumber));
3085 1639 : 3089 : perhash->hashGrpColIdxHash =
1640 : 3089 : palloc(perhash->numCols * sizeof(AttrNumber));
1641 : :
1642 : : /* Add all the grouping columns to colnos */
2298 1643 [ + + ]: 7761 : for (i = 0; i < perhash->numCols; i++)
1644 : 4672 : colnos = bms_add_member(colnos, grpColIdx[i]);
1645 : :
1646 : : /*
1647 : : * First build mapping for columns directly hashed. These are the
1648 : : * first, because they'll be accessed when computing hash values and
1649 : : * comparing tuples for exact matches. We also build simple mapping
1650 : : * for execGrouping, so it knows where to find the to-be-hashed /
1651 : : * compared columns in the input.
1652 : : */
3085 1653 [ + + ]: 7761 : for (i = 0; i < perhash->numCols; i++)
1654 : : {
1655 : 4672 : perhash->hashGrpColIdxInput[i] = grpColIdx[i];
1656 : 4672 : perhash->hashGrpColIdxHash[i] = i + 1;
1657 : 4672 : perhash->numhashGrpCols++;
1658 : : /* delete already mapped columns */
919 tgl@sss.pgh.pa.us 1659 : 4672 : colnos = bms_del_member(colnos, grpColIdx[i]);
1660 : : }
1661 : :
1662 : : /* and add the remaining columns */
1663 : 3089 : i = -1;
1664 [ + + ]: 3363 : while ((i = bms_next_member(colnos, i)) >= 0)
1665 : : {
3085 rhodiumtoad@postgres 1666 : 274 : perhash->hashGrpColIdxInput[perhash->numhashGrpCols] = i;
1667 : 274 : perhash->numhashGrpCols++;
1668 : : }
1669 : :
1670 : : /* and build a tuple descriptor for the hashtable */
1671 [ + + ]: 8035 : for (i = 0; i < perhash->numhashGrpCols; i++)
1672 : : {
1673 : 4946 : int varNumber = perhash->hashGrpColIdxInput[i] - 1;
1674 : :
1675 : 4946 : hashTlist = lappend(hashTlist, list_nth(outerTlist, varNumber));
1676 : 4946 : perhash->largestGrpColIdx =
1677 : 4946 : Max(varNumber + 1, perhash->largestGrpColIdx);
1678 : : }
1679 : :
2482 andres@anarazel.de 1680 : 3089 : hashDesc = ExecTypeFromTL(hashTlist);
1681 : :
2760 1682 : 3089 : execTuplesHashPrepare(perhash->numCols,
1683 : 3089 : perhash->aggnode->grpOperators,
1684 : : &perhash->eqfuncoids,
1685 : : &perhash->hashfunctions);
2759 1686 : 3089 : perhash->hashslot =
2487 1687 : 3089 : ExecAllocTableSlot(&estate->es_tupleTable, hashDesc,
1688 : : &TTSOpsMinimalTuple);
1689 : :
3085 rhodiumtoad@postgres 1690 : 3089 : list_free(hashTlist);
1691 : 3089 : bms_free(colnos);
1692 : : }
1693 : :
1694 : 2832 : bms_free(base_colnos);
8340 tgl@sss.pgh.pa.us 1695 : 2832 : }
1696 : :
1697 : : /*
1698 : : * Estimate per-hash-table-entry overhead.
1699 : : */
1700 : : Size
1982 jdavis@postgresql.or 1701 : 17835 : hash_agg_entry_size(int numTrans, Size tupleWidth, Size transitionSpace)
1702 : : {
1703 : : Size tupleChunkSize;
1704 : : Size pergroupChunkSize;
1705 : : Size transitionChunkSize;
1941 tgl@sss.pgh.pa.us 1706 : 17835 : Size tupleSize = (MAXALIGN(SizeofMinimalTupleHeader) +
1707 : : tupleWidth);
1708 : 17835 : Size pergroupSize = numTrans * sizeof(AggStatePerGroupData);
1709 : :
1710 : : /*
1711 : : * Entries use the Bump allocator, so the chunk sizes are the same as the
1712 : : * requested sizes.
1713 : : */
166 jdavis@postgresql.or 1714 : 17835 : tupleChunkSize = MAXALIGN(tupleSize);
1715 : 17835 : pergroupChunkSize = pergroupSize;
1716 : :
1717 : : /*
1718 : : * Transition values use AllocSet, which has a chunk header and also uses
1719 : : * power-of-two allocations.
1720 : : */
1982 1721 [ + + ]: 17835 : if (transitionSpace > 0)
166 1722 : 2387 : transitionChunkSize = CHUNKHDRSZ + pg_nextpower2_size_t(transitionSpace);
1723 : : else
1982 1724 : 15448 : transitionChunkSize = 0;
1725 : :
1726 : : return
166 1727 : 17835 : TupleHashEntrySize() +
1982 1728 : 17835 : tupleChunkSize +
1729 : 17835 : pergroupChunkSize +
1730 : : transitionChunkSize;
1731 : : }
1732 : :
1733 : : /*
1734 : : * hashagg_recompile_expressions()
1735 : : *
1736 : : * Identifies the right phase, compiles the right expression given the
1737 : : * arguments, and then sets phase->evalfunc to that expression.
1738 : : *
1739 : : * Different versions of the compiled expression are needed depending on
1740 : : * whether hash aggregation has spilled or not, and whether it's reading from
1741 : : * the outer plan or a tape. Before spilling to disk, the expression reads
1742 : : * from the outer plan and does not need to perform a NULL check. After
1743 : : * HashAgg begins to spill, new groups will not be created in the hash table,
1744 : : * and the AggStatePerGroup array may be NULL; therefore we need to add a null
1745 : : * pointer check to the expression. Then, when reading spilled data from a
1746 : : * tape, we change the outer slot type to be a fixed minimal tuple slot.
1747 : : *
1748 : : * It would be wasteful to recompile every time, so cache the compiled
1749 : : * expressions in the AggStatePerPhase, and reuse when appropriate.
1750 : : */
1751 : : static void
1998 1752 : 32844 : hashagg_recompile_expressions(AggState *aggstate, bool minslot, bool nullcheck)
1753 : : {
1754 : : AggStatePerPhase phase;
1941 tgl@sss.pgh.pa.us 1755 : 32844 : int i = minslot ? 1 : 0;
1756 : 32844 : int j = nullcheck ? 1 : 0;
1757 : :
1998 jdavis@postgresql.or 1758 [ + + - + ]: 32844 : Assert(aggstate->aggstrategy == AGG_HASHED ||
1759 : : aggstate->aggstrategy == AGG_MIXED);
1760 : :
1761 [ + + ]: 32844 : if (aggstate->aggstrategy == AGG_HASHED)
1762 : 6558 : phase = &aggstate->phases[0];
1763 : : else /* AGG_MIXED */
1764 : 26286 : phase = &aggstate->phases[1];
1765 : :
1766 [ + + ]: 32844 : if (phase->evaltrans_cache[i][j] == NULL)
1767 : : {
1941 tgl@sss.pgh.pa.us 1768 : 42 : const TupleTableSlotOps *outerops = aggstate->ss.ps.outerops;
1769 : 42 : bool outerfixed = aggstate->ss.ps.outeropsfixed;
1770 : 42 : bool dohash = true;
1715 jdavis@postgresql.or 1771 : 42 : bool dosort = false;
1772 : :
1773 : : /*
1774 : : * If minslot is true, that means we are processing a spilled batch
1775 : : * (inside agg_refill_hash_table()), and we must not advance the
1776 : : * sorted grouping sets.
1777 : : */
1778 [ + + + + ]: 42 : if (aggstate->aggstrategy == AGG_MIXED && !minslot)
1779 : 6 : dosort = true;
1780 : :
1781 : : /* temporarily change the outerops while compiling the expression */
1998 1782 [ + + ]: 42 : if (minslot)
1783 : : {
1784 : 21 : aggstate->ss.ps.outerops = &TTSOpsMinimalTuple;
1785 : 21 : aggstate->ss.ps.outeropsfixed = true;
1786 : : }
1787 : :
1941 tgl@sss.pgh.pa.us 1788 : 42 : phase->evaltrans_cache[i][j] = ExecBuildAggTrans(aggstate, phase,
1789 : : dosort, dohash,
1790 : : nullcheck);
1791 : :
1792 : : /* change back */
1998 jdavis@postgresql.or 1793 : 42 : aggstate->ss.ps.outerops = outerops;
1794 : 42 : aggstate->ss.ps.outeropsfixed = outerfixed;
1795 : : }
1796 : :
1797 : 32844 : phase->evaltrans = phase->evaltrans_cache[i][j];
1798 : 32844 : }
1799 : :
1800 : : /*
1801 : : * Set limits that trigger spilling to avoid exceeding hash_mem. Consider the
1802 : : * number of partitions we expect to create (if we do spill).
1803 : : *
1804 : : * There are two limits: a memory limit, and also an ngroups limit. The
1805 : : * ngroups limit becomes important when we expect transition values to grow
1806 : : * substantially larger than the initial value.
1807 : : */
1808 : : void
1866 1809 : 30127 : hash_agg_set_limits(double hashentrysize, double input_groups, int used_bits,
1810 : : Size *mem_limit, uint64 *ngroups_limit,
1811 : : int *num_partitions)
1812 : : {
1813 : : int npartitions;
1814 : : Size partition_mem;
1504 tgl@sss.pgh.pa.us 1815 : 30127 : Size hash_mem_limit = get_hash_memory_limit();
1816 : :
1817 : : /* if not expected to spill, use all of hash_mem */
1818 [ + + ]: 30127 : if (input_groups * hashentrysize <= hash_mem_limit)
1819 : : {
1988 jdavis@postgresql.or 1820 [ + + ]: 28912 : if (num_partitions != NULL)
1821 : 16612 : *num_partitions = 0;
1504 tgl@sss.pgh.pa.us 1822 : 28912 : *mem_limit = hash_mem_limit;
1823 : 28912 : *ngroups_limit = hash_mem_limit / hashentrysize;
1998 jdavis@postgresql.or 1824 : 28912 : return;
1825 : : }
1826 : :
1827 : : /*
1828 : : * Calculate expected memory requirements for spilling, which is the size
1829 : : * of the buffers needed for all the tapes that need to be open at once.
1830 : : * Then, subtract that from the memory available for holding hash tables.
1831 : : */
1832 : 1215 : npartitions = hash_choose_num_partitions(input_groups,
1833 : : hashentrysize,
1834 : : used_bits,
1835 : : NULL);
1836 [ + + ]: 1215 : if (num_partitions != NULL)
1837 : 48 : *num_partitions = npartitions;
1838 : :
1839 : 1215 : partition_mem =
1840 : 1215 : HASHAGG_READ_BUFFER_SIZE +
1841 : : HASHAGG_WRITE_BUFFER_SIZE * npartitions;
1842 : :
1843 : : /*
1844 : : * Don't set the limit below 3/4 of hash_mem. In that case, we are at the
1845 : : * minimum number of partitions, so we aren't going to dramatically exceed
1846 : : * work mem anyway.
1847 : : */
1504 tgl@sss.pgh.pa.us 1848 [ - + ]: 1215 : if (hash_mem_limit > 4 * partition_mem)
1504 tgl@sss.pgh.pa.us 1849 :UBC 0 : *mem_limit = hash_mem_limit - partition_mem;
1850 : : else
1504 tgl@sss.pgh.pa.us 1851 :CBC 1215 : *mem_limit = hash_mem_limit * 0.75;
1852 : :
1998 jdavis@postgresql.or 1853 [ + - ]: 1215 : if (*mem_limit > hashentrysize)
1854 : 1215 : *ngroups_limit = *mem_limit / hashentrysize;
1855 : : else
1998 jdavis@postgresql.or 1856 :UBC 0 : *ngroups_limit = 1;
1857 : : }
1858 : :
1859 : : /*
1860 : : * hash_agg_check_limits
1861 : : *
1862 : : * After adding a new group to the hash table, check whether we need to enter
1863 : : * spill mode. Allocations may happen without adding new groups (for instance,
1864 : : * if the transition state size grows), so this check is imperfect.
1865 : : */
1866 : : static void
1998 jdavis@postgresql.or 1867 :CBC 256810 : hash_agg_check_limits(AggState *aggstate)
1868 : : {
1941 tgl@sss.pgh.pa.us 1869 : 256810 : uint64 ngroups = aggstate->hash_ngroups_current;
1870 : 256810 : Size meta_mem = MemoryContextMemAllocated(aggstate->hash_metacxt,
1871 : : true);
166 jdavis@postgresql.or 1872 : 256810 : Size entry_mem = MemoryContextMemAllocated(aggstate->hash_tablecxt,
1873 : : true);
1874 : 256810 : Size tval_mem = MemoryContextMemAllocated(aggstate->hashcontext->ecxt_per_tuple_memory,
1875 : : true);
1876 : 256810 : Size total_mem = meta_mem + entry_mem + tval_mem;
207 1877 : 256810 : bool do_spill = false;
1878 : :
1879 : : #ifdef USE_INJECTION_POINTS
1880 : : if (ngroups >= 1000)
1881 : : {
1882 : : if (IS_INJECTION_POINT_ATTACHED("hash-aggregate-spill-1000"))
1883 : : {
1884 : : do_spill = true;
1885 : : INJECTION_POINT_CACHED("hash-aggregate-spill-1000", NULL);
1886 : : }
1887 : : }
1888 : : #endif
1889 : :
1890 : : /*
1891 : : * Don't spill unless there's at least one group in the hash table so we
1892 : : * can be sure to make progress even in edge cases.
1893 : : */
1998 1894 [ + - ]: 256810 : if (aggstate->hash_ngroups_current > 0 &&
166 1895 [ + + ]: 256810 : (total_mem > aggstate->hash_mem_limit ||
1998 1896 [ + + ]: 243613 : ngroups > aggstate->hash_ngroups_limit))
1897 : : {
207 1898 : 13224 : do_spill = true;
1899 : : }
1900 : :
1901 [ + + ]: 256810 : if (do_spill)
1902 : 13224 : hash_agg_enter_spill_mode(aggstate);
1998 1903 : 256810 : }
1904 : :
1905 : : /*
1906 : : * Enter "spill mode", meaning that no new groups are added to any of the hash
1907 : : * tables. Tuples that would create a new group are instead spilled, and
1908 : : * processed later.
1909 : : */
1910 : : static void
1911 : 13224 : hash_agg_enter_spill_mode(AggState *aggstate)
1912 : : {
1913 : : INJECTION_POINT("hash-aggregate-enter-spill-mode", NULL);
1914 : 13224 : aggstate->hash_spill_mode = true;
1915 : 13224 : hashagg_recompile_expressions(aggstate, aggstate->table_filled, true);
1916 : :
1917 [ + + ]: 13224 : if (!aggstate->hash_ever_spilled)
1918 : : {
1419 heikki.linnakangas@i 1919 [ - + ]: 30 : Assert(aggstate->hash_tapeset == NULL);
1998 jdavis@postgresql.or 1920 [ - + ]: 30 : Assert(aggstate->hash_spills == NULL);
1921 : :
1922 : 30 : aggstate->hash_ever_spilled = true;
1923 : :
1419 heikki.linnakangas@i 1924 : 30 : aggstate->hash_tapeset = LogicalTapeSetCreate(true, NULL, -1);
1925 : :
1941 tgl@sss.pgh.pa.us 1926 : 30 : aggstate->hash_spills = palloc(sizeof(HashAggSpill) * aggstate->num_hashes);
1927 : :
1998 jdavis@postgresql.or 1928 [ + + ]: 90 : for (int setno = 0; setno < aggstate->num_hashes; setno++)
1929 : : {
1941 tgl@sss.pgh.pa.us 1930 : 60 : AggStatePerHash perhash = &aggstate->perhash[setno];
1931 : 60 : HashAggSpill *spill = &aggstate->hash_spills[setno];
1932 : :
1419 heikki.linnakangas@i 1933 : 60 : hashagg_spill_init(spill, aggstate->hash_tapeset, 0,
1998 jdavis@postgresql.or 1934 : 60 : perhash->aggnode->numGroups,
1935 : : aggstate->hashentrysize);
1936 : : }
1937 : : }
1938 : 13224 : }
1939 : :
1940 : : /*
1941 : : * Update metrics after filling the hash table.
1942 : : *
1943 : : * If reading from the outer plan, from_tape should be false; if reading from
1944 : : * another tape, from_tape should be true.
1945 : : */
1946 : : static void
1947 : 21678 : hash_agg_update_metrics(AggState *aggstate, bool from_tape, int npartitions)
1948 : : {
1949 : : Size meta_mem;
1950 : : Size entry_mem;
1951 : : Size hashkey_mem;
1952 : : Size buffer_mem;
1953 : : Size total_mem;
1954 : :
1955 [ + + ]: 21678 : if (aggstate->aggstrategy != AGG_MIXED &&
1956 [ - + ]: 8475 : aggstate->aggstrategy != AGG_HASHED)
1998 jdavis@postgresql.or 1957 :UBC 0 : return;
1958 : :
1959 : : /* memory for the hash table itself */
1998 jdavis@postgresql.or 1960 :CBC 21678 : meta_mem = MemoryContextMemAllocated(aggstate->hash_metacxt, true);
1961 : :
1962 : : /* memory for hash entries */
166 1963 : 21678 : entry_mem = MemoryContextMemAllocated(aggstate->hash_tablecxt, true);
1964 : :
1965 : : /* memory for byref transition states */
1877 pg@bowt.ie 1966 : 21678 : hashkey_mem = MemoryContextMemAllocated(aggstate->hashcontext->ecxt_per_tuple_memory, true);
1967 : :
1968 : : /* memory for read/write tape buffers, if spilled */
1998 jdavis@postgresql.or 1969 : 21678 : buffer_mem = npartitions * HASHAGG_WRITE_BUFFER_SIZE;
1970 [ + + ]: 21678 : if (from_tape)
1971 : 13467 : buffer_mem += HASHAGG_READ_BUFFER_SIZE;
1972 : :
1973 : : /* update peak mem */
166 1974 : 21678 : total_mem = meta_mem + entry_mem + hashkey_mem + buffer_mem;
1998 1975 [ + + ]: 21678 : if (total_mem > aggstate->hash_mem_peak)
1976 : 2082 : aggstate->hash_mem_peak = total_mem;
1977 : :
1978 : : /* update disk usage */
1419 heikki.linnakangas@i 1979 [ + + ]: 21678 : if (aggstate->hash_tapeset != NULL)
1980 : : {
1981 : 13497 : uint64 disk_used = LogicalTapeSetBlocks(aggstate->hash_tapeset) * (BLCKSZ / 1024);
1982 : :
1998 jdavis@postgresql.or 1983 [ + + ]: 13497 : if (aggstate->hash_disk_used < disk_used)
1984 : 24 : aggstate->hash_disk_used = disk_used;
1985 : : }
1986 : :
1987 : : /* update hashentrysize estimate based on contents */
1988 [ + + ]: 21678 : if (aggstate->hash_ngroups_current > 0)
1989 : : {
1990 : 21613 : aggstate->hashentrysize =
166 1991 : 21613 : TupleHashEntrySize() +
1877 pg@bowt.ie 1992 : 21613 : (hashkey_mem / (double) aggstate->hash_ngroups_current);
1993 : : }
1994 : : }
1995 : :
1996 : : /*
1997 : : * Create memory contexts used for hash aggregation.
1998 : : */
1999 : : static void
166 jdavis@postgresql.or 2000 : 2832 : hash_create_memory(AggState *aggstate)
2001 : : {
2002 : 2832 : Size maxBlockSize = ALLOCSET_DEFAULT_MAXSIZE;
2003 : :
2004 : : /*
2005 : : * The hashcontext's per-tuple memory will be used for byref transition
2006 : : * values and returned by AggCheckCallContext().
2007 : : */
2008 : 2832 : aggstate->hashcontext = CreateWorkExprContext(aggstate->ss.ps.state);
2009 : :
2010 : : /*
2011 : : * The meta context will be used for the bucket array of
2012 : : * TupleHashEntryData (or arrays, in the case of grouping sets). As the
2013 : : * hash table grows, the bucket array will double in size and the old one
2014 : : * will be freed, so an AllocSet is appropriate. For large bucket arrays,
2015 : : * the large allocation path will be used, so it's not worth worrying
2016 : : * about wasting space due to power-of-two allocations.
2017 : : */
2018 : 2832 : aggstate->hash_metacxt = AllocSetContextCreate(aggstate->ss.ps.state->es_query_cxt,
2019 : : "HashAgg meta context",
2020 : : ALLOCSET_DEFAULT_SIZES);
2021 : :
2022 : : /*
2023 : : * The hash entries themselves, which include the grouping key
2024 : : * (firstTuple) and pergroup data, are stored in the table context. The
2025 : : * bump allocator can be used because the entries are not freed until the
2026 : : * entire hash table is reset. The bump allocator is faster for
2027 : : * allocations and avoids wasting space on the chunk header or
2028 : : * power-of-two allocations.
2029 : : *
2030 : : * Like CreateWorkExprContext(), use smaller sizings for smaller work_mem,
2031 : : * to avoid large jumps in memory usage.
2032 : : */
2033 : :
2034 : : /*
2035 : : * Like CreateWorkExprContext(), use smaller sizings for smaller work_mem,
2036 : : * to avoid large jumps in memory usage.
2037 : : */
2038 : 2832 : maxBlockSize = pg_prevpower2_size_t(work_mem * (Size) 1024 / 16);
2039 : :
2040 : : /* But no bigger than ALLOCSET_DEFAULT_MAXSIZE */
2041 : 2832 : maxBlockSize = Min(maxBlockSize, ALLOCSET_DEFAULT_MAXSIZE);
2042 : :
2043 : : /* and no smaller than ALLOCSET_DEFAULT_INITSIZE */
2044 : 2832 : maxBlockSize = Max(maxBlockSize, ALLOCSET_DEFAULT_INITSIZE);
2045 : :
2046 : 2832 : aggstate->hash_tablecxt = BumpContextCreate(aggstate->ss.ps.state->es_query_cxt,
2047 : : "HashAgg table context",
2048 : : ALLOCSET_DEFAULT_MINSIZE,
2049 : : ALLOCSET_DEFAULT_INITSIZE,
2050 : : maxBlockSize);
2051 : :
2052 : 2832 : }
2053 : :
2054 : : /*
2055 : : * Choose a reasonable number of buckets for the initial hash table size.
2056 : : */
2057 : : static long
1998 2058 : 2312 : hash_choose_num_buckets(double hashentrysize, long ngroups, Size memory)
2059 : : {
2060 : : long max_nbuckets;
1941 tgl@sss.pgh.pa.us 2061 : 2312 : long nbuckets = ngroups;
2062 : :
1998 jdavis@postgresql.or 2063 : 2312 : max_nbuckets = memory / hashentrysize;
2064 : :
2065 : : /*
2066 : : * Underestimating is better than overestimating. Too many buckets crowd
2067 : : * out space for group keys and transition state values.
2068 : : */
1993 2069 : 2312 : max_nbuckets >>= 1;
2070 : :
1998 2071 [ + + ]: 2312 : if (nbuckets > max_nbuckets)
2072 : 36 : nbuckets = max_nbuckets;
2073 : :
1916 2074 : 2312 : return Max(nbuckets, 1);
2075 : : }
2076 : :
2077 : : /*
2078 : : * Determine the number of partitions to create when spilling, which will
2079 : : * always be a power of two. If log2_npartitions is non-NULL, set
2080 : : * *log2_npartitions to the log2() of the number of partitions.
2081 : : */
2082 : : static int
1866 2083 : 7521 : hash_choose_num_partitions(double input_groups, double hashentrysize,
2084 : : int used_bits, int *log2_npartitions)
2085 : : {
1504 tgl@sss.pgh.pa.us 2086 : 7521 : Size hash_mem_limit = get_hash_memory_limit();
2087 : : double partition_limit;
2088 : : double mem_wanted;
2089 : : double dpartitions;
2090 : : int npartitions;
2091 : : int partition_bits;
2092 : :
2093 : : /*
2094 : : * Avoid creating so many partitions that the memory requirements of the
2095 : : * open partition files are greater than 1/4 of hash_mem.
2096 : : */
1998 jdavis@postgresql.or 2097 : 7521 : partition_limit =
1504 tgl@sss.pgh.pa.us 2098 : 7521 : (hash_mem_limit * 0.25 - HASHAGG_READ_BUFFER_SIZE) /
2099 : : HASHAGG_WRITE_BUFFER_SIZE;
2100 : :
1998 jdavis@postgresql.or 2101 : 7521 : mem_wanted = HASHAGG_PARTITION_FACTOR * input_groups * hashentrysize;
2102 : :
2103 : : /* make enough partitions so that each one is likely to fit in memory */
1504 tgl@sss.pgh.pa.us 2104 : 7521 : dpartitions = 1 + (mem_wanted / hash_mem_limit);
2105 : :
2106 [ + + ]: 7521 : if (dpartitions > partition_limit)
2107 : 7488 : dpartitions = partition_limit;
2108 : :
2109 [ + - ]: 7521 : if (dpartitions < HASHAGG_MIN_PARTITIONS)
2110 : 7521 : dpartitions = HASHAGG_MIN_PARTITIONS;
2111 [ - + ]: 7521 : if (dpartitions > HASHAGG_MAX_PARTITIONS)
1504 tgl@sss.pgh.pa.us 2112 :UBC 0 : dpartitions = HASHAGG_MAX_PARTITIONS;
2113 : :
2114 : : /* HASHAGG_MAX_PARTITIONS limit makes this safe */
1504 tgl@sss.pgh.pa.us 2115 :CBC 7521 : npartitions = (int) dpartitions;
2116 : :
2117 : : /* ceil(log2(npartitions)) */
1998 jdavis@postgresql.or 2118 : 7521 : partition_bits = my_log2(npartitions);
2119 : :
2120 : : /* make sure that we don't exhaust the hash bits */
2121 [ - + ]: 7521 : if (partition_bits + used_bits >= 32)
1998 jdavis@postgresql.or 2122 :UBC 0 : partition_bits = 32 - used_bits;
2123 : :
1998 jdavis@postgresql.or 2124 [ + + ]:CBC 7521 : if (log2_npartitions != NULL)
2125 : 6306 : *log2_npartitions = partition_bits;
2126 : :
2127 : : /* number of partitions will be a power of two */
1504 tgl@sss.pgh.pa.us 2128 : 7521 : npartitions = 1 << partition_bits;
2129 : :
1998 jdavis@postgresql.or 2130 : 7521 : return npartitions;
2131 : : }
2132 : :
2133 : : /*
2134 : : * Initialize a freshly-created TupleHashEntry.
2135 : : */
2136 : : static void
1868 2137 : 256810 : initialize_hash_entry(AggState *aggstate, TupleHashTable hashtable,
2138 : : TupleHashEntry entry)
2139 : : {
2140 : : AggStatePerGroup pergroup;
2141 : : int transno;
2142 : :
2143 : 256810 : aggstate->hash_ngroups_current++;
2144 : 256810 : hash_agg_check_limits(aggstate);
2145 : :
2146 : : /* no need to allocate or initialize per-group state */
2147 [ + + ]: 256810 : if (aggstate->numtrans == 0)
2148 : 100420 : return;
2149 : :
166 2150 : 156390 : pergroup = (AggStatePerGroup) TupleHashEntryGetAdditional(hashtable, entry);
2151 : :
2152 : : /*
2153 : : * Initialize aggregates for new tuple group, lookup_hash_entries()
2154 : : * already has selected the relevant grouping set.
2155 : : */
1868 2156 [ + + ]: 386527 : for (transno = 0; transno < aggstate->numtrans; transno++)
2157 : : {
2158 : 230137 : AggStatePerTrans pertrans = &aggstate->pertrans[transno];
2159 : 230137 : AggStatePerGroup pergroupstate = &pergroup[transno];
2160 : :
2161 : 230137 : initialize_aggregate(aggstate, pertrans, pergroupstate);
2162 : : }
2163 : : }
2164 : :
2165 : : /*
2166 : : * Look up hash entries for the current tuple in all hashed grouping sets.
2167 : : *
2168 : : * Some entries may be left NULL if we are in "spill mode". The same tuple
2169 : : * will belong to different groups for each grouping set, so may match a group
2170 : : * already in memory for one set and match a group not in memory for another
2171 : : * set. When in "spill mode", the tuple will be spilled for each grouping set
2172 : : * where it doesn't match a group in memory.
2173 : : *
2174 : : * NB: It's possible to spill the same tuple for several different grouping
2175 : : * sets. This may seem wasteful, but it's actually a trade-off: if we spill
2176 : : * the tuple multiple times for multiple grouping sets, it can be partitioned
2177 : : * for each grouping set, making the refilling of the hash table very
2178 : : * efficient.
2179 : : */
2180 : : static void
3085 rhodiumtoad@postgres 2181 : 2951796 : lookup_hash_entries(AggState *aggstate)
2182 : : {
2183 : 2951796 : AggStatePerGroup *pergroup = aggstate->hash_pergroup;
1868 jdavis@postgresql.or 2184 : 2951796 : TupleTableSlot *outerslot = aggstate->tmpcontext->ecxt_outertuple;
2185 : : int setno;
2186 : :
1998 2187 [ + + ]: 5970818 : for (setno = 0; setno < aggstate->num_hashes; setno++)
2188 : : {
1941 tgl@sss.pgh.pa.us 2189 : 3019022 : AggStatePerHash perhash = &aggstate->perhash[setno];
1868 jdavis@postgresql.or 2190 : 3019022 : TupleHashTable hashtable = perhash->hashtable;
2191 : 3019022 : TupleTableSlot *hashslot = perhash->hashslot;
2192 : : TupleHashEntry entry;
2193 : : uint32 hash;
2194 : 3019022 : bool isnew = false;
2195 : : bool *p_isnew;
2196 : :
2197 : : /* if hash table already spilled, don't create new entries */
2198 [ + + ]: 3019022 : p_isnew = aggstate->hash_spill_mode ? NULL : &isnew;
2199 : :
3085 rhodiumtoad@postgres 2200 : 3019022 : select_current_set(aggstate, setno, true);
1868 jdavis@postgresql.or 2201 : 3019022 : prepare_hash_slot(perhash,
2202 : : outerslot,
2203 : : hashslot);
2204 : :
2205 : 3019022 : entry = LookupTupleHashEntry(hashtable, hashslot,
2206 : : p_isnew, &hash);
2207 : :
2208 [ + + ]: 3019022 : if (entry != NULL)
2209 : : {
2210 [ + + ]: 2638556 : if (isnew)
2211 : 182608 : initialize_hash_entry(aggstate, hashtable, entry);
166 2212 : 2638556 : pergroup[setno] = TupleHashEntryGetAdditional(hashtable, entry);
2213 : : }
2214 : : else
2215 : : {
1941 tgl@sss.pgh.pa.us 2216 : 380466 : HashAggSpill *spill = &aggstate->hash_spills[setno];
2217 : 380466 : TupleTableSlot *slot = aggstate->tmpcontext->ecxt_outertuple;
2218 : :
1998 jdavis@postgresql.or 2219 [ - + ]: 380466 : if (spill->partitions == NULL)
1419 heikki.linnakangas@i 2220 :UBC 0 : hashagg_spill_init(spill, aggstate->hash_tapeset, 0,
1998 jdavis@postgresql.or 2221 : 0 : perhash->aggnode->numGroups,
2222 : : aggstate->hashentrysize);
2223 : :
1882 jdavis@postgresql.or 2224 :CBC 380466 : hashagg_spill_tuple(aggstate, spill, slot, hash);
1868 2225 : 380466 : pergroup[setno] = NULL;
2226 : : }
2227 : : }
3085 rhodiumtoad@postgres 2228 : 2951796 : }
2229 : :
2230 : : /*
2231 : : * ExecAgg -
2232 : : *
2233 : : * ExecAgg receives tuples from its outer subplan and aggregates over
2234 : : * the appropriate attribute for each aggregate function use (Aggref
2235 : : * node) appearing in the targetlist or qual of the node. The number
2236 : : * of tuples to aggregate over depends on whether grouped or plain
2237 : : * aggregation is selected. In grouped aggregation, we produce a result
2238 : : * row for each group; in plain aggregation there's a single result row
2239 : : * for the whole query. In either case, the value of each aggregate is
2240 : : * stored in the expression context to be used when ExecProject evaluates
2241 : : * the result tuple.
2242 : : */
2243 : : static TupleTableSlot *
2973 andres@anarazel.de 2244 : 401909 : ExecAgg(PlanState *pstate)
2245 : : {
2246 : 401909 : AggState *node = castNode(AggState, pstate);
3085 rhodiumtoad@postgres 2247 : 401909 : TupleTableSlot *result = NULL;
2248 : :
2965 andres@anarazel.de 2249 [ + + ]: 401909 : CHECK_FOR_INTERRUPTS();
2250 : :
3766 2251 [ + + ]: 401909 : if (!node->agg_done)
2252 : : {
2253 : : /* Dispatch based on strategy */
3085 rhodiumtoad@postgres 2254 [ + + + - ]: 370260 : switch (node->phase->aggstrategy)
2255 : : {
3766 andres@anarazel.de 2256 : 234758 : case AGG_HASHED:
2257 [ + + ]: 234758 : if (!node->table_filled)
2258 : 8139 : agg_fill_hash_table(node);
2259 : : /* FALLTHROUGH */
2260 : : case AGG_MIXED:
2261 : 248439 : result = agg_retrieve_hash_table(node);
2262 : 248439 : break;
3085 rhodiumtoad@postgres 2263 : 121821 : case AGG_PLAIN:
2264 : : case AGG_SORTED:
3766 andres@anarazel.de 2265 : 121821 : result = agg_retrieve_direct(node);
2266 : 121736 : break;
2267 : : }
2268 : :
2269 [ + + + - ]: 370175 : if (!TupIsNull(result))
2270 : 361223 : return result;
2271 : : }
2272 : :
2273 : 40601 : return NULL;
2274 : : }
2275 : :
2276 : : /*
2277 : : * ExecAgg for non-hashed case
2278 : : */
2279 : : static TupleTableSlot *
8311 tgl@sss.pgh.pa.us 2280 : 121821 : agg_retrieve_direct(AggState *aggstate)
2281 : : {
3766 andres@anarazel.de 2282 : 121821 : Agg *node = aggstate->phase->aggnode;
2283 : : ExprContext *econtext;
2284 : : ExprContext *tmpcontext;
2285 : : AggStatePerAgg peragg;
2286 : : AggStatePerGroup *pergroups;
2287 : : TupleTableSlot *outerslot;
2288 : : TupleTableSlot *firstSlot;
2289 : : TupleTableSlot *result;
2290 : 121821 : bool hasGroupingSets = aggstate->phase->numsets > 0;
2291 : 121821 : int numGroupingSets = Max(aggstate->phase->numsets, 1);
2292 : : int currentSet;
2293 : : int nextSetSize;
2294 : : int numReset;
2295 : : int i;
2296 : :
2297 : : /*
2298 : : * get state info from node
2299 : : *
2300 : : * econtext is the per-output-tuple expression context
2301 : : *
2302 : : * tmpcontext is the per-input-tuple expression context
2303 : : */
8311 tgl@sss.pgh.pa.us 2304 : 121821 : econtext = aggstate->ss.ps.ps_ExprContext;
8340 2305 : 121821 : tmpcontext = aggstate->tmpcontext;
2306 : :
9477 2307 : 121821 : peragg = aggstate->peragg;
2804 andres@anarazel.de 2308 : 121821 : pergroups = aggstate->pergroups;
8311 tgl@sss.pgh.pa.us 2309 : 121821 : firstSlot = aggstate->ss.ss_ScanTupleSlot;
2310 : :
2311 : : /*
2312 : : * We loop retrieving groups until we find one matching
2313 : : * aggstate->ss.ps.qual
2314 : : *
2315 : : * For grouping sets, we have the invariant that aggstate->projected_set
2316 : : * is either -1 (initial call) or the index (starting from 0) in
2317 : : * gset_lengths for the group we just completed (either by projecting a
2318 : : * row or by discarding it in the qual).
2319 : : */
7728 2320 [ + + ]: 157223 : while (!aggstate->agg_done)
2321 : : {
2322 : : /*
2323 : : * Clear the per-output-tuple context for each group, as well as
2324 : : * aggcontext (which contains any pass-by-ref transvalues of the old
2325 : : * group). Some aggregate functions store working state in child
2326 : : * contexts; those now get reset automatically without us needing to
2327 : : * do anything special.
2328 : : *
2329 : : * We use ReScanExprContext not just ResetExprContext because we want
2330 : : * any registered shutdown callbacks to be called. That allows
2331 : : * aggregate functions to ensure they've cleaned up any non-memory
2332 : : * resources.
2333 : : */
3766 andres@anarazel.de 2334 : 157120 : ReScanExprContext(econtext);
2335 : :
2336 : : /*
2337 : : * Determine how many grouping sets need to be reset at this boundary.
2338 : : */
2339 [ + + ]: 157120 : if (aggstate->projected_set >= 0 &&
2340 [ + + ]: 123557 : aggstate->projected_set < numGroupingSets)
2341 : 123554 : numReset = aggstate->projected_set + 1;
2342 : : else
2343 : 33566 : numReset = numGroupingSets;
2344 : :
2345 : : /*
2346 : : * numReset can change on a phase boundary, but that's OK; we want to
2347 : : * reset the contexts used in _this_ phase, and later, after possibly
2348 : : * changing phase, initialize the right number of aggregates for the
2349 : : * _new_ phase.
2350 : : */
2351 : :
2352 [ + + ]: 325367 : for (i = 0; i < numReset; i++)
2353 : : {
2354 : 168247 : ReScanExprContext(aggstate->aggcontexts[i]);
2355 : : }
2356 : :
2357 : : /*
2358 : : * Check if input is complete and there are no more groups to project
2359 : : * in this phase; move to next phase or mark as done.
2360 : : */
2361 [ + + ]: 157120 : if (aggstate->input_done == true &&
2362 [ + + ]: 789 : aggstate->projected_set >= (numGroupingSets - 1))
2363 : : {
2364 [ + + ]: 384 : if (aggstate->current_phase < aggstate->numphases - 1)
2365 : : {
2366 : 99 : initialize_phase(aggstate, aggstate->current_phase + 1);
2367 : 99 : aggstate->input_done = false;
2368 : 99 : aggstate->projected_set = -1;
2369 : 99 : numGroupingSets = Max(aggstate->phase->numsets, 1);
2370 : 99 : node = aggstate->phase->aggnode;
2371 : 99 : numReset = numGroupingSets;
2372 : : }
3085 rhodiumtoad@postgres 2373 [ + + ]: 285 : else if (aggstate->aggstrategy == AGG_MIXED)
2374 : : {
2375 : : /*
2376 : : * Mixed mode; we've output all the grouped stuff and have
2377 : : * full hashtables, so switch to outputting those.
2378 : : */
2379 : 78 : initialize_phase(aggstate, 0);
2380 : 78 : aggstate->table_filled = true;
2381 : 78 : ResetTupleHashIterator(aggstate->perhash[0].hashtable,
2382 : : &aggstate->perhash[0].hashiter);
2383 : 78 : select_current_set(aggstate, 0, true);
2384 : 78 : return agg_retrieve_hash_table(aggstate);
2385 : : }
2386 : : else
2387 : : {
8340 tgl@sss.pgh.pa.us 2388 : 207 : aggstate->agg_done = true;
3766 andres@anarazel.de 2389 : 207 : break;
2390 : : }
2391 : : }
2392 : :
2393 : : /*
2394 : : * Get the number of columns in the next grouping set after the last
2395 : : * projected one (if any). This is the number of columns to compare to
2396 : : * see if we reached the boundary of that set too.
2397 : : */
2398 [ + + ]: 156835 : if (aggstate->projected_set >= 0 &&
2399 [ + + ]: 123173 : aggstate->projected_set < (numGroupingSets - 1))
2400 : 13641 : nextSetSize = aggstate->phase->gset_lengths[aggstate->projected_set + 1];
2401 : : else
2402 : 143194 : nextSetSize = 0;
2403 : :
2404 : : /*----------
2405 : : * If a subgroup for the current grouping set is present, project it.
2406 : : *
2407 : : * We have a new group if:
2408 : : * - we're out of input but haven't projected all grouping sets
2409 : : * (checked above)
2410 : : * OR
2411 : : * - we already projected a row that wasn't from the last grouping
2412 : : * set
2413 : : * AND
2414 : : * - the next grouping set has at least one grouping column (since
2415 : : * empty grouping sets project only once input is exhausted)
2416 : : * AND
2417 : : * - the previous and pending rows differ on the grouping columns
2418 : : * of the next grouping set
2419 : : *----------
2420 : : */
2760 2421 : 156835 : tmpcontext->ecxt_innertuple = econtext->ecxt_outertuple;
3766 2422 [ + + ]: 156835 : if (aggstate->input_done ||
3085 rhodiumtoad@postgres 2423 [ + + ]: 156430 : (node->aggstrategy != AGG_PLAIN &&
3766 andres@anarazel.de 2424 [ + + ]: 123634 : aggstate->projected_set != -1 &&
2425 [ + + + + ]: 122768 : aggstate->projected_set < (numGroupingSets - 1) &&
2426 : 9970 : nextSetSize > 0 &&
2760 2427 [ + + ]: 9970 : !ExecQualAndReset(aggstate->phase->eqfunctions[nextSetSize - 1],
2428 : : tmpcontext)))
2429 : : {
3766 2430 : 7072 : aggstate->projected_set += 1;
2431 : :
2432 [ - + ]: 7072 : Assert(aggstate->projected_set < numGroupingSets);
2433 [ + + - + ]: 7072 : Assert(nextSetSize > 0 || aggstate->input_done);
2434 : : }
2435 : : else
2436 : : {
2437 : : /*
2438 : : * We no longer care what group we just projected, the next
2439 : : * projection will always be the first (or only) grouping set
2440 : : * (unless the input proves to be empty).
2441 : : */
2442 : 149763 : aggstate->projected_set = 0;
2443 : :
2444 : : /*
2445 : : * If we don't already have the first tuple of the new group,
2446 : : * fetch it from the outer plan.
2447 : : */
2448 [ + + ]: 149763 : if (aggstate->grp_firstTuple == NULL)
2449 : : {
2450 : 33662 : outerslot = fetch_input_tuple(aggstate);
2451 [ + + + + ]: 33638 : if (!TupIsNull(outerslot))
2452 : : {
2453 : : /*
2454 : : * Make a copy of the first input tuple; we will use this
2455 : : * for comparisons (in group mode) and for projection.
2456 : : */
2486 2457 : 27007 : aggstate->grp_firstTuple = ExecCopySlotHeapTuple(outerslot);
2458 : : }
2459 : : else
2460 : : {
2461 : : /* outer plan produced no tuples at all */
3766 2462 [ + + ]: 6631 : if (hasGroupingSets)
2463 : : {
2464 : : /*
2465 : : * If there was no input at all, we need to project
2466 : : * rows only if there are grouping sets of size 0.
2467 : : * Note that this implies that there can't be any
2468 : : * references to ungrouped Vars, which would otherwise
2469 : : * cause issues with the empty output slot.
2470 : : *
2471 : : * XXX: This is no longer true, we currently deal with
2472 : : * this in finalize_aggregates().
2473 : : */
2474 : 27 : aggstate->input_done = true;
2475 : :
2476 [ + + ]: 39 : while (aggstate->phase->gset_lengths[aggstate->projected_set] > 0)
2477 : : {
2478 : 15 : aggstate->projected_set += 1;
2479 [ + + ]: 15 : if (aggstate->projected_set >= numGroupingSets)
2480 : : {
2481 : : /*
2482 : : * We can't set agg_done here because we might
2483 : : * have more phases to do, even though the
2484 : : * input is empty. So we need to restart the
2485 : : * whole outer loop.
2486 : : */
2487 : 3 : break;
2488 : : }
2489 : : }
2490 : :
2491 [ + + ]: 27 : if (aggstate->projected_set >= numGroupingSets)
2492 : 3 : continue;
2493 : : }
2494 : : else
2495 : : {
2496 : 6604 : aggstate->agg_done = true;
2497 : : /* If we are grouping, we should produce no tuples too */
2498 [ + + ]: 6604 : if (node->aggstrategy != AGG_PLAIN)
2499 : 33 : return NULL;
2500 : : }
2501 : : }
2502 : : }
2503 : :
2504 : : /*
2505 : : * Initialize working state for a new input tuple group.
2506 : : */
2804 2507 : 149703 : initialize_aggregates(aggstate, pergroups, numReset);
2508 : :
3766 2509 [ + + ]: 149703 : if (aggstate->grp_firstTuple != NULL)
2510 : : {
2511 : : /*
2512 : : * Store the copied first input tuple in the tuple table slot
2513 : : * reserved for it. The tuple will be deleted when it is
2514 : : * cleared from the slot.
2515 : : */
2486 2516 : 143108 : ExecForceStoreHeapTuple(aggstate->grp_firstTuple,
2517 : : firstSlot, true);
2999 tgl@sss.pgh.pa.us 2518 : 143108 : aggstate->grp_firstTuple = NULL; /* don't keep two pointers */
2519 : :
2520 : : /* set up for first advance_aggregates call */
3766 andres@anarazel.de 2521 : 143108 : tmpcontext->ecxt_outertuple = firstSlot;
2522 : :
2523 : : /*
2524 : : * Process each outer-plan tuple, and then fetch the next one,
2525 : : * until we exhaust the outer plan or cross a group boundary.
2526 : : */
2527 : : for (;;)
2528 : : {
2529 : : /*
2530 : : * During phase 1 only of a mixed agg, we need to update
2531 : : * hashtables as well in advance_aggregates.
2532 : : */
3085 rhodiumtoad@postgres 2533 [ + + ]: 10683186 : if (aggstate->aggstrategy == AGG_MIXED &&
2534 [ + - ]: 19031 : aggstate->current_phase == 1)
2535 : : {
2797 andres@anarazel.de 2536 : 19031 : lookup_hash_entries(aggstate);
2537 : : }
2538 : :
2539 : : /* Advance the aggregates (or combine functions) */
2540 : 10683186 : advance_aggregates(aggstate);
2541 : :
2542 : : /* Reset per-input-tuple context after each tuple */
3766 2543 : 10683147 : ResetExprContext(tmpcontext);
2544 : :
2545 : 10683147 : outerslot = fetch_input_tuple(aggstate);
2546 [ + + + + ]: 10683137 : if (TupIsNull(outerslot))
2547 : : {
2548 : : /* no more outer-plan tuples available */
2549 : :
2550 : : /* if we built hash tables, finalize any spills */
1998 jdavis@postgresql.or 2551 [ + + ]: 26955 : if (aggstate->aggstrategy == AGG_MIXED &&
2552 [ + - ]: 72 : aggstate->current_phase == 1)
2553 : 72 : hashagg_finish_initial_spills(aggstate);
2554 : :
3766 andres@anarazel.de 2555 [ + + ]: 26955 : if (hasGroupingSets)
2556 : : {
2557 : 357 : aggstate->input_done = true;
2558 : 357 : break;
2559 : : }
2560 : : else
2561 : : {
2562 : 26598 : aggstate->agg_done = true;
2563 : 26598 : break;
2564 : : }
2565 : : }
2566 : : /* set up for next advance_aggregates call */
2567 : 10656182 : tmpcontext->ecxt_outertuple = outerslot;
2568 : :
2569 : : /*
2570 : : * If we are grouping, check whether we've crossed a group
2571 : : * boundary.
2572 : : */
962 tgl@sss.pgh.pa.us 2573 [ + + + + ]: 10656182 : if (node->aggstrategy != AGG_PLAIN && node->numCols > 0)
2574 : : {
2760 andres@anarazel.de 2575 : 1022097 : tmpcontext->ecxt_innertuple = firstSlot;
2576 [ + + ]: 1022097 : if (!ExecQual(aggstate->phase->eqfunctions[node->numCols - 1],
2577 : : tmpcontext))
2578 : : {
2486 2579 : 116104 : aggstate->grp_firstTuple = ExecCopySlotHeapTuple(outerslot);
3766 2580 : 116104 : break;
2581 : : }
2582 : : }
2583 : : }
2584 : : }
2585 : :
2586 : : /*
2587 : : * Use the representative input tuple for any references to
2588 : : * non-aggregated input columns in aggregate direct args, the node
2589 : : * qual, and the tlist. (If we are not grouping, and there are no
2590 : : * input rows at all, we will come here with an empty firstSlot
2591 : : * ... but if not grouping, there can't be any references to
2592 : : * non-aggregated input columns, so no problem.)
2593 : : */
2594 : 149654 : econtext->ecxt_outertuple = firstSlot;
2595 : : }
2596 : :
2597 [ - + ]: 156726 : Assert(aggstate->projected_set >= 0);
2598 : :
2599 : 156726 : currentSet = aggstate->projected_set;
2600 : :
2601 : 156726 : prepare_projection_slot(aggstate, econtext->ecxt_outertuple, currentSet);
2602 : :
3085 rhodiumtoad@postgres 2603 : 156726 : select_current_set(aggstate, currentSet, false);
2604 : :
2605 : 156726 : finalize_aggregates(aggstate,
2606 : : peragg,
2804 andres@anarazel.de 2607 : 156726 : pergroups[currentSet]);
2608 : :
2609 : : /*
2610 : : * If there's no row to project right now, we must continue rather
2611 : : * than returning a null since there might be more groups.
2612 : : */
3766 2613 : 156720 : result = project_aggregates(aggstate);
2614 [ + + ]: 156714 : if (result)
2615 : 121315 : return result;
2616 : : }
2617 : :
2618 : : /* No more groups */
7728 tgl@sss.pgh.pa.us 2619 : 310 : return NULL;
2620 : : }
2621 : :
2622 : : /*
2623 : : * ExecAgg for hashed case: read input and build hash table
2624 : : */
2625 : : static void
8311 2626 : 8139 : agg_fill_hash_table(AggState *aggstate)
2627 : : {
2628 : : TupleTableSlot *outerslot;
3085 rhodiumtoad@postgres 2629 : 8139 : ExprContext *tmpcontext = aggstate->tmpcontext;
2630 : :
2631 : : /*
2632 : : * Process each outer-plan tuple, and then fetch the next one, until we
2633 : : * exhaust the outer plan.
2634 : : */
2635 : : for (;;)
2636 : : {
3766 andres@anarazel.de 2637 : 2940904 : outerslot = fetch_input_tuple(aggstate);
8340 tgl@sss.pgh.pa.us 2638 [ + + + + ]: 2940904 : if (TupIsNull(outerslot))
2639 : : break;
2640 : :
2641 : : /* set up for lookup_hash_entries and advance_aggregates */
6771 2642 : 2932765 : tmpcontext->ecxt_outertuple = outerslot;
2643 : :
2644 : : /* Find or build hashtable entries */
2797 andres@anarazel.de 2645 : 2932765 : lookup_hash_entries(aggstate);
2646 : :
2647 : : /* Advance the aggregates (or combine functions) */
2648 : 2932765 : advance_aggregates(aggstate);
2649 : :
2650 : : /*
2651 : : * Reset per-input-tuple context after each tuple, but note that the
2652 : : * hash lookups do this too
2653 : : */
3085 rhodiumtoad@postgres 2654 : 2932765 : ResetExprContext(aggstate->tmpcontext);
2655 : : }
2656 : :
2657 : : /* finalize spills, if any */
1998 jdavis@postgresql.or 2658 : 8139 : hashagg_finish_initial_spills(aggstate);
2659 : :
8340 tgl@sss.pgh.pa.us 2660 : 8139 : aggstate->table_filled = true;
2661 : : /* Initialize to walk the first hash table */
3085 rhodiumtoad@postgres 2662 : 8139 : select_current_set(aggstate, 0, true);
2663 : 8139 : ResetTupleHashIterator(aggstate->perhash[0].hashtable,
2664 : : &aggstate->perhash[0].hashiter);
8340 tgl@sss.pgh.pa.us 2665 : 8139 : }
2666 : :
2667 : : /*
2668 : : * If any data was spilled during hash aggregation, reset the hash table and
2669 : : * reprocess one batch of spilled data. After reprocessing a batch, the hash
2670 : : * table will again contain data, ready to be consumed by
2671 : : * agg_retrieve_hash_table_in_memory().
2672 : : *
2673 : : * Should only be called after all in memory hash table entries have been
2674 : : * finalized and emitted.
2675 : : *
2676 : : * Return false when input is exhausted and there's no more work to be done;
2677 : : * otherwise return true.
2678 : : */
2679 : : static bool
1998 jdavis@postgresql.or 2680 : 22076 : agg_refill_hash_table(AggState *aggstate)
2681 : : {
2682 : : HashAggBatch *batch;
2683 : : AggStatePerHash perhash;
2684 : : HashAggSpill spill;
1419 heikki.linnakangas@i 2685 : 22076 : LogicalTapeSet *tapeset = aggstate->hash_tapeset;
1941 tgl@sss.pgh.pa.us 2686 : 22076 : bool spill_initialized = false;
2687 : :
1998 jdavis@postgresql.or 2688 [ + + ]: 22076 : if (aggstate->hash_batches == NIL)
2689 : 8609 : return false;
2690 : :
2691 : : /* hash_batches is a stack, with the top item at the end of the list */
1405 tgl@sss.pgh.pa.us 2692 : 13467 : batch = llast(aggstate->hash_batches);
2693 : 13467 : aggstate->hash_batches = list_delete_last(aggstate->hash_batches);
2694 : :
1866 jdavis@postgresql.or 2695 : 13467 : hash_agg_set_limits(aggstate->hashentrysize, batch->input_card,
2696 : : batch->used_bits, &aggstate->hash_mem_limit,
2697 : : &aggstate->hash_ngroups_limit, NULL);
2698 : :
2699 : : /*
2700 : : * Each batch only processes one grouping set; set the rest to NULL so
2701 : : * that advance_aggregates() knows to ignore them. We don't touch
2702 : : * pergroups for sorted grouping sets here, because they will be needed if
2703 : : * we rescan later. The expressions for sorted grouping sets will not be
2704 : : * evaluated after we recompile anyway.
2705 : : */
1715 2706 [ + - + - : 103704 : MemSet(aggstate->hash_pergroup, 0,
+ - + - +
+ ]
2707 : : sizeof(AggStatePerGroup) * aggstate->num_hashes);
2708 : :
2709 : : /* free memory and reset hash tables */
1998 2710 : 13467 : ReScanExprContext(aggstate->hashcontext);
166 2711 : 13467 : MemoryContextReset(aggstate->hash_tablecxt);
1998 2712 [ + + ]: 103704 : for (int setno = 0; setno < aggstate->num_hashes; setno++)
2713 : 90237 : ResetTupleHashTable(aggstate->perhash[setno].hashtable);
2714 : :
2715 : 13467 : aggstate->hash_ngroups_current = 0;
2716 : :
2717 : : /*
2718 : : * In AGG_MIXED mode, hash aggregation happens in phase 1 and the output
2719 : : * happens in phase 0. So, we switch to phase 1 when processing a batch,
2720 : : * and back to phase 0 after the batch is done.
2721 : : */
2722 [ - + ]: 13467 : Assert(aggstate->current_phase == 0);
2723 [ + + ]: 13467 : if (aggstate->phase->aggstrategy == AGG_MIXED)
2724 : : {
2725 : 13131 : aggstate->current_phase = 1;
2726 : 13131 : aggstate->phase = &aggstate->phases[aggstate->current_phase];
2727 : : }
2728 : :
2729 : 13467 : select_current_set(aggstate, batch->setno, true);
2730 : :
1868 2731 : 13467 : perhash = &aggstate->perhash[aggstate->current_set];
2732 : :
2733 : : /*
2734 : : * Spilled tuples are always read back as MinimalTuples, which may be
2735 : : * different from the outer plan, so recompile the aggregate expressions.
2736 : : *
2737 : : * We still need the NULL check, because we are only processing one
2738 : : * grouping set at a time and the rest will be NULL.
2739 : : */
1998 2740 : 13467 : hashagg_recompile_expressions(aggstate, true, true);
2741 : :
2742 : : INJECTION_POINT("hash-aggregate-process-batch", NULL);
2743 : : for (;;)
1941 tgl@sss.pgh.pa.us 2744 : 592812 : {
1868 jdavis@postgresql.or 2745 : 606279 : TupleTableSlot *spillslot = aggstate->hash_spill_rslot;
2746 : 606279 : TupleTableSlot *hashslot = perhash->hashslot;
166 2747 : 606279 : TupleHashTable hashtable = perhash->hashtable;
2748 : : TupleHashEntry entry;
2749 : : MinimalTuple tuple;
2750 : : uint32 hash;
1868 2751 : 606279 : bool isnew = false;
2752 [ + + ]: 606279 : bool *p_isnew = aggstate->hash_spill_mode ? NULL : &isnew;
2753 : :
1998 2754 [ - + ]: 606279 : CHECK_FOR_INTERRUPTS();
2755 : :
2756 : 606279 : tuple = hashagg_batch_read(batch, &hash);
2757 [ + + ]: 606279 : if (tuple == NULL)
2758 : 13467 : break;
2759 : :
1868 2760 : 592812 : ExecStoreMinimalTuple(tuple, spillslot, true);
2761 : 592812 : aggstate->tmpcontext->ecxt_outertuple = spillslot;
2762 : :
2763 : 592812 : prepare_hash_slot(perhash,
2764 : 592812 : aggstate->tmpcontext->ecxt_outertuple,
2765 : : hashslot);
166 2766 : 592812 : entry = LookupTupleHashEntryHash(hashtable, hashslot,
2767 : : p_isnew, hash);
2768 : :
1868 2769 [ + + ]: 592812 : if (entry != NULL)
2770 : : {
2771 [ + + ]: 380466 : if (isnew)
166 2772 : 74202 : initialize_hash_entry(aggstate, hashtable, entry);
2773 : 380466 : aggstate->hash_pergroup[batch->setno] = TupleHashEntryGetAdditional(hashtable, entry);
1998 2774 : 380466 : advance_aggregates(aggstate);
2775 : : }
2776 : : else
2777 : : {
2778 [ + + ]: 212346 : if (!spill_initialized)
2779 : : {
2780 : : /*
2781 : : * Avoid initializing the spill until we actually need it so
2782 : : * that we don't assign tapes that will never be used.
2783 : : */
2784 : 6246 : spill_initialized = true;
1419 heikki.linnakangas@i 2785 : 6246 : hashagg_spill_init(&spill, tapeset, batch->used_bits,
2786 : : batch->input_card, aggstate->hashentrysize);
2787 : : }
2788 : : /* no memory for a new group, spill */
1868 jdavis@postgresql.or 2789 : 212346 : hashagg_spill_tuple(aggstate, &spill, spillslot, hash);
2790 : :
2791 : 212346 : aggstate->hash_pergroup[batch->setno] = NULL;
2792 : : }
2793 : :
2794 : : /*
2795 : : * Reset per-input-tuple context after each tuple, but note that the
2796 : : * hash lookups do this too
2797 : : */
1998 2798 : 592812 : ResetExprContext(aggstate->tmpcontext);
2799 : : }
2800 : :
1419 heikki.linnakangas@i 2801 : 13467 : LogicalTapeClose(batch->input_tape);
2802 : :
2803 : : /* change back to phase 0 */
1998 jdavis@postgresql.or 2804 : 13467 : aggstate->current_phase = 0;
2805 : 13467 : aggstate->phase = &aggstate->phases[aggstate->current_phase];
2806 : :
2807 [ + + ]: 13467 : if (spill_initialized)
2808 : : {
2809 : 6246 : hashagg_spill_finish(aggstate, &spill, batch->setno);
1817 2810 : 6246 : hash_agg_update_metrics(aggstate, true, spill.npartitions);
2811 : : }
2812 : : else
1998 2813 : 7221 : hash_agg_update_metrics(aggstate, true, 0);
2814 : :
2815 : 13467 : aggstate->hash_spill_mode = false;
2816 : :
2817 : : /* prepare to walk the first hash table */
2818 : 13467 : select_current_set(aggstate, batch->setno, true);
2819 : 13467 : ResetTupleHashIterator(aggstate->perhash[batch->setno].hashtable,
2820 : : &aggstate->perhash[batch->setno].hashiter);
2821 : :
2822 : 13467 : pfree(batch);
2823 : :
2824 : 13467 : return true;
2825 : : }
2826 : :
2827 : : /*
2828 : : * ExecAgg for hashed case: retrieving groups from hash table
2829 : : *
2830 : : * After exhausting in-memory tuples, also try refilling the hash table using
2831 : : * previously-spilled tuples. Only returns NULL after all in-memory and
2832 : : * spilled tuples are exhausted.
2833 : : */
2834 : : static TupleTableSlot *
8311 tgl@sss.pgh.pa.us 2835 : 248517 : agg_retrieve_hash_table(AggState *aggstate)
2836 : : {
1998 jdavis@postgresql.or 2837 : 248517 : TupleTableSlot *result = NULL;
2838 : :
2839 [ + + ]: 501892 : while (result == NULL)
2840 : : {
2841 : 261984 : result = agg_retrieve_hash_table_in_memory(aggstate);
2842 [ + + ]: 261984 : if (result == NULL)
2843 : : {
2844 [ + + ]: 22076 : if (!agg_refill_hash_table(aggstate))
2845 : : {
2846 : 8609 : aggstate->agg_done = true;
2847 : 8609 : break;
2848 : : }
2849 : : }
2850 : : }
2851 : :
2852 : 248517 : return result;
2853 : : }
2854 : :
2855 : : /*
2856 : : * Retrieve the groups from the in-memory hash tables without considering any
2857 : : * spilled tuples.
2858 : : */
2859 : : static TupleTableSlot *
2860 : 261984 : agg_retrieve_hash_table_in_memory(AggState *aggstate)
2861 : : {
2862 : : ExprContext *econtext;
2863 : : AggStatePerAgg peragg;
2864 : : AggStatePerGroup pergroup;
2865 : : TupleHashEntry entry;
2866 : : TupleTableSlot *firstSlot;
2867 : : TupleTableSlot *result;
2868 : : AggStatePerHash perhash;
2869 : :
2870 : : /*
2871 : : * get state info from node.
2872 : : *
2873 : : * econtext is the per-output-tuple expression context.
2874 : : */
8311 tgl@sss.pgh.pa.us 2875 : 261984 : econtext = aggstate->ss.ps.ps_ExprContext;
8340 2876 : 261984 : peragg = aggstate->peragg;
8311 2877 : 261984 : firstSlot = aggstate->ss.ss_ScanTupleSlot;
2878 : :
2879 : : /*
2880 : : * Note that perhash (and therefore anything accessed through it) can
2881 : : * change inside the loop, as we change between grouping sets.
2882 : : */
3085 rhodiumtoad@postgres 2883 : 261984 : perhash = &aggstate->perhash[aggstate->current_set];
2884 : :
2885 : : /*
2886 : : * We loop retrieving groups until we find one satisfying
2887 : : * aggstate->ss.ps.qual
2888 : : */
2889 : : for (;;)
8340 tgl@sss.pgh.pa.us 2890 : 67977 : {
3085 rhodiumtoad@postgres 2891 : 329961 : TupleTableSlot *hashslot = perhash->hashslot;
166 jdavis@postgresql.or 2892 : 329961 : TupleHashTable hashtable = perhash->hashtable;
2893 : : int i;
2894 : :
2965 andres@anarazel.de 2895 [ - + ]: 329961 : CHECK_FOR_INTERRUPTS();
2896 : :
2897 : : /*
2898 : : * Find the next entry in the hash table
2899 : : */
166 jdavis@postgresql.or 2900 : 329961 : entry = ScanTupleHashTable(hashtable, &perhash->hashiter);
8275 tgl@sss.pgh.pa.us 2901 [ + + ]: 329961 : if (entry == NULL)
2902 : : {
3085 rhodiumtoad@postgres 2903 : 72222 : int nextset = aggstate->current_set + 1;
2904 : :
2905 [ + + ]: 72222 : if (nextset < aggstate->num_hashes)
2906 : : {
2907 : : /*
2908 : : * Switch to next grouping set, reinitialize, and restart the
2909 : : * loop.
2910 : : */
2911 : 50146 : select_current_set(aggstate, nextset, true);
2912 : :
2913 : 50146 : perhash = &aggstate->perhash[aggstate->current_set];
2914 : :
166 jdavis@postgresql.or 2915 : 50146 : ResetTupleHashIterator(hashtable, &perhash->hashiter);
2916 : :
3085 rhodiumtoad@postgres 2917 : 50146 : continue;
2918 : : }
2919 : : else
2920 : : {
2921 : 22076 : return NULL;
2922 : : }
2923 : : }
2924 : :
2925 : : /*
2926 : : * Clear the per-output-tuple context for each group
2927 : : *
2928 : : * We intentionally don't use ReScanExprContext here; if any aggs have
2929 : : * registered shutdown callbacks, they mustn't be called yet, since we
2930 : : * might not be done with that agg.
2931 : : */
8340 tgl@sss.pgh.pa.us 2932 : 257739 : ResetExprContext(econtext);
2933 : :
2934 : : /*
2935 : : * Transform representative tuple back into one with the right
2936 : : * columns.
2937 : : */
166 jdavis@postgresql.or 2938 : 257739 : ExecStoreMinimalTuple(TupleHashEntryGetTuple(entry), hashslot, false);
3202 andres@anarazel.de 2939 : 257739 : slot_getallattrs(hashslot);
2940 : :
2941 : 257739 : ExecClearTuple(firstSlot);
2942 : 257739 : memset(firstSlot->tts_isnull, true,
2943 : 257739 : firstSlot->tts_tupleDescriptor->natts * sizeof(bool));
2944 : :
3085 rhodiumtoad@postgres 2945 [ + + ]: 681274 : for (i = 0; i < perhash->numhashGrpCols; i++)
2946 : : {
2947 : 423535 : int varNumber = perhash->hashGrpColIdxInput[i] - 1;
2948 : :
3202 andres@anarazel.de 2949 : 423535 : firstSlot->tts_values[varNumber] = hashslot->tts_values[i];
2950 : 423535 : firstSlot->tts_isnull[varNumber] = hashslot->tts_isnull[i];
2951 : : }
2952 : 257739 : ExecStoreVirtualTuple(firstSlot);
2953 : :
166 jdavis@postgresql.or 2954 : 257739 : pergroup = (AggStatePerGroup) TupleHashEntryGetAdditional(hashtable, entry);
2955 : :
2956 : : /*
2957 : : * Use the representative input tuple for any references to
2958 : : * non-aggregated input columns in the qual and tlist.
2959 : : */
6771 tgl@sss.pgh.pa.us 2960 : 257739 : econtext->ecxt_outertuple = firstSlot;
2961 : :
3085 rhodiumtoad@postgres 2962 : 257739 : prepare_projection_slot(aggstate,
2963 : : econtext->ecxt_outertuple,
2964 : : aggstate->current_set);
2965 : :
2966 : 257739 : finalize_aggregates(aggstate, peragg, pergroup);
2967 : :
3766 andres@anarazel.de 2968 : 257739 : result = project_aggregates(aggstate);
2969 [ + + ]: 257739 : if (result)
2970 : 239908 : return result;
2971 : : }
2972 : :
2973 : : /* No more groups */
2974 : : return NULL;
2975 : : }
2976 : :
2977 : : /*
2978 : : * hashagg_spill_init
2979 : : *
2980 : : * Called after we determined that spilling is necessary. Chooses the number
2981 : : * of partitions to create, and initializes them.
2982 : : */
2983 : : static void
1419 heikki.linnakangas@i 2984 : 6306 : hashagg_spill_init(HashAggSpill *spill, LogicalTapeSet *tapeset, int used_bits,
2985 : : double input_groups, double hashentrysize)
2986 : : {
2987 : : int npartitions;
2988 : : int partition_bits;
2989 : :
1941 tgl@sss.pgh.pa.us 2990 : 6306 : npartitions = hash_choose_num_partitions(input_groups, hashentrysize,
2991 : : used_bits, &partition_bits);
2992 : :
2993 : : #ifdef USE_INJECTION_POINTS
2994 : : if (IS_INJECTION_POINT_ATTACHED("hash-aggregate-single-partition"))
2995 : : {
2996 : : npartitions = 1;
2997 : : partition_bits = 0;
2998 : : INJECTION_POINT_CACHED("hash-aggregate-single-partition", NULL);
2999 : : }
3000 : : #endif
3001 : :
1419 heikki.linnakangas@i 3002 : 6306 : spill->partitions = palloc0(sizeof(LogicalTape *) * npartitions);
1998 jdavis@postgresql.or 3003 : 6306 : spill->ntuples = palloc0(sizeof(int64) * npartitions);
1866 3004 : 6306 : spill->hll_card = palloc0(sizeof(hyperLogLogState) * npartitions);
3005 : :
1419 heikki.linnakangas@i 3006 [ + + ]: 31530 : for (int i = 0; i < npartitions; i++)
3007 : 25224 : spill->partitions[i] = LogicalTapeCreate(tapeset);
3008 : :
1998 jdavis@postgresql.or 3009 : 6306 : spill->shift = 32 - used_bits - partition_bits;
207 3010 [ + - ]: 6306 : if (spill->shift < 32)
3011 : 6306 : spill->mask = (npartitions - 1) << spill->shift;
3012 : : else
207 jdavis@postgresql.or 3013 :UBC 0 : spill->mask = 0;
1998 jdavis@postgresql.or 3014 :CBC 6306 : spill->npartitions = npartitions;
3015 : :
1866 3016 [ + + ]: 31530 : for (int i = 0; i < npartitions; i++)
3017 : 25224 : initHyperLogLog(&spill->hll_card[i], HASHAGG_HLL_BIT_WIDTH);
1998 3018 : 6306 : }
3019 : :
3020 : : /*
3021 : : * hashagg_spill_tuple
3022 : : *
3023 : : * No room for new groups in the hash table. Save for later in the appropriate
3024 : : * partition.
3025 : : */
3026 : : static Size
1882 3027 : 592812 : hashagg_spill_tuple(AggState *aggstate, HashAggSpill *spill,
3028 : : TupleTableSlot *inputslot, uint32 hash)
3029 : : {
3030 : : TupleTableSlot *spillslot;
3031 : : int partition;
3032 : : MinimalTuple tuple;
3033 : : LogicalTape *tape;
1941 tgl@sss.pgh.pa.us 3034 : 592812 : int total_written = 0;
3035 : : bool shouldFree;
3036 : :
1998 jdavis@postgresql.or 3037 [ - + ]: 592812 : Assert(spill->partitions != NULL);
3038 : :
3039 : : /* spill only attributes that we actually need */
1882 3040 [ + + ]: 592812 : if (!aggstate->all_cols_needed)
3041 : : {
3042 : 1938 : spillslot = aggstate->hash_spill_wslot;
3043 : 1938 : slot_getsomeattrs(inputslot, aggstate->max_colno_needed);
3044 : 1938 : ExecClearTuple(spillslot);
3045 [ + + ]: 5814 : for (int i = 0; i < spillslot->tts_tupleDescriptor->natts; i++)
3046 : : {
3047 [ + + ]: 3876 : if (bms_is_member(i + 1, aggstate->colnos_needed))
3048 : : {
3049 : 1938 : spillslot->tts_values[i] = inputslot->tts_values[i];
3050 : 1938 : spillslot->tts_isnull[i] = inputslot->tts_isnull[i];
3051 : : }
3052 : : else
3053 : 1938 : spillslot->tts_isnull[i] = true;
3054 : : }
3055 : 1938 : ExecStoreVirtualTuple(spillslot);
3056 : : }
3057 : : else
3058 : 590874 : spillslot = inputslot;
3059 : :
3060 : 592812 : tuple = ExecFetchSlotMinimalTuple(spillslot, &shouldFree);
3061 : :
207 3062 [ + - ]: 592812 : if (spill->shift < 32)
3063 : 592812 : partition = (hash & spill->mask) >> spill->shift;
3064 : : else
207 jdavis@postgresql.or 3065 :UBC 0 : partition = 0;
3066 : :
1998 jdavis@postgresql.or 3067 :CBC 592812 : spill->ntuples[partition]++;
3068 : :
3069 : : /*
3070 : : * All hash values destined for a given partition have some bits in
3071 : : * common, which causes bad HLL cardinality estimates. Hash the hash to
3072 : : * get a more uniform distribution.
3073 : : */
1866 3074 : 592812 : addHyperLogLog(&spill->hll_card[partition], hash_bytes_uint32(hash));
3075 : :
1419 heikki.linnakangas@i 3076 : 592812 : tape = spill->partitions[partition];
3077 : :
981 peter@eisentraut.org 3078 : 592812 : LogicalTapeWrite(tape, &hash, sizeof(uint32));
1998 jdavis@postgresql.or 3079 : 592812 : total_written += sizeof(uint32);
3080 : :
981 peter@eisentraut.org 3081 : 592812 : LogicalTapeWrite(tape, tuple, tuple->t_len);
1998 jdavis@postgresql.or 3082 : 592812 : total_written += tuple->t_len;
3083 : :
3084 [ + + ]: 592812 : if (shouldFree)
3085 : 380466 : pfree(tuple);
3086 : :
3087 : 592812 : return total_written;
3088 : : }
3089 : :
3090 : : /*
3091 : : * hashagg_batch_new
3092 : : *
3093 : : * Construct a HashAggBatch item, which represents one iteration of HashAgg to
3094 : : * be done.
3095 : : */
3096 : : static HashAggBatch *
1419 heikki.linnakangas@i 3097 : 13467 : hashagg_batch_new(LogicalTape *input_tape, int setno,
3098 : : int64 input_tuples, double input_card, int used_bits)
3099 : : {
1998 jdavis@postgresql.or 3100 : 13467 : HashAggBatch *batch = palloc0(sizeof(HashAggBatch));
3101 : :
3102 : 13467 : batch->setno = setno;
3103 : 13467 : batch->used_bits = used_bits;
1419 heikki.linnakangas@i 3104 : 13467 : batch->input_tape = input_tape;
1998 jdavis@postgresql.or 3105 : 13467 : batch->input_tuples = input_tuples;
1866 3106 : 13467 : batch->input_card = input_card;
3107 : :
1998 3108 : 13467 : return batch;
3109 : : }
3110 : :
3111 : : /*
3112 : : * hashagg_batch_read
3113 : : * read the next tuple from a batch's tape. Return NULL if no more.
3114 : : */
3115 : : static MinimalTuple
3116 : 606279 : hashagg_batch_read(HashAggBatch *batch, uint32 *hashp)
3117 : : {
1419 heikki.linnakangas@i 3118 : 606279 : LogicalTape *tape = batch->input_tape;
3119 : : MinimalTuple tuple;
3120 : : uint32 t_len;
3121 : : size_t nread;
3122 : : uint32 hash;
3123 : :
3124 : 606279 : nread = LogicalTapeRead(tape, &hash, sizeof(uint32));
1998 jdavis@postgresql.or 3125 [ + + ]: 606279 : if (nread == 0)
3126 : 13467 : return NULL;
3127 [ - + ]: 592812 : if (nread != sizeof(uint32))
1998 jdavis@postgresql.or 3128 [ # # ]:UBC 0 : ereport(ERROR,
3129 : : (errcode_for_file_access(),
3130 : : errmsg_internal("unexpected EOF for tape %p: requested %zu bytes, read %zu bytes",
3131 : : tape, sizeof(uint32), nread)));
1998 jdavis@postgresql.or 3132 [ + - ]:CBC 592812 : if (hashp != NULL)
3133 : 592812 : *hashp = hash;
3134 : :
1419 heikki.linnakangas@i 3135 : 592812 : nread = LogicalTapeRead(tape, &t_len, sizeof(t_len));
1998 jdavis@postgresql.or 3136 [ - + ]: 592812 : if (nread != sizeof(uint32))
1998 jdavis@postgresql.or 3137 [ # # ]:UBC 0 : ereport(ERROR,
3138 : : (errcode_for_file_access(),
3139 : : errmsg_internal("unexpected EOF for tape %p: requested %zu bytes, read %zu bytes",
3140 : : tape, sizeof(uint32), nread)));
3141 : :
1998 jdavis@postgresql.or 3142 :CBC 592812 : tuple = (MinimalTuple) palloc(t_len);
3143 : 592812 : tuple->t_len = t_len;
3144 : :
1419 heikki.linnakangas@i 3145 : 592812 : nread = LogicalTapeRead(tape,
3146 : : (char *) tuple + sizeof(uint32),
3147 : : t_len - sizeof(uint32));
1998 jdavis@postgresql.or 3148 [ - + ]: 592812 : if (nread != t_len - sizeof(uint32))
1998 jdavis@postgresql.or 3149 [ # # ]:UBC 0 : ereport(ERROR,
3150 : : (errcode_for_file_access(),
3151 : : errmsg_internal("unexpected EOF for tape %p: requested %zu bytes, read %zu bytes",
3152 : : tape, t_len - sizeof(uint32), nread)));
3153 : :
1998 jdavis@postgresql.or 3154 :CBC 592812 : return tuple;
3155 : : }
3156 : :
3157 : : /*
3158 : : * hashagg_finish_initial_spills
3159 : : *
3160 : : * After a HashAggBatch has been processed, it may have spilled tuples to
3161 : : * disk. If so, turn the spilled partitions into new batches that must later
3162 : : * be executed.
3163 : : */
3164 : : static void
3165 : 8211 : hashagg_finish_initial_spills(AggState *aggstate)
3166 : : {
3167 : : int setno;
1941 tgl@sss.pgh.pa.us 3168 : 8211 : int total_npartitions = 0;
3169 : :
1998 jdavis@postgresql.or 3170 [ + + ]: 8211 : if (aggstate->hash_spills != NULL)
3171 : : {
3172 [ + + ]: 90 : for (setno = 0; setno < aggstate->num_hashes; setno++)
3173 : : {
3174 : 60 : HashAggSpill *spill = &aggstate->hash_spills[setno];
3175 : :
3176 : 60 : total_npartitions += spill->npartitions;
3177 : 60 : hashagg_spill_finish(aggstate, spill, setno);
3178 : : }
3179 : :
3180 : : /*
3181 : : * We're not processing tuples from outer plan any more; only
3182 : : * processing batches of spilled tuples. The initial spill structures
3183 : : * are no longer needed.
3184 : : */
3185 : 30 : pfree(aggstate->hash_spills);
3186 : 30 : aggstate->hash_spills = NULL;
3187 : : }
3188 : :
3189 : 8211 : hash_agg_update_metrics(aggstate, false, total_npartitions);
3190 : 8211 : aggstate->hash_spill_mode = false;
3191 : 8211 : }
3192 : :
3193 : : /*
3194 : : * hashagg_spill_finish
3195 : : *
3196 : : * Transform spill partitions into new batches.
3197 : : */
3198 : : static void
3199 : 6306 : hashagg_spill_finish(AggState *aggstate, HashAggSpill *spill, int setno)
3200 : : {
3201 : : int i;
1941 tgl@sss.pgh.pa.us 3202 : 6306 : int used_bits = 32 - spill->shift;
3203 : :
1998 jdavis@postgresql.or 3204 [ - + ]: 6306 : if (spill->npartitions == 0)
1941 tgl@sss.pgh.pa.us 3205 :UBC 0 : return; /* didn't spill */
3206 : :
1998 jdavis@postgresql.or 3207 [ + + ]:CBC 31530 : for (i = 0; i < spill->npartitions; i++)
3208 : : {
1419 heikki.linnakangas@i 3209 : 25224 : LogicalTape *tape = spill->partitions[i];
3210 : : HashAggBatch *new_batch;
3211 : : double cardinality;
3212 : :
3213 : : /* if the partition is empty, don't create a new batch of work */
1998 jdavis@postgresql.or 3214 [ + + ]: 25224 : if (spill->ntuples[i] == 0)
3215 : 11757 : continue;
3216 : :
1866 3217 : 13467 : cardinality = estimateHyperLogLog(&spill->hll_card[i]);
3218 : 13467 : freeHyperLogLog(&spill->hll_card[i]);
3219 : :
3220 : : /* rewinding frees the buffer while not in use */
1419 heikki.linnakangas@i 3221 : 13467 : LogicalTapeRewindForRead(tape, HASHAGG_READ_BUFFER_SIZE);
3222 : :
3223 : 13467 : new_batch = hashagg_batch_new(tape, setno,
1817 jdavis@postgresql.or 3224 : 13467 : spill->ntuples[i], cardinality,
3225 : : used_bits);
1405 tgl@sss.pgh.pa.us 3226 : 13467 : aggstate->hash_batches = lappend(aggstate->hash_batches, new_batch);
1998 jdavis@postgresql.or 3227 : 13467 : aggstate->hash_batches_used++;
3228 : : }
3229 : :
3230 : 6306 : pfree(spill->ntuples);
1866 3231 : 6306 : pfree(spill->hll_card);
1998 3232 : 6306 : pfree(spill->partitions);
3233 : : }
3234 : :
3235 : : /*
3236 : : * Free resources related to a spilled HashAgg.
3237 : : */
3238 : : static void
3239 : 28106 : hashagg_reset_spill_state(AggState *aggstate)
3240 : : {
3241 : : /* free spills from initial pass */
3242 [ - + ]: 28106 : if (aggstate->hash_spills != NULL)
3243 : : {
3244 : : int setno;
3245 : :
1998 jdavis@postgresql.or 3246 [ # # ]:UBC 0 : for (setno = 0; setno < aggstate->num_hashes; setno++)
3247 : : {
3248 : 0 : HashAggSpill *spill = &aggstate->hash_spills[setno];
3249 : :
3250 : 0 : pfree(spill->ntuples);
3251 : 0 : pfree(spill->partitions);
3252 : : }
3253 : 0 : pfree(aggstate->hash_spills);
3254 : 0 : aggstate->hash_spills = NULL;
3255 : : }
3256 : :
3257 : : /* free batches */
1405 tgl@sss.pgh.pa.us 3258 :CBC 28106 : list_free_deep(aggstate->hash_batches);
1998 jdavis@postgresql.or 3259 : 28106 : aggstate->hash_batches = NIL;
3260 : :
3261 : : /* close tape set */
1419 heikki.linnakangas@i 3262 [ + + ]: 28106 : if (aggstate->hash_tapeset != NULL)
3263 : : {
3264 : 30 : LogicalTapeSetClose(aggstate->hash_tapeset);
3265 : 30 : aggstate->hash_tapeset = NULL;
3266 : : }
1998 jdavis@postgresql.or 3267 : 28106 : }
3268 : :
3269 : :
3270 : : /* -----------------
3271 : : * ExecInitAgg
3272 : : *
3273 : : * Creates the run-time information for the agg node produced by the
3274 : : * planner and initializes its outer subtree.
3275 : : *
3276 : : * -----------------
3277 : : */
3278 : : AggState *
7130 tgl@sss.pgh.pa.us 3279 : 22046 : ExecInitAgg(Agg *node, EState *estate, int eflags)
3280 : : {
3281 : : AggState *aggstate;
3282 : : AggStatePerAgg peraggs;
3283 : : AggStatePerTrans pertransstates;
3284 : : AggStatePerGroup *pergroups;
3285 : : Plan *outerPlan;
3286 : : ExprContext *econtext;
3287 : : TupleDesc scanDesc;
3288 : : int max_aggno;
3289 : : int max_transno;
3290 : : int numaggrefs;
3291 : : int numaggs;
3292 : : int numtrans;
3293 : : int phase;
3294 : : int phaseidx;
3295 : : ListCell *l;
3766 andres@anarazel.de 3296 : 22046 : Bitmapset *all_grouped_cols = NULL;
3297 : 22046 : int numGroupingSets = 1;
3298 : : int numPhases;
3299 : : int numHashes;
3300 : 22046 : int i = 0;
3301 : 22046 : int j = 0;
3085 rhodiumtoad@postgres 3302 [ + + ]: 41376 : bool use_hashing = (node->aggstrategy == AGG_HASHED ||
3303 [ + + ]: 19330 : node->aggstrategy == AGG_MIXED);
3304 : :
3305 : : /* check for unsupported flags */
7130 tgl@sss.pgh.pa.us 3306 [ - + ]: 22046 : Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK)));
3307 : :
3308 : : /*
3309 : : * create state structure
3310 : : */
10226 bruce@momjian.us 3311 : 22046 : aggstate = makeNode(AggState);
8311 tgl@sss.pgh.pa.us 3312 : 22046 : aggstate->ss.ps.plan = (Plan *) node;
3313 : 22046 : aggstate->ss.ps.state = estate;
2973 andres@anarazel.de 3314 : 22046 : aggstate->ss.ps.ExecProcNode = ExecAgg;
3315 : :
8311 tgl@sss.pgh.pa.us 3316 : 22046 : aggstate->aggs = NIL;
3317 : 22046 : aggstate->numaggs = 0;
3686 heikki.linnakangas@i 3318 : 22046 : aggstate->numtrans = 0;
3085 rhodiumtoad@postgres 3319 : 22046 : aggstate->aggstrategy = node->aggstrategy;
3359 tgl@sss.pgh.pa.us 3320 : 22046 : aggstate->aggsplit = node->aggsplit;
3766 andres@anarazel.de 3321 : 22046 : aggstate->maxsets = 0;
3322 : 22046 : aggstate->projected_set = -1;
3323 : 22046 : aggstate->current_set = 0;
8340 tgl@sss.pgh.pa.us 3324 : 22046 : aggstate->peragg = NULL;
3686 heikki.linnakangas@i 3325 : 22046 : aggstate->pertrans = NULL;
2886 tgl@sss.pgh.pa.us 3326 : 22046 : aggstate->curperagg = NULL;
3686 heikki.linnakangas@i 3327 : 22046 : aggstate->curpertrans = NULL;
3766 andres@anarazel.de 3328 : 22046 : aggstate->input_done = false;
3359 tgl@sss.pgh.pa.us 3329 : 22046 : aggstate->agg_done = false;
2804 andres@anarazel.de 3330 : 22046 : aggstate->pergroups = NULL;
8340 tgl@sss.pgh.pa.us 3331 : 22046 : aggstate->grp_firstTuple = NULL;
3766 andres@anarazel.de 3332 : 22046 : aggstate->sort_in = NULL;
3333 : 22046 : aggstate->sort_out = NULL;
3334 : :
3335 : : /*
3336 : : * phases[0] always exists, but is dummy in sorted/plain mode
3337 : : */
3085 rhodiumtoad@postgres 3338 [ + + ]: 22046 : numPhases = (use_hashing ? 1 : 2);
3339 : 22046 : numHashes = (use_hashing ? 1 : 0);
3340 : :
3341 : : /*
3342 : : * Calculate the maximum number of grouping sets in any phase; this
3343 : : * determines the size of some allocations. Also calculate the number of
3344 : : * phases, since all hashed/mixed nodes contribute to only a single phase.
3345 : : */
3766 andres@anarazel.de 3346 [ + + ]: 22046 : if (node->groupingSets)
3347 : : {
3348 : 430 : numGroupingSets = list_length(node->groupingSets);
3349 : :
3350 [ + + + + : 923 : foreach(l, node->chain)
+ + ]
3351 : : {
3759 bruce@momjian.us 3352 : 493 : Agg *agg = lfirst(l);
3353 : :
3766 andres@anarazel.de 3354 [ + + ]: 493 : numGroupingSets = Max(numGroupingSets,
3355 : : list_length(agg->groupingSets));
3356 : :
3357 : : /*
3358 : : * additional AGG_HASHED aggs become part of phase 0, but all
3359 : : * others add an extra phase.
3360 : : */
3085 rhodiumtoad@postgres 3361 [ + + ]: 493 : if (agg->aggstrategy != AGG_HASHED)
3362 : 236 : ++numPhases;
3363 : : else
3364 : 257 : ++numHashes;
3365 : : }
3366 : : }
3367 : :
3766 andres@anarazel.de 3368 : 22046 : aggstate->maxsets = numGroupingSets;
3085 rhodiumtoad@postgres 3369 : 22046 : aggstate->numphases = numPhases;
3370 : :
3766 andres@anarazel.de 3371 : 22046 : aggstate->aggcontexts = (ExprContext **)
3372 : 22046 : palloc0(sizeof(ExprContext *) * numGroupingSets);
3373 : :
3374 : : /*
3375 : : * Create expression contexts. We need three or more, one for
3376 : : * per-input-tuple processing, one for per-output-tuple processing, one
3377 : : * for all the hashtables, and one for each grouping set. The per-tuple
3378 : : * memory context of the per-grouping-set ExprContexts (aggcontexts)
3379 : : * replaces the standalone memory context formerly used to hold transition
3380 : : * values. We cheat a little by using ExecAssignExprContext() to build
3381 : : * all of them.
3382 : : *
3383 : : * NOTE: the details of what is stored in aggcontexts and what is stored
3384 : : * in the regular per-query memory context are driven by a simple
3385 : : * decision: we want to reset the aggcontext at group boundaries (if not
3386 : : * hashing) and in ExecReScanAgg to recover no-longer-wanted space.
3387 : : */
3388 : 22046 : ExecAssignExprContext(estate, &aggstate->ss.ps);
3389 : 22046 : aggstate->tmpcontext = aggstate->ss.ps.ps_ExprContext;
3390 : :
3391 [ + + ]: 44512 : for (i = 0; i < numGroupingSets; ++i)
3392 : : {
3393 : 22466 : ExecAssignExprContext(estate, &aggstate->ss.ps);
3394 : 22466 : aggstate->aggcontexts[i] = aggstate->ss.ps.ps_ExprContext;
3395 : : }
3396 : :
3085 rhodiumtoad@postgres 3397 [ + + ]: 22046 : if (use_hashing)
166 jdavis@postgresql.or 3398 : 2832 : hash_create_memory(aggstate);
3399 : :
3766 andres@anarazel.de 3400 : 22046 : ExecAssignExprContext(estate, &aggstate->ss.ps);
3401 : :
3402 : : /*
3403 : : * Initialize child nodes.
3404 : : *
3405 : : * If we are doing a hashed aggregation then the child plan does not need
3406 : : * to handle REWIND efficiently; see ExecReScanAgg.
3407 : : */
7130 tgl@sss.pgh.pa.us 3408 [ + + ]: 22046 : if (node->aggstrategy == AGG_HASHED)
3409 : 2716 : eflags &= ~EXEC_FLAG_REWIND;
8311 3410 : 22046 : outerPlan = outerPlan(node);
7130 3411 : 22046 : outerPlanState(aggstate) = ExecInitNode(outerPlan, estate, eflags);
3412 : :
3413 : : /*
3414 : : * initialize source tuple type.
3415 : : */
2235 andres@anarazel.de 3416 : 22046 : aggstate->ss.ps.outerops =
3417 : 22046 : ExecGetResultSlotOps(outerPlanState(&aggstate->ss),
3418 : : &aggstate->ss.ps.outeropsfixed);
3419 : 22046 : aggstate->ss.ps.outeropsset = true;
3420 : :
2487 3421 : 22046 : ExecCreateScanSlotFromOuterPlan(estate, &aggstate->ss,
3422 : : aggstate->ss.ps.outerops);
2760 3423 : 22046 : scanDesc = aggstate->ss.ss_ScanTupleSlot->tts_tupleDescriptor;
3424 : :
3425 : : /*
3426 : : * If there are more than two phases (including a potential dummy phase
3427 : : * 0), input will be resorted using tuplesort. Need a slot for that.
3428 : : */
2235 3429 [ + + ]: 22046 : if (numPhases > 2)
3430 : : {
2487 3431 : 99 : aggstate->sort_slot = ExecInitExtraTupleSlot(estate, scanDesc,
3432 : : &TTSOpsMinimalTuple);
3433 : :
3434 : : /*
3435 : : * The output of the tuplesort, and the output from the outer child
3436 : : * might not use the same type of slot. In most cases the child will
3437 : : * be a Sort, and thus return a TTSOpsMinimalTuple type slot - but the
3438 : : * input can also be presorted due an index, in which case it could be
3439 : : * a different type of slot.
3440 : : *
3441 : : * XXX: For efficiency it would be good to instead/additionally
3442 : : * generate expressions with corresponding settings of outerops* for
3443 : : * the individual phases - deforming is often a bottleneck for
3444 : : * aggregations with lots of rows per group. If there's multiple
3445 : : * sorts, we know that all but the first use TTSOpsMinimalTuple (via
3446 : : * the nodeAgg.c internal tuplesort).
3447 : : */
2235 3448 [ + - ]: 99 : if (aggstate->ss.ps.outeropsfixed &&
3449 [ + + ]: 99 : aggstate->ss.ps.outerops != &TTSOpsMinimalTuple)
3450 : 6 : aggstate->ss.ps.outeropsfixed = false;
3451 : : }
3452 : :
3453 : : /*
3454 : : * Initialize result type, slot and projection.
3455 : : */
2487 3456 : 22046 : ExecInitResultTupleSlotTL(&aggstate->ss.ps, &TTSOpsVirtual);
6791 tgl@sss.pgh.pa.us 3457 : 22046 : ExecAssignProjectionInfo(&aggstate->ss.ps, NULL);
3458 : :
3459 : : /*
3460 : : * initialize child expressions
3461 : : *
3462 : : * We expect the parser to have checked that no aggs contain other agg
3463 : : * calls in their arguments (and just to be sure, we verify it again while
3464 : : * initializing the plan node). This would make no sense under SQL
3465 : : * semantics, and it's forbidden by the spec. Because it is true, we
3466 : : * don't need to worry about evaluating the aggs in any particular order.
3467 : : *
3468 : : * Note: execExpr.c finds Aggrefs for us, and adds them to aggstate->aggs.
3469 : : * Aggrefs in the qual are found here; Aggrefs in the targetlist are found
3470 : : * during ExecAssignProjectionInfo, above.
3471 : : */
2759 andres@anarazel.de 3472 : 22046 : aggstate->ss.ps.qual =
3473 : 22046 : ExecInitQual(node->plan.qual, (PlanState *) aggstate);
3474 : :
3475 : : /*
3476 : : * We should now have found all Aggrefs in the targetlist and quals.
3477 : : */
1747 heikki.linnakangas@i 3478 : 22046 : numaggrefs = list_length(aggstate->aggs);
3479 : 22046 : max_aggno = -1;
3480 : 22046 : max_transno = -1;
3481 [ + + + + : 46910 : foreach(l, aggstate->aggs)
+ + ]
3482 : : {
3483 : 24864 : Aggref *aggref = (Aggref *) lfirst(l);
3484 : :
3485 : 24864 : max_aggno = Max(max_aggno, aggref->aggno);
3486 : 24864 : max_transno = Max(max_transno, aggref->aggtransno);
3487 : : }
242 jdavis@postgresql.or 3488 : 22046 : aggstate->numaggs = numaggs = max_aggno + 1;
3489 : 22046 : aggstate->numtrans = numtrans = max_transno + 1;
3490 : :
3491 : : /*
3492 : : * For each phase, prepare grouping set data and fmgr lookup data for
3493 : : * compare functions. Accumulate all_grouped_cols in passing.
3494 : : */
3766 andres@anarazel.de 3495 : 22046 : aggstate->phases = palloc0(numPhases * sizeof(AggStatePerPhaseData));
3496 : :
3085 rhodiumtoad@postgres 3497 : 22046 : aggstate->num_hashes = numHashes;
3498 [ + + ]: 22046 : if (numHashes)
3499 : : {
3500 : 2832 : aggstate->perhash = palloc0(sizeof(AggStatePerHashData) * numHashes);
3501 : 2832 : aggstate->phases[0].numsets = 0;
3502 : 2832 : aggstate->phases[0].gset_lengths = palloc(numHashes * sizeof(int));
3503 : 2832 : aggstate->phases[0].grouped_cols = palloc(numHashes * sizeof(Bitmapset *));
3504 : : }
3505 : :
3506 : 22046 : phase = 0;
3507 [ + + ]: 44585 : for (phaseidx = 0; phaseidx <= list_length(node->chain); ++phaseidx)
3508 : : {
3509 : : Agg *aggnode;
3510 : : Sort *sortnode;
3511 : :
3512 [ + + ]: 22539 : if (phaseidx > 0)
3513 : : {
3071 tgl@sss.pgh.pa.us 3514 : 493 : aggnode = list_nth_node(Agg, node->chain, phaseidx - 1);
1157 3515 : 493 : sortnode = castNode(Sort, outerPlan(aggnode));
3516 : : }
3517 : : else
3518 : : {
3766 andres@anarazel.de 3519 : 22046 : aggnode = node;
3520 : 22046 : sortnode = NULL;
3521 : : }
3522 : :
3085 rhodiumtoad@postgres 3523 [ + + - + ]: 22539 : Assert(phase <= 1 || sortnode);
3524 : :
3525 [ + + ]: 22539 : if (aggnode->aggstrategy == AGG_HASHED
3526 [ + + ]: 19566 : || aggnode->aggstrategy == AGG_MIXED)
3766 andres@anarazel.de 3527 : 3089 : {
3085 rhodiumtoad@postgres 3528 : 3089 : AggStatePerPhase phasedata = &aggstate->phases[0];
3529 : : AggStatePerHash perhash;
3530 : 3089 : Bitmapset *cols = NULL;
3531 : :
3532 [ - + ]: 3089 : Assert(phase == 0);
3533 : 3089 : i = phasedata->numsets++;
3534 : 3089 : perhash = &aggstate->perhash[i];
3535 : :
3536 : : /* phase 0 always points to the "real" Agg in the hash case */
3537 : 3089 : phasedata->aggnode = node;
3538 : 3089 : phasedata->aggstrategy = node->aggstrategy;
3539 : :
3540 : : /* but the actual Agg node representing this hash is saved here */
3541 : 3089 : perhash->aggnode = aggnode;
3542 : :
3543 : 3089 : phasedata->gset_lengths[i] = perhash->numCols = aggnode->numCols;
3544 : :
3545 [ + + ]: 7761 : for (j = 0; j < aggnode->numCols; ++j)
3546 : 4672 : cols = bms_add_member(cols, aggnode->grpColIdx[j]);
3547 : :
3548 : 3089 : phasedata->grouped_cols[i] = cols;
3549 : :
3550 : 3089 : all_grouped_cols = bms_add_members(all_grouped_cols, cols);
3551 : 3089 : continue;
3552 : : }
3553 : : else
3554 : : {
3555 : 19450 : AggStatePerPhase phasedata = &aggstate->phases[++phase];
3556 : : int num_sets;
3557 : :
3558 : 19450 : phasedata->numsets = num_sets = list_length(aggnode->groupingSets);
3559 : :
3560 [ + + ]: 19450 : if (num_sets)
3561 : : {
3562 : 467 : phasedata->gset_lengths = palloc(num_sets * sizeof(int));
3563 : 467 : phasedata->grouped_cols = palloc(num_sets * sizeof(Bitmapset *));
3564 : :
3565 : 467 : i = 0;
3566 [ + - + + : 1390 : foreach(l, aggnode->groupingSets)
+ + ]
3567 : : {
3568 : 923 : int current_length = list_length(lfirst(l));
3569 : 923 : Bitmapset *cols = NULL;
3570 : :
3571 : : /* planner forces this to be correct */
3572 [ + + ]: 1821 : for (j = 0; j < current_length; ++j)
3573 : 898 : cols = bms_add_member(cols, aggnode->grpColIdx[j]);
3574 : :
3575 : 923 : phasedata->grouped_cols[i] = cols;
3576 : 923 : phasedata->gset_lengths[i] = current_length;
3577 : :
3578 : 923 : ++i;
3579 : : }
3580 : :
3581 : 467 : all_grouped_cols = bms_add_members(all_grouped_cols,
2999 tgl@sss.pgh.pa.us 3582 : 467 : phasedata->grouped_cols[0]);
3583 : : }
3584 : : else
3585 : : {
3085 rhodiumtoad@postgres 3586 [ - + ]: 18983 : Assert(phaseidx == 0);
3587 : :
3588 : 18983 : phasedata->gset_lengths = NULL;
3589 : 18983 : phasedata->grouped_cols = NULL;
3590 : : }
3591 : :
3592 : : /*
3593 : : * If we are grouping, precompute fmgr lookup data for inner loop.
3594 : : */
3595 [ + + ]: 19450 : if (aggnode->aggstrategy == AGG_SORTED)
3596 : : {
3597 : : /*
3598 : : * Build a separate function for each subset of columns that
3599 : : * need to be compared.
3600 : : */
3601 : 1148 : phasedata->eqfunctions =
2760 andres@anarazel.de 3602 : 1148 : (ExprState **) palloc0(aggnode->numCols * sizeof(ExprState *));
3603 : :
3604 : : /* for each grouping set */
1067 drowley@postgresql.o 3605 [ + + ]: 1932 : for (int k = 0; k < phasedata->numsets; k++)
3606 : : {
3607 : 784 : int length = phasedata->gset_lengths[k];
3608 : :
3609 : : /* nothing to do for empty grouping set */
978 tgl@sss.pgh.pa.us 3610 [ + + ]: 784 : if (length == 0)
3611 : 163 : continue;
3612 : :
3613 : : /* if we already had one of this length, it'll do */
2760 andres@anarazel.de 3614 [ + + ]: 621 : if (phasedata->eqfunctions[length - 1] != NULL)
3615 : 69 : continue;
3616 : :
3617 : 552 : phasedata->eqfunctions[length - 1] =
3618 : 552 : execTuplesMatchPrepare(scanDesc,
3619 : : length,
3620 : 552 : aggnode->grpColIdx,
3621 : 552 : aggnode->grpOperators,
2360 peter@eisentraut.org 3622 : 552 : aggnode->grpCollations,
3623 : : (PlanState *) aggstate);
3624 : : }
3625 : :
3626 : : /* and for all grouped columns, unless already computed */
962 tgl@sss.pgh.pa.us 3627 [ + + ]: 1148 : if (aggnode->numCols > 0 &&
3628 [ + + ]: 1101 : phasedata->eqfunctions[aggnode->numCols - 1] == NULL)
3629 : : {
2760 andres@anarazel.de 3630 : 737 : phasedata->eqfunctions[aggnode->numCols - 1] =
3631 : 737 : execTuplesMatchPrepare(scanDesc,
3632 : : aggnode->numCols,
3633 : 737 : aggnode->grpColIdx,
3634 : 737 : aggnode->grpOperators,
2360 peter@eisentraut.org 3635 : 737 : aggnode->grpCollations,
3636 : : (PlanState *) aggstate);
3637 : : }
3638 : : }
3639 : :
3085 rhodiumtoad@postgres 3640 : 19450 : phasedata->aggnode = aggnode;
3641 : 19450 : phasedata->aggstrategy = aggnode->aggstrategy;
3642 : 19450 : phasedata->sortnode = sortnode;
3643 : : }
3644 : : }
3645 : :
3646 : : /*
3647 : : * Convert all_grouped_cols to a descending-order list.
3648 : : */
3766 andres@anarazel.de 3649 : 22046 : i = -1;
3650 [ + + ]: 26998 : while ((i = bms_next_member(all_grouped_cols, i)) >= 0)
3651 : 4952 : aggstate->all_grouped_cols = lcons_int(i, aggstate->all_grouped_cols);
3652 : :
3653 : : /*
3654 : : * Set up aggregate-result storage in the output expr context, and also
3655 : : * allocate my private per-agg working storage
3656 : : */
8311 tgl@sss.pgh.pa.us 3657 : 22046 : econtext = aggstate->ss.ps.ps_ExprContext;
8333 bruce@momjian.us 3658 : 22046 : econtext->ecxt_aggvalues = (Datum *) palloc0(sizeof(Datum) * numaggs);
3659 : 22046 : econtext->ecxt_aggnulls = (bool *) palloc0(sizeof(bool) * numaggs);
3660 : :
3686 heikki.linnakangas@i 3661 : 22046 : peraggs = (AggStatePerAgg) palloc0(sizeof(AggStatePerAggData) * numaggs);
1747 3662 : 22046 : pertransstates = (AggStatePerTrans) palloc0(sizeof(AggStatePerTransData) * numtrans);
3663 : :
3686 3664 : 22046 : aggstate->peragg = peraggs;
3665 : 22046 : aggstate->pertrans = pertransstates;
3666 : :
3667 : :
2797 andres@anarazel.de 3668 : 22046 : aggstate->all_pergroups =
3669 : 22046 : (AggStatePerGroup *) palloc0(sizeof(AggStatePerGroup)
3670 : 22046 : * (numGroupingSets + numHashes));
3671 : 22046 : pergroups = aggstate->all_pergroups;
3672 : :
3673 [ + + ]: 22046 : if (node->aggstrategy != AGG_HASHED)
3674 : : {
3675 [ + + ]: 39080 : for (i = 0; i < numGroupingSets; i++)
3676 : : {
3677 : 19750 : pergroups[i] = (AggStatePerGroup) palloc0(sizeof(AggStatePerGroupData)
3678 : : * numaggs);
3679 : : }
3680 : :
3681 : 19330 : aggstate->pergroups = pergroups;
3682 : 19330 : pergroups += numGroupingSets;
3683 : : }
3684 : :
3685 : : /*
3686 : : * Hashing can only appear in the initial phase.
3687 : : */
3085 rhodiumtoad@postgres 3688 [ + + ]: 22046 : if (use_hashing)
3689 : : {
1941 tgl@sss.pgh.pa.us 3690 : 2832 : Plan *outerplan = outerPlan(node);
3691 : 2832 : uint64 totalGroups = 0;
3692 : :
1882 jdavis@postgresql.or 3693 : 2832 : aggstate->hash_spill_rslot = ExecInitExtraTupleSlot(estate, scanDesc,
3694 : : &TTSOpsMinimalTuple);
3695 : 2832 : aggstate->hash_spill_wslot = ExecInitExtraTupleSlot(estate, scanDesc,
3696 : : &TTSOpsVirtual);
3697 : :
3698 : : /* this is an array of pointers, not structures */
2797 andres@anarazel.de 3699 : 2832 : aggstate->hash_pergroup = pergroups;
3700 : :
1941 tgl@sss.pgh.pa.us 3701 : 5664 : aggstate->hashentrysize = hash_agg_entry_size(aggstate->numtrans,
3702 : 2832 : outerplan->plan_width,
3703 : : node->transitionSpace);
3704 : :
3705 : : /*
3706 : : * Consider all of the grouping sets together when setting the limits
3707 : : * and estimating the number of partitions. This can be inaccurate
3708 : : * when there is more than one grouping set, but should still be
3709 : : * reasonable.
3710 : : */
1067 drowley@postgresql.o 3711 [ + + ]: 5921 : for (int k = 0; k < aggstate->num_hashes; k++)
3712 : 3089 : totalGroups += aggstate->perhash[k].aggnode->numGroups;
3713 : :
1998 jdavis@postgresql.or 3714 : 2832 : hash_agg_set_limits(aggstate->hashentrysize, totalGroups, 0,
3715 : : &aggstate->hash_mem_limit,
3716 : : &aggstate->hash_ngroups_limit,
3717 : : &aggstate->hash_planned_partitions);
3085 rhodiumtoad@postgres 3718 : 2832 : find_hash_columns(aggstate);
3719 : :
3720 : : /* Skip massive memory allocation if we are just doing EXPLAIN */
1753 heikki.linnakangas@i 3721 [ + + ]: 2832 : if (!(eflags & EXEC_FLAG_EXPLAIN_ONLY))
3722 : 2170 : build_hash_tables(aggstate);
3723 : :
8340 tgl@sss.pgh.pa.us 3724 : 2832 : aggstate->table_filled = false;
3725 : :
3726 : : /* Initialize this to 1, meaning nothing spilled, yet */
1865 drowley@postgresql.o 3727 : 2832 : aggstate->hash_batches_used = 1;
3728 : : }
3729 : :
3730 : : /*
3731 : : * Initialize current phase-dependent values to initial phase. The initial
3732 : : * phase is 1 (first sort pass) for all strategies that use sorting (if
3733 : : * hashing is being done too, then phase 0 is processed last); but if only
3734 : : * hashing is being done, then phase 0 is all there is.
3735 : : */
3085 rhodiumtoad@postgres 3736 [ + + ]: 22046 : if (node->aggstrategy == AGG_HASHED)
3737 : : {
3738 : 2716 : aggstate->current_phase = 0;
3739 : 2716 : initialize_phase(aggstate, 0);
3740 : 2716 : select_current_set(aggstate, 0, true);
3741 : : }
3742 : : else
3743 : : {
3744 : 19330 : aggstate->current_phase = 1;
3745 : 19330 : initialize_phase(aggstate, 1);
3746 : 19330 : select_current_set(aggstate, 0, false);
3747 : : }
3748 : :
3749 : : /*
3750 : : * Perform lookups of aggregate function info, and initialize the
3751 : : * unchanging fields of the per-agg and per-trans data.
3752 : : */
7773 neilc@samurai.com 3753 [ + + + + : 46907 : foreach(l, aggstate->aggs)
+ + ]
3754 : : {
1747 heikki.linnakangas@i 3755 : 24864 : Aggref *aggref = lfirst(l);
3756 : : AggStatePerAgg peragg;
3757 : : AggStatePerTrans pertrans;
3758 : : Oid aggTransFnInputTypes[FUNC_MAX_ARGS];
3759 : : int numAggTransFnArgs;
3760 : : int numDirectArgs;
3761 : : HeapTuple aggTuple;
3762 : : Form_pg_aggregate aggform;
3763 : : AclResult aclresult;
3764 : : Oid finalfn_oid;
3765 : : Oid serialfn_oid,
3766 : : deserialfn_oid;
3767 : : Oid aggOwner;
3768 : : Expr *finalfnexpr;
3769 : : Oid aggtranstype;
3770 : :
3771 : : /* Planner should have assigned aggregate to correct level */
8128 tgl@sss.pgh.pa.us 3772 [ - + ]: 24864 : Assert(aggref->agglevelsup == 0);
3773 : : /* ... and the split mode should match */
3359 3774 [ - + ]: 24864 : Assert(aggref->aggsplit == aggstate->aggsplit);
3775 : :
1747 heikki.linnakangas@i 3776 : 24864 : peragg = &peraggs[aggref->aggno];
3777 : :
3778 : : /* Check if we initialized the state for this aggregate already. */
3779 [ + + ]: 24864 : if (peragg->aggref != NULL)
8250 tgl@sss.pgh.pa.us 3780 : 242 : continue;
3781 : :
3686 heikki.linnakangas@i 3782 : 24622 : peragg->aggref = aggref;
1747 3783 : 24622 : peragg->transno = aggref->aggtransno;
3784 : :
3785 : : /* Fetch the pg_aggregate row */
5683 rhaas@postgresql.org 3786 : 24622 : aggTuple = SearchSysCache1(AGGFNOID,
3787 : : ObjectIdGetDatum(aggref->aggfnoid));
9477 tgl@sss.pgh.pa.us 3788 [ - + ]: 24622 : if (!HeapTupleIsValid(aggTuple))
8083 tgl@sss.pgh.pa.us 3789 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for aggregate %u",
3790 : : aggref->aggfnoid);
9477 tgl@sss.pgh.pa.us 3791 :CBC 24622 : aggform = (Form_pg_aggregate) GETSTRUCT(aggTuple);
3792 : :
3793 : : /* Check permission to call aggregate function */
1028 peter@eisentraut.org 3794 : 24622 : aclresult = object_aclcheck(ProcedureRelationId, aggref->aggfnoid, GetUserId(),
3795 : : ACL_EXECUTE);
8531 tgl@sss.pgh.pa.us 3796 [ + + ]: 24622 : if (aclresult != ACLCHECK_OK)
2835 peter_e@gmx.net 3797 : 3 : aclcheck_error(aclresult, OBJECT_AGGREGATE,
8072 tgl@sss.pgh.pa.us 3798 : 3 : get_func_name(aggref->aggfnoid));
4530 rhaas@postgresql.org 3799 [ - + ]: 24619 : InvokeFunctionExecuteHook(aggref->aggfnoid);
3800 : :
3801 : : /* planner recorded transition state type in the Aggref itself */
3368 tgl@sss.pgh.pa.us 3802 : 24619 : aggtranstype = aggref->aggtranstype;
3803 [ - + ]: 24619 : Assert(OidIsValid(aggtranstype));
3804 : :
3805 : : /* Final function only required if we're finalizing the aggregates */
3359 3806 [ + + ]: 24619 : if (DO_AGGSPLIT_SKIPFINAL(aggstate->aggsplit))
3517 rhaas@postgresql.org 3807 : 2128 : peragg->finalfn_oid = finalfn_oid = InvalidOid;
3808 : : else
3359 tgl@sss.pgh.pa.us 3809 : 22491 : peragg->finalfn_oid = finalfn_oid = aggform->aggfinalfn;
3810 : :
3448 rhaas@postgresql.org 3811 : 24619 : serialfn_oid = InvalidOid;
3812 : 24619 : deserialfn_oid = InvalidOid;
3813 : :
3814 : : /*
3815 : : * Check if serialization/deserialization is required. We only do it
3816 : : * for aggregates that have transtype INTERNAL.
3817 : : */
3359 tgl@sss.pgh.pa.us 3818 [ + + ]: 24619 : if (aggtranstype == INTERNALOID)
3819 : : {
3820 : : /*
3821 : : * The planner should only have generated a serialize agg node if
3822 : : * every aggregate with an INTERNAL state has a serialization
3823 : : * function. Verify that.
3824 : : */
3825 [ + + ]: 10350 : if (DO_AGGSPLIT_SERIALIZE(aggstate->aggsplit))
3826 : : {
3827 : : /* serialization only valid when not running finalfn */
3828 [ - + ]: 168 : Assert(DO_AGGSPLIT_SKIPFINAL(aggstate->aggsplit));
3829 : :
3830 [ - + ]: 168 : if (!OidIsValid(aggform->aggserialfn))
3359 tgl@sss.pgh.pa.us 3831 [ # # ]:UBC 0 : elog(ERROR, "serialfunc not provided for serialization aggregation");
3448 rhaas@postgresql.org 3832 :CBC 168 : serialfn_oid = aggform->aggserialfn;
3833 : : }
3834 : :
3835 : : /* Likewise for deserialization functions */
3359 tgl@sss.pgh.pa.us 3836 [ + + ]: 10350 : if (DO_AGGSPLIT_DESERIALIZE(aggstate->aggsplit))
3837 : : {
3838 : : /* deserialization only valid when combining states */
3839 [ - + ]: 60 : Assert(DO_AGGSPLIT_COMBINE(aggstate->aggsplit));
3840 : :
3841 [ - + ]: 60 : if (!OidIsValid(aggform->aggdeserialfn))
3359 tgl@sss.pgh.pa.us 3842 [ # # ]:UBC 0 : elog(ERROR, "deserialfunc not provided for deserialization aggregation");
3448 rhaas@postgresql.org 3843 :CBC 60 : deserialfn_oid = aggform->aggdeserialfn;
3844 : : }
3845 : : }
3846 : :
3847 : : /* Check that aggregate owner has permission to call component fns */
3848 : : {
3849 : : HeapTuple procTuple;
3850 : :
5683 3851 : 24619 : procTuple = SearchSysCache1(PROCOID,
3852 : : ObjectIdGetDatum(aggref->aggfnoid));
7527 tgl@sss.pgh.pa.us 3853 [ - + ]: 24619 : if (!HeapTupleIsValid(procTuple))
7527 tgl@sss.pgh.pa.us 3854 [ # # ]:UBC 0 : elog(ERROR, "cache lookup failed for function %u",
3855 : : aggref->aggfnoid);
7527 tgl@sss.pgh.pa.us 3856 :CBC 24619 : aggOwner = ((Form_pg_proc) GETSTRUCT(procTuple))->proowner;
3857 : 24619 : ReleaseSysCache(procTuple);
3858 : :
3859 [ + + ]: 24619 : if (OidIsValid(finalfn_oid))
3860 : : {
1028 peter@eisentraut.org 3861 : 11071 : aclresult = object_aclcheck(ProcedureRelationId, finalfn_oid, aggOwner,
3862 : : ACL_EXECUTE);
7527 tgl@sss.pgh.pa.us 3863 [ - + ]: 11071 : if (aclresult != ACLCHECK_OK)
2835 peter_e@gmx.net 3864 :UBC 0 : aclcheck_error(aclresult, OBJECT_FUNCTION,
7527 tgl@sss.pgh.pa.us 3865 : 0 : get_func_name(finalfn_oid));
4530 rhaas@postgresql.org 3866 [ - + ]:CBC 11071 : InvokeFunctionExecuteHook(finalfn_oid);
3867 : : }
3448 3868 [ + + ]: 24619 : if (OidIsValid(serialfn_oid))
3869 : : {
1028 peter@eisentraut.org 3870 : 168 : aclresult = object_aclcheck(ProcedureRelationId, serialfn_oid, aggOwner,
3871 : : ACL_EXECUTE);
3448 rhaas@postgresql.org 3872 [ - + ]: 168 : if (aclresult != ACLCHECK_OK)
2835 peter_e@gmx.net 3873 :UBC 0 : aclcheck_error(aclresult, OBJECT_FUNCTION,
3448 rhaas@postgresql.org 3874 : 0 : get_func_name(serialfn_oid));
3448 rhaas@postgresql.org 3875 [ - + ]:CBC 168 : InvokeFunctionExecuteHook(serialfn_oid);
3876 : : }
3877 [ + + ]: 24619 : if (OidIsValid(deserialfn_oid))
3878 : : {
1028 peter@eisentraut.org 3879 : 60 : aclresult = object_aclcheck(ProcedureRelationId, deserialfn_oid, aggOwner,
3880 : : ACL_EXECUTE);
3448 rhaas@postgresql.org 3881 [ - + ]: 60 : if (aclresult != ACLCHECK_OK)
2835 peter_e@gmx.net 3882 :UBC 0 : aclcheck_error(aclresult, OBJECT_FUNCTION,
3448 rhaas@postgresql.org 3883 : 0 : get_func_name(deserialfn_oid));
3448 rhaas@postgresql.org 3884 [ - + ]:CBC 60 : InvokeFunctionExecuteHook(deserialfn_oid);
3885 : : }
3886 : : }
3887 : :
3888 : : /*
3889 : : * Get actual datatypes of the (nominal) aggregate inputs. These
3890 : : * could be different from the agg's declared input types, when the
3891 : : * agg accepts ANY or a polymorphic type.
3892 : : */
1525 drowley@postgresql.o 3893 : 24619 : numAggTransFnArgs = get_aggregate_argtypes(aggref,
3894 : : aggTransFnInputTypes);
3895 : :
3896 : : /* Count the "direct" arguments, if any */
4275 tgl@sss.pgh.pa.us 3897 : 24619 : numDirectArgs = list_length(aggref->aggdirectargs);
3898 : :
3899 : : /* Detect how many arguments to pass to the finalfn */
3686 heikki.linnakangas@i 3900 [ + + ]: 24619 : if (aggform->aggfinalextra)
1525 drowley@postgresql.o 3901 : 7149 : peragg->numFinalArgs = numAggTransFnArgs + 1;
3902 : : else
3686 heikki.linnakangas@i 3903 : 17470 : peragg->numFinalArgs = numDirectArgs + 1;
3904 : :
3905 : : /* Initialize any direct-argument expressions */
2882 tgl@sss.pgh.pa.us 3906 : 24619 : peragg->aggdirectargs = ExecInitExprList(aggref->aggdirectargs,
3907 : : (PlanState *) aggstate);
3908 : :
3909 : : /*
3910 : : * build expression trees using actual argument & result types for the
3911 : : * finalfn, if it exists and is required.
3912 : : */
8103 3913 [ + + ]: 24619 : if (OidIsValid(finalfn_oid))
3914 : : {
1525 drowley@postgresql.o 3915 : 11071 : build_aggregate_finalfn_expr(aggTransFnInputTypes,
3916 : : peragg->numFinalArgs,
3917 : : aggtranstype,
3918 : : aggref->aggtype,
3919 : : aggref->inputcollid,
3920 : : finalfn_oid,
3921 : : &finalfnexpr);
3686 heikki.linnakangas@i 3922 : 11071 : fmgr_info(finalfn_oid, &peragg->finalfn);
3923 : 11071 : fmgr_info_set_expr((Node *) finalfnexpr, &peragg->finalfn);
3924 : : }
3925 : :
3926 : : /* get info about the output value's datatype */
3359 tgl@sss.pgh.pa.us 3927 : 24619 : get_typlenbyval(aggref->aggtype,
3928 : : &peragg->resulttypeLen,
3929 : : &peragg->resulttypeByVal);
3930 : :
3931 : : /*
3932 : : * Build working state for invoking the transition function, if we
3933 : : * haven't done it already.
3934 : : */
1747 heikki.linnakangas@i 3935 : 24619 : pertrans = &pertransstates[aggref->aggtransno];
3936 [ + + ]: 24619 : if (pertrans->aggref == NULL)
3937 : : {
3938 : : Datum textInitVal;
3939 : : Datum initValue;
3940 : : bool initValueIsNull;
3941 : : Oid transfn_oid;
3942 : :
3943 : : /*
3944 : : * If this aggregation is performing state combines, then instead
3945 : : * of using the transition function, we'll use the combine
3946 : : * function.
3947 : : */
3948 [ + + ]: 24490 : if (DO_AGGSPLIT_COMBINE(aggstate->aggsplit))
3949 : : {
3950 : 678 : transfn_oid = aggform->aggcombinefn;
3951 : :
3952 : : /* If not set then the planner messed up */
3953 [ - + ]: 678 : if (!OidIsValid(transfn_oid))
1747 heikki.linnakangas@i 3954 [ # # ]:UBC 0 : elog(ERROR, "combinefn not set for aggregate function");
3955 : : }
3956 : : else
1747 heikki.linnakangas@i 3957 :CBC 23812 : transfn_oid = aggform->aggtransfn;
3958 : :
1028 peter@eisentraut.org 3959 : 24490 : aclresult = object_aclcheck(ProcedureRelationId, transfn_oid, aggOwner, ACL_EXECUTE);
1747 heikki.linnakangas@i 3960 [ - + ]: 24490 : if (aclresult != ACLCHECK_OK)
1747 heikki.linnakangas@i 3961 :UBC 0 : aclcheck_error(aclresult, OBJECT_FUNCTION,
3962 : 0 : get_func_name(transfn_oid));
1747 heikki.linnakangas@i 3963 [ - + ]:CBC 24490 : InvokeFunctionExecuteHook(transfn_oid);
3964 : :
3965 : : /*
3966 : : * initval is potentially null, so don't try to access it as a
3967 : : * struct field. Must do it the hard way with SysCacheGetAttr.
3968 : : */
3969 : 24490 : textInitVal = SysCacheGetAttr(AGGFNOID, aggTuple,
3970 : : Anum_pg_aggregate_agginitval,
3971 : : &initValueIsNull);
3972 [ + + ]: 24490 : if (initValueIsNull)
3973 : 14486 : initValue = (Datum) 0;
3974 : : else
3975 : 10004 : initValue = GetAggInitVal(textInitVal, aggtranstype);
3976 : :
1525 drowley@postgresql.o 3977 [ + + ]: 24490 : if (DO_AGGSPLIT_COMBINE(aggstate->aggsplit))
3978 : : {
3979 : 678 : Oid combineFnInputTypes[] = {aggtranstype,
3980 : : aggtranstype};
3981 : :
3982 : : /*
3983 : : * When combining there's only one input, the to-be-combined
3984 : : * transition value. The transition value is not counted
3985 : : * here.
3986 : : */
3987 : 678 : pertrans->numTransInputs = 1;
3988 : :
3989 : : /* aggcombinefn always has two arguments of aggtranstype */
3990 : 678 : build_pertrans_for_aggref(pertrans, aggstate, estate,
3991 : : aggref, transfn_oid, aggtranstype,
3992 : : serialfn_oid, deserialfn_oid,
3993 : : initValue, initValueIsNull,
3994 : : combineFnInputTypes, 2);
3995 : :
3996 : : /*
3997 : : * Ensure that a combine function to combine INTERNAL states
3998 : : * is not strict. This should have been checked during CREATE
3999 : : * AGGREGATE, but the strict property could have been changed
4000 : : * since then.
4001 : : */
4002 [ + + - + ]: 678 : if (pertrans->transfn.fn_strict && aggtranstype == INTERNALOID)
1525 drowley@postgresql.o 4003 [ # # ]:UBC 0 : ereport(ERROR,
4004 : : (errcode(ERRCODE_INVALID_FUNCTION_DEFINITION),
4005 : : errmsg("combine function with transition type %s must not be declared STRICT",
4006 : : format_type_be(aggtranstype))));
4007 : : }
4008 : : else
4009 : : {
4010 : : /* Detect how many arguments to pass to the transfn */
1525 drowley@postgresql.o 4011 [ + + ]:CBC 23812 : if (AGGKIND_IS_ORDERED_SET(aggref->aggkind))
4012 : 126 : pertrans->numTransInputs = list_length(aggref->args);
4013 : : else
4014 : 23686 : pertrans->numTransInputs = numAggTransFnArgs;
4015 : :
4016 : 23812 : build_pertrans_for_aggref(pertrans, aggstate, estate,
4017 : : aggref, transfn_oid, aggtranstype,
4018 : : serialfn_oid, deserialfn_oid,
4019 : : initValue, initValueIsNull,
4020 : : aggTransFnInputTypes,
4021 : : numAggTransFnArgs);
4022 : :
4023 : : /*
4024 : : * If the transfn is strict and the initval is NULL, make sure
4025 : : * input type and transtype are the same (or at least
4026 : : * binary-compatible), so that it's OK to use the first
4027 : : * aggregated input value as the initial transValue. This
4028 : : * should have been checked at agg definition time, but we
4029 : : * must check again in case the transfn's strictness property
4030 : : * has been changed.
4031 : : */
4032 [ + + + + ]: 23812 : if (pertrans->transfn.fn_strict && pertrans->initValueIsNull)
4033 : : {
4034 [ + - ]: 2525 : if (numAggTransFnArgs <= numDirectArgs ||
4035 [ - + ]: 2525 : !IsBinaryCoercible(aggTransFnInputTypes[numDirectArgs],
4036 : : aggtranstype))
1525 drowley@postgresql.o 4037 [ # # ]:UBC 0 : ereport(ERROR,
4038 : : (errcode(ERRCODE_INVALID_FUNCTION_DEFINITION),
4039 : : errmsg("aggregate %u needs to have compatible input type and transition type",
4040 : : aggref->aggfnoid)));
4041 : : }
4042 : : }
4043 : : }
4044 : : else
1747 heikki.linnakangas@i 4045 :CBC 129 : pertrans->aggshared = true;
3686 4046 : 24619 : ReleaseSysCache(aggTuple);
4047 : : }
4048 : :
4049 : : /*
4050 : : * Last, check whether any more aggregates got added onto the node while
4051 : : * we processed the expressions for the aggregate arguments (including not
4052 : : * only the regular arguments and FILTER expressions handled immediately
4053 : : * above, but any direct arguments we might've handled earlier). If so,
4054 : : * we have nested aggregate functions, which is semantically nonsensical,
4055 : : * so complain. (This should have been caught by the parser, so we don't
4056 : : * need to work hard on a helpful error message; but we defend against it
4057 : : * here anyway, just to be sure.)
4058 : : */
1747 4059 [ - + ]: 22043 : if (numaggrefs != list_length(aggstate->aggs))
2797 andres@anarazel.de 4060 [ # # ]:UBC 0 : ereport(ERROR,
4061 : : (errcode(ERRCODE_GROUPING_ERROR),
4062 : : errmsg("aggregate function calls cannot be nested")));
4063 : :
4064 : : /*
4065 : : * Build expressions doing all the transition work at once. We build a
4066 : : * different one for each phase, as the number of transition function
4067 : : * invocation can differ between phases. Note this'll work both for
4068 : : * transition and combination functions (although there'll only be one
4069 : : * phase in the latter case).
4070 : : */
2797 andres@anarazel.de 4071 [ + + ]:CBC 63533 : for (phaseidx = 0; phaseidx < aggstate->numphases; phaseidx++)
4072 : : {
4073 : 41490 : AggStatePerPhase phase = &aggstate->phases[phaseidx];
4074 : 41490 : bool dohash = false;
4075 : 41490 : bool dosort = false;
4076 : :
4077 : : /* phase 0 doesn't necessarily exist */
4078 [ + + ]: 41490 : if (!phase->aggnode)
4079 : 19211 : continue;
4080 : :
4081 [ + + + + ]: 22279 : if (aggstate->aggstrategy == AGG_MIXED && phaseidx == 1)
4082 : : {
4083 : : /*
4084 : : * Phase one, and only phase one, in a mixed agg performs both
4085 : : * sorting and aggregation.
4086 : : */
4087 : 116 : dohash = true;
4088 : 116 : dosort = true;
4089 : : }
4090 [ + + + + ]: 22163 : else if (aggstate->aggstrategy == AGG_MIXED && phaseidx == 0)
4091 : : {
4092 : : /*
4093 : : * No need to compute a transition function for an AGG_MIXED phase
4094 : : * 0 - the contents of the hashtables will have been computed
4095 : : * during phase 1.
4096 : : */
4097 : 116 : continue;
4098 : : }
4099 [ + + ]: 22047 : else if (phase->aggstrategy == AGG_PLAIN ||
4100 [ + + ]: 3833 : phase->aggstrategy == AGG_SORTED)
4101 : : {
4102 : 19331 : dohash = false;
4103 : 19331 : dosort = true;
4104 : : }
4105 [ + - ]: 2716 : else if (phase->aggstrategy == AGG_HASHED)
4106 : : {
4107 : 2716 : dohash = true;
4108 : 2716 : dosort = false;
4109 : : }
4110 : : else
2797 andres@anarazel.de 4111 :UBC 0 : Assert(false);
4112 : :
2012 jdavis@postgresql.or 4113 :CBC 22163 : phase->evaltrans = ExecBuildAggTrans(aggstate, phase, dosort, dohash,
4114 : : false);
4115 : :
4116 : : /* cache compiled expression for outer slot without NULL check */
1998 4117 : 22163 : phase->evaltrans_cache[0][0] = phase->evaltrans;
4118 : : }
4119 : :
3686 heikki.linnakangas@i 4120 : 22043 : return aggstate;
4121 : : }
4122 : :
4123 : : /*
4124 : : * Build the state needed to calculate a state value for an aggregate.
4125 : : *
4126 : : * This initializes all the fields in 'pertrans'. 'aggref' is the aggregate
4127 : : * to initialize the state for. 'transfn_oid', 'aggtranstype', and the rest
4128 : : * of the arguments could be calculated from 'aggref', but the caller has
4129 : : * calculated them already, so might as well pass them.
4130 : : *
4131 : : * 'transfn_oid' may be either the Oid of the aggtransfn or the aggcombinefn.
4132 : : */
4133 : : static void
4134 : 24490 : build_pertrans_for_aggref(AggStatePerTrans pertrans,
4135 : : AggState *aggstate, EState *estate,
4136 : : Aggref *aggref,
4137 : : Oid transfn_oid, Oid aggtranstype,
4138 : : Oid aggserialfn, Oid aggdeserialfn,
4139 : : Datum initValue, bool initValueIsNull,
4140 : : Oid *inputTypes, int numArguments)
4141 : : {
4142 : 24490 : int numGroupingSets = Max(aggstate->maxsets, 1);
4143 : : Expr *transfnexpr;
4144 : : int numTransArgs;
3448 rhaas@postgresql.org 4145 : 24490 : Expr *serialfnexpr = NULL;
4146 : 24490 : Expr *deserialfnexpr = NULL;
4147 : : ListCell *lc;
4148 : : int numInputs;
4149 : : int numDirectArgs;
4150 : : List *sortlist;
4151 : : int numSortCols;
4152 : : int numDistinctCols;
4153 : : int i;
4154 : :
4155 : : /* Begin filling in the pertrans data */
3686 heikki.linnakangas@i 4156 : 24490 : pertrans->aggref = aggref;
2882 tgl@sss.pgh.pa.us 4157 : 24490 : pertrans->aggshared = false;
3686 heikki.linnakangas@i 4158 : 24490 : pertrans->aggCollation = aggref->inputcollid;
1525 drowley@postgresql.o 4159 : 24490 : pertrans->transfn_oid = transfn_oid;
3448 rhaas@postgresql.org 4160 : 24490 : pertrans->serialfn_oid = aggserialfn;
4161 : 24490 : pertrans->deserialfn_oid = aggdeserialfn;
3686 heikki.linnakangas@i 4162 : 24490 : pertrans->initValue = initValue;
4163 : 24490 : pertrans->initValueIsNull = initValueIsNull;
4164 : :
4165 : : /* Count the "direct" arguments, if any */
4166 : 24490 : numDirectArgs = list_length(aggref->aggdirectargs);
4167 : :
4168 : : /* Count the number of aggregated input columns */
4169 : 24490 : pertrans->numInputs = numInputs = list_length(aggref->args);
4170 : :
4171 : 24490 : pertrans->aggtranstype = aggtranstype;
4172 : :
4173 : : /* account for the current transition state */
1525 drowley@postgresql.o 4174 : 24490 : numTransArgs = pertrans->numTransInputs + 1;
4175 : :
4176 : : /*
4177 : : * Set up infrastructure for calling the transfn. Note that invtransfn is
4178 : : * not needed here.
4179 : : */
4180 : 24490 : build_aggregate_transfn_expr(inputTypes,
4181 : : numArguments,
4182 : : numDirectArgs,
4183 : 24490 : aggref->aggvariadic,
4184 : : aggtranstype,
4185 : : aggref->inputcollid,
4186 : : transfn_oid,
4187 : : InvalidOid,
4188 : : &transfnexpr,
4189 : : NULL);
4190 : :
4191 : 24490 : fmgr_info(transfn_oid, &pertrans->transfn);
4192 : 24490 : fmgr_info_set_expr((Node *) transfnexpr, &pertrans->transfn);
4193 : :
4194 : 24490 : pertrans->transfn_fcinfo =
4195 : 24490 : (FunctionCallInfo) palloc(SizeForFunctionCallInfo(numTransArgs));
4196 : 24490 : InitFunctionCallInfoData(*pertrans->transfn_fcinfo,
4197 : : &pertrans->transfn,
4198 : : numTransArgs,
4199 : : pertrans->aggCollation,
4200 : : (Node *) aggstate, NULL);
4201 : :
4202 : : /* get info about the state value's datatype */
3686 heikki.linnakangas@i 4203 : 24490 : get_typlenbyval(aggtranstype,
4204 : : &pertrans->transtypeLen,
4205 : : &pertrans->transtypeByVal);
4206 : :
3448 rhaas@postgresql.org 4207 [ + + ]: 24490 : if (OidIsValid(aggserialfn))
4208 : : {
3363 tgl@sss.pgh.pa.us 4209 : 168 : build_aggregate_serialfn_expr(aggserialfn,
4210 : : &serialfnexpr);
3448 rhaas@postgresql.org 4211 : 168 : fmgr_info(aggserialfn, &pertrans->serialfn);
4212 : 168 : fmgr_info_set_expr((Node *) serialfnexpr, &pertrans->serialfn);
4213 : :
2415 andres@anarazel.de 4214 : 168 : pertrans->serialfn_fcinfo =
4215 : 168 : (FunctionCallInfo) palloc(SizeForFunctionCallInfo(1));
4216 : 168 : InitFunctionCallInfoData(*pertrans->serialfn_fcinfo,
4217 : : &pertrans->serialfn,
4218 : : 1,
4219 : : InvalidOid,
4220 : : (Node *) aggstate, NULL);
4221 : : }
4222 : :
3448 rhaas@postgresql.org 4223 [ + + ]: 24490 : if (OidIsValid(aggdeserialfn))
4224 : : {
3363 tgl@sss.pgh.pa.us 4225 : 60 : build_aggregate_deserialfn_expr(aggdeserialfn,
4226 : : &deserialfnexpr);
3448 rhaas@postgresql.org 4227 : 60 : fmgr_info(aggdeserialfn, &pertrans->deserialfn);
4228 : 60 : fmgr_info_set_expr((Node *) deserialfnexpr, &pertrans->deserialfn);
4229 : :
2415 andres@anarazel.de 4230 : 60 : pertrans->deserialfn_fcinfo =
4231 : 60 : (FunctionCallInfo) palloc(SizeForFunctionCallInfo(2));
4232 : 60 : InitFunctionCallInfoData(*pertrans->deserialfn_fcinfo,
4233 : : &pertrans->deserialfn,
4234 : : 2,
4235 : : InvalidOid,
4236 : : (Node *) aggstate, NULL);
4237 : : }
4238 : :
4239 : : /*
4240 : : * If we're doing either DISTINCT or ORDER BY for a plain agg, then we
4241 : : * have a list of SortGroupClause nodes; fish out the data in them and
4242 : : * stick them into arrays. We ignore ORDER BY for an ordered-set agg,
4243 : : * however; the agg's transfn and finalfn are responsible for that.
4244 : : *
4245 : : * When the planner has set the aggpresorted flag, the input to the
4246 : : * aggregate is already correctly sorted. For ORDER BY aggregates we can
4247 : : * simply treat these as normal aggregates. For presorted DISTINCT
4248 : : * aggregates an extra step must be added to remove duplicate consecutive
4249 : : * inputs.
4250 : : *
4251 : : * Note that by construction, if there is a DISTINCT clause then the ORDER
4252 : : * BY clause is a prefix of it (see transformDistinctClause).
4253 : : */
3686 heikki.linnakangas@i 4254 [ + + ]: 24490 : if (AGGKIND_IS_ORDERED_SET(aggref->aggkind))
4255 : : {
4256 : 126 : sortlist = NIL;
4257 : 126 : numSortCols = numDistinctCols = 0;
1131 drowley@postgresql.o 4258 : 126 : pertrans->aggsortrequired = false;
4259 : : }
4260 [ + + + + ]: 24364 : else if (aggref->aggpresorted && aggref->aggdistinct == NIL)
4261 : : {
4262 : 1044 : sortlist = NIL;
4263 : 1044 : numSortCols = numDistinctCols = 0;
4264 : 1044 : pertrans->aggsortrequired = false;
4265 : : }
3686 heikki.linnakangas@i 4266 [ + + ]: 23320 : else if (aggref->aggdistinct)
4267 : : {
4268 : 291 : sortlist = aggref->aggdistinct;
4269 : 291 : numSortCols = numDistinctCols = list_length(sortlist);
4270 [ - + ]: 291 : Assert(numSortCols >= list_length(aggref->aggorder));
1131 drowley@postgresql.o 4271 : 291 : pertrans->aggsortrequired = !aggref->aggpresorted;
4272 : : }
4273 : : else
4274 : : {
3686 heikki.linnakangas@i 4275 : 23029 : sortlist = aggref->aggorder;
4276 : 23029 : numSortCols = list_length(sortlist);
4277 : 23029 : numDistinctCols = 0;
1131 drowley@postgresql.o 4278 : 23029 : pertrans->aggsortrequired = (numSortCols > 0);
4279 : : }
4280 : :
3686 heikki.linnakangas@i 4281 : 24490 : pertrans->numSortCols = numSortCols;
4282 : 24490 : pertrans->numDistinctCols = numDistinctCols;
4283 : :
4284 : : /*
4285 : : * If we have either sorting or filtering to do, create a tupledesc and
4286 : : * slot corresponding to the aggregated inputs (including sort
4287 : : * expressions) of the agg.
4288 : : */
2882 tgl@sss.pgh.pa.us 4289 [ + + + + ]: 24490 : if (numSortCols > 0 || aggref->aggfilter)
4290 : : {
2482 andres@anarazel.de 4291 : 715 : pertrans->sortdesc = ExecTypeFromTL(aggref->args);
2759 4292 : 715 : pertrans->sortslot =
2487 4293 : 715 : ExecInitExtraTupleSlot(estate, pertrans->sortdesc,
4294 : : &TTSOpsMinimalTuple);
4295 : : }
4296 : :
2882 tgl@sss.pgh.pa.us 4297 [ + + ]: 24490 : if (numSortCols > 0)
4298 : : {
4299 : : /*
4300 : : * We don't implement DISTINCT or ORDER BY aggs in the HASHED case
4301 : : * (yet)
4302 : : */
3085 rhodiumtoad@postgres 4303 [ + - - + ]: 360 : Assert(aggstate->aggstrategy != AGG_HASHED && aggstate->aggstrategy != AGG_MIXED);
4304 : :
4305 : : /* ORDER BY aggregates are not supported with partial aggregation */
1525 drowley@postgresql.o 4306 [ - + ]: 360 : Assert(!DO_AGGSPLIT_COMBINE(aggstate->aggsplit));
4307 : :
4308 : : /* If we have only one input, we need its len/byval info. */
3686 heikki.linnakangas@i 4309 [ + + ]: 360 : if (numInputs == 1)
4310 : : {
4311 : 285 : get_typlenbyval(inputTypes[numDirectArgs],
4312 : : &pertrans->inputtypeLen,
4313 : : &pertrans->inputtypeByVal);
4314 : : }
4315 [ + + ]: 75 : else if (numDistinctCols > 0)
4316 : : {
4317 : : /* we will need an extra slot to store prior values */
2759 andres@anarazel.de 4318 : 54 : pertrans->uniqslot =
2487 4319 : 54 : ExecInitExtraTupleSlot(estate, pertrans->sortdesc,
4320 : : &TTSOpsMinimalTuple);
4321 : : }
4322 : :
4323 : : /* Extract the sort information for use later */
3686 heikki.linnakangas@i 4324 : 360 : pertrans->sortColIdx =
4325 : 360 : (AttrNumber *) palloc(numSortCols * sizeof(AttrNumber));
4326 : 360 : pertrans->sortOperators =
4327 : 360 : (Oid *) palloc(numSortCols * sizeof(Oid));
4328 : 360 : pertrans->sortCollations =
4329 : 360 : (Oid *) palloc(numSortCols * sizeof(Oid));
4330 : 360 : pertrans->sortNullsFirst =
4331 : 360 : (bool *) palloc(numSortCols * sizeof(bool));
4332 : :
4333 : 360 : i = 0;
4334 [ + - + + : 819 : foreach(lc, sortlist)
+ + ]
4335 : : {
4336 : 459 : SortGroupClause *sortcl = (SortGroupClause *) lfirst(lc);
4337 : 459 : TargetEntry *tle = get_sortgroupclause_tle(sortcl, aggref->args);
4338 : :
4339 : : /* the parser should have made sure of this */
4340 [ - + ]: 459 : Assert(OidIsValid(sortcl->sortop));
4341 : :
4342 : 459 : pertrans->sortColIdx[i] = tle->resno;
4343 : 459 : pertrans->sortOperators[i] = sortcl->sortop;
4344 : 459 : pertrans->sortCollations[i] = exprCollation((Node *) tle->expr);
4345 : 459 : pertrans->sortNullsFirst[i] = sortcl->nulls_first;
4346 : 459 : i++;
4347 : : }
4348 [ - + ]: 360 : Assert(i == numSortCols);
4349 : : }
4350 : :
4351 [ + + ]: 24490 : if (aggref->aggdistinct)
4352 : : {
4353 : : Oid *ops;
4354 : :
4355 [ - + ]: 291 : Assert(numArguments > 0);
2760 andres@anarazel.de 4356 [ - + ]: 291 : Assert(list_length(aggref->aggdistinct) == numDistinctCols);
4357 : :
4358 : 291 : ops = palloc(numDistinctCols * sizeof(Oid));
4359 : :
3686 heikki.linnakangas@i 4360 : 291 : i = 0;
4361 [ + - + + : 672 : foreach(lc, aggref->aggdistinct)
+ + ]
2760 andres@anarazel.de 4362 : 381 : ops[i++] = ((SortGroupClause *) lfirst(lc))->eqop;
4363 : :
4364 : : /* lookup / build the necessary comparators */
4365 [ + + ]: 291 : if (numDistinctCols == 1)
4366 : 237 : fmgr_info(get_opcode(ops[0]), &pertrans->equalfnOne);
4367 : : else
4368 : 54 : pertrans->equalfnMulti =
4369 : 54 : execTuplesMatchPrepare(pertrans->sortdesc,
4370 : : numDistinctCols,
4371 : 54 : pertrans->sortColIdx,
4372 : : ops,
2360 peter@eisentraut.org 4373 : 54 : pertrans->sortCollations,
4374 : : &aggstate->ss.ps);
2760 andres@anarazel.de 4375 : 291 : pfree(ops);
4376 : : }
4377 : :
3686 heikki.linnakangas@i 4378 : 24490 : pertrans->sortstates = (Tuplesortstate **)
4379 : 24490 : palloc0(sizeof(Tuplesortstate *) * numGroupingSets);
10651 scrappy@hub.org 4380 : 24490 : }
4381 : :
4382 : :
4383 : : static Datum
8549 tgl@sss.pgh.pa.us 4384 : 10004 : GetAggInitVal(Datum textInitVal, Oid transtype)
4385 : : {
4386 : : Oid typinput,
4387 : : typioparam;
4388 : : char *strInitVal;
4389 : : Datum initVal;
4390 : :
7762 4391 : 10004 : getTypeInputInfo(transtype, &typinput, &typioparam);
6374 4392 : 10004 : strInitVal = TextDatumGetCString(textInitVal);
7095 4393 : 10004 : initVal = OidInputFunctionCall(typinput, strInitVal,
4394 : : typioparam, -1);
8549 4395 : 10004 : pfree(strInitVal);
4396 : 10004 : return initVal;
4397 : : }
4398 : :
4399 : : void
8311 4400 : 21953 : ExecEndAgg(AggState *node)
4401 : : {
4402 : : PlanState *outerPlan;
4403 : : int transno;
3766 andres@anarazel.de 4404 : 21953 : int numGroupingSets = Max(node->maxsets, 1);
4405 : : int setno;
4406 : :
4407 : : /*
4408 : : * When ending a parallel worker, copy the statistics gathered by the
4409 : : * worker back into shared memory so that it can be picked up by the main
4410 : : * process to report in EXPLAIN ANALYZE.
4411 : : */
1905 drowley@postgresql.o 4412 [ + + + + ]: 21953 : if (node->shared_info && IsParallelWorker())
4413 : : {
4414 : : AggregateInstrumentation *si;
4415 : :
4416 [ - + ]: 84 : Assert(ParallelWorkerNumber <= node->shared_info->num_workers);
4417 : 84 : si = &node->shared_info->sinstrument[ParallelWorkerNumber];
4418 : 84 : si->hash_batches_used = node->hash_batches_used;
4419 : 84 : si->hash_disk_used = node->hash_disk_used;
4420 : 84 : si->hash_mem_peak = node->hash_mem_peak;
4421 : : }
4422 : :
4423 : : /* Make sure we have closed any open tuplesorts */
4424 : :
3766 andres@anarazel.de 4425 [ + + ]: 21953 : if (node->sort_in)
4426 : 78 : tuplesort_end(node->sort_in);
4427 [ + + ]: 21953 : if (node->sort_out)
4428 : 21 : tuplesort_end(node->sort_out);
4429 : :
1998 jdavis@postgresql.or 4430 : 21953 : hashagg_reset_spill_state(node);
4431 : :
4432 [ + + ]: 21953 : if (node->hash_metacxt != NULL)
4433 : : {
4434 : 2828 : MemoryContextDelete(node->hash_metacxt);
4435 : 2828 : node->hash_metacxt = NULL;
4436 : : }
166 4437 [ + + ]: 21953 : if (node->hash_tablecxt != NULL)
4438 : : {
4439 : 2828 : MemoryContextDelete(node->hash_tablecxt);
4440 : 2828 : node->hash_tablecxt = NULL;
4441 : : }
4442 : :
4443 : :
3686 heikki.linnakangas@i 4444 [ + + ]: 46351 : for (transno = 0; transno < node->numtrans; transno++)
4445 : : {
4446 : 24398 : AggStatePerTrans pertrans = &node->pertrans[transno];
4447 : :
3766 andres@anarazel.de 4448 [ + + ]: 49303 : for (setno = 0; setno < numGroupingSets; setno++)
4449 : : {
3686 heikki.linnakangas@i 4450 [ - + ]: 24905 : if (pertrans->sortstates[setno])
3686 heikki.linnakangas@i 4451 :UBC 0 : tuplesort_end(pertrans->sortstates[setno]);
4452 : : }
4453 : : }
4454 : :
4455 : : /* And ensure any agg shutdown callbacks have been called */
3766 andres@anarazel.de 4456 [ + + ]:CBC 44326 : for (setno = 0; setno < numGroupingSets; setno++)
4457 : 22373 : ReScanExprContext(node->aggcontexts[setno]);
3085 rhodiumtoad@postgres 4458 [ + + ]: 21953 : if (node->hashcontext)
4459 : 2828 : ReScanExprContext(node->hashcontext);
4460 : :
8311 tgl@sss.pgh.pa.us 4461 : 21953 : outerPlan = outerPlanState(node);
4462 : 21953 : ExecEndNode(outerPlan);
10651 scrappy@hub.org 4463 : 21953 : }
4464 : :
4465 : : void
5535 tgl@sss.pgh.pa.us 4466 : 27209 : ExecReScanAgg(AggState *node)
4467 : : {
8311 4468 : 27209 : ExprContext *econtext = node->ss.ps.ps_ExprContext;
3759 bruce@momjian.us 4469 : 27209 : PlanState *outerPlan = outerPlanState(node);
3766 andres@anarazel.de 4470 : 27209 : Agg *aggnode = (Agg *) node->ss.ps.plan;
4471 : : int transno;
3759 bruce@momjian.us 4472 : 27209 : int numGroupingSets = Max(node->maxsets, 1);
4473 : : int setno;
4474 : :
8135 tgl@sss.pgh.pa.us 4475 : 27209 : node->agg_done = false;
4476 : :
3085 rhodiumtoad@postgres 4477 [ + + ]: 27209 : if (node->aggstrategy == AGG_HASHED)
4478 : : {
4479 : : /*
4480 : : * In the hashed case, if we haven't yet built the hash table then we
4481 : : * can just return; nothing done yet, so nothing to undo. If subnode's
4482 : : * chgParam is not NULL then it will be re-scanned by ExecProcNode,
4483 : : * else no reason to re-scan it at all.
4484 : : */
8135 tgl@sss.pgh.pa.us 4485 [ + + ]: 6647 : if (!node->table_filled)
4486 : 75 : return;
4487 : :
4488 : : /*
4489 : : * If we do have the hash table, and it never spilled, and the subplan
4490 : : * does not have any parameter changes, and none of our own parameter
4491 : : * changes affect input expressions of the aggregated functions, then
4492 : : * we can just rescan the existing hash table; no need to build it
4493 : : * again.
4494 : : */
1998 jdavis@postgresql.or 4495 [ + + + - ]: 6572 : if (outerPlan->chgParam == NULL && !node->hash_ever_spilled &&
3300 tgl@sss.pgh.pa.us 4496 [ + + ]: 446 : !bms_overlap(node->ss.ps.chgParam, aggnode->aggParams))
4497 : : {
3085 rhodiumtoad@postgres 4498 : 434 : ResetTupleHashIterator(node->perhash[0].hashtable,
4499 : : &node->perhash[0].hashiter);
4500 : 434 : select_current_set(node, 0, true);
8135 tgl@sss.pgh.pa.us 4501 : 434 : return;
4502 : : }
4503 : : }
4504 : :
4505 : : /* Make sure we have closed any open tuplesorts */
3686 heikki.linnakangas@i 4506 [ + + ]: 60816 : for (transno = 0; transno < node->numtrans; transno++)
4507 : : {
3766 andres@anarazel.de 4508 [ + + ]: 68250 : for (setno = 0; setno < numGroupingSets; setno++)
4509 : : {
3686 heikki.linnakangas@i 4510 : 34134 : AggStatePerTrans pertrans = &node->pertrans[transno];
4511 : :
4512 [ - + ]: 34134 : if (pertrans->sortstates[setno])
4513 : : {
3686 heikki.linnakangas@i 4514 :UBC 0 : tuplesort_end(pertrans->sortstates[setno]);
4515 : 0 : pertrans->sortstates[setno] = NULL;
4516 : : }
4517 : : }
4518 : : }
4519 : :
4520 : : /*
4521 : : * We don't need to ReScanExprContext the output tuple context here;
4522 : : * ExecReScan already did it. But we do need to reset our per-grouping-set
4523 : : * contexts, which may have transvalues stored in them. (We use rescan
4524 : : * rather than just reset because transfns may have registered callbacks
4525 : : * that need to be run now.) For the AGG_HASHED case, see below.
4526 : : */
4527 : :
3766 andres@anarazel.de 4528 [ + + ]:CBC 53418 : for (setno = 0; setno < numGroupingSets; setno++)
4529 : : {
4530 : 26718 : ReScanExprContext(node->aggcontexts[setno]);
4531 : : }
4532 : :
4533 : : /* Release first tuple of group, if we have made a copy */
8311 tgl@sss.pgh.pa.us 4534 [ - + ]: 26700 : if (node->grp_firstTuple != NULL)
4535 : : {
8311 tgl@sss.pgh.pa.us 4536 :UBC 0 : heap_freetuple(node->grp_firstTuple);
4537 : 0 : node->grp_firstTuple = NULL;
4538 : : }
3766 andres@anarazel.de 4539 :CBC 26700 : ExecClearTuple(node->ss.ss_ScanTupleSlot);
4540 : :
4541 : : /* Forget current agg values */
8311 tgl@sss.pgh.pa.us 4542 [ + - + - : 60816 : MemSet(econtext->ecxt_aggvalues, 0, sizeof(Datum) * node->numaggs);
+ - + - +
+ ]
4543 [ + - + + : 26700 : MemSet(econtext->ecxt_aggnulls, 0, sizeof(bool) * node->numaggs);
+ - + - -
+ ]
4544 : :
4545 : : /*
4546 : : * With AGG_HASHED/MIXED, the hash table is allocated in a sub-context of
4547 : : * the hashcontext. This used to be an issue, but now, resetting a context
4548 : : * automatically deletes sub-contexts too.
4549 : : */
3085 rhodiumtoad@postgres 4550 [ + + + + ]: 26700 : if (node->aggstrategy == AGG_HASHED || node->aggstrategy == AGG_MIXED)
4551 : : {
1998 jdavis@postgresql.or 4552 : 6153 : hashagg_reset_spill_state(node);
4553 : :
4554 : 6153 : node->hash_ever_spilled = false;
4555 : 6153 : node->hash_spill_mode = false;
4556 : 6153 : node->hash_ngroups_current = 0;
4557 : :
3085 rhodiumtoad@postgres 4558 : 6153 : ReScanExprContext(node->hashcontext);
166 jdavis@postgresql.or 4559 : 6153 : MemoryContextReset(node->hash_tablecxt);
4560 : : /* Rebuild an empty hash table */
2026 4561 : 6153 : build_hash_tables(node);
8311 tgl@sss.pgh.pa.us 4562 : 6153 : node->table_filled = false;
4563 : : /* iterator will be reset when the table is filled */
4564 : :
1998 jdavis@postgresql.or 4565 : 6153 : hashagg_recompile_expressions(node, false, false);
4566 : : }
4567 : :
3085 rhodiumtoad@postgres 4568 [ + + ]: 26700 : if (node->aggstrategy != AGG_HASHED)
4569 : : {
4570 : : /*
4571 : : * Reset the per-group state (in particular, mark transvalues null)
4572 : : */
2804 andres@anarazel.de 4573 [ + + ]: 41142 : for (setno = 0; setno < numGroupingSets; setno++)
4574 : : {
4575 [ + - + - : 88800 : MemSet(node->pergroups[setno], 0,
+ - + - +
+ ]
4576 : : sizeof(AggStatePerGroupData) * node->numaggs);
4577 : : }
4578 : :
4579 : : /* reset to phase 1 */
3085 rhodiumtoad@postgres 4580 : 20562 : initialize_phase(node, 1);
4581 : :
3766 andres@anarazel.de 4582 : 20562 : node->input_done = false;
4583 : 20562 : node->projected_set = -1;
4584 : : }
4585 : :
3778 rhaas@postgresql.org 4586 [ + + ]: 26700 : if (outerPlan->chgParam == NULL)
4587 : 94 : ExecReScan(outerPlan);
4588 : : }
4589 : :
4590 : :
4591 : : /***********************************************************************
4592 : : * API exposed to aggregate functions
4593 : : ***********************************************************************/
4594 : :
4595 : :
4596 : : /*
4597 : : * AggCheckCallContext - test if a SQL function is being called as an aggregate
4598 : : *
4599 : : * The transition and/or final functions of an aggregate may want to verify
4600 : : * that they are being called as aggregates, rather than as plain SQL
4601 : : * functions. They should use this function to do so. The return value
4602 : : * is nonzero if being called as an aggregate, or zero if not. (Specific
4603 : : * nonzero values are AGG_CONTEXT_AGGREGATE or AGG_CONTEXT_WINDOW, but more
4604 : : * values could conceivably appear in future.)
4605 : : *
4606 : : * If aggcontext isn't NULL, the function also stores at *aggcontext the
4607 : : * identity of the memory context that aggregate transition values are being
4608 : : * stored in. Note that the same aggregate call site (flinfo) may be called
4609 : : * interleaved on different transition values in different contexts, so it's
4610 : : * not kosher to cache aggcontext under fn_extra. It is, however, kosher to
4611 : : * cache it in the transvalue itself (for internal-type transvalues).
4612 : : */
4613 : : int
5689 tgl@sss.pgh.pa.us 4614 : 2676201 : AggCheckCallContext(FunctionCallInfo fcinfo, MemoryContext *aggcontext)
4615 : : {
4616 [ + + + + ]: 2676201 : if (fcinfo->context && IsA(fcinfo->context, AggState))
4617 : : {
4618 [ + + ]: 2670434 : if (aggcontext)
4619 : : {
3759 bruce@momjian.us 4620 : 1284173 : AggState *aggstate = ((AggState *) fcinfo->context);
3085 rhodiumtoad@postgres 4621 : 1284173 : ExprContext *cxt = aggstate->curaggcontext;
4622 : :
3766 andres@anarazel.de 4623 : 1284173 : *aggcontext = cxt->ecxt_per_tuple_memory;
4624 : : }
5689 tgl@sss.pgh.pa.us 4625 : 2670434 : return AGG_CONTEXT_AGGREGATE;
4626 : : }
4627 [ + + + - ]: 5767 : if (fcinfo->context && IsA(fcinfo->context, WindowAggState))
4628 : : {
4629 [ + + ]: 4830 : if (aggcontext)
4165 4630 : 355 : *aggcontext = ((WindowAggState *) fcinfo->context)->curaggcontext;
5689 4631 : 4830 : return AGG_CONTEXT_WINDOW;
4632 : : }
4633 : :
4634 : : /* this is just to prevent "uninitialized variable" warnings */
4635 [ + + ]: 937 : if (aggcontext)
4636 : 913 : *aggcontext = NULL;
4637 : 937 : return 0;
4638 : : }
4639 : :
4640 : : /*
4641 : : * AggGetAggref - allow an aggregate support function to get its Aggref
4642 : : *
4643 : : * If the function is being called as an aggregate support function,
4644 : : * return the Aggref node for the aggregate call. Otherwise, return NULL.
4645 : : *
4646 : : * Aggregates sharing the same inputs and transition functions can get
4647 : : * merged into a single transition calculation. If the transition function
4648 : : * calls AggGetAggref, it will get some one of the Aggrefs for which it is
4649 : : * executing. It must therefore not pay attention to the Aggref fields that
4650 : : * relate to the final function, as those are indeterminate. But if a final
4651 : : * function calls AggGetAggref, it will get a precise result.
4652 : : *
4653 : : * Note that if an aggregate is being used as a window function, this will
4654 : : * return NULL. We could provide a similar function to return the relevant
4655 : : * WindowFunc node in such cases, but it's not needed yet.
4656 : : */
4657 : : Aggref *
4275 4658 : 123 : AggGetAggref(FunctionCallInfo fcinfo)
4659 : : {
4660 [ + - + - ]: 123 : if (fcinfo->context && IsA(fcinfo->context, AggState))
4661 : : {
2882 4662 : 123 : AggState *aggstate = (AggState *) fcinfo->context;
4663 : : AggStatePerAgg curperagg;
4664 : : AggStatePerTrans curpertrans;
4665 : :
4666 : : /* check curperagg (valid when in a final function) */
4667 : 123 : curperagg = aggstate->curperagg;
4668 : :
2886 4669 [ - + ]: 123 : if (curperagg)
2886 tgl@sss.pgh.pa.us 4670 :UBC 0 : return curperagg->aggref;
4671 : :
4672 : : /* check curpertrans (valid when in a transition function) */
2882 tgl@sss.pgh.pa.us 4673 :CBC 123 : curpertrans = aggstate->curpertrans;
4674 : :
3686 heikki.linnakangas@i 4675 [ + - ]: 123 : if (curpertrans)
4676 : 123 : return curpertrans->aggref;
4677 : : }
4275 tgl@sss.pgh.pa.us 4678 :UBC 0 : return NULL;
4679 : : }
4680 : :
4681 : : /*
4682 : : * AggGetTempMemoryContext - fetch short-term memory context for aggregates
4683 : : *
4684 : : * This is useful in agg final functions; the context returned is one that
4685 : : * the final function can safely reset as desired. This isn't useful for
4686 : : * transition functions, since the context returned MAY (we don't promise)
4687 : : * be the same as the context those are called in.
4688 : : *
4689 : : * As above, this is currently not useful for aggs called as window functions.
4690 : : */
4691 : : MemoryContext
4083 4692 : 0 : AggGetTempMemoryContext(FunctionCallInfo fcinfo)
4693 : : {
4275 4694 [ # # # # ]: 0 : if (fcinfo->context && IsA(fcinfo->context, AggState))
4695 : : {
4696 : 0 : AggState *aggstate = (AggState *) fcinfo->context;
4697 : :
4083 4698 : 0 : return aggstate->tmpcontext->ecxt_per_tuple_memory;
4699 : : }
4275 4700 : 0 : return NULL;
4701 : : }
4702 : :
4703 : : /*
4704 : : * AggStateIsShared - find out whether transition state is shared
4705 : : *
4706 : : * If the function is being called as an aggregate support function,
4707 : : * return true if the aggregate's transition state is shared across
4708 : : * multiple aggregates, false if it is not.
4709 : : *
4710 : : * Returns true if not called as an aggregate support function.
4711 : : * This is intended as a conservative answer, ie "no you'd better not
4712 : : * scribble on your input". In particular, will return true if the
4713 : : * aggregate is being used as a window function, which is a scenario
4714 : : * in which changing the transition state is a bad idea. We might
4715 : : * want to refine the behavior for the window case in future.
4716 : : */
4717 : : bool
2882 tgl@sss.pgh.pa.us 4718 :CBC 123 : AggStateIsShared(FunctionCallInfo fcinfo)
4719 : : {
4720 [ + - + - ]: 123 : if (fcinfo->context && IsA(fcinfo->context, AggState))
4721 : : {
4722 : 123 : AggState *aggstate = (AggState *) fcinfo->context;
4723 : : AggStatePerAgg curperagg;
4724 : : AggStatePerTrans curpertrans;
4725 : :
4726 : : /* check curperagg (valid when in a final function) */
4727 : 123 : curperagg = aggstate->curperagg;
4728 : :
4729 [ - + ]: 123 : if (curperagg)
2882 tgl@sss.pgh.pa.us 4730 :UBC 0 : return aggstate->pertrans[curperagg->transno].aggshared;
4731 : :
4732 : : /* check curpertrans (valid when in a transition function) */
2882 tgl@sss.pgh.pa.us 4733 :CBC 123 : curpertrans = aggstate->curpertrans;
4734 : :
4735 [ + - ]: 123 : if (curpertrans)
4736 : 123 : return curpertrans->aggshared;
4737 : : }
2882 tgl@sss.pgh.pa.us 4738 :UBC 0 : return true;
4739 : : }
4740 : :
4741 : : /*
4742 : : * AggRegisterCallback - register a cleanup callback for an aggregate
4743 : : *
4744 : : * This is useful for aggs to register shutdown callbacks, which will ensure
4745 : : * that non-memory resources are freed. The callback will occur just before
4746 : : * the associated aggcontext (as returned by AggCheckCallContext) is reset,
4747 : : * either between groups or as a result of rescanning the query. The callback
4748 : : * will NOT be called on error paths. The typical use-case is for freeing of
4749 : : * tuplestores or tuplesorts maintained in aggcontext, or pins held by slots
4750 : : * created by the agg functions. (The callback will not be called until after
4751 : : * the result of the finalfn is no longer needed, so it's safe for the finalfn
4752 : : * to return data that will be freed by the callback.)
4753 : : *
4754 : : * As above, this is currently not useful for aggs called as window functions.
4755 : : */
4756 : : void
4083 tgl@sss.pgh.pa.us 4757 :CBC 330 : AggRegisterCallback(FunctionCallInfo fcinfo,
4758 : : ExprContextCallbackFunction func,
4759 : : Datum arg)
4760 : : {
4275 4761 [ + - + - ]: 330 : if (fcinfo->context && IsA(fcinfo->context, AggState))
4762 : : {
4763 : 330 : AggState *aggstate = (AggState *) fcinfo->context;
3085 rhodiumtoad@postgres 4764 : 330 : ExprContext *cxt = aggstate->curaggcontext;
4765 : :
3766 andres@anarazel.de 4766 : 330 : RegisterExprContextCallback(cxt, func, arg);
4767 : :
4083 tgl@sss.pgh.pa.us 4768 : 330 : return;
4769 : : }
4083 tgl@sss.pgh.pa.us 4770 [ # # ]:UBC 0 : elog(ERROR, "aggregate function cannot register a callback in this context");
4771 : : }
4772 : :
4773 : :
4774 : : /* ----------------------------------------------------------------
4775 : : * Parallel Query Support
4776 : : * ----------------------------------------------------------------
4777 : : */
4778 : :
4779 : : /* ----------------------------------------------------------------
4780 : : * ExecAggEstimate
4781 : : *
4782 : : * Estimate space required to propagate aggregate statistics.
4783 : : * ----------------------------------------------------------------
4784 : : */
4785 : : void
1905 drowley@postgresql.o 4786 :CBC 278 : ExecAggEstimate(AggState *node, ParallelContext *pcxt)
4787 : : {
4788 : : Size size;
4789 : :
4790 : : /* don't need this if not instrumenting or no workers */
4791 [ + + - + ]: 278 : if (!node->ss.ps.instrument || pcxt->nworkers == 0)
4792 : 227 : return;
4793 : :
4794 : 51 : size = mul_size(pcxt->nworkers, sizeof(AggregateInstrumentation));
4795 : 51 : size = add_size(size, offsetof(SharedAggInfo, sinstrument));
4796 : 51 : shm_toc_estimate_chunk(&pcxt->estimator, size);
4797 : 51 : shm_toc_estimate_keys(&pcxt->estimator, 1);
4798 : : }
4799 : :
4800 : : /* ----------------------------------------------------------------
4801 : : * ExecAggInitializeDSM
4802 : : *
4803 : : * Initialize DSM space for aggregate statistics.
4804 : : * ----------------------------------------------------------------
4805 : : */
4806 : : void
4807 : 278 : ExecAggInitializeDSM(AggState *node, ParallelContext *pcxt)
4808 : : {
4809 : : Size size;
4810 : :
4811 : : /* don't need this if not instrumenting or no workers */
4812 [ + + - + ]: 278 : if (!node->ss.ps.instrument || pcxt->nworkers == 0)
4813 : 227 : return;
4814 : :
4815 : 51 : size = offsetof(SharedAggInfo, sinstrument)
4816 : 51 : + pcxt->nworkers * sizeof(AggregateInstrumentation);
4817 : 51 : node->shared_info = shm_toc_allocate(pcxt->toc, size);
4818 : : /* ensure any unfilled slots will contain zeroes */
4819 : 51 : memset(node->shared_info, 0, size);
4820 : 51 : node->shared_info->num_workers = pcxt->nworkers;
4821 : 51 : shm_toc_insert(pcxt->toc, node->ss.ps.plan->plan_node_id,
4822 : 51 : node->shared_info);
4823 : : }
4824 : :
4825 : : /* ----------------------------------------------------------------
4826 : : * ExecAggInitializeWorker
4827 : : *
4828 : : * Attach worker to DSM space for aggregate statistics.
4829 : : * ----------------------------------------------------------------
4830 : : */
4831 : : void
4832 : 776 : ExecAggInitializeWorker(AggState *node, ParallelWorkerContext *pwcxt)
4833 : : {
4834 : 776 : node->shared_info =
4835 : 776 : shm_toc_lookup(pwcxt->toc, node->ss.ps.plan->plan_node_id, true);
4836 : 776 : }
4837 : :
4838 : : /* ----------------------------------------------------------------
4839 : : * ExecAggRetrieveInstrumentation
4840 : : *
4841 : : * Transfer aggregate statistics from DSM to private memory.
4842 : : * ----------------------------------------------------------------
4843 : : */
4844 : : void
4845 : 51 : ExecAggRetrieveInstrumentation(AggState *node)
4846 : : {
4847 : : Size size;
4848 : : SharedAggInfo *si;
4849 : :
4850 [ - + ]: 51 : if (node->shared_info == NULL)
1905 drowley@postgresql.o 4851 :UBC 0 : return;
4852 : :
1905 drowley@postgresql.o 4853 :CBC 51 : size = offsetof(SharedAggInfo, sinstrument)
4854 : 51 : + node->shared_info->num_workers * sizeof(AggregateInstrumentation);
4855 : 51 : si = palloc(size);
4856 : 51 : memcpy(si, node->shared_info, size);
4857 : 51 : node->shared_info = si;
4858 : : }
|
