| 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 |
| 457 |
select_current_set(AggState *aggstate, int setno, bool is_hash) |
| 458 |
{ |
| 459 |
/* |
| 460 |
* When changing this, also adapt ExecAggPlainTransByVal() and |
| 461 |
* ExecAggPlainTransByRef(). |
| 462 |
*/ |
| 463 |
if (is_hash) |
| 464 |
aggstate->curaggcontext = aggstate->hashcontext; |
| 465 |
else |
| 466 |
aggstate->curaggcontext = aggstate->aggcontexts[setno]; |
| 467 |
|
| 468 |
aggstate->current_set = setno; |
| 469 |
} |
| 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 |
| 479 |
initialize_phase(AggState *aggstate, int newphase) |
| 480 |
{ |
| 481 |
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 |
*/ |
| 487 |
if (aggstate->sort_in) |
| 488 |
{ |
| 489 |
tuplesort_end(aggstate->sort_in); |
| 490 |
aggstate->sort_in = NULL; |
| 491 |
} |
| 492 |
|
| 493 |
if (newphase <= 1) |
| 494 |
{ |
| 495 |
/* |
| 496 |
* Discard any existing output tuplesort. |
| 497 |
*/ |
| 498 |
if (aggstate->sort_out) |
| 499 |
{ |
| 500 |
tuplesort_end(aggstate->sort_out); |
| 501 |
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 |
aggstate->sort_in = aggstate->sort_out; |
| 511 |
aggstate->sort_out = NULL; |
| 512 |
Assert(aggstate->sort_in); |
| 513 |
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 |
*/ |
| 520 |
if (newphase > 0 && newphase < aggstate->numphases - 1) |
| 521 |
{ |
| 522 |
Sort *sortnode = aggstate->phases[newphase + 1].sortnode; |
| 523 |
PlanState *outerNode = outerPlanState(aggstate); |
| 524 |
TupleDesc tupDesc = ExecGetResultType(outerNode); |
| 525 |
|
| 526 |
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 |
aggstate->current_phase = newphase; |
| 537 |
aggstate->phase = &aggstate->phases[newphase]; |
| 538 |
} |
| 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 |
fetch_input_tuple(AggState *aggstate) |
| 550 |
{ |
| 551 |
TupleTableSlot *slot; |
| 552 |
|
| 553 |
if (aggstate->sort_in) |
| 554 |
{ |
| 555 |
/* make sure we check for interrupts in either path through here */ |
| 556 |
CHECK_FOR_INTERRUPTS(); |
| 557 |
if (!tuplesort_gettupleslot(aggstate->sort_in, true, false, |
| 558 |
aggstate->sort_slot, NULL)) |
| 559 |
return NULL; |
| 560 |
slot = aggstate->sort_slot; |
| 561 |
} |
| 562 |
else |
| 563 |
slot = ExecProcNode(outerPlanState(aggstate)); |
| 564 |
|
| 565 |
if (!TupIsNull(slot) && aggstate->sort_out) |
| 566 |
tuplesort_puttupleslot(aggstate->sort_out, slot); |
| 567 |
|
| 568 |
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 |
| 580 |
initialize_aggregate(AggState *aggstate, AggStatePerTrans pertrans, |
| 581 |
AggStatePerGroup pergroupstate) |
| 582 |
{ |
| 583 |
/* |
| 584 |
* Start a fresh sort operation for each DISTINCT/ORDER BY aggregate. |
| 585 |
*/ |
| 586 |
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 |
*/ |
| 592 |
if (pertrans->sortstates[aggstate->current_set]) |
| 593 |
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 |
*/ |
| 601 |
if (pertrans->numInputs == 1) |
| 602 |
{ |
| 603 |
Form_pg_attribute attr = TupleDescAttr(pertrans->sortdesc, 0); |
| 604 |
|
| 605 |
pertrans->sortstates[aggstate->current_set] = |
| 606 |
tuplesort_begin_datum(attr->atttypid, |
| 607 |
pertrans->sortOperators[0], |
| 608 |
pertrans->sortCollations[0], |
| 609 |
pertrans->sortNullsFirst[0], |
| 610 |
work_mem, NULL, TUPLESORT_NONE); |
| 611 |
} |
| 612 |
else |
| 613 |
pertrans->sortstates[aggstate->current_set] = |
| 614 |
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 |
*/ |
| 629 |
if (pertrans->initValueIsNull) |
| 630 |
pergroupstate->transValue = pertrans->initValue; |
| 631 |
else |
| 632 |
{ |
| 633 |
MemoryContext oldContext; |
| 634 |
|
| 635 |
oldContext = MemoryContextSwitchTo(aggstate->curaggcontext->ecxt_per_tuple_memory); |
| 636 |
pergroupstate->transValue = datumCopy(pertrans->initValue, |
| 637 |
pertrans->transtypeByVal, |
| 638 |
pertrans->transtypeLen); |
| 639 |
MemoryContextSwitchTo(oldContext); |
| 640 |
} |
| 641 |
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 |
pergroupstate->noTransValue = pertrans->initValueIsNull; |
| 651 |
} |
| 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 |
initialize_aggregates(AggState *aggstate, |
| 668 |
AggStatePerGroup *pergroups, |
| 669 |
int numReset) |
| 670 |
{ |
| 671 |
int transno; |
| 672 |
int numGroupingSets = Max(aggstate->phase->numsets, 1); |
| 673 |
int setno = 0; |
| 674 |
int numTrans = aggstate->numtrans; |
| 675 |
AggStatePerTrans transstates = aggstate->pertrans; |
| 676 |
|
| 677 |
if (numReset == 0) |
| 678 |
numReset = numGroupingSets; |
| 679 |
|
| 680 |
for (setno = 0; setno < numReset; setno++) |
| 681 |
{ |
| 682 |
AggStatePerGroup pergroup = pergroups[setno]; |
| 683 |
|
| 684 |
select_current_set(aggstate, setno, false); |
| 685 |
|
| 686 |
for (transno = 0; transno < numTrans; transno++) |
| 687 |
{ |
| 688 |
AggStatePerTrans pertrans = &transstates[transno]; |
| 689 |
AggStatePerGroup pergroupstate = &pergroup[transno]; |
| 690 |
|
| 691 |
initialize_aggregate(aggstate, pertrans, pergroupstate); |
| 692 |
} |
| 693 |
} |
| 694 |
} |
| 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 |
| 708 |
advance_transition_function(AggState *aggstate, |
| 709 |
AggStatePerTrans pertrans, |
| 710 |
AggStatePerGroup pergroupstate) |
| 711 |
{ |
| 712 |
FunctionCallInfo fcinfo = pertrans->transfn_fcinfo; |
| 713 |
MemoryContext oldContext; |
| 714 |
Datum newVal; |
| 715 |
|
| 716 |
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 |
int numTransInputs = pertrans->numTransInputs; |
| 723 |
int i; |
| 724 |
|
| 725 |
for (i = 1; i <= numTransInputs; i++) |
| 726 |
{ |
| 727 |
if (fcinfo->args[i].isnull) |
| 728 |
return; |
| 729 |
} |
| 730 |
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 |
*/ |
| 741 |
oldContext = MemoryContextSwitchTo(aggstate->curaggcontext->ecxt_per_tuple_memory); |
| 742 |
pergroupstate->transValue = datumCopy(fcinfo->args[1].value, |
| 743 |
pertrans->transtypeByVal, |
| 744 |
pertrans->transtypeLen); |
| 745 |
pergroupstate->transValueIsNull = false; |
| 746 |
pergroupstate->noTransValue = false; |
| 747 |
MemoryContextSwitchTo(oldContext); |
| 748 |
return; |
| 749 |
} |
| 750 |
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 |
*/ |
| 758 |
return; |
| 759 |
} |
| 760 |
} |
| 761 |
|
| 762 |
/* We run the transition functions in per-input-tuple memory context */ |
| 763 |
oldContext = MemoryContextSwitchTo(aggstate->tmpcontext->ecxt_per_tuple_memory); |
| 764 |
|
| 765 |
/* set up aggstate->curpertrans for AggGetAggref() */ |
| 766 |
aggstate->curpertrans = pertrans; |
| 767 |
|
| 768 |
/* |
| 769 |
* OK to call the transition function |
| 770 |
*/ |
| 771 |
fcinfo->args[0].value = pergroupstate->transValue; |
| 772 |
fcinfo->args[0].isnull = pergroupstate->transValueIsNull; |
| 773 |
fcinfo->isnull = false; /* just in case transfn doesn't set it */ |
| 774 |
|
| 775 |
newVal = FunctionCallInvoke(fcinfo); |
| 776 |
|
| 777 |
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 |
if (!pertrans->transtypeByVal && |
| 794 |
DatumGetPointer(newVal) != DatumGetPointer(pergroupstate->transValue)) |
| 795 |
newVal = ExecAggCopyTransValue(aggstate, pertrans, |
| 796 |
newVal, fcinfo->isnull, |
| 797 |
pergroupstate->transValue, |
| 798 |
pergroupstate->transValueIsNull); |
| 799 |
|
| 800 |
pergroupstate->transValue = newVal; |
| 801 |
pergroupstate->transValueIsNull = fcinfo->isnull; |
| 802 |
|
| 803 |
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 |
| 818 |
advance_aggregates(AggState *aggstate) |
| 819 |
{ |
| 820 |
ExecEvalExprNoReturnSwitchContext(aggstate->phase->evaltrans, |
| 821 |
aggstate->tmpcontext); |
| 822 |
} |
| 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 |
| 847 |
process_ordered_aggregate_single(AggState *aggstate, |
| 848 |
AggStatePerTrans pertrans, |
| 849 |
AggStatePerGroup pergroupstate) |
| 850 |
{ |
| 851 |
Datum oldVal = (Datum) 0; |
| 852 |
bool oldIsNull = true; |
| 853 |
bool haveOldVal = false; |
| 854 |
MemoryContext workcontext = aggstate->tmpcontext->ecxt_per_tuple_memory; |
| 855 |
MemoryContext oldContext; |
| 856 |
bool isDistinct = (pertrans->numDistinctCols > 0); |
| 857 |
Datum newAbbrevVal = (Datum) 0; |
| 858 |
Datum oldAbbrevVal = (Datum) 0; |
| 859 |
FunctionCallInfo fcinfo = pertrans->transfn_fcinfo; |
| 860 |
Datum *newVal; |
| 861 |
bool *isNull; |
| 862 |
|
| 863 |
Assert(pertrans->numDistinctCols < 2); |
| 864 |
|
| 865 |
tuplesort_performsort(pertrans->sortstates[aggstate->current_set]); |
| 866 |
|
| 867 |
/* Load the column into argument 1 (arg 0 will be transition value) */ |
| 868 |
newVal = &fcinfo->args[1].value; |
| 869 |
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 |
|
| 877 |
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 |
*/ |
| 884 |
MemoryContextReset(workcontext); |
| 885 |
oldContext = MemoryContextSwitchTo(workcontext); |
| 886 |
|
| 887 |
/* |
| 888 |
* If DISTINCT mode, and not distinct from prior, skip it. |
| 889 |
*/ |
| 890 |
if (isDistinct && |
| 891 |
haveOldVal && |
| 892 |
((oldIsNull && *isNull) || |
| 893 |
(!oldIsNull && !*isNull && |
| 894 |
oldAbbrevVal == newAbbrevVal && |
| 895 |
DatumGetBool(FunctionCall2Coll(&pertrans->equalfnOne, |
| 896 |
pertrans->aggCollation, |
| 897 |
oldVal, *newVal))))) |
| 898 |
{ |
| 899 |
MemoryContextSwitchTo(oldContext); |
| 900 |
continue; |
| 901 |
} |
| 902 |
else |
| 903 |
{ |
| 904 |
advance_transition_function(aggstate, pertrans, pergroupstate); |
| 905 |
|
| 906 |
MemoryContextSwitchTo(oldContext); |
| 907 |
|
| 908 |
/* |
| 909 |
* Forget the old value, if any, and remember the new one for |
| 910 |
* subsequent equality checks. |
| 911 |
*/ |
| 912 |
if (!pertrans->inputtypeByVal) |
| 913 |
{ |
| 914 |
if (!oldIsNull) |
| 915 |
pfree(DatumGetPointer(oldVal)); |
| 916 |
if (!*isNull) |
| 917 |
oldVal = datumCopy(*newVal, pertrans->inputtypeByVal, |
| 918 |
pertrans->inputtypeLen); |
| 919 |
} |
| 920 |
else |
| 921 |
oldVal = *newVal; |
| 922 |
oldAbbrevVal = newAbbrevVal; |
| 923 |
oldIsNull = *isNull; |
| 924 |
haveOldVal = true; |
| 925 |
} |
| 926 |
} |
| 927 |
|
| 928 |
if (!oldIsNull && !pertrans->inputtypeByVal) |
| 929 |
pfree(DatumGetPointer(oldVal)); |
| 930 |
|
| 931 |
tuplesort_end(pertrans->sortstates[aggstate->current_set]); |
| 932 |
pertrans->sortstates[aggstate->current_set] = NULL; |
| 933 |
} |
| 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 |
| 948 |
process_ordered_aggregate_multi(AggState *aggstate, |
| 949 |
AggStatePerTrans pertrans, |
| 950 |
AggStatePerGroup pergroupstate) |
| 951 |
{ |
| 952 |
ExprContext *tmpcontext = aggstate->tmpcontext; |
| 953 |
FunctionCallInfo fcinfo = pertrans->transfn_fcinfo; |
| 954 |
TupleTableSlot *slot1 = pertrans->sortslot; |
| 955 |
TupleTableSlot *slot2 = pertrans->uniqslot; |
| 956 |
int numTransInputs = pertrans->numTransInputs; |
| 957 |
int numDistinctCols = pertrans->numDistinctCols; |
| 958 |
Datum newAbbrevVal = (Datum) 0; |
| 959 |
Datum oldAbbrevVal = (Datum) 0; |
| 960 |
bool haveOldValue = false; |
| 961 |
TupleTableSlot *save = aggstate->tmpcontext->ecxt_outertuple; |
| 962 |
int i; |
| 963 |
|
| 964 |
tuplesort_performsort(pertrans->sortstates[aggstate->current_set]); |
| 965 |
|
| 966 |
ExecClearTuple(slot1); |
| 967 |
if (slot2) |
| 968 |
ExecClearTuple(slot2); |
| 969 |
|
| 970 |
while (tuplesort_gettupleslot(pertrans->sortstates[aggstate->current_set], |
| 971 |
true, true, slot1, &newAbbrevVal)) |
| 972 |
{ |
| 973 |
CHECK_FOR_INTERRUPTS(); |
| 974 |
|
| 975 |
tmpcontext->ecxt_outertuple = slot1; |
| 976 |
tmpcontext->ecxt_innertuple = slot2; |
| 977 |
|
| 978 |
if (numDistinctCols == 0 || |
| 979 |
!haveOldValue || |
| 980 |
newAbbrevVal != oldAbbrevVal || |
| 981 |
!ExecQual(pertrans->equalfnMulti, tmpcontext)) |
| 982 |
{ |
| 983 |
/* |
| 984 |
* Extract the first numTransInputs columns as datums to pass to |
| 985 |
* the transfn. |
| 986 |
*/ |
| 987 |
slot_getsomeattrs(slot1, numTransInputs); |
| 988 |
|
| 989 |
/* Load values into fcinfo */ |
| 990 |
/* Start from 1, since the 0th arg will be the transition value */ |
| 991 |
for (i = 0; i < numTransInputs; i++) |
| 992 |
{ |
| 993 |
fcinfo->args[i + 1].value = slot1->tts_values[i]; |
| 994 |
fcinfo->args[i + 1].isnull = slot1->tts_isnull[i]; |
| 995 |
} |
| 996 |
|
| 997 |
advance_transition_function(aggstate, pertrans, pergroupstate); |
| 998 |
|
| 999 |
if (numDistinctCols > 0) |
| 1000 |
{ |
| 1001 |
/* swap the slot pointers to retain the current tuple */ |
| 1002 |
TupleTableSlot *tmpslot = slot2; |
| 1003 |
|
| 1004 |
slot2 = slot1; |
| 1005 |
slot1 = tmpslot; |
| 1006 |
/* avoid ExecQual() calls by reusing abbreviated keys */ |
| 1007 |
oldAbbrevVal = newAbbrevVal; |
| 1008 |
haveOldValue = true; |
| 1009 |
} |
| 1010 |
} |
| 1011 |
|
| 1012 |
/* Reset context each time */ |
| 1013 |
ResetExprContext(tmpcontext); |
| 1014 |
|
| 1015 |
ExecClearTuple(slot1); |
| 1016 |
} |
| 1017 |
|
| 1018 |
if (slot2) |
| 1019 |
ExecClearTuple(slot2); |
| 1020 |
|
| 1021 |
tuplesort_end(pertrans->sortstates[aggstate->current_set]); |
| 1022 |
pertrans->sortstates[aggstate->current_set] = NULL; |
| 1023 |
|
| 1024 |
/* restore previous slot, potentially in use for grouping sets */ |
| 1025 |
tmpcontext->ecxt_outertuple = save; |
| 1026 |
} |
| 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 |
| 1045 |
finalize_aggregate(AggState *aggstate, |
| 1046 |
AggStatePerAgg peragg, |
| 1047 |
AggStatePerGroup pergroupstate, |
| 1048 |
Datum *resultVal, bool *resultIsNull) |
| 1049 |
{ |
| 1050 |
LOCAL_FCINFO(fcinfo, FUNC_MAX_ARGS); |
| 1051 |
bool anynull = false; |
| 1052 |
MemoryContext oldContext; |
| 1053 |
int i; |
| 1054 |
ListCell *lc; |
| 1055 |
AggStatePerTrans pertrans = &aggstate->pertrans[peragg->transno]; |
| 1056 |
|
| 1057 |
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 |
*/ |
| 1065 |
i = 1; |
| 1066 |
foreach(lc, peragg->aggdirectargs) |
| 1067 |
{ |
| 1068 |
ExprState *expr = (ExprState *) lfirst(lc); |
| 1069 |
|
| 1070 |
fcinfo->args[i].value = ExecEvalExpr(expr, |
| 1071 |
aggstate->ss.ps.ps_ExprContext, |
| 1072 |
&fcinfo->args[i].isnull); |
| 1073 |
anynull |= fcinfo->args[i].isnull; |
| 1074 |
i++; |
| 1075 |
} |
| 1076 |
|
| 1077 |
/* |
| 1078 |
* Apply the agg's finalfn if one is provided, else return transValue. |
| 1079 |
*/ |
| 1080 |
if (OidIsValid(peragg->finalfn_oid)) |
| 1081 |
{ |
| 1082 |
int numFinalArgs = peragg->numFinalArgs; |
| 1083 |
|
| 1084 |
/* set up aggstate->curperagg for AggGetAggref() */ |
| 1085 |
aggstate->curperagg = peragg; |
| 1086 |
|
| 1087 |
InitFunctionCallInfoData(*fcinfo, &peragg->finalfn, |
| 1088 |
numFinalArgs, |
| 1089 |
pertrans->aggCollation, |
| 1090 |
(Node *) aggstate, NULL); |
| 1091 |
|
| 1092 |
/* Fill in the transition state value */ |
| 1093 |
fcinfo->args[0].value = |
| 1094 |
MakeExpandedObjectReadOnly(pergroupstate->transValue, |
| 1095 |
pergroupstate->transValueIsNull, |
| 1096 |
pertrans->transtypeLen); |
| 1097 |
fcinfo->args[0].isnull = pergroupstate->transValueIsNull; |
| 1098 |
anynull |= pergroupstate->transValueIsNull; |
| 1099 |
|
| 1100 |
/* Fill any remaining argument positions with nulls */ |
| 1101 |
for (; i < numFinalArgs; i++) |
| 1102 |
{ |
| 1103 |
fcinfo->args[i].value = (Datum) 0; |
| 1104 |
fcinfo->args[i].isnull = true; |
| 1105 |
anynull = true; |
| 1106 |
} |
| 1107 |
|
| 1108 |
if (fcinfo->flinfo->fn_strict && anynull) |
| 1109 |
{ |
| 1110 |
/* don't call a strict function with NULL inputs */ |
| 1111 |
*resultVal = (Datum) 0; |
| 1112 |
*resultIsNull = true; |
| 1113 |
} |
| 1114 |
else |
| 1115 |
{ |
| 1116 |
Datum result; |
| 1117 |
|
| 1118 |
result = FunctionCallInvoke(fcinfo); |
| 1119 |
*resultIsNull = fcinfo->isnull; |
| 1120 |
*resultVal = MakeExpandedObjectReadOnly(result, |
| 1121 |
fcinfo->isnull, |
| 1122 |
peragg->resulttypeLen); |
| 1123 |
} |
| 1124 |
aggstate->curperagg = NULL; |
| 1125 |
} |
| 1126 |
else |
| 1127 |
{ |
| 1128 |
*resultVal = |
| 1129 |
MakeExpandedObjectReadOnly(pergroupstate->transValue, |
| 1130 |
pergroupstate->transValueIsNull, |
| 1131 |
pertrans->transtypeLen); |
| 1132 |
*resultIsNull = pergroupstate->transValueIsNull; |
| 1133 |
} |
| 1134 |
|
| 1135 |
MemoryContextSwitchTo(oldContext); |
| 1136 |
} |
| 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 |
| 1145 |
finalize_partialaggregate(AggState *aggstate, |
| 1146 |
AggStatePerAgg peragg, |
| 1147 |
AggStatePerGroup pergroupstate, |
| 1148 |
Datum *resultVal, bool *resultIsNull) |
| 1149 |
{ |
| 1150 |
AggStatePerTrans pertrans = &aggstate->pertrans[peragg->transno]; |
| 1151 |
MemoryContext oldContext; |
| 1152 |
|
| 1153 |
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 |
if (OidIsValid(pertrans->serialfn_oid)) |
| 1160 |
{ |
| 1161 |
/* Don't call a strict serialization function with NULL input. */ |
| 1162 |
if (pertrans->serialfn.fn_strict && pergroupstate->transValueIsNull) |
| 1163 |
{ |
| 1164 |
*resultVal = (Datum) 0; |
| 1165 |
*resultIsNull = true; |
| 1166 |
} |
| 1167 |
else |
| 1168 |
{ |
| 1169 |
FunctionCallInfo fcinfo = pertrans->serialfn_fcinfo; |
| 1170 |
Datum result; |
| 1171 |
|
| 1172 |
fcinfo->args[0].value = |
| 1173 |
MakeExpandedObjectReadOnly(pergroupstate->transValue, |
| 1174 |
pergroupstate->transValueIsNull, |
| 1175 |
pertrans->transtypeLen); |
| 1176 |
fcinfo->args[0].isnull = pergroupstate->transValueIsNull; |
| 1177 |
fcinfo->isnull = false; |
| 1178 |
|
| 1179 |
result = FunctionCallInvoke(fcinfo); |
| 1180 |
*resultIsNull = fcinfo->isnull; |
| 1181 |
*resultVal = MakeExpandedObjectReadOnly(result, |
| 1182 |
fcinfo->isnull, |
| 1183 |
peragg->resulttypeLen); |
| 1184 |
} |
| 1185 |
} |
| 1186 |
else |
| 1187 |
{ |
| 1188 |
*resultVal = |
| 1189 |
MakeExpandedObjectReadOnly(pergroupstate->transValue, |
| 1190 |
pergroupstate->transValueIsNull, |
| 1191 |
pertrans->transtypeLen); |
| 1192 |
*resultIsNull = pergroupstate->transValueIsNull; |
| 1193 |
} |
| 1194 |
|
| 1195 |
MemoryContextSwitchTo(oldContext); |
| 1196 |
} |
| 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 |
| 1203 |
prepare_hash_slot(AggStatePerHash perhash, |
| 1204 |
TupleTableSlot *inputslot, |
| 1205 |
TupleTableSlot *hashslot) |
| 1206 |
{ |
| 1207 |
int i; |
| 1208 |
|
| 1209 |
/* transfer just the needed columns into hashslot */ |
| 1210 |
slot_getsomeattrs(inputslot, perhash->largestGrpColIdx); |
| 1211 |
ExecClearTuple(hashslot); |
| 1212 |
|
| 1213 |
for (i = 0; i < perhash->numhashGrpCols; i++) |
| 1214 |
{ |
| 1215 |
int varNumber = perhash->hashGrpColIdxInput[i] - 1; |
| 1216 |
|
| 1217 |
hashslot->tts_values[i] = inputslot->tts_values[varNumber]; |
| 1218 |
hashslot->tts_isnull[i] = inputslot->tts_isnull[varNumber]; |
| 1219 |
} |
| 1220 |
ExecStoreVirtualTuple(hashslot); |
| 1221 |
} |
| 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 |
| 1248 |
prepare_projection_slot(AggState *aggstate, TupleTableSlot *slot, int currentSet) |
| 1249 |
{ |
| 1250 |
if (aggstate->phase->grouped_cols) |
| 1251 |
{ |
| 1252 |
Bitmapset *grouped_cols = aggstate->phase->grouped_cols[currentSet]; |
| 1253 |
|
| 1254 |
aggstate->grouped_cols = grouped_cols; |
| 1255 |
|
| 1256 |
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 |
*/ |
| 1263 |
ExecStoreAllNullTuple(slot); |
| 1264 |
} |
| 1265 |
else if (aggstate->all_grouped_cols) |
| 1266 |
{ |
| 1267 |
ListCell *lc; |
| 1268 |
|
| 1269 |
/* all_grouped_cols is arranged in desc order */ |
| 1270 |
slot_getsomeattrs(slot, linitial_int(aggstate->all_grouped_cols)); |
| 1271 |
|
| 1272 |
foreach(lc, aggstate->all_grouped_cols) |
| 1273 |
{ |
| 1274 |
int attnum = lfirst_int(lc); |
| 1275 |
|
| 1276 |
if (!bms_is_member(attnum, grouped_cols)) |
| 1277 |
slot->tts_isnull[attnum - 1] = true; |
| 1278 |
} |
| 1279 |
} |
| 1280 |
} |
| 1281 |
} |
| 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 |
finalize_aggregates(AggState *aggstate, |
| 1294 |
AggStatePerAgg peraggs, |
| 1295 |
AggStatePerGroup pergroup) |
| 1296 |
{ |
| 1297 |
ExprContext *econtext = aggstate->ss.ps.ps_ExprContext; |
| 1298 |
Datum *aggvalues = econtext->ecxt_aggvalues; |
| 1299 |
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 |
*/ |
| 1306 |
for (int transno = 0; transno < aggstate->numtrans; transno++) |
| 1307 |
{ |
| 1308 |
AggStatePerTrans pertrans = &aggstate->pertrans[transno]; |
| 1309 |
AggStatePerGroup pergroupstate; |
| 1310 |
|
| 1311 |
pergroupstate = &pergroup[transno]; |
| 1312 |
|
| 1313 |
if (pertrans->aggsortrequired) |
| 1314 |
{ |
| 1315 |
Assert(aggstate->aggstrategy != AGG_HASHED && |
| 1316 |
aggstate->aggstrategy != AGG_MIXED); |
| 1317 |
|
| 1318 |
if (pertrans->numInputs == 1) |
| 1319 |
process_ordered_aggregate_single(aggstate, |
| 1320 |
pertrans, |
| 1321 |
pergroupstate); |
| 1322 |
else |
| 1323 |
process_ordered_aggregate_multi(aggstate, |
| 1324 |
pertrans, |
| 1325 |
pergroupstate); |
| 1326 |
} |
| 1327 |
else if (pertrans->numDistinctCols > 0 && pertrans->haslast) |
| 1328 |
{ |
| 1329 |
pertrans->haslast = false; |
| 1330 |
|
| 1331 |
if (pertrans->numDistinctCols == 1) |
| 1332 |
{ |
| 1333 |
if (!pertrans->inputtypeByVal && !pertrans->lastisnull) |
| 1334 |
pfree(DatumGetPointer(pertrans->lastdatum)); |
| 1335 |
|
| 1336 |
pertrans->lastisnull = false; |
| 1337 |
pertrans->lastdatum = (Datum) 0; |
| 1338 |
} |
| 1339 |
else |
| 1340 |
ExecClearTuple(pertrans->uniqslot); |
| 1341 |
} |
| 1342 |
} |
| 1343 |
|
| 1344 |
/* |
| 1345 |
* Run the final functions. |
| 1346 |
*/ |
| 1347 |
for (aggno = 0; aggno < aggstate->numaggs; aggno++) |
| 1348 |
{ |
| 1349 |
AggStatePerAgg peragg = &peraggs[aggno]; |
| 1350 |
int transno = peragg->transno; |
| 1351 |
AggStatePerGroup pergroupstate; |
| 1352 |
|
| 1353 |
pergroupstate = &pergroup[transno]; |
| 1354 |
|
| 1355 |
if (DO_AGGSPLIT_SKIPFINAL(aggstate->aggsplit)) |
| 1356 |
finalize_partialaggregate(aggstate, peragg, pergroupstate, |
| 1357 |
&aggvalues[aggno], &aggnulls[aggno]); |
| 1358 |
else |
| 1359 |
finalize_aggregate(aggstate, peragg, pergroupstate, |
| 1360 |
&aggvalues[aggno], &aggnulls[aggno]); |
| 1361 |
} |
| 1362 |
} |
| 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 |
project_aggregates(AggState *aggstate) |
| 1371 |
{ |
| 1372 |
ExprContext *econtext = aggstate->ss.ps.ps_ExprContext; |
| 1373 |
|
| 1374 |
/* |
| 1375 |
* Check the qual (HAVING clause); if the group does not match, ignore it. |
| 1376 |
*/ |
| 1377 |
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 |
*/ |
| 1383 |
return ExecProject(aggstate->ss.ps.ps_ProjInfo); |
| 1384 |
} |
| 1385 |
else |
| 1386 |
InstrCountFiltered1(aggstate, 1); |
| 1387 |
|
| 1388 |
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 |
| 1396 |
find_cols(AggState *aggstate, Bitmapset **aggregated, Bitmapset **unaggregated) |
| 1397 |
{ |
| 1398 |
Agg *agg = (Agg *) aggstate->ss.ps.plan; |
| 1399 |
FindColsContext context; |
| 1400 |
|
| 1401 |
context.is_aggref = false; |
| 1402 |
context.aggregated = NULL; |
| 1403 |
context.unaggregated = NULL; |
| 1404 |
|
| 1405 |
/* Examine tlist and quals */ |
| 1406 |
(void) find_cols_walker((Node *) agg->plan.targetlist, &context); |
| 1407 |
(void) find_cols_walker((Node *) agg->plan.qual, &context); |
| 1408 |
|
| 1409 |
/* In some cases, grouping columns will not appear in the tlist */ |
| 1410 |
for (int i = 0; i < agg->numCols; i++) |
| 1411 |
context.unaggregated = bms_add_member(context.unaggregated, |
| 1412 |
agg->grpColIdx[i]); |
| 1413 |
|
| 1414 |
*aggregated = context.aggregated; |
| 1415 |
*unaggregated = context.unaggregated; |
| 1416 |
} |
| 1417 |
|
| 1418 |
static bool |
| 1419 |
find_cols_walker(Node *node, FindColsContext *context) |
| 1420 |
{ |
| 1421 |
if (node == NULL) |
| 1422 |
return false; |
| 1423 |
if (IsA(node, Var)) |
| 1424 |
{ |
| 1425 |
Var *var = (Var *) node; |
| 1426 |
|
| 1427 |
/* setrefs.c should have set the varno to OUTER_VAR */ |
| 1428 |
Assert(var->varno == OUTER_VAR); |
| 1429 |
Assert(var->varlevelsup == 0); |
| 1430 |
if (context->is_aggref) |
| 1431 |
context->aggregated = bms_add_member(context->aggregated, |
| 1432 |
var->varattno); |
| 1433 |
else |
| 1434 |
context->unaggregated = bms_add_member(context->unaggregated, |
| 1435 |
var->varattno); |
| 1436 |
return false; |
| 1437 |
} |
| 1438 |
if (IsA(node, Aggref)) |
| 1439 |
{ |
| 1440 |
Assert(!context->is_aggref); |
| 1441 |
context->is_aggref = true; |
| 1442 |
expression_tree_walker(node, find_cols_walker, context); |
| 1443 |
context->is_aggref = false; |
| 1444 |
return false; |
| 1445 |
} |
| 1446 |
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 |
| 1466 |
build_hash_tables(AggState *aggstate) |
| 1467 |
{ |
| 1468 |
int setno; |
| 1469 |
|
| 1470 |
for (setno = 0; setno < aggstate->num_hashes; ++setno) |
| 1471 |
{ |
| 1472 |
AggStatePerHash perhash = &aggstate->perhash[setno]; |
| 1473 |
long nbuckets; |
| 1474 |
Size memory; |
| 1475 |
|
| 1476 |
if (perhash->hashtable != NULL) |
| 1477 |
{ |
| 1478 |
ResetTupleHashTable(perhash->hashtable); |
| 1479 |
continue; |
| 1480 |
} |
| 1481 |
|
| 1482 |
Assert(perhash->aggnode->numGroups > 0); |
| 1483 |
|
| 1484 |
memory = aggstate->hash_mem_limit / aggstate->num_hashes; |
| 1485 |
|
| 1486 |
/* choose reasonable number of buckets per hashtable */ |
| 1487 |
nbuckets = hash_choose_num_buckets(aggstate->hashentrysize, |
| 1488 |
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 |
|
| 1499 |
build_hash_table(aggstate, setno, nbuckets); |
| 1500 |
} |
| 1501 |
|
| 1502 |
aggstate->hash_ngroups_current = 0; |
| 1503 |
} |
| 1504 |
|
| 1505 |
/* |
| 1506 |
* Build a single hashtable for this grouping set. |
| 1507 |
*/ |
| 1508 |
static void |
| 1509 |
build_hash_table(AggState *aggstate, int setno, long nbuckets) |
| 1510 |
{ |
| 1511 |
AggStatePerHash perhash = &aggstate->perhash[setno]; |
| 1512 |
MemoryContext metacxt = aggstate->hash_metacxt; |
| 1513 |
MemoryContext tablecxt = aggstate->hash_tablecxt; |
| 1514 |
MemoryContext tmpcxt = aggstate->tmpcontext->ecxt_per_tuple_memory; |
| 1515 |
Size additionalsize; |
| 1516 |
|
| 1517 |
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 |
additionalsize = aggstate->numtrans * sizeof(AggStatePerGroupData); |
| 1527 |
|
| 1528 |
perhash->hashtable = BuildTupleHashTable(&aggstate->ss.ps, |
| 1529 |
perhash->hashslot->tts_tupleDescriptor, |
| 1530 |
perhash->hashslot->tts_ops, |
| 1531 |
perhash->numCols, |
| 1532 |
perhash->hashGrpColIdxHash, |
| 1533 |
perhash->eqfuncoids, |
| 1534 |
perhash->hashfunctions, |
| 1535 |
perhash->aggnode->grpCollations, |
| 1536 |
nbuckets, |
| 1537 |
additionalsize, |
| 1538 |
metacxt, |
| 1539 |
tablecxt, |
| 1540 |
tmpcxt, |
| 1541 |
DO_AGGSPLIT_SKIPFINAL(aggstate->aggsplit)); |
| 1542 |
} |
| 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 |
| 1570 |
find_hash_columns(AggState *aggstate) |
| 1571 |
{ |
| 1572 |
Bitmapset *base_colnos; |
| 1573 |
Bitmapset *aggregated_colnos; |
| 1574 |
TupleDesc scanDesc = aggstate->ss.ss_ScanTupleSlot->tts_tupleDescriptor; |
| 1575 |
List *outerTlist = outerPlanState(aggstate)->plan->targetlist; |
| 1576 |
int numHashes = aggstate->num_hashes; |
| 1577 |
EState *estate = aggstate->ss.ps.state; |
| 1578 |
int j; |
| 1579 |
|
| 1580 |
/* Find Vars that will be needed in tlist and qual */ |
| 1581 |
find_cols(aggstate, &aggregated_colnos, &base_colnos); |
| 1582 |
aggstate->colnos_needed = bms_union(base_colnos, aggregated_colnos); |
| 1583 |
aggstate->max_colno_needed = 0; |
| 1584 |
aggstate->all_cols_needed = true; |
| 1585 |
|
| 1586 |
for (int i = 0; i < scanDesc->natts; i++) |
| 1587 |
{ |
| 1588 |
int colno = i + 1; |
| 1589 |
|
| 1590 |
if (bms_is_member(colno, aggstate->colnos_needed)) |
| 1591 |
aggstate->max_colno_needed = colno; |
| 1592 |
else |
| 1593 |
aggstate->all_cols_needed = false; |
| 1594 |
} |
| 1595 |
|
| 1596 |
for (j = 0; j < numHashes; ++j) |
| 1597 |
{ |
| 1598 |
AggStatePerHash perhash = &aggstate->perhash[j]; |
| 1599 |
Bitmapset *colnos = bms_copy(base_colnos); |
| 1600 |
AttrNumber *grpColIdx = perhash->aggnode->grpColIdx; |
| 1601 |
List *hashTlist = NIL; |
| 1602 |
TupleDesc hashDesc; |
| 1603 |
int maxCols; |
| 1604 |
int i; |
| 1605 |
|
| 1606 |
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 |
if (aggstate->phases[0].grouped_cols) |
| 1616 |
{ |
| 1617 |
Bitmapset *grouped_cols = aggstate->phases[0].grouped_cols[j]; |
| 1618 |
ListCell *lc; |
| 1619 |
|
| 1620 |
foreach(lc, aggstate->all_grouped_cols) |
| 1621 |
{ |
| 1622 |
int attnum = lfirst_int(lc); |
| 1623 |
|
| 1624 |
if (!bms_is_member(attnum, grouped_cols)) |
| 1625 |
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 |
*/ |
| 1635 |
maxCols = bms_num_members(colnos) + perhash->numCols; |
| 1636 |
|
| 1637 |
perhash->hashGrpColIdxInput = |
| 1638 |
palloc(maxCols * sizeof(AttrNumber)); |
| 1639 |
perhash->hashGrpColIdxHash = |
| 1640 |
palloc(perhash->numCols * sizeof(AttrNumber)); |
| 1641 |
|
| 1642 |
/* Add all the grouping columns to colnos */ |
| 1643 |
for (i = 0; i < perhash->numCols; i++) |
| 1644 |
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 |
*/ |
| 1653 |
for (i = 0; i < perhash->numCols; i++) |
| 1654 |
{ |
| 1655 |
perhash->hashGrpColIdxInput[i] = grpColIdx[i]; |
| 1656 |
perhash->hashGrpColIdxHash[i] = i + 1; |
| 1657 |
perhash->numhashGrpCols++; |
| 1658 |
/* delete already mapped columns */ |
| 1659 |
colnos = bms_del_member(colnos, grpColIdx[i]); |
| 1660 |
} |
| 1661 |
|
| 1662 |
/* and add the remaining columns */ |
| 1663 |
i = -1; |
| 1664 |
while ((i = bms_next_member(colnos, i)) >= 0) |
| 1665 |
{ |
| 1666 |
perhash->hashGrpColIdxInput[perhash->numhashGrpCols] = i; |
| 1667 |
perhash->numhashGrpCols++; |
| 1668 |
} |
| 1669 |
|
| 1670 |
/* and build a tuple descriptor for the hashtable */ |
| 1671 |
for (i = 0; i < perhash->numhashGrpCols; i++) |
| 1672 |
{ |
| 1673 |
int varNumber = perhash->hashGrpColIdxInput[i] - 1; |
| 1674 |
|
| 1675 |
hashTlist = lappend(hashTlist, list_nth(outerTlist, varNumber)); |
| 1676 |
perhash->largestGrpColIdx = |
| 1677 |
Max(varNumber + 1, perhash->largestGrpColIdx); |
| 1678 |
} |
| 1679 |
|
| 1680 |
hashDesc = ExecTypeFromTL(hashTlist); |
| 1681 |
|
| 1682 |
execTuplesHashPrepare(perhash->numCols, |
| 1683 |
perhash->aggnode->grpOperators, |
| 1684 |
&perhash->eqfuncoids, |
| 1685 |
&perhash->hashfunctions); |
| 1686 |
perhash->hashslot = |
| 1687 |
ExecAllocTableSlot(&estate->es_tupleTable, hashDesc, |
| 1688 |
&TTSOpsMinimalTuple); |
| 1689 |
|
| 1690 |
list_free(hashTlist); |
| 1691 |
bms_free(colnos); |
| 1692 |
} |
| 1693 |
|
| 1694 |
bms_free(base_colnos); |
| 1695 |
} |
| 1696 |
|
| 1697 |
/* |
| 1698 |
* Estimate per-hash-table-entry overhead. |
| 1699 |
*/ |
| 1700 |
Size |
| 1701 |
hash_agg_entry_size(int numTrans, Size tupleWidth, Size transitionSpace) |
| 1702 |
{ |
| 1703 |
Size tupleChunkSize; |
| 1704 |
Size pergroupChunkSize; |
| 1705 |
Size transitionChunkSize; |
| 1706 |
Size tupleSize = (MAXALIGN(SizeofMinimalTupleHeader) + |
| 1707 |
tupleWidth); |
| 1708 |
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 |
*/ |
| 1714 |
tupleChunkSize = MAXALIGN(tupleSize); |
| 1715 |
pergroupChunkSize = pergroupSize; |
| 1716 |
|
| 1717 |
/* |
| 1718 |
* Transition values use AllocSet, which has a chunk header and also uses |
| 1719 |
* power-of-two allocations. |
| 1720 |
*/ |
| 1721 |
if (transitionSpace > 0) |
| 1722 |
transitionChunkSize = CHUNKHDRSZ + pg_nextpower2_size_t(transitionSpace); |
| 1723 |
else |
| 1724 |
transitionChunkSize = 0; |
| 1725 |
|
| 1726 |
return |
| 1727 |
TupleHashEntrySize() + |
| 1728 |
tupleChunkSize + |
| 1729 |
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 |
| 1752 |
hashagg_recompile_expressions(AggState *aggstate, bool minslot, bool nullcheck) |
| 1753 |
{ |
| 1754 |
AggStatePerPhase phase; |
| 1755 |
int i = minslot ? 1 : 0; |
| 1756 |
int j = nullcheck ? 1 : 0; |
| 1757 |
|
| 1758 |
Assert(aggstate->aggstrategy == AGG_HASHED || |
| 1759 |
aggstate->aggstrategy == AGG_MIXED); |
| 1760 |
|
| 1761 |
if (aggstate->aggstrategy == AGG_HASHED) |
| 1762 |
phase = &aggstate->phases[0]; |
| 1763 |
else /* AGG_MIXED */ |
| 1764 |
phase = &aggstate->phases[1]; |
| 1765 |
|
| 1766 |
if (phase->evaltrans_cache[i][j] == NULL) |
| 1767 |
{ |
| 1768 |
const TupleTableSlotOps *outerops = aggstate->ss.ps.outerops; |
| 1769 |
bool outerfixed = aggstate->ss.ps.outeropsfixed; |
| 1770 |
bool dohash = true; |
| 1771 |
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 |
if (aggstate->aggstrategy == AGG_MIXED && !minslot) |
| 1779 |
dosort = true; |
| 1780 |
|
| 1781 |
/* temporarily change the outerops while compiling the expression */ |
| 1782 |
if (minslot) |
| 1783 |
{ |
| 1784 |
aggstate->ss.ps.outerops = &TTSOpsMinimalTuple; |
| 1785 |
aggstate->ss.ps.outeropsfixed = true; |
| 1786 |
} |
| 1787 |
|
| 1788 |
phase->evaltrans_cache[i][j] = ExecBuildAggTrans(aggstate, phase, |
| 1789 |
dosort, dohash, |
| 1790 |
nullcheck); |
| 1791 |
|
| 1792 |
/* change back */ |
| 1793 |
aggstate->ss.ps.outerops = outerops; |
| 1794 |
aggstate->ss.ps.outeropsfixed = outerfixed; |
| 1795 |
} |
| 1796 |
|
| 1797 |
phase->evaltrans = phase->evaltrans_cache[i][j]; |
| 1798 |
} |
| 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 |
| 1809 |
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; |
| 1815 |
Size hash_mem_limit = get_hash_memory_limit(); |
| 1816 |
|
| 1817 |
/* if not expected to spill, use all of hash_mem */ |
| 1818 |
if (input_groups * hashentrysize <= hash_mem_limit) |
| 1819 |
{ |
| 1820 |
if (num_partitions != NULL) |
| 1821 |
*num_partitions = 0; |
| 1822 |
*mem_limit = hash_mem_limit; |
| 1823 |
*ngroups_limit = hash_mem_limit / hashentrysize; |
| 1824 |
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 |
npartitions = hash_choose_num_partitions(input_groups, |
| 1833 |
hashentrysize, |
| 1834 |
used_bits, |
| 1835 |
NULL); |
| 1836 |
if (num_partitions != NULL) |
| 1837 |
*num_partitions = npartitions; |
| 1838 |
|
| 1839 |
partition_mem = |
| 1840 |
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 |
*/ |
| 1848 |
if (hash_mem_limit > 4 * partition_mem) |
| 1849 |
*mem_limit = hash_mem_limit - partition_mem; |
| 1850 |
else |
| 1851 |
*mem_limit = hash_mem_limit * 0.75; |
| 1852 |
|
| 1853 |
if (*mem_limit > hashentrysize) |
| 1854 |
*ngroups_limit = *mem_limit / hashentrysize; |
| 1855 |
else |
| 1856 |
*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 |
| 1867 |
hash_agg_check_limits(AggState *aggstate) |
| 1868 |
{ |
| 1869 |
uint64 ngroups = aggstate->hash_ngroups_current; |
| 1870 |
Size meta_mem = MemoryContextMemAllocated(aggstate->hash_metacxt, |
| 1871 |
true); |
| 1872 |
Size entry_mem = MemoryContextMemAllocated(aggstate->hash_tablecxt, |
| 1873 |
true); |
| 1874 |
Size tval_mem = MemoryContextMemAllocated(aggstate->hashcontext->ecxt_per_tuple_memory, |
| 1875 |
true); |
| 1876 |
Size total_mem = meta_mem + entry_mem + tval_mem; |
| 1877 |
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 |
*/ |
| 1894 |
if (aggstate->hash_ngroups_current > 0 && |
| 1895 |
(total_mem > aggstate->hash_mem_limit || |
| 1896 |
ngroups > aggstate->hash_ngroups_limit)) |
| 1897 |
{ |
| 1898 |
do_spill = true; |
| 1899 |
} |
| 1900 |
|
| 1901 |
if (do_spill) |
| 1902 |
hash_agg_enter_spill_mode(aggstate); |
| 1903 |
} |
| 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 |
hash_agg_enter_spill_mode(AggState *aggstate) |
| 1912 |
{ |
| 1913 |
INJECTION_POINT("hash-aggregate-enter-spill-mode", NULL); |
| 1914 |
aggstate->hash_spill_mode = true; |
| 1915 |
hashagg_recompile_expressions(aggstate, aggstate->table_filled, true); |
| 1916 |
|
| 1917 |
if (!aggstate->hash_ever_spilled) |
| 1918 |
{ |
| 1919 |
Assert(aggstate->hash_tapeset == NULL); |
| 1920 |
Assert(aggstate->hash_spills == NULL); |
| 1921 |
|
| 1922 |
aggstate->hash_ever_spilled = true; |
| 1923 |
|
| 1924 |
aggstate->hash_tapeset = LogicalTapeSetCreate(true, NULL, -1); |
| 1925 |
|
| 1926 |
aggstate->hash_spills = palloc(sizeof(HashAggSpill) * aggstate->num_hashes); |
| 1927 |
|
| 1928 |
for (int setno = 0; setno < aggstate->num_hashes; setno++) |
| 1929 |
{ |
| 1930 |
AggStatePerHash perhash = &aggstate->perhash[setno]; |
| 1931 |
HashAggSpill *spill = &aggstate->hash_spills[setno]; |
| 1932 |
|
| 1933 |
hashagg_spill_init(spill, aggstate->hash_tapeset, 0, |
| 1934 |
perhash->aggnode->numGroups, |
| 1935 |
aggstate->hashentrysize); |
| 1936 |
} |
| 1937 |
} |
| 1938 |
} |
| 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 |
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 |
if (aggstate->aggstrategy != AGG_MIXED && |
| 1956 |
aggstate->aggstrategy != AGG_HASHED) |
| 1957 |
return; |
| 1958 |
|
| 1959 |
/* memory for the hash table itself */ |
| 1960 |
meta_mem = MemoryContextMemAllocated(aggstate->hash_metacxt, true); |
| 1961 |
|
| 1962 |
/* memory for hash entries */ |
| 1963 |
entry_mem = MemoryContextMemAllocated(aggstate->hash_tablecxt, true); |
| 1964 |
|
| 1965 |
/* memory for byref transition states */ |
| 1966 |
hashkey_mem = MemoryContextMemAllocated(aggstate->hashcontext->ecxt_per_tuple_memory, true); |
| 1967 |
|
| 1968 |
/* memory for read/write tape buffers, if spilled */ |
| 1969 |
buffer_mem = npartitions * HASHAGG_WRITE_BUFFER_SIZE; |
| 1970 |
if (from_tape) |
| 1971 |
buffer_mem += HASHAGG_READ_BUFFER_SIZE; |
| 1972 |
|
| 1973 |
/* update peak mem */ |
| 1974 |
total_mem = meta_mem + entry_mem + hashkey_mem + buffer_mem; |
| 1975 |
if (total_mem > aggstate->hash_mem_peak) |
| 1976 |
aggstate->hash_mem_peak = total_mem; |
| 1977 |
|
| 1978 |
/* update disk usage */ |
| 1979 |
if (aggstate->hash_tapeset != NULL) |
| 1980 |
{ |
| 1981 |
uint64 disk_used = LogicalTapeSetBlocks(aggstate->hash_tapeset) * (BLCKSZ / 1024); |
| 1982 |
|
| 1983 |
if (aggstate->hash_disk_used < disk_used) |
| 1984 |
aggstate->hash_disk_used = disk_used; |
| 1985 |
} |
| 1986 |
|
| 1987 |
/* update hashentrysize estimate based on contents */ |
| 1988 |
if (aggstate->hash_ngroups_current > 0) |
| 1989 |
{ |
| 1990 |
aggstate->hashentrysize = |
| 1991 |
TupleHashEntrySize() + |
| 1992 |
(hashkey_mem / (double) aggstate->hash_ngroups_current); |
| 1993 |
} |
| 1994 |
} |
| 1995 |
|
| 1996 |
/* |
| 1997 |
* Create memory contexts used for hash aggregation. |
| 1998 |
*/ |
| 1999 |
static void |
| 2000 |
hash_create_memory(AggState *aggstate) |
| 2001 |
{ |
| 2002 |
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 |
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 |
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 |
maxBlockSize = pg_prevpower2_size_t(work_mem * (Size) 1024 / 16); |
| 2039 |
|
| 2040 |
/* But no bigger than ALLOCSET_DEFAULT_MAXSIZE */ |
| 2041 |
maxBlockSize = Min(maxBlockSize, ALLOCSET_DEFAULT_MAXSIZE); |
| 2042 |
|
| 2043 |
/* and no smaller than ALLOCSET_DEFAULT_INITSIZE */ |
| 2044 |
maxBlockSize = Max(maxBlockSize, ALLOCSET_DEFAULT_INITSIZE); |
| 2045 |
|
| 2046 |
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 |
} |
| 2053 |
|
| 2054 |
/* |
| 2055 |
* Choose a reasonable number of buckets for the initial hash table size. |
| 2056 |
*/ |
| 2057 |
static long |
| 2058 |
hash_choose_num_buckets(double hashentrysize, long ngroups, Size memory) |
| 2059 |
{ |
| 2060 |
long max_nbuckets; |
| 2061 |
long nbuckets = ngroups; |
| 2062 |
|
| 2063 |
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 |
*/ |
| 2069 |
max_nbuckets >>= 1; |
| 2070 |
|
| 2071 |
if (nbuckets > max_nbuckets) |
| 2072 |
nbuckets = max_nbuckets; |
| 2073 |
|
| 2074 |
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 |
| 2083 |
hash_choose_num_partitions(double input_groups, double hashentrysize, |
| 2084 |
int used_bits, int *log2_npartitions) |
| 2085 |
{ |
| 2086 |
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 |
*/ |
| 2097 |
partition_limit = |
| 2098 |
(hash_mem_limit * 0.25 - HASHAGG_READ_BUFFER_SIZE) / |
| 2099 |
HASHAGG_WRITE_BUFFER_SIZE; |
| 2100 |
|
| 2101 |
mem_wanted = HASHAGG_PARTITION_FACTOR * input_groups * hashentrysize; |
| 2102 |
|
| 2103 |
/* make enough partitions so that each one is likely to fit in memory */ |
| 2104 |
dpartitions = 1 + (mem_wanted / hash_mem_limit); |
| 2105 |
|
| 2106 |
if (dpartitions > partition_limit) |
| 2107 |
dpartitions = partition_limit; |
| 2108 |
|
| 2109 |
if (dpartitions < HASHAGG_MIN_PARTITIONS) |
| 2110 |
dpartitions = HASHAGG_MIN_PARTITIONS; |
| 2111 |
if (dpartitions > HASHAGG_MAX_PARTITIONS) |
| 2112 |
dpartitions = HASHAGG_MAX_PARTITIONS; |
| 2113 |
|
| 2114 |
/* HASHAGG_MAX_PARTITIONS limit makes this safe */ |
| 2115 |
npartitions = (int) dpartitions; |
| 2116 |
|
| 2117 |
/* ceil(log2(npartitions)) */ |
| 2118 |
partition_bits = my_log2(npartitions); |
| 2119 |
|
| 2120 |
/* make sure that we don't exhaust the hash bits */ |
| 2121 |
if (partition_bits + used_bits >= 32) |
| 2122 |
partition_bits = 32 - used_bits; |
| 2123 |
|
| 2124 |
if (log2_npartitions != NULL) |
| 2125 |
*log2_npartitions = partition_bits; |
| 2126 |
|
| 2127 |
/* number of partitions will be a power of two */ |
| 2128 |
npartitions = 1 << partition_bits; |
| 2129 |
|
| 2130 |
return npartitions; |
| 2131 |
} |
| 2132 |
|
| 2133 |
/* |
| 2134 |
* Initialize a freshly-created TupleHashEntry. |
| 2135 |
*/ |
| 2136 |
static void |
| 2137 |
initialize_hash_entry(AggState *aggstate, TupleHashTable hashtable, |
| 2138 |
TupleHashEntry entry) |
| 2139 |
{ |
| 2140 |
AggStatePerGroup pergroup; |
| 2141 |
int transno; |
| 2142 |
|
| 2143 |
aggstate->hash_ngroups_current++; |
| 2144 |
hash_agg_check_limits(aggstate); |
| 2145 |
|
| 2146 |
/* no need to allocate or initialize per-group state */ |
| 2147 |
if (aggstate->numtrans == 0) |
| 2148 |
return; |
| 2149 |
|
| 2150 |
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 |
*/ |
| 2156 |
for (transno = 0; transno < aggstate->numtrans; transno++) |
| 2157 |
{ |
| 2158 |
AggStatePerTrans pertrans = &aggstate->pertrans[transno]; |
| 2159 |
AggStatePerGroup pergroupstate = &pergroup[transno]; |
| 2160 |
|
| 2161 |
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 |
| 2181 |
lookup_hash_entries(AggState *aggstate) |
| 2182 |
{ |
| 2183 |
AggStatePerGroup *pergroup = aggstate->hash_pergroup; |
| 2184 |
TupleTableSlot *outerslot = aggstate->tmpcontext->ecxt_outertuple; |
| 2185 |
int setno; |
| 2186 |
|
| 2187 |
for (setno = 0; setno < aggstate->num_hashes; setno++) |
| 2188 |
{ |
| 2189 |
AggStatePerHash perhash = &aggstate->perhash[setno]; |
| 2190 |
TupleHashTable hashtable = perhash->hashtable; |
| 2191 |
TupleTableSlot *hashslot = perhash->hashslot; |
| 2192 |
TupleHashEntry entry; |
| 2193 |
uint32 hash; |
| 2194 |
bool isnew = false; |
| 2195 |
bool *p_isnew; |
| 2196 |
|
| 2197 |
/* if hash table already spilled, don't create new entries */ |
| 2198 |
p_isnew = aggstate->hash_spill_mode ? NULL : &isnew; |
| 2199 |
|
| 2200 |
select_current_set(aggstate, setno, true); |
| 2201 |
prepare_hash_slot(perhash, |
| 2202 |
outerslot, |
| 2203 |
hashslot); |
| 2204 |
|
| 2205 |
entry = LookupTupleHashEntry(hashtable, hashslot, |
| 2206 |
p_isnew, &hash); |
| 2207 |
|
| 2208 |
if (entry != NULL) |
| 2209 |
{ |
| 2210 |
if (isnew) |
| 2211 |
initialize_hash_entry(aggstate, hashtable, entry); |
| 2212 |
pergroup[setno] = TupleHashEntryGetAdditional(hashtable, entry); |
| 2213 |
} |
| 2214 |
else |
| 2215 |
{ |
| 2216 |
HashAggSpill *spill = &aggstate->hash_spills[setno]; |
| 2217 |
TupleTableSlot *slot = aggstate->tmpcontext->ecxt_outertuple; |
| 2218 |
|
| 2219 |
if (spill->partitions == NULL) |
| 2220 |
hashagg_spill_init(spill, aggstate->hash_tapeset, 0, |
| 2221 |
perhash->aggnode->numGroups, |
| 2222 |
aggstate->hashentrysize); |
| 2223 |
|
| 2224 |
hashagg_spill_tuple(aggstate, spill, slot, hash); |
| 2225 |
pergroup[setno] = NULL; |
| 2226 |
} |
| 2227 |
} |
| 2228 |
} |
| 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 * |
| 2244 |
ExecAgg(PlanState *pstate) |
| 2245 |
{ |
| 2246 |
AggState *node = castNode(AggState, pstate); |
| 2247 |
TupleTableSlot *result = NULL; |
| 2248 |
|
| 2249 |
CHECK_FOR_INTERRUPTS(); |
| 2250 |
|
| 2251 |
if (!node->agg_done) |
| 2252 |
{ |
| 2253 |
/* Dispatch based on strategy */ |
| 2254 |
switch (node->phase->aggstrategy) |
| 2255 |
{ |
| 2256 |
case AGG_HASHED: |
| 2257 |
if (!node->table_filled) |
| 2258 |
agg_fill_hash_table(node); |
| 2259 |
/* FALLTHROUGH */ |
| 2260 |
case AGG_MIXED: |
| 2261 |
result = agg_retrieve_hash_table(node); |
| 2262 |
break; |
| 2263 |
case AGG_PLAIN: |
| 2264 |
case AGG_SORTED: |
| 2265 |
result = agg_retrieve_direct(node); |
| 2266 |
break; |
| 2267 |
} |
| 2268 |
|
| 2269 |
if (!TupIsNull(result)) |
| 2270 |
return result; |
| 2271 |
} |
| 2272 |
|
| 2273 |
return NULL; |
| 2274 |
} |
| 2275 |
|
| 2276 |
/* |
| 2277 |
* ExecAgg for non-hashed case |
| 2278 |
*/ |
| 2279 |
static TupleTableSlot * |
| 2280 |
agg_retrieve_direct(AggState *aggstate) |
| 2281 |
{ |
| 2282 |
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 |
bool hasGroupingSets = aggstate->phase->numsets > 0; |
| 2291 |
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 |
*/ |
| 2304 |
econtext = aggstate->ss.ps.ps_ExprContext; |
| 2305 |
tmpcontext = aggstate->tmpcontext; |
| 2306 |
|
| 2307 |
peragg = aggstate->peragg; |
| 2308 |
pergroups = aggstate->pergroups; |
| 2309 |
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 |
*/ |
| 2320 |
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 |
*/ |
| 2334 |
ReScanExprContext(econtext); |
| 2335 |
|
| 2336 |
/* |
| 2337 |
* Determine how many grouping sets need to be reset at this boundary. |
| 2338 |
*/ |
| 2339 |
if (aggstate->projected_set >= 0 && |
| 2340 |
aggstate->projected_set < numGroupingSets) |
| 2341 |
numReset = aggstate->projected_set + 1; |
| 2342 |
else |
| 2343 |
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 |
for (i = 0; i < numReset; i++) |
| 2353 |
{ |
| 2354 |
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 |
if (aggstate->input_done == true && |
| 2362 |
aggstate->projected_set >= (numGroupingSets - 1)) |
| 2363 |
{ |
| 2364 |
if (aggstate->current_phase < aggstate->numphases - 1) |
| 2365 |
{ |
| 2366 |
initialize_phase(aggstate, aggstate->current_phase + 1); |
| 2367 |
aggstate->input_done = false; |
| 2368 |
aggstate->projected_set = -1; |
| 2369 |
numGroupingSets = Max(aggstate->phase->numsets, 1); |
| 2370 |
node = aggstate->phase->aggnode; |
| 2371 |
numReset = numGroupingSets; |
| 2372 |
} |
| 2373 |
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 |
initialize_phase(aggstate, 0); |
| 2380 |
aggstate->table_filled = true; |
| 2381 |
ResetTupleHashIterator(aggstate->perhash[0].hashtable, |
| 2382 |
&aggstate->perhash[0].hashiter); |
| 2383 |
select_current_set(aggstate, 0, true); |
| 2384 |
return agg_retrieve_hash_table(aggstate); |
| 2385 |
} |
| 2386 |
else |
| 2387 |
{ |
| 2388 |
aggstate->agg_done = true; |
| 2389 |
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 |
if (aggstate->projected_set >= 0 && |
| 2399 |
aggstate->projected_set < (numGroupingSets - 1)) |
| 2400 |
nextSetSize = aggstate->phase->gset_lengths[aggstate->projected_set + 1]; |
| 2401 |
else |
| 2402 |
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 |
*/ |
| 2421 |
tmpcontext->ecxt_innertuple = econtext->ecxt_outertuple; |
| 2422 |
if (aggstate->input_done || |
| 2423 |
(node->aggstrategy != AGG_PLAIN && |
| 2424 |
aggstate->projected_set != -1 && |
| 2425 |
aggstate->projected_set < (numGroupingSets - 1) && |
| 2426 |
nextSetSize > 0 && |
| 2427 |
!ExecQualAndReset(aggstate->phase->eqfunctions[nextSetSize - 1], |
| 2428 |
tmpcontext))) |
| 2429 |
{ |
| 2430 |
aggstate->projected_set += 1; |
| 2431 |
|
| 2432 |
Assert(aggstate->projected_set < numGroupingSets); |
| 2433 |
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 |
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 |
if (aggstate->grp_firstTuple == NULL) |
| 2449 |
{ |
| 2450 |
outerslot = fetch_input_tuple(aggstate); |
| 2451 |
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 |
*/ |
| 2457 |
aggstate->grp_firstTuple = ExecCopySlotHeapTuple(outerslot); |
| 2458 |
} |
| 2459 |
else |
| 2460 |
{ |
| 2461 |
/* outer plan produced no tuples at all */ |
| 2462 |
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 |
aggstate->input_done = true; |
| 2475 |
|
| 2476 |
while (aggstate->phase->gset_lengths[aggstate->projected_set] > 0) |
| 2477 |
{ |
| 2478 |
aggstate->projected_set += 1; |
| 2479 |
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 |
break; |
| 2488 |
} |
| 2489 |
} |
| 2490 |
|
| 2491 |
if (aggstate->projected_set >= numGroupingSets) |
| 2492 |
continue; |
| 2493 |
} |
| 2494 |
else |
| 2495 |
{ |
| 2496 |
aggstate->agg_done = true; |
| 2497 |
/* If we are grouping, we should produce no tuples too */ |
| 2498 |
if (node->aggstrategy != AGG_PLAIN) |
| 2499 |
return NULL; |
| 2500 |
} |
| 2501 |
} |
| 2502 |
} |
| 2503 |
|
| 2504 |
/* |
| 2505 |
* Initialize working state for a new input tuple group. |
| 2506 |
*/ |
| 2507 |
initialize_aggregates(aggstate, pergroups, numReset); |
| 2508 |
|
| 2509 |
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 |
*/ |
| 2516 |
ExecForceStoreHeapTuple(aggstate->grp_firstTuple, |
| 2517 |
firstSlot, true); |
| 2518 |
aggstate->grp_firstTuple = NULL; /* don't keep two pointers */ |
| 2519 |
|
| 2520 |
/* set up for first advance_aggregates call */ |
| 2521 |
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 |
*/ |
| 2533 |
if (aggstate->aggstrategy == AGG_MIXED && |
| 2534 |
aggstate->current_phase == 1) |
| 2535 |
{ |
| 2536 |
lookup_hash_entries(aggstate); |
| 2537 |
} |
| 2538 |
|
| 2539 |
/* Advance the aggregates (or combine functions) */ |
| 2540 |
advance_aggregates(aggstate); |
| 2541 |
|
| 2542 |
/* Reset per-input-tuple context after each tuple */ |
| 2543 |
ResetExprContext(tmpcontext); |
| 2544 |
|
| 2545 |
outerslot = fetch_input_tuple(aggstate); |
| 2546 |
if (TupIsNull(outerslot)) |
| 2547 |
{ |
| 2548 |
/* no more outer-plan tuples available */ |
| 2549 |
|
| 2550 |
/* if we built hash tables, finalize any spills */ |
| 2551 |
if (aggstate->aggstrategy == AGG_MIXED && |
| 2552 |
aggstate->current_phase == 1) |
| 2553 |
hashagg_finish_initial_spills(aggstate); |
| 2554 |
|
| 2555 |
if (hasGroupingSets) |
| 2556 |
{ |
| 2557 |
aggstate->input_done = true; |
| 2558 |
break; |
| 2559 |
} |
| 2560 |
else |
| 2561 |
{ |
| 2562 |
aggstate->agg_done = true; |
| 2563 |
break; |
| 2564 |
} |
| 2565 |
} |
| 2566 |
/* set up for next advance_aggregates call */ |
| 2567 |
tmpcontext->ecxt_outertuple = outerslot; |
| 2568 |
|
| 2569 |
/* |
| 2570 |
* If we are grouping, check whether we've crossed a group |
| 2571 |
* boundary. |
| 2572 |
*/ |
| 2573 |
if (node->aggstrategy != AGG_PLAIN && node->numCols > 0) |
| 2574 |
{ |
| 2575 |
tmpcontext->ecxt_innertuple = firstSlot; |
| 2576 |
if (!ExecQual(aggstate->phase->eqfunctions[node->numCols - 1], |
| 2577 |
tmpcontext)) |
| 2578 |
{ |
| 2579 |
aggstate->grp_firstTuple = ExecCopySlotHeapTuple(outerslot); |
| 2580 |
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 |
econtext->ecxt_outertuple = firstSlot; |
| 2595 |
} |
| 2596 |
|
| 2597 |
Assert(aggstate->projected_set >= 0); |
| 2598 |
|
| 2599 |
currentSet = aggstate->projected_set; |
| 2600 |
|
| 2601 |
prepare_projection_slot(aggstate, econtext->ecxt_outertuple, currentSet); |
| 2602 |
|
| 2603 |
select_current_set(aggstate, currentSet, false); |
| 2604 |
|
| 2605 |
finalize_aggregates(aggstate, |
| 2606 |
peragg, |
| 2607 |
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 |
*/ |
| 2613 |
result = project_aggregates(aggstate); |
| 2614 |
if (result) |
| 2615 |
return result; |
| 2616 |
} |
| 2617 |
|
| 2618 |
/* No more groups */ |
| 2619 |
return NULL; |
| 2620 |
} |
| 2621 |
|
| 2622 |
/* |
| 2623 |
* ExecAgg for hashed case: read input and build hash table |
| 2624 |
*/ |
| 2625 |
static void |
| 2626 |
agg_fill_hash_table(AggState *aggstate) |
| 2627 |
{ |
| 2628 |
TupleTableSlot *outerslot; |
| 2629 |
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 |
{ |
| 2637 |
outerslot = fetch_input_tuple(aggstate); |
| 2638 |
if (TupIsNull(outerslot)) |
| 2639 |
break; |
| 2640 |
|
| 2641 |
/* set up for lookup_hash_entries and advance_aggregates */ |
| 2642 |
tmpcontext->ecxt_outertuple = outerslot; |
| 2643 |
|
| 2644 |
/* Find or build hashtable entries */ |
| 2645 |
lookup_hash_entries(aggstate); |
| 2646 |
|
| 2647 |
/* Advance the aggregates (or combine functions) */ |
| 2648 |
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 |
*/ |
| 2654 |
ResetExprContext(aggstate->tmpcontext); |
| 2655 |
} |
| 2656 |
|
| 2657 |
/* finalize spills, if any */ |
| 2658 |
hashagg_finish_initial_spills(aggstate); |
| 2659 |
|
| 2660 |
aggstate->table_filled = true; |
| 2661 |
/* Initialize to walk the first hash table */ |
| 2662 |
select_current_set(aggstate, 0, true); |
| 2663 |
ResetTupleHashIterator(aggstate->perhash[0].hashtable, |
| 2664 |
&aggstate->perhash[0].hashiter); |
| 2665 |
} |
| 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 |
| 2680 |
agg_refill_hash_table(AggState *aggstate) |
| 2681 |
{ |
| 2682 |
HashAggBatch *batch; |
| 2683 |
AggStatePerHash perhash; |
| 2684 |
HashAggSpill spill; |
| 2685 |
LogicalTapeSet *tapeset = aggstate->hash_tapeset; |
| 2686 |
bool spill_initialized = false; |
| 2687 |
|
| 2688 |
if (aggstate->hash_batches == NIL) |
| 2689 |
return false; |
| 2690 |
|
| 2691 |
/* hash_batches is a stack, with the top item at the end of the list */ |
| 2692 |
batch = llast(aggstate->hash_batches); |
| 2693 |
aggstate->hash_batches = list_delete_last(aggstate->hash_batches); |
| 2694 |
|
| 2695 |
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 |
*/ |
| 2706 |
MemSet(aggstate->hash_pergroup, 0, |
| 2707 |
sizeof(AggStatePerGroup) * aggstate->num_hashes); |
| 2708 |
|
| 2709 |
/* free memory and reset hash tables */ |
| 2710 |
ReScanExprContext(aggstate->hashcontext); |
| 2711 |
MemoryContextReset(aggstate->hash_tablecxt); |
| 2712 |
for (int setno = 0; setno < aggstate->num_hashes; setno++) |
| 2713 |
ResetTupleHashTable(aggstate->perhash[setno].hashtable); |
| 2714 |
|
| 2715 |
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 |
Assert(aggstate->current_phase == 0); |
| 2723 |
if (aggstate->phase->aggstrategy == AGG_MIXED) |
| 2724 |
{ |
| 2725 |
aggstate->current_phase = 1; |
| 2726 |
aggstate->phase = &aggstate->phases[aggstate->current_phase]; |
| 2727 |
} |
| 2728 |
|
| 2729 |
select_current_set(aggstate, batch->setno, true); |
| 2730 |
|
| 2731 |
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 |
*/ |
| 2740 |
hashagg_recompile_expressions(aggstate, true, true); |
| 2741 |
|
| 2742 |
INJECTION_POINT("hash-aggregate-process-batch", NULL); |
| 2743 |
for (;;) |
| 2744 |
{ |
| 2745 |
TupleTableSlot *spillslot = aggstate->hash_spill_rslot; |
| 2746 |
TupleTableSlot *hashslot = perhash->hashslot; |
| 2747 |
TupleHashTable hashtable = perhash->hashtable; |
| 2748 |
TupleHashEntry entry; |
| 2749 |
MinimalTuple tuple; |
| 2750 |
uint32 hash; |
| 2751 |
bool isnew = false; |
| 2752 |
bool *p_isnew = aggstate->hash_spill_mode ? NULL : &isnew; |
| 2753 |
|
| 2754 |
CHECK_FOR_INTERRUPTS(); |
| 2755 |
|
| 2756 |
tuple = hashagg_batch_read(batch, &hash); |
| 2757 |
if (tuple == NULL) |
| 2758 |
break; |
| 2759 |
|
| 2760 |
ExecStoreMinimalTuple(tuple, spillslot, true); |
| 2761 |
aggstate->tmpcontext->ecxt_outertuple = spillslot; |
| 2762 |
|
| 2763 |
prepare_hash_slot(perhash, |
| 2764 |
aggstate->tmpcontext->ecxt_outertuple, |
| 2765 |
hashslot); |
| 2766 |
entry = LookupTupleHashEntryHash(hashtable, hashslot, |
| 2767 |
p_isnew, hash); |
| 2768 |
|
| 2769 |
if (entry != NULL) |
| 2770 |
{ |
| 2771 |
if (isnew) |
| 2772 |
initialize_hash_entry(aggstate, hashtable, entry); |
| 2773 |
aggstate->hash_pergroup[batch->setno] = TupleHashEntryGetAdditional(hashtable, entry); |
| 2774 |
advance_aggregates(aggstate); |
| 2775 |
} |
| 2776 |
else |
| 2777 |
{ |
| 2778 |
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 |
spill_initialized = true; |
| 2785 |
hashagg_spill_init(&spill, tapeset, batch->used_bits, |
| 2786 |
batch->input_card, aggstate->hashentrysize); |
| 2787 |
} |
| 2788 |
/* no memory for a new group, spill */ |
| 2789 |
hashagg_spill_tuple(aggstate, &spill, spillslot, hash); |
| 2790 |
|
| 2791 |
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 |
*/ |
| 2798 |
ResetExprContext(aggstate->tmpcontext); |
| 2799 |
} |
| 2800 |
|
| 2801 |
LogicalTapeClose(batch->input_tape); |
| 2802 |
|
| 2803 |
/* change back to phase 0 */ |
| 2804 |
aggstate->current_phase = 0; |
| 2805 |
aggstate->phase = &aggstate->phases[aggstate->current_phase]; |
| 2806 |
|
| 2807 |
if (spill_initialized) |
| 2808 |
{ |
| 2809 |
hashagg_spill_finish(aggstate, &spill, batch->setno); |
| 2810 |
hash_agg_update_metrics(aggstate, true, spill.npartitions); |
| 2811 |
} |
| 2812 |
else |
| 2813 |
hash_agg_update_metrics(aggstate, true, 0); |
| 2814 |
|
| 2815 |
aggstate->hash_spill_mode = false; |
| 2816 |
|
| 2817 |
/* prepare to walk the first hash table */ |
| 2818 |
select_current_set(aggstate, batch->setno, true); |
| 2819 |
ResetTupleHashIterator(aggstate->perhash[batch->setno].hashtable, |
| 2820 |
&aggstate->perhash[batch->setno].hashiter); |
| 2821 |
|
| 2822 |
pfree(batch); |
| 2823 |
|
| 2824 |
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 * |
| 2835 |
agg_retrieve_hash_table(AggState *aggstate) |
| 2836 |
{ |
| 2837 |
TupleTableSlot *result = NULL; |
| 2838 |
|
| 2839 |
while (result == NULL) |
| 2840 |
{ |
| 2841 |
result = agg_retrieve_hash_table_in_memory(aggstate); |
| 2842 |
if (result == NULL) |
| 2843 |
{ |
| 2844 |
if (!agg_refill_hash_table(aggstate)) |
| 2845 |
{ |
| 2846 |
aggstate->agg_done = true; |
| 2847 |
break; |
| 2848 |
} |
| 2849 |
} |
| 2850 |
} |
| 2851 |
|
| 2852 |
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 |
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 |
*/ |
| 2875 |
econtext = aggstate->ss.ps.ps_ExprContext; |
| 2876 |
peragg = aggstate->peragg; |
| 2877 |
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 |
*/ |
| 2883 |
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 (;;) |
| 2890 |
{ |
| 2891 |
TupleTableSlot *hashslot = perhash->hashslot; |
| 2892 |
TupleHashTable hashtable = perhash->hashtable; |
| 2893 |
int i; |
| 2894 |
|
| 2895 |
CHECK_FOR_INTERRUPTS(); |
| 2896 |
|
| 2897 |
/* |
| 2898 |
* Find the next entry in the hash table |
| 2899 |
*/ |
| 2900 |
entry = ScanTupleHashTable(hashtable, &perhash->hashiter); |
| 2901 |
if (entry == NULL) |
| 2902 |
{ |
| 2903 |
int nextset = aggstate->current_set + 1; |
| 2904 |
|
| 2905 |
if (nextset < aggstate->num_hashes) |
| 2906 |
{ |
| 2907 |
/* |
| 2908 |
* Switch to next grouping set, reinitialize, and restart the |
| 2909 |
* loop. |
| 2910 |
*/ |
| 2911 |
select_current_set(aggstate, nextset, true); |
| 2912 |
|
| 2913 |
perhash = &aggstate->perhash[aggstate->current_set]; |
| 2914 |
|
| 2915 |
ResetTupleHashIterator(perhash->hashtable, &perhash->hashiter); |
| 2916 |
|
| 2917 |
continue; |
| 2918 |
} |
| 2919 |
else |
| 2920 |
{ |
| 2921 |
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 |
*/ |
| 2932 |
ResetExprContext(econtext); |
| 2933 |
|
| 2934 |
/* |
| 2935 |
* Transform representative tuple back into one with the right |
| 2936 |
* columns. |
| 2937 |
*/ |
| 2938 |
ExecStoreMinimalTuple(TupleHashEntryGetTuple(entry), hashslot, false); |
| 2939 |
slot_getallattrs(hashslot); |
| 2940 |
|
| 2941 |
ExecClearTuple(firstSlot); |
| 2942 |
memset(firstSlot->tts_isnull, true, |
| 2943 |
firstSlot->tts_tupleDescriptor->natts * sizeof(bool)); |
| 2944 |
|
| 2945 |
for (i = 0; i < perhash->numhashGrpCols; i++) |
| 2946 |
{ |
| 2947 |
int varNumber = perhash->hashGrpColIdxInput[i] - 1; |
| 2948 |
|
| 2949 |
firstSlot->tts_values[varNumber] = hashslot->tts_values[i]; |
| 2950 |
firstSlot->tts_isnull[varNumber] = hashslot->tts_isnull[i]; |
| 2951 |
} |
| 2952 |
ExecStoreVirtualTuple(firstSlot); |
| 2953 |
|
| 2954 |
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 |
*/ |
| 2960 |
econtext->ecxt_outertuple = firstSlot; |
| 2961 |
|
| 2962 |
prepare_projection_slot(aggstate, |
| 2963 |
econtext->ecxt_outertuple, |
| 2964 |
aggstate->current_set); |
| 2965 |
|
| 2966 |
finalize_aggregates(aggstate, peragg, pergroup); |
| 2967 |
|
| 2968 |
result = project_aggregates(aggstate); |
| 2969 |
if (result) |
| 2970 |
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 |
| 2984 |
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 |
|
| 2990 |
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 |
|
| 3002 |
spill->partitions = palloc0(sizeof(LogicalTape *) * npartitions); |
| 3003 |
spill->ntuples = palloc0(sizeof(int64) * npartitions); |
| 3004 |
spill->hll_card = palloc0(sizeof(hyperLogLogState) * npartitions); |
| 3005 |
|
| 3006 |
for (int i = 0; i < npartitions; i++) |
| 3007 |
spill->partitions[i] = LogicalTapeCreate(tapeset); |
| 3008 |
|
| 3009 |
spill->shift = 32 - used_bits - partition_bits; |
| 3010 |
if (spill->shift < 32) |
| 3011 |
spill->mask = (npartitions - 1) << spill->shift; |
| 3012 |
else |
| 3013 |
spill->mask = 0; |
| 3014 |
spill->npartitions = npartitions; |
| 3015 |
|
| 3016 |
for (int i = 0; i < npartitions; i++) |
| 3017 |
initHyperLogLog(&spill->hll_card[i], HASHAGG_HLL_BIT_WIDTH); |
| 3018 |
} |
| 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 |
| 3027 |
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; |
| 3034 |
int total_written = 0; |
| 3035 |
bool shouldFree; |
| 3036 |
|
| 3037 |
Assert(spill->partitions != NULL); |
| 3038 |
|
| 3039 |
/* spill only attributes that we actually need */ |
| 3040 |
if (!aggstate->all_cols_needed) |
| 3041 |
{ |
| 3042 |
spillslot = aggstate->hash_spill_wslot; |
| 3043 |
slot_getsomeattrs(inputslot, aggstate->max_colno_needed); |
| 3044 |
ExecClearTuple(spillslot); |
| 3045 |
for (int i = 0; i < spillslot->tts_tupleDescriptor->natts; i++) |
| 3046 |
{ |
| 3047 |
if (bms_is_member(i + 1, aggstate->colnos_needed)) |
| 3048 |
{ |
| 3049 |
spillslot->tts_values[i] = inputslot->tts_values[i]; |
| 3050 |
spillslot->tts_isnull[i] = inputslot->tts_isnull[i]; |
| 3051 |
} |
| 3052 |
else |
| 3053 |
spillslot->tts_isnull[i] = true; |
| 3054 |
} |
| 3055 |
ExecStoreVirtualTuple(spillslot); |
| 3056 |
} |
| 3057 |
else |
| 3058 |
spillslot = inputslot; |
| 3059 |
|
| 3060 |
tuple = ExecFetchSlotMinimalTuple(spillslot, &shouldFree); |
| 3061 |
|
| 3062 |
if (spill->shift < 32) |
| 3063 |
partition = (hash & spill->mask) >> spill->shift; |
| 3064 |
else |
| 3065 |
partition = 0; |
| 3066 |
|
| 3067 |
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 |
*/ |
| 3074 |
addHyperLogLog(&spill->hll_card[partition], hash_bytes_uint32(hash)); |
| 3075 |
|
| 3076 |
tape = spill->partitions[partition]; |
| 3077 |
|
| 3078 |
LogicalTapeWrite(tape, &hash, sizeof(uint32)); |
| 3079 |
total_written += sizeof(uint32); |
| 3080 |
|
| 3081 |
LogicalTapeWrite(tape, tuple, tuple->t_len); |
| 3082 |
total_written += tuple->t_len; |
| 3083 |
|
| 3084 |
if (shouldFree) |
| 3085 |
pfree(tuple); |
| 3086 |
|
| 3087 |
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 * |
| 3097 |
hashagg_batch_new(LogicalTape *input_tape, int setno, |
| 3098 |
int64 input_tuples, double input_card, int used_bits) |
| 3099 |
{ |
| 3100 |
HashAggBatch *batch = palloc0(sizeof(HashAggBatch)); |
| 3101 |
|
| 3102 |
batch->setno = setno; |
| 3103 |
batch->used_bits = used_bits; |
| 3104 |
batch->input_tape = input_tape; |
| 3105 |
batch->input_tuples = input_tuples; |
| 3106 |
batch->input_card = input_card; |
| 3107 |
|
| 3108 |
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 |
hashagg_batch_read(HashAggBatch *batch, uint32 *hashp) |
| 3117 |
{ |
| 3118 |
LogicalTape *tape = batch->input_tape; |
| 3119 |
MinimalTuple tuple; |
| 3120 |
uint32 t_len; |
| 3121 |
size_t nread; |
| 3122 |
uint32 hash; |
| 3123 |
|
| 3124 |
nread = LogicalTapeRead(tape, &hash, sizeof(uint32)); |
| 3125 |
if (nread == 0) |
| 3126 |
return NULL; |
| 3127 |
if (nread != sizeof(uint32)) |
| 3128 |
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))); |
| 3132 |
if (hashp != NULL) |
| 3133 |
*hashp = hash; |
| 3134 |
|
| 3135 |
nread = LogicalTapeRead(tape, &t_len, sizeof(t_len)); |
| 3136 |
if (nread != sizeof(uint32)) |
| 3137 |
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 |
|
| 3142 |
tuple = (MinimalTuple) palloc(t_len); |
| 3143 |
tuple->t_len = t_len; |
| 3144 |
|
| 3145 |
nread = LogicalTapeRead(tape, |
| 3146 |
(char *) tuple + sizeof(uint32), |
| 3147 |
t_len - sizeof(uint32)); |
| 3148 |
if (nread != t_len - sizeof(uint32)) |
| 3149 |
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 |
|
| 3154 |
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 |
hashagg_finish_initial_spills(AggState *aggstate) |
| 3166 |
{ |
| 3167 |
int setno; |
| 3168 |
int total_npartitions = 0; |
| 3169 |
|
| 3170 |
if (aggstate->hash_spills != NULL) |
| 3171 |
{ |
| 3172 |
for (setno = 0; setno < aggstate->num_hashes; setno++) |
| 3173 |
{ |
| 3174 |
HashAggSpill *spill = &aggstate->hash_spills[setno]; |
| 3175 |
|
| 3176 |
total_npartitions += spill->npartitions; |
| 3177 |
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 |
pfree(aggstate->hash_spills); |
| 3186 |
aggstate->hash_spills = NULL; |
| 3187 |
} |
| 3188 |
|
| 3189 |
hash_agg_update_metrics(aggstate, false, total_npartitions); |
| 3190 |
aggstate->hash_spill_mode = false; |
| 3191 |
} |
| 3192 |
|
| 3193 |
/* |
| 3194 |
* hashagg_spill_finish |
| 3195 |
* |
| 3196 |
* Transform spill partitions into new batches. |
| 3197 |
*/ |
| 3198 |
static void |
| 3199 |
hashagg_spill_finish(AggState *aggstate, HashAggSpill *spill, int setno) |
| 3200 |
{ |
| 3201 |
int i; |
| 3202 |
int used_bits = 32 - spill->shift; |
| 3203 |
|
| 3204 |
if (spill->npartitions == 0) |
| 3205 |
return; /* didn't spill */ |
| 3206 |
|
| 3207 |
for (i = 0; i < spill->npartitions; i++) |
| 3208 |
{ |
| 3209 |
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 */ |
| 3214 |
if (spill->ntuples[i] == 0) |
| 3215 |
continue; |
| 3216 |
|
| 3217 |
cardinality = estimateHyperLogLog(&spill->hll_card[i]); |
| 3218 |
freeHyperLogLog(&spill->hll_card[i]); |
| 3219 |
|
| 3220 |
/* rewinding frees the buffer while not in use */ |
| 3221 |
LogicalTapeRewindForRead(tape, HASHAGG_READ_BUFFER_SIZE); |
| 3222 |
|
| 3223 |
new_batch = hashagg_batch_new(tape, setno, |
| 3224 |
spill->ntuples[i], cardinality, |
| 3225 |
used_bits); |
| 3226 |
aggstate->hash_batches = lappend(aggstate->hash_batches, new_batch); |
| 3227 |
aggstate->hash_batches_used++; |
| 3228 |
} |
| 3229 |
|
| 3230 |
pfree(spill->ntuples); |
| 3231 |
pfree(spill->hll_card); |
| 3232 |
pfree(spill->partitions); |
| 3233 |
} |
| 3234 |
|
| 3235 |
/* |
| 3236 |
* Free resources related to a spilled HashAgg. |
| 3237 |
*/ |
| 3238 |
static void |
| 3239 |
hashagg_reset_spill_state(AggState *aggstate) |
| 3240 |
{ |
| 3241 |
/* free spills from initial pass */ |
| 3242 |
if (aggstate->hash_spills != NULL) |
| 3243 |
{ |
| 3244 |
int setno; |
| 3245 |
|
| 3246 |
for (setno = 0; setno < aggstate->num_hashes; setno++) |
| 3247 |
{ |
| 3248 |
HashAggSpill *spill = &aggstate->hash_spills[setno]; |
| 3249 |
|
| 3250 |
pfree(spill->ntuples); |
| 3251 |
pfree(spill->partitions); |
| 3252 |
} |
| 3253 |
pfree(aggstate->hash_spills); |
| 3254 |
aggstate->hash_spills = NULL; |
| 3255 |
} |
| 3256 |
|
| 3257 |
/* free batches */ |
| 3258 |
list_free_deep(aggstate->hash_batches); |
| 3259 |
aggstate->hash_batches = NIL; |
| 3260 |
|
| 3261 |
/* close tape set */ |
| 3262 |
if (aggstate->hash_tapeset != NULL) |
| 3263 |
{ |
| 3264 |
LogicalTapeSetClose(aggstate->hash_tapeset); |
| 3265 |
aggstate->hash_tapeset = NULL; |
| 3266 |
} |
| 3267 |
} |
| 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 * |
| 3279 |
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; |
| 3296 |
Bitmapset *all_grouped_cols = NULL; |
| 3297 |
int numGroupingSets = 1; |
| 3298 |
int numPhases; |
| 3299 |
int numHashes; |
| 3300 |
int i = 0; |
| 3301 |
int j = 0; |
| 3302 |
bool use_hashing = (node->aggstrategy == AGG_HASHED || |
| 3303 |
node->aggstrategy == AGG_MIXED); |
| 3304 |
|
| 3305 |
/* check for unsupported flags */ |
| 3306 |
Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK))); |
| 3307 |
|
| 3308 |
/* |
| 3309 |
* create state structure |
| 3310 |
*/ |
| 3311 |
aggstate = makeNode(AggState); |
| 3312 |
aggstate->ss.ps.plan = (Plan *) node; |
| 3313 |
aggstate->ss.ps.state = estate; |
| 3314 |
aggstate->ss.ps.ExecProcNode = ExecAgg; |
| 3315 |
|
| 3316 |
aggstate->aggs = NIL; |
| 3317 |
aggstate->numaggs = 0; |
| 3318 |
aggstate->numtrans = 0; |
| 3319 |
aggstate->aggstrategy = node->aggstrategy; |
| 3320 |
aggstate->aggsplit = node->aggsplit; |
| 3321 |
aggstate->maxsets = 0; |
| 3322 |
aggstate->projected_set = -1; |
| 3323 |
aggstate->current_set = 0; |
| 3324 |
aggstate->peragg = NULL; |
| 3325 |
aggstate->pertrans = NULL; |
| 3326 |
aggstate->curperagg = NULL; |
| 3327 |
aggstate->curpertrans = NULL; |
| 3328 |
aggstate->input_done = false; |
| 3329 |
aggstate->agg_done = false; |
| 3330 |
aggstate->pergroups = NULL; |
| 3331 |
aggstate->grp_firstTuple = NULL; |
| 3332 |
aggstate->sort_in = NULL; |
| 3333 |
aggstate->sort_out = NULL; |
| 3334 |
|
| 3335 |
/* |
| 3336 |
* phases[0] always exists, but is dummy in sorted/plain mode |
| 3337 |
*/ |
| 3338 |
numPhases = (use_hashing ? 1 : 2); |
| 3339 |
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 |
*/ |
| 3346 |
if (node->groupingSets) |
| 3347 |
{ |
| 3348 |
numGroupingSets = list_length(node->groupingSets); |
| 3349 |
|
| 3350 |
foreach(l, node->chain) |
| 3351 |
{ |
| 3352 |
Agg *agg = lfirst(l); |
| 3353 |
|
| 3354 |
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 |
*/ |
| 3361 |
if (agg->aggstrategy != AGG_HASHED) |
| 3362 |
++numPhases; |
| 3363 |
else |
| 3364 |
++numHashes; |
| 3365 |
} |
| 3366 |
} |
| 3367 |
|
| 3368 |
aggstate->maxsets = numGroupingSets; |
| 3369 |
aggstate->numphases = numPhases; |
| 3370 |
|
| 3371 |
aggstate->aggcontexts = (ExprContext **) |
| 3372 |
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 |
ExecAssignExprContext(estate, &aggstate->ss.ps); |
| 3389 |
aggstate->tmpcontext = aggstate->ss.ps.ps_ExprContext; |
| 3390 |
|
| 3391 |
for (i = 0; i < numGroupingSets; ++i) |
| 3392 |
{ |
| 3393 |
ExecAssignExprContext(estate, &aggstate->ss.ps); |
| 3394 |
aggstate->aggcontexts[i] = aggstate->ss.ps.ps_ExprContext; |
| 3395 |
} |
| 3396 |
|
| 3397 |
if (use_hashing) |
| 3398 |
hash_create_memory(aggstate); |
| 3399 |
|
| 3400 |
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 |
*/ |
| 3408 |
if (node->aggstrategy == AGG_HASHED) |
| 3409 |
eflags &= ~EXEC_FLAG_REWIND; |
| 3410 |
outerPlan = outerPlan(node); |
| 3411 |
outerPlanState(aggstate) = ExecInitNode(outerPlan, estate, eflags); |
| 3412 |
|
| 3413 |
/* |
| 3414 |
* initialize source tuple type. |
| 3415 |
*/ |
| 3416 |
aggstate->ss.ps.outerops = |
| 3417 |
ExecGetResultSlotOps(outerPlanState(&aggstate->ss), |
| 3418 |
&aggstate->ss.ps.outeropsfixed); |
| 3419 |
aggstate->ss.ps.outeropsset = true; |
| 3420 |
|
| 3421 |
ExecCreateScanSlotFromOuterPlan(estate, &aggstate->ss, |
| 3422 |
aggstate->ss.ps.outerops); |
| 3423 |
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 |
*/ |
| 3429 |
if (numPhases > 2) |
| 3430 |
{ |
| 3431 |
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 |
*/ |
| 3448 |
if (aggstate->ss.ps.outeropsfixed && |
| 3449 |
aggstate->ss.ps.outerops != &TTSOpsMinimalTuple) |
| 3450 |
aggstate->ss.ps.outeropsfixed = false; |
| 3451 |
} |
| 3452 |
|
| 3453 |
/* |
| 3454 |
* Initialize result type, slot and projection. |
| 3455 |
*/ |
| 3456 |
ExecInitResultTupleSlotTL(&aggstate->ss.ps, &TTSOpsVirtual); |
| 3457 |
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 |
*/ |
| 3472 |
aggstate->ss.ps.qual = |
| 3473 |
ExecInitQual(node->plan.qual, (PlanState *) aggstate); |
| 3474 |
|
| 3475 |
/* |
| 3476 |
* We should now have found all Aggrefs in the targetlist and quals. |
| 3477 |
*/ |
| 3478 |
numaggrefs = list_length(aggstate->aggs); |
| 3479 |
max_aggno = -1; |
| 3480 |
max_transno = -1; |
| 3481 |
foreach(l, aggstate->aggs) |
| 3482 |
{ |
| 3483 |
Aggref *aggref = (Aggref *) lfirst(l); |
| 3484 |
|
| 3485 |
max_aggno = Max(max_aggno, aggref->aggno); |
| 3486 |
max_transno = Max(max_transno, aggref->aggtransno); |
| 3487 |
} |
| 3488 |
aggstate->numaggs = numaggs = max_aggno + 1; |
| 3489 |
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 |
*/ |
| 3495 |
aggstate->phases = palloc0(numPhases * sizeof(AggStatePerPhaseData)); |
| 3496 |
|
| 3497 |
aggstate->num_hashes = numHashes; |
| 3498 |
if (numHashes) |
| 3499 |
{ |
| 3500 |
aggstate->perhash = palloc0(sizeof(AggStatePerHashData) * numHashes); |
| 3501 |
aggstate->phases[0].numsets = 0; |
| 3502 |
aggstate->phases[0].gset_lengths = palloc(numHashes * sizeof(int)); |
| 3503 |
aggstate->phases[0].grouped_cols = palloc(numHashes * sizeof(Bitmapset *)); |
| 3504 |
} |
| 3505 |
|
| 3506 |
phase = 0; |
| 3507 |
for (phaseidx = 0; phaseidx <= list_length(node->chain); ++phaseidx) |
| 3508 |
{ |
| 3509 |
Agg *aggnode; |
| 3510 |
Sort *sortnode; |
| 3511 |
|
| 3512 |
if (phaseidx > 0) |
| 3513 |
{ |
| 3514 |
aggnode = list_nth_node(Agg, node->chain, phaseidx - 1); |
| 3515 |
sortnode = castNode(Sort, outerPlan(aggnode)); |
| 3516 |
} |
| 3517 |
else |
| 3518 |
{ |
| 3519 |
aggnode = node; |
| 3520 |
sortnode = NULL; |
| 3521 |
} |
| 3522 |
|
| 3523 |
Assert(phase <= 1 || sortnode); |
| 3524 |
|
| 3525 |
if (aggnode->aggstrategy == AGG_HASHED |
| 3526 |
|| aggnode->aggstrategy == AGG_MIXED) |
| 3527 |
{ |
| 3528 |
AggStatePerPhase phasedata = &aggstate->phases[0]; |
| 3529 |
AggStatePerHash perhash; |
| 3530 |
Bitmapset *cols = NULL; |
| 3531 |
|
| 3532 |
Assert(phase == 0); |
| 3533 |
i = phasedata->numsets++; |
| 3534 |
perhash = &aggstate->perhash[i]; |
| 3535 |
|
| 3536 |
/* phase 0 always points to the "real" Agg in the hash case */ |
| 3537 |
phasedata->aggnode = node; |
| 3538 |
phasedata->aggstrategy = node->aggstrategy; |
| 3539 |
|
| 3540 |
/* but the actual Agg node representing this hash is saved here */ |
| 3541 |
perhash->aggnode = aggnode; |
| 3542 |
|
| 3543 |
phasedata->gset_lengths[i] = perhash->numCols = aggnode->numCols; |
| 3544 |
|
| 3545 |
for (j = 0; j < aggnode->numCols; ++j) |
| 3546 |
cols = bms_add_member(cols, aggnode->grpColIdx[j]); |
| 3547 |
|
| 3548 |
phasedata->grouped_cols[i] = cols; |
| 3549 |
|
| 3550 |
all_grouped_cols = bms_add_members(all_grouped_cols, cols); |
| 3551 |
continue; |
| 3552 |
} |
| 3553 |
else |
| 3554 |
{ |
| 3555 |
AggStatePerPhase phasedata = &aggstate->phases[++phase]; |
| 3556 |
int num_sets; |
| 3557 |
|
| 3558 |
phasedata->numsets = num_sets = list_length(aggnode->groupingSets); |
| 3559 |
|
| 3560 |
if (num_sets) |
| 3561 |
{ |
| 3562 |
phasedata->gset_lengths = palloc(num_sets * sizeof(int)); |
| 3563 |
phasedata->grouped_cols = palloc(num_sets * sizeof(Bitmapset *)); |
| 3564 |
|
| 3565 |
i = 0; |
| 3566 |
foreach(l, aggnode->groupingSets) |
| 3567 |
{ |
| 3568 |
int current_length = list_length(lfirst(l)); |
| 3569 |
Bitmapset *cols = NULL; |
| 3570 |
|
| 3571 |
/* planner forces this to be correct */ |
| 3572 |
for (j = 0; j < current_length; ++j) |
| 3573 |
cols = bms_add_member(cols, aggnode->grpColIdx[j]); |
| 3574 |
|
| 3575 |
phasedata->grouped_cols[i] = cols; |
| 3576 |
phasedata->gset_lengths[i] = current_length; |
| 3577 |
|
| 3578 |
++i; |
| 3579 |
} |
| 3580 |
|
| 3581 |
all_grouped_cols = bms_add_members(all_grouped_cols, |
| 3582 |
phasedata->grouped_cols[0]); |
| 3583 |
} |
| 3584 |
else |
| 3585 |
{ |
| 3586 |
Assert(phaseidx == 0); |
| 3587 |
|
| 3588 |
phasedata->gset_lengths = NULL; |
| 3589 |
phasedata->grouped_cols = NULL; |
| 3590 |
} |
| 3591 |
|
| 3592 |
/* |
| 3593 |
* If we are grouping, precompute fmgr lookup data for inner loop. |
| 3594 |
*/ |
| 3595 |
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 |
phasedata->eqfunctions = |
| 3602 |
(ExprState **) palloc0(aggnode->numCols * sizeof(ExprState *)); |
| 3603 |
|
| 3604 |
/* for each grouping set */ |
| 3605 |
for (int k = 0; k < phasedata->numsets; k++) |
| 3606 |
{ |
| 3607 |
int length = phasedata->gset_lengths[k]; |
| 3608 |
|
| 3609 |
/* nothing to do for empty grouping set */ |
| 3610 |
if (length == 0) |
| 3611 |
continue; |
| 3612 |
|
| 3613 |
/* if we already had one of this length, it'll do */ |
| 3614 |
if (phasedata->eqfunctions[length - 1] != NULL) |
| 3615 |
continue; |
| 3616 |
|
| 3617 |
phasedata->eqfunctions[length - 1] = |
| 3618 |
execTuplesMatchPrepare(scanDesc, |
| 3619 |
length, |
| 3620 |
aggnode->grpColIdx, |
| 3621 |
aggnode->grpOperators, |
| 3622 |
aggnode->grpCollations, |
| 3623 |
(PlanState *) aggstate); |
| 3624 |
} |
| 3625 |
|
| 3626 |
/* and for all grouped columns, unless already computed */ |
| 3627 |
if (aggnode->numCols > 0 && |
| 3628 |
phasedata->eqfunctions[aggnode->numCols - 1] == NULL) |
| 3629 |
{ |
| 3630 |
phasedata->eqfunctions[aggnode->numCols - 1] = |
| 3631 |
execTuplesMatchPrepare(scanDesc, |
| 3632 |
aggnode->numCols, |
| 3633 |
aggnode->grpColIdx, |
| 3634 |
aggnode->grpOperators, |
| 3635 |
aggnode->grpCollations, |
| 3636 |
(PlanState *) aggstate); |
| 3637 |
} |
| 3638 |
} |
| 3639 |
|
| 3640 |
phasedata->aggnode = aggnode; |
| 3641 |
phasedata->aggstrategy = aggnode->aggstrategy; |
| 3642 |
phasedata->sortnode = sortnode; |
| 3643 |
} |
| 3644 |
} |
| 3645 |
|
| 3646 |
/* |
| 3647 |
* Convert all_grouped_cols to a descending-order list. |
| 3648 |
*/ |
| 3649 |
i = -1; |
| 3650 |
while ((i = bms_next_member(all_grouped_cols, i)) >= 0) |
| 3651 |
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 |
*/ |
| 3657 |
econtext = aggstate->ss.ps.ps_ExprContext; |
| 3658 |
econtext->ecxt_aggvalues = (Datum *) palloc0(sizeof(Datum) * numaggs); |
| 3659 |
econtext->ecxt_aggnulls = (bool *) palloc0(sizeof(bool) * numaggs); |
| 3660 |
|
| 3661 |
peraggs = (AggStatePerAgg) palloc0(sizeof(AggStatePerAggData) * numaggs); |
| 3662 |
pertransstates = (AggStatePerTrans) palloc0(sizeof(AggStatePerTransData) * numtrans); |
| 3663 |
|
| 3664 |
aggstate->peragg = peraggs; |
| 3665 |
aggstate->pertrans = pertransstates; |
| 3666 |
|
| 3667 |
|
| 3668 |
aggstate->all_pergroups = |
| 3669 |
(AggStatePerGroup *) palloc0(sizeof(AggStatePerGroup) |
| 3670 |
* (numGroupingSets + numHashes)); |
| 3671 |
pergroups = aggstate->all_pergroups; |
| 3672 |
|
| 3673 |
if (node->aggstrategy != AGG_HASHED) |
| 3674 |
{ |
| 3675 |
for (i = 0; i < numGroupingSets; i++) |
| 3676 |
{ |
| 3677 |
pergroups[i] = (AggStatePerGroup) palloc0(sizeof(AggStatePerGroupData) |
| 3678 |
* numaggs); |
| 3679 |
} |
| 3680 |
|
| 3681 |
aggstate->pergroups = pergroups; |
| 3682 |
pergroups += numGroupingSets; |
| 3683 |
} |
| 3684 |
|
| 3685 |
/* |
| 3686 |
* Hashing can only appear in the initial phase. |
| 3687 |
*/ |
| 3688 |
if (use_hashing) |
| 3689 |
{ |
| 3690 |
Plan *outerplan = outerPlan(node); |
| 3691 |
uint64 totalGroups = 0; |
| 3692 |
|
| 3693 |
aggstate->hash_spill_rslot = ExecInitExtraTupleSlot(estate, scanDesc, |
| 3694 |
&TTSOpsMinimalTuple); |
| 3695 |
aggstate->hash_spill_wslot = ExecInitExtraTupleSlot(estate, scanDesc, |
| 3696 |
&TTSOpsVirtual); |
| 3697 |
|
| 3698 |
/* this is an array of pointers, not structures */ |
| 3699 |
aggstate->hash_pergroup = pergroups; |
| 3700 |
|
| 3701 |
aggstate->hashentrysize = hash_agg_entry_size(aggstate->numtrans, |
| 3702 |
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 |
*/ |
| 3711 |
for (int k = 0; k < aggstate->num_hashes; k++) |
| 3712 |
totalGroups += aggstate->perhash[k].aggnode->numGroups; |
| 3713 |
|
| 3714 |
hash_agg_set_limits(aggstate->hashentrysize, totalGroups, 0, |
| 3715 |
&aggstate->hash_mem_limit, |
| 3716 |
&aggstate->hash_ngroups_limit, |
| 3717 |
&aggstate->hash_planned_partitions); |
| 3718 |
find_hash_columns(aggstate); |
| 3719 |
|
| 3720 |
/* Skip massive memory allocation if we are just doing EXPLAIN */ |
| 3721 |
if (!(eflags & EXEC_FLAG_EXPLAIN_ONLY)) |
| 3722 |
build_hash_tables(aggstate); |
| 3723 |
|
| 3724 |
aggstate->table_filled = false; |
| 3725 |
|
| 3726 |
/* Initialize this to 1, meaning nothing spilled, yet */ |
| 3727 |
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 |
*/ |
| 3736 |
if (node->aggstrategy == AGG_HASHED) |
| 3737 |
{ |
| 3738 |
aggstate->current_phase = 0; |
| 3739 |
initialize_phase(aggstate, 0); |
| 3740 |
select_current_set(aggstate, 0, true); |
| 3741 |
} |
| 3742 |
else |
| 3743 |
{ |
| 3744 |
aggstate->current_phase = 1; |
| 3745 |
initialize_phase(aggstate, 1); |
| 3746 |
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 |
*/ |
| 3753 |
foreach(l, aggstate->aggs) |
| 3754 |
{ |
| 3755 |
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 */ |
| 3772 |
Assert(aggref->agglevelsup == 0); |
| 3773 |
/* ... and the split mode should match */ |
| 3774 |
Assert(aggref->aggsplit == aggstate->aggsplit); |
| 3775 |
|
| 3776 |
peragg = &peraggs[aggref->aggno]; |
| 3777 |
|
| 3778 |
/* Check if we initialized the state for this aggregate already. */ |
| 3779 |
if (peragg->aggref != NULL) |
| 3780 |
continue; |
| 3781 |
|
| 3782 |
peragg->aggref = aggref; |
| 3783 |
peragg->transno = aggref->aggtransno; |
| 3784 |
|
| 3785 |
/* Fetch the pg_aggregate row */ |
| 3786 |
aggTuple = SearchSysCache1(AGGFNOID, |
| 3787 |
ObjectIdGetDatum(aggref->aggfnoid)); |
| 3788 |
if (!HeapTupleIsValid(aggTuple)) |
| 3789 |
elog(ERROR, "cache lookup failed for aggregate %u", |
| 3790 |
aggref->aggfnoid); |
| 3791 |
aggform = (Form_pg_aggregate) GETSTRUCT(aggTuple); |
| 3792 |
|
| 3793 |
/* Check permission to call aggregate function */ |
| 3794 |
aclresult = object_aclcheck(ProcedureRelationId, aggref->aggfnoid, GetUserId(), |
| 3795 |
ACL_EXECUTE); |
| 3796 |
if (aclresult != ACLCHECK_OK) |
| 3797 |
aclcheck_error(aclresult, OBJECT_AGGREGATE, |
| 3798 |
get_func_name(aggref->aggfnoid)); |
| 3799 |
InvokeFunctionExecuteHook(aggref->aggfnoid); |
| 3800 |
|
| 3801 |
/* planner recorded transition state type in the Aggref itself */ |
| 3802 |
aggtranstype = aggref->aggtranstype; |
| 3803 |
Assert(OidIsValid(aggtranstype)); |
| 3804 |
|
| 3805 |
/* Final function only required if we're finalizing the aggregates */ |
| 3806 |
if (DO_AGGSPLIT_SKIPFINAL(aggstate->aggsplit)) |
| 3807 |
peragg->finalfn_oid = finalfn_oid = InvalidOid; |
| 3808 |
else |
| 3809 |
peragg->finalfn_oid = finalfn_oid = aggform->aggfinalfn; |
| 3810 |
|
| 3811 |
serialfn_oid = InvalidOid; |
| 3812 |
deserialfn_oid = InvalidOid; |
| 3813 |
|
| 3814 |
/* |
| 3815 |
* Check if serialization/deserialization is required. We only do it |
| 3816 |
* for aggregates that have transtype INTERNAL. |
| 3817 |
*/ |
| 3818 |
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 |
if (DO_AGGSPLIT_SERIALIZE(aggstate->aggsplit)) |
| 3826 |
{ |
| 3827 |
/* serialization only valid when not running finalfn */ |
| 3828 |
Assert(DO_AGGSPLIT_SKIPFINAL(aggstate->aggsplit)); |
| 3829 |
|
| 3830 |
if (!OidIsValid(aggform->aggserialfn)) |
| 3831 |
elog(ERROR, "serialfunc not provided for serialization aggregation"); |
| 3832 |
serialfn_oid = aggform->aggserialfn; |
| 3833 |
} |
| 3834 |
|
| 3835 |
/* Likewise for deserialization functions */ |
| 3836 |
if (DO_AGGSPLIT_DESERIALIZE(aggstate->aggsplit)) |
| 3837 |
{ |
| 3838 |
/* deserialization only valid when combining states */ |
| 3839 |
Assert(DO_AGGSPLIT_COMBINE(aggstate->aggsplit)); |
| 3840 |
|
| 3841 |
if (!OidIsValid(aggform->aggdeserialfn)) |
| 3842 |
elog(ERROR, "deserialfunc not provided for deserialization aggregation"); |
| 3843 |
deserialfn_oid = aggform->aggdeserialfn; |
| 3844 |
} |
| 3845 |
} |
| 3846 |
|
| 3847 |
/* Check that aggregate owner has permission to call component fns */ |
| 3848 |
{ |
| 3849 |
HeapTuple procTuple; |
| 3850 |
|
| 3851 |
procTuple = SearchSysCache1(PROCOID, |
| 3852 |
ObjectIdGetDatum(aggref->aggfnoid)); |
| 3853 |
if (!HeapTupleIsValid(procTuple)) |
| 3854 |
elog(ERROR, "cache lookup failed for function %u", |
| 3855 |
aggref->aggfnoid); |
| 3856 |
aggOwner = ((Form_pg_proc) GETSTRUCT(procTuple))->proowner; |
| 3857 |
ReleaseSysCache(procTuple); |
| 3858 |
|
| 3859 |
if (OidIsValid(finalfn_oid)) |
| 3860 |
{ |
| 3861 |
aclresult = object_aclcheck(ProcedureRelationId, finalfn_oid, aggOwner, |
| 3862 |
ACL_EXECUTE); |
| 3863 |
if (aclresult != ACLCHECK_OK) |
| 3864 |
aclcheck_error(aclresult, OBJECT_FUNCTION, |
| 3865 |
get_func_name(finalfn_oid)); |
| 3866 |
InvokeFunctionExecuteHook(finalfn_oid); |
| 3867 |
} |
| 3868 |
if (OidIsValid(serialfn_oid)) |
| 3869 |
{ |
| 3870 |
aclresult = object_aclcheck(ProcedureRelationId, serialfn_oid, aggOwner, |
| 3871 |
ACL_EXECUTE); |
| 3872 |
if (aclresult != ACLCHECK_OK) |
| 3873 |
aclcheck_error(aclresult, OBJECT_FUNCTION, |
| 3874 |
get_func_name(serialfn_oid)); |
| 3875 |
InvokeFunctionExecuteHook(serialfn_oid); |
| 3876 |
} |
| 3877 |
if (OidIsValid(deserialfn_oid)) |
| 3878 |
{ |
| 3879 |
aclresult = object_aclcheck(ProcedureRelationId, deserialfn_oid, aggOwner, |
| 3880 |
ACL_EXECUTE); |
| 3881 |
if (aclresult != ACLCHECK_OK) |
| 3882 |
aclcheck_error(aclresult, OBJECT_FUNCTION, |
| 3883 |
get_func_name(deserialfn_oid)); |
| 3884 |
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 |
*/ |
| 3893 |
numAggTransFnArgs = get_aggregate_argtypes(aggref, |
| 3894 |
aggTransFnInputTypes); |
| 3895 |
|
| 3896 |
/* Count the "direct" arguments, if any */ |
| 3897 |
numDirectArgs = list_length(aggref->aggdirectargs); |
| 3898 |
|
| 3899 |
/* Detect how many arguments to pass to the finalfn */ |
| 3900 |
if (aggform->aggfinalextra) |
| 3901 |
peragg->numFinalArgs = numAggTransFnArgs + 1; |
| 3902 |
else |
| 3903 |
peragg->numFinalArgs = numDirectArgs + 1; |
| 3904 |
|
| 3905 |
/* Initialize any direct-argument expressions */ |
| 3906 |
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 |
*/ |
| 3913 |
if (OidIsValid(finalfn_oid)) |
| 3914 |
{ |
| 3915 |
build_aggregate_finalfn_expr(aggTransFnInputTypes, |
| 3916 |
peragg->numFinalArgs, |
| 3917 |
aggtranstype, |
| 3918 |
aggref->aggtype, |
| 3919 |
aggref->inputcollid, |
| 3920 |
finalfn_oid, |
| 3921 |
&finalfnexpr); |
| 3922 |
fmgr_info(finalfn_oid, &peragg->finalfn); |
| 3923 |
fmgr_info_set_expr((Node *) finalfnexpr, &peragg->finalfn); |
| 3924 |
} |
| 3925 |
|
| 3926 |
/* get info about the output value's datatype */ |
| 3927 |
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 |
*/ |
| 3935 |
pertrans = &pertransstates[aggref->aggtransno]; |
| 3936 |
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 |
if (DO_AGGSPLIT_COMBINE(aggstate->aggsplit)) |
| 3949 |
{ |
| 3950 |
transfn_oid = aggform->aggcombinefn; |
| 3951 |
|
| 3952 |
/* If not set then the planner messed up */ |
| 3953 |
if (!OidIsValid(transfn_oid)) |
| 3954 |
elog(ERROR, "combinefn not set for aggregate function"); |
| 3955 |
} |
| 3956 |
else |
| 3957 |
transfn_oid = aggform->aggtransfn; |
| 3958 |
|
| 3959 |
aclresult = object_aclcheck(ProcedureRelationId, transfn_oid, aggOwner, ACL_EXECUTE); |
| 3960 |
if (aclresult != ACLCHECK_OK) |
| 3961 |
aclcheck_error(aclresult, OBJECT_FUNCTION, |
| 3962 |
get_func_name(transfn_oid)); |
| 3963 |
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 |
textInitVal = SysCacheGetAttr(AGGFNOID, aggTuple, |
| 3970 |
Anum_pg_aggregate_agginitval, |
| 3971 |
&initValueIsNull); |
| 3972 |
if (initValueIsNull) |
| 3973 |
initValue = (Datum) 0; |
| 3974 |
else |
| 3975 |
initValue = GetAggInitVal(textInitVal, aggtranstype); |
| 3976 |
|
| 3977 |
if (DO_AGGSPLIT_COMBINE(aggstate->aggsplit)) |
| 3978 |
{ |
| 3979 |
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 |
pertrans->numTransInputs = 1; |
| 3988 |
|
| 3989 |
/* aggcombinefn always has two arguments of aggtranstype */ |
| 3990 |
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 |
if (pertrans->transfn.fn_strict && aggtranstype == INTERNALOID) |
| 4003 |
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 */ |
| 4011 |
if (AGGKIND_IS_ORDERED_SET(aggref->aggkind)) |
| 4012 |
pertrans->numTransInputs = list_length(aggref->args); |
| 4013 |
else |
| 4014 |
pertrans->numTransInputs = numAggTransFnArgs; |
| 4015 |
|
| 4016 |
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 |
if (pertrans->transfn.fn_strict && pertrans->initValueIsNull) |
| 4033 |
{ |
| 4034 |
if (numAggTransFnArgs <= numDirectArgs || |
| 4035 |
!IsBinaryCoercible(aggTransFnInputTypes[numDirectArgs], |
| 4036 |
aggtranstype)) |
| 4037 |
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 |
| 4045 |
pertrans->aggshared = true; |
| 4046 |
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 |
*/ |
| 4059 |
if (numaggrefs != list_length(aggstate->aggs)) |
| 4060 |
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 |
*/ |
| 4071 |
for (phaseidx = 0; phaseidx < aggstate->numphases; phaseidx++) |
| 4072 |
{ |
| 4073 |
AggStatePerPhase phase = &aggstate->phases[phaseidx]; |
| 4074 |
bool dohash = false; |
| 4075 |
bool dosort = false; |
| 4076 |
|
| 4077 |
/* phase 0 doesn't necessarily exist */ |
| 4078 |
if (!phase->aggnode) |
| 4079 |
continue; |
| 4080 |
|
| 4081 |
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 |
dohash = true; |
| 4088 |
dosort = true; |
| 4089 |
} |
| 4090 |
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 |
continue; |
| 4098 |
} |
| 4099 |
else if (phase->aggstrategy == AGG_PLAIN || |
| 4100 |
phase->aggstrategy == AGG_SORTED) |
| 4101 |
{ |
| 4102 |
dohash = false; |
| 4103 |
dosort = true; |
| 4104 |
} |
| 4105 |
else if (phase->aggstrategy == AGG_HASHED) |
| 4106 |
{ |
| 4107 |
dohash = true; |
| 4108 |
dosort = false; |
| 4109 |
} |
| 4110 |
else |
| 4111 |
Assert(false); |
| 4112 |
|
| 4113 |
phase->evaltrans = ExecBuildAggTrans(aggstate, phase, dosort, dohash, |
| 4114 |
false); |
| 4115 |
|
| 4116 |
/* cache compiled expression for outer slot without NULL check */ |
| 4117 |
phase->evaltrans_cache[0][0] = phase->evaltrans; |
| 4118 |
} |
| 4119 |
|
| 4120 |
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 |
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 |
int numGroupingSets = Max(aggstate->maxsets, 1); |
| 4143 |
Expr *transfnexpr; |
| 4144 |
int numTransArgs; |
| 4145 |
Expr *serialfnexpr = NULL; |
| 4146 |
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 */ |
| 4156 |
pertrans->aggref = aggref; |
| 4157 |
pertrans->aggshared = false; |
| 4158 |
pertrans->aggCollation = aggref->inputcollid; |
| 4159 |
pertrans->transfn_oid = transfn_oid; |
| 4160 |
pertrans->serialfn_oid = aggserialfn; |
| 4161 |
pertrans->deserialfn_oid = aggdeserialfn; |
| 4162 |
pertrans->initValue = initValue; |
| 4163 |
pertrans->initValueIsNull = initValueIsNull; |
| 4164 |
|
| 4165 |
/* Count the "direct" arguments, if any */ |
| 4166 |
numDirectArgs = list_length(aggref->aggdirectargs); |
| 4167 |
|
| 4168 |
/* Count the number of aggregated input columns */ |
| 4169 |
pertrans->numInputs = numInputs = list_length(aggref->args); |
| 4170 |
|
| 4171 |
pertrans->aggtranstype = aggtranstype; |
| 4172 |
|
| 4173 |
/* account for the current transition state */ |
| 4174 |
numTransArgs = pertrans->numTransInputs + 1; |
| 4175 |
|
| 4176 |
/* |
| 4177 |
* Set up infrastructure for calling the transfn. Note that invtransfn is |
| 4178 |
* not needed here. |
| 4179 |
*/ |
| 4180 |
build_aggregate_transfn_expr(inputTypes, |
| 4181 |
numArguments, |
| 4182 |
numDirectArgs, |
| 4183 |
aggref->aggvariadic, |
| 4184 |
aggtranstype, |
| 4185 |
aggref->inputcollid, |
| 4186 |
transfn_oid, |
| 4187 |
InvalidOid, |
| 4188 |
&transfnexpr, |
| 4189 |
NULL); |
| 4190 |
|
| 4191 |
fmgr_info(transfn_oid, &pertrans->transfn); |
| 4192 |
fmgr_info_set_expr((Node *) transfnexpr, &pertrans->transfn); |
| 4193 |
|
| 4194 |
pertrans->transfn_fcinfo = |
| 4195 |
(FunctionCallInfo) palloc(SizeForFunctionCallInfo(numTransArgs)); |
| 4196 |
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 */ |
| 4203 |
get_typlenbyval(aggtranstype, |
| 4204 |
&pertrans->transtypeLen, |
| 4205 |
&pertrans->transtypeByVal); |
| 4206 |
|
| 4207 |
if (OidIsValid(aggserialfn)) |
| 4208 |
{ |
| 4209 |
build_aggregate_serialfn_expr(aggserialfn, |
| 4210 |
&serialfnexpr); |
| 4211 |
fmgr_info(aggserialfn, &pertrans->serialfn); |
| 4212 |
fmgr_info_set_expr((Node *) serialfnexpr, &pertrans->serialfn); |
| 4213 |
|
| 4214 |
pertrans->serialfn_fcinfo = |
| 4215 |
(FunctionCallInfo) palloc(SizeForFunctionCallInfo(1)); |
| 4216 |
InitFunctionCallInfoData(*pertrans->serialfn_fcinfo, |
| 4217 |
&pertrans->serialfn, |
| 4218 |
1, |
| 4219 |
InvalidOid, |
| 4220 |
(Node *) aggstate, NULL); |
| 4221 |
} |
| 4222 |
|
| 4223 |
if (OidIsValid(aggdeserialfn)) |
| 4224 |
{ |
| 4225 |
build_aggregate_deserialfn_expr(aggdeserialfn, |
| 4226 |
&deserialfnexpr); |
| 4227 |
fmgr_info(aggdeserialfn, &pertrans->deserialfn); |
| 4228 |
fmgr_info_set_expr((Node *) deserialfnexpr, &pertrans->deserialfn); |
| 4229 |
|
| 4230 |
pertrans->deserialfn_fcinfo = |
| 4231 |
(FunctionCallInfo) palloc(SizeForFunctionCallInfo(2)); |
| 4232 |
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 |
*/ |
| 4254 |
if (AGGKIND_IS_ORDERED_SET(aggref->aggkind)) |
| 4255 |
{ |
| 4256 |
sortlist = NIL; |
| 4257 |
numSortCols = numDistinctCols = 0; |
| 4258 |
pertrans->aggsortrequired = false; |
| 4259 |
} |
| 4260 |
else if (aggref->aggpresorted && aggref->aggdistinct == NIL) |
| 4261 |
{ |
| 4262 |
sortlist = NIL; |
| 4263 |
numSortCols = numDistinctCols = 0; |
| 4264 |
pertrans->aggsortrequired = false; |
| 4265 |
} |
| 4266 |
else if (aggref->aggdistinct) |
| 4267 |
{ |
| 4268 |
sortlist = aggref->aggdistinct; |
| 4269 |
numSortCols = numDistinctCols = list_length(sortlist); |
| 4270 |
Assert(numSortCols >= list_length(aggref->aggorder)); |
| 4271 |
pertrans->aggsortrequired = !aggref->aggpresorted; |
| 4272 |
} |
| 4273 |
else |
| 4274 |
{ |
| 4275 |
sortlist = aggref->aggorder; |
| 4276 |
numSortCols = list_length(sortlist); |
| 4277 |
numDistinctCols = 0; |
| 4278 |
pertrans->aggsortrequired = (numSortCols > 0); |
| 4279 |
} |
| 4280 |
|
| 4281 |
pertrans->numSortCols = numSortCols; |
| 4282 |
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 |
*/ |
| 4289 |
if (numSortCols > 0 || aggref->aggfilter) |
| 4290 |
{ |
| 4291 |
pertrans->sortdesc = ExecTypeFromTL(aggref->args); |
| 4292 |
pertrans->sortslot = |
| 4293 |
ExecInitExtraTupleSlot(estate, pertrans->sortdesc, |
| 4294 |
&TTSOpsMinimalTuple); |
| 4295 |
} |
| 4296 |
|
| 4297 |
if (numSortCols > 0) |
| 4298 |
{ |
| 4299 |
/* |
| 4300 |
* We don't implement DISTINCT or ORDER BY aggs in the HASHED case |
| 4301 |
* (yet) |
| 4302 |
*/ |
| 4303 |
Assert(aggstate->aggstrategy != AGG_HASHED && aggstate->aggstrategy != AGG_MIXED); |
| 4304 |
|
| 4305 |
/* ORDER BY aggregates are not supported with partial aggregation */ |
| 4306 |
Assert(!DO_AGGSPLIT_COMBINE(aggstate->aggsplit)); |
| 4307 |
|
| 4308 |
/* If we have only one input, we need its len/byval info. */ |
| 4309 |
if (numInputs == 1) |
| 4310 |
{ |
| 4311 |
get_typlenbyval(inputTypes[numDirectArgs], |
| 4312 |
&pertrans->inputtypeLen, |
| 4313 |
&pertrans->inputtypeByVal); |
| 4314 |
} |
| 4315 |
else if (numDistinctCols > 0) |
| 4316 |
{ |
| 4317 |
/* we will need an extra slot to store prior values */ |
| 4318 |
pertrans->uniqslot = |
| 4319 |
ExecInitExtraTupleSlot(estate, pertrans->sortdesc, |
| 4320 |
&TTSOpsMinimalTuple); |
| 4321 |
} |
| 4322 |
|
| 4323 |
/* Extract the sort information for use later */ |
| 4324 |
pertrans->sortColIdx = |
| 4325 |
(AttrNumber *) palloc(numSortCols * sizeof(AttrNumber)); |
| 4326 |
pertrans->sortOperators = |
| 4327 |
(Oid *) palloc(numSortCols * sizeof(Oid)); |
| 4328 |
pertrans->sortCollations = |
| 4329 |
(Oid *) palloc(numSortCols * sizeof(Oid)); |
| 4330 |
pertrans->sortNullsFirst = |
| 4331 |
(bool *) palloc(numSortCols * sizeof(bool)); |
| 4332 |
|
| 4333 |
i = 0; |
| 4334 |
foreach(lc, sortlist) |
| 4335 |
{ |
| 4336 |
SortGroupClause *sortcl = (SortGroupClause *) lfirst(lc); |
| 4337 |
TargetEntry *tle = get_sortgroupclause_tle(sortcl, aggref->args); |
| 4338 |
|
| 4339 |
/* the parser should have made sure of this */ |
| 4340 |
Assert(OidIsValid(sortcl->sortop)); |
| 4341 |
|
| 4342 |
pertrans->sortColIdx[i] = tle->resno; |
| 4343 |
pertrans->sortOperators[i] = sortcl->sortop; |
| 4344 |
pertrans->sortCollations[i] = exprCollation((Node *) tle->expr); |
| 4345 |
pertrans->sortNullsFirst[i] = sortcl->nulls_first; |
| 4346 |
i++; |
| 4347 |
} |
| 4348 |
Assert(i == numSortCols); |
| 4349 |
} |
| 4350 |
|
| 4351 |
if (aggref->aggdistinct) |
| 4352 |
{ |
| 4353 |
Oid *ops; |
| 4354 |
|
| 4355 |
Assert(numArguments > 0); |
| 4356 |
Assert(list_length(aggref->aggdistinct) == numDistinctCols); |
| 4357 |
|
| 4358 |
ops = palloc(numDistinctCols * sizeof(Oid)); |
| 4359 |
|
| 4360 |
i = 0; |
| 4361 |
foreach(lc, aggref->aggdistinct) |
| 4362 |
ops[i++] = ((SortGroupClause *) lfirst(lc))->eqop; |
| 4363 |
|
| 4364 |
/* lookup / build the necessary comparators */ |
| 4365 |
if (numDistinctCols == 1) |
| 4366 |
fmgr_info(get_opcode(ops[0]), &pertrans->equalfnOne); |
| 4367 |
else |
| 4368 |
pertrans->equalfnMulti = |
| 4369 |
execTuplesMatchPrepare(pertrans->sortdesc, |
| 4370 |
numDistinctCols, |
| 4371 |
pertrans->sortColIdx, |
| 4372 |
ops, |
| 4373 |
pertrans->sortCollations, |
| 4374 |
&aggstate->ss.ps); |
| 4375 |
pfree(ops); |
| 4376 |
} |
| 4377 |
|
| 4378 |
pertrans->sortstates = (Tuplesortstate **) |
| 4379 |
palloc0(sizeof(Tuplesortstate *) * numGroupingSets); |
| 4380 |
} |
| 4381 |
|
| 4382 |
|
| 4383 |
static Datum |
| 4384 |
GetAggInitVal(Datum textInitVal, Oid transtype) |
| 4385 |
{ |
| 4386 |
Oid typinput, |
| 4387 |
typioparam; |
| 4388 |
char *strInitVal; |
| 4389 |
Datum initVal; |
| 4390 |
|
| 4391 |
getTypeInputInfo(transtype, &typinput, &typioparam); |
| 4392 |
strInitVal = TextDatumGetCString(textInitVal); |
| 4393 |
initVal = OidInputFunctionCall(typinput, strInitVal, |
| 4394 |
typioparam, -1); |
| 4395 |
pfree(strInitVal); |
| 4396 |
return initVal; |
| 4397 |
} |
| 4398 |
|
| 4399 |
void |
| 4400 |
ExecEndAgg(AggState *node) |
| 4401 |
{ |
| 4402 |
PlanState *outerPlan; |
| 4403 |
int transno; |
| 4404 |
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 |
*/ |
| 4412 |
if (node->shared_info && IsParallelWorker()) |
| 4413 |
{ |
| 4414 |
AggregateInstrumentation *si; |
| 4415 |
|
| 4416 |
Assert(ParallelWorkerNumber < node->shared_info->num_workers); |
| 4417 |
si = &node->shared_info->sinstrument[ParallelWorkerNumber]; |
| 4418 |
si->hash_batches_used = node->hash_batches_used; |
| 4419 |
si->hash_disk_used = node->hash_disk_used; |
| 4420 |
si->hash_mem_peak = node->hash_mem_peak; |
| 4421 |
} |
| 4422 |
|
| 4423 |
/* Make sure we have closed any open tuplesorts */ |
| 4424 |
|
| 4425 |
if (node->sort_in) |
| 4426 |
tuplesort_end(node->sort_in); |
| 4427 |
if (node->sort_out) |
| 4428 |
tuplesort_end(node->sort_out); |
| 4429 |
|
| 4430 |
hashagg_reset_spill_state(node); |
| 4431 |
|
| 4432 |
if (node->hash_metacxt != NULL) |
| 4433 |
{ |
| 4434 |
MemoryContextDelete(node->hash_metacxt); |
| 4435 |
node->hash_metacxt = NULL; |
| 4436 |
} |
| 4437 |
if (node->hash_tablecxt != NULL) |
| 4438 |
{ |
| 4439 |
MemoryContextDelete(node->hash_tablecxt); |
| 4440 |
node->hash_tablecxt = NULL; |
| 4441 |
} |
| 4442 |
|
| 4443 |
|
| 4444 |
for (transno = 0; transno < node->numtrans; transno++) |
| 4445 |
{ |
| 4446 |
AggStatePerTrans pertrans = &node->pertrans[transno]; |
| 4447 |
|
| 4448 |
for (setno = 0; setno < numGroupingSets; setno++) |
| 4449 |
{ |
| 4450 |
if (pertrans->sortstates[setno]) |
| 4451 |
tuplesort_end(pertrans->sortstates[setno]); |
| 4452 |
} |
| 4453 |
} |
| 4454 |
|
| 4455 |
/* And ensure any agg shutdown callbacks have been called */ |
| 4456 |
for (setno = 0; setno < numGroupingSets; setno++) |
| 4457 |
ReScanExprContext(node->aggcontexts[setno]); |
| 4458 |
if (node->hashcontext) |
| 4459 |
ReScanExprContext(node->hashcontext); |
| 4460 |
|
| 4461 |
outerPlan = outerPlanState(node); |
| 4462 |
ExecEndNode(outerPlan); |
| 4463 |
} |
| 4464 |
|
| 4465 |
void |
| 4466 |
ExecReScanAgg(AggState *node) |
| 4467 |
{ |
| 4468 |
ExprContext *econtext = node->ss.ps.ps_ExprContext; |
| 4469 |
PlanState *outerPlan = outerPlanState(node); |
| 4470 |
Agg *aggnode = (Agg *) node->ss.ps.plan; |
| 4471 |
int transno; |
| 4472 |
int numGroupingSets = Max(node->maxsets, 1); |
| 4473 |
int setno; |
| 4474 |
|
| 4475 |
node->agg_done = false; |
| 4476 |
|
| 4477 |
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 |
*/ |
| 4485 |
if (!node->table_filled) |
| 4486 |
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 |
*/ |
| 4495 |
if (outerPlan->chgParam == NULL && !node->hash_ever_spilled && |
| 4496 |
!bms_overlap(node->ss.ps.chgParam, aggnode->aggParams)) |
| 4497 |
{ |
| 4498 |
ResetTupleHashIterator(node->perhash[0].hashtable, |
| 4499 |
&node->perhash[0].hashiter); |
| 4500 |
select_current_set(node, 0, true); |
| 4501 |
return; |
| 4502 |
} |
| 4503 |
} |
| 4504 |
|
| 4505 |
/* Make sure we have closed any open tuplesorts */ |
| 4506 |
for (transno = 0; transno < node->numtrans; transno++) |
| 4507 |
{ |
| 4508 |
for (setno = 0; setno < numGroupingSets; setno++) |
| 4509 |
{ |
| 4510 |
AggStatePerTrans pertrans = &node->pertrans[transno]; |
| 4511 |
|
| 4512 |
if (pertrans->sortstates[setno]) |
| 4513 |
{ |
| 4514 |
tuplesort_end(pertrans->sortstates[setno]); |
| 4515 |
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 |
|
| 4528 |
for (setno = 0; setno < numGroupingSets; setno++) |
| 4529 |
{ |
| 4530 |
ReScanExprContext(node->aggcontexts[setno]); |
| 4531 |
} |
| 4532 |
|
| 4533 |
/* Release first tuple of group, if we have made a copy */ |
| 4534 |
if (node->grp_firstTuple != NULL) |
| 4535 |
{ |
| 4536 |
heap_freetuple(node->grp_firstTuple); |
| 4537 |
node->grp_firstTuple = NULL; |
| 4538 |
} |
| 4539 |
ExecClearTuple(node->ss.ss_ScanTupleSlot); |
| 4540 |
|
| 4541 |
/* Forget current agg values */ |
| 4542 |
MemSet(econtext->ecxt_aggvalues, 0, sizeof(Datum) * node->numaggs); |
| 4543 |
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 |
*/ |
| 4550 |
if (node->aggstrategy == AGG_HASHED || node->aggstrategy == AGG_MIXED) |
| 4551 |
{ |
| 4552 |
hashagg_reset_spill_state(node); |
| 4553 |
|
| 4554 |
node->hash_ever_spilled = false; |
| 4555 |
node->hash_spill_mode = false; |
| 4556 |
node->hash_ngroups_current = 0; |
| 4557 |
|
| 4558 |
ReScanExprContext(node->hashcontext); |
| 4559 |
MemoryContextReset(node->hash_tablecxt); |
| 4560 |
/* Rebuild an empty hash table */ |
| 4561 |
build_hash_tables(node); |
| 4562 |
node->table_filled = false; |
| 4563 |
/* iterator will be reset when the table is filled */ |
| 4564 |
|
| 4565 |
hashagg_recompile_expressions(node, false, false); |
| 4566 |
} |
| 4567 |
|
| 4568 |
if (node->aggstrategy != AGG_HASHED) |
| 4569 |
{ |
| 4570 |
/* |
| 4571 |
* Reset the per-group state (in particular, mark transvalues null) |
| 4572 |
*/ |
| 4573 |
for (setno = 0; setno < numGroupingSets; setno++) |
| 4574 |
{ |
| 4575 |
MemSet(node->pergroups[setno], 0, |
| 4576 |
sizeof(AggStatePerGroupData) * node->numaggs); |
| 4577 |
} |
| 4578 |
|
| 4579 |
/* reset to phase 1 */ |
| 4580 |
initialize_phase(node, 1); |
| 4581 |
|
| 4582 |
node->input_done = false; |
| 4583 |
node->projected_set = -1; |
| 4584 |
} |
| 4585 |
|
| 4586 |
if (outerPlan->chgParam == NULL) |
| 4587 |
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 |
| 4614 |
AggCheckCallContext(FunctionCallInfo fcinfo, MemoryContext *aggcontext) |
| 4615 |
{ |
| 4616 |
if (fcinfo->context && IsA(fcinfo->context, AggState)) |
| 4617 |
{ |
| 4618 |
if (aggcontext) |
| 4619 |
{ |
| 4620 |
AggState *aggstate = ((AggState *) fcinfo->context); |
| 4621 |
ExprContext *cxt = aggstate->curaggcontext; |
| 4622 |
|
| 4623 |
*aggcontext = cxt->ecxt_per_tuple_memory; |
| 4624 |
} |
| 4625 |
return AGG_CONTEXT_AGGREGATE; |
| 4626 |
} |
| 4627 |
if (fcinfo->context && IsA(fcinfo->context, WindowAggState)) |
| 4628 |
{ |
| 4629 |
if (aggcontext) |
| 4630 |
*aggcontext = ((WindowAggState *) fcinfo->context)->curaggcontext; |
| 4631 |
return AGG_CONTEXT_WINDOW; |
| 4632 |
} |
| 4633 |
|
| 4634 |
/* this is just to prevent "uninitialized variable" warnings */ |
| 4635 |
if (aggcontext) |
| 4636 |
*aggcontext = NULL; |
| 4637 |
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 * |
| 4658 |
AggGetAggref(FunctionCallInfo fcinfo) |
| 4659 |
{ |
| 4660 |
if (fcinfo->context && IsA(fcinfo->context, AggState)) |
| 4661 |
{ |
| 4662 |
AggState *aggstate = (AggState *) fcinfo->context; |
| 4663 |
AggStatePerAgg curperagg; |
| 4664 |
AggStatePerTrans curpertrans; |
| 4665 |
|
| 4666 |
/* check curperagg (valid when in a final function) */ |
| 4667 |
curperagg = aggstate->curperagg; |
| 4668 |
|
| 4669 |
if (curperagg) |
| 4670 |
return curperagg->aggref; |
| 4671 |
|
| 4672 |
/* check curpertrans (valid when in a transition function) */ |
| 4673 |
curpertrans = aggstate->curpertrans; |
| 4674 |
|
| 4675 |
if (curpertrans) |
| 4676 |
return curpertrans->aggref; |
| 4677 |
} |
| 4678 |
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 |
| 4692 |
AggGetTempMemoryContext(FunctionCallInfo fcinfo) |
| 4693 |
{ |
| 4694 |
if (fcinfo->context && IsA(fcinfo->context, AggState)) |
| 4695 |
{ |
| 4696 |
AggState *aggstate = (AggState *) fcinfo->context; |
| 4697 |
|
| 4698 |
return aggstate->tmpcontext->ecxt_per_tuple_memory; |
| 4699 |
} |
| 4700 |
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 |
| 4718 |
AggStateIsShared(FunctionCallInfo fcinfo) |
| 4719 |
{ |
| 4720 |
if (fcinfo->context && IsA(fcinfo->context, AggState)) |
| 4721 |
{ |
| 4722 |
AggState *aggstate = (AggState *) fcinfo->context; |
| 4723 |
AggStatePerAgg curperagg; |
| 4724 |
AggStatePerTrans curpertrans; |
| 4725 |
|
| 4726 |
/* check curperagg (valid when in a final function) */ |
| 4727 |
curperagg = aggstate->curperagg; |
| 4728 |
|
| 4729 |
if (curperagg) |
| 4730 |
return aggstate->pertrans[curperagg->transno].aggshared; |
| 4731 |
|
| 4732 |
/* check curpertrans (valid when in a transition function) */ |
| 4733 |
curpertrans = aggstate->curpertrans; |
| 4734 |
|
| 4735 |
if (curpertrans) |
| 4736 |
return curpertrans->aggshared; |
| 4737 |
} |
| 4738 |
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 |
| 4757 |
AggRegisterCallback(FunctionCallInfo fcinfo, |
| 4758 |
ExprContextCallbackFunction func, |
| 4759 |
Datum arg) |
| 4760 |
{ |
| 4761 |
if (fcinfo->context && IsA(fcinfo->context, AggState)) |
| 4762 |
{ |
| 4763 |
AggState *aggstate = (AggState *) fcinfo->context; |
| 4764 |
ExprContext *cxt = aggstate->curaggcontext; |
| 4765 |
|
| 4766 |
RegisterExprContextCallback(cxt, func, arg); |
| 4767 |
|
| 4768 |
return; |
| 4769 |
} |
| 4770 |
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 |
| 4786 |
ExecAggEstimate(AggState *node, ParallelContext *pcxt) |
| 4787 |
{ |
| 4788 |
Size size; |
| 4789 |
|
| 4790 |
/* don't need this if not instrumenting or no workers */ |
| 4791 |
if (!node->ss.ps.instrument || pcxt->nworkers == 0) |
| 4792 |
return; |
| 4793 |
|
| 4794 |
size = mul_size(pcxt->nworkers, sizeof(AggregateInstrumentation)); |
| 4795 |
size = add_size(size, offsetof(SharedAggInfo, sinstrument)); |
| 4796 |
shm_toc_estimate_chunk(&pcxt->estimator, size); |
| 4797 |
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 |
ExecAggInitializeDSM(AggState *node, ParallelContext *pcxt) |
| 4808 |
{ |
| 4809 |
Size size; |
| 4810 |
|
| 4811 |
/* don't need this if not instrumenting or no workers */ |
| 4812 |
if (!node->ss.ps.instrument || pcxt->nworkers == 0) |
| 4813 |
return; |
| 4814 |
|
| 4815 |
size = offsetof(SharedAggInfo, sinstrument) |
| 4816 |
+ pcxt->nworkers * sizeof(AggregateInstrumentation); |
| 4817 |
node->shared_info = shm_toc_allocate(pcxt->toc, size); |
| 4818 |
/* ensure any unfilled slots will contain zeroes */ |
| 4819 |
memset(node->shared_info, 0, size); |
| 4820 |
node->shared_info->num_workers = pcxt->nworkers; |
| 4821 |
shm_toc_insert(pcxt->toc, node->ss.ps.plan->plan_node_id, |
| 4822 |
node->shared_info); |
| 4823 |
} |
| 4824 |
|
| 4825 |
/* ---------------------------------------------------------------- |
| 4826 |
* ExecAggInitializeWorker |
| 4827 |
* |
| 4828 |
* Attach worker to DSM space for aggregate statistics. |
| 4829 |
* ---------------------------------------------------------------- |
| 4830 |
*/ |
| 4831 |
void |
| 4832 |
ExecAggInitializeWorker(AggState *node, ParallelWorkerContext *pwcxt) |
| 4833 |
{ |
| 4834 |
node->shared_info = |
| 4835 |
shm_toc_lookup(pwcxt->toc, node->ss.ps.plan->plan_node_id, true); |
| 4836 |
} |
| 4837 |
|
| 4838 |
/* ---------------------------------------------------------------- |
| 4839 |
* ExecAggRetrieveInstrumentation |
| 4840 |
* |
| 4841 |
* Transfer aggregate statistics from DSM to private memory. |
| 4842 |
* ---------------------------------------------------------------- |
| 4843 |
*/ |
| 4844 |
void |
| 4845 |
ExecAggRetrieveInstrumentation(AggState *node) |
| 4846 |
{ |
| 4847 |
Size size; |
| 4848 |
SharedAggInfo *si; |
| 4849 |
|
| 4850 |
if (node->shared_info == NULL) |
| 4851 |
return; |
| 4852 |
|
| 4853 |
size = offsetof(SharedAggInfo, sinstrument) |
| 4854 |
+ node->shared_info->num_workers * sizeof(AggregateInstrumentation); |
| 4855 |
si = palloc(size); |
| 4856 |
memcpy(si, node->shared_info, size); |
| 4857 |
node->shared_info = si; |
| 4858 |
} |