| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* rangetypes_typanalyze.c |
| 4 |
* Functions for gathering statistics from range columns |
| 5 |
* |
| 6 |
* For a range type column, histograms of lower and upper bounds, and |
| 7 |
* the fraction of NULL and empty ranges are collected. |
| 8 |
* |
| 9 |
* Both histograms have the same length, and they are combined into a |
| 10 |
* single array of ranges. This has the same shape as the histogram that |
| 11 |
* std_typanalyze would collect, but the values are different. Each range |
| 12 |
* in the array is a valid range, even though the lower and upper bounds |
| 13 |
* come from different tuples. In theory, the standard scalar selectivity |
| 14 |
* functions could be used with the combined histogram. |
| 15 |
* |
| 16 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 17 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 18 |
* |
| 19 |
* |
| 20 |
* IDENTIFICATION |
| 21 |
* src/backend/utils/adt/rangetypes_typanalyze.c |
| 22 |
* |
| 23 |
*------------------------------------------------------------------------- |
| 24 |
*/ |
| 25 |
#include "postgres.h" |
| 26 |
|
| 27 |
#include "catalog/pg_operator.h" |
| 28 |
#include "commands/vacuum.h" |
| 29 |
#include "utils/float.h" |
| 30 |
#include "utils/fmgrprotos.h" |
| 31 |
#include "utils/lsyscache.h" |
| 32 |
#include "utils/multirangetypes.h" |
| 33 |
#include "utils/rangetypes.h" |
| 34 |
#include "varatt.h" |
| 35 |
|
| 36 |
static int float8_qsort_cmp(const void *a1, const void *a2, void *arg); |
| 37 |
static int range_bound_qsort_cmp(const void *a1, const void *a2, void *arg); |
| 38 |
static void compute_range_stats(VacAttrStats *stats, |
| 39 |
AnalyzeAttrFetchFunc fetchfunc, int samplerows, |
| 40 |
double totalrows); |
| 41 |
|
| 42 |
/* |
| 43 |
* range_typanalyze -- typanalyze function for range columns |
| 44 |
*/ |
| 45 |
Datum |
| 46 |
range_typanalyze(PG_FUNCTION_ARGS) |
| 47 |
{ |
| 48 |
VacAttrStats *stats = (VacAttrStats *) PG_GETARG_POINTER(0); |
| 49 |
TypeCacheEntry *typcache; |
| 50 |
|
| 51 |
/* Get information about range type; note column might be a domain */ |
| 52 |
typcache = range_get_typcache(fcinfo, getBaseType(stats->attrtypid)); |
| 53 |
|
| 54 |
if (stats->attstattarget < 0) |
| 55 |
stats->attstattarget = default_statistics_target; |
| 56 |
|
| 57 |
stats->compute_stats = compute_range_stats; |
| 58 |
stats->extra_data = typcache; |
| 59 |
/* same as in std_typanalyze */ |
| 60 |
stats->minrows = 300 * stats->attstattarget; |
| 61 |
|
| 62 |
PG_RETURN_BOOL(true); |
| 63 |
} |
| 64 |
|
| 65 |
/* |
| 66 |
* multirange_typanalyze -- typanalyze function for multirange columns |
| 67 |
* |
| 68 |
* We do the same analysis as for ranges, but on the smallest range that |
| 69 |
* completely includes the multirange. |
| 70 |
*/ |
| 71 |
Datum |
| 72 |
multirange_typanalyze(PG_FUNCTION_ARGS) |
| 73 |
{ |
| 74 |
VacAttrStats *stats = (VacAttrStats *) PG_GETARG_POINTER(0); |
| 75 |
TypeCacheEntry *typcache; |
| 76 |
|
| 77 |
/* Get information about multirange type; note column might be a domain */ |
| 78 |
typcache = multirange_get_typcache(fcinfo, getBaseType(stats->attrtypid)); |
| 79 |
|
| 80 |
if (stats->attstattarget < 0) |
| 81 |
stats->attstattarget = default_statistics_target; |
| 82 |
|
| 83 |
stats->compute_stats = compute_range_stats; |
| 84 |
stats->extra_data = typcache; |
| 85 |
/* same as in std_typanalyze */ |
| 86 |
stats->minrows = 300 * stats->attstattarget; |
| 87 |
|
| 88 |
PG_RETURN_BOOL(true); |
| 89 |
} |
| 90 |
|
| 91 |
/* |
| 92 |
* Comparison function for sorting float8s, used for range lengths. |
| 93 |
*/ |
| 94 |
static int |
| 95 |
float8_qsort_cmp(const void *a1, const void *a2, void *arg) |
| 96 |
{ |
| 97 |
const float8 *f1 = (const float8 *) a1; |
| 98 |
const float8 *f2 = (const float8 *) a2; |
| 99 |
|
| 100 |
if (*f1 < *f2) |
| 101 |
return -1; |
| 102 |
else if (*f1 == *f2) |
| 103 |
return 0; |
| 104 |
else |
| 105 |
return 1; |
| 106 |
} |
| 107 |
|
| 108 |
/* |
| 109 |
* Comparison function for sorting RangeBounds. |
| 110 |
*/ |
| 111 |
static int |
| 112 |
range_bound_qsort_cmp(const void *a1, const void *a2, void *arg) |
| 113 |
{ |
| 114 |
RangeBound *b1 = (RangeBound *) a1; |
| 115 |
RangeBound *b2 = (RangeBound *) a2; |
| 116 |
TypeCacheEntry *typcache = (TypeCacheEntry *) arg; |
| 117 |
|
| 118 |
return range_cmp_bounds(typcache, b1, b2); |
| 119 |
} |
| 120 |
|
| 121 |
/* |
| 122 |
* compute_range_stats() -- compute statistics for a range column |
| 123 |
*/ |
| 124 |
static void |
| 125 |
compute_range_stats(VacAttrStats *stats, AnalyzeAttrFetchFunc fetchfunc, |
| 126 |
int samplerows, double totalrows) |
| 127 |
{ |
| 128 |
TypeCacheEntry *typcache = (TypeCacheEntry *) stats->extra_data; |
| 129 |
TypeCacheEntry *mltrng_typcache = NULL; |
| 130 |
bool has_subdiff; |
| 131 |
int null_cnt = 0; |
| 132 |
int non_null_cnt = 0; |
| 133 |
int non_empty_cnt = 0; |
| 134 |
int empty_cnt = 0; |
| 135 |
int range_no; |
| 136 |
int slot_idx; |
| 137 |
int num_bins = stats->attstattarget; |
| 138 |
int num_hist; |
| 139 |
float8 *lengths; |
| 140 |
RangeBound *lowers, |
| 141 |
*uppers; |
| 142 |
double total_width = 0; |
| 143 |
|
| 144 |
if (typcache->typtype == TYPTYPE_MULTIRANGE) |
| 145 |
{ |
| 146 |
mltrng_typcache = typcache; |
| 147 |
typcache = typcache->rngtype; |
| 148 |
} |
| 149 |
else |
| 150 |
Assert(typcache->typtype == TYPTYPE_RANGE); |
| 151 |
has_subdiff = OidIsValid(typcache->rng_subdiff_finfo.fn_oid); |
| 152 |
|
| 153 |
/* Allocate memory to hold range bounds and lengths of the sample ranges. */ |
| 154 |
lowers = (RangeBound *) palloc(sizeof(RangeBound) * samplerows); |
| 155 |
uppers = (RangeBound *) palloc(sizeof(RangeBound) * samplerows); |
| 156 |
lengths = (float8 *) palloc(sizeof(float8) * samplerows); |
| 157 |
|
| 158 |
/* Loop over the sample ranges. */ |
| 159 |
for (range_no = 0; range_no < samplerows; range_no++) |
| 160 |
{ |
| 161 |
Datum value; |
| 162 |
bool isnull, |
| 163 |
empty; |
| 164 |
MultirangeType *multirange; |
| 165 |
RangeType *range; |
| 166 |
RangeBound lower, |
| 167 |
upper; |
| 168 |
float8 length; |
| 169 |
|
| 170 |
vacuum_delay_point(true); |
| 171 |
|
| 172 |
value = fetchfunc(stats, range_no, &isnull); |
| 173 |
if (isnull) |
| 174 |
{ |
| 175 |
/* range is null, just count that */ |
| 176 |
null_cnt++; |
| 177 |
continue; |
| 178 |
} |
| 179 |
|
| 180 |
/* |
| 181 |
* XXX: should we ignore wide values, like std_typanalyze does, to |
| 182 |
* avoid bloating the statistics table? |
| 183 |
*/ |
| 184 |
total_width += VARSIZE_ANY(DatumGetPointer(value)); |
| 185 |
|
| 186 |
/* Get range and deserialize it for further analysis. */ |
| 187 |
if (mltrng_typcache != NULL) |
| 188 |
{ |
| 189 |
/* Treat multiranges like a big range without gaps. */ |
| 190 |
multirange = DatumGetMultirangeTypeP(value); |
| 191 |
if (!MultirangeIsEmpty(multirange)) |
| 192 |
{ |
| 193 |
RangeBound tmp; |
| 194 |
|
| 195 |
multirange_get_bounds(typcache, multirange, 0, |
| 196 |
&lower, &tmp); |
| 197 |
multirange_get_bounds(typcache, multirange, |
| 198 |
multirange->rangeCount - 1, |
| 199 |
&tmp, &upper); |
| 200 |
empty = false; |
| 201 |
} |
| 202 |
else |
| 203 |
{ |
| 204 |
empty = true; |
| 205 |
} |
| 206 |
} |
| 207 |
else |
| 208 |
{ |
| 209 |
range = DatumGetRangeTypeP(value); |
| 210 |
range_deserialize(typcache, range, &lower, &upper, &empty); |
| 211 |
} |
| 212 |
|
| 213 |
if (!empty) |
| 214 |
{ |
| 215 |
/* Remember bounds and length for further usage in histograms */ |
| 216 |
lowers[non_empty_cnt] = lower; |
| 217 |
uppers[non_empty_cnt] = upper; |
| 218 |
|
| 219 |
if (lower.infinite || upper.infinite) |
| 220 |
{ |
| 221 |
/* Length of any kind of an infinite range is infinite */ |
| 222 |
length = get_float8_infinity(); |
| 223 |
} |
| 224 |
else if (has_subdiff) |
| 225 |
{ |
| 226 |
/* |
| 227 |
* For an ordinary range, use subdiff function between upper |
| 228 |
* and lower bound values. |
| 229 |
*/ |
| 230 |
length = DatumGetFloat8(FunctionCall2Coll(&typcache->rng_subdiff_finfo, |
| 231 |
typcache->rng_collation, |
| 232 |
upper.val, lower.val)); |
| 233 |
} |
| 234 |
else |
| 235 |
{ |
| 236 |
/* Use default value of 1.0 if no subdiff is available. */ |
| 237 |
length = 1.0; |
| 238 |
} |
| 239 |
lengths[non_empty_cnt] = length; |
| 240 |
|
| 241 |
non_empty_cnt++; |
| 242 |
} |
| 243 |
else |
| 244 |
empty_cnt++; |
| 245 |
|
| 246 |
non_null_cnt++; |
| 247 |
} |
| 248 |
|
| 249 |
slot_idx = 0; |
| 250 |
|
| 251 |
/* We can only compute real stats if we found some non-null values. */ |
| 252 |
if (non_null_cnt > 0) |
| 253 |
{ |
| 254 |
Datum *bound_hist_values; |
| 255 |
Datum *length_hist_values; |
| 256 |
int pos, |
| 257 |
posfrac, |
| 258 |
delta, |
| 259 |
deltafrac, |
| 260 |
i; |
| 261 |
MemoryContext old_cxt; |
| 262 |
float4 *emptyfrac; |
| 263 |
|
| 264 |
stats->stats_valid = true; |
| 265 |
/* Do the simple null-frac and width stats */ |
| 266 |
stats->stanullfrac = (double) null_cnt / (double) samplerows; |
| 267 |
stats->stawidth = total_width / (double) non_null_cnt; |
| 268 |
|
| 269 |
/* Estimate that non-null values are unique */ |
| 270 |
stats->stadistinct = -1.0 * (1.0 - stats->stanullfrac); |
| 271 |
|
| 272 |
/* Must copy the target values into anl_context */ |
| 273 |
old_cxt = MemoryContextSwitchTo(stats->anl_context); |
| 274 |
|
| 275 |
/* |
| 276 |
* Generate a bounds histogram slot entry if there are at least two |
| 277 |
* values. |
| 278 |
*/ |
| 279 |
if (non_empty_cnt >= 2) |
| 280 |
{ |
| 281 |
/* Sort bound values */ |
| 282 |
qsort_interruptible(lowers, non_empty_cnt, sizeof(RangeBound), |
| 283 |
range_bound_qsort_cmp, typcache); |
| 284 |
qsort_interruptible(uppers, non_empty_cnt, sizeof(RangeBound), |
| 285 |
range_bound_qsort_cmp, typcache); |
| 286 |
|
| 287 |
num_hist = non_empty_cnt; |
| 288 |
if (num_hist > num_bins) |
| 289 |
num_hist = num_bins + 1; |
| 290 |
|
| 291 |
bound_hist_values = (Datum *) palloc(num_hist * sizeof(Datum)); |
| 292 |
|
| 293 |
/* |
| 294 |
* The object of this loop is to construct ranges from first and |
| 295 |
* last entries in lowers[] and uppers[] along with evenly-spaced |
| 296 |
* values in between. So the i'th value is a range of lowers[(i * |
| 297 |
* (nvals - 1)) / (num_hist - 1)] and uppers[(i * (nvals - 1)) / |
| 298 |
* (num_hist - 1)]. But computing that subscript directly risks |
| 299 |
* integer overflow when the stats target is more than a couple |
| 300 |
* thousand. Instead we add (nvals - 1) / (num_hist - 1) to pos |
| 301 |
* at each step, tracking the integral and fractional parts of the |
| 302 |
* sum separately. |
| 303 |
*/ |
| 304 |
delta = (non_empty_cnt - 1) / (num_hist - 1); |
| 305 |
deltafrac = (non_empty_cnt - 1) % (num_hist - 1); |
| 306 |
pos = posfrac = 0; |
| 307 |
|
| 308 |
for (i = 0; i < num_hist; i++) |
| 309 |
{ |
| 310 |
bound_hist_values[i] = PointerGetDatum(range_serialize(typcache, |
| 311 |
&lowers[pos], |
| 312 |
&uppers[pos], |
| 313 |
false, |
| 314 |
NULL)); |
| 315 |
pos += delta; |
| 316 |
posfrac += deltafrac; |
| 317 |
if (posfrac >= (num_hist - 1)) |
| 318 |
{ |
| 319 |
/* fractional part exceeds 1, carry to integer part */ |
| 320 |
pos++; |
| 321 |
posfrac -= (num_hist - 1); |
| 322 |
} |
| 323 |
} |
| 324 |
|
| 325 |
stats->stakind[slot_idx] = STATISTIC_KIND_BOUNDS_HISTOGRAM; |
| 326 |
stats->stavalues[slot_idx] = bound_hist_values; |
| 327 |
stats->numvalues[slot_idx] = num_hist; |
| 328 |
|
| 329 |
/* Store ranges even if we're analyzing a multirange column */ |
| 330 |
stats->statypid[slot_idx] = typcache->type_id; |
| 331 |
stats->statyplen[slot_idx] = typcache->typlen; |
| 332 |
stats->statypbyval[slot_idx] = typcache->typbyval; |
| 333 |
stats->statypalign[slot_idx] = typcache->typalign; |
| 334 |
|
| 335 |
slot_idx++; |
| 336 |
} |
| 337 |
|
| 338 |
/* |
| 339 |
* Generate a length histogram slot entry if there are at least two |
| 340 |
* values. |
| 341 |
*/ |
| 342 |
if (non_empty_cnt >= 2) |
| 343 |
{ |
| 344 |
/* |
| 345 |
* Ascending sort of range lengths for further filling of |
| 346 |
* histogram |
| 347 |
*/ |
| 348 |
qsort_interruptible(lengths, non_empty_cnt, sizeof(float8), |
| 349 |
float8_qsort_cmp, NULL); |
| 350 |
|
| 351 |
num_hist = non_empty_cnt; |
| 352 |
if (num_hist > num_bins) |
| 353 |
num_hist = num_bins + 1; |
| 354 |
|
| 355 |
length_hist_values = (Datum *) palloc(num_hist * sizeof(Datum)); |
| 356 |
|
| 357 |
/* |
| 358 |
* The object of this loop is to copy the first and last lengths[] |
| 359 |
* entries along with evenly-spaced values in between. So the i'th |
| 360 |
* value is lengths[(i * (nvals - 1)) / (num_hist - 1)]. But |
| 361 |
* computing that subscript directly risks integer overflow when |
| 362 |
* the stats target is more than a couple thousand. Instead we |
| 363 |
* add (nvals - 1) / (num_hist - 1) to pos at each step, tracking |
| 364 |
* the integral and fractional parts of the sum separately. |
| 365 |
*/ |
| 366 |
delta = (non_empty_cnt - 1) / (num_hist - 1); |
| 367 |
deltafrac = (non_empty_cnt - 1) % (num_hist - 1); |
| 368 |
pos = posfrac = 0; |
| 369 |
|
| 370 |
for (i = 0; i < num_hist; i++) |
| 371 |
{ |
| 372 |
length_hist_values[i] = Float8GetDatum(lengths[pos]); |
| 373 |
pos += delta; |
| 374 |
posfrac += deltafrac; |
| 375 |
if (posfrac >= (num_hist - 1)) |
| 376 |
{ |
| 377 |
/* fractional part exceeds 1, carry to integer part */ |
| 378 |
pos++; |
| 379 |
posfrac -= (num_hist - 1); |
| 380 |
} |
| 381 |
} |
| 382 |
} |
| 383 |
else |
| 384 |
{ |
| 385 |
/* |
| 386 |
* Even when we don't create the histogram, store an empty array |
| 387 |
* to mean "no histogram". We can't just leave stavalues NULL, |
| 388 |
* because get_attstatsslot() errors if you ask for stavalues, and |
| 389 |
* it's NULL. We'll still store the empty fraction in stanumbers. |
| 390 |
*/ |
| 391 |
length_hist_values = palloc(0); |
| 392 |
num_hist = 0; |
| 393 |
} |
| 394 |
stats->staop[slot_idx] = Float8LessOperator; |
| 395 |
stats->stacoll[slot_idx] = InvalidOid; |
| 396 |
stats->stavalues[slot_idx] = length_hist_values; |
| 397 |
stats->numvalues[slot_idx] = num_hist; |
| 398 |
stats->statypid[slot_idx] = FLOAT8OID; |
| 399 |
stats->statyplen[slot_idx] = sizeof(float8); |
| 400 |
stats->statypbyval[slot_idx] = FLOAT8PASSBYVAL; |
| 401 |
stats->statypalign[slot_idx] = 'd'; |
| 402 |
|
| 403 |
/* Store the fraction of empty ranges */ |
| 404 |
emptyfrac = (float4 *) palloc(sizeof(float4)); |
| 405 |
*emptyfrac = ((double) empty_cnt) / ((double) non_null_cnt); |
| 406 |
stats->stanumbers[slot_idx] = emptyfrac; |
| 407 |
stats->numnumbers[slot_idx] = 1; |
| 408 |
|
| 409 |
stats->stakind[slot_idx] = STATISTIC_KIND_RANGE_LENGTH_HISTOGRAM; |
| 410 |
slot_idx++; |
| 411 |
|
| 412 |
MemoryContextSwitchTo(old_cxt); |
| 413 |
} |
| 414 |
else if (null_cnt > 0) |
| 415 |
{ |
| 416 |
/* We found only nulls; assume the column is entirely null */ |
| 417 |
stats->stats_valid = true; |
| 418 |
stats->stanullfrac = 1.0; |
| 419 |
stats->stawidth = 0; /* "unknown" */ |
| 420 |
stats->stadistinct = 0.0; /* "unknown" */ |
| 421 |
} |
| 422 |
|
| 423 |
/* |
| 424 |
* We don't need to bother cleaning up any of our temporary palloc's. The |
| 425 |
* hashtable should also go away, as it used a child memory context. |
| 426 |
*/ |
| 427 |
} |