1
/*-------------------------------------------------------------------------
3
* ragetypes_typanalyze.c
4
* Functions for gathering statistics from range columns
6
* For a range type column, histograms of lower and upper bounds, and
7
* the fraction of NULL and empty ranges are collected.
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.
16
* Portions Copyright (c) 1996-2013, PostgreSQL Global Development Group
17
* Portions Copyright (c) 1994, Regents of the University of California
21
* src/backend/utils/adt/rangetypes_typanalyze.c
23
*-------------------------------------------------------------------------
27
#include "catalog/pg_operator.h"
28
#include "commands/vacuum.h"
29
#include "utils/builtins.h"
30
#include "utils/rangetypes.h"
32
static int float8_qsort_cmp(const void *a1, const void *a2);
33
static int range_bound_qsort_cmp(const void *a1, const void *a2, void *arg);
34
static void compute_range_stats(VacAttrStats *stats,
35
AnalyzeAttrFetchFunc fetchfunc, int samplerows, double totalrows);
38
* range_typanalyze -- typanalyze function for range columns
41
range_typanalyze(PG_FUNCTION_ARGS)
43
VacAttrStats *stats = (VacAttrStats *) PG_GETARG_POINTER(0);
44
TypeCacheEntry *typcache;
45
Form_pg_attribute attr = stats->attr;
47
/* Get information about range type */
48
typcache = range_get_typcache(fcinfo, stats->attrtypid);
50
if (attr->attstattarget < 0)
51
attr->attstattarget = default_statistics_target;
53
stats->compute_stats = compute_range_stats;
54
stats->extra_data = typcache;
55
/* same as in std_typanalyze */
56
stats->minrows = 300 * attr->attstattarget;
62
* Comparison function for sorting float8s, used for range lengths.
65
float8_qsort_cmp(const void *a1, const void *a2)
67
const float8 *f1 = (const float8 *) a1;
68
const float8 *f2 = (const float8 *) a2;
79
* Comparison function for sorting RangeBounds.
82
range_bound_qsort_cmp(const void *a1, const void *a2, void *arg)
84
RangeBound *b1 = (RangeBound *)a1;
85
RangeBound *b2 = (RangeBound *)a2;
86
TypeCacheEntry *typcache = (TypeCacheEntry *)arg;
88
return range_cmp_bounds(typcache, b1, b2);
92
* compute_range_stats() -- compute statistics for a range column
95
compute_range_stats(VacAttrStats *stats, AnalyzeAttrFetchFunc fetchfunc,
96
int samplerows, double totalrows)
98
TypeCacheEntry *typcache = (TypeCacheEntry *) stats->extra_data;
99
bool has_subdiff = OidIsValid(typcache->rng_subdiff_finfo.fn_oid);
101
int non_null_cnt = 0;
102
int non_empty_cnt = 0;
106
int num_bins = stats->attr->attstattarget;
109
RangeBound *lowers, *uppers;
110
double total_width = 0;
112
/* Allocate memory to hold range bounds and lengths of the sample ranges. */
113
lowers = (RangeBound *) palloc(sizeof(RangeBound) * samplerows);
114
uppers = (RangeBound *) palloc(sizeof(RangeBound) * samplerows);
115
lengths = (float8 *) palloc(sizeof(float8) * samplerows);
117
/* Loop over the sample ranges. */
118
for (range_no = 0; range_no < samplerows; range_no++)
128
vacuum_delay_point();
130
value = fetchfunc(stats, range_no, &isnull);
133
/* range is null, just count that */
139
* XXX: should we ignore wide values, like std_typanalyze does, to
140
* avoid bloating the statistics table?
142
total_width += VARSIZE_ANY(DatumGetPointer(value));
144
/* Get range and deserialize it for further analysis. */
145
range = DatumGetRangeType(value);
146
range_deserialize(typcache, range, &lower, &upper, &empty);
150
/* Remember bounds and length for further usage in histograms */
151
lowers[non_empty_cnt] = lower;
152
uppers[non_empty_cnt] = upper;
154
if (lower.infinite || upper.infinite)
156
/* Length of any kind of an infinite range is infinite */
157
length = get_float8_infinity();
159
else if (has_subdiff)
162
* For an ordinary range, use subdiff function between upper
163
* and lower bound values.
165
length = DatumGetFloat8(FunctionCall2Coll(
166
&typcache->rng_subdiff_finfo,
167
typcache->rng_collation,
168
upper.val, lower.val));
172
/* Use default value of 1.0 if no subdiff is available. */
175
lengths[non_empty_cnt] = length;
187
/* We can only compute real stats if we found some non-null values. */
188
if (non_null_cnt > 0)
190
Datum *bound_hist_values;
191
Datum *length_hist_values;
197
MemoryContext old_cxt;
200
stats->stats_valid = true;
201
/* Do the simple null-frac and width stats */
202
stats->stanullfrac = (double) null_cnt / (double) samplerows;
203
stats->stawidth = total_width / (double) non_null_cnt;
204
stats->stadistinct = -1.0;
206
/* Must copy the target values into anl_context */
207
old_cxt = MemoryContextSwitchTo(stats->anl_context);
210
* Generate a bounds histogram slot entry if there are at least two
213
if (non_empty_cnt >= 2)
215
/* Sort bound values */
216
qsort_arg(lowers, non_empty_cnt, sizeof(RangeBound),
217
range_bound_qsort_cmp, typcache);
218
qsort_arg(uppers, non_empty_cnt, sizeof(RangeBound),
219
range_bound_qsort_cmp, typcache);
221
num_hist = non_empty_cnt;
222
if (num_hist > num_bins)
223
num_hist = num_bins + 1;
225
bound_hist_values = (Datum *) palloc(num_hist * sizeof(Datum));
228
* The object of this loop is to construct ranges from first and
229
* last entries in lowers[] and uppers[] along with evenly-spaced
230
* values in between. So the i'th value is a range of
231
* lowers[(i * (nvals - 1)) / (num_hist - 1)] and
232
* uppers[(i * (nvals - 1)) / (num_hist - 1)]. But computing that
233
* subscript directly risks integer overflow when the stats target
234
* is more than a couple thousand. Instead we add
235
* (nvals - 1) / (num_hist - 1) to pos at each step, tracking the
236
* integral and fractional parts of the sum separately.
238
delta = (non_empty_cnt - 1) / (num_hist - 1);
239
deltafrac = (non_empty_cnt - 1) % (num_hist - 1);
242
for (i = 0; i < num_hist; i++)
244
bound_hist_values[i] = PointerGetDatum(range_serialize(
245
typcache, &lowers[pos], &uppers[pos], false));
247
posfrac += deltafrac;
248
if (posfrac >= (num_hist - 1))
250
/* fractional part exceeds 1, carry to integer part */
252
posfrac -= (num_hist - 1);
256
stats->stakind[slot_idx] = STATISTIC_KIND_BOUNDS_HISTOGRAM;
257
stats->stavalues[slot_idx] = bound_hist_values;
258
stats->numvalues[slot_idx] = num_hist;
263
* Generate a length histogram slot entry if there are at least two
266
if (non_empty_cnt >= 2)
269
* Ascending sort of range lengths for further filling of
272
qsort(lengths, non_empty_cnt, sizeof(float8), float8_qsort_cmp);
274
num_hist = non_empty_cnt;
275
if (num_hist > num_bins)
276
num_hist = num_bins + 1;
278
length_hist_values = (Datum *) palloc(num_hist * sizeof(Datum));
281
* The object of this loop is to copy the first and last lengths[]
282
* entries along with evenly-spaced values in between. So the i'th
283
* value is lengths[(i * (nvals - 1)) / (num_hist - 1)]. But
284
* computing that subscript directly risks integer overflow when the
285
* stats target is more than a couple thousand. Instead we add
286
* (nvals - 1) / (num_hist - 1) to pos at each step, tracking the
287
* integral and fractional parts of the sum separately.
289
delta = (non_empty_cnt - 1) / (num_hist - 1);
290
deltafrac = (non_empty_cnt - 1) % (num_hist - 1);
293
for (i = 0; i < num_hist; i++)
295
length_hist_values[i] = Float8GetDatum(lengths[pos]);
297
posfrac += deltafrac;
298
if (posfrac >= (num_hist - 1))
300
/* fractional part exceeds 1, carry to integer part */
302
posfrac -= (num_hist - 1);
309
* Even when we don't create the histogram, store an empty array
310
* to mean "no histogram". We can't just leave stavalues NULL,
311
* because get_attstatsslot() errors if you ask for stavalues, and
312
* it's NULL. We'll still store the empty fraction in stanumbers.
314
length_hist_values = palloc(0);
317
stats->staop[slot_idx] = Float8LessOperator;
318
stats->stavalues[slot_idx] = length_hist_values;
319
stats->numvalues[slot_idx] = num_hist;
320
stats->statypid[slot_idx] = FLOAT8OID;
321
stats->statyplen[slot_idx] = sizeof(float8);
322
#ifdef USE_FLOAT8_BYVAL
323
stats->statypbyval[slot_idx] = true;
325
stats->statypbyval[slot_idx] = false;
327
stats->statypalign[slot_idx] = 'd';
329
/* Store the fraction of empty ranges */
330
emptyfrac = (float4 *) palloc(sizeof(float4));
331
*emptyfrac = ((double) empty_cnt) / ((double) non_null_cnt);
332
stats->stanumbers[slot_idx] = emptyfrac;
333
stats->numnumbers[slot_idx] = 1;
335
stats->stakind[slot_idx] = STATISTIC_KIND_RANGE_LENGTH_HISTOGRAM;
338
MemoryContextSwitchTo(old_cxt);
340
else if (null_cnt > 0)
342
/* We found only nulls; assume the column is entirely null */
343
stats->stats_valid = true;
344
stats->stanullfrac = 1.0;
345
stats->stawidth = 0; /* "unknown" */
346
stats->stadistinct = 0.0; /* "unknown" */
349
* We don't need to bother cleaning up any of our temporary palloc's. The
350
* hashtable should also go away, as it used a child memory context.