summaryrefslogtreecommitdiffstats
path: root/contrib/tsm_system_time/tsm_system_time.c
blob: 2fda572e6522d686cdae4032297c9efa1cf52449 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
/*-------------------------------------------------------------------------
 *
 * tsm_system_time.c
 *	  support routines for SYSTEM_TIME tablesample method
 *
 * The desire here is to produce a random sample with as many rows as possible
 * in no more than the specified amount of time.  We use a block-sampling
 * approach.  To ensure that the whole relation will be visited if necessary,
 * we start at a randomly chosen block and then advance with a stride that
 * is randomly chosen but is relatively prime to the relation's nblocks.
 *
 * Because of the time dependence, this method is necessarily unrepeatable.
 * However, we do what we can to reduce surprising behavior by selecting
 * the sampling pattern just once per query, much as in tsm_system_rows.
 *
 * Portions Copyright (c) 1996-2020, PostgreSQL Global Development Group
 * Portions Copyright (c) 1994, Regents of the University of California
 *
 * IDENTIFICATION
 *	  contrib/tsm_system_time/tsm_system_time.c
 *
 *-------------------------------------------------------------------------
 */

#include "postgres.h"

#include <math.h>

#include "access/relscan.h"
#include "access/tsmapi.h"
#include "catalog/pg_type.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "utils/sampling.h"
#include "utils/spccache.h"

PG_MODULE_MAGIC;

PG_FUNCTION_INFO_V1(tsm_system_time_handler);


/* Private state */
typedef struct
{
	uint32		seed;			/* random seed */
	double		millis;			/* time limit for sampling */
	instr_time	start_time;		/* scan start time */
	OffsetNumber lt;			/* last tuple returned from current block */
	BlockNumber doneblocks;		/* number of already-scanned blocks */
	BlockNumber lb;				/* last block visited */
	/* these three values are not changed during a rescan: */
	BlockNumber nblocks;		/* number of blocks in relation */
	BlockNumber firstblock;		/* first block to sample from */
	BlockNumber step;			/* step size, or 0 if not set yet */
} SystemTimeSamplerData;

static void system_time_samplescangetsamplesize(PlannerInfo *root,
												RelOptInfo *baserel,
												List *paramexprs,
												BlockNumber *pages,
												double *tuples);
static void system_time_initsamplescan(SampleScanState *node,
									   int eflags);
static void system_time_beginsamplescan(SampleScanState *node,
										Datum *params,
										int nparams,
										uint32 seed);
static BlockNumber system_time_nextsampleblock(SampleScanState *node, BlockNumber nblocks);
static OffsetNumber system_time_nextsampletuple(SampleScanState *node,
												BlockNumber blockno,
												OffsetNumber maxoffset);
static uint32 random_relative_prime(uint32 n, SamplerRandomState randstate);


/*
 * Create a TsmRoutine descriptor for the SYSTEM_TIME method.
 */
Datum
tsm_system_time_handler(PG_FUNCTION_ARGS)
{
	TsmRoutine *tsm = makeNode(TsmRoutine);

	tsm->parameterTypes = list_make1_oid(FLOAT8OID);

	/* See notes at head of file */
	tsm->repeatable_across_queries = false;
	tsm->repeatable_across_scans = false;

	tsm->SampleScanGetSampleSize = system_time_samplescangetsamplesize;
	tsm->InitSampleScan = system_time_initsamplescan;
	tsm->BeginSampleScan = system_time_beginsamplescan;
	tsm->NextSampleBlock = system_time_nextsampleblock;
	tsm->NextSampleTuple = system_time_nextsampletuple;
	tsm->EndSampleScan = NULL;

	PG_RETURN_POINTER(tsm);
}

/*
 * Sample size estimation.
 */
static void
system_time_samplescangetsamplesize(PlannerInfo *root,
									RelOptInfo *baserel,
									List *paramexprs,
									BlockNumber *pages,
									double *tuples)
{
	Node	   *limitnode;
	double		millis;
	double		spc_random_page_cost;
	double		npages;
	double		ntuples;

	/* Try to extract an estimate for the limit time spec */
	limitnode = (Node *) linitial(paramexprs);
	limitnode = estimate_expression_value(root, limitnode);

	if (IsA(limitnode, Const) &&
		!((Const *) limitnode)->constisnull)
	{
		millis = DatumGetFloat8(((Const *) limitnode)->constvalue);
		if (millis < 0 || isnan(millis))
		{
			/* Default millis if the value is bogus */
			millis = 1000;
		}
	}
	else
	{
		/* Default millis if we didn't obtain a non-null Const */
		millis = 1000;
	}

	/* Get the planner's idea of cost per page read */
	get_tablespace_page_costs(baserel->reltablespace,
							  &spc_random_page_cost,
							  NULL);

	/*
	 * Estimate the number of pages we can read by assuming that the cost
	 * figure is expressed in milliseconds.  This is completely, unmistakably
	 * bogus, but we have to do something to produce an estimate and there's
	 * no better answer.
	 */
	if (spc_random_page_cost > 0)
		npages = millis / spc_random_page_cost;
	else
		npages = millis;		/* even more bogus, but whatcha gonna do? */

	/* Clamp to sane value */
	npages = clamp_row_est(Min((double) baserel->pages, npages));

	if (baserel->tuples > 0 && baserel->pages > 0)
	{
		/* Estimate number of tuples returned based on tuple density */
		double		density = baserel->tuples / (double) baserel->pages;

		ntuples = npages * density;
	}
	else
	{
		/* For lack of data, assume one tuple per page */
		ntuples = npages;
	}

	/* Clamp to the estimated relation size */
	ntuples = clamp_row_est(Min(baserel->tuples, ntuples));

	*pages = npages;
	*tuples = ntuples;
}

/*
 * Initialize during executor setup.
 */
static void
system_time_initsamplescan(SampleScanState *node, int eflags)
{
	node->tsm_state = palloc0(sizeof(SystemTimeSamplerData));
	/* Note the above leaves tsm_state->step equal to zero */
}

/*
 * Examine parameters and prepare for a sample scan.
 */
static void
system_time_beginsamplescan(SampleScanState *node,
							Datum *params,
							int nparams,
							uint32 seed)
{
	SystemTimeSamplerData *sampler = (SystemTimeSamplerData *) node->tsm_state;
	double		millis = DatumGetFloat8(params[0]);

	if (millis < 0 || isnan(millis))
		ereport(ERROR,
				(errcode(ERRCODE_INVALID_TABLESAMPLE_ARGUMENT),
				 errmsg("sample collection time must not be negative")));

	sampler->seed = seed;
	sampler->millis = millis;
	sampler->lt = InvalidOffsetNumber;
	sampler->doneblocks = 0;
	/* start_time, lb will be initialized during first NextSampleBlock call */
	/* we intentionally do not change nblocks/firstblock/step here */
}

/*
 * Select next block to sample.
 *
 * Uses linear probing algorithm for picking next block.
 */
static BlockNumber
system_time_nextsampleblock(SampleScanState *node, BlockNumber nblocks)
{
	SystemTimeSamplerData *sampler = (SystemTimeSamplerData *) node->tsm_state;
	instr_time	cur_time;

	/* First call within scan? */
	if (sampler->doneblocks == 0)
	{
		/* First scan within query? */
		if (sampler->step == 0)
		{
			/* Initialize now that we have scan descriptor */
			SamplerRandomState randstate;

			/* If relation is empty, there's nothing to scan */
			if (nblocks == 0)
				return InvalidBlockNumber;

			/* We only need an RNG during this setup step */
			sampler_random_init_state(sampler->seed, randstate);

			/* Compute nblocks/firstblock/step only once per query */
			sampler->nblocks = nblocks;

			/* Choose random starting block within the relation */
			/* (Actually this is the predecessor of the first block visited) */
			sampler->firstblock = sampler_random_fract(randstate) *
				sampler->nblocks;

			/* Find relative prime as step size for linear probing */
			sampler->step = random_relative_prime(sampler->nblocks, randstate);
		}

		/* Reinitialize lb and start_time */
		sampler->lb = sampler->firstblock;
		INSTR_TIME_SET_CURRENT(sampler->start_time);
	}

	/* If we've read all blocks in relation, we're done */
	if (++sampler->doneblocks > sampler->nblocks)
		return InvalidBlockNumber;

	/* If we've used up all the allotted time, we're done */
	INSTR_TIME_SET_CURRENT(cur_time);
	INSTR_TIME_SUBTRACT(cur_time, sampler->start_time);
	if (INSTR_TIME_GET_MILLISEC(cur_time) >= sampler->millis)
		return InvalidBlockNumber;

	/*
	 * It's probably impossible for scan->rs_nblocks to decrease between scans
	 * within a query; but just in case, loop until we select a block number
	 * less than scan->rs_nblocks.  We don't care if scan->rs_nblocks has
	 * increased since the first scan.
	 */
	do
	{
		/* Advance lb, using uint64 arithmetic to forestall overflow */
		sampler->lb = ((uint64) sampler->lb + sampler->step) % sampler->nblocks;
	} while (sampler->lb >= nblocks);

	return sampler->lb;
}

/*
 * Select next sampled tuple in current block.
 *
 * In block sampling, we just want to sample all the tuples in each selected
 * block.
 *
 * When we reach end of the block, return InvalidOffsetNumber which tells
 * SampleScan to go to next block.
 */
static OffsetNumber
system_time_nextsampletuple(SampleScanState *node,
							BlockNumber blockno,
							OffsetNumber maxoffset)
{
	SystemTimeSamplerData *sampler = (SystemTimeSamplerData *) node->tsm_state;
	OffsetNumber tupoffset = sampler->lt;

	/* Advance to next possible offset on page */
	if (tupoffset == InvalidOffsetNumber)
		tupoffset = FirstOffsetNumber;
	else
		tupoffset++;

	/* Done? */
	if (tupoffset > maxoffset)
		tupoffset = InvalidOffsetNumber;

	sampler->lt = tupoffset;

	return tupoffset;
}

/*
 * Compute greatest common divisor of two uint32's.
 */
static uint32
gcd(uint32 a, uint32 b)
{
	uint32		c;

	while (a != 0)
	{
		c = a;
		a = b % a;
		b = c;
	}

	return b;
}

/*
 * Pick a random value less than and relatively prime to n, if possible
 * (else return 1).
 */
static uint32
random_relative_prime(uint32 n, SamplerRandomState randstate)
{
	uint32		r;

	/* Safety check to avoid infinite loop or zero result for small n. */
	if (n <= 1)
		return 1;

	/*
	 * This should only take 2 or 3 iterations as the probability of 2 numbers
	 * being relatively prime is ~61%; but just in case, we'll include a
	 * CHECK_FOR_INTERRUPTS in the loop.
	 */
	do
	{
		CHECK_FOR_INTERRUPTS();
		r = (uint32) (sampler_random_fract(randstate) * n);
	} while (r == 0 || gcd(r, n) > 1);

	return r;
}