Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * nodeSamplescan.c
4 : * Support routines for sample scans of relations (table sampling).
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/executor/nodeSamplescan.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include "access/relscan.h"
18 : #include "access/tableam.h"
19 : #include "access/tsmapi.h"
20 : #include "common/pg_prng.h"
21 : #include "executor/executor.h"
22 : #include "executor/nodeSamplescan.h"
23 : #include "utils/fmgrprotos.h"
24 : #include "utils/rel.h"
25 :
26 : static TupleTableSlot *SampleNext(SampleScanState *node);
27 : static void tablesample_init(SampleScanState *scanstate);
28 : static TupleTableSlot *tablesample_getnext(SampleScanState *scanstate);
29 :
30 : /* ----------------------------------------------------------------
31 : * Scan Support
32 : * ----------------------------------------------------------------
33 : */
34 :
35 : /* ----------------------------------------------------------------
36 : * SampleNext
37 : *
38 : * This is a workhorse for ExecSampleScan
39 : * ----------------------------------------------------------------
40 : */
41 : static TupleTableSlot *
42 120685 : SampleNext(SampleScanState *node)
43 : {
44 : /*
45 : * if this is first call within a scan, initialize
46 : */
47 120685 : if (!node->begun)
48 108 : tablesample_init(node);
49 :
50 : /*
51 : * get the next tuple, and store it in our result slot
52 : */
53 120665 : return tablesample_getnext(node);
54 : }
55 :
56 : /*
57 : * SampleRecheck -- access method routine to recheck a tuple in EvalPlanQual
58 : */
59 : static bool
60 0 : SampleRecheck(SampleScanState *node, TupleTableSlot *slot)
61 : {
62 : /*
63 : * No need to recheck for SampleScan, since like SeqScan we don't pass any
64 : * checkable keys to heap_beginscan.
65 : */
66 0 : return true;
67 : }
68 :
69 : /* ----------------------------------------------------------------
70 : * ExecSampleScan(node)
71 : *
72 : * Scans the relation using the sampling method and returns
73 : * the next qualifying tuple.
74 : * We call the ExecScan() routine and pass it the appropriate
75 : * access method functions.
76 : * ----------------------------------------------------------------
77 : */
78 : static TupleTableSlot *
79 120682 : ExecSampleScan(PlanState *pstate)
80 : {
81 120682 : SampleScanState *node = castNode(SampleScanState, pstate);
82 :
83 120682 : return ExecScan(&node->ss,
84 : (ExecScanAccessMtd) SampleNext,
85 : (ExecScanRecheckMtd) SampleRecheck);
86 : }
87 :
88 : /* ----------------------------------------------------------------
89 : * ExecInitSampleScan
90 : * ----------------------------------------------------------------
91 : */
92 : SampleScanState *
93 153 : ExecInitSampleScan(SampleScan *node, EState *estate, int eflags)
94 : {
95 : SampleScanState *scanstate;
96 153 : TableSampleClause *tsc = node->tablesample;
97 : TsmRoutine *tsm;
98 :
99 : Assert(outerPlan(node) == NULL);
100 : Assert(innerPlan(node) == NULL);
101 :
102 : /*
103 : * create state structure
104 : */
105 153 : scanstate = makeNode(SampleScanState);
106 153 : scanstate->ss.ps.plan = (Plan *) node;
107 153 : scanstate->ss.ps.state = estate;
108 153 : scanstate->ss.ps.ExecProcNode = ExecSampleScan;
109 :
110 : /*
111 : * Miscellaneous initialization
112 : *
113 : * create expression context for node
114 : */
115 153 : ExecAssignExprContext(estate, &scanstate->ss.ps);
116 :
117 : /*
118 : * open the scan relation
119 : */
120 153 : scanstate->ss.ss_currentRelation =
121 153 : ExecOpenScanRelation(estate,
122 : node->scan.scanrelid,
123 : eflags);
124 :
125 : /* we won't set up the HeapScanDesc till later */
126 153 : scanstate->ss.ss_currentScanDesc = NULL;
127 :
128 : /* and create slot with appropriate rowtype */
129 153 : ExecInitScanTupleSlot(estate, &scanstate->ss,
130 153 : RelationGetDescr(scanstate->ss.ss_currentRelation),
131 : table_slot_callbacks(scanstate->ss.ss_currentRelation));
132 :
133 : /*
134 : * Initialize result type and projection.
135 : */
136 153 : ExecInitResultTypeTL(&scanstate->ss.ps);
137 153 : ExecAssignScanProjectionInfo(&scanstate->ss);
138 :
139 : /*
140 : * initialize child expressions
141 : */
142 153 : scanstate->ss.ps.qual =
143 153 : ExecInitQual(node->scan.plan.qual, (PlanState *) scanstate);
144 :
145 153 : scanstate->args = ExecInitExprList(tsc->args, (PlanState *) scanstate);
146 153 : scanstate->repeatable =
147 153 : ExecInitExpr(tsc->repeatable, (PlanState *) scanstate);
148 :
149 : /*
150 : * If we don't have a REPEATABLE clause, select a random seed. We want to
151 : * do this just once, since the seed shouldn't change over rescans.
152 : */
153 153 : if (tsc->repeatable == NULL)
154 87 : scanstate->seed = pg_prng_uint32(&pg_global_prng_state);
155 :
156 : /*
157 : * Finally, initialize the TABLESAMPLE method handler.
158 : */
159 153 : tsm = GetTsmRoutine(tsc->tsmhandler);
160 153 : scanstate->tsmroutine = tsm;
161 153 : scanstate->tsm_state = NULL;
162 :
163 153 : if (tsm->InitSampleScan)
164 153 : tsm->InitSampleScan(scanstate, eflags);
165 :
166 : /* We'll do BeginSampleScan later; we can't evaluate params yet */
167 153 : scanstate->begun = false;
168 :
169 153 : return scanstate;
170 : }
171 :
172 : /* ----------------------------------------------------------------
173 : * ExecEndSampleScan
174 : *
175 : * frees any storage allocated through C routines.
176 : * ----------------------------------------------------------------
177 : */
178 : void
179 133 : ExecEndSampleScan(SampleScanState *node)
180 : {
181 : /*
182 : * Tell sampling function that we finished the scan.
183 : */
184 133 : if (node->tsmroutine->EndSampleScan)
185 0 : node->tsmroutine->EndSampleScan(node);
186 :
187 : /*
188 : * close heap scan
189 : */
190 133 : if (node->ss.ss_currentScanDesc)
191 73 : table_endscan(node->ss.ss_currentScanDesc);
192 133 : }
193 :
194 : /* ----------------------------------------------------------------
195 : * ExecReScanSampleScan
196 : *
197 : * Rescans the relation.
198 : *
199 : * ----------------------------------------------------------------
200 : */
201 : void
202 29 : ExecReScanSampleScan(SampleScanState *node)
203 : {
204 : /* Remember we need to do BeginSampleScan again (if we did it at all) */
205 29 : node->begun = false;
206 29 : node->done = false;
207 29 : node->haveblock = false;
208 29 : node->donetuples = 0;
209 :
210 29 : ExecScanReScan(&node->ss);
211 29 : }
212 :
213 :
214 : /*
215 : * Initialize the TABLESAMPLE method: evaluate params and call BeginSampleScan.
216 : */
217 : static void
218 108 : tablesample_init(SampleScanState *scanstate)
219 : {
220 108 : TsmRoutine *tsm = scanstate->tsmroutine;
221 108 : ExprContext *econtext = scanstate->ss.ps.ps_ExprContext;
222 : Datum *params;
223 : Datum datum;
224 : bool isnull;
225 : uint32 seed;
226 : bool allow_sync;
227 : int i;
228 : ListCell *arg;
229 :
230 108 : scanstate->donetuples = 0;
231 108 : params = palloc_array(Datum, list_length(scanstate->args));
232 :
233 108 : i = 0;
234 213 : foreach(arg, scanstate->args)
235 : {
236 108 : ExprState *argstate = (ExprState *) lfirst(arg);
237 :
238 108 : params[i] = ExecEvalExprSwitchContext(argstate,
239 : econtext,
240 : &isnull);
241 108 : if (isnull)
242 3 : ereport(ERROR,
243 : (errcode(ERRCODE_INVALID_TABLESAMPLE_ARGUMENT),
244 : errmsg("TABLESAMPLE parameter cannot be null")));
245 105 : i++;
246 : }
247 :
248 105 : if (scanstate->repeatable)
249 : {
250 36 : datum = ExecEvalExprSwitchContext(scanstate->repeatable,
251 : econtext,
252 : &isnull);
253 36 : if (isnull)
254 3 : ereport(ERROR,
255 : (errcode(ERRCODE_INVALID_TABLESAMPLE_REPEAT),
256 : errmsg("TABLESAMPLE REPEATABLE parameter cannot be null")));
257 :
258 : /*
259 : * The REPEATABLE parameter has been coerced to float8 by the parser.
260 : * The reason for using float8 at the SQL level is that it will
261 : * produce unsurprising results both for users used to databases that
262 : * accept only integers in the REPEATABLE clause and for those who
263 : * might expect that REPEATABLE works like setseed() (a float in the
264 : * range from -1 to 1).
265 : *
266 : * We use hashfloat8() to convert the supplied value into a suitable
267 : * seed. For regression-testing purposes, that has the convenient
268 : * property that REPEATABLE(0) gives a machine-independent result.
269 : */
270 33 : seed = DatumGetUInt32(DirectFunctionCall1(hashfloat8, datum));
271 : }
272 : else
273 : {
274 : /* Use the seed selected by ExecInitSampleScan */
275 69 : seed = scanstate->seed;
276 : }
277 :
278 : /* Set default values for params that BeginSampleScan can adjust */
279 102 : scanstate->use_bulkread = true;
280 102 : scanstate->use_pagemode = true;
281 :
282 : /* Let tablesample method do its thing */
283 102 : tsm->BeginSampleScan(scanstate,
284 : params,
285 102 : list_length(scanstate->args),
286 : seed);
287 :
288 : /* We'll use syncscan if there's no NextSampleBlock function */
289 88 : allow_sync = (tsm->NextSampleBlock == NULL);
290 :
291 : /* Now we can create or reset the HeapScanDesc */
292 88 : if (scanstate->ss.ss_currentScanDesc == NULL)
293 : {
294 73 : scanstate->ss.ss_currentScanDesc =
295 73 : table_beginscan_sampling(scanstate->ss.ss_currentRelation,
296 73 : scanstate->ss.ps.state->es_snapshot,
297 : 0, NULL,
298 73 : scanstate->use_bulkread,
299 : allow_sync,
300 73 : scanstate->use_pagemode);
301 : }
302 : else
303 : {
304 15 : table_rescan_set_params(scanstate->ss.ss_currentScanDesc, NULL,
305 15 : scanstate->use_bulkread,
306 : allow_sync,
307 15 : scanstate->use_pagemode);
308 : }
309 :
310 88 : pfree(params);
311 :
312 : /* And we're initialized. */
313 88 : scanstate->begun = true;
314 88 : }
315 :
316 : /*
317 : * Get next tuple from TABLESAMPLE method.
318 : */
319 : static TupleTableSlot *
320 120665 : tablesample_getnext(SampleScanState *scanstate)
321 : {
322 120665 : TableScanDesc scan = scanstate->ss.ss_currentScanDesc;
323 120665 : TupleTableSlot *slot = scanstate->ss.ss_ScanTupleSlot;
324 :
325 120665 : ExecClearTuple(slot);
326 :
327 120665 : if (scanstate->done)
328 0 : return NULL;
329 :
330 : for (;;)
331 : {
332 127033 : if (!scanstate->haveblock)
333 : {
334 6456 : if (!table_scan_sample_next_block(scan, scanstate))
335 : {
336 85 : scanstate->haveblock = false;
337 85 : scanstate->done = true;
338 :
339 : /* exhausted relation */
340 85 : return NULL;
341 : }
342 :
343 6371 : scanstate->haveblock = true;
344 : }
345 :
346 126948 : if (!table_scan_sample_next_tuple(scan, scanstate, slot))
347 : {
348 : /*
349 : * If we get here, it means we've exhausted the items on this page
350 : * and it's time to move to the next.
351 : */
352 6368 : scanstate->haveblock = false;
353 6368 : continue;
354 : }
355 :
356 : /* Found visible tuple, return it. */
357 120580 : break;
358 : }
359 :
360 120580 : scanstate->donetuples++;
361 :
362 120580 : return slot;
363 : }
|
