Line data Source code
1 : /*
2 : * SPDX-License-Identifier: MPL-2.0
3 : *
4 : * This Source Code Form is subject to the terms of the Mozilla Public
5 : * License, v. 2.0. If a copy of the MPL was not distributed with this
6 : * file, You can obtain one at http://mozilla.org/MPL/2.0/.
7 : *
8 : * Copyright 2024 MonetDB Foundation;
9 : * Copyright August 2008 - 2023 MonetDB B.V.;
10 : * Copyright 1997 - July 2008 CWI.
11 : */
12 :
13 : #include "monetdb_config.h"
14 : #include "opt_mitosis.h"
15 : #include "mal_interpreter.h"
16 : #include "gdk_utils.h"
17 :
18 : #define MIN_PART_SIZE 100000 /* minimal record count per partition */
19 : #define MAX_PARTS2THREADS_RATIO 4 /* There should be at most this multiple more of partitions then threads */
20 :
21 :
22 : str
23 364238 : OPTmitosisImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk,
24 : InstrPtr pci)
25 : {
26 364238 : int i, j, limit, slimit, estimate = 0, pieces = 1, mito_parts = 0,
27 364238 : mito_size = 0, row_size = 0, mt = -1, nr_cols = 0, nr_aggrs = 0,
28 364238 : nr_maps = 0;
29 364238 : str schema = 0, table = 0;
30 364238 : BUN r = 0, rowcnt = 0; /* table should be sizeable to consider parallel execution */
31 364238 : InstrPtr p, q, *old, target = 0;
32 364238 : size_t argsize = 6 * sizeof(lng), m = 0;
33 : /* estimate size per operator estimate: 4 args + 2 res */
34 364238 : int threads = GDKnr_threads ? GDKnr_threads : 1, maxparts = MAXSLICES;
35 364238 : str msg = MAL_SUCCEED;
36 :
37 : /* if the user has associated limitation on the number of threads, respect it in the
38 : * generation of the number of partitions. Beware, they may lead to larger pieces, it only
39 : * limits the CPU power */
40 364238 : if (cntxt->workerlimit)
41 0 : threads = cntxt->workerlimit;
42 364238 : (void) cntxt;
43 364238 : (void) stk;
44 :
45 364238 : old = mb->stmt;
46 14782697 : for (i = 1; i < mb->stop; i++) {
47 14421071 : InstrPtr p = old[i];
48 :
49 14421071 : if (getModuleId(p) == sqlRef && getFunctionId(p) == assertRef
50 9208 : && p->argc > 2 && getArgType(mb, p, 2) == TYPE_str
51 9208 : && isVarConstant(mb, getArg(p, 2))
52 9208 : && getVarConstant(mb, getArg(p, 2)).val.sval != NULL
53 9208 : &&
54 9208 : (strstr(getVarConstant(mb, getArg(p, 2)).val.sval,
55 : "PRIMARY KEY constraint")
56 8003 : || strstr(getVarConstant(mb, getArg(p, 2)).val.sval,
57 : "UNIQUE constraint"))) {
58 1273 : pieces = 0;
59 1273 : goto bailout;
60 : }
61 :
62 : /* mitosis/mergetable bailout conditions */
63 : /* Crude protection against self join explosion */
64 14419798 : if (p->retc == 2 && isMatJoinOp(p))
65 14419806 : maxparts = threads;
66 :
67 14419806 : nr_aggrs += (p->argc > 2 && getModuleId(p) == aggrRef);
68 14419806 : nr_maps += (isMapOp(p));
69 :
70 14419796 : if (p->argc > 2 && getModuleId(p) == aggrRef
71 7091 : && getFunctionId(p) != subcountRef && getFunctionId(p) != subminRef
72 3076 : && getFunctionId(p) != submaxRef && getFunctionId(p) != subavgRef
73 2557 : && getFunctionId(p) != subsumRef && getFunctionId(p) != subprodRef
74 931 : && getFunctionId(p) != countRef && getFunctionId(p) != minRef
75 311 : && getFunctionId(p) != maxRef && getFunctionId(p) != avgRef
76 259 : && getFunctionId(p) != sumRef && getFunctionId(p) != prodRef) {
77 259 : pieces = 0;
78 259 : goto bailout;
79 : }
80 :
81 : /* rtree functions should not be optimized by mitosis
82 : * (single-threaded execution) */
83 14419537 : if (getModuleId(p) == rtreeRef) {
84 0 : pieces = 0;
85 0 : goto bailout;
86 : }
87 :
88 : /* do not split up floating point bat that is being summed */
89 14419537 : if (p->retc == 1 &&
90 13774307 : (((p->argc == 5 || p->argc == 6)
91 952289 : && getModuleId(p) == aggrRef
92 6064 : && getFunctionId(p) == subsumRef)
93 13772685 : || (p->argc == 4
94 597198 : && getModuleId(p) == aggrRef
95 0 : && getFunctionId(p) == sumRef))
96 1622 : && isaBatType(getArgType(mb, p, p->retc))
97 1622 : && (getBatType(getArgType(mb, p, p->retc)) == TYPE_flt
98 : || getBatType(getArgType(mb, p, p->retc)) == TYPE_dbl)) {
99 14 : pieces = 0;
100 14 : goto bailout;
101 : }
102 :
103 14419523 : if (p->argc > 2
104 7811210 : && (getModuleId(p) == capiRef || getModuleId(p) == rapiRef
105 7811179 : || getModuleId(p) == pyapi3Ref)
106 143 : && getFunctionId(p) == subeval_aggrRef) {
107 26 : pieces = 0;
108 26 : goto bailout;
109 : }
110 :
111 : /* Mergetable cannot handle intersect/except's for now */
112 14419497 : if (getModuleId(p) == algebraRef && getFunctionId(p) == groupbyRef) {
113 1038 : pieces = 0;
114 1038 : goto bailout;
115 : }
116 :
117 : /* locate the largest non-partitioned table */
118 14418459 : if (getModuleId(p) != sqlRef
119 1183341 : || (getFunctionId(p) != bindRef && getFunctionId(p) != bindidxRef
120 663186 : && getFunctionId(p) != tidRef))
121 13767334 : continue;
122 : /* don't split insert BATs */
123 651125 : if (p->argc > 5 && getVarConstant(mb, getArg(p, 5)).val.ival == 1)
124 0 : continue;
125 651125 : if (p->argc > 6)
126 142610 : continue; /* already partitioned */
127 : /*
128 : * The SQL optimizer already collects the counts of the base
129 : * table and passes them on as a row property. All pieces for a
130 : * single subplan should ideally fit together.
131 : */
132 508515 : r = getRowCnt(mb, getArg(p, 0));
133 508515 : if (r == rowcnt)
134 257043 : nr_cols++;
135 508515 : if (r > rowcnt) {
136 : /* the rowsize depends on the column types, assume void-headed */
137 50892 : row_size = ATOMsize(getBatType(getArgType(mb, p, 0)));
138 50892 : rowcnt = r;
139 50892 : nr_cols = 1;
140 50892 : target = p;
141 50892 : estimate++;
142 50892 : r = 0;
143 : }
144 : }
145 361626 : if (target == 0) {
146 313855 : pieces = 0;
147 313855 : goto bailout;
148 : }
149 : /*
150 : * The number of pieces should be based on the footprint of the
151 : * query plan, such that preferably it can be handled without
152 : * swapping intermediates. For the time being we just go for pieces
153 : * that fit into memory in isolation. A fictive rowcount is derived
154 : * based on argument types, such that all pieces would fit into
155 : * memory conveniently for processing. We attempt to use not more
156 : * threads than strictly needed.
157 : * Experience shows that the pieces should not be too small.
158 : * If we should limit to |threads| is still an open issue.
159 : *
160 : * Take into account the number of client connections,
161 : * because all user together are responsible for resource contentions
162 : */
163 47771 : MT_lock_set(&mal_contextLock);
164 47785 : cntxt->idle = 0; // this one is definitely not idle
165 47785 : MT_lock_unset(&mal_contextLock);
166 :
167 : /* improve memory usage estimation */
168 47785 : if (nr_cols > 1 || nr_aggrs > 1 || nr_maps > 1)
169 47684 : argsize = (nr_cols + nr_aggrs + nr_maps) * sizeof(lng);
170 : /* We haven't assigned the number of pieces.
171 : * Determine the memory available for this client
172 : */
173 :
174 : /* respect the memory limit size set for the user
175 : * and determine the column part size
176 : */
177 47785 : m = GDK_mem_maxsize / MCactiveClients(); /* use temporarily */
178 47785 : if (cntxt->memorylimit > 0 && (size_t) cntxt->memorylimit << 20 < m)
179 0 : m = ((size_t) cntxt->memorylimit << 20) / argsize;
180 47785 : else if (cntxt->maxmem > 0 && cntxt->maxmem < (lng) m)
181 0 : m = (size_t) (cntxt->maxmem / argsize);
182 : else
183 47785 : m = m / argsize;
184 :
185 : /* if data exceeds memory size,
186 : * i.e., (rowcnt*argsize > GDK_mem_maxsize),
187 : * i.e., (rowcnt > GDK_mem_maxsize/argsize = m) */
188 47785 : if (rowcnt > m && m / threads > 0) {
189 : /* create |pieces| > |threads| partitions such that
190 : * |threads| partitions at a time fit in memory,
191 : * i.e., (threads*(rowcnt/pieces) <= m),
192 : * i.e., (rowcnt/pieces <= m/threads),
193 : * i.e., (pieces => rowcnt/(m/threads))
194 : * (assuming that (m > threads*MIN_PART_SIZE)) */
195 : /* the number of pieces affects SF-100, going beyond 8x increases
196 : * the optimizer costs beyond the execution time
197 : */
198 0 : pieces = ((int) ceil((double) rowcnt / (m / threads)));
199 0 : if (pieces <= threads)
200 : pieces = threads;
201 47785 : } else if (rowcnt > MIN_PART_SIZE) {
202 : /* exploit parallelism, but ensure minimal partition size to
203 : * limit overhead */
204 193 : pieces = MIN((int) ceil((double) rowcnt / MIN_PART_SIZE),
205 : MAX_PARTS2THREADS_RATIO * threads);
206 : }
207 :
208 : /* when testing, always aim for full parallelism, but avoid
209 : * empty pieces */
210 47785 : FORCEMITODEBUG if (pieces < threads)
211 47515 : pieces = (int) MIN((BUN) threads, rowcnt);
212 : /* prevent plan explosion */
213 47785 : if (pieces > maxparts)
214 : pieces = maxparts;
215 : /* to enable experimentation we introduce the option to set
216 : * the number of partitions required and/or the size of each chunk (in K)
217 : */
218 47785 : mito_parts = GDKgetenv_int("mito_parts", 0);
219 47785 : if (mito_parts > 0)
220 0 : pieces = mito_parts;
221 47785 : mito_size = GDKgetenv_int("mito_size", 0);
222 47785 : if (mito_size > 0)
223 0 : pieces = (int) ((rowcnt * row_size) / (mito_size * 1024));
224 :
225 47785 : if (pieces <= 1) {
226 389 : pieces = 0;
227 389 : goto bailout;
228 : }
229 :
230 : /* at this stage we have identified the #chunks to be used for the largest table */
231 47396 : limit = mb->stop;
232 47396 : slimit = mb->ssize;
233 47396 : if (newMalBlkStmt(mb, mb->stop + 2 * estimate) < 0)
234 0 : throw(MAL, "optimizer.mitosis", SQLSTATE(HY013) MAL_MALLOC_FAIL);
235 47396 : estimate = 0;
236 :
237 47396 : schema = getVarConstant(mb, getArg(target, 2)).val.sval;
238 47396 : table = getVarConstant(mb, getArg(target, 3)).val.sval;
239 4383696 : for (i = 0; mb->errors == NULL && i < limit; i++) {
240 4336302 : int upd = 0, qtpe, rtpe = 0, qv, rv;
241 4336302 : InstrPtr matq, matr = NULL;
242 4336302 : p = old[i];
243 :
244 4336302 : if (getModuleId(p) != sqlRef
245 920440 : || !(getFunctionId(p) == bindRef || getFunctionId(p) == bindidxRef
246 429837 : || getFunctionId(p) == tidRef)) {
247 3725889 : pushInstruction(mb, p);
248 3725875 : continue;
249 : }
250 : /* don't split insert BATs */
251 610413 : if (p->argc == 6 && getVarConstant(mb, getArg(p, 5)).val.ival == 1) {
252 0 : pushInstruction(mb, p);
253 0 : continue;
254 : }
255 610413 : r = getRowCnt(mb, getArg(p, 0));
256 610413 : if (r < rowcnt) {
257 314592 : pushInstruction(mb, p);
258 314592 : continue;
259 : }
260 : /* Don't split the (index) bat if we already have identified a range */
261 : /* This will happen if we inline separately optimized routines */
262 295821 : if (p->argc > 7) {
263 0 : pushInstruction(mb, p);
264 0 : continue;
265 : }
266 295821 : if (p->retc == 2)
267 62347 : upd = 1;
268 296309 : if (mt < 0
269 295821 : && (strcmp(schema, getVarConstant(mb, getArg(p, 2 + upd)).val.sval)
270 295821 : || strcmp(table,
271 295821 : getVarConstant(mb, getArg(p, 3 + upd)).val.sval))) {
272 486 : pushInstruction(mb, p);
273 488 : continue;
274 : }
275 : /* we keep the original bind operation, because it allows for
276 : * easy undo when the mergtable can not do something */
277 : // pushInstruction(mb, p);
278 :
279 295335 : qtpe = getVarType(mb, getArg(p, 0));
280 :
281 295335 : matq = newInstructionArgs(NULL, matRef, newRef, pieces + 1);
282 295326 : if (matq == NULL) {
283 0 : msg = createException(MAL, "optimizer.mitosis",
284 : SQLSTATE(HY013) MAL_MALLOC_FAIL);
285 0 : break;
286 : }
287 295326 : getArg(matq, 0) = getArg(p, 0);
288 :
289 295326 : if (upd) {
290 62347 : matr = newInstructionArgs(NULL, matRef, newRef, pieces + 1);
291 62347 : if (matr == NULL) {
292 0 : freeInstruction(matq);
293 0 : msg = createException(MAL, "optimizer.mitosis",
294 : SQLSTATE(HY013) MAL_MALLOC_FAIL);
295 0 : break;
296 : }
297 62347 : getArg(matr, 0) = getArg(p, 1);
298 62347 : rtpe = getVarType(mb, getArg(p, 1));
299 : }
300 :
301 2519896 : for (j = 0; j < pieces; j++) {
302 2224570 : q = copyInstruction(p);
303 2224802 : if (q == NULL) {
304 0 : freeInstruction(matr);
305 0 : freeInstruction(matq);
306 0 : for (; i < limit; i++)
307 0 : if (old[i])
308 0 : pushInstruction(mb, old[i]);
309 0 : GDKfree(old);
310 0 : throw(MAL, "optimizer.mitosis",
311 : SQLSTATE(HY013) MAL_MALLOC_FAIL);
312 : }
313 2224802 : q = pushInt(mb, q, j);
314 2223912 : q = pushInt(mb, q, pieces);
315 :
316 2224075 : qv = getArg(q, 0) = newTmpVariable(mb, qtpe);
317 2224540 : if (upd) {
318 458095 : rv = getArg(q, 1) = newTmpVariable(mb, rtpe);
319 : }
320 2224540 : pushInstruction(mb, q);
321 2224551 : matq = pushArgument(mb, matq, qv);
322 2224570 : if (upd)
323 458095 : matr = pushArgument(mb, matr, rv);
324 : }
325 295326 : pushInstruction(mb, matq);
326 295328 : if (upd)
327 62347 : pushInstruction(mb, matr);
328 295328 : freeInstruction(p);
329 : }
330 9505732 : for (; i < slimit; i++)
331 9458337 : if (old[i])
332 0 : pushInstruction(mb, old[i]);
333 47395 : GDKfree(old);
334 :
335 : /* Defense line against incorrect plans */
336 47396 : if (msg == MAL_SUCCEED) {
337 47396 : msg = chkTypes(cntxt->usermodule, mb, FALSE);
338 47396 : if (msg == MAL_SUCCEED) {
339 47396 : msg = chkFlow(mb);
340 47393 : if (msg == MAL_SUCCEED)
341 47394 : msg = chkDeclarations(mb);
342 : }
343 : }
344 0 : bailout:
345 : /* keep actions taken as a fake argument */
346 364248 : (void) pushInt(mb, pci, pieces);
347 364248 : return msg;
348 : }
|
