Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * tuplesort.c
4 : * Generalized tuple sorting routines.
5 : *
6 : * This module provides a generalized facility for tuple sorting, which can be
7 : * applied to different kinds of sortable objects. Implementation of
8 : * the particular sorting variants is given in tuplesortvariants.c.
9 : * This module works efficiently for both small and large amounts
10 : * of data. Small amounts are sorted in-memory using qsort(). Large
11 : * amounts are sorted using temporary files and a standard external sort
12 : * algorithm.
13 : *
14 : * See Knuth, volume 3, for more than you want to know about external
15 : * sorting algorithms. The algorithm we use is a balanced k-way merge.
16 : * Before PostgreSQL 15, we used the polyphase merge algorithm (Knuth's
17 : * Algorithm 5.4.2D), but with modern hardware, a straightforward balanced
18 : * merge is better. Knuth is assuming that tape drives are expensive
19 : * beasts, and in particular that there will always be many more runs than
20 : * tape drives. The polyphase merge algorithm was good at keeping all the
21 : * tape drives busy, but in our implementation a "tape drive" doesn't cost
22 : * much more than a few Kb of memory buffers, so we can afford to have
23 : * lots of them. In particular, if we can have as many tape drives as
24 : * sorted runs, we can eliminate any repeated I/O at all.
25 : *
26 : * Historically, we divided the input into sorted runs using replacement
27 : * selection, in the form of a priority tree implemented as a heap
28 : * (essentially Knuth's Algorithm 5.2.3H), but now we always use quicksort
29 : * for run generation.
30 : *
31 : * The approximate amount of memory allowed for any one sort operation
32 : * is specified in kilobytes by the caller (most pass work_mem). Initially,
33 : * we absorb tuples and simply store them in an unsorted array as long as
34 : * we haven't exceeded workMem. If we reach the end of the input without
35 : * exceeding workMem, we sort the array using qsort() and subsequently return
36 : * tuples just by scanning the tuple array sequentially. If we do exceed
37 : * workMem, we begin to emit tuples into sorted runs in temporary tapes.
38 : * When tuples are dumped in batch after quicksorting, we begin a new run
39 : * with a new output tape. If we reach the max number of tapes, we write
40 : * subsequent runs on the existing tapes in a round-robin fashion. We will
41 : * need multiple merge passes to finish the merge in that case. After the
42 : * end of the input is reached, we dump out remaining tuples in memory into
43 : * a final run, then merge the runs.
44 : *
45 : * When merging runs, we use a heap containing just the frontmost tuple from
46 : * each source run; we repeatedly output the smallest tuple and replace it
47 : * with the next tuple from its source tape (if any). When the heap empties,
48 : * the merge is complete. The basic merge algorithm thus needs very little
49 : * memory --- only M tuples for an M-way merge, and M is constrained to a
50 : * small number. However, we can still make good use of our full workMem
51 : * allocation by pre-reading additional blocks from each source tape. Without
52 : * prereading, our access pattern to the temporary file would be very erratic;
53 : * on average we'd read one block from each of M source tapes during the same
54 : * time that we're writing M blocks to the output tape, so there is no
55 : * sequentiality of access at all, defeating the read-ahead methods used by
56 : * most Unix kernels. Worse, the output tape gets written into a very random
57 : * sequence of blocks of the temp file, ensuring that things will be even
58 : * worse when it comes time to read that tape. A straightforward merge pass
59 : * thus ends up doing a lot of waiting for disk seeks. We can improve matters
60 : * by prereading from each source tape sequentially, loading about workMem/M
61 : * bytes from each tape in turn, and making the sequential blocks immediately
62 : * available for reuse. This approach helps to localize both read and write
63 : * accesses. The pre-reading is handled by logtape.c, we just tell it how
64 : * much memory to use for the buffers.
65 : *
66 : * In the current code we determine the number of input tapes M on the basis
67 : * of workMem: we want workMem/M to be large enough that we read a fair
68 : * amount of data each time we read from a tape, so as to maintain the
69 : * locality of access described above. Nonetheless, with large workMem we
70 : * can have many tapes. The logical "tapes" are implemented by logtape.c,
71 : * which avoids space wastage by recycling disk space as soon as each block
72 : * is read from its "tape".
73 : *
74 : * When the caller requests random access to the sort result, we form
75 : * the final sorted run on a logical tape which is then "frozen", so
76 : * that we can access it randomly. When the caller does not need random
77 : * access, we return from tuplesort_performsort() as soon as we are down
78 : * to one run per logical tape. The final merge is then performed
79 : * on-the-fly as the caller repeatedly calls tuplesort_getXXX; this
80 : * saves one cycle of writing all the data out to disk and reading it in.
81 : *
82 : * This module supports parallel sorting. Parallel sorts involve coordination
83 : * among one or more worker processes, and a leader process, each with its own
84 : * tuplesort state. The leader process (or, more accurately, the
85 : * Tuplesortstate associated with a leader process) creates a full tapeset
86 : * consisting of worker tapes with one run to merge; a run for every
87 : * worker process. This is then merged. Worker processes are guaranteed to
88 : * produce exactly one output run from their partial input.
89 : *
90 : *
91 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
92 : * Portions Copyright (c) 1994, Regents of the University of California
93 : *
94 : * IDENTIFICATION
95 : * src/backend/utils/sort/tuplesort.c
96 : *
97 : *-------------------------------------------------------------------------
98 : */
99 :
100 : #include "postgres.h"
101 :
102 : #include <limits.h>
103 :
104 : #include "commands/tablespace.h"
105 : #include "miscadmin.h"
106 : #include "pg_trace.h"
107 : #include "storage/shmem.h"
108 : #include "utils/guc.h"
109 : #include "utils/memutils.h"
110 : #include "utils/pg_rusage.h"
111 : #include "utils/tuplesort.h"
112 :
113 : /*
114 : * Initial size of memtuples array. This must be more than
115 : * ALLOCSET_SEPARATE_THRESHOLD; see comments in grow_memtuples(). Clamp at
116 : * 1024 elements to avoid excessive reallocs.
117 : */
118 : #define INITIAL_MEMTUPSIZE Max(1024, \
119 : ALLOCSET_SEPARATE_THRESHOLD / sizeof(SortTuple) + 1)
120 :
121 : /* GUC variables */
122 : bool trace_sort = false;
123 :
124 : #ifdef DEBUG_BOUNDED_SORT
125 : bool optimize_bounded_sort = true;
126 : #endif
127 :
128 :
129 : /*
130 : * During merge, we use a pre-allocated set of fixed-size slots to hold
131 : * tuples. To avoid palloc/pfree overhead.
132 : *
133 : * Merge doesn't require a lot of memory, so we can afford to waste some,
134 : * by using gratuitously-sized slots. If a tuple is larger than 1 kB, the
135 : * palloc() overhead is not significant anymore.
136 : *
137 : * 'nextfree' is valid when this chunk is in the free list. When in use, the
138 : * slot holds a tuple.
139 : */
140 : #define SLAB_SLOT_SIZE 1024
141 :
142 : typedef union SlabSlot
143 : {
144 : union SlabSlot *nextfree;
145 : char buffer[SLAB_SLOT_SIZE];
146 : } SlabSlot;
147 :
148 : /*
149 : * Possible states of a Tuplesort object. These denote the states that
150 : * persist between calls of Tuplesort routines.
151 : */
152 : typedef enum
153 : {
154 : TSS_INITIAL, /* Loading tuples; still within memory limit */
155 : TSS_BOUNDED, /* Loading tuples into bounded-size heap */
156 : TSS_BUILDRUNS, /* Loading tuples; writing to tape */
157 : TSS_SORTEDINMEM, /* Sort completed entirely in memory */
158 : TSS_SORTEDONTAPE, /* Sort completed, final run is on tape */
159 : TSS_FINALMERGE, /* Performing final merge on-the-fly */
160 : } TupSortStatus;
161 :
162 : /*
163 : * Parameters for calculation of number of tapes to use --- see inittapes()
164 : * and tuplesort_merge_order().
165 : *
166 : * In this calculation we assume that each tape will cost us about 1 blocks
167 : * worth of buffer space. This ignores the overhead of all the other data
168 : * structures needed for each tape, but it's probably close enough.
169 : *
170 : * MERGE_BUFFER_SIZE is how much buffer space we'd like to allocate for each
171 : * input tape, for pre-reading (see discussion at top of file). This is *in
172 : * addition to* the 1 block already included in TAPE_BUFFER_OVERHEAD.
173 : */
174 : #define MINORDER 6 /* minimum merge order */
175 : #define MAXORDER 500 /* maximum merge order */
176 : #define TAPE_BUFFER_OVERHEAD BLCKSZ
177 : #define MERGE_BUFFER_SIZE (BLCKSZ * 32)
178 :
179 :
180 : /*
181 : * Private state of a Tuplesort operation.
182 : */
183 : struct Tuplesortstate
184 : {
185 : TuplesortPublic base;
186 : TupSortStatus status; /* enumerated value as shown above */
187 : bool bounded; /* did caller specify a maximum number of
188 : * tuples to return? */
189 : bool boundUsed; /* true if we made use of a bounded heap */
190 : int bound; /* if bounded, the maximum number of tuples */
191 : int64 tupleMem; /* memory consumed by individual tuples.
192 : * storing this separately from what we track
193 : * in availMem allows us to subtract the
194 : * memory consumed by all tuples when dumping
195 : * tuples to tape */
196 : int64 availMem; /* remaining memory available, in bytes */
197 : int64 allowedMem; /* total memory allowed, in bytes */
198 : int maxTapes; /* max number of input tapes to merge in each
199 : * pass */
200 : int64 maxSpace; /* maximum amount of space occupied among sort
201 : * of groups, either in-memory or on-disk */
202 : bool isMaxSpaceDisk; /* true when maxSpace tracks on-disk space,
203 : * false means in-memory */
204 : TupSortStatus maxSpaceStatus; /* sort status when maxSpace was reached */
205 : LogicalTapeSet *tapeset; /* logtape.c object for tapes in a temp file */
206 :
207 : /*
208 : * This array holds the tuples now in sort memory. If we are in state
209 : * INITIAL, the tuples are in no particular order; if we are in state
210 : * SORTEDINMEM, the tuples are in final sorted order; in states BUILDRUNS
211 : * and FINALMERGE, the tuples are organized in "heap" order per Algorithm
212 : * H. In state SORTEDONTAPE, the array is not used.
213 : */
214 : SortTuple *memtuples; /* array of SortTuple structs */
215 : int memtupcount; /* number of tuples currently present */
216 : int memtupsize; /* allocated length of memtuples array */
217 : bool growmemtuples; /* memtuples' growth still underway? */
218 :
219 : /*
220 : * Memory for tuples is sometimes allocated using a simple slab allocator,
221 : * rather than with palloc(). Currently, we switch to slab allocation
222 : * when we start merging. Merging only needs to keep a small, fixed
223 : * number of tuples in memory at any time, so we can avoid the
224 : * palloc/pfree overhead by recycling a fixed number of fixed-size slots
225 : * to hold the tuples.
226 : *
227 : * For the slab, we use one large allocation, divided into SLAB_SLOT_SIZE
228 : * slots. The allocation is sized to have one slot per tape, plus one
229 : * additional slot. We need that many slots to hold all the tuples kept
230 : * in the heap during merge, plus the one we have last returned from the
231 : * sort, with tuplesort_gettuple.
232 : *
233 : * Initially, all the slots are kept in a linked list of free slots. When
234 : * a tuple is read from a tape, it is put to the next available slot, if
235 : * it fits. If the tuple is larger than SLAB_SLOT_SIZE, it is palloc'd
236 : * instead.
237 : *
238 : * When we're done processing a tuple, we return the slot back to the free
239 : * list, or pfree() if it was palloc'd. We know that a tuple was
240 : * allocated from the slab, if its pointer value is between
241 : * slabMemoryBegin and -End.
242 : *
243 : * When the slab allocator is used, the USEMEM/LACKMEM mechanism of
244 : * tracking memory usage is not used.
245 : */
246 : bool slabAllocatorUsed;
247 :
248 : char *slabMemoryBegin; /* beginning of slab memory arena */
249 : char *slabMemoryEnd; /* end of slab memory arena */
250 : SlabSlot *slabFreeHead; /* head of free list */
251 :
252 : /* Memory used for input and output tape buffers. */
253 : size_t tape_buffer_mem;
254 :
255 : /*
256 : * When we return a tuple to the caller in tuplesort_gettuple_XXX, that
257 : * came from a tape (that is, in TSS_SORTEDONTAPE or TSS_FINALMERGE
258 : * modes), we remember the tuple in 'lastReturnedTuple', so that we can
259 : * recycle the memory on next gettuple call.
260 : */
261 : void *lastReturnedTuple;
262 :
263 : /*
264 : * While building initial runs, this is the current output run number.
265 : * Afterwards, it is the number of initial runs we made.
266 : */
267 : int currentRun;
268 :
269 : /*
270 : * Logical tapes, for merging.
271 : *
272 : * The initial runs are written in the output tapes. In each merge pass,
273 : * the output tapes of the previous pass become the input tapes, and new
274 : * output tapes are created as needed. When nInputTapes equals
275 : * nInputRuns, there is only one merge pass left.
276 : */
277 : LogicalTape **inputTapes;
278 : int nInputTapes;
279 : int nInputRuns;
280 :
281 : LogicalTape **outputTapes;
282 : int nOutputTapes;
283 : int nOutputRuns;
284 :
285 : LogicalTape *destTape; /* current output tape */
286 :
287 : /*
288 : * These variables are used after completion of sorting to keep track of
289 : * the next tuple to return. (In the tape case, the tape's current read
290 : * position is also critical state.)
291 : */
292 : LogicalTape *result_tape; /* actual tape of finished output */
293 : int current; /* array index (only used if SORTEDINMEM) */
294 : bool eof_reached; /* reached EOF (needed for cursors) */
295 :
296 : /* markpos_xxx holds marked position for mark and restore */
297 : int64 markpos_block; /* tape block# (only used if SORTEDONTAPE) */
298 : int markpos_offset; /* saved "current", or offset in tape block */
299 : bool markpos_eof; /* saved "eof_reached" */
300 :
301 : /*
302 : * These variables are used during parallel sorting.
303 : *
304 : * worker is our worker identifier. Follows the general convention that
305 : * -1 value relates to a leader tuplesort, and values >= 0 worker
306 : * tuplesorts. (-1 can also be a serial tuplesort.)
307 : *
308 : * shared is mutable shared memory state, which is used to coordinate
309 : * parallel sorts.
310 : *
311 : * nParticipants is the number of worker Tuplesortstates known by the
312 : * leader to have actually been launched, which implies that they must
313 : * finish a run that the leader needs to merge. Typically includes a
314 : * worker state held by the leader process itself. Set in the leader
315 : * Tuplesortstate only.
316 : */
317 : int worker;
318 : Sharedsort *shared;
319 : int nParticipants;
320 :
321 : /*
322 : * Additional state for managing "abbreviated key" sortsupport routines
323 : * (which currently may be used by all cases except the hash index case).
324 : * Tracks the intervals at which the optimization's effectiveness is
325 : * tested.
326 : */
327 : int64 abbrevNext; /* Tuple # at which to next check
328 : * applicability */
329 :
330 : /*
331 : * Resource snapshot for time of sort start.
332 : */
333 : PGRUsage ru_start;
334 : };
335 :
336 : /*
337 : * Private mutable state of tuplesort-parallel-operation. This is allocated
338 : * in shared memory.
339 : */
340 : struct Sharedsort
341 : {
342 : /* mutex protects all fields prior to tapes */
343 : slock_t mutex;
344 :
345 : /*
346 : * currentWorker generates ordinal identifier numbers for parallel sort
347 : * workers. These start from 0, and are always gapless.
348 : *
349 : * Workers increment workersFinished to indicate having finished. If this
350 : * is equal to state.nParticipants within the leader, leader is ready to
351 : * merge worker runs.
352 : */
353 : int currentWorker;
354 : int workersFinished;
355 :
356 : /* Temporary file space */
357 : SharedFileSet fileset;
358 :
359 : /* Size of tapes flexible array */
360 : int nTapes;
361 :
362 : /*
363 : * Tapes array used by workers to report back information needed by the
364 : * leader to concatenate all worker tapes into one for merging
365 : */
366 : TapeShare tapes[FLEXIBLE_ARRAY_MEMBER];
367 : };
368 :
369 : /*
370 : * Is the given tuple allocated from the slab memory arena?
371 : */
372 : #define IS_SLAB_SLOT(state, tuple) \
373 : ((char *) (tuple) >= (state)->slabMemoryBegin && \
374 : (char *) (tuple) < (state)->slabMemoryEnd)
375 :
376 : /*
377 : * Return the given tuple to the slab memory free list, or free it
378 : * if it was palloc'd.
379 : */
380 : #define RELEASE_SLAB_SLOT(state, tuple) \
381 : do { \
382 : SlabSlot *buf = (SlabSlot *) tuple; \
383 : \
384 : if (IS_SLAB_SLOT((state), buf)) \
385 : { \
386 : buf->nextfree = (state)->slabFreeHead; \
387 : (state)->slabFreeHead = buf; \
388 : } else \
389 : pfree(buf); \
390 : } while(0)
391 :
392 : #define REMOVEABBREV(state,stup,count) ((*(state)->base.removeabbrev) (state, stup, count))
393 : #define COMPARETUP(state,a,b) ((*(state)->base.comparetup) (a, b, state))
394 : #define WRITETUP(state,tape,stup) ((*(state)->base.writetup) (state, tape, stup))
395 : #define READTUP(state,stup,tape,len) ((*(state)->base.readtup) (state, stup, tape, len))
396 : #define FREESTATE(state) ((state)->base.freestate ? (*(state)->base.freestate) (state) : (void) 0)
397 : #define LACKMEM(state) ((state)->availMem < 0 && !(state)->slabAllocatorUsed)
398 : #define USEMEM(state,amt) ((state)->availMem -= (amt))
399 : #define FREEMEM(state,amt) ((state)->availMem += (amt))
400 : #define SERIAL(state) ((state)->shared == NULL)
401 : #define WORKER(state) ((state)->shared && (state)->worker != -1)
402 : #define LEADER(state) ((state)->shared && (state)->worker == -1)
403 :
404 : /*
405 : * NOTES about on-tape representation of tuples:
406 : *
407 : * We require the first "unsigned int" of a stored tuple to be the total size
408 : * on-tape of the tuple, including itself (so it is never zero; an all-zero
409 : * unsigned int is used to delimit runs). The remainder of the stored tuple
410 : * may or may not match the in-memory representation of the tuple ---
411 : * any conversion needed is the job of the writetup and readtup routines.
412 : *
413 : * If state->sortopt contains TUPLESORT_RANDOMACCESS, then the stored
414 : * representation of the tuple must be followed by another "unsigned int" that
415 : * is a copy of the length --- so the total tape space used is actually
416 : * sizeof(unsigned int) more than the stored length value. This allows
417 : * read-backwards. When the random access flag was not specified, the
418 : * write/read routines may omit the extra length word.
419 : *
420 : * writetup is expected to write both length words as well as the tuple
421 : * data. When readtup is called, the tape is positioned just after the
422 : * front length word; readtup must read the tuple data and advance past
423 : * the back length word (if present).
424 : *
425 : * The write/read routines can make use of the tuple description data
426 : * stored in the Tuplesortstate record, if needed. They are also expected
427 : * to adjust state->availMem by the amount of memory space (not tape space!)
428 : * released or consumed. There is no error return from either writetup
429 : * or readtup; they should ereport() on failure.
430 : *
431 : *
432 : * NOTES about memory consumption calculations:
433 : *
434 : * We count space allocated for tuples against the workMem limit, plus
435 : * the space used by the variable-size memtuples array. Fixed-size space
436 : * is not counted; it's small enough to not be interesting.
437 : *
438 : * Note that we count actual space used (as shown by GetMemoryChunkSpace)
439 : * rather than the originally-requested size. This is important since
440 : * palloc can add substantial overhead. It's not a complete answer since
441 : * we won't count any wasted space in palloc allocation blocks, but it's
442 : * a lot better than what we were doing before 7.3. As of 9.6, a
443 : * separate memory context is used for caller passed tuples. Resetting
444 : * it at certain key increments significantly ameliorates fragmentation.
445 : * readtup routines use the slab allocator (they cannot use
446 : * the reset context because it gets deleted at the point that merging
447 : * begins).
448 : */
449 :
450 :
451 : static void tuplesort_begin_batch(Tuplesortstate *state);
452 : static bool consider_abort_common(Tuplesortstate *state);
453 : static void inittapes(Tuplesortstate *state, bool mergeruns);
454 : static void inittapestate(Tuplesortstate *state, int maxTapes);
455 : static void selectnewtape(Tuplesortstate *state);
456 : static void init_slab_allocator(Tuplesortstate *state, int numSlots);
457 : static void mergeruns(Tuplesortstate *state);
458 : static void mergeonerun(Tuplesortstate *state);
459 : static void beginmerge(Tuplesortstate *state);
460 : static bool mergereadnext(Tuplesortstate *state, LogicalTape *srcTape, SortTuple *stup);
461 : static void dumptuples(Tuplesortstate *state, bool alltuples);
462 : static void make_bounded_heap(Tuplesortstate *state);
463 : static void sort_bounded_heap(Tuplesortstate *state);
464 : static void tuplesort_sort_memtuples(Tuplesortstate *state);
465 : static void tuplesort_heap_insert(Tuplesortstate *state, SortTuple *tuple);
466 : static void tuplesort_heap_replace_top(Tuplesortstate *state, SortTuple *tuple);
467 : static void tuplesort_heap_delete_top(Tuplesortstate *state);
468 : static void reversedirection(Tuplesortstate *state);
469 : static unsigned int getlen(LogicalTape *tape, bool eofOK);
470 : static void markrunend(LogicalTape *tape);
471 : static int worker_get_identifier(Tuplesortstate *state);
472 : static void worker_freeze_result_tape(Tuplesortstate *state);
473 : static void worker_nomergeruns(Tuplesortstate *state);
474 : static void leader_takeover_tapes(Tuplesortstate *state);
475 : static void free_sort_tuple(Tuplesortstate *state, SortTuple *stup);
476 : static void tuplesort_free(Tuplesortstate *state);
477 : static void tuplesort_updatemax(Tuplesortstate *state);
478 :
479 : /*
480 : * Specialized comparators that we can inline into specialized sorts. The goal
481 : * is to try to sort two tuples without having to follow the pointers to the
482 : * comparator or the tuple.
483 : *
484 : * XXX: For now, there is no specialization for cases where datum1 is
485 : * authoritative and we don't even need to fall back to a callback at all (that
486 : * would be true for types like int4/int8/timestamp/date, but not true for
487 : * abbreviations of text or multi-key sorts. There could be! Is it worth it?
488 : */
489 :
490 : /* Used if first key's comparator is ssup_datum_unsigned_cmp */
491 : static pg_attribute_always_inline int
492 46637480 : qsort_tuple_unsigned_compare(SortTuple *a, SortTuple *b, Tuplesortstate *state)
493 : {
494 : int compare;
495 :
496 46637480 : compare = ApplyUnsignedSortComparator(a->datum1, a->isnull1,
497 46637480 : b->datum1, b->isnull1,
498 : &state->base.sortKeys[0]);
499 46637480 : if (compare != 0)
500 42070550 : return compare;
501 :
502 : /*
503 : * No need to waste effort calling the tiebreak function when there are no
504 : * other keys to sort on.
505 : */
506 4566930 : if (state->base.onlyKey != NULL)
507 0 : return 0;
508 :
509 4566930 : return state->base.comparetup_tiebreak(a, b, state);
510 : }
511 :
512 : /* Used if first key's comparator is ssup_datum_signed_cmp */
513 : static pg_attribute_always_inline int
514 9710082 : qsort_tuple_signed_compare(SortTuple *a, SortTuple *b, Tuplesortstate *state)
515 : {
516 : int compare;
517 :
518 9710082 : compare = ApplySignedSortComparator(a->datum1, a->isnull1,
519 9710082 : b->datum1, b->isnull1,
520 : &state->base.sortKeys[0]);
521 :
522 9710082 : if (compare != 0)
523 9600384 : return compare;
524 :
525 : /*
526 : * No need to waste effort calling the tiebreak function when there are no
527 : * other keys to sort on.
528 : */
529 109698 : if (state->base.onlyKey != NULL)
530 95264 : return 0;
531 :
532 14434 : return state->base.comparetup_tiebreak(a, b, state);
533 : }
534 :
535 : /* Used if first key's comparator is ssup_datum_int32_cmp */
536 : static pg_attribute_always_inline int
537 56096930 : qsort_tuple_int32_compare(SortTuple *a, SortTuple *b, Tuplesortstate *state)
538 : {
539 : int compare;
540 :
541 56096930 : compare = ApplyInt32SortComparator(a->datum1, a->isnull1,
542 56096930 : b->datum1, b->isnull1,
543 : &state->base.sortKeys[0]);
544 :
545 56096930 : if (compare != 0)
546 39971326 : return compare;
547 :
548 : /*
549 : * No need to waste effort calling the tiebreak function when there are no
550 : * other keys to sort on.
551 : */
552 16125604 : if (state->base.onlyKey != NULL)
553 2335442 : return 0;
554 :
555 13790162 : return state->base.comparetup_tiebreak(a, b, state);
556 : }
557 :
558 : /*
559 : * Special versions of qsort just for SortTuple objects. qsort_tuple() sorts
560 : * any variant of SortTuples, using the appropriate comparetup function.
561 : * qsort_ssup() is specialized for the case where the comparetup function
562 : * reduces to ApplySortComparator(), that is single-key MinimalTuple sorts
563 : * and Datum sorts. qsort_tuple_{unsigned,signed,int32} are specialized for
564 : * common comparison functions on pass-by-value leading datums.
565 : */
566 :
567 : #define ST_SORT qsort_tuple_unsigned
568 : #define ST_ELEMENT_TYPE SortTuple
569 : #define ST_COMPARE(a, b, state) qsort_tuple_unsigned_compare(a, b, state)
570 : #define ST_COMPARE_ARG_TYPE Tuplesortstate
571 : #define ST_CHECK_FOR_INTERRUPTS
572 : #define ST_SCOPE static
573 : #define ST_DEFINE
574 : #include "lib/sort_template.h"
575 :
576 : #define ST_SORT qsort_tuple_signed
577 : #define ST_ELEMENT_TYPE SortTuple
578 : #define ST_COMPARE(a, b, state) qsort_tuple_signed_compare(a, b, state)
579 : #define ST_COMPARE_ARG_TYPE Tuplesortstate
580 : #define ST_CHECK_FOR_INTERRUPTS
581 : #define ST_SCOPE static
582 : #define ST_DEFINE
583 : #include "lib/sort_template.h"
584 :
585 : #define ST_SORT qsort_tuple_int32
586 : #define ST_ELEMENT_TYPE SortTuple
587 : #define ST_COMPARE(a, b, state) qsort_tuple_int32_compare(a, b, state)
588 : #define ST_COMPARE_ARG_TYPE Tuplesortstate
589 : #define ST_CHECK_FOR_INTERRUPTS
590 : #define ST_SCOPE static
591 : #define ST_DEFINE
592 : #include "lib/sort_template.h"
593 :
594 : #define ST_SORT qsort_tuple
595 : #define ST_ELEMENT_TYPE SortTuple
596 : #define ST_COMPARE_RUNTIME_POINTER
597 : #define ST_COMPARE_ARG_TYPE Tuplesortstate
598 : #define ST_CHECK_FOR_INTERRUPTS
599 : #define ST_SCOPE static
600 : #define ST_DECLARE
601 : #define ST_DEFINE
602 : #include "lib/sort_template.h"
603 :
604 : #define ST_SORT qsort_ssup
605 : #define ST_ELEMENT_TYPE SortTuple
606 : #define ST_COMPARE(a, b, ssup) \
607 : ApplySortComparator((a)->datum1, (a)->isnull1, \
608 : (b)->datum1, (b)->isnull1, (ssup))
609 : #define ST_COMPARE_ARG_TYPE SortSupportData
610 : #define ST_CHECK_FOR_INTERRUPTS
611 : #define ST_SCOPE static
612 : #define ST_DEFINE
613 : #include "lib/sort_template.h"
614 :
615 : /*
616 : * tuplesort_begin_xxx
617 : *
618 : * Initialize for a tuple sort operation.
619 : *
620 : * After calling tuplesort_begin, the caller should call tuplesort_putXXX
621 : * zero or more times, then call tuplesort_performsort when all the tuples
622 : * have been supplied. After performsort, retrieve the tuples in sorted
623 : * order by calling tuplesort_getXXX until it returns false/NULL. (If random
624 : * access was requested, rescan, markpos, and restorepos can also be called.)
625 : * Call tuplesort_end to terminate the operation and release memory/disk space.
626 : *
627 : * Each variant of tuplesort_begin has a workMem parameter specifying the
628 : * maximum number of kilobytes of RAM to use before spilling data to disk.
629 : * (The normal value of this parameter is work_mem, but some callers use
630 : * other values.) Each variant also has a sortopt which is a bitmask of
631 : * sort options. See TUPLESORT_* definitions in tuplesort.h
632 : */
633 :
634 : Tuplesortstate *
635 278896 : tuplesort_begin_common(int workMem, SortCoordinate coordinate, int sortopt)
636 : {
637 : Tuplesortstate *state;
638 : MemoryContext maincontext;
639 : MemoryContext sortcontext;
640 : MemoryContext oldcontext;
641 :
642 : /* See leader_takeover_tapes() remarks on random access support */
643 278896 : if (coordinate && (sortopt & TUPLESORT_RANDOMACCESS))
644 0 : elog(ERROR, "random access disallowed under parallel sort");
645 :
646 : /*
647 : * Memory context surviving tuplesort_reset. This memory context holds
648 : * data which is useful to keep while sorting multiple similar batches.
649 : */
650 278896 : maincontext = AllocSetContextCreate(CurrentMemoryContext,
651 : "TupleSort main",
652 : ALLOCSET_DEFAULT_SIZES);
653 :
654 : /*
655 : * Create a working memory context for one sort operation. The content of
656 : * this context is deleted by tuplesort_reset.
657 : */
658 278896 : sortcontext = AllocSetContextCreate(maincontext,
659 : "TupleSort sort",
660 : ALLOCSET_DEFAULT_SIZES);
661 :
662 : /*
663 : * Additionally a working memory context for tuples is setup in
664 : * tuplesort_begin_batch.
665 : */
666 :
667 : /*
668 : * Make the Tuplesortstate within the per-sortstate context. This way, we
669 : * don't need a separate pfree() operation for it at shutdown.
670 : */
671 278896 : oldcontext = MemoryContextSwitchTo(maincontext);
672 :
673 278896 : state = palloc0_object(Tuplesortstate);
674 :
675 278896 : if (trace_sort)
676 0 : pg_rusage_init(&state->ru_start);
677 :
678 278896 : state->base.sortopt = sortopt;
679 278896 : state->base.tuples = true;
680 278896 : state->abbrevNext = 10;
681 :
682 : /*
683 : * workMem is forced to be at least 64KB, the current minimum valid value
684 : * for the work_mem GUC. This is a defense against parallel sort callers
685 : * that divide out memory among many workers in a way that leaves each
686 : * with very little memory.
687 : */
688 278896 : state->allowedMem = Max(workMem, 64) * (int64) 1024;
689 278896 : state->base.sortcontext = sortcontext;
690 278896 : state->base.maincontext = maincontext;
691 :
692 278896 : state->memtupsize = INITIAL_MEMTUPSIZE;
693 278896 : state->memtuples = NULL;
694 :
695 : /*
696 : * After all of the other non-parallel-related state, we setup all of the
697 : * state needed for each batch.
698 : */
699 278896 : tuplesort_begin_batch(state);
700 :
701 : /*
702 : * Initialize parallel-related state based on coordination information
703 : * from caller
704 : */
705 278896 : if (!coordinate)
706 : {
707 : /* Serial sort */
708 278068 : state->shared = NULL;
709 278068 : state->worker = -1;
710 278068 : state->nParticipants = -1;
711 : }
712 828 : else if (coordinate->isWorker)
713 : {
714 : /* Parallel worker produces exactly one final run from all input */
715 564 : state->shared = coordinate->sharedsort;
716 564 : state->worker = worker_get_identifier(state);
717 564 : state->nParticipants = -1;
718 : }
719 : else
720 : {
721 : /* Parallel leader state only used for final merge */
722 264 : state->shared = coordinate->sharedsort;
723 264 : state->worker = -1;
724 264 : state->nParticipants = coordinate->nParticipants;
725 : Assert(state->nParticipants >= 1);
726 : }
727 :
728 278896 : MemoryContextSwitchTo(oldcontext);
729 :
730 278896 : return state;
731 : }
732 :
733 : /*
734 : * tuplesort_begin_batch
735 : *
736 : * Setup, or reset, all state need for processing a new set of tuples with this
737 : * sort state. Called both from tuplesort_begin_common (the first time sorting
738 : * with this sort state) and tuplesort_reset (for subsequent usages).
739 : */
740 : static void
741 282232 : tuplesort_begin_batch(Tuplesortstate *state)
742 : {
743 : MemoryContext oldcontext;
744 :
745 282232 : oldcontext = MemoryContextSwitchTo(state->base.maincontext);
746 :
747 : /*
748 : * Caller tuple (e.g. IndexTuple) memory context.
749 : *
750 : * A dedicated child context used exclusively for caller passed tuples
751 : * eases memory management. Resetting at key points reduces
752 : * fragmentation. Note that the memtuples array of SortTuples is allocated
753 : * in the parent context, not this context, because there is no need to
754 : * free memtuples early. For bounded sorts, tuples may be pfreed in any
755 : * order, so we use a regular aset.c context so that it can make use of
756 : * free'd memory. When the sort is not bounded, we make use of a bump.c
757 : * context as this keeps allocations more compact with less wastage.
758 : * Allocations are also slightly more CPU efficient.
759 : */
760 282232 : if (TupleSortUseBumpTupleCxt(state->base.sortopt))
761 280866 : state->base.tuplecontext = BumpContextCreate(state->base.sortcontext,
762 : "Caller tuples",
763 : ALLOCSET_DEFAULT_SIZES);
764 : else
765 1366 : state->base.tuplecontext = AllocSetContextCreate(state->base.sortcontext,
766 : "Caller tuples",
767 : ALLOCSET_DEFAULT_SIZES);
768 :
769 :
770 282232 : state->status = TSS_INITIAL;
771 282232 : state->bounded = false;
772 282232 : state->boundUsed = false;
773 :
774 282232 : state->availMem = state->allowedMem;
775 :
776 282232 : state->tapeset = NULL;
777 :
778 282232 : state->memtupcount = 0;
779 :
780 282232 : state->growmemtuples = true;
781 282232 : state->slabAllocatorUsed = false;
782 282232 : if (state->memtuples != NULL && state->memtupsize != INITIAL_MEMTUPSIZE)
783 : {
784 72 : pfree(state->memtuples);
785 72 : state->memtuples = NULL;
786 72 : state->memtupsize = INITIAL_MEMTUPSIZE;
787 : }
788 282232 : if (state->memtuples == NULL)
789 : {
790 278968 : state->memtuples = (SortTuple *) palloc(state->memtupsize * sizeof(SortTuple));
791 278968 : USEMEM(state, GetMemoryChunkSpace(state->memtuples));
792 : }
793 :
794 : /* workMem must be large enough for the minimal memtuples array */
795 282232 : if (LACKMEM(state))
796 0 : elog(ERROR, "insufficient memory allowed for sort");
797 :
798 282232 : state->currentRun = 0;
799 :
800 : /*
801 : * Tape variables (inputTapes, outputTapes, etc.) will be initialized by
802 : * inittapes(), if needed.
803 : */
804 :
805 282232 : state->result_tape = NULL; /* flag that result tape has not been formed */
806 :
807 282232 : MemoryContextSwitchTo(oldcontext);
808 282232 : }
809 :
810 : /*
811 : * tuplesort_set_bound
812 : *
813 : * Advise tuplesort that at most the first N result tuples are required.
814 : *
815 : * Must be called before inserting any tuples. (Actually, we could allow it
816 : * as long as the sort hasn't spilled to disk, but there seems no need for
817 : * delayed calls at the moment.)
818 : *
819 : * This is a hint only. The tuplesort may still return more tuples than
820 : * requested. Parallel leader tuplesorts will always ignore the hint.
821 : */
822 : void
823 1232 : tuplesort_set_bound(Tuplesortstate *state, int64 bound)
824 : {
825 : /* Assert we're called before loading any tuples */
826 : Assert(state->status == TSS_INITIAL && state->memtupcount == 0);
827 : /* Assert we allow bounded sorts */
828 : Assert(state->base.sortopt & TUPLESORT_ALLOWBOUNDED);
829 : /* Can't set the bound twice, either */
830 : Assert(!state->bounded);
831 : /* Also, this shouldn't be called in a parallel worker */
832 : Assert(!WORKER(state));
833 :
834 : /* Parallel leader allows but ignores hint */
835 1232 : if (LEADER(state))
836 0 : return;
837 :
838 : #ifdef DEBUG_BOUNDED_SORT
839 : /* Honor GUC setting that disables the feature (for easy testing) */
840 : if (!optimize_bounded_sort)
841 : return;
842 : #endif
843 :
844 : /* We want to be able to compute bound * 2, so limit the setting */
845 1232 : if (bound > (int64) (INT_MAX / 2))
846 0 : return;
847 :
848 1232 : state->bounded = true;
849 1232 : state->bound = (int) bound;
850 :
851 : /*
852 : * Bounded sorts are not an effective target for abbreviated key
853 : * optimization. Disable by setting state to be consistent with no
854 : * abbreviation support.
855 : */
856 1232 : state->base.sortKeys->abbrev_converter = NULL;
857 1232 : if (state->base.sortKeys->abbrev_full_comparator)
858 16 : state->base.sortKeys->comparator = state->base.sortKeys->abbrev_full_comparator;
859 :
860 : /* Not strictly necessary, but be tidy */
861 1232 : state->base.sortKeys->abbrev_abort = NULL;
862 1232 : state->base.sortKeys->abbrev_full_comparator = NULL;
863 : }
864 :
865 : /*
866 : * tuplesort_used_bound
867 : *
868 : * Allow callers to find out if the sort state was able to use a bound.
869 : */
870 : bool
871 380 : tuplesort_used_bound(Tuplesortstate *state)
872 : {
873 380 : return state->boundUsed;
874 : }
875 :
876 : /*
877 : * tuplesort_free
878 : *
879 : * Internal routine for freeing resources of tuplesort.
880 : */
881 : static void
882 281970 : tuplesort_free(Tuplesortstate *state)
883 : {
884 : /* context swap probably not needed, but let's be safe */
885 281970 : MemoryContext oldcontext = MemoryContextSwitchTo(state->base.sortcontext);
886 : int64 spaceUsed;
887 :
888 281970 : if (state->tapeset)
889 898 : spaceUsed = LogicalTapeSetBlocks(state->tapeset);
890 : else
891 281072 : spaceUsed = (state->allowedMem - state->availMem + 1023) / 1024;
892 :
893 : /*
894 : * Delete temporary "tape" files, if any.
895 : *
896 : * We don't bother to destroy the individual tapes here. They will go away
897 : * with the sortcontext. (In TSS_FINALMERGE state, we have closed
898 : * finished tapes already.)
899 : */
900 281970 : if (state->tapeset)
901 898 : LogicalTapeSetClose(state->tapeset);
902 :
903 281970 : if (trace_sort)
904 : {
905 0 : if (state->tapeset)
906 0 : elog(LOG, "%s of worker %d ended, %" PRId64 " disk blocks used: %s",
907 : SERIAL(state) ? "external sort" : "parallel external sort",
908 : state->worker, spaceUsed, pg_rusage_show(&state->ru_start));
909 : else
910 0 : elog(LOG, "%s of worker %d ended, %" PRId64 " KB used: %s",
911 : SERIAL(state) ? "internal sort" : "unperformed parallel sort",
912 : state->worker, spaceUsed, pg_rusage_show(&state->ru_start));
913 : }
914 :
915 : TRACE_POSTGRESQL_SORT_DONE(state->tapeset != NULL, spaceUsed);
916 :
917 281970 : FREESTATE(state);
918 281970 : MemoryContextSwitchTo(oldcontext);
919 :
920 : /*
921 : * Free the per-sort memory context, thereby releasing all working memory.
922 : */
923 281970 : MemoryContextReset(state->base.sortcontext);
924 281970 : }
925 :
926 : /*
927 : * tuplesort_end
928 : *
929 : * Release resources and clean up.
930 : *
931 : * NOTE: after calling this, any pointers returned by tuplesort_getXXX are
932 : * pointing to garbage. Be careful not to attempt to use or free such
933 : * pointers afterwards!
934 : */
935 : void
936 278634 : tuplesort_end(Tuplesortstate *state)
937 : {
938 278634 : tuplesort_free(state);
939 :
940 : /*
941 : * Free the main memory context, including the Tuplesortstate struct
942 : * itself.
943 : */
944 278634 : MemoryContextDelete(state->base.maincontext);
945 278634 : }
946 :
947 : /*
948 : * tuplesort_updatemax
949 : *
950 : * Update maximum resource usage statistics.
951 : */
952 : static void
953 3732 : tuplesort_updatemax(Tuplesortstate *state)
954 : {
955 : int64 spaceUsed;
956 : bool isSpaceDisk;
957 :
958 : /*
959 : * Note: it might seem we should provide both memory and disk usage for a
960 : * disk-based sort. However, the current code doesn't track memory space
961 : * accurately once we have begun to return tuples to the caller (since we
962 : * don't account for pfree's the caller is expected to do), so we cannot
963 : * rely on availMem in a disk sort. This does not seem worth the overhead
964 : * to fix. Is it worth creating an API for the memory context code to
965 : * tell us how much is actually used in sortcontext?
966 : */
967 3732 : if (state->tapeset)
968 : {
969 6 : isSpaceDisk = true;
970 6 : spaceUsed = LogicalTapeSetBlocks(state->tapeset) * BLCKSZ;
971 : }
972 : else
973 : {
974 3726 : isSpaceDisk = false;
975 3726 : spaceUsed = state->allowedMem - state->availMem;
976 : }
977 :
978 : /*
979 : * Sort evicts data to the disk when it wasn't able to fit that data into
980 : * main memory. This is why we assume space used on the disk to be more
981 : * important for tracking resource usage than space used in memory. Note
982 : * that the amount of space occupied by some tupleset on the disk might be
983 : * less than amount of space occupied by the same tupleset in memory due
984 : * to more compact representation.
985 : */
986 3732 : if ((isSpaceDisk && !state->isMaxSpaceDisk) ||
987 3726 : (isSpaceDisk == state->isMaxSpaceDisk && spaceUsed > state->maxSpace))
988 : {
989 506 : state->maxSpace = spaceUsed;
990 506 : state->isMaxSpaceDisk = isSpaceDisk;
991 506 : state->maxSpaceStatus = state->status;
992 : }
993 3732 : }
994 :
995 : /*
996 : * tuplesort_reset
997 : *
998 : * Reset the tuplesort. Reset all the data in the tuplesort, but leave the
999 : * meta-information in. After tuplesort_reset, tuplesort is ready to start
1000 : * a new sort. This allows avoiding recreation of tuple sort states (and
1001 : * save resources) when sorting multiple small batches.
1002 : */
1003 : void
1004 3336 : tuplesort_reset(Tuplesortstate *state)
1005 : {
1006 3336 : tuplesort_updatemax(state);
1007 3336 : tuplesort_free(state);
1008 :
1009 : /*
1010 : * After we've freed up per-batch memory, re-setup all of the state common
1011 : * to both the first batch and any subsequent batch.
1012 : */
1013 3336 : tuplesort_begin_batch(state);
1014 :
1015 3336 : state->lastReturnedTuple = NULL;
1016 3336 : state->slabMemoryBegin = NULL;
1017 3336 : state->slabMemoryEnd = NULL;
1018 3336 : state->slabFreeHead = NULL;
1019 3336 : }
1020 :
1021 : /*
1022 : * Grow the memtuples[] array, if possible within our memory constraint. We
1023 : * must not exceed INT_MAX tuples in memory or the caller-provided memory
1024 : * limit. Return true if we were able to enlarge the array, false if not.
1025 : *
1026 : * Normally, at each increment we double the size of the array. When doing
1027 : * that would exceed a limit, we attempt one last, smaller increase (and then
1028 : * clear the growmemtuples flag so we don't try any more). That allows us to
1029 : * use memory as fully as permitted; sticking to the pure doubling rule could
1030 : * result in almost half going unused. Because availMem moves around with
1031 : * tuple addition/removal, we need some rule to prevent making repeated small
1032 : * increases in memtupsize, which would just be useless thrashing. The
1033 : * growmemtuples flag accomplishes that and also prevents useless
1034 : * recalculations in this function.
1035 : */
1036 : static bool
1037 8504 : grow_memtuples(Tuplesortstate *state)
1038 : {
1039 : int newmemtupsize;
1040 8504 : int memtupsize = state->memtupsize;
1041 8504 : int64 memNowUsed = state->allowedMem - state->availMem;
1042 :
1043 : /* Forget it if we've already maxed out memtuples, per comment above */
1044 8504 : if (!state->growmemtuples)
1045 140 : return false;
1046 :
1047 : /* Select new value of memtupsize */
1048 8364 : if (memNowUsed <= state->availMem)
1049 : {
1050 : /*
1051 : * We've used no more than half of allowedMem; double our usage,
1052 : * clamping at INT_MAX tuples.
1053 : */
1054 8222 : if (memtupsize < INT_MAX / 2)
1055 8222 : newmemtupsize = memtupsize * 2;
1056 : else
1057 : {
1058 0 : newmemtupsize = INT_MAX;
1059 0 : state->growmemtuples = false;
1060 : }
1061 : }
1062 : else
1063 : {
1064 : /*
1065 : * This will be the last increment of memtupsize. Abandon doubling
1066 : * strategy and instead increase as much as we safely can.
1067 : *
1068 : * To stay within allowedMem, we can't increase memtupsize by more
1069 : * than availMem / sizeof(SortTuple) elements. In practice, we want
1070 : * to increase it by considerably less, because we need to leave some
1071 : * space for the tuples to which the new array slots will refer. We
1072 : * assume the new tuples will be about the same size as the tuples
1073 : * we've already seen, and thus we can extrapolate from the space
1074 : * consumption so far to estimate an appropriate new size for the
1075 : * memtuples array. The optimal value might be higher or lower than
1076 : * this estimate, but it's hard to know that in advance. We again
1077 : * clamp at INT_MAX tuples.
1078 : *
1079 : * This calculation is safe against enlarging the array so much that
1080 : * LACKMEM becomes true, because the memory currently used includes
1081 : * the present array; thus, there would be enough allowedMem for the
1082 : * new array elements even if no other memory were currently used.
1083 : *
1084 : * We do the arithmetic in float8, because otherwise the product of
1085 : * memtupsize and allowedMem could overflow. Any inaccuracy in the
1086 : * result should be insignificant; but even if we computed a
1087 : * completely insane result, the checks below will prevent anything
1088 : * really bad from happening.
1089 : */
1090 : double grow_ratio;
1091 :
1092 142 : grow_ratio = (double) state->allowedMem / (double) memNowUsed;
1093 142 : if (memtupsize * grow_ratio < INT_MAX)
1094 142 : newmemtupsize = (int) (memtupsize * grow_ratio);
1095 : else
1096 0 : newmemtupsize = INT_MAX;
1097 :
1098 : /* We won't make any further enlargement attempts */
1099 142 : state->growmemtuples = false;
1100 : }
1101 :
1102 : /* Must enlarge array by at least one element, else report failure */
1103 8364 : if (newmemtupsize <= memtupsize)
1104 0 : goto noalloc;
1105 :
1106 : /*
1107 : * On a 32-bit machine, allowedMem could exceed MaxAllocHugeSize. Clamp
1108 : * to ensure our request won't be rejected. Note that we can easily
1109 : * exhaust address space before facing this outcome. (This is presently
1110 : * impossible due to guc.c's MAX_KILOBYTES limitation on work_mem, but
1111 : * don't rely on that at this distance.)
1112 : */
1113 8364 : if ((Size) newmemtupsize >= MaxAllocHugeSize / sizeof(SortTuple))
1114 : {
1115 0 : newmemtupsize = (int) (MaxAllocHugeSize / sizeof(SortTuple));
1116 0 : state->growmemtuples = false; /* can't grow any more */
1117 : }
1118 :
1119 : /*
1120 : * We need to be sure that we do not cause LACKMEM to become true, else
1121 : * the space management algorithm will go nuts. The code above should
1122 : * never generate a dangerous request, but to be safe, check explicitly
1123 : * that the array growth fits within availMem. (We could still cause
1124 : * LACKMEM if the memory chunk overhead associated with the memtuples
1125 : * array were to increase. That shouldn't happen because we chose the
1126 : * initial array size large enough to ensure that palloc will be treating
1127 : * both old and new arrays as separate chunks. But we'll check LACKMEM
1128 : * explicitly below just in case.)
1129 : */
1130 8364 : if (state->availMem < (int64) ((newmemtupsize - memtupsize) * sizeof(SortTuple)))
1131 0 : goto noalloc;
1132 :
1133 : /* OK, do it */
1134 8364 : FREEMEM(state, GetMemoryChunkSpace(state->memtuples));
1135 8364 : state->memtupsize = newmemtupsize;
1136 8364 : state->memtuples = (SortTuple *)
1137 8364 : repalloc_huge(state->memtuples,
1138 8364 : state->memtupsize * sizeof(SortTuple));
1139 8364 : USEMEM(state, GetMemoryChunkSpace(state->memtuples));
1140 8364 : if (LACKMEM(state))
1141 0 : elog(ERROR, "unexpected out-of-memory situation in tuplesort");
1142 8364 : return true;
1143 :
1144 0 : noalloc:
1145 : /* If for any reason we didn't realloc, shut off future attempts */
1146 0 : state->growmemtuples = false;
1147 0 : return false;
1148 : }
1149 :
1150 : /*
1151 : * Shared code for tuple and datum cases.
1152 : */
1153 : void
1154 31245408 : tuplesort_puttuple_common(Tuplesortstate *state, SortTuple *tuple,
1155 : bool useAbbrev, Size tuplen)
1156 : {
1157 31245408 : MemoryContext oldcontext = MemoryContextSwitchTo(state->base.sortcontext);
1158 :
1159 : Assert(!LEADER(state));
1160 :
1161 : /* account for the memory used for this tuple */
1162 31245408 : USEMEM(state, tuplen);
1163 31245408 : state->tupleMem += tuplen;
1164 :
1165 31245408 : if (!useAbbrev)
1166 : {
1167 : /*
1168 : * Leave ordinary Datum representation, or NULL value. If there is a
1169 : * converter it won't expect NULL values, and cost model is not
1170 : * required to account for NULL, so in that case we avoid calling
1171 : * converter and just set datum1 to zeroed representation (to be
1172 : * consistent, and to support cheap inequality tests for NULL
1173 : * abbreviated keys).
1174 : */
1175 : }
1176 4445810 : else if (!consider_abort_common(state))
1177 : {
1178 : /* Store abbreviated key representation */
1179 4445714 : tuple->datum1 = state->base.sortKeys->abbrev_converter(tuple->datum1,
1180 : state->base.sortKeys);
1181 : }
1182 : else
1183 : {
1184 : /*
1185 : * Set state to be consistent with never trying abbreviation.
1186 : *
1187 : * Alter datum1 representation in already-copied tuples, so as to
1188 : * ensure a consistent representation (current tuple was just
1189 : * handled). It does not matter if some dumped tuples are already
1190 : * sorted on tape, since serialized tuples lack abbreviated keys
1191 : * (TSS_BUILDRUNS state prevents control reaching here in any case).
1192 : */
1193 96 : REMOVEABBREV(state, state->memtuples, state->memtupcount);
1194 : }
1195 :
1196 31245408 : switch (state->status)
1197 : {
1198 26382942 : case TSS_INITIAL:
1199 :
1200 : /*
1201 : * Save the tuple into the unsorted array. First, grow the array
1202 : * as needed. Note that we try to grow the array when there is
1203 : * still one free slot remaining --- if we fail, there'll still be
1204 : * room to store the incoming tuple, and then we'll switch to
1205 : * tape-based operation.
1206 : */
1207 26382942 : if (state->memtupcount >= state->memtupsize - 1)
1208 : {
1209 8504 : (void) grow_memtuples(state);
1210 : Assert(state->memtupcount < state->memtupsize);
1211 : }
1212 26382942 : state->memtuples[state->memtupcount++] = *tuple;
1213 :
1214 : /*
1215 : * Check if it's time to switch over to a bounded heapsort. We do
1216 : * so if the input tuple count exceeds twice the desired tuple
1217 : * count (this is a heuristic for where heapsort becomes cheaper
1218 : * than a quicksort), or if we've just filled workMem and have
1219 : * enough tuples to meet the bound.
1220 : *
1221 : * Note that once we enter TSS_BOUNDED state we will always try to
1222 : * complete the sort that way. In the worst case, if later input
1223 : * tuples are larger than earlier ones, this might cause us to
1224 : * exceed workMem significantly.
1225 : */
1226 26382942 : if (state->bounded &&
1227 47316 : (state->memtupcount > state->bound * 2 ||
1228 46908 : (state->memtupcount > state->bound && LACKMEM(state))))
1229 : {
1230 408 : if (trace_sort)
1231 0 : elog(LOG, "switching to bounded heapsort at %d tuples: %s",
1232 : state->memtupcount,
1233 : pg_rusage_show(&state->ru_start));
1234 408 : make_bounded_heap(state);
1235 408 : MemoryContextSwitchTo(oldcontext);
1236 408 : return;
1237 : }
1238 :
1239 : /*
1240 : * Done if we still fit in available memory and have array slots.
1241 : */
1242 26382534 : if (state->memtupcount < state->memtupsize && !LACKMEM(state))
1243 : {
1244 26382394 : MemoryContextSwitchTo(oldcontext);
1245 26382394 : return;
1246 : }
1247 :
1248 : /*
1249 : * Nope; time to switch to tape-based operation.
1250 : */
1251 140 : inittapes(state, true);
1252 :
1253 : /*
1254 : * Dump all tuples.
1255 : */
1256 140 : dumptuples(state, false);
1257 140 : break;
1258 :
1259 3750180 : case TSS_BOUNDED:
1260 :
1261 : /*
1262 : * We don't want to grow the array here, so check whether the new
1263 : * tuple can be discarded before putting it in. This should be a
1264 : * good speed optimization, too, since when there are many more
1265 : * input tuples than the bound, most input tuples can be discarded
1266 : * with just this one comparison. Note that because we currently
1267 : * have the sort direction reversed, we must check for <= not >=.
1268 : */
1269 3750180 : if (COMPARETUP(state, tuple, &state->memtuples[0]) <= 0)
1270 : {
1271 : /* new tuple <= top of the heap, so we can discard it */
1272 3247194 : free_sort_tuple(state, tuple);
1273 3247194 : CHECK_FOR_INTERRUPTS();
1274 : }
1275 : else
1276 : {
1277 : /* discard top of heap, replacing it with the new tuple */
1278 502986 : free_sort_tuple(state, &state->memtuples[0]);
1279 502986 : tuplesort_heap_replace_top(state, tuple);
1280 : }
1281 3750180 : break;
1282 :
1283 1112286 : case TSS_BUILDRUNS:
1284 :
1285 : /*
1286 : * Save the tuple into the unsorted array (there must be space)
1287 : */
1288 1112286 : state->memtuples[state->memtupcount++] = *tuple;
1289 :
1290 : /*
1291 : * If we are over the memory limit, dump all tuples.
1292 : */
1293 1112286 : dumptuples(state, false);
1294 1112286 : break;
1295 :
1296 0 : default:
1297 0 : elog(ERROR, "invalid tuplesort state");
1298 : break;
1299 : }
1300 4862606 : MemoryContextSwitchTo(oldcontext);
1301 : }
1302 :
1303 : static bool
1304 4445810 : consider_abort_common(Tuplesortstate *state)
1305 : {
1306 : Assert(state->base.sortKeys[0].abbrev_converter != NULL);
1307 : Assert(state->base.sortKeys[0].abbrev_abort != NULL);
1308 : Assert(state->base.sortKeys[0].abbrev_full_comparator != NULL);
1309 :
1310 : /*
1311 : * Check effectiveness of abbreviation optimization. Consider aborting
1312 : * when still within memory limit.
1313 : */
1314 4445810 : if (state->status == TSS_INITIAL &&
1315 3973380 : state->memtupcount >= state->abbrevNext)
1316 : {
1317 5144 : state->abbrevNext *= 2;
1318 :
1319 : /*
1320 : * Check opclass-supplied abbreviation abort routine. It may indicate
1321 : * that abbreviation should not proceed.
1322 : */
1323 5144 : if (!state->base.sortKeys->abbrev_abort(state->memtupcount,
1324 : state->base.sortKeys))
1325 5048 : return false;
1326 :
1327 : /*
1328 : * Finally, restore authoritative comparator, and indicate that
1329 : * abbreviation is not in play by setting abbrev_converter to NULL
1330 : */
1331 96 : state->base.sortKeys[0].comparator = state->base.sortKeys[0].abbrev_full_comparator;
1332 96 : state->base.sortKeys[0].abbrev_converter = NULL;
1333 : /* Not strictly necessary, but be tidy */
1334 96 : state->base.sortKeys[0].abbrev_abort = NULL;
1335 96 : state->base.sortKeys[0].abbrev_full_comparator = NULL;
1336 :
1337 : /* Give up - expect original pass-by-value representation */
1338 96 : return true;
1339 : }
1340 :
1341 4440666 : return false;
1342 : }
1343 :
1344 : /*
1345 : * All tuples have been provided; finish the sort.
1346 : */
1347 : void
1348 239304 : tuplesort_performsort(Tuplesortstate *state)
1349 : {
1350 239304 : MemoryContext oldcontext = MemoryContextSwitchTo(state->base.sortcontext);
1351 :
1352 239304 : if (trace_sort)
1353 0 : elog(LOG, "performsort of worker %d starting: %s",
1354 : state->worker, pg_rusage_show(&state->ru_start));
1355 :
1356 239304 : switch (state->status)
1357 : {
1358 238756 : case TSS_INITIAL:
1359 :
1360 : /*
1361 : * We were able to accumulate all the tuples within the allowed
1362 : * amount of memory, or leader to take over worker tapes
1363 : */
1364 238756 : if (SERIAL(state))
1365 : {
1366 : /* Just qsort 'em and we're done */
1367 237998 : tuplesort_sort_memtuples(state);
1368 237908 : state->status = TSS_SORTEDINMEM;
1369 : }
1370 758 : else if (WORKER(state))
1371 : {
1372 : /*
1373 : * Parallel workers must still dump out tuples to tape. No
1374 : * merge is required to produce single output run, though.
1375 : */
1376 564 : inittapes(state, false);
1377 564 : dumptuples(state, true);
1378 564 : worker_nomergeruns(state);
1379 564 : state->status = TSS_SORTEDONTAPE;
1380 : }
1381 : else
1382 : {
1383 : /*
1384 : * Leader will take over worker tapes and merge worker runs.
1385 : * Note that mergeruns sets the correct state->status.
1386 : */
1387 194 : leader_takeover_tapes(state);
1388 194 : mergeruns(state);
1389 : }
1390 238666 : state->current = 0;
1391 238666 : state->eof_reached = false;
1392 238666 : state->markpos_block = 0L;
1393 238666 : state->markpos_offset = 0;
1394 238666 : state->markpos_eof = false;
1395 238666 : break;
1396 :
1397 408 : case TSS_BOUNDED:
1398 :
1399 : /*
1400 : * We were able to accumulate all the tuples required for output
1401 : * in memory, using a heap to eliminate excess tuples. Now we
1402 : * have to transform the heap to a properly-sorted array. Note
1403 : * that sort_bounded_heap sets the correct state->status.
1404 : */
1405 408 : sort_bounded_heap(state);
1406 408 : state->current = 0;
1407 408 : state->eof_reached = false;
1408 408 : state->markpos_offset = 0;
1409 408 : state->markpos_eof = false;
1410 408 : break;
1411 :
1412 140 : case TSS_BUILDRUNS:
1413 :
1414 : /*
1415 : * Finish tape-based sort. First, flush all tuples remaining in
1416 : * memory out to tape; then merge until we have a single remaining
1417 : * run (or, if !randomAccess and !WORKER(), one run per tape).
1418 : * Note that mergeruns sets the correct state->status.
1419 : */
1420 140 : dumptuples(state, true);
1421 140 : mergeruns(state);
1422 140 : state->eof_reached = false;
1423 140 : state->markpos_block = 0L;
1424 140 : state->markpos_offset = 0;
1425 140 : state->markpos_eof = false;
1426 140 : break;
1427 :
1428 0 : default:
1429 0 : elog(ERROR, "invalid tuplesort state");
1430 : break;
1431 : }
1432 :
1433 239214 : if (trace_sort)
1434 : {
1435 0 : if (state->status == TSS_FINALMERGE)
1436 0 : elog(LOG, "performsort of worker %d done (except %d-way final merge): %s",
1437 : state->worker, state->nInputTapes,
1438 : pg_rusage_show(&state->ru_start));
1439 : else
1440 0 : elog(LOG, "performsort of worker %d done: %s",
1441 : state->worker, pg_rusage_show(&state->ru_start));
1442 : }
1443 :
1444 239214 : MemoryContextSwitchTo(oldcontext);
1445 239214 : }
1446 :
1447 : /*
1448 : * Internal routine to fetch the next tuple in either forward or back
1449 : * direction into *stup. Returns false if no more tuples.
1450 : * Returned tuple belongs to tuplesort memory context, and must not be freed
1451 : * by caller. Note that fetched tuple is stored in memory that may be
1452 : * recycled by any future fetch.
1453 : */
1454 : bool
1455 28869564 : tuplesort_gettuple_common(Tuplesortstate *state, bool forward,
1456 : SortTuple *stup)
1457 : {
1458 : unsigned int tuplen;
1459 : size_t nmoved;
1460 :
1461 : Assert(!WORKER(state));
1462 :
1463 28869564 : switch (state->status)
1464 : {
1465 23884570 : case TSS_SORTEDINMEM:
1466 : Assert(forward || state->base.sortopt & TUPLESORT_RANDOMACCESS);
1467 : Assert(!state->slabAllocatorUsed);
1468 23884570 : if (forward)
1469 : {
1470 23884504 : if (state->current < state->memtupcount)
1471 : {
1472 23647506 : *stup = state->memtuples[state->current++];
1473 23647506 : return true;
1474 : }
1475 236998 : state->eof_reached = true;
1476 :
1477 : /*
1478 : * Complain if caller tries to retrieve more tuples than
1479 : * originally asked for in a bounded sort. This is because
1480 : * returning EOF here might be the wrong thing.
1481 : */
1482 236998 : if (state->bounded && state->current >= state->bound)
1483 0 : elog(ERROR, "retrieved too many tuples in a bounded sort");
1484 :
1485 236998 : return false;
1486 : }
1487 : else
1488 : {
1489 66 : if (state->current <= 0)
1490 0 : return false;
1491 :
1492 : /*
1493 : * if all tuples are fetched already then we return last
1494 : * tuple, else - tuple before last returned.
1495 : */
1496 66 : if (state->eof_reached)
1497 12 : state->eof_reached = false;
1498 : else
1499 : {
1500 54 : state->current--; /* last returned tuple */
1501 54 : if (state->current <= 0)
1502 6 : return false;
1503 : }
1504 60 : *stup = state->memtuples[state->current - 1];
1505 60 : return true;
1506 : }
1507 : break;
1508 :
1509 303000 : case TSS_SORTEDONTAPE:
1510 : Assert(forward || state->base.sortopt & TUPLESORT_RANDOMACCESS);
1511 : Assert(state->slabAllocatorUsed);
1512 :
1513 : /*
1514 : * The slot that held the tuple that we returned in previous
1515 : * gettuple call can now be reused.
1516 : */
1517 303000 : if (state->lastReturnedTuple)
1518 : {
1519 152850 : RELEASE_SLAB_SLOT(state, state->lastReturnedTuple);
1520 152850 : state->lastReturnedTuple = NULL;
1521 : }
1522 :
1523 303000 : if (forward)
1524 : {
1525 302970 : if (state->eof_reached)
1526 0 : return false;
1527 :
1528 302970 : if ((tuplen = getlen(state->result_tape, true)) != 0)
1529 : {
1530 302940 : READTUP(state, stup, state->result_tape, tuplen);
1531 :
1532 : /*
1533 : * Remember the tuple we return, so that we can recycle
1534 : * its memory on next call. (This can be NULL, in the
1535 : * !state->tuples case).
1536 : */
1537 302940 : state->lastReturnedTuple = stup->tuple;
1538 :
1539 302940 : return true;
1540 : }
1541 : else
1542 : {
1543 30 : state->eof_reached = true;
1544 30 : return false;
1545 : }
1546 : }
1547 :
1548 : /*
1549 : * Backward.
1550 : *
1551 : * if all tuples are fetched already then we return last tuple,
1552 : * else - tuple before last returned.
1553 : */
1554 30 : if (state->eof_reached)
1555 : {
1556 : /*
1557 : * Seek position is pointing just past the zero tuplen at the
1558 : * end of file; back up to fetch last tuple's ending length
1559 : * word. If seek fails we must have a completely empty file.
1560 : */
1561 12 : nmoved = LogicalTapeBackspace(state->result_tape,
1562 : 2 * sizeof(unsigned int));
1563 12 : if (nmoved == 0)
1564 0 : return false;
1565 12 : else if (nmoved != 2 * sizeof(unsigned int))
1566 0 : elog(ERROR, "unexpected tape position");
1567 12 : state->eof_reached = false;
1568 : }
1569 : else
1570 : {
1571 : /*
1572 : * Back up and fetch previously-returned tuple's ending length
1573 : * word. If seek fails, assume we are at start of file.
1574 : */
1575 18 : nmoved = LogicalTapeBackspace(state->result_tape,
1576 : sizeof(unsigned int));
1577 18 : if (nmoved == 0)
1578 0 : return false;
1579 18 : else if (nmoved != sizeof(unsigned int))
1580 0 : elog(ERROR, "unexpected tape position");
1581 18 : tuplen = getlen(state->result_tape, false);
1582 :
1583 : /*
1584 : * Back up to get ending length word of tuple before it.
1585 : */
1586 18 : nmoved = LogicalTapeBackspace(state->result_tape,
1587 : tuplen + 2 * sizeof(unsigned int));
1588 18 : if (nmoved == tuplen + sizeof(unsigned int))
1589 : {
1590 : /*
1591 : * We backed up over the previous tuple, but there was no
1592 : * ending length word before it. That means that the prev
1593 : * tuple is the first tuple in the file. It is now the
1594 : * next to read in forward direction (not obviously right,
1595 : * but that is what in-memory case does).
1596 : */
1597 6 : return false;
1598 : }
1599 12 : else if (nmoved != tuplen + 2 * sizeof(unsigned int))
1600 0 : elog(ERROR, "bogus tuple length in backward scan");
1601 : }
1602 :
1603 24 : tuplen = getlen(state->result_tape, false);
1604 :
1605 : /*
1606 : * Now we have the length of the prior tuple, back up and read it.
1607 : * Note: READTUP expects we are positioned after the initial
1608 : * length word of the tuple, so back up to that point.
1609 : */
1610 24 : nmoved = LogicalTapeBackspace(state->result_tape,
1611 : tuplen);
1612 24 : if (nmoved != tuplen)
1613 0 : elog(ERROR, "bogus tuple length in backward scan");
1614 24 : READTUP(state, stup, state->result_tape, tuplen);
1615 :
1616 : /*
1617 : * Remember the tuple we return, so that we can recycle its memory
1618 : * on next call. (This can be NULL, in the Datum case).
1619 : */
1620 24 : state->lastReturnedTuple = stup->tuple;
1621 :
1622 24 : return true;
1623 :
1624 4681994 : case TSS_FINALMERGE:
1625 : Assert(forward);
1626 : /* We are managing memory ourselves, with the slab allocator. */
1627 : Assert(state->slabAllocatorUsed);
1628 :
1629 : /*
1630 : * The slab slot holding the tuple that we returned in previous
1631 : * gettuple call can now be reused.
1632 : */
1633 4681994 : if (state->lastReturnedTuple)
1634 : {
1635 4561652 : RELEASE_SLAB_SLOT(state, state->lastReturnedTuple);
1636 4561652 : state->lastReturnedTuple = NULL;
1637 : }
1638 :
1639 : /*
1640 : * This code should match the inner loop of mergeonerun().
1641 : */
1642 4681994 : if (state->memtupcount > 0)
1643 : {
1644 4681700 : int srcTapeIndex = state->memtuples[0].srctape;
1645 4681700 : LogicalTape *srcTape = state->inputTapes[srcTapeIndex];
1646 : SortTuple newtup;
1647 :
1648 4681700 : *stup = state->memtuples[0];
1649 :
1650 : /*
1651 : * Remember the tuple we return, so that we can recycle its
1652 : * memory on next call. (This can be NULL, in the Datum case).
1653 : */
1654 4681700 : state->lastReturnedTuple = stup->tuple;
1655 :
1656 : /*
1657 : * Pull next tuple from tape, and replace the returned tuple
1658 : * at top of the heap with it.
1659 : */
1660 4681700 : if (!mergereadnext(state, srcTape, &newtup))
1661 : {
1662 : /*
1663 : * If no more data, we've reached end of run on this tape.
1664 : * Remove the top node from the heap.
1665 : */
1666 430 : tuplesort_heap_delete_top(state);
1667 430 : state->nInputRuns--;
1668 :
1669 : /*
1670 : * Close the tape. It'd go away at the end of the sort
1671 : * anyway, but better to release the memory early.
1672 : */
1673 430 : LogicalTapeClose(srcTape);
1674 430 : return true;
1675 : }
1676 4681270 : newtup.srctape = srcTapeIndex;
1677 4681270 : tuplesort_heap_replace_top(state, &newtup);
1678 4681270 : return true;
1679 : }
1680 294 : return false;
1681 :
1682 0 : default:
1683 0 : elog(ERROR, "invalid tuplesort state");
1684 : return false; /* keep compiler quiet */
1685 : }
1686 : }
1687 :
1688 :
1689 : /*
1690 : * Advance over N tuples in either forward or back direction,
1691 : * without returning any data. N==0 is a no-op.
1692 : * Returns true if successful, false if ran out of tuples.
1693 : */
1694 : bool
1695 392 : tuplesort_skiptuples(Tuplesortstate *state, int64 ntuples, bool forward)
1696 : {
1697 : MemoryContext oldcontext;
1698 :
1699 : /*
1700 : * We don't actually support backwards skip yet, because no callers need
1701 : * it. The API is designed to allow for that later, though.
1702 : */
1703 : Assert(forward);
1704 : Assert(ntuples >= 0);
1705 : Assert(!WORKER(state));
1706 :
1707 392 : switch (state->status)
1708 : {
1709 368 : case TSS_SORTEDINMEM:
1710 368 : if (state->memtupcount - state->current >= ntuples)
1711 : {
1712 368 : state->current += ntuples;
1713 368 : return true;
1714 : }
1715 0 : state->current = state->memtupcount;
1716 0 : state->eof_reached = true;
1717 :
1718 : /*
1719 : * Complain if caller tries to retrieve more tuples than
1720 : * originally asked for in a bounded sort. This is because
1721 : * returning EOF here might be the wrong thing.
1722 : */
1723 0 : if (state->bounded && state->current >= state->bound)
1724 0 : elog(ERROR, "retrieved too many tuples in a bounded sort");
1725 :
1726 0 : return false;
1727 :
1728 24 : case TSS_SORTEDONTAPE:
1729 : case TSS_FINALMERGE:
1730 :
1731 : /*
1732 : * We could probably optimize these cases better, but for now it's
1733 : * not worth the trouble.
1734 : */
1735 24 : oldcontext = MemoryContextSwitchTo(state->base.sortcontext);
1736 240132 : while (ntuples-- > 0)
1737 : {
1738 : SortTuple stup;
1739 :
1740 240108 : if (!tuplesort_gettuple_common(state, forward, &stup))
1741 : {
1742 0 : MemoryContextSwitchTo(oldcontext);
1743 0 : return false;
1744 : }
1745 240108 : CHECK_FOR_INTERRUPTS();
1746 : }
1747 24 : MemoryContextSwitchTo(oldcontext);
1748 24 : return true;
1749 :
1750 0 : default:
1751 0 : elog(ERROR, "invalid tuplesort state");
1752 : return false; /* keep compiler quiet */
1753 : }
1754 : }
1755 :
1756 : /*
1757 : * tuplesort_merge_order - report merge order we'll use for given memory
1758 : * (note: "merge order" just means the number of input tapes in the merge).
1759 : *
1760 : * This is exported for use by the planner. allowedMem is in bytes.
1761 : */
1762 : int
1763 17674 : tuplesort_merge_order(int64 allowedMem)
1764 : {
1765 : int mOrder;
1766 :
1767 : /*----------
1768 : * In the merge phase, we need buffer space for each input and output tape.
1769 : * Each pass in the balanced merge algorithm reads from M input tapes, and
1770 : * writes to N output tapes. Each tape consumes TAPE_BUFFER_OVERHEAD bytes
1771 : * of memory. In addition to that, we want MERGE_BUFFER_SIZE workspace per
1772 : * input tape.
1773 : *
1774 : * totalMem = M * (TAPE_BUFFER_OVERHEAD + MERGE_BUFFER_SIZE) +
1775 : * N * TAPE_BUFFER_OVERHEAD
1776 : *
1777 : * Except for the last and next-to-last merge passes, where there can be
1778 : * fewer tapes left to process, M = N. We choose M so that we have the
1779 : * desired amount of memory available for the input buffers
1780 : * (TAPE_BUFFER_OVERHEAD + MERGE_BUFFER_SIZE), given the total memory
1781 : * available for the tape buffers (allowedMem).
1782 : *
1783 : * Note: you might be thinking we need to account for the memtuples[]
1784 : * array in this calculation, but we effectively treat that as part of the
1785 : * MERGE_BUFFER_SIZE workspace.
1786 : *----------
1787 : */
1788 17674 : mOrder = allowedMem /
1789 : (2 * TAPE_BUFFER_OVERHEAD + MERGE_BUFFER_SIZE);
1790 :
1791 : /*
1792 : * Even in minimum memory, use at least a MINORDER merge. On the other
1793 : * hand, even when we have lots of memory, do not use more than a MAXORDER
1794 : * merge. Tapes are pretty cheap, but they're not entirely free. Each
1795 : * additional tape reduces the amount of memory available to build runs,
1796 : * which in turn can cause the same sort to need more runs, which makes
1797 : * merging slower even if it can still be done in a single pass. Also,
1798 : * high order merges are quite slow due to CPU cache effects; it can be
1799 : * faster to pay the I/O cost of a multi-pass merge than to perform a
1800 : * single merge pass across many hundreds of tapes.
1801 : */
1802 17674 : mOrder = Max(mOrder, MINORDER);
1803 17674 : mOrder = Min(mOrder, MAXORDER);
1804 :
1805 17674 : return mOrder;
1806 : }
1807 :
1808 : /*
1809 : * Helper function to calculate how much memory to allocate for the read buffer
1810 : * of each input tape in a merge pass.
1811 : *
1812 : * 'avail_mem' is the amount of memory available for the buffers of all the
1813 : * tapes, both input and output.
1814 : * 'nInputTapes' and 'nInputRuns' are the number of input tapes and runs.
1815 : * 'maxOutputTapes' is the max. number of output tapes we should produce.
1816 : */
1817 : static int64
1818 364 : merge_read_buffer_size(int64 avail_mem, int nInputTapes, int nInputRuns,
1819 : int maxOutputTapes)
1820 : {
1821 : int nOutputRuns;
1822 : int nOutputTapes;
1823 :
1824 : /*
1825 : * How many output tapes will we produce in this pass?
1826 : *
1827 : * This is nInputRuns / nInputTapes, rounded up.
1828 : */
1829 364 : nOutputRuns = (nInputRuns + nInputTapes - 1) / nInputTapes;
1830 :
1831 364 : nOutputTapes = Min(nOutputRuns, maxOutputTapes);
1832 :
1833 : /*
1834 : * Each output tape consumes TAPE_BUFFER_OVERHEAD bytes of memory. All
1835 : * remaining memory is divided evenly between the input tapes.
1836 : *
1837 : * This also follows from the formula in tuplesort_merge_order, but here
1838 : * we derive the input buffer size from the amount of memory available,
1839 : * and M and N.
1840 : */
1841 364 : return Max((avail_mem - TAPE_BUFFER_OVERHEAD * nOutputTapes) / nInputTapes, 0);
1842 : }
1843 :
1844 : /*
1845 : * inittapes - initialize for tape sorting.
1846 : *
1847 : * This is called only if we have found we won't sort in memory.
1848 : */
1849 : static void
1850 704 : inittapes(Tuplesortstate *state, bool mergeruns)
1851 : {
1852 : Assert(!LEADER(state));
1853 :
1854 704 : if (mergeruns)
1855 : {
1856 : /* Compute number of input tapes to use when merging */
1857 140 : state->maxTapes = tuplesort_merge_order(state->allowedMem);
1858 : }
1859 : else
1860 : {
1861 : /* Workers can sometimes produce single run, output without merge */
1862 : Assert(WORKER(state));
1863 564 : state->maxTapes = MINORDER;
1864 : }
1865 :
1866 704 : if (trace_sort)
1867 0 : elog(LOG, "worker %d switching to external sort with %d tapes: %s",
1868 : state->worker, state->maxTapes, pg_rusage_show(&state->ru_start));
1869 :
1870 : /* Create the tape set */
1871 704 : inittapestate(state, state->maxTapes);
1872 704 : state->tapeset =
1873 704 : LogicalTapeSetCreate(false,
1874 704 : state->shared ? &state->shared->fileset : NULL,
1875 : state->worker);
1876 :
1877 704 : state->currentRun = 0;
1878 :
1879 : /*
1880 : * Initialize logical tape arrays.
1881 : */
1882 704 : state->inputTapes = NULL;
1883 704 : state->nInputTapes = 0;
1884 704 : state->nInputRuns = 0;
1885 :
1886 704 : state->outputTapes = palloc0(state->maxTapes * sizeof(LogicalTape *));
1887 704 : state->nOutputTapes = 0;
1888 704 : state->nOutputRuns = 0;
1889 :
1890 704 : state->status = TSS_BUILDRUNS;
1891 :
1892 704 : selectnewtape(state);
1893 704 : }
1894 :
1895 : /*
1896 : * inittapestate - initialize generic tape management state
1897 : */
1898 : static void
1899 898 : inittapestate(Tuplesortstate *state, int maxTapes)
1900 : {
1901 : int64 tapeSpace;
1902 :
1903 : /*
1904 : * Decrease availMem to reflect the space needed for tape buffers; but
1905 : * don't decrease it to the point that we have no room for tuples. (That
1906 : * case is only likely to occur if sorting pass-by-value Datums; in all
1907 : * other scenarios the memtuples[] array is unlikely to occupy more than
1908 : * half of allowedMem. In the pass-by-value case it's not important to
1909 : * account for tuple space, so we don't care if LACKMEM becomes
1910 : * inaccurate.)
1911 : */
1912 898 : tapeSpace = (int64) maxTapes * TAPE_BUFFER_OVERHEAD;
1913 :
1914 898 : if (tapeSpace + GetMemoryChunkSpace(state->memtuples) < state->allowedMem)
1915 776 : USEMEM(state, tapeSpace);
1916 :
1917 : /*
1918 : * Make sure that the temp file(s) underlying the tape set are created in
1919 : * suitable temp tablespaces. For parallel sorts, this should have been
1920 : * called already, but it doesn't matter if it is called a second time.
1921 : */
1922 898 : PrepareTempTablespaces();
1923 898 : }
1924 :
1925 : /*
1926 : * selectnewtape -- select next tape to output to.
1927 : *
1928 : * This is called after finishing a run when we know another run
1929 : * must be started. This is used both when building the initial
1930 : * runs, and during merge passes.
1931 : */
1932 : static void
1933 1808 : selectnewtape(Tuplesortstate *state)
1934 : {
1935 : /*
1936 : * At the beginning of each merge pass, nOutputTapes and nOutputRuns are
1937 : * both zero. On each call, we create a new output tape to hold the next
1938 : * run, until maxTapes is reached. After that, we assign new runs to the
1939 : * existing tapes in a round robin fashion.
1940 : */
1941 1808 : if (state->nOutputTapes < state->maxTapes)
1942 : {
1943 : /* Create a new tape to hold the next run */
1944 : Assert(state->outputTapes[state->nOutputRuns] == NULL);
1945 : Assert(state->nOutputRuns == state->nOutputTapes);
1946 1226 : state->destTape = LogicalTapeCreate(state->tapeset);
1947 1226 : state->outputTapes[state->nOutputTapes] = state->destTape;
1948 1226 : state->nOutputTapes++;
1949 1226 : state->nOutputRuns++;
1950 : }
1951 : else
1952 : {
1953 : /*
1954 : * We have reached the max number of tapes. Append to an existing
1955 : * tape.
1956 : */
1957 582 : state->destTape = state->outputTapes[state->nOutputRuns % state->nOutputTapes];
1958 582 : state->nOutputRuns++;
1959 : }
1960 1808 : }
1961 :
1962 : /*
1963 : * Initialize the slab allocation arena, for the given number of slots.
1964 : */
1965 : static void
1966 334 : init_slab_allocator(Tuplesortstate *state, int numSlots)
1967 : {
1968 334 : if (numSlots > 0)
1969 : {
1970 : char *p;
1971 : int i;
1972 :
1973 306 : state->slabMemoryBegin = palloc(numSlots * SLAB_SLOT_SIZE);
1974 306 : state->slabMemoryEnd = state->slabMemoryBegin +
1975 306 : numSlots * SLAB_SLOT_SIZE;
1976 306 : state->slabFreeHead = (SlabSlot *) state->slabMemoryBegin;
1977 306 : USEMEM(state, numSlots * SLAB_SLOT_SIZE);
1978 :
1979 306 : p = state->slabMemoryBegin;
1980 1170 : for (i = 0; i < numSlots - 1; i++)
1981 : {
1982 864 : ((SlabSlot *) p)->nextfree = (SlabSlot *) (p + SLAB_SLOT_SIZE);
1983 864 : p += SLAB_SLOT_SIZE;
1984 : }
1985 306 : ((SlabSlot *) p)->nextfree = NULL;
1986 : }
1987 : else
1988 : {
1989 28 : state->slabMemoryBegin = state->slabMemoryEnd = NULL;
1990 28 : state->slabFreeHead = NULL;
1991 : }
1992 334 : state->slabAllocatorUsed = true;
1993 334 : }
1994 :
1995 : /*
1996 : * mergeruns -- merge all the completed initial runs.
1997 : *
1998 : * This implements the Balanced k-Way Merge Algorithm. All input data has
1999 : * already been written to initial runs on tape (see dumptuples).
2000 : */
2001 : static void
2002 334 : mergeruns(Tuplesortstate *state)
2003 : {
2004 : int tapenum;
2005 :
2006 : Assert(state->status == TSS_BUILDRUNS);
2007 : Assert(state->memtupcount == 0);
2008 :
2009 334 : if (state->base.sortKeys != NULL && state->base.sortKeys->abbrev_converter != NULL)
2010 : {
2011 : /*
2012 : * If there are multiple runs to be merged, when we go to read back
2013 : * tuples from disk, abbreviated keys will not have been stored, and
2014 : * we don't care to regenerate them. Disable abbreviation from this
2015 : * point on.
2016 : */
2017 30 : state->base.sortKeys->abbrev_converter = NULL;
2018 30 : state->base.sortKeys->comparator = state->base.sortKeys->abbrev_full_comparator;
2019 :
2020 : /* Not strictly necessary, but be tidy */
2021 30 : state->base.sortKeys->abbrev_abort = NULL;
2022 30 : state->base.sortKeys->abbrev_full_comparator = NULL;
2023 : }
2024 :
2025 : /*
2026 : * Reset tuple memory. We've freed all the tuples that we previously
2027 : * allocated. We will use the slab allocator from now on.
2028 : */
2029 334 : MemoryContextResetOnly(state->base.tuplecontext);
2030 :
2031 : /*
2032 : * We no longer need a large memtuples array. (We will allocate a smaller
2033 : * one for the heap later.)
2034 : */
2035 334 : FREEMEM(state, GetMemoryChunkSpace(state->memtuples));
2036 334 : pfree(state->memtuples);
2037 334 : state->memtuples = NULL;
2038 :
2039 : /*
2040 : * Initialize the slab allocator. We need one slab slot per input tape,
2041 : * for the tuples in the heap, plus one to hold the tuple last returned
2042 : * from tuplesort_gettuple. (If we're sorting pass-by-val Datums,
2043 : * however, we don't need to do allocate anything.)
2044 : *
2045 : * In a multi-pass merge, we could shrink this allocation for the last
2046 : * merge pass, if it has fewer tapes than previous passes, but we don't
2047 : * bother.
2048 : *
2049 : * From this point on, we no longer use the USEMEM()/LACKMEM() mechanism
2050 : * to track memory usage of individual tuples.
2051 : */
2052 334 : if (state->base.tuples)
2053 306 : init_slab_allocator(state, state->nOutputTapes + 1);
2054 : else
2055 28 : init_slab_allocator(state, 0);
2056 :
2057 : /*
2058 : * Allocate a new 'memtuples' array, for the heap. It will hold one tuple
2059 : * from each input tape.
2060 : *
2061 : * We could shrink this, too, between passes in a multi-pass merge, but we
2062 : * don't bother. (The initial input tapes are still in outputTapes. The
2063 : * number of input tapes will not increase between passes.)
2064 : */
2065 334 : state->memtupsize = state->nOutputTapes;
2066 668 : state->memtuples = (SortTuple *) MemoryContextAlloc(state->base.maincontext,
2067 334 : state->nOutputTapes * sizeof(SortTuple));
2068 334 : USEMEM(state, GetMemoryChunkSpace(state->memtuples));
2069 :
2070 : /*
2071 : * Use all the remaining memory we have available for tape buffers among
2072 : * all the input tapes. At the beginning of each merge pass, we will
2073 : * divide this memory between the input and output tapes in the pass.
2074 : */
2075 334 : state->tape_buffer_mem = state->availMem;
2076 334 : USEMEM(state, state->tape_buffer_mem);
2077 334 : if (trace_sort)
2078 0 : elog(LOG, "worker %d using %zu KB of memory for tape buffers",
2079 : state->worker, state->tape_buffer_mem / 1024);
2080 :
2081 : for (;;)
2082 : {
2083 : /*
2084 : * On the first iteration, or if we have read all the runs from the
2085 : * input tapes in a multi-pass merge, it's time to start a new pass.
2086 : * Rewind all the output tapes, and make them inputs for the next
2087 : * pass.
2088 : */
2089 472 : if (state->nInputRuns == 0)
2090 : {
2091 : int64 input_buffer_size;
2092 :
2093 : /* Close the old, emptied, input tapes */
2094 364 : if (state->nInputTapes > 0)
2095 : {
2096 210 : for (tapenum = 0; tapenum < state->nInputTapes; tapenum++)
2097 180 : LogicalTapeClose(state->inputTapes[tapenum]);
2098 30 : pfree(state->inputTapes);
2099 : }
2100 :
2101 : /* Previous pass's outputs become next pass's inputs. */
2102 364 : state->inputTapes = state->outputTapes;
2103 364 : state->nInputTapes = state->nOutputTapes;
2104 364 : state->nInputRuns = state->nOutputRuns;
2105 :
2106 : /*
2107 : * Reset output tape variables. The actual LogicalTapes will be
2108 : * created as needed, here we only allocate the array to hold
2109 : * them.
2110 : */
2111 364 : state->outputTapes = palloc0(state->nInputTapes * sizeof(LogicalTape *));
2112 364 : state->nOutputTapes = 0;
2113 364 : state->nOutputRuns = 0;
2114 :
2115 : /*
2116 : * Redistribute the memory allocated for tape buffers, among the
2117 : * new input and output tapes.
2118 : */
2119 364 : input_buffer_size = merge_read_buffer_size(state->tape_buffer_mem,
2120 : state->nInputTapes,
2121 : state->nInputRuns,
2122 : state->maxTapes);
2123 :
2124 364 : if (trace_sort)
2125 0 : elog(LOG, "starting merge pass of %d input runs on %d tapes, " INT64_FORMAT " KB of memory for each input tape: %s",
2126 : state->nInputRuns, state->nInputTapes, input_buffer_size / 1024,
2127 : pg_rusage_show(&state->ru_start));
2128 :
2129 : /* Prepare the new input tapes for merge pass. */
2130 1428 : for (tapenum = 0; tapenum < state->nInputTapes; tapenum++)
2131 1064 : LogicalTapeRewindForRead(state->inputTapes[tapenum], input_buffer_size);
2132 :
2133 : /*
2134 : * If there's just one run left on each input tape, then only one
2135 : * merge pass remains. If we don't have to produce a materialized
2136 : * sorted tape, we can stop at this point and do the final merge
2137 : * on-the-fly.
2138 : */
2139 364 : if ((state->base.sortopt & TUPLESORT_RANDOMACCESS) == 0
2140 342 : && state->nInputRuns <= state->nInputTapes
2141 312 : && !WORKER(state))
2142 : {
2143 : /* Tell logtape.c we won't be writing anymore */
2144 312 : LogicalTapeSetForgetFreeSpace(state->tapeset);
2145 : /* Initialize for the final merge pass */
2146 312 : beginmerge(state);
2147 312 : state->status = TSS_FINALMERGE;
2148 312 : return;
2149 : }
2150 : }
2151 :
2152 : /* Select an output tape */
2153 160 : selectnewtape(state);
2154 :
2155 : /* Merge one run from each input tape. */
2156 160 : mergeonerun(state);
2157 :
2158 : /*
2159 : * If the input tapes are empty, and we output only one output run,
2160 : * we're done. The current output tape contains the final result.
2161 : */
2162 160 : if (state->nInputRuns == 0 && state->nOutputRuns <= 1)
2163 22 : break;
2164 : }
2165 :
2166 : /*
2167 : * Done. The result is on a single run on a single tape.
2168 : */
2169 22 : state->result_tape = state->outputTapes[0];
2170 22 : if (!WORKER(state))
2171 22 : LogicalTapeFreeze(state->result_tape, NULL);
2172 : else
2173 0 : worker_freeze_result_tape(state);
2174 22 : state->status = TSS_SORTEDONTAPE;
2175 :
2176 : /* Close all the now-empty input tapes, to release their read buffers. */
2177 114 : for (tapenum = 0; tapenum < state->nInputTapes; tapenum++)
2178 92 : LogicalTapeClose(state->inputTapes[tapenum]);
2179 : }
2180 :
2181 : /*
2182 : * Merge one run from each input tape.
2183 : */
2184 : static void
2185 160 : mergeonerun(Tuplesortstate *state)
2186 : {
2187 : int srcTapeIndex;
2188 : LogicalTape *srcTape;
2189 :
2190 : /*
2191 : * Start the merge by loading one tuple from each active source tape into
2192 : * the heap.
2193 : */
2194 160 : beginmerge(state);
2195 :
2196 : Assert(state->slabAllocatorUsed);
2197 :
2198 : /*
2199 : * Execute merge by repeatedly extracting lowest tuple in heap, writing it
2200 : * out, and replacing it with next tuple from same tape (if there is
2201 : * another one).
2202 : */
2203 875592 : while (state->memtupcount > 0)
2204 : {
2205 : SortTuple stup;
2206 :
2207 : /* write the tuple to destTape */
2208 875432 : srcTapeIndex = state->memtuples[0].srctape;
2209 875432 : srcTape = state->inputTapes[srcTapeIndex];
2210 875432 : WRITETUP(state, state->destTape, &state->memtuples[0]);
2211 :
2212 : /* recycle the slot of the tuple we just wrote out, for the next read */
2213 875432 : if (state->memtuples[0].tuple)
2214 735348 : RELEASE_SLAB_SLOT(state, state->memtuples[0].tuple);
2215 :
2216 : /*
2217 : * pull next tuple from the tape, and replace the written-out tuple in
2218 : * the heap with it.
2219 : */
2220 875432 : if (mergereadnext(state, srcTape, &stup))
2221 : {
2222 874578 : stup.srctape = srcTapeIndex;
2223 874578 : tuplesort_heap_replace_top(state, &stup);
2224 : }
2225 : else
2226 : {
2227 854 : tuplesort_heap_delete_top(state);
2228 854 : state->nInputRuns--;
2229 : }
2230 : }
2231 :
2232 : /*
2233 : * When the heap empties, we're done. Write an end-of-run marker on the
2234 : * output tape.
2235 : */
2236 160 : markrunend(state->destTape);
2237 160 : }
2238 :
2239 : /*
2240 : * beginmerge - initialize for a merge pass
2241 : *
2242 : * Fill the merge heap with the first tuple from each input tape.
2243 : */
2244 : static void
2245 472 : beginmerge(Tuplesortstate *state)
2246 : {
2247 : int activeTapes;
2248 : int srcTapeIndex;
2249 :
2250 : /* Heap should be empty here */
2251 : Assert(state->memtupcount == 0);
2252 :
2253 472 : activeTapes = Min(state->nInputTapes, state->nInputRuns);
2254 :
2255 2118 : for (srcTapeIndex = 0; srcTapeIndex < activeTapes; srcTapeIndex++)
2256 : {
2257 : SortTuple tup;
2258 :
2259 1646 : if (mergereadnext(state, state->inputTapes[srcTapeIndex], &tup))
2260 : {
2261 1332 : tup.srctape = srcTapeIndex;
2262 1332 : tuplesort_heap_insert(state, &tup);
2263 : }
2264 : }
2265 472 : }
2266 :
2267 : /*
2268 : * mergereadnext - read next tuple from one merge input tape
2269 : *
2270 : * Returns false on EOF.
2271 : */
2272 : static bool
2273 5558778 : mergereadnext(Tuplesortstate *state, LogicalTape *srcTape, SortTuple *stup)
2274 : {
2275 : unsigned int tuplen;
2276 :
2277 : /* read next tuple, if any */
2278 5558778 : if ((tuplen = getlen(srcTape, true)) == 0)
2279 1598 : return false;
2280 5557180 : READTUP(state, stup, srcTape, tuplen);
2281 :
2282 5557180 : return true;
2283 : }
2284 :
2285 : /*
2286 : * dumptuples - remove tuples from memtuples and write initial run to tape
2287 : *
2288 : * When alltuples = true, dump everything currently in memory. (This case is
2289 : * only used at end of input data.)
2290 : */
2291 : static void
2292 1113130 : dumptuples(Tuplesortstate *state, bool alltuples)
2293 : {
2294 : int memtupwrite;
2295 : int i;
2296 :
2297 : /*
2298 : * Nothing to do if we still fit in available memory and have array slots,
2299 : * unless this is the final call during initial run generation.
2300 : */
2301 1113130 : if (state->memtupcount < state->memtupsize && !LACKMEM(state) &&
2302 1112186 : !alltuples)
2303 1111482 : return;
2304 :
2305 : /*
2306 : * Final call might require no sorting, in rare cases where we just so
2307 : * happen to have previously LACKMEM()'d at the point where exactly all
2308 : * remaining tuples are loaded into memory, just before input was
2309 : * exhausted. In general, short final runs are quite possible, but avoid
2310 : * creating a completely empty run. In a worker, though, we must produce
2311 : * at least one tape, even if it's empty.
2312 : */
2313 1648 : if (state->memtupcount == 0 && state->currentRun > 0)
2314 0 : return;
2315 :
2316 : Assert(state->status == TSS_BUILDRUNS);
2317 :
2318 : /*
2319 : * It seems unlikely that this limit will ever be exceeded, but take no
2320 : * chances
2321 : */
2322 1648 : if (state->currentRun == INT_MAX)
2323 0 : ereport(ERROR,
2324 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
2325 : errmsg("cannot have more than %d runs for an external sort",
2326 : INT_MAX)));
2327 :
2328 1648 : if (state->currentRun > 0)
2329 944 : selectnewtape(state);
2330 :
2331 1648 : state->currentRun++;
2332 :
2333 1648 : if (trace_sort)
2334 0 : elog(LOG, "worker %d starting quicksort of run %d: %s",
2335 : state->worker, state->currentRun,
2336 : pg_rusage_show(&state->ru_start));
2337 :
2338 : /*
2339 : * Sort all tuples accumulated within the allowed amount of memory for
2340 : * this run using quicksort
2341 : */
2342 1648 : tuplesort_sort_memtuples(state);
2343 :
2344 1648 : if (trace_sort)
2345 0 : elog(LOG, "worker %d finished quicksort of run %d: %s",
2346 : state->worker, state->currentRun,
2347 : pg_rusage_show(&state->ru_start));
2348 :
2349 1648 : memtupwrite = state->memtupcount;
2350 5198600 : for (i = 0; i < memtupwrite; i++)
2351 : {
2352 5196952 : SortTuple *stup = &state->memtuples[i];
2353 :
2354 5196952 : WRITETUP(state, state->destTape, stup);
2355 : }
2356 :
2357 1648 : state->memtupcount = 0;
2358 :
2359 : /*
2360 : * Reset tuple memory. We've freed all of the tuples that we previously
2361 : * allocated. It's important to avoid fragmentation when there is a stark
2362 : * change in the sizes of incoming tuples. In bounded sorts,
2363 : * fragmentation due to AllocSetFree's bucketing by size class might be
2364 : * particularly bad if this step wasn't taken.
2365 : */
2366 1648 : MemoryContextReset(state->base.tuplecontext);
2367 :
2368 : /*
2369 : * Now update the memory accounting to subtract the memory used by the
2370 : * tuple.
2371 : */
2372 1648 : FREEMEM(state, state->tupleMem);
2373 1648 : state->tupleMem = 0;
2374 :
2375 1648 : markrunend(state->destTape);
2376 :
2377 1648 : if (trace_sort)
2378 0 : elog(LOG, "worker %d finished writing run %d to tape %d: %s",
2379 : state->worker, state->currentRun, (state->currentRun - 1) % state->nOutputTapes + 1,
2380 : pg_rusage_show(&state->ru_start));
2381 : }
2382 :
2383 : /*
2384 : * tuplesort_rescan - rewind and replay the scan
2385 : */
2386 : void
2387 60 : tuplesort_rescan(Tuplesortstate *state)
2388 : {
2389 60 : MemoryContext oldcontext = MemoryContextSwitchTo(state->base.sortcontext);
2390 :
2391 : Assert(state->base.sortopt & TUPLESORT_RANDOMACCESS);
2392 :
2393 60 : switch (state->status)
2394 : {
2395 52 : case TSS_SORTEDINMEM:
2396 52 : state->current = 0;
2397 52 : state->eof_reached = false;
2398 52 : state->markpos_offset = 0;
2399 52 : state->markpos_eof = false;
2400 52 : break;
2401 8 : case TSS_SORTEDONTAPE:
2402 8 : LogicalTapeRewindForRead(state->result_tape, 0);
2403 8 : state->eof_reached = false;
2404 8 : state->markpos_block = 0L;
2405 8 : state->markpos_offset = 0;
2406 8 : state->markpos_eof = false;
2407 8 : break;
2408 0 : default:
2409 0 : elog(ERROR, "invalid tuplesort state");
2410 : break;
2411 : }
2412 :
2413 60 : MemoryContextSwitchTo(oldcontext);
2414 60 : }
2415 :
2416 : /*
2417 : * tuplesort_markpos - saves current position in the merged sort file
2418 : */
2419 : void
2420 584210 : tuplesort_markpos(Tuplesortstate *state)
2421 : {
2422 584210 : MemoryContext oldcontext = MemoryContextSwitchTo(state->base.sortcontext);
2423 :
2424 : Assert(state->base.sortopt & TUPLESORT_RANDOMACCESS);
2425 :
2426 584210 : switch (state->status)
2427 : {
2428 575402 : case TSS_SORTEDINMEM:
2429 575402 : state->markpos_offset = state->current;
2430 575402 : state->markpos_eof = state->eof_reached;
2431 575402 : break;
2432 8808 : case TSS_SORTEDONTAPE:
2433 8808 : LogicalTapeTell(state->result_tape,
2434 : &state->markpos_block,
2435 : &state->markpos_offset);
2436 8808 : state->markpos_eof = state->eof_reached;
2437 8808 : break;
2438 0 : default:
2439 0 : elog(ERROR, "invalid tuplesort state");
2440 : break;
2441 : }
2442 :
2443 584210 : MemoryContextSwitchTo(oldcontext);
2444 584210 : }
2445 :
2446 : /*
2447 : * tuplesort_restorepos - restores current position in merged sort file to
2448 : * last saved position
2449 : */
2450 : void
2451 38722 : tuplesort_restorepos(Tuplesortstate *state)
2452 : {
2453 38722 : MemoryContext oldcontext = MemoryContextSwitchTo(state->base.sortcontext);
2454 :
2455 : Assert(state->base.sortopt & TUPLESORT_RANDOMACCESS);
2456 :
2457 38722 : switch (state->status)
2458 : {
2459 32530 : case TSS_SORTEDINMEM:
2460 32530 : state->current = state->markpos_offset;
2461 32530 : state->eof_reached = state->markpos_eof;
2462 32530 : break;
2463 6192 : case TSS_SORTEDONTAPE:
2464 6192 : LogicalTapeSeek(state->result_tape,
2465 : state->markpos_block,
2466 : state->markpos_offset);
2467 6192 : state->eof_reached = state->markpos_eof;
2468 6192 : break;
2469 0 : default:
2470 0 : elog(ERROR, "invalid tuplesort state");
2471 : break;
2472 : }
2473 :
2474 38722 : MemoryContextSwitchTo(oldcontext);
2475 38722 : }
2476 :
2477 : /*
2478 : * tuplesort_get_stats - extract summary statistics
2479 : *
2480 : * This can be called after tuplesort_performsort() finishes to obtain
2481 : * printable summary information about how the sort was performed.
2482 : */
2483 : void
2484 396 : tuplesort_get_stats(Tuplesortstate *state,
2485 : TuplesortInstrumentation *stats)
2486 : {
2487 : /*
2488 : * Note: it might seem we should provide both memory and disk usage for a
2489 : * disk-based sort. However, the current code doesn't track memory space
2490 : * accurately once we have begun to return tuples to the caller (since we
2491 : * don't account for pfree's the caller is expected to do), so we cannot
2492 : * rely on availMem in a disk sort. This does not seem worth the overhead
2493 : * to fix. Is it worth creating an API for the memory context code to
2494 : * tell us how much is actually used in sortcontext?
2495 : */
2496 396 : tuplesort_updatemax(state);
2497 :
2498 396 : if (state->isMaxSpaceDisk)
2499 6 : stats->spaceType = SORT_SPACE_TYPE_DISK;
2500 : else
2501 390 : stats->spaceType = SORT_SPACE_TYPE_MEMORY;
2502 396 : stats->spaceUsed = (state->maxSpace + 1023) / 1024;
2503 :
2504 396 : switch (state->maxSpaceStatus)
2505 : {
2506 390 : case TSS_SORTEDINMEM:
2507 390 : if (state->boundUsed)
2508 42 : stats->sortMethod = SORT_TYPE_TOP_N_HEAPSORT;
2509 : else
2510 348 : stats->sortMethod = SORT_TYPE_QUICKSORT;
2511 390 : break;
2512 0 : case TSS_SORTEDONTAPE:
2513 0 : stats->sortMethod = SORT_TYPE_EXTERNAL_SORT;
2514 0 : break;
2515 6 : case TSS_FINALMERGE:
2516 6 : stats->sortMethod = SORT_TYPE_EXTERNAL_MERGE;
2517 6 : break;
2518 0 : default:
2519 0 : stats->sortMethod = SORT_TYPE_STILL_IN_PROGRESS;
2520 0 : break;
2521 : }
2522 396 : }
2523 :
2524 : /*
2525 : * Convert TuplesortMethod to a string.
2526 : */
2527 : const char *
2528 294 : tuplesort_method_name(TuplesortMethod m)
2529 : {
2530 294 : switch (m)
2531 : {
2532 0 : case SORT_TYPE_STILL_IN_PROGRESS:
2533 0 : return "still in progress";
2534 42 : case SORT_TYPE_TOP_N_HEAPSORT:
2535 42 : return "top-N heapsort";
2536 246 : case SORT_TYPE_QUICKSORT:
2537 246 : return "quicksort";
2538 0 : case SORT_TYPE_EXTERNAL_SORT:
2539 0 : return "external sort";
2540 6 : case SORT_TYPE_EXTERNAL_MERGE:
2541 6 : return "external merge";
2542 : }
2543 :
2544 0 : return "unknown";
2545 : }
2546 :
2547 : /*
2548 : * Convert TuplesortSpaceType to a string.
2549 : */
2550 : const char *
2551 258 : tuplesort_space_type_name(TuplesortSpaceType t)
2552 : {
2553 : Assert(t == SORT_SPACE_TYPE_DISK || t == SORT_SPACE_TYPE_MEMORY);
2554 258 : return t == SORT_SPACE_TYPE_DISK ? "Disk" : "Memory";
2555 : }
2556 :
2557 :
2558 : /*
2559 : * Heap manipulation routines, per Knuth's Algorithm 5.2.3H.
2560 : */
2561 :
2562 : /*
2563 : * Convert the existing unordered array of SortTuples to a bounded heap,
2564 : * discarding all but the smallest "state->bound" tuples.
2565 : *
2566 : * When working with a bounded heap, we want to keep the largest entry
2567 : * at the root (array entry zero), instead of the smallest as in the normal
2568 : * sort case. This allows us to discard the largest entry cheaply.
2569 : * Therefore, we temporarily reverse the sort direction.
2570 : */
2571 : static void
2572 408 : make_bounded_heap(Tuplesortstate *state)
2573 : {
2574 408 : int tupcount = state->memtupcount;
2575 : int i;
2576 :
2577 : Assert(state->status == TSS_INITIAL);
2578 : Assert(state->bounded);
2579 : Assert(tupcount >= state->bound);
2580 : Assert(SERIAL(state));
2581 :
2582 : /* Reverse sort direction so largest entry will be at root */
2583 408 : reversedirection(state);
2584 :
2585 408 : state->memtupcount = 0; /* make the heap empty */
2586 37980 : for (i = 0; i < tupcount; i++)
2587 : {
2588 37572 : if (state->memtupcount < state->bound)
2589 : {
2590 : /* Insert next tuple into heap */
2591 : /* Must copy source tuple to avoid possible overwrite */
2592 18582 : SortTuple stup = state->memtuples[i];
2593 :
2594 18582 : tuplesort_heap_insert(state, &stup);
2595 : }
2596 : else
2597 : {
2598 : /*
2599 : * The heap is full. Replace the largest entry with the new
2600 : * tuple, or just discard it, if it's larger than anything already
2601 : * in the heap.
2602 : */
2603 18990 : if (COMPARETUP(state, &state->memtuples[i], &state->memtuples[0]) <= 0)
2604 : {
2605 9800 : free_sort_tuple(state, &state->memtuples[i]);
2606 9800 : CHECK_FOR_INTERRUPTS();
2607 : }
2608 : else
2609 9190 : tuplesort_heap_replace_top(state, &state->memtuples[i]);
2610 : }
2611 : }
2612 :
2613 : Assert(state->memtupcount == state->bound);
2614 408 : state->status = TSS_BOUNDED;
2615 408 : }
2616 :
2617 : /*
2618 : * Convert the bounded heap to a properly-sorted array
2619 : */
2620 : static void
2621 408 : sort_bounded_heap(Tuplesortstate *state)
2622 : {
2623 408 : int tupcount = state->memtupcount;
2624 :
2625 : Assert(state->status == TSS_BOUNDED);
2626 : Assert(state->bounded);
2627 : Assert(tupcount == state->bound);
2628 : Assert(SERIAL(state));
2629 :
2630 : /*
2631 : * We can unheapify in place because each delete-top call will remove the
2632 : * largest entry, which we can promptly store in the newly freed slot at
2633 : * the end. Once we're down to a single-entry heap, we're done.
2634 : */
2635 18582 : while (state->memtupcount > 1)
2636 : {
2637 18174 : SortTuple stup = state->memtuples[0];
2638 :
2639 : /* this sifts-up the next-largest entry and decreases memtupcount */
2640 18174 : tuplesort_heap_delete_top(state);
2641 18174 : state->memtuples[state->memtupcount] = stup;
2642 : }
2643 408 : state->memtupcount = tupcount;
2644 :
2645 : /*
2646 : * Reverse sort direction back to the original state. This is not
2647 : * actually necessary but seems like a good idea for tidiness.
2648 : */
2649 408 : reversedirection(state);
2650 :
2651 408 : state->status = TSS_SORTEDINMEM;
2652 408 : state->boundUsed = true;
2653 408 : }
2654 :
2655 : /*
2656 : * Sort all memtuples using specialized qsort() routines.
2657 : *
2658 : * Quicksort is used for small in-memory sorts, and external sort runs.
2659 : */
2660 : static void
2661 239646 : tuplesort_sort_memtuples(Tuplesortstate *state)
2662 : {
2663 : Assert(!LEADER(state));
2664 :
2665 239646 : if (state->memtupcount > 1)
2666 : {
2667 : /*
2668 : * Do we have the leading column's value or abbreviation in datum1,
2669 : * and is there a specialization for its comparator?
2670 : */
2671 68782 : if (state->base.haveDatum1 && state->base.sortKeys)
2672 : {
2673 68670 : if (state->base.sortKeys[0].comparator == ssup_datum_unsigned_cmp)
2674 : {
2675 3538 : qsort_tuple_unsigned(state->memtuples,
2676 3538 : state->memtupcount,
2677 : state);
2678 3522 : return;
2679 : }
2680 65132 : else if (state->base.sortKeys[0].comparator == ssup_datum_signed_cmp)
2681 : {
2682 1386 : qsort_tuple_signed(state->memtuples,
2683 1386 : state->memtupcount,
2684 : state);
2685 1386 : return;
2686 : }
2687 63746 : else if (state->base.sortKeys[0].comparator == ssup_datum_int32_cmp)
2688 : {
2689 39928 : qsort_tuple_int32(state->memtuples,
2690 39928 : state->memtupcount,
2691 : state);
2692 39868 : return;
2693 : }
2694 : }
2695 :
2696 : /* Can we use the single-key sort function? */
2697 23930 : if (state->base.onlyKey != NULL)
2698 : {
2699 10246 : qsort_ssup(state->memtuples, state->memtupcount,
2700 10246 : state->base.onlyKey);
2701 : }
2702 : else
2703 : {
2704 13684 : qsort_tuple(state->memtuples,
2705 13684 : state->memtupcount,
2706 : state->base.comparetup,
2707 : state);
2708 : }
2709 : }
2710 : }
2711 :
2712 : /*
2713 : * Insert a new tuple into an empty or existing heap, maintaining the
2714 : * heap invariant. Caller is responsible for ensuring there's room.
2715 : *
2716 : * Note: For some callers, tuple points to a memtuples[] entry above the
2717 : * end of the heap. This is safe as long as it's not immediately adjacent
2718 : * to the end of the heap (ie, in the [memtupcount] array entry) --- if it
2719 : * is, it might get overwritten before being moved into the heap!
2720 : */
2721 : static void
2722 19914 : tuplesort_heap_insert(Tuplesortstate *state, SortTuple *tuple)
2723 : {
2724 : SortTuple *memtuples;
2725 : int j;
2726 :
2727 19914 : memtuples = state->memtuples;
2728 : Assert(state->memtupcount < state->memtupsize);
2729 :
2730 19914 : CHECK_FOR_INTERRUPTS();
2731 :
2732 : /*
2733 : * Sift-up the new entry, per Knuth 5.2.3 exercise 16. Note that Knuth is
2734 : * using 1-based array indexes, not 0-based.
2735 : */
2736 19914 : j = state->memtupcount++;
2737 57792 : while (j > 0)
2738 : {
2739 50952 : int i = (j - 1) >> 1;
2740 :
2741 50952 : if (COMPARETUP(state, tuple, &memtuples[i]) >= 0)
2742 13074 : break;
2743 37878 : memtuples[j] = memtuples[i];
2744 37878 : j = i;
2745 : }
2746 19914 : memtuples[j] = *tuple;
2747 19914 : }
2748 :
2749 : /*
2750 : * Remove the tuple at state->memtuples[0] from the heap. Decrement
2751 : * memtupcount, and sift up to maintain the heap invariant.
2752 : *
2753 : * The caller has already free'd the tuple the top node points to,
2754 : * if necessary.
2755 : */
2756 : static void
2757 19458 : tuplesort_heap_delete_top(Tuplesortstate *state)
2758 : {
2759 19458 : SortTuple *memtuples = state->memtuples;
2760 : SortTuple *tuple;
2761 :
2762 19458 : if (--state->memtupcount <= 0)
2763 332 : return;
2764 :
2765 : /*
2766 : * Remove the last tuple in the heap, and re-insert it, by replacing the
2767 : * current top node with it.
2768 : */
2769 19126 : tuple = &memtuples[state->memtupcount];
2770 19126 : tuplesort_heap_replace_top(state, tuple);
2771 : }
2772 :
2773 : /*
2774 : * Replace the tuple at state->memtuples[0] with a new tuple. Sift up to
2775 : * maintain the heap invariant.
2776 : *
2777 : * This corresponds to Knuth's "sift-up" algorithm (Algorithm 5.2.3H,
2778 : * Heapsort, steps H3-H8).
2779 : */
2780 : static void
2781 6087150 : tuplesort_heap_replace_top(Tuplesortstate *state, SortTuple *tuple)
2782 : {
2783 6087150 : SortTuple *memtuples = state->memtuples;
2784 : unsigned int i,
2785 : n;
2786 :
2787 : Assert(state->memtupcount >= 1);
2788 :
2789 6087150 : CHECK_FOR_INTERRUPTS();
2790 :
2791 : /*
2792 : * state->memtupcount is "int", but we use "unsigned int" for i, j, n.
2793 : * This prevents overflow in the "2 * i + 1" calculation, since at the top
2794 : * of the loop we must have i < n <= INT_MAX <= UINT_MAX/2.
2795 : */
2796 6087150 : n = state->memtupcount;
2797 6087150 : i = 0; /* i is where the "hole" is */
2798 : for (;;)
2799 1778286 : {
2800 7865436 : unsigned int j = 2 * i + 1;
2801 :
2802 7865436 : if (j >= n)
2803 1238184 : break;
2804 9076228 : if (j + 1 < n &&
2805 2448976 : COMPARETUP(state, &memtuples[j], &memtuples[j + 1]) > 0)
2806 967166 : j++;
2807 6627252 : if (COMPARETUP(state, tuple, &memtuples[j]) <= 0)
2808 4848966 : break;
2809 1778286 : memtuples[i] = memtuples[j];
2810 1778286 : i = j;
2811 : }
2812 6087150 : memtuples[i] = *tuple;
2813 6087150 : }
2814 :
2815 : /*
2816 : * Function to reverse the sort direction from its current state
2817 : *
2818 : * It is not safe to call this when performing hash tuplesorts
2819 : */
2820 : static void
2821 816 : reversedirection(Tuplesortstate *state)
2822 : {
2823 816 : SortSupport sortKey = state->base.sortKeys;
2824 : int nkey;
2825 :
2826 1992 : for (nkey = 0; nkey < state->base.nKeys; nkey++, sortKey++)
2827 : {
2828 1176 : sortKey->ssup_reverse = !sortKey->ssup_reverse;
2829 1176 : sortKey->ssup_nulls_first = !sortKey->ssup_nulls_first;
2830 : }
2831 816 : }
2832 :
2833 :
2834 : /*
2835 : * Tape interface routines
2836 : */
2837 :
2838 : static unsigned int
2839 5861790 : getlen(LogicalTape *tape, bool eofOK)
2840 : {
2841 : unsigned int len;
2842 :
2843 5861790 : if (LogicalTapeRead(tape,
2844 : &len, sizeof(len)) != sizeof(len))
2845 0 : elog(ERROR, "unexpected end of tape");
2846 5861790 : if (len == 0 && !eofOK)
2847 0 : elog(ERROR, "unexpected end of data");
2848 5861790 : return len;
2849 : }
2850 :
2851 : static void
2852 1808 : markrunend(LogicalTape *tape)
2853 : {
2854 1808 : unsigned int len = 0;
2855 :
2856 1808 : LogicalTapeWrite(tape, &len, sizeof(len));
2857 1808 : }
2858 :
2859 : /*
2860 : * Get memory for tuple from within READTUP() routine.
2861 : *
2862 : * We use next free slot from the slab allocator, or palloc() if the tuple
2863 : * is too large for that.
2864 : */
2865 : void *
2866 5449916 : tuplesort_readtup_alloc(Tuplesortstate *state, Size tuplen)
2867 : {
2868 : SlabSlot *buf;
2869 :
2870 : /*
2871 : * We pre-allocate enough slots in the slab arena that we should never run
2872 : * out.
2873 : */
2874 : Assert(state->slabFreeHead);
2875 :
2876 5449916 : if (tuplen > SLAB_SLOT_SIZE || !state->slabFreeHead)
2877 6 : return MemoryContextAlloc(state->base.sortcontext, tuplen);
2878 : else
2879 : {
2880 5449910 : buf = state->slabFreeHead;
2881 : /* Reuse this slot */
2882 5449910 : state->slabFreeHead = buf->nextfree;
2883 :
2884 5449910 : return buf;
2885 : }
2886 : }
2887 :
2888 :
2889 : /*
2890 : * Parallel sort routines
2891 : */
2892 :
2893 : /*
2894 : * tuplesort_estimate_shared - estimate required shared memory allocation
2895 : *
2896 : * nWorkers is an estimate of the number of workers (it's the number that
2897 : * will be requested).
2898 : */
2899 : Size
2900 196 : tuplesort_estimate_shared(int nWorkers)
2901 : {
2902 : Size tapesSize;
2903 :
2904 : Assert(nWorkers > 0);
2905 :
2906 : /* Make sure that BufFile shared state is MAXALIGN'd */
2907 196 : tapesSize = mul_size(sizeof(TapeShare), nWorkers);
2908 196 : tapesSize = MAXALIGN(add_size(tapesSize, offsetof(Sharedsort, tapes)));
2909 :
2910 196 : return tapesSize;
2911 : }
2912 :
2913 : /*
2914 : * tuplesort_initialize_shared - initialize shared tuplesort state
2915 : *
2916 : * Must be called from leader process before workers are launched, to
2917 : * establish state needed up-front for worker tuplesortstates. nWorkers
2918 : * should match the argument passed to tuplesort_estimate_shared().
2919 : */
2920 : void
2921 266 : tuplesort_initialize_shared(Sharedsort *shared, int nWorkers, dsm_segment *seg)
2922 : {
2923 : int i;
2924 :
2925 : Assert(nWorkers > 0);
2926 :
2927 266 : SpinLockInit(&shared->mutex);
2928 266 : shared->currentWorker = 0;
2929 266 : shared->workersFinished = 0;
2930 266 : SharedFileSetInit(&shared->fileset, seg);
2931 266 : shared->nTapes = nWorkers;
2932 838 : for (i = 0; i < nWorkers; i++)
2933 : {
2934 572 : shared->tapes[i].firstblocknumber = 0L;
2935 : }
2936 266 : }
2937 :
2938 : /*
2939 : * tuplesort_attach_shared - attach to shared tuplesort state
2940 : *
2941 : * Must be called by all worker processes.
2942 : */
2943 : void
2944 300 : tuplesort_attach_shared(Sharedsort *shared, dsm_segment *seg)
2945 : {
2946 : /* Attach to SharedFileSet */
2947 300 : SharedFileSetAttach(&shared->fileset, seg);
2948 300 : }
2949 :
2950 : /*
2951 : * worker_get_identifier - Assign and return ordinal identifier for worker
2952 : *
2953 : * The order in which these are assigned is not well defined, and should not
2954 : * matter; worker numbers across parallel sort participants need only be
2955 : * distinct and gapless. logtape.c requires this.
2956 : *
2957 : * Note that the identifiers assigned from here have no relation to
2958 : * ParallelWorkerNumber number, to avoid making any assumption about
2959 : * caller's requirements. However, we do follow the ParallelWorkerNumber
2960 : * convention of representing a non-worker with worker number -1. This
2961 : * includes the leader, as well as serial Tuplesort processes.
2962 : */
2963 : static int
2964 564 : worker_get_identifier(Tuplesortstate *state)
2965 : {
2966 564 : Sharedsort *shared = state->shared;
2967 : int worker;
2968 :
2969 : Assert(WORKER(state));
2970 :
2971 564 : SpinLockAcquire(&shared->mutex);
2972 564 : worker = shared->currentWorker++;
2973 564 : SpinLockRelease(&shared->mutex);
2974 :
2975 564 : return worker;
2976 : }
2977 :
2978 : /*
2979 : * worker_freeze_result_tape - freeze worker's result tape for leader
2980 : *
2981 : * This is called by workers just after the result tape has been determined,
2982 : * instead of calling LogicalTapeFreeze() directly. They do so because
2983 : * workers require a few additional steps over similar serial
2984 : * TSS_SORTEDONTAPE external sort cases, which also happen here. The extra
2985 : * steps are around freeing now unneeded resources, and representing to
2986 : * leader that worker's input run is available for its merge.
2987 : *
2988 : * There should only be one final output run for each worker, which consists
2989 : * of all tuples that were originally input into worker.
2990 : */
2991 : static void
2992 564 : worker_freeze_result_tape(Tuplesortstate *state)
2993 : {
2994 564 : Sharedsort *shared = state->shared;
2995 : TapeShare output;
2996 :
2997 : Assert(WORKER(state));
2998 : Assert(state->result_tape != NULL);
2999 : Assert(state->memtupcount == 0);
3000 :
3001 : /*
3002 : * Free most remaining memory, in case caller is sensitive to our holding
3003 : * on to it. memtuples may not be a tiny merge heap at this point.
3004 : */
3005 564 : pfree(state->memtuples);
3006 : /* Be tidy */
3007 564 : state->memtuples = NULL;
3008 564 : state->memtupsize = 0;
3009 :
3010 : /*
3011 : * Parallel worker requires result tape metadata, which is to be stored in
3012 : * shared memory for leader
3013 : */
3014 564 : LogicalTapeFreeze(state->result_tape, &output);
3015 :
3016 : /* Store properties of output tape, and update finished worker count */
3017 564 : SpinLockAcquire(&shared->mutex);
3018 564 : shared->tapes[state->worker] = output;
3019 564 : shared->workersFinished++;
3020 564 : SpinLockRelease(&shared->mutex);
3021 564 : }
3022 :
3023 : /*
3024 : * worker_nomergeruns - dump memtuples in worker, without merging
3025 : *
3026 : * This called as an alternative to mergeruns() with a worker when no
3027 : * merging is required.
3028 : */
3029 : static void
3030 564 : worker_nomergeruns(Tuplesortstate *state)
3031 : {
3032 : Assert(WORKER(state));
3033 : Assert(state->result_tape == NULL);
3034 : Assert(state->nOutputRuns == 1);
3035 :
3036 564 : state->result_tape = state->destTape;
3037 564 : worker_freeze_result_tape(state);
3038 564 : }
3039 :
3040 : /*
3041 : * leader_takeover_tapes - create tapeset for leader from worker tapes
3042 : *
3043 : * So far, leader Tuplesortstate has performed no actual sorting. By now, all
3044 : * sorting has occurred in workers, all of which must have already returned
3045 : * from tuplesort_performsort().
3046 : *
3047 : * When this returns, leader process is left in a state that is virtually
3048 : * indistinguishable from it having generated runs as a serial external sort
3049 : * might have.
3050 : */
3051 : static void
3052 194 : leader_takeover_tapes(Tuplesortstate *state)
3053 : {
3054 194 : Sharedsort *shared = state->shared;
3055 194 : int nParticipants = state->nParticipants;
3056 : int workersFinished;
3057 : int j;
3058 :
3059 : Assert(LEADER(state));
3060 : Assert(nParticipants >= 1);
3061 :
3062 194 : SpinLockAcquire(&shared->mutex);
3063 194 : workersFinished = shared->workersFinished;
3064 194 : SpinLockRelease(&shared->mutex);
3065 :
3066 194 : if (nParticipants != workersFinished)
3067 0 : elog(ERROR, "cannot take over tapes before all workers finish");
3068 :
3069 : /*
3070 : * Create the tapeset from worker tapes, including a leader-owned tape at
3071 : * the end. Parallel workers are far more expensive than logical tapes,
3072 : * so the number of tapes allocated here should never be excessive.
3073 : */
3074 194 : inittapestate(state, nParticipants);
3075 194 : state->tapeset = LogicalTapeSetCreate(false, &shared->fileset, -1);
3076 :
3077 : /*
3078 : * Set currentRun to reflect the number of runs we will merge (it's not
3079 : * used for anything, this is just pro forma)
3080 : */
3081 194 : state->currentRun = nParticipants;
3082 :
3083 : /*
3084 : * Initialize the state to look the same as after building the initial
3085 : * runs.
3086 : *
3087 : * There will always be exactly 1 run per worker, and exactly one input
3088 : * tape per run, because workers always output exactly 1 run, even when
3089 : * there were no input tuples for workers to sort.
3090 : */
3091 194 : state->inputTapes = NULL;
3092 194 : state->nInputTapes = 0;
3093 194 : state->nInputRuns = 0;
3094 :
3095 194 : state->outputTapes = palloc0(nParticipants * sizeof(LogicalTape *));
3096 194 : state->nOutputTapes = nParticipants;
3097 194 : state->nOutputRuns = nParticipants;
3098 :
3099 618 : for (j = 0; j < nParticipants; j++)
3100 : {
3101 424 : state->outputTapes[j] = LogicalTapeImport(state->tapeset, j, &shared->tapes[j]);
3102 : }
3103 :
3104 194 : state->status = TSS_BUILDRUNS;
3105 194 : }
3106 :
3107 : /*
3108 : * Convenience routine to free a tuple previously loaded into sort memory
3109 : */
3110 : static void
3111 3759980 : free_sort_tuple(Tuplesortstate *state, SortTuple *stup)
3112 : {
3113 3759980 : if (stup->tuple)
3114 : {
3115 3599702 : FREEMEM(state, GetMemoryChunkSpace(stup->tuple));
3116 3599702 : pfree(stup->tuple);
3117 3599702 : stup->tuple = NULL;
3118 : }
3119 3759980 : }
3120 :
3121 : int
3122 0 : ssup_datum_unsigned_cmp(Datum x, Datum y, SortSupport ssup)
3123 : {
3124 0 : if (x < y)
3125 0 : return -1;
3126 0 : else if (x > y)
3127 0 : return 1;
3128 : else
3129 0 : return 0;
3130 : }
3131 :
3132 : int
3133 1143736 : ssup_datum_signed_cmp(Datum x, Datum y, SortSupport ssup)
3134 : {
3135 1143736 : int64 xx = DatumGetInt64(x);
3136 1143736 : int64 yy = DatumGetInt64(y);
3137 :
3138 1143736 : if (xx < yy)
3139 425186 : return -1;
3140 718550 : else if (xx > yy)
3141 358896 : return 1;
3142 : else
3143 359654 : return 0;
3144 : }
3145 :
3146 : int
3147 191136420 : ssup_datum_int32_cmp(Datum x, Datum y, SortSupport ssup)
3148 : {
3149 191136420 : int32 xx = DatumGetInt32(x);
3150 191136420 : int32 yy = DatumGetInt32(y);
3151 :
3152 191136420 : if (xx < yy)
3153 46327704 : return -1;
3154 144808716 : else if (xx > yy)
3155 43622344 : return 1;
3156 : else
3157 101186372 : return 0;
3158 : }
|
