Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * read_stream.c
4 : * Mechanism for accessing buffered relation data with look-ahead
5 : *
6 : * Code that needs to access relation data typically pins blocks one at a
7 : * time, often in a predictable order that might be sequential or data-driven.
8 : * Calling the simple ReadBuffer() function for each block is inefficient,
9 : * because blocks that are not yet in the buffer pool require I/O operations
10 : * that are small and might stall waiting for storage. This mechanism looks
11 : * into the future and calls StartReadBuffers() and WaitReadBuffers() to read
12 : * neighboring blocks together and ahead of time, with an adaptive look-ahead
13 : * distance.
14 : *
15 : * A user-provided callback generates a stream of block numbers that is used
16 : * to form reads of up to io_combine_limit, by attempting to merge them with a
17 : * pending read. When that isn't possible, the existing pending read is sent
18 : * to StartReadBuffers() so that a new one can begin to form.
19 : *
20 : * The algorithm for controlling the look-ahead distance is based on recent
21 : * cache hit and miss history. When no I/O is necessary, there is no benefit
22 : * in looking ahead more than one block. This is the default initial
23 : * assumption, but when blocks needing I/O are streamed, the distance is
24 : * increased rapidly to try to benefit from I/O combining and concurrency. It
25 : * is reduced gradually when cached blocks are streamed.
26 : *
27 : * The main data structure is a circular queue of buffers of size
28 : * max_pinned_buffers plus some extra space for technical reasons, ready to be
29 : * returned by read_stream_next_buffer(). Each buffer also has an optional
30 : * variable sized object that is passed from the callback to the consumer of
31 : * buffers.
32 : *
33 : * Parallel to the queue of buffers, there is a circular queue of in-progress
34 : * I/Os that have been started with StartReadBuffers(), and for which
35 : * WaitReadBuffers() must be called before returning the buffer.
36 : *
37 : * For example, if the callback returns block numbers 10, 42, 43, 44, 60 in
38 : * successive calls, then these data structures might appear as follows:
39 : *
40 : * buffers buf/data ios
41 : *
42 : * +----+ +-----+ +--------+
43 : * | | | | +----+ 42..44 | <- oldest_io_index
44 : * +----+ +-----+ | +--------+
45 : * oldest_buffer_index -> | 10 | | ? | | +--+ 60..60 |
46 : * +----+ +-----+ | | +--------+
47 : * | 42 | | ? |<-+ | | | <- next_io_index
48 : * +----+ +-----+ | +--------+
49 : * | 43 | | ? | | | |
50 : * +----+ +-----+ | +--------+
51 : * | 44 | | ? | | | |
52 : * +----+ +-----+ | +--------+
53 : * | 60 | | ? |<---+
54 : * +----+ +-----+
55 : * next_buffer_index -> | | | |
56 : * +----+ +-----+
57 : *
58 : * In the example, 5 buffers are pinned, and the next buffer to be streamed to
59 : * the client is block 10. Block 10 was a hit and has no associated I/O, but
60 : * the range 42..44 requires an I/O wait before its buffers are returned, as
61 : * does block 60.
62 : *
63 : *
64 : * Portions Copyright (c) 2024-2025, PostgreSQL Global Development Group
65 : * Portions Copyright (c) 1994, Regents of the University of California
66 : *
67 : * IDENTIFICATION
68 : * src/backend/storage/aio/read_stream.c
69 : *
70 : *-------------------------------------------------------------------------
71 : */
72 : #include "postgres.h"
73 :
74 : #include "miscadmin.h"
75 : #include "storage/aio.h"
76 : #include "storage/fd.h"
77 : #include "storage/smgr.h"
78 : #include "storage/read_stream.h"
79 : #include "utils/memdebug.h"
80 : #include "utils/rel.h"
81 : #include "utils/spccache.h"
82 :
83 : typedef struct InProgressIO
84 : {
85 : int16 buffer_index;
86 : ReadBuffersOperation op;
87 : } InProgressIO;
88 :
89 : /*
90 : * State for managing a stream of reads.
91 : */
92 : struct ReadStream
93 : {
94 : int16 max_ios;
95 : int16 io_combine_limit;
96 : int16 ios_in_progress;
97 : int16 queue_size;
98 : int16 max_pinned_buffers;
99 : int16 forwarded_buffers;
100 : int16 pinned_buffers;
101 : int16 distance;
102 : int16 initialized_buffers;
103 : int read_buffers_flags;
104 : bool sync_mode; /* using io_method=sync */
105 : bool batch_mode; /* READ_STREAM_USE_BATCHING */
106 : bool advice_enabled;
107 : bool temporary;
108 :
109 : /*
110 : * One-block buffer to support 'ungetting' a block number, to resolve flow
111 : * control problems when I/Os are split.
112 : */
113 : BlockNumber buffered_blocknum;
114 :
115 : /*
116 : * The callback that will tell us which block numbers to read, and an
117 : * opaque pointer that will be pass to it for its own purposes.
118 : */
119 : ReadStreamBlockNumberCB callback;
120 : void *callback_private_data;
121 :
122 : /* Next expected block, for detecting sequential access. */
123 : BlockNumber seq_blocknum;
124 : BlockNumber seq_until_processed;
125 :
126 : /* The read operation we are currently preparing. */
127 : BlockNumber pending_read_blocknum;
128 : int16 pending_read_nblocks;
129 :
130 : /* Space for buffers and optional per-buffer private data. */
131 : size_t per_buffer_data_size;
132 : void *per_buffer_data;
133 :
134 : /* Read operations that have been started but not waited for yet. */
135 : InProgressIO *ios;
136 : int16 oldest_io_index;
137 : int16 next_io_index;
138 :
139 : bool fast_path;
140 :
141 : /* Circular queue of buffers. */
142 : int16 oldest_buffer_index; /* Next pinned buffer to return */
143 : int16 next_buffer_index; /* Index of next buffer to pin */
144 : Buffer buffers[FLEXIBLE_ARRAY_MEMBER];
145 : };
146 :
147 : /*
148 : * Return a pointer to the per-buffer data by index.
149 : */
150 : static inline void *
151 6746210 : get_per_buffer_data(ReadStream *stream, int16 buffer_index)
152 : {
153 13492420 : return (char *) stream->per_buffer_data +
154 6746210 : stream->per_buffer_data_size * buffer_index;
155 : }
156 :
157 : /*
158 : * General-use ReadStreamBlockNumberCB for block range scans. Loops over the
159 : * blocks [current_blocknum, last_exclusive).
160 : */
161 : BlockNumber
162 626106 : block_range_read_stream_cb(ReadStream *stream,
163 : void *callback_private_data,
164 : void *per_buffer_data)
165 : {
166 626106 : BlockRangeReadStreamPrivate *p = callback_private_data;
167 :
168 626106 : if (p->current_blocknum < p->last_exclusive)
169 502816 : return p->current_blocknum++;
170 :
171 123290 : return InvalidBlockNumber;
172 : }
173 :
174 : /*
175 : * Ask the callback which block it would like us to read next, with a one block
176 : * buffer in front to allow read_stream_unget_block() to work.
177 : */
178 : static inline BlockNumber
179 9327576 : read_stream_get_block(ReadStream *stream, void *per_buffer_data)
180 : {
181 : BlockNumber blocknum;
182 :
183 9327576 : blocknum = stream->buffered_blocknum;
184 9327576 : if (blocknum != InvalidBlockNumber)
185 8 : stream->buffered_blocknum = InvalidBlockNumber;
186 : else
187 : {
188 : /*
189 : * Tell Valgrind that the per-buffer data is undefined. That replaces
190 : * the "noaccess" state that was set when the consumer moved past this
191 : * entry last time around the queue, and should also catch callbacks
192 : * that fail to initialize data that the buffer consumer later
193 : * accesses. On the first go around, it is undefined already.
194 : */
195 : VALGRIND_MAKE_MEM_UNDEFINED(per_buffer_data,
196 : stream->per_buffer_data_size);
197 9327568 : blocknum = stream->callback(stream,
198 : stream->callback_private_data,
199 : per_buffer_data);
200 : }
201 :
202 9327576 : return blocknum;
203 : }
204 :
205 : /*
206 : * In order to deal with buffer shortages and I/O limits after short reads, we
207 : * sometimes need to defer handling of a block we've already consumed from the
208 : * registered callback until later.
209 : */
210 : static inline void
211 8 : read_stream_unget_block(ReadStream *stream, BlockNumber blocknum)
212 : {
213 : /* We shouldn't ever unget more than one block. */
214 : Assert(stream->buffered_blocknum == InvalidBlockNumber);
215 : Assert(blocknum != InvalidBlockNumber);
216 8 : stream->buffered_blocknum = blocknum;
217 8 : }
218 :
219 : /*
220 : * Start as much of the current pending read as we can. If we have to split it
221 : * because of the per-backend buffer limit, or the buffer manager decides to
222 : * split it, then the pending read is adjusted to hold the remaining portion.
223 : *
224 : * We can always start a read of at least size one if we have no progress yet.
225 : * Otherwise it's possible that we can't start a read at all because of a lack
226 : * of buffers, and then false is returned. Buffer shortages also reduce the
227 : * distance to a level that prevents look-ahead until buffers are released.
228 : */
229 : static bool
230 3506716 : read_stream_start_pending_read(ReadStream *stream)
231 : {
232 : bool need_wait;
233 : int requested_nblocks;
234 : int nblocks;
235 : int flags;
236 : int forwarded;
237 : int16 io_index;
238 : int16 overflow;
239 : int16 buffer_index;
240 : int16 buffer_limit;
241 :
242 : /* This should only be called with a pending read. */
243 : Assert(stream->pending_read_nblocks > 0);
244 : Assert(stream->pending_read_nblocks <= stream->io_combine_limit);
245 :
246 : /* We had better not exceed the per-stream buffer limit with this read. */
247 : Assert(stream->pinned_buffers + stream->pending_read_nblocks <=
248 : stream->max_pinned_buffers);
249 :
250 : /* We had better not be overwriting an existing pinned buffer. */
251 3506716 : if (stream->pinned_buffers > 0)
252 : Assert(stream->next_buffer_index != stream->oldest_buffer_index);
253 : else
254 : Assert(stream->next_buffer_index == stream->oldest_buffer_index);
255 :
256 : /* Do we need to issue read-ahead advice? */
257 3506716 : flags = stream->read_buffers_flags;
258 3506716 : if (stream->advice_enabled)
259 : {
260 0 : if (stream->pending_read_blocknum == stream->seq_blocknum)
261 : {
262 : /*
263 : * Sequential: Issue advice until the preadv() calls have caught
264 : * up with the first advice issued for this sequential region, and
265 : * then stay of the way of the kernel's own read-ahead.
266 : */
267 0 : if (stream->seq_until_processed != InvalidBlockNumber)
268 0 : flags |= READ_BUFFERS_ISSUE_ADVICE;
269 : }
270 : else
271 : {
272 : /*
273 : * Random jump: Note the starting location of a new potential
274 : * sequential region and start issuing advice. Skip it this time
275 : * if the preadv() follows immediately, eg first block in stream.
276 : */
277 0 : stream->seq_until_processed = stream->pending_read_blocknum;
278 0 : if (stream->pinned_buffers > 0)
279 0 : flags |= READ_BUFFERS_ISSUE_ADVICE;
280 : }
281 : }
282 :
283 : /*
284 : * How many more buffers is this backend allowed?
285 : *
286 : * Forwarded buffers are already pinned and map to the leading blocks of
287 : * the pending read (the remaining portion of an earlier short read that
288 : * we're about to continue). They are not counted in pinned_buffers, but
289 : * they are counted as pins already held by this backend according to the
290 : * buffer manager, so they must be added to the limit it grants us.
291 : */
292 3506716 : if (stream->temporary)
293 23550 : buffer_limit = Min(GetAdditionalLocalPinLimit(), PG_INT16_MAX);
294 : else
295 3483166 : buffer_limit = Min(GetAdditionalPinLimit(), PG_INT16_MAX);
296 : Assert(stream->forwarded_buffers <= stream->pending_read_nblocks);
297 3506716 : buffer_limit += stream->forwarded_buffers;
298 3506716 : if (buffer_limit == 0 && stream->pinned_buffers == 0)
299 1262968 : buffer_limit = 1; /* guarantee progress */
300 :
301 : /* Does the per-backend limit affect this read? */
302 3506716 : nblocks = stream->pending_read_nblocks;
303 3506716 : if (buffer_limit < nblocks)
304 : {
305 : int16 new_distance;
306 :
307 : /* Shrink distance: no more look-ahead until buffers are released. */
308 4962 : new_distance = stream->pinned_buffers + buffer_limit;
309 4962 : if (stream->distance > new_distance)
310 3464 : stream->distance = new_distance;
311 :
312 : /* Unless we have nothing to give the consumer, stop here. */
313 4962 : if (stream->pinned_buffers > 0)
314 2444 : return false;
315 :
316 : /* A short read is required to make progress. */
317 2518 : nblocks = buffer_limit;
318 : }
319 :
320 : /*
321 : * We say how many blocks we want to read, but it may be smaller on return
322 : * if the buffer manager decides to shorten the read. Initialize buffers
323 : * to InvalidBuffer (= not a forwarded buffer) as input on first use only,
324 : * and keep the original nblocks number so we can check for forwarded
325 : * buffers as output, below.
326 : */
327 3504272 : buffer_index = stream->next_buffer_index;
328 3504272 : io_index = stream->next_io_index;
329 5469398 : while (stream->initialized_buffers < buffer_index + nblocks)
330 1965126 : stream->buffers[stream->initialized_buffers++] = InvalidBuffer;
331 3504272 : requested_nblocks = nblocks;
332 3504272 : need_wait = StartReadBuffers(&stream->ios[io_index].op,
333 3504272 : &stream->buffers[buffer_index],
334 : stream->pending_read_blocknum,
335 : &nblocks,
336 : flags);
337 3504260 : stream->pinned_buffers += nblocks;
338 :
339 : /* Remember whether we need to wait before returning this buffer. */
340 3504260 : if (!need_wait)
341 : {
342 : /* Look-ahead distance decays, no I/O necessary. */
343 2337552 : if (stream->distance > 1)
344 32820 : stream->distance--;
345 : }
346 : else
347 : {
348 : /*
349 : * Remember to call WaitReadBuffers() before returning head buffer.
350 : * Look-ahead distance will be adjusted after waiting.
351 : */
352 1166708 : stream->ios[io_index].buffer_index = buffer_index;
353 1166708 : if (++stream->next_io_index == stream->max_ios)
354 51306 : stream->next_io_index = 0;
355 : Assert(stream->ios_in_progress < stream->max_ios);
356 1166708 : stream->ios_in_progress++;
357 1166708 : stream->seq_blocknum = stream->pending_read_blocknum + nblocks;
358 : }
359 :
360 : /*
361 : * How many pins were acquired but forwarded to the next call? These need
362 : * to be passed to the next StartReadBuffers() call by leaving them
363 : * exactly where they are in the queue, or released if the stream ends
364 : * early. We need the number for accounting purposes, since they are not
365 : * counted in stream->pinned_buffers but we already hold them.
366 : */
367 3504260 : forwarded = 0;
368 3507784 : while (nblocks + forwarded < requested_nblocks &&
369 100894 : stream->buffers[buffer_index + nblocks + forwarded] != InvalidBuffer)
370 3524 : forwarded++;
371 3504260 : stream->forwarded_buffers = forwarded;
372 :
373 : /*
374 : * We gave a contiguous range of buffer space to StartReadBuffers(), but
375 : * we want it to wrap around at queue_size. Copy overflowing buffers to
376 : * the front of the array where they'll be consumed, but also leave a copy
377 : * in the overflow zone which the I/O operation has a pointer to (it needs
378 : * a contiguous array). Both copies will be cleared when the buffers are
379 : * handed to the consumer.
380 : */
381 3504260 : overflow = (buffer_index + nblocks + forwarded) - stream->queue_size;
382 3504260 : if (overflow > 0)
383 : {
384 : Assert(overflow < stream->queue_size); /* can't overlap */
385 2070 : memcpy(&stream->buffers[0],
386 2070 : &stream->buffers[stream->queue_size],
387 : sizeof(stream->buffers[0]) * overflow);
388 : }
389 :
390 : /* Compute location of start of next read, without using % operator. */
391 3504260 : buffer_index += nblocks;
392 3504260 : if (buffer_index >= stream->queue_size)
393 597020 : buffer_index -= stream->queue_size;
394 : Assert(buffer_index >= 0 && buffer_index < stream->queue_size);
395 3504260 : stream->next_buffer_index = buffer_index;
396 :
397 : /* Adjust the pending read to cover the remaining portion, if any. */
398 3504260 : stream->pending_read_blocknum += nblocks;
399 3504260 : stream->pending_read_nblocks -= nblocks;
400 :
401 3504260 : return true;
402 : }
403 :
404 : static void
405 5801432 : read_stream_look_ahead(ReadStream *stream)
406 : {
407 : /*
408 : * Allow amortizing the cost of submitting IO over multiple IOs. This
409 : * requires that we don't do any operations that could lead to a deadlock
410 : * with staged-but-unsubmitted IO. The callback needs to opt-in to being
411 : * careful.
412 : */
413 5801432 : if (stream->batch_mode)
414 4553450 : pgaio_enter_batchmode();
415 :
416 9635544 : while (stream->ios_in_progress < stream->max_ios &&
417 9635506 : stream->pinned_buffers + stream->pending_read_nblocks < stream->distance)
418 : {
419 : BlockNumber blocknum;
420 : int16 buffer_index;
421 : void *per_buffer_data;
422 :
423 5628006 : if (stream->pending_read_nblocks == stream->io_combine_limit)
424 : {
425 8008 : read_stream_start_pending_read(stream);
426 8008 : continue;
427 : }
428 :
429 : /*
430 : * See which block the callback wants next in the stream. We need to
431 : * compute the index of the Nth block of the pending read including
432 : * wrap-around, but we don't want to use the expensive % operator.
433 : */
434 5619998 : buffer_index = stream->next_buffer_index + stream->pending_read_nblocks;
435 5619998 : if (buffer_index >= stream->queue_size)
436 28000 : buffer_index -= stream->queue_size;
437 : Assert(buffer_index >= 0 && buffer_index < stream->queue_size);
438 5619998 : per_buffer_data = get_per_buffer_data(stream, buffer_index);
439 5619998 : blocknum = read_stream_get_block(stream, per_buffer_data);
440 5619998 : if (blocknum == InvalidBlockNumber)
441 : {
442 : /* End of stream. */
443 1793886 : stream->distance = 0;
444 1793886 : break;
445 : }
446 :
447 : /* Can we merge it with the pending read? */
448 3826112 : if (stream->pending_read_nblocks > 0 &&
449 422962 : stream->pending_read_blocknum + stream->pending_read_nblocks == blocknum)
450 : {
451 422902 : stream->pending_read_nblocks++;
452 422902 : continue;
453 : }
454 :
455 : /* We have to start the pending read before we can build another. */
456 3403282 : while (stream->pending_read_nblocks > 0)
457 : {
458 80 : if (!read_stream_start_pending_read(stream) ||
459 72 : stream->ios_in_progress == stream->max_ios)
460 : {
461 : /* We've hit the buffer or I/O limit. Rewind and stop here. */
462 8 : read_stream_unget_block(stream, blocknum);
463 8 : if (stream->batch_mode)
464 0 : pgaio_exit_batchmode();
465 8 : return;
466 : }
467 : }
468 :
469 : /* This is the start of a new pending read. */
470 3403202 : stream->pending_read_blocknum = blocknum;
471 3403202 : stream->pending_read_nblocks = 1;
472 : }
473 :
474 : /*
475 : * We don't start the pending read just because we've hit the distance
476 : * limit, preferring to give it another chance to grow to full
477 : * io_combine_limit size once more buffers have been consumed. However,
478 : * if we've already reached io_combine_limit, or we've reached the
479 : * distance limit and there isn't anything pinned yet, or the callback has
480 : * signaled end-of-stream, we start the read immediately. Note that the
481 : * pending read can exceed the distance goal, if the latter was reduced
482 : * after hitting the per-backend buffer limit.
483 : */
484 5801424 : if (stream->pending_read_nblocks > 0 &&
485 3597584 : (stream->pending_read_nblocks == stream->io_combine_limit ||
486 3580736 : (stream->pending_read_nblocks >= stream->distance &&
487 3481780 : stream->pinned_buffers == 0) ||
488 111880 : stream->distance == 0) &&
489 3498628 : stream->ios_in_progress < stream->max_ios)
490 3498628 : read_stream_start_pending_read(stream);
491 :
492 : /*
493 : * There should always be something pinned when we leave this function,
494 : * whether started by this call or not, unless we've hit the end of the
495 : * stream. In the worst case we can always make progress one buffer at a
496 : * time.
497 : */
498 : Assert(stream->pinned_buffers > 0 || stream->distance == 0);
499 :
500 5801412 : if (stream->batch_mode)
501 4553438 : pgaio_exit_batchmode();
502 : }
503 :
504 : /*
505 : * Create a new read stream object that can be used to perform the equivalent
506 : * of a series of ReadBuffer() calls for one fork of one relation.
507 : * Internally, it generates larger vectored reads where possible by looking
508 : * ahead. The callback should return block numbers or InvalidBlockNumber to
509 : * signal end-of-stream, and if per_buffer_data_size is non-zero, it may also
510 : * write extra data for each block into the space provided to it. It will
511 : * also receive callback_private_data for its own purposes.
512 : */
513 : static ReadStream *
514 963792 : read_stream_begin_impl(int flags,
515 : BufferAccessStrategy strategy,
516 : Relation rel,
517 : SMgrRelation smgr,
518 : char persistence,
519 : ForkNumber forknum,
520 : ReadStreamBlockNumberCB callback,
521 : void *callback_private_data,
522 : size_t per_buffer_data_size)
523 : {
524 : ReadStream *stream;
525 : size_t size;
526 : int16 queue_size;
527 : int16 queue_overflow;
528 : int max_ios;
529 : int strategy_pin_limit;
530 : uint32 max_pinned_buffers;
531 : uint32 max_possible_buffer_limit;
532 : Oid tablespace_id;
533 :
534 : /*
535 : * Decide how many I/Os we will allow to run at the same time. That
536 : * currently means advice to the kernel to tell it that we will soon read.
537 : * This number also affects how far we look ahead for opportunities to
538 : * start more I/Os.
539 : */
540 963792 : tablespace_id = smgr->smgr_rlocator.locator.spcOid;
541 963792 : if (!OidIsValid(MyDatabaseId) ||
542 1117570 : (rel && IsCatalogRelation(rel)) ||
543 284134 : IsCatalogRelationOid(smgr->smgr_rlocator.locator.relNumber))
544 : {
545 : /*
546 : * Avoid circularity while trying to look up tablespace settings or
547 : * before spccache.c is ready.
548 : */
549 785672 : max_ios = effective_io_concurrency;
550 : }
551 178120 : else if (flags & READ_STREAM_MAINTENANCE)
552 19730 : max_ios = get_tablespace_maintenance_io_concurrency(tablespace_id);
553 : else
554 158390 : max_ios = get_tablespace_io_concurrency(tablespace_id);
555 :
556 : /* Cap to INT16_MAX to avoid overflowing below */
557 963792 : max_ios = Min(max_ios, PG_INT16_MAX);
558 :
559 : /*
560 : * If starting a multi-block I/O near the end of the queue, we might
561 : * temporarily need extra space for overflowing buffers before they are
562 : * moved to regular circular position. This is the maximum extra space we
563 : * could need.
564 : */
565 963792 : queue_overflow = io_combine_limit - 1;
566 :
567 : /*
568 : * Choose the maximum number of buffers we're prepared to pin. We try to
569 : * pin fewer if we can, though. We add one so that we can make progress
570 : * even if max_ios is set to 0 (see also further down). For max_ios > 0,
571 : * this also allows an extra full I/O's worth of buffers: after an I/O
572 : * finishes we don't want to have to wait for its buffers to be consumed
573 : * before starting a new one.
574 : *
575 : * Be careful not to allow int16 to overflow. That is possible with the
576 : * current GUC range limits, so this is an artificial limit of ~32k
577 : * buffers and we'd need to adjust the types to exceed that. We also have
578 : * to allow for the spare entry and the overflow space.
579 : */
580 963792 : max_pinned_buffers = (max_ios + 1) * io_combine_limit;
581 963792 : max_pinned_buffers = Min(max_pinned_buffers,
582 : PG_INT16_MAX - queue_overflow - 1);
583 :
584 : /* Give the strategy a chance to limit the number of buffers we pin. */
585 963792 : strategy_pin_limit = GetAccessStrategyPinLimit(strategy);
586 963792 : max_pinned_buffers = Min(strategy_pin_limit, max_pinned_buffers);
587 :
588 : /*
589 : * Also limit our queue to the maximum number of pins we could ever be
590 : * allowed to acquire according to the buffer manager. We may not really
591 : * be able to use them all due to other pins held by this backend, but
592 : * we'll check that later in read_stream_start_pending_read().
593 : */
594 963792 : if (SmgrIsTemp(smgr))
595 13368 : max_possible_buffer_limit = GetLocalPinLimit();
596 : else
597 950424 : max_possible_buffer_limit = GetPinLimit();
598 963792 : max_pinned_buffers = Min(max_pinned_buffers, max_possible_buffer_limit);
599 :
600 : /*
601 : * The limit might be zero on a system configured with too few buffers for
602 : * the number of connections. We need at least one to make progress.
603 : */
604 963792 : max_pinned_buffers = Max(1, max_pinned_buffers);
605 :
606 : /*
607 : * We need one extra entry for buffers and per-buffer data, because users
608 : * of per-buffer data have access to the object until the next call to
609 : * read_stream_next_buffer(), so we need a gap between the head and tail
610 : * of the queue so that we don't clobber it.
611 : */
612 963792 : queue_size = max_pinned_buffers + 1;
613 :
614 : /*
615 : * Allocate the object, the buffers, the ios and per_buffer_data space in
616 : * one big chunk. Though we have queue_size buffers, we want to be able
617 : * to assume that all the buffers for a single read are contiguous (i.e.
618 : * don't wrap around halfway through), so we allow temporary overflows of
619 : * up to the maximum possible overflow size.
620 : */
621 963792 : size = offsetof(ReadStream, buffers);
622 963792 : size += sizeof(Buffer) * (queue_size + queue_overflow);
623 963792 : size += sizeof(InProgressIO) * Max(1, max_ios);
624 963792 : size += per_buffer_data_size * queue_size;
625 963792 : size += MAXIMUM_ALIGNOF * 2;
626 963792 : stream = (ReadStream *) palloc(size);
627 963792 : memset(stream, 0, offsetof(ReadStream, buffers));
628 963792 : stream->ios = (InProgressIO *)
629 963792 : MAXALIGN(&stream->buffers[queue_size + queue_overflow]);
630 963792 : if (per_buffer_data_size > 0)
631 140272 : stream->per_buffer_data = (void *)
632 140272 : MAXALIGN(&stream->ios[Max(1, max_ios)]);
633 :
634 963792 : stream->sync_mode = io_method == IOMETHOD_SYNC;
635 963792 : stream->batch_mode = flags & READ_STREAM_USE_BATCHING;
636 :
637 : #ifdef USE_PREFETCH
638 :
639 : /*
640 : * Read-ahead advice simulating asynchronous I/O with synchronous calls.
641 : * Issue advice only if AIO is not used, direct I/O isn't enabled, the
642 : * caller hasn't promised sequential access (overriding our detection
643 : * heuristics), and max_ios hasn't been set to zero.
644 : */
645 963792 : if (stream->sync_mode &&
646 0 : (io_direct_flags & IO_DIRECT_DATA) == 0 &&
647 0 : (flags & READ_STREAM_SEQUENTIAL) == 0 &&
648 : max_ios > 0)
649 0 : stream->advice_enabled = true;
650 : #endif
651 :
652 : /*
653 : * Setting max_ios to zero disables AIO and advice-based pseudo AIO, but
654 : * we still need to allocate space to combine and run one I/O. Bump it up
655 : * to one, and remember to ask for synchronous I/O only.
656 : */
657 963792 : if (max_ios == 0)
658 : {
659 0 : max_ios = 1;
660 0 : stream->read_buffers_flags = READ_BUFFERS_SYNCHRONOUSLY;
661 : }
662 :
663 : /*
664 : * Capture stable values for these two GUC-derived numbers for the
665 : * lifetime of this stream, so we don't have to worry about the GUCs
666 : * changing underneath us beyond this point.
667 : */
668 963792 : stream->max_ios = max_ios;
669 963792 : stream->io_combine_limit = io_combine_limit;
670 :
671 963792 : stream->per_buffer_data_size = per_buffer_data_size;
672 963792 : stream->max_pinned_buffers = max_pinned_buffers;
673 963792 : stream->queue_size = queue_size;
674 963792 : stream->callback = callback;
675 963792 : stream->callback_private_data = callback_private_data;
676 963792 : stream->buffered_blocknum = InvalidBlockNumber;
677 963792 : stream->seq_blocknum = InvalidBlockNumber;
678 963792 : stream->seq_until_processed = InvalidBlockNumber;
679 963792 : stream->temporary = SmgrIsTemp(smgr);
680 :
681 : /*
682 : * Skip the initial ramp-up phase if the caller says we're going to be
683 : * reading the whole relation. This way we start out assuming we'll be
684 : * doing full io_combine_limit sized reads.
685 : */
686 963792 : if (flags & READ_STREAM_FULL)
687 124924 : stream->distance = Min(max_pinned_buffers, stream->io_combine_limit);
688 : else
689 838868 : stream->distance = 1;
690 :
691 : /*
692 : * Since we always access the same relation, we can initialize parts of
693 : * the ReadBuffersOperation objects and leave them that way, to avoid
694 : * wasting CPU cycles writing to them for each read.
695 : */
696 16427372 : for (int i = 0; i < max_ios; ++i)
697 : {
698 15463580 : stream->ios[i].op.rel = rel;
699 15463580 : stream->ios[i].op.smgr = smgr;
700 15463580 : stream->ios[i].op.persistence = persistence;
701 15463580 : stream->ios[i].op.forknum = forknum;
702 15463580 : stream->ios[i].op.strategy = strategy;
703 : }
704 :
705 963792 : return stream;
706 : }
707 :
708 : /*
709 : * Create a new read stream for reading a relation.
710 : * See read_stream_begin_impl() for the detailed explanation.
711 : */
712 : ReadStream *
713 850102 : read_stream_begin_relation(int flags,
714 : BufferAccessStrategy strategy,
715 : Relation rel,
716 : ForkNumber forknum,
717 : ReadStreamBlockNumberCB callback,
718 : void *callback_private_data,
719 : size_t per_buffer_data_size)
720 : {
721 850102 : return read_stream_begin_impl(flags,
722 : strategy,
723 : rel,
724 : RelationGetSmgr(rel),
725 850102 : rel->rd_rel->relpersistence,
726 : forknum,
727 : callback,
728 : callback_private_data,
729 : per_buffer_data_size);
730 : }
731 :
732 : /*
733 : * Create a new read stream for reading a SMgr relation.
734 : * See read_stream_begin_impl() for the detailed explanation.
735 : */
736 : ReadStream *
737 113690 : read_stream_begin_smgr_relation(int flags,
738 : BufferAccessStrategy strategy,
739 : SMgrRelation smgr,
740 : char smgr_persistence,
741 : ForkNumber forknum,
742 : ReadStreamBlockNumberCB callback,
743 : void *callback_private_data,
744 : size_t per_buffer_data_size)
745 : {
746 113690 : return read_stream_begin_impl(flags,
747 : strategy,
748 : NULL,
749 : smgr,
750 : smgr_persistence,
751 : forknum,
752 : callback,
753 : callback_private_data,
754 : per_buffer_data_size);
755 : }
756 :
757 : /*
758 : * Pull one pinned buffer out of a stream. Each call returns successive
759 : * blocks in the order specified by the callback. If per_buffer_data_size was
760 : * set to a non-zero size, *per_buffer_data receives a pointer to the extra
761 : * per-buffer data that the callback had a chance to populate, which remains
762 : * valid until the next call to read_stream_next_buffer(). When the stream
763 : * runs out of data, InvalidBuffer is returned. The caller may decide to end
764 : * the stream early at any time by calling read_stream_end().
765 : */
766 : Buffer
767 11217350 : read_stream_next_buffer(ReadStream *stream, void **per_buffer_data)
768 : {
769 : Buffer buffer;
770 : int16 oldest_buffer_index;
771 :
772 : #ifndef READ_STREAM_DISABLE_FAST_PATH
773 :
774 : /*
775 : * A fast path for all-cached scans. This is the same as the usual
776 : * algorithm, but it is specialized for no I/O and no per-buffer data, so
777 : * we can skip the queue management code, stay in the same buffer slot and
778 : * use singular StartReadBuffer().
779 : */
780 11217350 : if (likely(stream->fast_path))
781 : {
782 : BlockNumber next_blocknum;
783 :
784 : /* Fast path assumptions. */
785 : Assert(stream->ios_in_progress == 0);
786 : Assert(stream->forwarded_buffers == 0);
787 : Assert(stream->pinned_buffers == 1);
788 : Assert(stream->distance == 1);
789 : Assert(stream->pending_read_nblocks == 0);
790 : Assert(stream->per_buffer_data_size == 0);
791 : Assert(stream->initialized_buffers > stream->oldest_buffer_index);
792 :
793 : /* We're going to return the buffer we pinned last time. */
794 3707578 : oldest_buffer_index = stream->oldest_buffer_index;
795 : Assert((oldest_buffer_index + 1) % stream->queue_size ==
796 : stream->next_buffer_index);
797 3707578 : buffer = stream->buffers[oldest_buffer_index];
798 : Assert(buffer != InvalidBuffer);
799 :
800 : /* Choose the next block to pin. */
801 3707578 : next_blocknum = read_stream_get_block(stream, NULL);
802 :
803 3707578 : if (likely(next_blocknum != InvalidBlockNumber))
804 : {
805 3550688 : int flags = stream->read_buffers_flags;
806 :
807 3550688 : if (stream->advice_enabled)
808 0 : flags |= READ_BUFFERS_ISSUE_ADVICE;
809 :
810 : /*
811 : * Pin a buffer for the next call. Same buffer entry, and
812 : * arbitrary I/O entry (they're all free). We don't have to
813 : * adjust pinned_buffers because we're transferring one to caller
814 : * but pinning one more.
815 : *
816 : * In the fast path we don't need to check the pin limit. We're
817 : * always allowed at least one pin so that progress can be made,
818 : * and that's all we need here. Although two pins are momentarily
819 : * held at the same time, the model used here is that the stream
820 : * holds only one, and the other now belongs to the caller.
821 : */
822 3550688 : if (likely(!StartReadBuffer(&stream->ios[0].op,
823 : &stream->buffers[oldest_buffer_index],
824 : next_blocknum,
825 : flags)))
826 : {
827 : /* Fast return. */
828 3527232 : return buffer;
829 : }
830 :
831 : /* Next call must wait for I/O for the newly pinned buffer. */
832 23456 : stream->oldest_io_index = 0;
833 23456 : stream->next_io_index = stream->max_ios > 1 ? 1 : 0;
834 23456 : stream->ios_in_progress = 1;
835 23456 : stream->ios[0].buffer_index = oldest_buffer_index;
836 23456 : stream->seq_blocknum = next_blocknum + 1;
837 : }
838 : else
839 : {
840 : /* No more blocks, end of stream. */
841 156890 : stream->distance = 0;
842 156890 : stream->oldest_buffer_index = stream->next_buffer_index;
843 156890 : stream->pinned_buffers = 0;
844 156890 : stream->buffers[oldest_buffer_index] = InvalidBuffer;
845 : }
846 :
847 180346 : stream->fast_path = false;
848 180346 : return buffer;
849 : }
850 : #endif
851 :
852 7509772 : if (unlikely(stream->pinned_buffers == 0))
853 : {
854 : Assert(stream->oldest_buffer_index == stream->next_buffer_index);
855 :
856 : /* End of stream reached? */
857 4991748 : if (stream->distance == 0)
858 2857752 : return InvalidBuffer;
859 :
860 : /*
861 : * The usual order of operations is that we look ahead at the bottom
862 : * of this function after potentially finishing an I/O and making
863 : * space for more, but if we're just starting up we'll need to crank
864 : * the handle to get started.
865 : */
866 2133996 : read_stream_look_ahead(stream);
867 :
868 : /* End of stream reached? */
869 2133996 : if (stream->pinned_buffers == 0)
870 : {
871 : Assert(stream->distance == 0);
872 984584 : return InvalidBuffer;
873 : }
874 : }
875 :
876 : /* Grab the oldest pinned buffer and associated per-buffer data. */
877 : Assert(stream->pinned_buffers > 0);
878 3667436 : oldest_buffer_index = stream->oldest_buffer_index;
879 : Assert(oldest_buffer_index >= 0 &&
880 : oldest_buffer_index < stream->queue_size);
881 3667436 : buffer = stream->buffers[oldest_buffer_index];
882 3667436 : if (per_buffer_data)
883 1126212 : *per_buffer_data = get_per_buffer_data(stream, oldest_buffer_index);
884 :
885 : Assert(BufferIsValid(buffer));
886 :
887 : /* Do we have to wait for an associated I/O first? */
888 3667436 : if (stream->ios_in_progress > 0 &&
889 1360714 : stream->ios[stream->oldest_io_index].buffer_index == oldest_buffer_index)
890 : {
891 1189570 : int16 io_index = stream->oldest_io_index;
892 : int32 distance; /* wider temporary value, clamped below */
893 :
894 : /* Sanity check that we still agree on the buffers. */
895 : Assert(stream->ios[io_index].op.buffers ==
896 : &stream->buffers[oldest_buffer_index]);
897 :
898 1189570 : WaitReadBuffers(&stream->ios[io_index].op);
899 :
900 : Assert(stream->ios_in_progress > 0);
901 1189570 : stream->ios_in_progress--;
902 1189570 : if (++stream->oldest_io_index == stream->max_ios)
903 51306 : stream->oldest_io_index = 0;
904 :
905 : /* Look-ahead distance ramps up rapidly after we do I/O. */
906 1189570 : distance = stream->distance * 2;
907 1189570 : distance = Min(distance, stream->max_pinned_buffers);
908 1189570 : stream->distance = distance;
909 :
910 : /*
911 : * If we've reached the first block of a sequential region we're
912 : * issuing advice for, cancel that until the next jump. The kernel
913 : * will see the sequential preadv() pattern starting here.
914 : */
915 1189570 : if (stream->advice_enabled &&
916 0 : stream->ios[io_index].op.blocknum == stream->seq_until_processed)
917 0 : stream->seq_until_processed = InvalidBlockNumber;
918 : }
919 :
920 : /*
921 : * We must zap this queue entry, or else it would appear as a forwarded
922 : * buffer. If it's potentially in the overflow zone (ie from a
923 : * multi-block I/O that wrapped around the queue), also zap the copy.
924 : */
925 3667436 : stream->buffers[oldest_buffer_index] = InvalidBuffer;
926 3667436 : if (oldest_buffer_index < stream->io_combine_limit - 1)
927 2642170 : stream->buffers[stream->queue_size + oldest_buffer_index] =
928 : InvalidBuffer;
929 :
930 : #if defined(CLOBBER_FREED_MEMORY) || defined(USE_VALGRIND)
931 :
932 : /*
933 : * The caller will get access to the per-buffer data, until the next call.
934 : * We wipe the one before, which is never occupied because queue_size
935 : * allowed one extra element. This will hopefully trip up client code
936 : * that is holding a dangling pointer to it.
937 : */
938 : if (stream->per_buffer_data)
939 : {
940 : void *per_buffer_data;
941 :
942 : per_buffer_data = get_per_buffer_data(stream,
943 : oldest_buffer_index == 0 ?
944 : stream->queue_size - 1 :
945 : oldest_buffer_index - 1);
946 :
947 : #if defined(CLOBBER_FREED_MEMORY)
948 : /* This also tells Valgrind the memory is "noaccess". */
949 : wipe_mem(per_buffer_data, stream->per_buffer_data_size);
950 : #elif defined(USE_VALGRIND)
951 : /* Tell it ourselves. */
952 : VALGRIND_MAKE_MEM_NOACCESS(per_buffer_data,
953 : stream->per_buffer_data_size);
954 : #endif
955 : }
956 : #endif
957 :
958 : /* Pin transferred to caller. */
959 : Assert(stream->pinned_buffers > 0);
960 3667436 : stream->pinned_buffers--;
961 :
962 : /* Advance oldest buffer, with wrap-around. */
963 3667436 : stream->oldest_buffer_index++;
964 3667436 : if (stream->oldest_buffer_index == stream->queue_size)
965 583670 : stream->oldest_buffer_index = 0;
966 :
967 : /* Prepare for the next call. */
968 3667436 : read_stream_look_ahead(stream);
969 :
970 : #ifndef READ_STREAM_DISABLE_FAST_PATH
971 : /* See if we can take the fast path for all-cached scans next time. */
972 3667424 : if (stream->ios_in_progress == 0 &&
973 2464858 : stream->forwarded_buffers == 0 &&
974 2464386 : stream->pinned_buffers == 1 &&
975 1357960 : stream->distance == 1 &&
976 1231506 : stream->pending_read_nblocks == 0 &&
977 1229242 : stream->per_buffer_data_size == 0)
978 : {
979 353480 : stream->fast_path = true;
980 : }
981 : #endif
982 :
983 3667424 : return buffer;
984 : }
985 :
986 : /*
987 : * Transitional support for code that would like to perform or skip reads
988 : * itself, without using the stream. Returns, and consumes, the next block
989 : * number that would be read by the stream's look-ahead algorithm, or
990 : * InvalidBlockNumber if the end of the stream is reached. Also reports the
991 : * strategy that would be used to read it.
992 : */
993 : BlockNumber
994 0 : read_stream_next_block(ReadStream *stream, BufferAccessStrategy *strategy)
995 : {
996 0 : *strategy = stream->ios[0].op.strategy;
997 0 : return read_stream_get_block(stream, NULL);
998 : }
999 :
1000 : /*
1001 : * Reset a read stream by releasing any queued up buffers, allowing the stream
1002 : * to be used again for different blocks. This can be used to clear an
1003 : * end-of-stream condition and start again, or to throw away blocks that were
1004 : * speculatively read and read some different blocks instead.
1005 : */
1006 : void
1007 2135264 : read_stream_reset(ReadStream *stream)
1008 : {
1009 : int16 index;
1010 : Buffer buffer;
1011 :
1012 : /* Stop looking ahead. */
1013 2135264 : stream->distance = 0;
1014 :
1015 : /* Forget buffered block number and fast path state. */
1016 2135264 : stream->buffered_blocknum = InvalidBlockNumber;
1017 2135264 : stream->fast_path = false;
1018 :
1019 : /* Unpin anything that wasn't consumed. */
1020 2358658 : while ((buffer = read_stream_next_buffer(stream, NULL)) != InvalidBuffer)
1021 223394 : ReleaseBuffer(buffer);
1022 :
1023 : /* Unpin any unused forwarded buffers. */
1024 2135264 : index = stream->next_buffer_index;
1025 2135264 : while (index < stream->initialized_buffers &&
1026 435004 : (buffer = stream->buffers[index]) != InvalidBuffer)
1027 : {
1028 : Assert(stream->forwarded_buffers > 0);
1029 0 : stream->forwarded_buffers--;
1030 0 : ReleaseBuffer(buffer);
1031 :
1032 0 : stream->buffers[index] = InvalidBuffer;
1033 0 : if (index < stream->io_combine_limit - 1)
1034 0 : stream->buffers[stream->queue_size + index] = InvalidBuffer;
1035 :
1036 0 : if (++index == stream->queue_size)
1037 0 : index = 0;
1038 : }
1039 :
1040 : Assert(stream->forwarded_buffers == 0);
1041 : Assert(stream->pinned_buffers == 0);
1042 : Assert(stream->ios_in_progress == 0);
1043 :
1044 : /* Start off assuming data is cached. */
1045 2135264 : stream->distance = 1;
1046 2135264 : }
1047 :
1048 : /*
1049 : * Release and free a read stream.
1050 : */
1051 : void
1052 959336 : read_stream_end(ReadStream *stream)
1053 : {
1054 959336 : read_stream_reset(stream);
1055 959336 : pfree(stream);
1056 959336 : }
|
