PostgreSQL Source Code  git master
sharedtuplestore.c File Reference
#include "postgres.h"
#include "access/htup.h"
#include "access/htup_details.h"
#include "storage/buffile.h"
#include "storage/lwlock.h"
#include "storage/sharedfileset.h"
#include "utils/sharedtuplestore.h"
Include dependency graph for sharedtuplestore.c:

Go to the source code of this file.

Data Structures

struct  SharedTuplestoreChunk
 
struct  SharedTuplestoreParticipant
 
struct  SharedTuplestore
 
struct  SharedTuplestoreAccessor
 

Macros

#define STS_CHUNK_PAGES   4
 
#define STS_CHUNK_HEADER_SIZE   offsetof(SharedTuplestoreChunk, data)
 
#define STS_CHUNK_DATA_SIZE   (STS_CHUNK_PAGES * BLCKSZ - STS_CHUNK_HEADER_SIZE)
 

Typedefs

typedef struct SharedTuplestoreChunk SharedTuplestoreChunk
 
typedef struct SharedTuplestoreParticipant SharedTuplestoreParticipant
 

Functions

static void sts_filename (char *name, SharedTuplestoreAccessor *accessor, int participant)
 
size_t sts_estimate (int participants)
 
SharedTuplestoreAccessorsts_initialize (SharedTuplestore *sts, int participants, int my_participant_number, size_t meta_data_size, int flags, SharedFileSet *fileset, const char *name)
 
SharedTuplestoreAccessorsts_attach (SharedTuplestore *sts, int my_participant_number, SharedFileSet *fileset)
 
static void sts_flush_chunk (SharedTuplestoreAccessor *accessor)
 
void sts_end_write (SharedTuplestoreAccessor *accessor)
 
void sts_reinitialize (SharedTuplestoreAccessor *accessor)
 
void sts_begin_parallel_scan (SharedTuplestoreAccessor *accessor)
 
void sts_end_parallel_scan (SharedTuplestoreAccessor *accessor)
 
void sts_puttuple (SharedTuplestoreAccessor *accessor, void *meta_data, MinimalTuple tuple)
 
static MinimalTuple sts_read_tuple (SharedTuplestoreAccessor *accessor, void *meta_data)
 
MinimalTuple sts_parallel_scan_next (SharedTuplestoreAccessor *accessor, void *meta_data)
 

Macro Definition Documentation

◆ STS_CHUNK_DATA_SIZE

#define STS_CHUNK_DATA_SIZE   (STS_CHUNK_PAGES * BLCKSZ - STS_CHUNK_HEADER_SIZE)

Definition at line 39 of file sharedtuplestore.c.

◆ STS_CHUNK_HEADER_SIZE

#define STS_CHUNK_HEADER_SIZE   offsetof(SharedTuplestoreChunk, data)

Definition at line 38 of file sharedtuplestore.c.

◆ STS_CHUNK_PAGES

#define STS_CHUNK_PAGES   4

Definition at line 37 of file sharedtuplestore.c.

Typedef Documentation

◆ SharedTuplestoreChunk

◆ SharedTuplestoreParticipant

Function Documentation

◆ sts_attach()

SharedTuplestoreAccessor* sts_attach ( SharedTuplestore sts,
int  my_participant_number,
SharedFileSet fileset 
)

Definition at line 178 of file sharedtuplestore.c.

181 {
182  SharedTuplestoreAccessor *accessor;
183 
184  Assert(my_participant_number < sts->nparticipants);
185 
186  accessor = palloc0(sizeof(SharedTuplestoreAccessor));
187  accessor->participant = my_participant_number;
188  accessor->sts = sts;
189  accessor->fileset = fileset;
190  accessor->context = CurrentMemoryContext;
191 
192  return accessor;
193 }
#define Assert(condition)
Definition: c.h:861
void * palloc0(Size size)
Definition: mcxt.c:1347
MemoryContext CurrentMemoryContext
Definition: mcxt.c:143
SharedTuplestore * sts

References Assert, SharedTuplestoreAccessor::context, CurrentMemoryContext, SharedTuplestoreAccessor::fileset, palloc0(), SharedTuplestoreAccessor::participant, and SharedTuplestoreAccessor::sts.

Referenced by ExecParallelHashEnsureBatchAccessors(), and ExecParallelHashRepartitionRest().

◆ sts_begin_parallel_scan()

void sts_begin_parallel_scan ( SharedTuplestoreAccessor accessor)

Definition at line 253 of file sharedtuplestore.c.

254 {
256 
257  /* End any existing scan that was in progress. */
258  sts_end_parallel_scan(accessor);
259 
260  /*
261  * Any backend that might have written into this shared tuplestore must
262  * have called sts_end_write(), so that all buffers are flushed and the
263  * files have stopped growing.
264  */
265  for (i = 0; i < accessor->sts->nparticipants; ++i)
266  Assert(!accessor->sts->participants[i].writing);
267 
268  /*
269  * We will start out reading the file that THIS backend wrote. There may
270  * be some caching locality advantage to that.
271  */
272  accessor->read_participant = accessor->participant;
273  accessor->read_file = NULL;
274  accessor->read_next_page = 0;
275 }
#define PG_USED_FOR_ASSERTS_ONLY
Definition: c.h:185
int i
Definition: isn.c:73
void sts_end_parallel_scan(SharedTuplestoreAccessor *accessor)
SharedTuplestoreParticipant participants[FLEXIBLE_ARRAY_MEMBER]

References Assert, i, SharedTuplestore::nparticipants, SharedTuplestoreAccessor::participant, SharedTuplestore::participants, PG_USED_FOR_ASSERTS_ONLY, SharedTuplestoreAccessor::read_file, SharedTuplestoreAccessor::read_next_page, SharedTuplestoreAccessor::read_participant, SharedTuplestoreAccessor::sts, sts_end_parallel_scan(), and SharedTuplestoreParticipant::writing.

Referenced by ExecParallelHashJoinNewBatch(), and ExecParallelHashRepartitionRest().

◆ sts_end_parallel_scan()

void sts_end_parallel_scan ( SharedTuplestoreAccessor accessor)

Definition at line 281 of file sharedtuplestore.c.

282 {
283  /*
284  * Here we could delete all files if SHARED_TUPLESTORE_SINGLE_PASS, but
285  * we'd probably need a reference count of current parallel scanners so we
286  * could safely do it only when the reference count reaches zero.
287  */
288  if (accessor->read_file != NULL)
289  {
290  BufFileClose(accessor->read_file);
291  accessor->read_file = NULL;
292  }
293 }
void BufFileClose(BufFile *file)
Definition: buffile.c:412

References BufFileClose(), and SharedTuplestoreAccessor::read_file.

Referenced by ExecHashTableDetach(), ExecHashTableDetachBatch(), ExecParallelHashCloseBatchAccessors(), ExecParallelHashJoinNewBatch(), ExecParallelHashRepartitionRest(), ExecParallelPrepHashTableForUnmatched(), and sts_begin_parallel_scan().

◆ sts_end_write()

void sts_end_write ( SharedTuplestoreAccessor accessor)

Definition at line 213 of file sharedtuplestore.c.

214 {
215  if (accessor->write_file != NULL)
216  {
217  sts_flush_chunk(accessor);
218  BufFileClose(accessor->write_file);
219  pfree(accessor->write_chunk);
220  accessor->write_chunk = NULL;
221  accessor->write_file = NULL;
222  accessor->sts->participants[accessor->participant].writing = false;
223  }
224 }
void pfree(void *pointer)
Definition: mcxt.c:1521
static void sts_flush_chunk(SharedTuplestoreAccessor *accessor)
SharedTuplestoreChunk * write_chunk

References BufFileClose(), SharedTuplestoreAccessor::participant, SharedTuplestore::participants, pfree(), SharedTuplestoreAccessor::sts, sts_flush_chunk(), SharedTuplestoreAccessor::write_chunk, SharedTuplestoreAccessor::write_file, and SharedTuplestoreParticipant::writing.

Referenced by ExecHashTableDetach(), ExecParallelHashCloseBatchAccessors(), ExecParallelHashJoinPartitionOuter(), and MultiExecParallelHash().

◆ sts_estimate()

size_t sts_estimate ( int  participants)

Definition at line 104 of file sharedtuplestore.c.

105 {
106  return offsetof(SharedTuplestore, participants) +
107  sizeof(SharedTuplestoreParticipant) * participants;
108 }
struct SharedTuplestoreParticipant SharedTuplestoreParticipant

◆ sts_filename()

static void sts_filename ( char *  name,
SharedTuplestoreAccessor accessor,
int  participant 
)
static

Definition at line 598 of file sharedtuplestore.c.

599 {
600  snprintf(name, MAXPGPATH, "%s.p%d", accessor->sts->name, participant);
601 }
#define MAXPGPATH
#define snprintf
Definition: port.h:238
char name[NAMEDATALEN]
const char * name

References MAXPGPATH, name, SharedTuplestore::name, snprintf, and SharedTuplestoreAccessor::sts.

Referenced by sts_parallel_scan_next(), and sts_puttuple().

◆ sts_flush_chunk()

static void sts_flush_chunk ( SharedTuplestoreAccessor accessor)
static

Definition at line 196 of file sharedtuplestore.c.

197 {
198  size_t size;
199 
200  size = STS_CHUNK_PAGES * BLCKSZ;
201  BufFileWrite(accessor->write_file, accessor->write_chunk, size);
202  memset(accessor->write_chunk, 0, size);
203  accessor->write_pointer = &accessor->write_chunk->data[0];
204  accessor->sts->participants[accessor->participant].npages +=
206 }
void BufFileWrite(BufFile *file, const void *ptr, size_t size)
Definition: buffile.c:676
#define STS_CHUNK_PAGES
static pg_noinline void Size size
Definition: slab.c:607
char data[FLEXIBLE_ARRAY_MEMBER]

References BufFileWrite(), SharedTuplestoreChunk::data, SharedTuplestoreParticipant::npages, SharedTuplestoreAccessor::participant, SharedTuplestore::participants, size, SharedTuplestoreAccessor::sts, STS_CHUNK_PAGES, SharedTuplestoreAccessor::write_chunk, SharedTuplestoreAccessor::write_file, and SharedTuplestoreAccessor::write_pointer.

Referenced by sts_end_write(), and sts_puttuple().

◆ sts_initialize()

SharedTuplestoreAccessor* sts_initialize ( SharedTuplestore sts,
int  participants,
int  my_participant_number,
size_t  meta_data_size,
int  flags,
SharedFileSet fileset,
const char *  name 
)

Definition at line 126 of file sharedtuplestore.c.

132 {
133  SharedTuplestoreAccessor *accessor;
134  int i;
135 
136  Assert(my_participant_number < participants);
137 
138  sts->nparticipants = participants;
139  sts->meta_data_size = meta_data_size;
140  sts->flags = flags;
141 
142  if (strlen(name) > sizeof(sts->name) - 1)
143  elog(ERROR, "SharedTuplestore name too long");
144  strcpy(sts->name, name);
145 
146  /*
147  * Limit meta-data so it + tuple size always fits into a single chunk.
148  * sts_puttuple() and sts_read_tuple() could be made to support scenarios
149  * where that's not the case, but it's not currently required. If so,
150  * meta-data size probably should be made variable, too.
151  */
152  if (meta_data_size + sizeof(uint32) >= STS_CHUNK_DATA_SIZE)
153  elog(ERROR, "meta-data too long");
154 
155  for (i = 0; i < participants; ++i)
156  {
159  sts->participants[i].read_page = 0;
160  sts->participants[i].npages = 0;
161  sts->participants[i].writing = false;
162  }
163 
164  accessor = palloc0(sizeof(SharedTuplestoreAccessor));
165  accessor->participant = my_participant_number;
166  accessor->sts = sts;
167  accessor->fileset = fileset;
168  accessor->context = CurrentMemoryContext;
169 
170  return accessor;
171 }
unsigned int uint32
Definition: c.h:509
#define ERROR
Definition: elog.h:39
#define elog(elevel,...)
Definition: elog.h:225
void LWLockInitialize(LWLock *lock, int tranche_id)
Definition: lwlock.c:707
@ LWTRANCHE_SHARED_TUPLESTORE
Definition: lwlock.h:199
#define STS_CHUNK_DATA_SIZE

References Assert, SharedTuplestoreAccessor::context, CurrentMemoryContext, elog, ERROR, SharedTuplestoreAccessor::fileset, SharedTuplestore::flags, i, SharedTuplestoreParticipant::lock, LWLockInitialize(), LWTRANCHE_SHARED_TUPLESTORE, SharedTuplestore::meta_data_size, name, SharedTuplestore::name, SharedTuplestoreParticipant::npages, SharedTuplestore::nparticipants, palloc0(), SharedTuplestoreAccessor::participant, SharedTuplestore::participants, SharedTuplestoreParticipant::read_page, SharedTuplestoreAccessor::sts, STS_CHUNK_DATA_SIZE, and SharedTuplestoreParticipant::writing.

Referenced by ExecParallelHashJoinSetUpBatches().

◆ sts_parallel_scan_next()

MinimalTuple sts_parallel_scan_next ( SharedTuplestoreAccessor accessor,
void *  meta_data 
)

Definition at line 495 of file sharedtuplestore.c.

496 {
498  BlockNumber read_page;
499  bool eof;
500 
501  for (;;)
502  {
503  /* Can we read more tuples from the current chunk? */
504  if (accessor->read_ntuples < accessor->read_ntuples_available)
505  return sts_read_tuple(accessor, meta_data);
506 
507  /* Find the location of a new chunk to read. */
508  p = &accessor->sts->participants[accessor->read_participant];
509 
511  /* We can skip directly past overflow pages we know about. */
512  if (p->read_page < accessor->read_next_page)
513  p->read_page = accessor->read_next_page;
514  eof = p->read_page >= p->npages;
515  if (!eof)
516  {
517  /* Claim the next chunk. */
518  read_page = p->read_page;
519  /* Advance the read head for the next reader. */
521  accessor->read_next_page = p->read_page;
522  }
523  LWLockRelease(&p->lock);
524 
525  if (!eof)
526  {
527  SharedTuplestoreChunk chunk_header;
528 
529  /* Make sure we have the file open. */
530  if (accessor->read_file == NULL)
531  {
532  char name[MAXPGPATH];
533  MemoryContext oldcxt;
534 
535  sts_filename(name, accessor, accessor->read_participant);
536 
537  oldcxt = MemoryContextSwitchTo(accessor->context);
538  accessor->read_file =
539  BufFileOpenFileSet(&accessor->fileset->fs, name, O_RDONLY,
540  false);
541  MemoryContextSwitchTo(oldcxt);
542  }
543 
544  /* Seek and load the chunk header. */
545  if (BufFileSeekBlock(accessor->read_file, read_page) != 0)
546  ereport(ERROR,
548  errmsg("could not seek to block %u in shared tuplestore temporary file",
549  read_page)));
550  BufFileReadExact(accessor->read_file, &chunk_header, STS_CHUNK_HEADER_SIZE);
551 
552  /*
553  * If this is an overflow chunk, we skip it and any following
554  * overflow chunks all at once.
555  */
556  if (chunk_header.overflow > 0)
557  {
558  accessor->read_next_page = read_page +
559  chunk_header.overflow * STS_CHUNK_PAGES;
560  continue;
561  }
562 
563  accessor->read_ntuples = 0;
564  accessor->read_ntuples_available = chunk_header.ntuples;
565  accessor->read_bytes = STS_CHUNK_HEADER_SIZE;
566 
567  /* Go around again, so we can get a tuple from this chunk. */
568  }
569  else
570  {
571  if (accessor->read_file != NULL)
572  {
573  BufFileClose(accessor->read_file);
574  accessor->read_file = NULL;
575  }
576 
577  /*
578  * Try the next participant's file. If we've gone full circle,
579  * we're done.
580  */
581  accessor->read_participant = (accessor->read_participant + 1) %
582  accessor->sts->nparticipants;
583  if (accessor->read_participant == accessor->participant)
584  break;
585  accessor->read_next_page = 0;
586 
587  /* Go around again, so we can get a chunk from this file. */
588  }
589  }
590 
591  return NULL;
592 }
uint32 BlockNumber
Definition: block.h:31
int BufFileSeekBlock(BufFile *file, int64 blknum)
Definition: buffile.c:851
void BufFileReadExact(BufFile *file, void *ptr, size_t size)
Definition: buffile.c:654
BufFile * BufFileOpenFileSet(FileSet *fileset, const char *name, int mode, bool missing_ok)
Definition: buffile.c:291
int errcode_for_file_access(void)
Definition: elog.c:876
int errmsg(const char *fmt,...)
Definition: elog.c:1070
#define ereport(elevel,...)
Definition: elog.h:149
if(TABLE==NULL||TABLE_index==NULL)
Definition: isn.c:77
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1168
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1781
@ LW_EXCLUSIVE
Definition: lwlock.h:114
MemoryContextSwitchTo(old_ctx)
static MinimalTuple sts_read_tuple(SharedTuplestoreAccessor *accessor, void *meta_data)
#define STS_CHUNK_HEADER_SIZE
static void sts_filename(char *name, SharedTuplestoreAccessor *accessor, int participant)

References BufFileClose(), BufFileOpenFileSet(), BufFileReadExact(), BufFileSeekBlock(), SharedTuplestoreAccessor::context, ereport, errcode_for_file_access(), errmsg(), ERROR, SharedTuplestoreAccessor::fileset, SharedFileSet::fs, if(), SharedTuplestoreParticipant::lock, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MAXPGPATH, MemoryContextSwitchTo(), name, SharedTuplestoreParticipant::npages, SharedTuplestore::nparticipants, SharedTuplestoreChunk::ntuples, SharedTuplestoreChunk::overflow, SharedTuplestoreAccessor::participant, SharedTuplestore::participants, SharedTuplestoreAccessor::read_bytes, SharedTuplestoreAccessor::read_file, SharedTuplestoreAccessor::read_next_page, SharedTuplestoreAccessor::read_ntuples, SharedTuplestoreAccessor::read_ntuples_available, SharedTuplestoreParticipant::read_page, SharedTuplestoreAccessor::read_participant, SharedTuplestoreAccessor::sts, STS_CHUNK_HEADER_SIZE, STS_CHUNK_PAGES, sts_filename(), and sts_read_tuple().

Referenced by ExecParallelHashJoinNewBatch(), ExecParallelHashJoinOuterGetTuple(), and ExecParallelHashRepartitionRest().

◆ sts_puttuple()

void sts_puttuple ( SharedTuplestoreAccessor accessor,
void *  meta_data,
MinimalTuple  tuple 
)

Definition at line 300 of file sharedtuplestore.c.

302 {
303  size_t size;
304 
305  /* Do we have our own file yet? */
306  if (accessor->write_file == NULL)
307  {
308  SharedTuplestoreParticipant *participant;
309  char name[MAXPGPATH];
310  MemoryContext oldcxt;
311 
312  /* Create one. Only this backend will write into it. */
313  sts_filename(name, accessor, accessor->participant);
314 
315  oldcxt = MemoryContextSwitchTo(accessor->context);
316  accessor->write_file =
317  BufFileCreateFileSet(&accessor->fileset->fs, name);
318  MemoryContextSwitchTo(oldcxt);
319 
320  /* Set up the shared state for this backend's file. */
321  participant = &accessor->sts->participants[accessor->participant];
322  participant->writing = true; /* for assertions only */
323  }
324 
325  /* Do we have space? */
326  size = accessor->sts->meta_data_size + tuple->t_len;
327  if (accessor->write_pointer + size > accessor->write_end)
328  {
329  if (accessor->write_chunk == NULL)
330  {
331  /* First time through. Allocate chunk. */
332  accessor->write_chunk = (SharedTuplestoreChunk *)
334  STS_CHUNK_PAGES * BLCKSZ);
335  accessor->write_chunk->ntuples = 0;
336  accessor->write_pointer = &accessor->write_chunk->data[0];
337  accessor->write_end = (char *)
338  accessor->write_chunk + STS_CHUNK_PAGES * BLCKSZ;
339  }
340  else
341  {
342  /* See if flushing helps. */
343  sts_flush_chunk(accessor);
344  }
345 
346  /* It may still not be enough in the case of a gigantic tuple. */
347  if (accessor->write_pointer + size > accessor->write_end)
348  {
349  size_t written;
350 
351  /*
352  * We'll write the beginning of the oversized tuple, and then
353  * write the rest in some number of 'overflow' chunks.
354  *
355  * sts_initialize() verifies that the size of the tuple +
356  * meta-data always fits into a chunk. Because the chunk has been
357  * flushed above, we can be sure to have all of a chunk's usable
358  * space available.
359  */
360  Assert(accessor->write_pointer + accessor->sts->meta_data_size +
361  sizeof(uint32) < accessor->write_end);
362 
363  /* Write the meta-data as one chunk. */
364  if (accessor->sts->meta_data_size > 0)
365  memcpy(accessor->write_pointer, meta_data,
366  accessor->sts->meta_data_size);
367 
368  /*
369  * Write as much of the tuple as we can fit. This includes the
370  * tuple's size at the start.
371  */
372  written = accessor->write_end - accessor->write_pointer -
373  accessor->sts->meta_data_size;
374  memcpy(accessor->write_pointer + accessor->sts->meta_data_size,
375  tuple, written);
376  ++accessor->write_chunk->ntuples;
377  size -= accessor->sts->meta_data_size;
378  size -= written;
379  /* Now write as many overflow chunks as we need for the rest. */
380  while (size > 0)
381  {
382  size_t written_this_chunk;
383 
384  sts_flush_chunk(accessor);
385 
386  /*
387  * How many overflow chunks to go? This will allow readers to
388  * skip all of them at once instead of reading each one.
389  */
390  accessor->write_chunk->overflow = (size + STS_CHUNK_DATA_SIZE - 1) /
392  written_this_chunk =
393  Min(accessor->write_end - accessor->write_pointer, size);
394  memcpy(accessor->write_pointer, (char *) tuple + written,
395  written_this_chunk);
396  accessor->write_pointer += written_this_chunk;
397  size -= written_this_chunk;
398  written += written_this_chunk;
399  }
400  return;
401  }
402  }
403 
404  /* Copy meta-data and tuple into buffer. */
405  if (accessor->sts->meta_data_size > 0)
406  memcpy(accessor->write_pointer, meta_data,
407  accessor->sts->meta_data_size);
408  memcpy(accessor->write_pointer + accessor->sts->meta_data_size, tuple,
409  tuple->t_len);
410  accessor->write_pointer += size;
411  ++accessor->write_chunk->ntuples;
412 }
BufFile * BufFileCreateFileSet(FileSet *fileset, const char *name)
Definition: buffile.c:267
#define Min(x, y)
Definition: c.h:1007
void * MemoryContextAllocZero(MemoryContext context, Size size)
Definition: mcxt.c:1215

References Assert, BufFileCreateFileSet(), SharedTuplestoreAccessor::context, SharedTuplestoreChunk::data, SharedTuplestoreAccessor::fileset, SharedFileSet::fs, MAXPGPATH, MemoryContextAllocZero(), MemoryContextSwitchTo(), SharedTuplestore::meta_data_size, Min, name, SharedTuplestoreChunk::ntuples, SharedTuplestoreChunk::overflow, SharedTuplestoreAccessor::participant, SharedTuplestore::participants, size, SharedTuplestoreAccessor::sts, STS_CHUNK_DATA_SIZE, STS_CHUNK_PAGES, sts_filename(), sts_flush_chunk(), MinimalTupleData::t_len, SharedTuplestoreAccessor::write_chunk, SharedTuplestoreAccessor::write_end, SharedTuplestoreAccessor::write_file, SharedTuplestoreAccessor::write_pointer, and SharedTuplestoreParticipant::writing.

Referenced by ExecParallelHashJoinPartitionOuter(), ExecParallelHashRepartitionFirst(), ExecParallelHashRepartitionRest(), and ExecParallelHashTableInsert().

◆ sts_read_tuple()

static MinimalTuple sts_read_tuple ( SharedTuplestoreAccessor accessor,
void *  meta_data 
)
static

Definition at line 415 of file sharedtuplestore.c.

416 {
417  MinimalTuple tuple;
418  uint32 size;
419  size_t remaining_size;
420  size_t this_chunk_size;
421  char *destination;
422 
423  /*
424  * We'll keep track of bytes read from this chunk so that we can detect an
425  * overflowing tuple and switch to reading overflow pages.
426  */
427  if (accessor->sts->meta_data_size > 0)
428  {
429  BufFileReadExact(accessor->read_file, meta_data, accessor->sts->meta_data_size);
430  accessor->read_bytes += accessor->sts->meta_data_size;
431  }
432  BufFileReadExact(accessor->read_file, &size, sizeof(size));
433  accessor->read_bytes += sizeof(size);
434  if (size > accessor->read_buffer_size)
435  {
436  size_t new_read_buffer_size;
437 
438  if (accessor->read_buffer != NULL)
439  pfree(accessor->read_buffer);
440  new_read_buffer_size = Max(size, accessor->read_buffer_size * 2);
441  accessor->read_buffer =
442  MemoryContextAlloc(accessor->context, new_read_buffer_size);
443  accessor->read_buffer_size = new_read_buffer_size;
444  }
445  remaining_size = size - sizeof(uint32);
446  this_chunk_size = Min(remaining_size,
447  BLCKSZ * STS_CHUNK_PAGES - accessor->read_bytes);
448  destination = accessor->read_buffer + sizeof(uint32);
449  BufFileReadExact(accessor->read_file, destination, this_chunk_size);
450  accessor->read_bytes += this_chunk_size;
451  remaining_size -= this_chunk_size;
452  destination += this_chunk_size;
453  ++accessor->read_ntuples;
454 
455  /* Check if we need to read any overflow chunks. */
456  while (remaining_size > 0)
457  {
458  /* We are now positioned at the start of an overflow chunk. */
459  SharedTuplestoreChunk chunk_header;
460 
461  BufFileReadExact(accessor->read_file, &chunk_header, STS_CHUNK_HEADER_SIZE);
462  accessor->read_bytes = STS_CHUNK_HEADER_SIZE;
463  if (chunk_header.overflow == 0)
464  ereport(ERROR,
466  errmsg("unexpected chunk in shared tuplestore temporary file"),
467  errdetail_internal("Expected overflow chunk.")));
468  accessor->read_next_page += STS_CHUNK_PAGES;
469  this_chunk_size = Min(remaining_size,
470  BLCKSZ * STS_CHUNK_PAGES -
472  BufFileReadExact(accessor->read_file, destination, this_chunk_size);
473  accessor->read_bytes += this_chunk_size;
474  remaining_size -= this_chunk_size;
475  destination += this_chunk_size;
476 
477  /*
478  * These will be used to count regular tuples following the oversized
479  * tuple that spilled into this overflow chunk.
480  */
481  accessor->read_ntuples = 0;
482  accessor->read_ntuples_available = chunk_header.ntuples;
483  }
484 
485  tuple = (MinimalTuple) accessor->read_buffer;
486  tuple->t_len = size;
487 
488  return tuple;
489 }
#define Max(x, y)
Definition: c.h:1001
int errdetail_internal(const char *fmt,...)
Definition: elog.c:1230
MinimalTupleData * MinimalTuple
Definition: htup.h:27
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:1181

References BufFileReadExact(), SharedTuplestoreAccessor::context, ereport, errcode_for_file_access(), errdetail_internal(), errmsg(), ERROR, Max, MemoryContextAlloc(), SharedTuplestore::meta_data_size, Min, SharedTuplestoreChunk::ntuples, SharedTuplestoreChunk::overflow, pfree(), SharedTuplestoreAccessor::read_buffer, SharedTuplestoreAccessor::read_buffer_size, SharedTuplestoreAccessor::read_bytes, SharedTuplestoreAccessor::read_file, SharedTuplestoreAccessor::read_next_page, SharedTuplestoreAccessor::read_ntuples, SharedTuplestoreAccessor::read_ntuples_available, size, SharedTuplestoreAccessor::sts, STS_CHUNK_HEADER_SIZE, STS_CHUNK_PAGES, and MinimalTupleData::t_len.

Referenced by sts_parallel_scan_next().

◆ sts_reinitialize()

void sts_reinitialize ( SharedTuplestoreAccessor accessor)

Definition at line 234 of file sharedtuplestore.c.

235 {
236  int i;
237 
238  /*
239  * Reset the shared read head for all participants' files. Also set the
240  * initial chunk size to the minimum (any increases from that size will be
241  * recorded in chunk_expansion_log).
242  */
243  for (i = 0; i < accessor->sts->nparticipants; ++i)
244  {
245  accessor->sts->participants[i].read_page = 0;
246  }
247 }

References i, SharedTuplestore::nparticipants, SharedTuplestore::participants, SharedTuplestoreParticipant::read_page, and SharedTuplestoreAccessor::sts.