-
Notifications
You must be signed in to change notification settings - Fork 66
/
DataReader.cpp
2954 lines (2541 loc) · 94.6 KB
/
DataReader.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*++
Module Name:
DataReader.cpp
Abstract:
Concrete implementation classes for DataReader and DataSupplier.
These are completely opaque, and are only exposed through static supplier objects
defined in DataReader.h
Environment:
User mode service.
--*/
#include "stdafx.h"
#include "BigAlloc.h"
#include "Compat.h"
#include "RangeSplitter.h"
#include "ParallelTask.h"
#include "DataReader.h"
#include "Bam.h"
#include "zlib.h"
#include "exit.h"
#include "Error.h"
#include "Util.h"
using std::max;
using std::min;
using std::map;
using std::string;
//#define VALIDATE_STDIO
//
// Read-Based
//
//
// A data reader that uses a read-type call to get its data (as opposed to memory mapping).
// This class contains the generic implementation, it must be subclassed to implement
// startIo() and waitForBuffer(), which do the actual IO.
//
class ReadBasedDataReader : public DataReader
{
public:
ReadBasedDataReader(unsigned i_nBuffers, _int64 i_overflowBytes, double extraFactor, size_t bufferSpace = 0);
virtual ~ReadBasedDataReader();
virtual bool init(const char* fileName) = 0;
char* readHeader(_int64* io_headerSize);
virtual void reinit(_int64 startingOffset, _int64 amountOfFileToProcess);
virtual bool getData(char** o_buffer, _int64* o_validBytes, _int64* o_startBytes = NULL);
virtual void advance(_int64 bytes);
virtual void nextBatch();
virtual bool isEOF();
virtual DataBatch getBatch();
virtual void holdBatch(DataBatch batch);
virtual bool releaseBatch(DataBatch batch);
virtual _int64 getFileOffset();
virtual void getExtra(char** o_extra, _int64* o_length);
virtual const char* getFilename() = 0;
protected:
// must hold the lock to call
virtual void startIo() = 0;
// must hold the lock to call
virtual void waitForBuffer(unsigned bufferNumber) = 0;
// must hold the lock to call
virtual void addBuffer();
static const unsigned BUFFER_SIZE = 4 * 1024 * 1024 - 4096;
enum BufferState {Empty, Reading, Full, InUse};
struct BufferInfo
{
char *buffer;
BufferState state;
size_t validBytes;
unsigned nBytesThatMayBeginARead;
bool isEOF;
unsigned offset; // How far has the consumer gotten in current buffer
_int64 fileOffset;
_uint32 batchID;
int holds;
char* extra;
int next, previous; // index of next/previous in free/ready list, -1 if end
bool headerBuffer; // Set if this is a special buffer that holds the rewound header. These get read once and deallocated.g
#ifdef VALIDATE_STDIO
_uint32 dataHash;
#endif
void operator=(BufferInfo &peer) {
buffer = peer.buffer;
state = peer.state;
validBytes = peer.validBytes;
nBytesThatMayBeginARead = peer.nBytesThatMayBeginARead;
isEOF = peer.isEOF;
offset = peer.offset;
fileOffset = peer.fileOffset;
batchID = peer.batchID;
holds = peer.holds;
extra = peer.extra;
next = peer.next;
previous = peer.previous;
headerBuffer = peer.headerBuffer;
}
};
unsigned nBuffers;
const unsigned maxBuffers;
int headerBuffersOutstanding;
bool startedReadingHeader;
_int64 extraBytes;
_int64 overflowBytes;
BufferInfo* bufferInfo;
_uint32 nextBatchID;
int nextBufferForReader; // list head (singly linked), -1 if empty
int nextBufferForConsumer; // list head (doubly linked), -1 if empty
int lastBufferForConsumer; // list tail, -1 if empty
EventObject releaseEvent;
_int64 releaseWaitInMillis;
ExclusiveLock lock;
private:
virtual bool getDataInternal(char** o_buffer, _int64* o_validBytes, _int64* o_startBytes = NULL);
//
// Stuff for handling the header read. We allow arbitrarily large header reads, and service them by copying data from the underlying
// data reader into a local buffer. We might wind up reading more than the actual header, so we serve reads out of the header buffer
// until it's used up.
//
char *headerBuffer;
_int64 headerBufferSize;
char *headerExtra; // Allocated in one go with the headerBuffer
_int64 headerExtraSize;
_int64 amountAdvancedThroughUnderlyingStoreByUs;
unsigned nHeaderBuffersAllocated;
bool hitEOFReadingHeader;
protected:
const size_t bufferSize;
};
ReadBasedDataReader::ReadBasedDataReader(
unsigned i_nBuffers,
_int64 i_overflowBytes,
double extraFactor,
size_t i_bufferSpace)
: DataReader(), nBuffers(i_nBuffers), overflowBytes(i_overflowBytes),
maxBuffers(i_nBuffers * (i_nBuffers == 1 ? 2 : 4)),
bufferSize(i_bufferSpace > 0 ? i_bufferSpace / ((_int64)i_nBuffers * 2) : BUFFER_SIZE),
headerBuffer(NULL), headerBufferSize(0), amountAdvancedThroughUnderlyingStoreByUs(0),
headerExtra(NULL), headerExtraSize(0), startedReadingHeader(false), headerBuffersOutstanding(0), nHeaderBuffersAllocated(0),
hitEOFReadingHeader(false)
{
//
// Initialize the buffer info struct.
//
// allocate all the data in one big block
// NOTE: buffers are not null-terminated (since memmap version can't do it)
_ASSERT(extraFactor >= 0 && i_nBuffers > 0);
bufferInfo = new BufferInfo[maxBuffers];
extraBytes = max((_int64) 0, (_int64) ((bufferSize + overflowBytes) * extraFactor));
char* allocated = (char*) BigReserve(maxBuffers * (bufferSize + extraBytes + overflowBytes));
BigCommit(allocated, nBuffers * (bufferSize + extraBytes + overflowBytes));
if (NULL == allocated) {
WriteErrorMessage("ReadBasedDataReader: unable to allocate IO buffer\n");
soft_exit(1);
}
for (unsigned i = 0 ; i < nBuffers; i++) {
bufferInfo[i].buffer = allocated;
allocated += bufferSize + overflowBytes;
bufferInfo[i].extra = extraBytes > 0 ? allocated : NULL;
allocated += extraBytes;
bufferInfo[i].state = Empty;
bufferInfo[i].isEOF = false;
bufferInfo[i].offset = 0;
bufferInfo[i].next = i < nBuffers - 1 ? i + 1 : -1;
bufferInfo[i].previous = i > 0 ? i - 1 : -1;
bufferInfo[i].holds = 0;
bufferInfo[i].headerBuffer = false;
}
nextBatchID = 1;
nextBufferForConsumer = -1;
lastBufferForConsumer = -1;
//fprintf(stderr, "DataReader.cpp:%d nextBufferForReader %d -> %d\n", __LINE__, nextBufferForReader, 0);
nextBufferForReader = 0;
CreateEventObject(&releaseEvent);
releaseWaitInMillis = 5; // wait up to 5 ms before allocating a new buffer
InitializeExclusiveLock(&lock);
}
ReadBasedDataReader::~ReadBasedDataReader()
{
BigDealloc(bufferInfo[0].buffer);
for (unsigned i = 0; i < nBuffers; i++) {
bufferInfo[i].buffer = bufferInfo[i].extra = NULL;
}
if (NULL != headerBuffer) {
delete[] headerBuffer;
headerBuffer = NULL;
}
if (NULL != headerExtra) {
delete[] headerExtra;
headerExtra = NULL;
}
DestroyExclusiveLock(&lock);
DestroyEventObject(&releaseEvent);
}
char *
ReadBasedDataReader::readHeader(_int64* io_headerSize)
{
_ASSERT(!startedReadingHeader);
_int64 validBytesInHeader;
if (NULL != headerBuffer) {
if (*io_headerSize <= headerBufferSize) {
return headerBuffer;
}
//
// We need more data for the header. Reallocate the buffer, copy the data from the old buffer into the new, and then get more
// data from the underlying reader.
//
char *newHeaderBuffer = new char[*io_headerSize];
memcpy(newHeaderBuffer, headerBuffer, headerBufferSize);
delete[] headerBuffer;
headerBuffer = newHeaderBuffer;
validBytesInHeader = headerBufferSize;
headerBufferSize = *io_headerSize;
} else {
reinit(0, 0);
headerBufferSize = *io_headerSize;
headerBuffer = new char[headerBufferSize];
validBytesInHeader = 0;
}
//
// Run through the underlying data provider getting data until we've filled the header buffer or hit EOF.
//
_int64 bytesLeftToGet = headerBufferSize - validBytesInHeader;
_ASSERT(bytesLeftToGet);
while (bytesLeftToGet != 0) {
if (amountAdvancedThroughUnderlyingStoreByUs < validBytesInHeader) {
_int64 amountToAdvance = validBytesInHeader - amountAdvancedThroughUnderlyingStoreByUs - overflowBytes; // Leave overflowBytes left over, since we
if (amountToAdvance <= 0) {
//
// We're probably almost at EOF. Consume the overflow bytes, too.
//
amountToAdvance = validBytesInHeader - amountAdvancedThroughUnderlyingStoreByUs;
}
advance(amountToAdvance);
amountAdvancedThroughUnderlyingStoreByUs += amountToAdvance;
}
char *dataFromUnderlyingStore;
_int64 dataSizeFromUnderlyingStore;
if (!getDataInternal(&dataFromUnderlyingStore, &dataSizeFromUnderlyingStore)) {
nextBatch();
if (!getDataInternal(&dataFromUnderlyingStore, &dataSizeFromUnderlyingStore)) {
//
// Hit EOF while reading header.
//
hitEOFReadingHeader = true;
headerBufferSize = *io_headerSize = validBytesInHeader;
return headerBuffer;
}
}
//
// Adjust for the fact that we don't advance as far as we've read, so that we leave some overlap for
// subsequent readers who want the data, not the header.
//
_ASSERT(amountAdvancedThroughUnderlyingStoreByUs <= validBytesInHeader); // We haven't advanced over something we need.
_int64 offsetIntoBuffer = validBytesInHeader - amountAdvancedThroughUnderlyingStoreByUs;
_ASSERT(dataSizeFromUnderlyingStore >= offsetIntoBuffer);
dataSizeFromUnderlyingStore -= offsetIntoBuffer;
_int64 bytesToCopy = __min(dataSizeFromUnderlyingStore, bytesLeftToGet);
memcpy(headerBuffer + validBytesInHeader, dataFromUnderlyingStore + offsetIntoBuffer, bytesToCopy);
bytesLeftToGet -= bytesToCopy;
validBytesInHeader += bytesToCopy;
}
return headerBuffer; // No need to reset *io_headerSize, we read as much as was requested
}
//
// This gets called only for subclasses that can't implement their own. It's able to put the header reads
// back on the queue, but can't seek anywhere else.
//
void
ReadBasedDataReader::reinit(
_int64 startingOffset,
_int64 amountOfFileToProcess)
{
AcquireExclusiveLock(&lock);
if (0 != amountOfFileToProcess) {
WriteErrorMessage("ReadBasedDataReader:reinit called with non-zero amountOfFileToProcess (%lld, %lld)\n", startingOffset, amountOfFileToProcess);
soft_exit(1);
}
if (startedReadingHeader || 0 != headerBuffersOutstanding) {
WriteErrorMessage("ReadBasedDataReader:reinit called after reading some data (%lld, %lld)\n", startingOffset, amountOfFileToProcess);
soft_exit(1);
}
//
// We've already read a bunch of data from the underlying reader during header read. Create some new virtual buffers that point into the
// header buffer, and stick them at the head of the "already read" list.
//
_ASSERT(!startedReadingHeader && 0 == headerBuffersOutstanding);
if (0 != headerBufferSize) {
startedReadingHeader = true;
}
//
// First let any pending IO complete.
//
for (unsigned i = 0; i < nBuffers; i++) {
if (bufferInfo[i].state == Reading) {
waitForBuffer(i);
}
}
_ASSERT(amountAdvancedThroughUnderlyingStoreByUs <= headerBufferSize);
if (amountAdvancedThroughUnderlyingStoreByUs == 0) {
nHeaderBuffersAllocated = 0;
} else {
nHeaderBuffersAllocated = (int)((amountAdvancedThroughUnderlyingStoreByUs + bufferSize - 1) / (bufferSize - overflowBytes)); // Round up, in case we read the last buffer.
}
int totalBuffersNeeded = (int)(maxBuffers + nHeaderBuffersAllocated);
//
// Reallocate the buffers array.
//
BufferInfo *newBuffers = new BufferInfo[totalBuffersNeeded];
for (unsigned i = 0; i < maxBuffers; i++) {
newBuffers[i] = bufferInfo[i];
}
delete[] bufferInfo;
bufferInfo = newBuffers;
//
// Don't increase maxBuffers, so the buffer adder won't use the headerBuffers for anything.
//
//
// Now construct the header buffers.
//
headerExtraSize = extraBytes * nHeaderBuffersAllocated;
headerExtra = new char[headerExtraSize];
char *headerPointer = headerBuffer;
char *headerExtraPointer = headerExtra;
_int64 fileOffset = 0;
_int64 bytesRemaining = amountAdvancedThroughUnderlyingStoreByUs;
for (int i = maxBuffers; i < totalBuffersNeeded; i++) {
bufferInfo[i].state = Full;
bufferInfo[i].isEOF = false;
bufferInfo[i].offset = 0;
bufferInfo[i].next = (i == totalBuffersNeeded - 1) ? nextBufferForConsumer : i + 1;
bufferInfo[i].previous = (i == maxBuffers) ? -1 : i - 1;
bufferInfo[i].holds = 0;
bufferInfo[i].headerBuffer = true;
bufferInfo[i].validBytes = (int)__min(bytesRemaining, (_int64) bufferSize);
bufferInfo[i].nBytesThatMayBeginARead = (int)((bytesRemaining <= (_int64) bufferSize) ? bytesRemaining : __max(bufferInfo[i].validBytes - overflowBytes, 0));
bufferInfo[i].offset = 0;
bufferInfo[i].fileOffset = fileOffset;
bufferInfo[i].batchID = nextBatchID++;
bufferInfo[i].buffer = headerPointer;
headerPointer += bufferInfo[i].nBytesThatMayBeginARead; // NB: don't add overflowBytes; these buffers overlap
fileOffset += bufferInfo[i].nBytesThatMayBeginARead;
bufferInfo[i].extra = headerExtraPointer;
headerExtraPointer += extraBytes;
headerBuffersOutstanding++;
bytesRemaining -= bufferInfo[i].nBytesThatMayBeginARead;
}
if (nHeaderBuffersAllocated > 0) {
_ASSERT(bufferInfo[nextBufferForConsumer].previous == -1);
bufferInfo[nextBufferForConsumer].previous = totalBuffersNeeded - 1;
nextBufferForConsumer = maxBuffers;
if (hitEOFReadingHeader) {
bufferInfo[totalBuffersNeeded - 1].isEOF = true;
}
}
//
// Kick off IO, wait for the first buffer to be read
//
startIo();
waitForBuffer(nextBufferForConsumer);
ReleaseExclusiveLock(&lock);
//
// Now, consume data until we've gotten to startingOffset.
//
_int64 bytesToSkip = startingOffset;
while (bytesToSkip > 0) {
char *p;
_int64 valid, start;
bool ok = getData(&p, &valid, &start);
if (!ok) {
WriteErrorMessage("ReadBasedDataReader::init() failure getting data\n");
soft_exit(1);
}
_int64 bytesToSkipThisTime = __min(valid, bytesToSkip);
advance(bytesToSkipThisTime);
if (bytesToSkipThisTime > start) {
nextBatch();
}
getData(&p, &valid, &start);
bytesToSkip -= bytesToSkipThisTime;
}
}
bool
ReadBasedDataReader::getData(
char** o_buffer,
_int64* o_validBytes,
_int64* o_startBytes)
{
if (NULL != headerBuffer && !startedReadingHeader) {
delete[] headerBuffer;
headerBuffer = NULL;
_ASSERT(NULL == headerExtra);
}
return getDataInternal(o_buffer, o_validBytes, o_startBytes);
}
bool
ReadBasedDataReader::getDataInternal(
char** o_buffer,
_int64* o_validBytes,
_int64* o_startBytes)
{
_ASSERT(nextBufferForConsumer >= 0);
BufferInfo *info = &bufferInfo[nextBufferForConsumer];
if (info->isEOF && info->offset >= info->validBytes) {
//
// EOF.
//
return false;
}
if (info->offset >= info->nBytesThatMayBeginARead) {
//
// Past the end of our section.
//
return false;
}
if (info->state != Full) {
_ASSERT(info->state != InUse);
AcquireExclusiveLock(&lock);
waitForBuffer(nextBufferForConsumer);
ReleaseExclusiveLock(&lock);
}
*o_buffer = info->buffer + info->offset;
*o_validBytes = info->validBytes - info->offset;
if (o_startBytes != NULL) {
*o_startBytes = info->nBytesThatMayBeginARead - info->offset;
}
return true;
}
void
ReadBasedDataReader::advance(
_int64 bytes)
{
BufferInfo* info = &bufferInfo[nextBufferForConsumer];
_ASSERT(info->validBytes >= info->offset && bytes >= 0 && bytes <= info->validBytes - info->offset);
info->offset += __min(info->validBytes - info->offset, (unsigned)__max((_int64)0, bytes));
}
void
ReadBasedDataReader::nextBatch()
{
AcquireExclusiveLock(&lock);
_ASSERT(nextBufferForConsumer >= 0);
BufferInfo* info = &bufferInfo[nextBufferForConsumer];
if (info->isEOF) {
ReleaseExclusiveLock(&lock);
if (info->holds == 0) {
releaseBatch(DataBatch(info->batchID));
}
return;
}
DataBatch priorBatch = DataBatch(info->batchID);
info->state = InUse;
_uint32 overflow = max((unsigned) info->offset, info->nBytesThatMayBeginARead) - info->nBytesThatMayBeginARead;
_int64 nextStart = info->fileOffset + info->nBytesThatMayBeginARead; // for validation
//fprintf(stderr, "ReadBasedDataReader:nextBatch() finished buffer %d at %llu, starting buffer %d at %llu\n", nextBufferForConsumer, info->fileOffset, info->next, bufferInfo[info->next].fileOffset);
//fprintf(stderr, "ReadBasedDataReader:nextBatch() skipping %u overflow bytes used in previous batch\n", overflow);
_ASSERT(info->next == -1 || bufferInfo[info->next].fileOffset == nextStart);
//fprintf(stderr, "DataReader.cpp:%d nextBufferForConsumer %d -> %d\n", __LINE__, nextBufferForConsumer, info->next);
nextBufferForConsumer = info->next;
bool first = true;
while (nextBufferForConsumer == -1) {
nextStart = 0; // can no longer count on getting sequential buffers from file
ReleaseExclusiveLock(&lock);
if (! first) {
//fprintf(stderr, "ReadBasedDataReader::nextBatch thread %d wait for release\n", GetCurrentThreadId());
_int64 start = timeInNanos();
bool waitSucceeded = WaitForEventWithTimeout(&releaseEvent, releaseWaitInMillis);
InterlockedAdd64AndReturnNewValue(&ReleaseWaitTime, timeInNanos() - start);
//fprintf(stderr, "ReadBasedDataReader::nextBatch thread %d released\n", GetCurrentThreadId());
if (!waitSucceeded) {
AcquireExclusiveLock(&lock);
addBuffer();
ReleaseExclusiveLock(&lock);
}
}
first = false;
AcquireExclusiveLock(&lock);
startIo();
}
if (bufferInfo[nextBufferForConsumer].state != Full) {
waitForBuffer(nextBufferForConsumer);
}
bufferInfo[nextBufferForConsumer].offset = overflow;
bufferInfo[nextBufferForConsumer].holds = 0;
//fprintf(stderr,"emitting buffer starting at 0x%llx\n", info->fileOffset);
//if (nextStart != 0) fprintf(stderr, "checking NextStart 0x%llx\n", nextStart);
_ASSERT(nextStart == 0 || nextStart == bufferInfo[nextBufferForConsumer].fileOffset || bufferInfo[nextBufferForConsumer].isEOF);
ReleaseExclusiveLock(&lock);
if (info->holds == 0) {
releaseBatch(priorBatch);
}
}
bool
ReadBasedDataReader::isEOF()
{
BufferInfo* info = &bufferInfo[nextBufferForConsumer];
return info->isEOF && info->offset >= info->validBytes;
}
DataBatch
ReadBasedDataReader::getBatch()
{
return DataBatch(bufferInfo[nextBufferForConsumer].batchID);
}
void
ReadBasedDataReader::holdBatch(
DataBatch batch)
{
AcquireExclusiveLock(&lock);
for (unsigned i = 0; i < maxBuffers + nHeaderBuffersAllocated; i = (i == nBuffers - 1) ? maxBuffers : i+1) { // Goofy loop is because headerBuffers get tacked on beyond maxBuffers
BufferInfo *info = &bufferInfo[i];
if (info->batchID == batch.batchID) {
//fprintf(stderr, "%x holdBatch batch 0x%x, holds on buffer %d now %d\n", (unsigned) this, batch.batchID, i, info->holds);
info->holds++;
}
}
ReleaseExclusiveLock(&lock);
}
bool
ReadBasedDataReader::releaseBatch(
DataBatch batch)
{
AcquireExclusiveLock(&lock);
bool released = false;
bool result = true;
for (unsigned i = 0; i < maxBuffers + nHeaderBuffersAllocated; i = (i == nBuffers - 1) ? maxBuffers : i + 1) { // Goofy loop is because headerBuffers get tacked on beyond maxBuffers
BufferInfo* info = &bufferInfo[i];
if (info->batchID == batch.batchID) {
switch (info->state) {
case Empty:
// should never happen
break;
case Reading:
// should never happen
_ASSERT(false);
break;
case InUse:
released = info->holds <= 1;
// fall through
case Full:
if (info->holds > 0) {
info->holds--;
}
if (info->holds == 0 && (i != nextBufferForConsumer || info->isEOF)) {
//fprintf(stderr,"ReadBasedDataReader:releaseBatch batch %d, releasing %s buffer %d\n", batch.batchID, info->state == InUse ? "InUse" : "Full", i);
info->state = Empty;
// remove from ready list
if (i == lastBufferForConsumer) {
lastBufferForConsumer = info->previous;
}
if (info->next != -1) {
bufferInfo[info->next].previous = info->previous;
}
if (info->previous != -1) {
bufferInfo[info->previous].next = info->next;
}
if (info->headerBuffer) {
// Header buffers never get reused. Just get rid of it.
info->buffer = NULL;
info->extra = NULL;
_ASSERT(headerBuffersOutstanding > 0);
if (headerBuffersOutstanding > 0) {
headerBuffersOutstanding--;
if (0 == headerBuffersOutstanding) {
delete[] headerBuffer;
delete[] headerExtra;
headerBuffer = headerExtra = NULL;
nHeaderBuffersAllocated = 0;
}
}
} else {
// add to head of free list
info->next = nextBufferForReader;
info->batchID = 0;
#ifdef _DEBUG
//memset(info->buffer, 0xde, bufferSize + extraBytes);
#endif
//fprintf(stderr, "DataReader.cpp:%d nextBufferForReader %d -> %d\n", __LINE__, nextBufferForReader, i);
nextBufferForReader = i;
}
result = true;
} else {
//fprintf(stderr,"releaseBatch batch %d, holds on buffer %d now %d\n", batch.batchID, i, info->holds);
result = false;
}
break;
default:
WriteErrorMessage("ReadBasedDataReader::releaseBatch():invalid enum\n");
soft_exit(1);
}
}
}
startIo();
if (released) {
//fprintf(stderr, "releaseBatch set releaseEvent\n");
AllowEventWaitersToProceed(&releaseEvent);
}
ReleaseExclusiveLock(&lock);
return result;
}
_int64
ReadBasedDataReader::getFileOffset()
{
return bufferInfo[nextBufferForConsumer].fileOffset + bufferInfo[nextBufferForConsumer].offset;
}
void
ReadBasedDataReader::getExtra(
char** o_extra,
_int64* o_length)
{
// hack: return valid buffer even when no consumer buffers - this may happen when reading header
*o_extra = bufferInfo[max(0, nextBufferForConsumer)].extra;
*o_length = extraBytes;
}
void
ReadBasedDataReader::addBuffer()
{
if (nBuffers == maxBuffers) {
//fprintf(stderr, "ReadBasedDataReader: addBuffer at limit\n");
return;
}
_ASSERT(nBuffers < maxBuffers);
//fprintf(stderr, "ReadBasedDataReader: addBuffer %d of %d\n", nBuffers, maxBuffers);
size_t bytes = bufferSize + extraBytes + overflowBytes;
bufferInfo[nBuffers].buffer = bufferInfo[nBuffers-1].buffer + bytes;
if (! BigCommit(bufferInfo[nBuffers].buffer, bytes)) {
WriteErrorMessage("ReadBasedDataReader: unable to commit IO buffer\n");
soft_exit(1);
}
bufferInfo[nBuffers].extra = extraBytes > 0 ? bufferInfo[nBuffers].buffer + bytes - extraBytes : NULL;
bufferInfo[nBuffers].state = Empty;
bufferInfo[nBuffers].isEOF= false;
bufferInfo[nBuffers].offset = 0;
bufferInfo[nBuffers].next = nextBufferForReader;
bufferInfo[nBuffers].previous = -1;
bufferInfo[nBuffers].headerBuffer = false;
//fprintf(stderr, "DataReader.cpp:%d nextBufferForReader %d -> %d\n", __LINE__, nextBufferForReader, nBuffers);
nextBufferForReader = nBuffers;
nBuffers++;
_ASSERT(nBuffers <= maxBuffers);
if (nBuffers == maxBuffers) {
releaseWaitInMillis = 1000 * 3600 * 24 * 7; // A week
}
}
class StdioDataReader : public ReadBasedDataReader
{
public:
StdioDataReader(unsigned i_nBuffers, _int64 i_overflowBytes, double extraFactor);
~StdioDataReader();
virtual bool init(const char* i_fileName);
virtual const char* getFilename()
{ return "-"; }
protected:
// must hold the lock to call
virtual void startIo();
// must hold the lock to call
virtual void waitForBuffer(unsigned bufferNumber);
private:
//
// Because reads don't necessarily divide evenly into buffers, we have to assure that
// the buffers that we read can overlap. In file-IO based readers, we do this by reading
// a buffer's worth of data each time, but advancing the file pointer only by
// bufferSize - overflowBytes, so each buffer ovelaps with its predecessor by a little.
// That doesn't work for stdio, since it can't rewind. So, instead, we allocate
// storage on the side to hold a copy of the last overflowBytes
// and then just copy those bytes into the beginning of the next buffer to read.
// We also use this buffer to hold the header (the first read), and to allow
// reading the header plus some extra data, parsing the header, and then seeking
// backward to the actual end of the header.
//
char *overflowBuffer;
bool overflowBufferFilled; // For the very first read, there may be no overlap buffer data.
bool started;
bool hitEOF;
_int64 readOffset;
};
StdioDataReader::StdioDataReader(unsigned i_nBuffers, _int64 i_overflowBytes, double extraFactor) :
ReadBasedDataReader(i_nBuffers, i_overflowBytes, extraFactor), started(false), hitEOF(false), overflowBufferFilled(false),
readOffset(0), overflowBuffer(NULL)
{
}
StdioDataReader::~StdioDataReader()
{
BigDealloc(overflowBuffer);
overflowBuffer = NULL;
}
bool
StdioDataReader::init(const char * i_fileName)
{
if (strcmp(i_fileName, "-")) {
WriteErrorMessage("StdioDataReader: must have filename of '-', got '%s'\n", i_fileName);
soft_exit(1);
}
#ifdef _MSC_VER
int result = _setmode( _fileno( stdin ), _O_BINARY ); // puts stdin in to non-translated mode, so if we're reading compressed data windows' CRLF processing doesn't destroy it.
if (-1 == result) {
WriteErrorMessage("StdioDataReader::freopen to change to untranslated mode failed\n");
soft_exit(1);
}
#endif // _MSC_VER
return true;
}
void
StdioDataReader::startIo()
{
AssertExclusiveLockHeld(&lock);
started = true;
//
// Synchronously read data into whatever buffers are ready.
//
while (nextBufferForReader != -1) {
// remove from free list
BufferInfo* info = &bufferInfo[nextBufferForReader];
_ASSERT(info->state == Empty);
int index = nextBufferForReader;
//fprintf(stderr, "DataReader.cpp:%d nextBufferForReader %d -> %d\n", __LINE__, nextBufferForReader, info->next);
nextBufferForReader = info->next;
info->batchID = nextBatchID++;
// add to end of consumer list
if (lastBufferForConsumer != -1) {
_ASSERT(bufferInfo[lastBufferForConsumer].next == -1);
bufferInfo[lastBufferForConsumer].next = index;
}
info->next = -1;
info->previous = lastBufferForConsumer;
lastBufferForConsumer = index;
if (nextBufferForConsumer == -1) {
//fprintf(stderr, "StdioDataReader::startIo set nextBufferForConsumder -1 -> %d\n", index);
nextBufferForConsumer = index;
}
if (hitEOF) {
info->validBytes = 0;
info->buffer[0] = '\0';
info->nBytesThatMayBeginARead = 0;
info->isEOF = true;
info->state = Full;
return;
}
size_t amountToRead;
size_t bufferOffset;
if (overflowBufferFilled) {
//
// Copy the bytes from the overflow buffer into our buffer.
//
memcpy(info->buffer, overflowBuffer, overflowBytes);
bufferOffset = overflowBytes;
amountToRead = bufferSize - overflowBytes;
info->fileOffset = readOffset - overflowBytes;
} else {
amountToRead = bufferSize;
bufferOffset = 0;
info->fileOffset = readOffset;
}
//
// We have to run this holding the lock, because otherwise there's no way to make the overflow buffer work properly.
//
size_t bytesRead = fread(info->buffer + bufferOffset, 1, amountToRead, stdin);
//fprintf(stderr,"StdioDataReader:startIO(): Read offset 0x%llx into buffer %d at 0x%llx, size %d, copied 0x%x overflow bytes, start at 0x%llx, tid %d\n", readOffset, info-bufferInfo, info->buffer, bytesRead, bufferOffset, readOffset - bufferOffset, GetThreadId());
readOffset += bytesRead;
if (bytesRead != amountToRead) {
if (feof(stdin)) {
info->isEOF = true;
hitEOF = true;
} else {
WriteErrorMessage("StdinDataReader: Error reading stdin (but not EOF).\n");
soft_exit(1);
}
} else {
info->isEOF = false;
}
info->validBytes = (unsigned)(bytesRead + bufferOffset);
#ifdef VALIDATE_STDIO
info->dataHash = util::hash(info->buffer, info->validBytes);
#endif
if (hitEOF) {
info->nBytesThatMayBeginARead = (unsigned)(bytesRead + bufferOffset);
overflowBufferFilled = false;
} else {
info->nBytesThatMayBeginARead = (unsigned)(bytesRead + bufferOffset - overflowBytes);
//
// Fill the overflow buffer with the last bytes from this buffer.
//
if (NULL == overflowBuffer) {
//
// We can get here if we never called readHeader(). If so, we know we never will and so
// we can just allocate the overflow buffer to be the size of the header.
//
overflowBuffer = (char *)BigAlloc(overflowBytes);
}
memcpy(overflowBuffer, info->buffer + bufferOffset + bytesRead - overflowBytes, overflowBytes);
overflowBufferFilled = true;
}
info->state = Full;
}
if (nextBufferForConsumer == -1) {
//fprintf(stderr, "startIo thread %x reset releaseEvent\n", GetThreadId());
PreventEventWaitersFromProceeding(&releaseEvent);
}
}
void
StdioDataReader::waitForBuffer(
unsigned bufferNumber)
{
_ASSERT(bufferNumber >= 0 && (bufferNumber < nBuffers || bufferNumber >= maxBuffers && 0 != headerBuffersOutstanding));
BufferInfo *info = &bufferInfo[bufferNumber];
while (info->state == InUse) {
//fprintf(stderr, "StdioDataReader::waitForBuffer %d InUse...\n", bufferNumber);
// must already have lock to call, release & wait & reacquire
ReleaseExclusiveLock(&lock);
// TODO: implement timed wait on Linux
#ifdef _MSC_VER
_int64 start = timeInNanos();
_uint32 waitTime;
if (releaseWaitInMillis > 0xffffffff) {
waitTime = INFINITE;
} else {
waitTime = (_uint32)releaseWaitInMillis;
}
_uint32 result = WaitForSingleObject(releaseEvent, waitTime);
InterlockedAdd64AndReturnNewValue(&ReleaseWaitTime, timeInNanos() - start);
#else
WaitForEvent(&releaseEvent);
#endif
AcquireExclusiveLock(&lock);
#ifdef _MSC_VER
if (result == WAIT_TIMEOUT) {
// this isn't going to directly make this buffer available, but will reduce pressure
addBuffer();
}
#endif
}
if (info->state == Full) {
#ifdef VALIDATE_STDIO
if (info->dataHash != util::hash(info->buffer, info->validBytes)) {
WriteErrorMessage("Buffer contents modified\n");
soft_exit(1);
}
#endif
return;
}
_ASSERT(info->state != Reading); // We're synchronous, we don't use Reading
startIo();
#ifdef VALIDATE_STDIO
if (info->dataHash != util::hash(info->buffer, info->validBytes)) {
WriteErrorMessage("Buffer contents modified\n");
soft_exit(1);
}
#endif
info->state = Full;
info->buffer[info->validBytes] = 0;
}
class StdioDataSupplier : public DataSupplier