ha_archive.cc 31.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/* Copyright (C) 2003 MySQL AB

  This program is free software; you can redistribute it and/or modify
  it under the terms of the GNU General Public License as published by
  the Free Software Foundation; either version 2 of the License, or
  (at your option) any later version.

  This program is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  GNU General Public License for more details.

  You should have received a copy of the GNU General Public License
  along with this program; if not, write to the Free Software
  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */

kent@mysql.com's avatar
kent@mysql.com committed
17
#ifdef USE_PRAGMA_IMPLEMENTATION
18 19 20
#pragma implementation        // gcc: Class implementation
#endif

21
#include "mysql_priv.h"
22

23
#if defined(HAVE_ARCHIVE_DB)
24
#include "ha_archive.h"
25
#include <my_dir.h>
26 27 28 29 30 31 32

/*
  First, if you want to understand storage engines you should look at 
  ha_example.cc and ha_example.h. 
  This example was written as a test case for a customer who needed
  a storage engine without indexes that could compress data very well.
  So, welcome to a completely compressed storage engine. This storage
33
  engine only does inserts. No replace, deletes, or updates. All reads are 
34 35 36 37 38 39 40 41 42 43 44 45
  complete table scans. Compression is done through gzip (bzip compresses
  better, but only marginally, if someone asks I could add support for
  it too, but beaware that it costs a lot more in CPU time then gzip).
  
  We keep a file pointer open for each instance of ha_archive for each read
  but for writes we keep one open file handle just for that. We flush it
  only if we have a read occur. gzip handles compressing lots of records
  at once much better then doing lots of little records between writes.
  It is possible to not lock on writes but this would then mean we couldn't
  handle bulk inserts as well (that is if someone was trying to read at
  the same time since we would want to flush).

46 47 48 49 50 51 52 53 54 55
  A "meta" file is kept alongside the data file. This file serves two purpose.
  The first purpose is to track the number of rows in the table. The second 
  purpose is to determine if the table was closed properly or not. When the 
  meta file is first opened it is marked as dirty. It is opened when the table 
  itself is opened for writing. When the table is closed the new count for rows 
  is written to the meta file and the file is marked as clean. If the meta file 
  is opened and it is marked as dirty, it is assumed that a crash occured. At 
  this point an error occurs and the user is told to rebuild the file.
  A rebuild scans the rows and rewrites the meta file. If corruption is found
  in the data file then the meta file is not repaired.
56

57
  At some point a recovery method for such a drastic case needs to be divised.
58

59
  Locks are row level, and you will get a consistant read. 
60 61 62 63 64 65 66 67 68 69

  For performance as far as table scans go it is quite fast. I don't have
  good numbers but locally it has out performed both Innodb and MyISAM. For
  Innodb the question will be if the table can be fit into the buffer
  pool. For MyISAM its a question of how much the file system caches the
  MyISAM file. With enough free memory MyISAM is faster. Its only when the OS
  doesn't have enough memory to cache entire table that archive turns out 
  to be any faster. For writes it is always a bit slower then MyISAM. It has no
  internal limits though for row length.

70
  Examples between MyISAM (packed) and Archive.
71 72 73 74 75 76 77 78 79 80 81

  Table with 76695844 identical rows:
  29680807 a_archive.ARZ
  920350317 a.MYD


  Table with 8991478 rows (all of Slashdot's comments):
  1922964506 comment_archive.ARZ
  2944970297 comment_text.MYD


82 83 84 85 86 87 88 89
  TODO:
   Add bzip optional support.
   Allow users to set compression level.
   Add truncate table command.
   Implement versioning, should be easy.
   Allow for errors, find a way to mark bad rows.
   Talk to the gzip guys, come up with a writable format so that updates are doable
     without switching to a block method.
90
   Add optional feature so that rows can be flushed at interval (which will cause less
91 92 93 94 95
     compression but may speed up ordered searches).
   Checkpoint the meta file to allow for faster rebuilds.
   Dirty open (right now the meta file is repaired if a crash occured).
   Option to allow for dirty reads, this would lower the sync calls, which would make
     inserts a lot faster, but would mean highly arbitrary reads.
96 97 98

    -Brian
*/
99 100 101 102 103 104 105 106 107
/*
  Notes on file formats.
  The Meta file is layed out as:
  check - Just an int of 254 to make sure that the the file we are opening was
          never corrupted.
  version - The current version of the file format.
  rows - This is an unsigned long long which is the number of rows in the data
         file.
  check point - Reserved for future use
108 109
  dirty - Status of the file, whether or not its values are the latest. This
          flag is what causes a repair to occur
110 111 112 113 114 115

  The data file:
  check - Just an int of 254 to make sure that the the file we are opening was
          never corrupted.
  version - The current version of the file format.
  data - The data is stored in a "row +blobs" format.
116
*/
117

118
/* If the archive storage engine has been inited */
119
static bool archive_inited= FALSE;
120 121 122 123 124
/* Variables for archive share methods */
pthread_mutex_t archive_mutex;
static HASH archive_open_tables;

/* The file extension */
125 126 127 128 129 130 131 132 133 134 135 136
#define ARZ ".ARZ"               // The data file
#define ARN ".ARN"               // Files used during an optimize call
#define ARM ".ARM"               // Meta file
/*
  uchar + uchar + ulonglong + ulonglong + uchar
*/
#define META_BUFFER_SIZE 19      // Size of the data used in the meta file
/*
  uchar + uchar
*/
#define DATA_BUFFER_SIZE 2       // Size of the data used in the data file
#define ARCHIVE_CHECK_HEADER 254 // The number we use to determine corruption
137

138 139 140 141 142 143 144
/* 
  Number of rows that will force a bulk insert.
*/
#define ARCHIVE_MIN_ROWS_TO_USE_BULK_INSERT 2



serg@serg.mylan's avatar
serg@serg.mylan committed
145
/* dummy handlerton - only to have something to return from archive_db_init */
146
handlerton archive_hton = {
147
  "ARCHIVE",
148 149 150 151
  SHOW_OPTION_YES,
  "Archive storage engine", 
  DB_TYPE_ARCHIVE_DB,
  archive_db_init,
serg@serg.mylan's avatar
serg@serg.mylan committed
152 153
  0,       /* slot */
  0,       /* savepoint size. */
154 155 156 157 158 159 160 161 162 163 164 165 166
  NULL,    /* close_connection */
  NULL,    /* savepoint */
  NULL,    /* rollback to savepoint */
  NULL,    /* releas savepoint */
  NULL,    /* commit */
  NULL,    /* rollback */
  NULL,    /* prepare */
  NULL,    /* recover */
  NULL,    /* commit_by_xid */
  NULL,    /* rollback_by_xid */
  NULL,    /* create_cursor_read_view */
  NULL,    /* set_cursor_read_view */
  NULL,    /* close_cursor_read_view */
167
  HTON_NO_FLAGS
serg@serg.mylan's avatar
serg@serg.mylan committed
168 169 170
};


171 172 173 174 175 176 177 178 179 180
/*
  Used for hash table that tracks open tables.
*/
static byte* archive_get_key(ARCHIVE_SHARE *share,uint *length,
                             my_bool not_used __attribute__((unused)))
{
  *length=share->table_name_length;
  return (byte*) share->table_name;
}

181 182 183 184 185 186 187 188 189

/*
  Initialize the archive handler.

  SYNOPSIS
    archive_db_init()
    void

  RETURN
190 191
    FALSE       OK
    TRUE        Error
192 193
*/

194
bool archive_db_init()
195
{
196 197 198
  DBUG_ENTER("archive_db_init");
  if (pthread_mutex_init(&archive_mutex, MY_MUTEX_INIT_FAST))
    goto error;
serg@serg.mylan's avatar
serg@serg.mylan committed
199 200
  if (hash_init(&archive_open_tables, system_charset_info, 32, 0, 0,
                (hash_get_key) archive_get_key, 0, 0))
201 202 203 204 205 206 207 208 209 210 211
  {
    VOID(pthread_mutex_destroy(&archive_mutex));
  }
  else
  {
    archive_inited= TRUE;
    DBUG_RETURN(FALSE);
  }
error:
  have_archive_db= SHOW_OPTION_DISABLED;	// If we couldn't use handler
  DBUG_RETURN(TRUE);
212 213 214 215 216 217 218 219 220 221 222 223 224 225 226
}

/*
  Release the archive handler.

  SYNOPSIS
    archive_db_end()
    void

  RETURN
    FALSE       OK
*/

bool archive_db_end()
{
227 228 229 230 231 232
  if (archive_inited)
  {
    hash_free(&archive_open_tables);
    VOID(pthread_mutex_destroy(&archive_mutex));
  }
  archive_inited= 0;
233 234 235
  return FALSE;
}

236 237 238 239 240 241 242
ha_archive::ha_archive(TABLE *table_arg)
  :handler(&archive_hton, table_arg), delayed_insert(0), bulk_insert(0)
{
  /* Set our original buffer from pre-allocated memory */
  buffer.set((char *)byte_buffer, IO_SIZE, system_charset_info);

  /* The size of the offset value we will use for position() */
243 244
  ref_length = 2 << ((zlibCompileFlags() >> 6) & 3);
  DBUG_ASSERT(ref_length <= sizeof(z_off_t));
245
}
246

247 248 249 250 251
/*
  This method reads the header of a datafile and returns whether or not it was successful.
*/
int ha_archive::read_data_header(gzFile file_to_read)
{
252
  uchar data_buffer[DATA_BUFFER_SIZE];
253 254 255 256 257
  DBUG_ENTER("ha_archive::read_data_header");

  if (gzrewind(file_to_read) == -1)
    DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);

258
  if (gzread(file_to_read, data_buffer, DATA_BUFFER_SIZE) != DATA_BUFFER_SIZE)
259
    DBUG_RETURN(errno ? errno : -1);
260 261 262 263 264 265
  
  DBUG_PRINT("ha_archive::read_data_header", ("Check %u", data_buffer[0]));
  DBUG_PRINT("ha_archive::read_data_header", ("Version %u", data_buffer[1]));
  
  if ((data_buffer[0] != (uchar)ARCHIVE_CHECK_HEADER) &&  
      (data_buffer[1] != (uchar)ARCHIVE_VERSION))
266 267 268 269
    DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);

  DBUG_RETURN(0);
}
270 271

/*
272 273 274 275
  This method writes out the header of a datafile and returns whether or not it was successful.
*/
int ha_archive::write_data_header(gzFile file_to_write)
{
276
  uchar data_buffer[DATA_BUFFER_SIZE];
277 278
  DBUG_ENTER("ha_archive::write_data_header");

279 280 281 282
  data_buffer[0]= (uchar)ARCHIVE_CHECK_HEADER;
  data_buffer[1]= (uchar)ARCHIVE_VERSION;

  if (gzwrite(file_to_write, &data_buffer, DATA_BUFFER_SIZE) != 
283
      DATA_BUFFER_SIZE)
284
    goto error;
285 286
  DBUG_PRINT("ha_archive::write_data_header", ("Check %u", (uint)data_buffer[0]));
  DBUG_PRINT("ha_archive::write_data_header", ("Version %u", (uint)data_buffer[1]));
287 288 289 290 291 292 293 294 295 296

  DBUG_RETURN(0);
error:
  DBUG_RETURN(errno);
}

/*
  This method reads the header of a meta file and returns whether or not it was successful.
  *rows will contain the current number of rows in the data file upon success.
*/
297
int ha_archive::read_meta_file(File meta_file, ha_rows *rows)
298
{
299
  uchar meta_buffer[META_BUFFER_SIZE];
300 301 302 303 304
  ulonglong check_point;

  DBUG_ENTER("ha_archive::read_meta_file");

  VOID(my_seek(meta_file, 0, MY_SEEK_SET, MYF(0)));
305
  if (my_read(meta_file, (byte*)meta_buffer, META_BUFFER_SIZE, 0) != META_BUFFER_SIZE)
306 307 308 309 310
    DBUG_RETURN(-1);
  
  /*
    Parse out the meta data, we ignore version at the moment
  */
311
  *rows= (ha_rows)uint8korr(meta_buffer + 2);
312 313 314 315 316 317 318 319 320 321 322
  check_point= uint8korr(meta_buffer + 10);

  DBUG_PRINT("ha_archive::read_meta_file", ("Check %d", (uint)meta_buffer[0]));
  DBUG_PRINT("ha_archive::read_meta_file", ("Version %d", (uint)meta_buffer[1]));
  DBUG_PRINT("ha_archive::read_meta_file", ("Rows %lld", *rows));
  DBUG_PRINT("ha_archive::read_meta_file", ("Checkpoint %lld", check_point));
  DBUG_PRINT("ha_archive::read_meta_file", ("Dirty %d", (int)meta_buffer[18]));

  if ((meta_buffer[0] != (uchar)ARCHIVE_CHECK_HEADER) || 
      ((bool)meta_buffer[18] == TRUE))
    DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);
323 324 325 326 327 328 329 330 331

  my_sync(meta_file, MYF(MY_WME));

  DBUG_RETURN(0);
}

/*
  This method writes out the header of a meta file and returns whether or not it was successful.
  By setting dirty you say whether or not the file represents the actual state of the data file.
332
  Upon ::open() we set to dirty, and upon ::close() we set to clean.
333
*/
334
int ha_archive::write_meta_file(File meta_file, ha_rows rows, bool dirty)
335
{
336 337 338
  uchar meta_buffer[META_BUFFER_SIZE];
  ulonglong check_point= 0; //Reserved for the future

339 340
  DBUG_ENTER("ha_archive::write_meta_file");

341 342
  meta_buffer[0]= (uchar)ARCHIVE_CHECK_HEADER;
  meta_buffer[1]= (uchar)ARCHIVE_VERSION;
343
  int8store(meta_buffer + 2, (ulonglong)rows); 
344 345 346 347
  int8store(meta_buffer + 10, check_point); 
  *(meta_buffer + 18)= (uchar)dirty;
  DBUG_PRINT("ha_archive::write_meta_file", ("Check %d", (uint)ARCHIVE_CHECK_HEADER));
  DBUG_PRINT("ha_archive::write_meta_file", ("Version %d", (uint)ARCHIVE_VERSION));
348
  DBUG_PRINT("ha_archive::write_meta_file", ("Rows %llu", (ulonglong)rows));
349 350
  DBUG_PRINT("ha_archive::write_meta_file", ("Checkpoint %llu", check_point));
  DBUG_PRINT("ha_archive::write_meta_file", ("Dirty %d", (uint)dirty));
351 352

  VOID(my_seek(meta_file, 0, MY_SEEK_SET, MYF(0)));
353
  if (my_write(meta_file, (byte *)meta_buffer, META_BUFFER_SIZE, 0) != META_BUFFER_SIZE)
354 355 356 357 358 359 360 361 362 363
    DBUG_RETURN(-1);
  
  my_sync(meta_file, MYF(MY_WME));

  DBUG_RETURN(0);
}


/*
  We create the shared memory space that we will use for the open table. 
364 365 366
  No matter what we try to get or create a share. This is so that a repair
  table operation can occur. 

367
  See ha_example.cc for a longer description.
368
*/
369
ARCHIVE_SHARE *ha_archive::get_share(const char *table_name, TABLE *table)
370 371
{
  ARCHIVE_SHARE *share;
372
  char meta_file_name[FN_REFLEN];
373 374 375 376 377 378 379 380 381 382
  uint length;
  char *tmp_name;

  pthread_mutex_lock(&archive_mutex);
  length=(uint) strlen(table_name);

  if (!(share=(ARCHIVE_SHARE*) hash_search(&archive_open_tables,
                                           (byte*) table_name,
                                           length)))
  {
383
    if (!my_multi_malloc(MYF(MY_WME | MY_ZEROFILL),
384 385
                          &share, sizeof(*share),
                          &tmp_name, length+1,
386
                          NullS)) 
387 388 389 390 391
    {
      pthread_mutex_unlock(&archive_mutex);
      return NULL;
    }

392 393 394
    share->use_count= 0;
    share->table_name_length= length;
    share->table_name= tmp_name;
395
    share->crashed= FALSE;
396
    fn_format(share->data_file_name,table_name,"",ARZ,MY_REPLACE_EXT|MY_UNPACK_FILENAME);
397
    fn_format(meta_file_name,table_name,"",ARM,MY_REPLACE_EXT|MY_UNPACK_FILENAME);
398
    strmov(share->table_name,table_name);
399 400 401 402 403
    /*
      We will use this lock for rows.
    */
    VOID(pthread_mutex_init(&share->mutex,MY_MUTEX_INIT_FAST));
    if ((share->meta_file= my_open(meta_file_name, O_RDWR, MYF(0))) == -1)
404
      share->crashed= TRUE;
405 406 407
    
    /*
      After we read, we set the file to dirty. When we close, we will do the 
408 409
      opposite. If the meta file will not open we assume it is crashed and
      leave it up to the user to fix.
410
    */
411 412 413 414
    if (read_meta_file(share->meta_file, &share->rows_recorded))
      share->crashed= TRUE;
    else
      (void)write_meta_file(share->meta_file, share->rows_recorded, TRUE);
415

416
    /* 
417 418 419
      It is expensive to open and close the data files and since you can't have
      a gzip file that can be both read and written we keep a writer open
      that is shared amoung all open tables.
420
    */
421
    if ((share->archive_write= gzopen(share->data_file_name, "ab")) == NULL)
422 423
      share->crashed= TRUE;
    VOID(my_hash_insert(&archive_open_tables, (byte*) share));
424
    thr_lock_init(&share->lock);
425 426 427 428 429 430 431 432 433
  }
  share->use_count++;
  pthread_mutex_unlock(&archive_mutex);

  return share;
}


/* 
434
  Free the share.
435 436
  See ha_example.cc for a description.
*/
437
int ha_archive::free_share(ARCHIVE_SHARE *share)
438 439 440 441 442 443 444
{
  int rc= 0;
  pthread_mutex_lock(&archive_mutex);
  if (!--share->use_count)
  {
    hash_delete(&archive_open_tables, (byte*) share);
    thr_lock_delete(&share->lock);
445 446
    VOID(pthread_mutex_destroy(&share->mutex));
    (void)write_meta_file(share->meta_file, share->rows_recorded, FALSE);
447
    if (gzclose(share->archive_write) == Z_ERRNO)
448
      rc= 1;
449 450
    if (my_close(share->meta_file, MYF(0)))
      rc= 1;
bar@mysql.com's avatar
bar@mysql.com committed
451
    my_free((gptr) share, MYF(0));
452 453 454 455 456 457 458
  }
  pthread_mutex_unlock(&archive_mutex);

  return rc;
}


459
/*
460 461
  We just implement one additional file extension.
*/
462 463 464 465 466 467
static const char *ha_archive_exts[] = {
  ARZ,
  ARM,
  NullS
};

468
const char **ha_archive::bas_ext() const
469 470 471
{
  return ha_archive_exts;
}
472 473 474 475 476 477 478 479 480 481 482 483


/* 
  When opening a file we:
  Create/get our shared structure.
  Init out lock.
  We open the file we will read from.
*/
int ha_archive::open(const char *name, int mode, uint test_if_locked)
{
  DBUG_ENTER("ha_archive::open");

484
  if (!(share= get_share(name, table)))
485
    DBUG_RETURN(HA_ERR_OUT_OF_MEM); // Not handled well by calling code!
486 487
  thr_lock_data_init(&share->lock,&lock,NULL);

488 489
  if ((archive= gzopen(share->data_file_name, "rb")) == NULL)
  {
490 491 492
    if (errno == EROFS || errno == EACCES)
      DBUG_RETURN(my_errno= errno);
    DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);
493
  }
494 495 496 497 498 499

  DBUG_RETURN(0);
}


/*
500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515
  Closes the file.

  SYNOPSIS
    close();
  
  IMPLEMENTATION:

  We first close this storage engines file handle to the archive and
  then remove our reference count to the table (and possibly free it
  as well).

  RETURN
    0  ok
    1  Error
*/

516 517
int ha_archive::close(void)
{
518
  int rc= 0;
519
  DBUG_ENTER("ha_archive::close");
520 521 522 523 524 525 526 527

  /* First close stream */
  if (gzclose(archive) == Z_ERRNO)
    rc= 1;
  /* then also close share */
  rc|= free_share(share);

  DBUG_RETURN(rc);
528 529 530 531
}


/*
532 533 534 535 536 537
  We create our data file here. The format is pretty simple. 
  You can read about the format of the data file above.
  Unlike other storage engines we do not "pack" our data. Since we 
  are about to do a general compression, packing would just be a waste of 
  CPU time. If the table has blobs they are written after the row in the order 
  of creation.
538 539 540 541
*/

int ha_archive::create(const char *name, TABLE *table_arg,
                       HA_CREATE_INFO *create_info)
542
{
543
  File create_file;  // We use to create the datafile and the metafile
544
  char name_buff[FN_REFLEN];
545
  int error;
546 547
  DBUG_ENTER("ha_archive::create");

548 549 550 551 552 553 554 555 556 557 558 559 560
  if ((create_file= my_create(fn_format(name_buff,name,"",ARM,
                                        MY_REPLACE_EXT|MY_UNPACK_FILENAME),0,
                              O_RDWR | O_TRUNC,MYF(MY_WME))) < 0)
  {
    error= my_errno;
    goto error;
  }
  write_meta_file(create_file, 0, FALSE);
  my_close(create_file,MYF(0));

  /* 
    We reuse name_buff since it is available.
  */
561 562 563 564 565
  if ((create_file= my_create(fn_format(name_buff,name,"",ARZ,
                                        MY_REPLACE_EXT|MY_UNPACK_FILENAME),0,
                              O_RDWR | O_TRUNC,MYF(MY_WME))) < 0)
  {
    error= my_errno;
566
    goto error;
567
  }
568
  if ((archive= gzdopen(create_file, "wb")) == NULL)
569
  {
570
    error= errno;
571
    goto error2;
572
  }
573
  if (write_data_header(archive))
574
  {
575 576
    error= errno;
    goto error3;
577
  }
578 579

  if (gzclose(archive))
jimw@mysql.com's avatar
jimw@mysql.com committed
580
  {
581
    error= errno;
582
    goto error2;
583 584 585
  }

  my_close(create_file, MYF(0));
586

587
  DBUG_RETURN(0);
588

589 590 591
error3:
  /* We already have an error, so ignore results of gzclose. */
  (void)gzclose(archive);
592
error2:
593 594
  my_close(create_file, MYF(0));
  delete_table(name);
595
error:
596 597
  /* Return error number, if we got one */
  DBUG_RETURN(error ? error : -1);
598 599
}

600 601
/*
  This is where the actual row is written out.
602
*/
603
int ha_archive::real_write_row(byte *buf, gzFile writer)
604 605
{
  z_off_t written;
606
  uint *ptr, *end;
607
  DBUG_ENTER("ha_archive::real_write_row");
608

609 610
  written= gzwrite(writer, buf, table->s->reclength);
  DBUG_PRINT("ha_archive::real_write_row", ("Wrote %d bytes expected %d", written, table->s->reclength));
611
  if (!delayed_insert || !bulk_insert)
612 613
    share->dirty= TRUE;

614
  if (written != (z_off_t)table->s->reclength)
615
    DBUG_RETURN(errno ? errno : -1);
616 617 618 619
  /*
    We should probably mark the table as damagaged if the record is written
    but the blob fails.
  */
monty@mysql.com's avatar
monty@mysql.com committed
620
  for (ptr= table->s->blob_field, end= ptr + table->s->blob_fields ;
621 622
       ptr != end ;
       ptr++)
623
  {
624
    char *data_ptr;
625
    uint32 size= ((Field_blob*) table->field[*ptr])->get_length();
626

627 628
    if (size)
    {
629
      ((Field_blob*) table->field[*ptr])->get_ptr(&data_ptr);
630
      written= gzwrite(writer, data_ptr, (unsigned)size);
631
      if (written != (z_off_t)size)
632
        DBUG_RETURN(errno ? errno : -1);
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
  DBUG_RETURN(0);
}


/* 
  Look at ha_archive::open() for an explanation of the row format.
  Here we just write out the row.

  Wondering about start_bulk_insert()? We don't implement it for
  archive since it optimizes for lots of writes. The only save
  for implementing start_bulk_insert() is that we could skip 
  setting dirty to true each time.
*/
int ha_archive::write_row(byte *buf)
{
  int rc;
  DBUG_ENTER("ha_archive::write_row");

  if (share->crashed)
      DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);

  statistic_increment(table->in_use->status_var.ha_write_count, &LOCK_status);
  if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_INSERT)
    table->timestamp_field->set_time();
  pthread_mutex_lock(&share->mutex);
660
  share->rows_recorded++;
661
  rc= real_write_row(buf, share->archive_write);
662
  pthread_mutex_unlock(&share->mutex);
663

664
  DBUG_RETURN(rc);
665 666 667 668 669 670 671
}

/*
  All calls that need to scan the table start with this method. If we are told
  that it is a table scan we rewind the file to the beginning, otherwise
  we assume the position will be set.
*/
672

673 674 675
int ha_archive::rnd_init(bool scan)
{
  DBUG_ENTER("ha_archive::rnd_init");
676 677 678
  
  if (share->crashed)
      DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);
679

680
  /* We rewind the file so that we can read from the beginning if scan */
681
  if (scan)
682
  {
683
    scan_rows= share->rows_recorded;
684 685
    records= 0;

686 687 688 689
    /* 
      If dirty, we lock, and then reset/flush the data.
      I found that just calling gzflush() doesn't always work.
    */
690
    if (share->dirty == TRUE)
691
    {
692 693 694 695 696 697 698
      pthread_mutex_lock(&share->mutex);
      if (share->dirty == TRUE)
      {
        gzflush(share->archive_write, Z_SYNC_FLUSH);
        share->dirty= FALSE;
      }
      pthread_mutex_unlock(&share->mutex);
699
    }
700 701 702

    if (read_data_header(archive))
      DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);
703 704
  }

705 706 707 708 709 710 711 712
  DBUG_RETURN(0);
}


/*
  This is the method that is used to read a row. It assumes that the row is 
  positioned where you want it.
*/
713
int ha_archive::get_row(gzFile file_to_read, byte *buf)
714 715
{
  int read; // Bytes read, gzread() returns int
716
  uint *ptr, *end;
717 718
  char *last;
  size_t total_blob_length= 0;
719
  DBUG_ENTER("ha_archive::get_row");
720

721 722
  read= gzread(file_to_read, buf, table->s->reclength);
  DBUG_PRINT("ha_archive::get_row", ("Read %d bytes expected %d", read, table->s->reclength));
723 724 725

  if (read == Z_STREAM_ERROR)
    DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);
726 727 728 729 730

  /* If we read nothing we are at the end of the file */
  if (read == 0)
    DBUG_RETURN(HA_ERR_END_OF_FILE);

731 732 733
  /* 
    If the record is the wrong size, the file is probably damaged, unless 
    we are dealing with a delayed insert or a bulk insert.
734
  */
735
  if ((ulong) read != table->s->reclength)
736
    DBUG_RETURN(HA_ERR_END_OF_FILE);
737 738

  /* Calculate blob length, we use this for our buffer */
739 740 741 742
  for (ptr= table->s->blob_field, end=ptr + table->s->blob_fields ;
       ptr != end ;
       ptr++)
    total_blob_length += ((Field_blob*) table->field[*ptr])->get_length();
743 744 745

  /* Adjust our row buffer if we need be */
  buffer.alloc(total_blob_length);
746
  last= (char *)buffer.ptr();
747

748
  /* Loop through our blobs and read them */
749 750 751
  for (ptr= table->s->blob_field, end=ptr + table->s->blob_fields ;
       ptr != end ;
       ptr++)
752
  {
753
    size_t size= ((Field_blob*) table->field[*ptr])->get_length();
754 755 756 757
    if (size)
    {
      read= gzread(file_to_read, last, size);
      if ((size_t) read != size)
758
        DBUG_RETURN(HA_ERR_END_OF_FILE);
759
      ((Field_blob*) table->field[*ptr])->set_ptr(size, last);
760 761
      last += size;
    }
762 763 764 765
  }
  DBUG_RETURN(0);
}

766

767 768 769 770
/* 
  Called during ORDER BY. Its position is either from being called sequentially
  or by having had ha_archive::rnd_pos() called before it is called.
*/
771

772 773 774
int ha_archive::rnd_next(byte *buf)
{
  int rc;
775
  DBUG_ENTER("ha_archive::rnd_next");
776

777 778 779
  if (share->crashed)
      DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);

780 781 782 783
  if (!scan_rows)
    DBUG_RETURN(HA_ERR_END_OF_FILE);
  scan_rows--;

784 785
  statistic_increment(table->in_use->status_var.ha_read_rnd_next_count,
		      &LOCK_status);
786
  current_position= gztell(archive);
787 788 789
  rc= get_row(archive, buf);


790
  if (rc != HA_ERR_END_OF_FILE)
791 792 793 794 795 796
    records++;

  DBUG_RETURN(rc);
}


797
/*
798 799 800 801
  Thanks to the table flag HA_REC_NOT_IN_SEQ this will be called after
  each call to ha_archive::rnd_next() if an ordering of the rows is
  needed.
*/
802

803 804 805
void ha_archive::position(const byte *record)
{
  DBUG_ENTER("ha_archive::position");
806
  my_store_ptr(ref, ref_length, current_position);
807 808 809 810 811
  DBUG_VOID_RETURN;
}


/*
812 813 814 815
  This is called after a table scan for each row if the results of the
  scan need to be ordered. It will take *pos and use it to move the
  cursor in the file so that the next row that is called is the
  correctly ordered row.
816
*/
817

818 819 820
int ha_archive::rnd_pos(byte * buf, byte *pos)
{
  DBUG_ENTER("ha_archive::rnd_pos");
821 822
  statistic_increment(table->in_use->status_var.ha_read_rnd_next_count,
		      &LOCK_status);
823
  current_position= (z_off_t)my_get_ptr(pos, ref_length);
824
  (void)gzseek(archive, current_position, SEEK_SET);
825

826 827 828 829
  DBUG_RETURN(get_row(archive, buf));
}

/*
830
  This method repairs the meta file. It does this by walking the datafile and 
831 832
  rewriting the meta file. Currently it does this by calling optimize with
  the extended flag.
833
*/
834
int ha_archive::repair(THD* thd, HA_CHECK_OPT* check_opt)
835
{
836
  DBUG_ENTER("ha_archive::repair");
837 838
  check_opt->flags= T_EXTEND;
  int rc= optimize(thd, check_opt);
839

840 841
  if (rc)
    DBUG_RETURN(HA_ERR_CRASHED_ON_REPAIR);
842

843
  share->crashed= FALSE;
844
  DBUG_RETURN(0);
845 846
}

847 848 849
/*
  The table can become fragmented if data was inserted, read, and then
  inserted again. What we do is open up the file and recompress it completely. 
850
*/
851 852 853
int ha_archive::optimize(THD* thd, HA_CHECK_OPT* check_opt)
{
  DBUG_ENTER("ha_archive::optimize");
854 855
  int rc;
  gzFile writer;
856 857
  char writer_filename[FN_REFLEN];

858 859
  /* Flush any waiting data */
  gzflush(share->archive_write, Z_SYNC_FLUSH);
860

861
  /* Lets create a file to contain the new data */
862 863
  fn_format(writer_filename, share->table_name, "", ARN, 
            MY_REPLACE_EXT|MY_UNPACK_FILENAME);
864 865

  if ((writer= gzopen(writer_filename, "wb")) == NULL)
866 867 868 869 870 871 872 873
    DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE); 

  /* 
    An extended rebuild is a lot more effort. We open up each row and re-record it. 
    Any dead rows are removed (aka rows that may have been partially recorded). 
  */

  if (check_opt->flags == T_EXTEND)
874
  {
875
    byte *buf; 
876

877 878 879 880 881 882 883 884 885
    /* 
      First we create a buffer that we can use for reading rows, and can pass
      to get_row().
    */
    if (!(buf= (byte*) my_malloc(table->s->reclength, MYF(MY_WME))))
    {
      rc= HA_ERR_OUT_OF_MEM;
      goto error;
    }
886

887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904
    /*
      Now we will rewind the archive file so that we are positioned at the 
      start of the file.
    */
    rc= read_data_header(archive);
    
    /*
      Assuming now error from rewinding the archive file, we now write out the 
      new header for out data file.
    */
    if (!rc)
      rc= write_data_header(writer);

    /* 
      On success of writing out the new header, we now fetch each row and
      insert it into the new archive file. 
    */
    if (!rc)
905 906
    {
      share->rows_recorded= 0;
907
      while (!(rc= get_row(archive, buf)))
908
      {
909
        real_write_row(buf, writer);
910 911 912
        share->rows_recorded++;
      }
    }
913

reggie@linux.site's avatar
reggie@linux.site committed
914
    my_free((char*)buf, MYF(0));
915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937
    if (rc && rc != HA_ERR_END_OF_FILE)
      goto error;
  } 
  else
  {
    /* 
      The quick method is to just read the data raw, and then compress it directly.
    */
    int read; // Bytes read, gzread() returns int
    char block[IO_SIZE];
    if (gzrewind(archive) == -1)
    {
      rc= HA_ERR_CRASHED_ON_USAGE;
      goto error;
    }

    while ((read= gzread(archive, block, IO_SIZE)))
      gzwrite(writer, block, read);
  }

  gzflush(writer, Z_SYNC_FLUSH);
  gzclose(share->archive_write);
  share->archive_write= writer; 
938 939 940 941 942

  my_rename(writer_filename,share->data_file_name,MYF(0));

  DBUG_RETURN(0); 

943 944 945 946 947
error:
  gzclose(writer);

  DBUG_RETURN(rc); 
}
948 949 950 951 952 953 954 955

/* 
  Below is an example of how to setup row level locking.
*/
THR_LOCK_DATA **ha_archive::store_lock(THD *thd,
                                       THR_LOCK_DATA **to,
                                       enum thr_lock_type lock_type)
{
956 957 958 959 960
  if (lock_type == TL_WRITE_DELAYED)
    delayed_insert= TRUE;
  else
    delayed_insert= FALSE;

961 962
  if (lock_type != TL_IGNORE && lock.type == TL_UNLOCK) 
  {
963 964 965 966 967 968 969 970 971
    /* 
      Here is where we get into the guts of a row level lock.
      If TL_UNLOCK is set 
      If we are not doing a LOCK TABLE or DISCARD/IMPORT
      TABLESPACE, then allow multiple writers 
    */

    if ((lock_type >= TL_WRITE_CONCURRENT_INSERT &&
         lock_type <= TL_WRITE) && !thd->in_lock_tables
972
        && !thd->tablespace_op)
973 974 975 976 977 978 979 980 981 982
      lock_type = TL_WRITE_ALLOW_WRITE;

    /* 
      In queries of type INSERT INTO t1 SELECT ... FROM t2 ...
      MySQL would use the lock TL_READ_NO_INSERT on t2, and that
      would conflict with TL_WRITE_ALLOW_WRITE, blocking all inserts
      to t2. Convert the lock to a normal read lock to allow
      concurrent inserts to t2. 
    */

983
    if (lock_type == TL_READ_NO_INSERT && !thd->in_lock_tables) 
984 985 986 987 988 989 990 991 992 993
      lock_type = TL_READ;

    lock.type=lock_type;
  }

  *to++= &lock;

  return to;
}

994 995 996 997

/*
  Hints for optimizer, see ha_tina for more information
*/
998 999 1000
void ha_archive::info(uint flag)
{
  DBUG_ENTER("ha_archive::info");
1001 1002 1003 1004
  /* 
    This should be an accurate number now, though bulk and delayed inserts can
    cause the number to be inaccurate.
  */
1005 1006
  records= share->rows_recorded;
  deleted= 0;
1007 1008 1009 1010 1011 1012 1013
  /* Costs quite a bit more to get all information */
  if (flag & HA_STATUS_TIME)
  {
    MY_STAT file_stat;  // Stat information for the data file

    VOID(my_stat(share->data_file_name, &file_stat, MYF(MY_WME)));

1014
    mean_rec_length= table->s->reclength + buffer.alloced_length();
1015
    data_file_length= file_stat.st_size;
1016 1017
    create_time= file_stat.st_ctime;
    update_time= file_stat.st_mtime;
1018
    max_data_file_length= share->rows_recorded * mean_rec_length;
1019 1020 1021
  }
  delete_length= 0;
  index_file_length=0;
1022

1023 1024
  DBUG_VOID_RETURN;
}
1025 1026 1027 1028 1029 1030 1031 1032 1033 1034


/*
  This method tells us that a bulk insert operation is about to occur. We set
  a flag which will keep write_row from saying that its data is dirty. This in
  turn will keep selects from causing a sync to occur.
  Basically, yet another optimizations to keep compression working well.
*/
void ha_archive::start_bulk_insert(ha_rows rows)
{
1035
  DBUG_ENTER("ha_archive::start_bulk_insert");
1036 1037
  if (!rows || rows >= ARCHIVE_MIN_ROWS_TO_USE_BULK_INSERT)
    bulk_insert= TRUE;
1038 1039 1040 1041 1042 1043 1044 1045 1046 1047
  DBUG_VOID_RETURN;
}


/* 
  Other side of start_bulk_insert, is end_bulk_insert. Here we turn off the bulk insert
  flag, and set the share dirty so that the next select will call sync for us.
*/
int ha_archive::end_bulk_insert()
{
1048
  DBUG_ENTER("ha_archive::end_bulk_insert");
1049 1050 1051 1052
  bulk_insert= FALSE;
  share->dirty= TRUE;
  DBUG_RETURN(0);
}
1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063

/*
  We cancel a truncate command. The only way to delete an archive table is to drop it.
  This is done for security reasons. In a later version we will enable this by 
  allowing the user to select a different row format.
*/
int ha_archive::delete_all_rows()
{
  DBUG_ENTER("ha_archive::delete_all_rows");
  DBUG_RETURN(0);
}
1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140

/*
  We just return state if asked.
*/
bool ha_archive::is_crashed() const 
{
  return share->crashed; 
}

/*
  Simple scan of the tables to make sure everything is ok.
*/

int ha_archive::check(THD* thd, HA_CHECK_OPT* check_opt)
{
  int rc= 0;
  byte *buf; 
  const char *old_proc_info=thd->proc_info;
  ha_rows count= share->rows_recorded;
  DBUG_ENTER("ha_archive::check");

  thd->proc_info= "Checking table";
  /* Flush any waiting data */
  gzflush(share->archive_write, Z_SYNC_FLUSH);

  /* 
    First we create a buffer that we can use for reading rows, and can pass
    to get_row().
  */
  if (!(buf= (byte*) my_malloc(table->s->reclength, MYF(MY_WME))))
    rc= HA_ERR_OUT_OF_MEM;

  /*
    Now we will rewind the archive file so that we are positioned at the 
    start of the file.
  */
  if (!rc)
    read_data_header(archive);

  if (!rc)
    while (!(rc= get_row(archive, buf)))
      count--;

  my_free((char*)buf, MYF(0));

  thd->proc_info= old_proc_info;

  if ((rc && rc != HA_ERR_END_OF_FILE) || count)  
  {
    share->crashed= FALSE;
    DBUG_RETURN(HA_ADMIN_CORRUPT);
  }
  else
  {
    DBUG_RETURN(HA_ADMIN_OK);
  }
}

/*
  Check and repair the table if needed.
*/
bool ha_archive::check_and_repair(THD *thd) 
{
  HA_CHECK_OPT check_opt;
  DBUG_ENTER("ha_archive::check_and_repair");

  check_opt.init();

  if (check(thd, &check_opt) == HA_ADMIN_CORRUPT)
  {
    DBUG_RETURN(repair(thd, &check_opt));
  }
  else
  {
    DBUG_RETURN(HA_ADMIN_OK);
  }
}
1141
#endif /* HAVE_ARCHIVE_DB */