1<?xml version="1.0" encoding="UTF-8" standalone="no"?> 2<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"> 3<html xmlns="http://www.w3.org/1999/xhtml"> 4 <head> 5 <meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /> 6 <title>Isolation</title> 7 <link rel="stylesheet" href="gettingStarted.css" type="text/css" /> 8 <meta name="generator" content="DocBook XSL Stylesheets V1.73.2" /> 9 <link rel="start" href="index.html" title="Getting Started with Berkeley DB Transaction Processing" /> 10 <link rel="up" href="txnconcurrency.html" title="Chapter 4. Concurrency" /> 11 <link rel="prev" href="lockingsubsystem.html" title="The Locking Subsystem" /> 12 <link rel="next" href="txn_ccursor.html" title="Transactional Cursors and Concurrent Applications" /> 13 </head> 14 <body> 15 <div class="navheader"> 16 <table width="100%" summary="Navigation header"> 17 <tr> 18 <th colspan="3" align="center">Isolation</th> 19 </tr> 20 <tr> 21 <td width="20%" align="left"><a accesskey="p" href="lockingsubsystem.html">Prev</a> </td> 22 <th width="60%" align="center">Chapter 4. Concurrency</th> 23 <td width="20%" align="right"> <a accesskey="n" href="txn_ccursor.html">Next</a></td> 24 </tr> 25 </table> 26 <hr /> 27 </div> 28 <div class="sect1" lang="en" xml:lang="en"> 29 <div class="titlepage"> 30 <div> 31 <div> 32 <h2 class="title" style="clear: both"><a id="isolation"></a>Isolation</h2> 33 </div> 34 </div> 35 </div> 36 <div class="toc"> 37 <dl> 38 <dt> 39 <span class="sect2"> 40 <a href="isolation.html#degreesofisolation">Supported Degrees of Isolation</a> 41 </span> 42 </dt> 43 <dt> 44 <span class="sect2"> 45 <a href="isolation.html#dirtyreads">Reading Uncommitted Data</a> 46 </span> 47 </dt> 48 <dt> 49 <span class="sect2"> 50 <a href="isolation.html#readcommitted">Committed Reads</a> 51 </span> 52 </dt> 53 <dt> 54 <span class="sect2"> 55 <a href="isolation.html#snapshot_isolation">Using Snapshot Isolation</a> 56 </span> 57 </dt> 58 </dl> 59 </div> 60 <p> 61 Isolation guarantees are an important aspect of transactional 62 protection. Transactions 63 ensure the data your transaction is working with will not be changed by some other transaction. 64 Moreover, the modifications made by a transaction will never be viewable outside of that transaction until 65 the changes have been committed. 66 </p> 67 <p> 68 That said, there are different degrees of isolation, and you can choose to relax your isolation 69 guarantees to one degree or another depending on your application's requirements. The primary reason why 70 you might want to do this is because of performance; the more isolation you ask your transactions to 71 provide, the more locking that your application must do. With more locking comes a greater chance of 72 blocking, which in turn causes your threads to pause while waiting for a lock. Therefore, by relaxing 73 your isolation guarantees, you can <span class="emphasis"><em>potentially</em></span> improve your application's throughput. 74 Whether you actually see any improvement depends, of course, on 75 the nature of your application's data and transactions. 76 </p> 77 <div class="sect2" lang="en" xml:lang="en"> 78 <div class="titlepage"> 79 <div> 80 <div> 81 <h3 class="title"><a id="degreesofisolation"></a>Supported Degrees of Isolation</h3> 82 </div> 83 </div> 84 </div> 85 <p> 86 DB supports the following levels of isolation: 87 </p> 88 <div class="informaltable"> 89 <table border="1" width="80%"> 90 <colgroup> 91 <col /> 92 <col /> 93 <col /> 94 </colgroup> 95 <thead> 96 <tr> 97 <th>Degree</th> 98 <th>ANSI Term</th> 99 <th>Definition</th> 100 </tr> 101 </thead> 102 <tbody> 103 <tr> 104 <td>1</td> 105 <td>READ UNCOMMITTED</td> 106 <td> 107 Uncommitted reads means that one transaction will never 108 overwrite another transaction's dirty data. Dirty data is 109 data that a transaction has modified but not yet committed 110 to the underlying data store. However, uncommitted reads allows a 111 transaction to see data dirtied by another 112 transaction. In addition, a transaction may read data 113 dirtied by another transaction, but which subsequently 114 is aborted by that other transaction. In this latter 115 case, the reading transaction may be reading data that 116 never really existed in the database. 117 </td> 118 </tr> 119 <tr> 120 <td>2</td> 121 <td>READ COMMITTED</td> 122 <td> 123 <p> 124 Committed read isolation means that degree 1 is observed, except that dirty data is never read. 125 </p> 126 <p> 127 In addition, this isolation level guarantees that data will never change so long as 128 it is addressed by the cursor, but the data may change before the reading cursor is closed. 129 In the case of a transaction, data at the current 130 cursor position will not change, but once the cursor 131 moves, the previous referenced data can change. This 132 means that readers release read locks before the cursor 133 is closed, and therefore, before the transaction 134 completes. Note that this level of isolation causes the 135 cursor to operate in exactly the same way as it does in 136 the absence of a transaction. 137 </p> 138 </td> 139 </tr> 140 <tr> 141 <td>3</td> 142 <td>SERIALIZABLE</td> 143 <td> 144 <p> 145 Committed read is observed, plus the data read by a transaction, T, 146 will never be dirtied by another transaction before T completes. 147 This means that both read and write locks are not 148 released until the transaction completes. 149 </p> 150 <p> 151 <span> 152 In addition, 153 </span> 154 155 156 157 no transactions will see phantoms. Phantoms are records 158 returned as a result of a search, but which were not seen by 159 the same transaction when the identical 160 search criteria was previously used. 161 </p> 162 <p> 163 This is DB's default isolation guarantee. 164 </p> 165 </td> 166 </tr> 167 </tbody> 168 </table> 169 </div> 170 <p> 171 172 By default, DB transactions and transactional cursors offer 173 <span> 174 serializable isolation. 175 </span> 176 177 178 You can optionally reduce your isolation level by configuring DB to use 179 uncommitted read isolation. See 180 <a class="xref" href="isolation.html#dirtyreads" title="Reading Uncommitted Data">Reading Uncommitted Data</a> 181 for more information. 182 183 You can also configure DB to use committed read isolation. See 184 <a class="xref" href="isolation.html#readcommitted" title="Committed Reads">Committed Reads</a> 185 for more information. 186 187 </p> 188 <p> 189 Finally, in addition to DB's normal degrees of isolation, you 190 can also use <span class="emphasis"><em>snapshot isolation</em></span>. This allows 191 you to avoid the read locks that serializable isolation requires. See 192 <a class="xref" href="isolation.html#snapshot_isolation" title="Using Snapshot Isolation">Using Snapshot Isolation</a> 193 for details. 194 </p> 195 </div> 196 <div class="sect2" lang="en" xml:lang="en"> 197 <div class="titlepage"> 198 <div> 199 <div> 200 <h3 class="title"><a id="dirtyreads"></a>Reading Uncommitted Data</h3> 201 </div> 202 </div> 203 </div> 204 <p> 205 Berkeley DB allows you to configure your application to read data that has been modified but not yet 206 committed by another transaction; that is, dirty data. When you do this, you 207 may see a performance benefit by allowing your 208 application to not have to block waiting for write locks. On the other hand, the data that your 209 application is reading may change before the transaction has completed. 210 </p> 211 <p> 212 When used with transactions, uncommitted reads means that one transaction can see data 213 modified but not yet committed by another transaction. When 214 used with transactional cursors, uncommitted reads means 215 that any database reader can see data modified by the 216 cursor before the cursor's transaction has committed. 217 </p> 218 <p> 219 Because of this, uncommitted reads allow a transaction to read data 220 that may subsequently be aborted by another transaction. In 221 this case, the reading transaction will have read data that 222 never really existed in the database. 223 </p> 224 <p> 225 To configure your application to read uncommitted data: 226 </p> 227 <div class="orderedlist"> 228 <ol type="1"> 229 <li> 230 <p> 231 Open your database such that it will allow uncommitted reads. You do this by 232 <span> 233 specifying <code class="literal">DB_READ_UNCOMMITTED</code> when you open your database. 234 </span> 235 236 </p> 237 </li> 238 <li> 239 <p> 240 <span> 241 Specify <code class="literal">DB_READ_UNCOMMITTED</code> 242 when you create the transaction, 243 open the cursor, or read a record from the database. 244 </span> 245 246 247 </p> 248 </li> 249 </ol> 250 </div> 251 <p> 252 For example, the following opens the database such that it supports uncommitted reads, and then creates a 253 transaction that causes all reads performed within it to use uncommitted reads. Remember that simply opening 254 the database to support uncommitted reads is not enough; you must also declare your read operations to be 255 performed using uncommitted reads. 256 </p> 257 <pre class="programlisting">#include "db_cxx.h" 258 259... 260 261int main(void) 262{ 263 u_int32_t env_flags = DB_CREATE | // If the environment does not 264 // exist, create it. 265 DB_INIT_LOCK | // Initialize locking 266 DB_INIT_LOG | // Initialize logging 267 DB_INIT_MPOOL | // Initialize the cache 268 DB_THREAD | // Free-thread the env handle 269 DB_INIT_TXN; // Initialize transactions 270 271 u_int32_t db_flags = DB_CREATE | // Create the db if it does 272 // not exist 273 DB_AUTO_COMMIT | // Enable auto commit 274 DB_READ_UNCOMMITTED; // Enable uncommitted reads 275 276 Db *dbp = NULL; 277 const char *file_name = "mydb.db"; 278 const char *keystr ="thekey"; 279 const char *datastr = "thedata"; 280 281 std::string envHome("/export1/testEnv"); 282 DbEnv myEnv(0); 283 284 try { 285 286 myEnv.open(envHome.c_str(), env_flags, 0); 287 dbp = new Db(&myEnv, 0); 288 dbp->open(NULL, // Txn pointer 289 file_name, // File name 290 NULL, // Logical db name 291 DB_BTREE, // Database type (using btree) 292 db_flags, // Open flags 293 0); // File mode. Using defaults 294 295 DbTxn *txn = NULL; 296 myEnv.txn_begin(NULL, &txn, DB_READ_UNCOMMITTED); 297 298 // From here, you perform your database reads and writes as normal, 299 // committing and aborting the transactions as is necessary, and 300 // testing for deadlock exceptions as normal (omitted for brevity). 301 302 ...</pre> 303 </div> 304 <div class="sect2" lang="en" xml:lang="en"> 305 <div class="titlepage"> 306 <div> 307 <div> 308 <h3 class="title"><a id="readcommitted"></a>Committed Reads</h3> 309 </div> 310 </div> 311 </div> 312 <p> 313 You can configure your transaction so that the data being 314 read by a transactional cursor is consistent so long as it 315 is being addressed by the cursor. However, once the cursor is done reading the 316 317 <span> 318 record (that is, reading records from the page that it currently has locked), 319 </span> 320 321 the cursor releases its lock on that 322 323 <span> 324 record or page. 325 </span> 326 327 This means that the data the cursor has read and released 328 may change before the cursor's transaction has completed. 329 </p> 330 <p> 331 For example, 332 suppose you have two transactions, <code class="literal">Ta</code> and <code class="literal">Tb</code>. Suppose further that 333 <code class="literal">Ta</code> has a cursor that reads <code class="literal">record R</code>, but does not modify it. Normally, 334 <code class="literal">Tb</code> would then be unable to write <code class="literal">record R</code> because 335 <code class="literal">Ta</code> would be holding a read lock on it. But when you configure your transaction for 336 committed reads, <code class="literal">Tb</code> <span class="emphasis"><em>can</em></span> modify <code class="literal">record 337 R</code> before <code class="literal">Ta</code> completes, so long as the reading cursor is no longer 338 addressing the 339 340 <span> 341 record or page. 342 </span> 343 344 </p> 345 <p> 346 When you configure your application for this level of isolation, you may see better performance 347 throughput because there are fewer read locks being held by your transactions. 348 Read committed isolation is most useful when you have a cursor that is reading and/or writing records in 349 a single direction, and that does not ever have to go back to re-read those same records. In this case, 350 you can allow DB to release read locks as it goes, rather than hold them for the life of the 351 transaction. 352 </p> 353 <p> 354 To configure your application to use committed reads, do one of the following: 355 </p> 356 <div class="itemizedlist"> 357 <ul type="disc"> 358 <li> 359 <p> 360 Create your transaction such that it allows committed reads. You do this by 361 <span> 362 specifying <code class="literal">DB_READ_COMMITTED</code> when you open the transaction. 363 </span> 364 365 </p> 366 </li> 367 <li> 368 <p> 369 <span> 370 Specify <code class="literal">DB_READ_COMMITTED</code> 371 when you open the cursor. 372 </span> 373 374 </p> 375 </li> 376 </ul> 377 </div> 378 <p> 379 For example, the following creates a transaction that allows committed reads: 380 </p> 381 <pre class="programlisting">#include "db_cxx.h" 382 383... 384 385int main(void) 386{ 387 u_int32_t env_flags = DB_CREATE | // If the environment does not 388 // exist, create it. 389 DB_INIT_LOCK | // Initialize locking 390 DB_INIT_LOG | // Initialize logging 391 DB_INIT_MPOOL | // Initialize the cache 392 DB_THREAD | // Free-thread the env handle 393 DB_INIT_TXN; // Initialize transactions 394 395 // Notice that we do not have to specify any flags to the database to 396 // allow committed reads (this is as opposed to uncommitted reads 397 // where we DO have to specify a flag on the database open. 398 u_int32_t db_flags = DB_CREATE | DB_AUTO_COMMIT; 399 Db *dbp = NULL; 400 const char *file_name = "mydb.db"; 401 402 std::string envHome("/export1/testEnv"); 403 DbEnv myEnv(0); 404 405 try { 406 407 myEnv.open(envHome.c_str(), env_flags, 0); 408 dbp = new Db(&myEnv, 0); 409 dbp->open(NULL, // Txn pointer 410 file_name, // File name 411 NULL, // Logical db name 412 DB_BTREE, // Database type (using btree) 413 db_flags, // Open flags 414 0); // File mode. Using defaults 415 416 DbTxn *txn = NULL; 417 418 // Open the transaction and enable committed reads. All cursors 419 // open with this transaction handle will use read committed 420 // isolation. 421 myEnv.txn_begin(NULL, &txn, DB_READ_COMMITTED); 422 423 // From here, you perform your database reads and writes as normal, 424 // committing and aborting the transactions as is necessary, 425 // testing for deadlock exceptions as normal (omitted for brevity). 426 427 // Using transactional cursors with concurrent applications is 428 // described in more detail in the following section. 429 430 ...</pre> 431 </div> 432 <div class="sect2" lang="en" xml:lang="en"> 433 <div class="titlepage"> 434 <div> 435 <div> 436 <h3 class="title"><a id="snapshot_isolation"></a>Using Snapshot Isolation</h3> 437 </div> 438 </div> 439 </div> 440 <p> 441 By default DB uses serializable isolation. An 442 important side effect of this isolation level is that 443 read operations obtain read locks on database pages, 444 and then hold those locks until the read operation is 445 completed. 446 447 <span> 448 When you are using transactional cursors, this 449 means that read locks are held until the transaction commits or 450 aborts. In that case, over time a transactional cursor 451 can gradually block all other transactions from writing 452 to the database. 453 </span> 454 </p> 455 <p> 456 You can avoid this by using snapshot isolation. 457 Snapshot isolation uses <span class="emphasis"><em>multiversion 458 concurrency control</em></span> to guarantee 459 repeatable reads. What this means is that every time a 460 writer would take a read lock on a page, instead a copy of 461 the page is made and the writer operates on that page 462 copy. This frees other writers from blocking due to a 463 read lock held on the page. 464 </p> 465 <div class="note" style="margin-left: 0.5in; margin-right: 0.5in;"> 466 <h3 class="title">Note</h3> 467 <p> 468 Snapshot isolation is strongly recommended for read-only threads when writer 469 threads are also running, as this will eliminate read-write contention and 470 greatly improve transaction throughput for your writer threads. However, in 471 order for snapshot isolation to work for your reader-only threads, you must 472 of course use transactions for your DB reads. 473 </p> 474 </div> 475 <div class="sect3" lang="en" xml:lang="en"> 476 <div class="titlepage"> 477 <div> 478 <div> 479 <h4 class="title"><a id="sisolation_cost"></a>Snapshot Isolation Cost</h4> 480 </div> 481 </div> 482 </div> 483 <p> 484 Snapshot isolation does not come without a cost. 485 Because pages are being duplicated before being 486 operated upon, the cache will fill up faster. This 487 means that you might need a larger cache in order to 488 hold the entire working set in memory. 489 </p> 490 <p> 491 If the cache becomes full of page copies before old 492 copies can be discarded, additional I/O will occur as 493 pages are written to temporary "freezer" files on disk. 494 This can substantially reduce throughput, and should be 495 avoided if possible by configuring a large cache and 496 keeping snapshot isolation transactions short. 497 </p> 498 <p> 499 You can estimate how large your cache should be by 500 taking a checkpoint, followed by a call to the 501 502 <code class="methodname">DbEnv::log_archive()</code> 503 504 method. The amount of cache required is approximately 505 double the size of the remaining log files (that is, 506 the log files that cannot be archived). 507 </p> 508 </div> 509 <div class="sect3" lang="en" xml:lang="en"> 510 <div class="titlepage"> 511 <div> 512 <div> 513 <h4 class="title"><a id="sisolation_maxtxn"></a>Snapshot Isolation Transactional Requirements</h4> 514 </div> 515 </div> 516 </div> 517 <p> 518 In addition to an increased cache size, you may also 519 need to increase the maximum number of transactions 520 that your application supports. (See 521 <a class="xref" href="maxtxns.html" title="Configuring the Transaction Subsystem">Configuring the Transaction Subsystem</a> 522 for details on how to set this.) 523 In the worst case scenario, you might need to configure your application for one 524 more transaction for every page in the cache. This is 525 because transactions are retained until the last page 526 they created is evicted from the cache. 527 </p> 528 </div> 529 <div class="sect3" lang="en" xml:lang="en"> 530 <div class="titlepage"> 531 <div> 532 <div> 533 <h4 class="title"><a id="sisolation_whenuse"></a>When to Use Snapshot Isolation</h4> 534 </div> 535 </div> 536 </div> 537 <p> 538 Snapshot isolation is best used when all or most 539 of the following conditions are true: 540 </p> 541 <div class="itemizedlist"> 542 <ul type="disc"> 543 <li> 544 <p> 545 You can have a large cache relative to your working data set size. 546 </p> 547 </li> 548 <li> 549 <p> 550 You require repeatable reads. 551 </p> 552 </li> 553 <li> 554 <p> 555 You will be using transactions that routinely work on 556 the entire database, or more commonly, 557 there is data in your database that will be very 558 frequently written by more than one transaction. 559 </p> 560 </li> 561 <li> 562 <p> 563 Read/write contention is 564 limiting your application's 565 throughput, or the application 566 is all or mostly read-only and 567 contention for the lock manager 568 mutex is limiting throughput. 569 </p> 570 </li> 571 </ul> 572 </div> 573 </div> 574 <div class="sect3" lang="en" xml:lang="en"> 575 <div class="titlepage"> 576 <div> 577 <div> 578 <h4 class="title"><a id="sisolation_howuse"></a>How to use Snapshot Isolation</h4> 579 </div> 580 </div> 581 </div> 582 <p> 583 You use snapshot isolation by: 584 </p> 585 <div class="itemizedlist"> 586 <ul type="disc"> 587 <li> 588 <p> 589 Opening the database with 590 multiversion support. You can 591 configure this either when you 592 open your environment or when 593 you open your 594 595 <span> 596 database. 597 </span> 598 599 600 <span> 601 Use the 602 <code class="literal">DB_MULTIVERSION</code> 603 flag to configure this support. 604 </span> 605 606 <span> 607 Use the 608 <code class="methodname">XmlContainerConfig::setMultiversion()</code> 609 option to configure this support when you open your 610 container. To configure multiversion support when you 611 open your environment, use the 612 <code class="literal">DB_MULTIVERSION</code> flag on the 613 environment open. 614 </span> 615 616 617 </p> 618 </li> 619 <li> 620 <p> 621 Configure your <span>cursor or</span> 622 transaction to use snapshot isolation. 623 </p> 624 <p> 625 To do this, 626 627 <span> 628 pass the <code class="literal">DB_TXN_SNAPSHOT</code> flag 629 when you 630 631 <span> 632 open the cursor or 633 </span> 634 635 create the transaction. 636 637 <span> 638 If configured for the transaction, 639 then this flag is not required 640 when the cursor is opened. 641 </span> 642 643 </span> 644 645 646 </p> 647 </li> 648 </ul> 649 </div> 650 <p> 651 The simplest way to take advantage of snapshot 652 isolation is for queries: keep update 653 transactions using full read/write locking and 654 use snapshot isolation on read-only transactions or 655 cursors. This should minimize blocking of 656 snapshot isolation transactions and will avoid 657 deadlock errors. 658 </p> 659 <p> 660 If the application has update transactions which 661 read many items and only update a small set (for 662 example, scanning until a desired record is 663 found, then modifying it), throughput may be 664 improved by running some updates at snapshot 665 isolation as well. But doing this means that 666 you must manage deadlock errors. 667 See 668 <a class="xref" href="lockingsubsystem.html#deadlockresolve" title="Resolving Deadlocks">Resolving Deadlocks</a> 669 for details. 670 </p> 671 <p> 672 The following code fragment turns 673 on snapshot isolation for a transaction: 674 </p> 675 <pre class="programlisting">#include "db_cxx.h" 676 677... 678 679int main(void) 680{ 681 u_int32_t env_flags = DB_CREATE | // If the environment does not 682 // exist, create it. 683 DB_INIT_LOCK | // Initialize locking 684 DB_INIT_LOG | // Initialize logging 685 DB_INIT_MPOOL | // Initialize the cache 686 DB_INIT_TXN | // Initialize transactions 687 <strong class="userinput"><code>DB_MULTIVERSION; // Support snapshot isolation.</code></strong> 688 689 // Note that no special flags are required here for snapshot isolation. 690 // This is because it is already enabled at the environment level. 691 u_int32_t db_flags = DB_CREATE | DB_AUTO_COMMIT; 692 Db *dbp = NULL; 693 const char *file_name = "mydb.db"; 694 695 std::string envHome("/export1/testEnv"); 696 DbEnv myEnv(0); 697 698 try { 699 700 myEnv.open(envHome.c_str(), env_flags, 0); 701 dbp = new Db(&myEnv, 0); 702 dbp->open(NULL, // Txn pointer 703 file_name, // File name 704 NULL, // Logical db name 705 DB_BTREE, // Database type (using btree) 706 db_flags, // Open flags 707 0); // File mode. Using defaults 708 709 } catch(DbException &e) { 710 std::cerr << "Error opening database and environment: " 711 << file_name << ", " 712 << envHome << std::endl; 713 std::cerr << e.what() << std::endl; 714 } 715 716 .... 717 718 envp->txn_begin(NULL, txn, <strong class="userinput"><code>DB_TXN_SNAPSHOT</code></strong>); 719 720 // Remainder of program omitted for brevity. 721 722 </pre> 723 </div> 724 </div> 725 </div> 726 <div class="navfooter"> 727 <hr /> 728 <table width="100%" summary="Navigation footer"> 729 <tr> 730 <td width="40%" align="left"><a accesskey="p" href="lockingsubsystem.html">Prev</a> </td> 731 <td width="20%" align="center"> 732 <a accesskey="u" href="txnconcurrency.html">Up</a> 733 </td> 734 <td width="40%" align="right"> <a accesskey="n" href="txn_ccursor.html">Next</a></td> 735 </tr> 736 <tr> 737 <td width="40%" align="left" valign="top">The Locking Subsystem </td> 738 <td width="20%" align="center"> 739 <a accesskey="h" href="index.html">Home</a> 740 </td> 741 <td width="40%" align="right" valign="top"> Transactional Cursors and Concurrent Applications</td> 742 </tr> 743 </table> 744 </div> 745 </body> 746</html> 747