1<!--$Id: 4.6.html,v 1.38 2007/09/27 15:30:04 bostic Exp $--> 2<html> 3<head> 4<title>The Berkeley DB Package: DB 4.6.21 Change Log</title> 5<meta name="description" content="Berkeley DB: A database programmatic toolkit."> 6<meta name="keywords" content="embedded,database,programmatic,toolkit,b+tree,btree,hash,hashing,transaction,transactions,locking,logging,access method,access methods"> 7</head> 8<body bgcolor=white> 9 10<h3 align=center>Berkeley DB 4.6.21 Change Log</h3> 11<h4>Patches:</h4> 12<ol> 13 14<li>Fix a bug where mutex contention in database environments configured 15for hybrid mutex support could result in performance degradation. 16[#15646] 17<li>Fix a bug where closing a database handle after aborting a 18transaction which included a failed open of that database handle could 19result in application failure. [#15650] 20<li>Fix multiple MVCC bugs including a race which <b>could result in 21incorrect data being returned</b> to the application. [#15653] 22<li>Fix a bug where a database store into a Hash database could 23self-deadlock in a database environment configured for the Berkeley DB 24Concurrent Data Store product and with a free-threaded DB_ENV or DB 25handle. [#15718] 26<li>Fix an installation bug where Berkeley DB's PHP header file was not 27installed in the correct place. 28 29</ol> 30<h3 align=center>Berkeley DB 4.6.19 Change Log</h3> 31<h4>Patches:</h4> 32<ol> 33 34<li>Fix a bug where a client in a two-site replication group could 35become master, after failure of the existing master, even if the client 36had priority 0. [#15388] 37<li>Fix a bug where 32-bit builds on 64-bit machines could immediately 38core dump because of a misaligned access. [#15643] 39<li>Fix a bug where attempts to configure a database for MVCC in the 40Java API were silently ignored. [#15644] 41<li>Fix a bug where database environments configured for replication and 42verbose output could drop core. [#15651] 43 44</ol> 45<h3 align=center>Berkeley DB 4.6.18 Change Log</h3> 46 47<h4>Database or Log File On-Disk Format Changes:</h4> 48<ol> 49 50<li>The on-disk log format has changed. 51<li>The format of Hash database pages was changed in the Berkeley DB 4.6 52release, and items are now stored in sorted order. 53<b>The format changes are entirely backward-compatible, and no database 54upgrades are needed.</b> 55However, upgrading existing databases can offer significant performance 56improvements. Note that databases created using the 4.6 release may not 57be usable with earlier Berkeley DB releases. 58 59</ol> 60<h4>New Features:</h4> 61<ol> 62 63<li>Add support for a cursor DB_PREV_DUP flag, which moves the cursor 64to the previous key/data pair if it's a duplicate of the current 65key/data pair. [#4801] 66<li>Add the ability to set cache page priority on a database or cursor 67handle. [#11886] 68<li>Add verbose output tracing for filesystem operations. [#13760] 69<li>Port Berkeley DB to Qualcomm's Binary Runtime Environment for 70Wireless (BREW). [#14562] 71<li>Port Berkeley DB to WinCE. [#15312] 72<li>Port Berkeley DB to S60. [#15371] 73<li>Add a key_exists method to the DB handle. [#15374] 74<li>Applications may now begin processing new transactions while previously 75prepared, but unresolved, transactions are still pending. [#14754] 76<li>Significant performance improvements in the Hash access method. [#15017] 77 78</ol> 79<h4>Database Environment Changes:</h4> 80<ol> 81 82<li>Add support to close open file handles in the case of catastrophic 83database environment failure so applications that do not exit and 84restart on failure won't leak file handles. [#6538] 85<li>Replace the Berkeley DB shared memory allocator with a new 86implementation, intended to decrease the performance drop-off seen in 87database environments having working sets that are larger than the 88cache, especially database environments with multiple cache page sizes. 89[#13122] 90<li>Fix a bug that would incorrectly cause a thread to appear to be 91in the Berkeley DB API after a call to db_create. [#14562] 92<li>Allow database close prior to resolving all transactions updating 93the database. [#14785] 94<li>Fix a bug where the db_stat utility -Z flag and the statistics 95method's DB_STAT_CLEAR flag could clear mutex statistics too quickly, 96leading to incorrect values being displayed. [#15032] 97<li>Fix a bug where removal of a file after and open/close pair spanning 98the most recent checkpoint log-sequence-numbers made recovery fail. 99[#15092] 100<li>Fix a bug that could leave an environment unrecoverable if FTRUNCATE 101was not set and a roll-forward to a timestamp was interrupted between the 102truncation of the log and the recording of aborted allocations. [#15108] 103<li> Fix a bug where recovery of a rename operation could fail if the 104rename occurred in a directory that no longer existed. [#15119] 105<li>Fix a bug that could cause recovery to report a "File exists" 106error if a committed create was partially recovered by a previously 107failed recovery operation. [#15151] 108<li>Fix a bug where the DbEnv.get_thread_count method implementation was 109missing from the Berkeley DB 4.5 release. [#15201] 110<li>Fix a bug where replication operations were not reported properly 111when the DbEnv.failchk method was called. [#15094] 112<li> Fixed a bug that caused SEQ->remove not to use a transaction 113if the sequence was opened on a transactional database handle but no 114transaction was specified on the call. [#15235] 115<li>Fix a bug where accesses to the database environment reference count 116could race, causing the DB_ENV->remove method to incorrectly remove 117or not remove a database environment. [#15240] 118<li>Fix a bug that could cause a recovery failure if a partial record 119was written near the end of a log file before a crash and then never 120overwritten after recovery runs and before a log file switch occurs. 121[#15302] 122<li>Fix a bug that could fire a diagnostic assertion if an error occurred 123during a database environment open. [#15309] 124<li> Fix a bug where memp_trickle attempts to flush an infinite number 125of buffers. [#15342] 126<li>Cause application updates of the DB_ENV->set_mp_max_write values 127to affect already running cache flush operations. [#15342] 128<li>Fix a bug which could cause system hang if a checkpoint happened at 129the same time as a database file create or rename. [#15346] 130<li>Fix a bug which could cause application failure if the open of a 131subdatabase failed while other database opens were happening. [#15346] 132<li>Fix a bug that could cause recovery to not process a transaction 133properly if the transaction was started before the transaction IDs were 134reset but did not put its first record into the log until after the 135txn_recycle record. [#15400] 136<li>Fix a bug that could cause a thread in cache allocation to loop 137infinitely. [#15406] 138<li>Fix a bug that could cause recovery to report a Log Sequence Error 139on systems without the ftruncate system call where a page allocation 140occurred and the database metadata page was forced out of cache without 141being marked dirty and then had to be recovered. [#15441] 142<li>Fix a bug on systems lacking the ftruncate system call, where a page 143may be improperly linked into the free list if archive recovery was done 144in multiple steps, that is, applying additional logs to the same 145databases. [#15557] 146 147</ol> 148<h4>Concurrent Data Store Changes:</h4> 149 150None. 151 152<h4>General Access Method Changes:</h4> 153<ol> 154 155<li>Add a feature where applications can specify a custom comparison 156function for the Hash access method [#4109] 157<li>Open, create, close and removal of non-transactional databases is 158are longer logged in transactional database environments unless debug 159logging is enabled. [#8037] 160<li>Add the ability to set cache page priority on a database or cursor 161handle. [#11886] 162<li>fix a bug where the DB_ENV->fileid_reset method failed when 163called on on encrypted or check-summed databases. [#13990] 164<li>Fix a bug where the DB->fd method failed when called on 165in-memory databases. [#14157] 166<li>Fix a bug where an attempt to open a Recno database with a backing 167file that does not exist could report an error because it couldn't 168remove a temporary file. [#14160] 169<li>Reverse a change found in previous releases which disallowed setting 170"partial" flags on key DBTs for DB and DbCursor put method calls. [#14520] 171<li>Fix a bug where transactional file operations, such as remove or 172rename, could leak file handles. [#15222] 173<li>Fix a bug that could cause the in-memory sorted freelist used by the 174DB->compact method not to be freed if transaction or lock timeouts 175were set in the environment. [#15292] 176<li>Add the DB->get_multiple method, which returns if the DB handle 177references a "master" database in the physical file. [#15352] 178<li>Fix a bug that could cause an DB_INORDER, DB->get method 179DB_CONSUME operation to loop if the Queue database was missing a record 180due to a rollback by a writer or a non-queue insert in the queue. 181[#15452] 182<li>Fix a bug preventing database removal after application or system 183failure in a database environment configured for in-memory logging. 184[#15459] 185 186</ol> 187<h4>Btree Access Method Changes:</h4> 188 189None. 190 191<h4>Hash Access Method Changes:</h4> 192<ol> 193 194<li>Change the internal format of Hash database pages, storing items in 195sorted order. There are no externally visible changes, and hash 196databases using historic on-page formats do not require an explicit 197upgrade. (However, upgrading existing databases can offer significant 198performance improvements.) [#15017] 199<li>Fix a bug preventing LSNs from being reset on hash databases when 200the databases were configured with a non-standard hash function. 201[#15567] 202 203</ol> 204<h4>Queue Access Method Changes:</h4> 205<ol> 206 207<li>Fix a bug which could cause a Queue extent file to be incorrectly 208removed if an empty extent file was being closed by one thread and being 209updated by another thread (which was using random access operations). 210[#9101] 211 212</ol> 213<h4>Recno Access Method Changes:</h4> 214 215None. 216 217<h4>C++-specific API Changes:</h4> 218 219None. 220 221<h4>Java-specific API Changes:</h4> 222<ol> 223 224<li>Add a feature where an exception is thrown by the Java API, the 225Berkeley DB error message is now included in the exception object. 226[#11870] 227<li>Fix a bug which can cause a JVM crash when doing a partial get 228operation. [#15143] 229<li>Fix a bug which prevented the use of Berkeley DB sequences from 230Java. [#15220] 231<li>Fix multiple bugs where DBTs were not being copied correctly in the 232Java replication APIs. [#15223] 233<li>Add transaction.commitWriteNoSync to the Java API. [#15376] 234 235</ol> 236<h4>Java collections and bind API Changes:</h4> 237<ol> 238 239<li>Change SerialBinding to use the current thread's context class 240loader when loading application classes. This allows the JE jar file 241to be deployed in application servers and other containers as a shared 242library rather than as an application jar. [#15447] 243<li>Tuple bindings now support the java.math.BigInteger type. Like other 244tuple binding values, BigInteger values are sorted in natural integer 245order by default, without using a custom comparator. For details please 246see the Javadoc for: com.sleepycat.bind.tuple.TupleInput.readBigInteger 247com.sleepycat.bind.tuple.TupleOutput.writeBigInteger 248com.sleepycat.bind.tuple.BigIntegerBinding [#15244] 249<li>Add tuple binding methods for reading and writing packed int and 250long values. Packed integer values take less space, but take slightly 251more processing time to read and write. See: TupleInput.readPackedInt 252TupleInput.getPackedIntByteLength TupleInput.readPackedLong 253TupleInput.getPackedLongByteLength TupleOutput.writePackedInt 254TupleOutput.writePackedLong PackedInteger [#15422] 255<li>The Collections API has been enhanced so that auto-commit works for 256the standard Java Iterator.remove(), set() and add() methods. Previously 257it was necessary to explicitly begin and commit a transaction in order 258to call these methods, when the underlying Database was transactional. 259Note that starting a transaction is still necessary when calling these 260methods if the StoredCollection.storedIterator method is used. [#15401] 261<li>Fix a bug that causes a memory leak for applications where both of 262the following are true: many Environment objects are opened and closed, 263and the CurrentTransaction or TransactionRunner class is used. [#15444] 264 265</ol> 266<h4>Tcl-specific API Changes:</h4> 267 268None. 269 270<h4>RPC-specific Client/Server Changes:</h4> 271 272None. 273 274<h4>Replication Changes:</h4> 275<ol> 276 277<li>Fix a bug where transactions could be rolled-back if an existing 278replication group master was partitioned and unable to participate in 279an election. [#14752] 280<li>Add a new event when a replication manager framework master fails 281to send and confirm receipt by clients of a "permanent" message. 282[#14775] 283<li>Fix a race where multiple threads might attempt to process a 284LOGREADY condition. [#14902] 285<li>Change the DB_VERB_REPLICATION flag to no longer require the 286Berkeley DB library be built with the --enable-diagnostic configuration 287option to output additional replication logging information. [#14991] 288<li>Fix a bug with elections occurring during internal init of a 289replication client site. [#15057] 290<li>Fix lockout code to lockout message threads and API separately. 291Send indication that log requests is for internal init. [#15067] 292<li>Replication manager changed to retry host-name look-up failures, 293since they could be caused by transient name server outage. 294[#15081] 295<li>Fix a bug which led to memory corruption when the sending of a bulk 296buffer resulted in an error. [#15100] 297<li>A throttling limit of 10 megabytes is now set by default in a newly 298created database environment (see the DbEnv.rep_set_limit method). 299[#15115] 300<li>Fix a bug in ALL_REQ handling where master could get a DB_NOTFOUND. 301[#15116] 302<li>Fix a bug which could lead to client sites repeatedly but unproductively 303calling for an election, when a master site already exists. [#15128] 304<li>Modify gap processing algorithms so XXX_MORE messages ask for data 305beyond what it just processed, not an earlier gap that might exist. 306[#15136] 307<li>Fixed a bug in the ex_rep example application which could cause the 308last few transactions to disappear when shutting down the sites of the 309replication group gracefully. [#15162] 310<li>Fix a bug where if a client crashed during internal init, its 311database environment would be left in a confused state, making it 312impossible to synchronize again with the master. [#15177] 313<li>Fix a bug where election flags are not cleared atomically with the 314setting of the new master ID. [#15186] 315<li>Fix a bug which would cause Berkeley DB to crash if an internal init 316happened when there were no database files at the master. [#15227] 317<li>It is now guaranteed that the DB_EVENT_REP_STARTUPDONE event will 318be presented to the application after the corresponding 319DB_EVENT_REP_NEWMASTER event, even in the face of extreme scheduling 320anomalies. [#15265] 321<li>Fix minor memory leaks in the replication manager. [#15239] [#15256] 322<li>Fix a bug which caused the replication manager to lose track of a 323failed connection, resulting in the inability to accept a replacement 324connection. [#15311] 325<li>Fix a bug where a client starting an election when the rest of the 326replication group already had an established master could confuse 327replication management at the other client sites, leading to failure to 328properly acknowledge PERM transactions from the master. [#15428] 329<li>Add support for reporting Replication Manager statistics. [#15430] 330<li>Fix a bug where a send failure during processing of a request 331message from a client could erroneously appear to the application as an 332EPERM system error. [#15436] 333<li>Client now sets STARTUPDONE at the end of the synchronization phase 334when it has caught up to the end of the master's transaction log, without 335requiring ongoing transactions at the master. [#15542] 336<li>Fix a bug in sleep-time calculation which could cause a Replication 337Manager failure. [#15552] 338 339</ol> 340<h4>XA Resource Manager Changes:</h4> 341 342None. 343 344<h4>Locking Subsystem Changes:</h4> 345<ol> 346 347<li>Change the DB_ENV->lock_detect method to return the number of 348transactions timed out in addition to those were rejected due to 349deadlock. [#15281] 350 351</ol> 352<h4>Logging Subsystem Changes:</h4> 353 354None. 355 356<h4>Memory Pool Subsystem Changes:</h4> 357<ol> 358 359<li>Fix a bug that could cause a checkpoint to hang if a database was 360closed while the checkpoint was forcing that file to disk and all the 361pages for that database were replaced in the cache. [#15135] 362<li>Fix a bug where a system error in closing a file could result in a 363core dump. [#15137] 364<li>Fix MVCC statistics counts for private database environments. [#15218] 365 366</ol> 367<h4>Transaction Subsystem Changes:</h4> 368<ol> 369 370<li>Fix a bug where creating a database with the DB_TXN_NOTDURABLE flag 371set would still write a log record. [#15386] 372<li>Change transaction checkpoint to wait only for pages being updated 373during the checkpoint. [#14710] 374 375</ol> 376<h4>Utility Changes:</h4> 377<ol> 378 379<li>Fix a bug that prevented db_load from handling subdatabase names 380that were of zero length. [#8204] 381<li>Fix a bug where the db_hotbackup utility did not clean out and record 382the log file numbers in the backup directory when both the -u and -D flags 383were specified. [#15395] 384 385</ol> 386<h4>Configuration, Documentation, Portability and Build Changes:</h4> 387<ol> 388 389<li>Berkeley DB no longer supports process-shared database environments 390on Windows 9X platforms; the DB_PRIVATE flag must always be specified 391to the DB_ENV->open method. [#13766] 392<li>Port Berkeley DB to Qualcomm's Binary Runtime Environment for 393Wireless (BREW). [#14562] 394<li>Compile SWIG-generated code with the -fno-strict-aliasing flag when 395using the GNU gcc compiler. [#14953] 396<li>Changed include files so ENOENT is resolved on Windows. [#15078] 397<li>Port Berkeley DB to WinCE. [#15312] 398<li>Port Berkeley DB to S60. [#15371] 399<li>Add the db_hotbackup executable to the Windows MSI installer. [#15372] 400<li>Change the db_hotbackup utility to use the Berkeley DB library 401portability layer. [#15415] 402<li>Re-write the GNU gcc mutex implementation on the x86 platform to 403avoid compiler errors. [#15461] 404<li>Fix a bug with non-HFS filesystems under OS X which could affect 405data durability. [#15501] 406 407</ol> 408</body> 409</html> 410