Berkeley DB 4.4.16 Change Log
Database or Log File On-Disk Format Changes:
- The on-disk log format has changed.
New Features:
- Add support to compact an existing Btree database. [#6750]
- Add support for named in-memory databases. [#9927]
- Add support for database environment recovery serialization. This
simplifies multiprocess application architectures. Add DB_REGISTER flag
to DB_ENV->open(). [#11511]
- Add utility for performing hot backups of a database environment.
[#11536]
- Add replication configuration API. [#12110]
- Add replication support to return error instead of waiting for
client sync to complete. [#12110]
- Add replication support for delayed client synchronization. [#12110]
- Add replication support for client-to-client synchronization.
[#12110]
- Add replication support for bulk transfer. [#12110]
- Add new flags DB_DSYNC_DB and DB_DSYNC_LOG [12941]
- Add DbEnv.log_printf, a new DbEnv method which logs printf style
formatted strings into the Berkeley DB database environment log.
[#13241]
Database Environment Changes:
- Add a feature to support arbitrary alignment of mutexes in order to
minimize cache line collisions. [#9580]
- Change cache regions on 64-bit machines to allow regions larger than
4GB. [#10668]
- Fix a bug where a loop could occur if the application or system
failed during modification of the linked list of shared regions.
[#11532]
- Fix mutex alignment on Linux/PA-RISC, add test-and-set mutexes for
MIPS and x86_64. [#11575]
- Fix a bug where private database environments (DB_PRIVATE) on 64-bit
machines would core dump because of 64-bit address truncation. [#11983]
- Fix a bug where freed memory is accessed when DB_PRIVATE
environments are closed. This can happen on systems where the operating
system holds mutex resources that must be freed when the mutex is
destroyed. [#12591]
- Fix a bug where the DbEnv.stat_print method could self-deadlock and
hang. The DbEnv.stat_print method no longer displays statistics for any
of the database environments databases. [#12039]
- Fix a bug where Berkeley DB could create fragmented
filesystem-backed shared region files. [#12125]
- Fix a bug where Berkeley DB stat calls could report a cache size of
0 after the statistics were cleared. [#12307]
- Threads of control joining database environments are now configured
for all of the subsystems (lock, log, cache, or transaction) for which
the environment was originally configured, it is now an error to attempt
configuration of additional subsystems after an environment is created.
[#12422]
- Fix a bug where negative percentages could be displayed in
statistics output. [#12673]
- Fix a bug that could cause a panic if the cache is filled with
non-logging updated pages. [#12763]
- Fix a bug that could cause an unreported deadlock if the application
was using the DB_DIRTY_READ flag and the record was an off page
duplicate record. [#12893]
- Fix a bug where a handle lock could be incorrectly retained during
a delete or rename operation. [#12906]
Concurrent Data Store Changes:
- Lock upgrades and downgrades are now accounted for separately from
lock requests and releases. [#11155]
- Fix a bug where a second process joining a Concurrent Data Store
environment, with the DB_CDB_ALLDB flag set, would fail. This would
happen if the first thread were not entirely finished with
initialization. [#12277]
General Access Method Changes:
- Fix a bug where filesystem operations are improperly synchronized.
[#10564]
- Add support for database files larger than 2GB on Windows. [#11839]
- Rename DB_DEGREE_2 (and all related flags) to DB_READ_COMMITTED.
Rename DB_DIRTY (and all related flags) to DB_READ_UNCOMMITTED. [#11776]
- Fix a bug where wrapping of sequences was incorrect when the cache
size is smaller than the range of the maximum value minus the minimum
value. [#11997]
- Fix a bug that could result in a hot backup having a page missing
from a database file if a file truncation was in progress during the
backup but was then aborted. [#12017]
- Fix a bug where a long filename could cause one too few bytes to
be allocated when opening a file. [#12085]
- Fix a bug in secondary cursor code if a write lock is not granted.
[#12109]
- Fix a bug in secondary cursors where the current record would change
on error. [#12141]
- Fix a bug in Db->truncate where the method was not checking to see
if the handle was opened read-only. [#12179]
- Fix a bug in sequences so that they are now platform independent,
taking into account little-endian and big-endian architectures. They
will be automatically upgraded in 4.4. [#12202]
- Fix a bug with non-wrapping sequences when initial value was
INT64_MIN. [#12390]
- Add a retry for operating system operations that return EIO (IO
Error) to better support NFS mounted filesystems. [#12426]
- Fix sequence wrapping at INT64 limits. [#12520]
- Fix a bug where errors during DB->associate could leave secondaries
half associated. [#13173]
- Fix a bug so that we no longer will update in CDS and DS if the file
size limit will be exceeded. [#13222]
Btree Access Method Changes:
- Remove maxkey configuration. [#8904]
- Fix a memory leak in operations on large Btrees. [#12000]
Hash Access Method Changes:
- Fix a bug where access to HASH or encrypted database pages might be
blocked during a checkpoint. [#11031]
- Fix a bug where recovery would fail when a database has a hash page
on the free list and that hash page was freed without using transactions
and later allocated and aborted within a transaction. [#11214]
- Fix a bug in hash duplicates where if the caller left garbage in the
partial length field, we were using it. Fix a bug where a replacement
of a hash item that should have gone on an overflow page, did not.
[#11966]
- Fix a bug where free space was miscalculated when adding the first
duplicate to an existing item and the existing item plus the new item
does not fit on a page. [#12270]
- Fix a bug where allocations of hash buckets are not recovered
correctly. [#12846]
Queue Access Method Changes:
- Improve performance of deletes from a QUEUE database that does not
have a secondary index. [#11538]
- Fix a bug where updates that do not use transactions, but do enable
locking, failed to release locks. [#11669]
- Fix a bug where a transaction might not be rolled forward if the
site was performing hot backups and an application aborted a prepared
but not committed transaction. [#12181]
- Fix a bug with queue extents not being reclaimed. [#12249]
- Fix a bug where a record being inserted before the head of the queue
could appear missing if DB_CONSUME is not specified. [#12919]
- Fix a bug that might cause recovery to move the head or tail of the
queue to exclude a record that was deleted but whose transaction did not
commit. [#13256]
- Fix a bug that could cause recovery to move the head or tail pointer
beyond a record that was aborted but was rolled backward by recovery.
[#13318]
Recno Access Method Changes
None.
C++-specific API Changes:
- Fix a bug so that a DbMemoryException will be raised during a
DB_BUFFER_SMALL error. [#13273]
Java-specific API Changes:
- Add VersionMismatchException to map the DB_VERSION_MISMATCH error.
[#11429]
- Fix a bug in Environment.getConfiguration() method in non-crypto
builds. [#11752]
- Fix a bug that caused a NullPointerException when using the
MultipleDataEntry default constructor. [#11753]
- Fix handling of replication errors. [#11822]
- Remove EnvironmentConfig.setReadOnly() method. [#11882]
- Fix a bug where prefix strings in the error handler may be
corrupted. [#11967]
- Fix a bug so that nested exceptions will appear in stack traces.
[#11992]
- Fix a bug on LogSequenceNumber objects in the Java API. [#12223]
- Fix a bug when no files are returned from a call to
DB_ENV->log_archive. [#12383]
- Fix a bug when multiple verbose flags are set. [#12383]
- Fix a bug so that an OutOfMemoryError is thrown when allocation
fails in the JNI layer. [#13434]
Java collections and bind API Changes:
- Binding performance has been improved by using System.arraycopy in
the FastOutputStream and FastInputStream utility classes. [#12002]
- The objectToEntry method is now implemented in all TupleBinding
subclasses (IntegerBinding, etc) so that tuple bindings are fully
nestable. An example of this usage is a custom binding that dynamically
discovers the data types of each of the properties of a Java bean class.
For each property, it calls TupleBinding.getPrimitiveBinding using the
property's type (class). When the custom binding's objectToEntry method
is called, it in turn calls the objectToEntry method of the nested
bindings for each property. [#12124]
- The getCause method for IOExceptionWrapper and
RuntimeExceptionWrapper is now defined so that nested exceptions appear
in stack traces for exceptions thrown by the collections API. [#11992]
- TupleBinding.getPrimitiveBinding can now be passed a primitive type
class as well as a primitive wrapper class. The return value for
Integer.TYPE and Integer.class, for example, will be the same binding.
[#12035]
- Improvements have been made to prevent the buffer used in serial and
tuple bindings from growing inefficiently, and to provide more
alternatives for the application to specify the desired size. For
details see com.sleepycat.bind.serial.SerialBase and
com.sleepycat.bind.tuple.TupleBase. [#12398]
- Add StoredContainer.getCursorConfig, deprecate isDirtyRead.
Deprecate StoredCollections.dirtyReadMap (dirtyReadSet, etc) which is
replaced by configuredMap (configuredSet, etc). Deprecated
StoredContainer.isDirtyReadAllowed with no replacement (please use
DatabaseConfig.getDirtyRead). Also note that
StoredCollections.configuredMap (configuredSet, etc) can be used to
configure read committed and write lock containers, as well as read
uncommitted containers, since all CursorConfig properties are supported.
[#11776]
- Add the protected method SerialBinding.getClassLoader so that
subclasses may return a specific or dynamically determined class loader.
Useful for applications which use multiple class loaders, including
applications that serialize Groovy-defined classes. [#12764] [#12749]
Tcl-specific API Changes:
- Fix a bug that could cause a memory leak in the replication test
code. [#13436]
RPC-specific Client/Server Changes:
- Fix double-free in RPC server when handling an out-of-memory error.
[#11852]
Replication Changes:
- Fix race condition (introduced in 4.3) in rep_start function.
[#11030]
- Changed internal initialization to no longer store records. [#11090]
- Add support for replication bulk transfer. [#11099]
- Berkeley DB now calls check_doreq function for MASTER_REQ messages.
[#11207]
- Fix a bug where transactions could be counted incorrectly during
txn_recover. [#11257]
- Add DB_REP_IGNORE flag so that old messages (especially PERM
messages) can be ignored by applications. [#11585]
- Fix a bug where op_timestamp was not initialized. [#11795]
- Fix a bug in db_refresh where a client would write a log record on
closing a file. [#11892]
- Fix backward arguments in C++ rep_elect API. [#11906]
- Fix a bug where a race condition could happen between downgrading a
master and a database update operation. [#11955]
- Fix a bug on VERIFY_REQ. We now honor wait recs/rcvd. [#12097]
- Fix a bug in rebroadcast of verify_req by initializing lp->wait_recs
when finding a new master. [#12097]
- Fix a bug by adding lockout checking to __env_rep_enter since
rename/remove now call it. [#12192]
- Fix a bug so that we now skip __db_chk_meta if we are a rep client.
[#12316]
- Fix a replication failure on Windows. [#12331]
- Remove master discovery phase from rep_elect as a performance
improvement to speed up elections. [#12551]
- Fix a bug to avoid multiple data streams when issuing al ALL_REQ.
[#12595]
- Fix a bug to request the remaining gap again if the gap record is
dropped after we receive the singleton. [#12974]
- Fix a bug in internal initialization when master changes in the
middle of initializing. [#13074]
- Fix a bug in replication/archiving with internal init. [#13110]
- Fix pp handling of db_truncate. [#13115]
- Fix a bug where rep_timestamp could be updated when it should not
be updated. [#13331]
- Fix a bug with bulk transfer when toggling during updates. [#13339]
- Change EINVAL error return to DB_REP_JOIN_FAILURE. [#12110]
- Add C++ exception for DB_REP_HANDLE_DEAD. [#13361]
- Fix a bug where starting an election concurrently with processing a
NEWMASTER message could cause the send function to be called with an
invalid eid. [#13403]
XA Resource Manager Changes:
None.
Locking Subsystem Changes:
None.
Logging Subsystem Changes:
- Add set_log_filemode for applications that need to set an absolute
file mode on log files. [#8747]
- Fix a bug that caused Not Found to be returned if a log file exists
but is not readable. [#11185]
- Removed checksum of records with an in-memory log buffer. [#11280]
- Fix a bug so that the DB_LOG_INMEMORY flag can no longer be set
after calling DB_ENV->open. [#11436]
- Fix a bug introduced after release 4.0 where two simultaneous
checkpoints could cause ckp_lsn values to be out of order. [#12094]
- Fix a bug when in debug mode and using the DEBUG_ROP which will now
log read operations in __dbc_logging. [#12303]
- Fix a bug where failing to write a log record on a file close would
result in a core dump later. [#12460]
- Fix a bug where automatic log file removal, or the return of log
files using an absolute path, could fail silently if the applications
current working directory could not be reached using the systems getcwd
library call. [#12505]
- Avoid locking the log region if we are not going to flush the log.
This can improve performance for some write-intensive application
workloads. [#13090]
- Fix a bug with a possible segment fault when memp_stat_print is
called on a temporary database. [#13315]
- Fix a bug where log_stat_print could deadlock with threads during a
checkpoint. [#13315]
Memory Pool Subsystem Changes:
- Fix a bug where modified database pages might not be flushed if
recovery were run and all pages from a database were found in the system
cache and up to date, followed by a system crash. [#11654]
Transaction Subsystem Changes:
- Add new DbTxn class methods allowing applications to set/get a
descriptive name associated with a transaction. The descriptive name
is also displayed by the db_stat utility. [#0382]
- Fix a bug where aborting a transaction with a large number of nested
transactions could take a long time. [#10972]
- Add support to allow the TXN_WRITE_NOSYNC flag to be specified on
the transaction handle. [#11151]
- Fix a bug that could cause a page to be on the free list twice if
it was originally put on the free list by a non-transactional update and
then reallocated in a transaction that aborts. [#11159]
- Remove the requirement for the DB_AUTO_COMMIT flag to make database
operations transactional. Specifying the database environment as
transactional or opening the database handle transactionally is
sufficient. [#11302]
- Fix a bug so that environments created from errant programs that
called dbp->close while transactions were still active can now be
recovered. [#11384]
- Fix a bug that caused free pages at the end of a file to be
truncated during recovery rather than placed on the free page list.
[#11643]
- Fix a bug that caused a page to have the wrong type if the truncate
of a BREE or RECNO database needed to be rolled forward. [#11670]
- Fix a bug when manually undoing a subdb create, dont try to free a
root page that has not been allocated. [#11925]
- Add a check on database open to see if log files were incorrectly
removed by system administration mistakes. [#12178]
- Fix a bug when calling DB->pget and then specifying the
DB_READ_COMMITTED (DB_DEGREE_2) on a cursor. If followed by a
DBC->c_pget, the primary database would incorrectly remain locked.
[#12410]
- Fix a bug where the abort of a transaction in which a sub database
was opened with the DB_TXN_NOT_DURABLE flag could fail. [#12420]
- Fix a bug that could cause an abort transaction that allocated new
pages to a file that were not flushed to disk prior to the abort
transaction to report out of disk space. [#12743]
- Fix a bug that could prevent multiple creates and destroys of the
same file to be recovered correctly. [#13026]
- Fix a bug when recovery previously handled a section of the log that
did not contain any transactions. [#13139]
- Fix a bug that could result in the loss of durability in
Transactional Environments on Mac OS X. [#13149]
- Fix a bug that could cause the improper reuse of a transaction id
when recovery restores prepared transactions. [#13256]
Utility Changes:
- Add utility for performing hot backups of a database environment.
[#11536]
- Change the Verify utility to now identify any nodes that have
incorrect record counts. [#11934]
- Fix a bug in the 1.85 compatibility code supporting per-application
Btree comparison and prefix compression functions. The functions would
not work on big-endian 64-bit hardware. [#13316]
Configuration, Documentation, Portability and Build Changes:
- Change the ex_tpcb sample application to no longer displays
intermediate results. It displays results at the end of the run.
[#11259]
- Change the Visual Studio projects on Windows so that each is in an
intermediate directory. [#11441]
- Fix errors in test subdb011. [#11799]
- Fix a bug that could cause applications using gcc on Power PC
platforms to hang. [#12233]
- Fix a bug where installation will fail if a true program cannot be
found. [#12278]
- Fix a bug that prevented C++ applications from configuring XA
[#12300].
- Fix a race condition in the Windows mutex implementation found on
8-way Itanium systems. [#12417]
- Add pthread mutex support for IBM OS/390 platform (z/OS or MVS).
[#12639]
- Fix a bug where the Tcl API did not configure on OS X 10.4. [#12699]
- Fix portability issues with queue or recno primary databases.
[#12872]
- Fix a bug where utility attempted to send replication message.
[#13446]