Skip to content
Go to file
2 contributors

Users who have contributed to this file

@swaldman @wbrauneis
1601 lines (1590 sloc) 109 KB
-- Properly implement the JDBC 4.1 abort method. Thanks to Andrew Johnson for calling attention
to this issue.
-- Disabling entity expansions, as we did in v. turns out not to be sufficient to prevent all
XML-config parsing related attacks (if an attacker can control the XML config file that will be
parsed). We now make XML parsing much more restrictove by default, but allow users to revert to the
old, permissive pre- behavior by setting config property 'com.mchange.v2.c3p0.cfg.xml.usePermissiveParser'
to true. That property replaces and leaves deprecated the 'com.mchange.v2.c3p0.cfg.xml.expandEntityReferences'
property introduced on Many thanks to Aaron Massey (amassey) at HackerOne for calling attention
to the continued vulnerability of XML parsing to these kinds of attacks.
-- Address situation where a throwable during forceKillAcquires() left the force_kill_acquires flag
set to true, making it impossible for the pool to restart acquisition attempts on recovery. We
now unset the flag under any circumstance, but log interrupts or unexpected throwables, and make
a best effort to complete the intended expiration of waiting clients by throwing InterruptException
Many thanks to Stefan Cordes (rscadrde on github), Vipin Nair (swvist on github), and Łukasz Jąder
(ljader on github) for their work on this issue.
-- Address CVE-2018-20433, re liberal parsing of
XML config. By default, c3p0 no longer expands entity references in XML config files. This
behavior can be overridden via config property 'com.mchange.v2.c3p0.cfg.xml.expandEntityReferences'
by applications that understand the security concerns but wish to make use of entity references.
Thanks to user zhutougg on GitHub for calling attention to and suggesting a fix for this issue.
-- Upgrade dependency to mchange-commons-java 0.2.15, which includes support for log4j2 (implemented
in mchange-commons-java by GitHub user fireandfuel. Many thanks!
-- Fix a bug in MLog bridge to slf4j logging, in which loggability of levels of wrapped loggers
was misreported, leading to useless allocation of log Strings below the logging threshold. Grr.
[change is in mchange-commons-java 0.2.11]. Many thanks to Lewis Wong on Stack Exchange for calling
attention to this issue.
-- Embed last acquistion failure as nested Exception in CannotAcquireResourceException. Thanks to
nigam on github for this addition.
-- Implemented configuration property com.mchange.v2.c3p0.impl.DefaultConnectionTester.isValidTimeout
to define timeouts on tests based on Connection.isValid(...). Many thanks to james-hu on github
for suggesting this.
-- Added a forceSynchronousCheckins config param, which can be a significant performance boost
if no tests are performed on checkin and no long work is performed in ConnectionCustomizer.onCheckIn(...).
The parameter is particularly useful for installations in which the Thread pool is under stress,
as it permits prompt checkins without use of the Thread pool, and helps reduce Thread pool congestion.
Many thanks to Vamsi Krishna for calling attention to this issue.
-- Removed embarrassingly frivolous nested locking in C3P0Registry which provoked rare deadlocks
when users lookup extensions. Many thanks to Vlad Skarzhevskyy for calling attention to this
-- ScatteredAcquireTask now tests to see if a BasicResourcePool has already been closed before
attempting an acquisition. Thanks to kenicky on github for calling attention to this issue.
-- Fixed small typo in log message, thanks Jaran Nilsen
-- DriverManagerDataSource: Added synchronization missing for the very rare case of an on-the-fly
reset of 'driverClass'; minor optimization, properly marked driver class loaded if an instantiation
of the driver has succeeded.
-- Removed support for try-with-resources (implementation of marker interface java.lang.AutoCloseable)
as it breaks Java6 support. I'll save it for 0.9.6 (unless I get a lot of complaints).
Thanks again to dwscott on github.
-- Added config parameter that allows users to
have c3p0 generate predictable, reproducible full JMX ObjectNames (at the cost of extra user
responsibility in ensuring the uniqueness of those names). Updated and expanded documentation
with respect to JMX support. Thanks to dwscott on github for suggesting this improvement.
-- In mchange-commons-java, modified log library to consider slf4j library unfound if it gets bound to its
no-op ('NOP') back-end.
-- Gave proxies a toString() method that embeds the toString() method of the objects they are wrapping, including
Statement text if the underlying driver supports that. Thanks to Ketan Padegaonkar for the suggestion.
-- Added API [in mchange-commons-java] to make it easier for users to create a MultiPropertiesConfig object from a
Properties object, which will enable use of the new C3P0Config.refreshMainConfig( overrides, overridesDescription )
method, which may enable some useful dynamic configuarion possibilities, see e.g.
Thanks to Vlad Skarzhevskyy on SourceForge for the proposed use case.
-- Modified mchange-commons-java (and c3p0 docs) to prioritize slf4j over log4j if both are available in the application
classpath and com.mchange.v2.log.MLog has not been explicitly set. Thanks to Steve Ramage for the suggestion.
-- If a problem occurs while acquiring a cached Statement from the Statement pool, the Exception we throw includes
both the actual problem (which occurs in a helper thread) and the stack trace of the client which provoked the
failed acquisition. The problem that occurred within the helper thread is captured as the root cause Throwable.
Many thanks to Michele Rossi for the suggestion.
-- Finally wrapped StackOverflowError that comes in the rare, pathological case when every attempt to checkout
a resource fails some test or check until eventually recursive retries blow the stack. A NoGoodResourcesException
is now thrown in this case. Thanks to Ibenda on github for calling attention to the issue (and others annoyed by
it long ago). [Note: The StackOverflowError remains accessible, wrapped as the root cause Exception. So I expect
continuing bug reports / complaints about a StackOverflowException from users who don't notice it is caught and
-- Some documentation updates and fixes.
-- Made PooledDataSource AutoCloseable, so that try with resources can be used to clean up after them.
-- Added some warnings in BasicResourcePool's constructor if min, max, and start are misordered.
Misordered values are reordered. Thanks to bronzenose on github for the suggestion.
-- Overloaded AbstractPoolBackedDataSource / AbstractComboPoolBackedDataSource toString,
so we've a toString( boolean show_config ). The full reflective config is now shown only when logging
pool manager initialization. Otherwise we revert to a fast, nonreflective toString(). We'll see whether anyone
minds the abbreviation. Thanks to kibertoad and Xerosigma on github for the suggestion.
-- Fixed implementations of wrap( ... ) / isWrapperFor( ... ) to properly conform to spec when given an interface implemented
directly by the receiver of the wrap( ... ) call. Many thanks to baracil on github for calling attention to this issue.
-- Resolved a Serialization problem introduced in c3p0-0.9.5-pre7 (Broke serializability of DefaultConnectionTester
and therefore typical instances of WrapperConnectionPoolDataSource.) Many thanks to Petar Tachiev for calling attention to this issue.
-- [in mchange-commons-java] Fixed a latent issue where decoding of JNDI References would fail if the Object factory
was not loaded by the System ClassLoader. (This issue was revealed when Serialization failures provoked fallback
to a Reference Serialization strategy.) Many thanks to Petar Tachiev for calling attention to this issue.
-- Cleaned up (and exposed a configuration property to tweak) selection of Connection tests when
no query is provided. Do we use Connection.isValid(...), the traditional DatabaseMetaData-based test,
something else? Now, by default we try isValid(...) but we back-off to the traditional test on error.
But this behavior is configurable via the property 'com.mchange.v2.c3p0.impl.DefaultConnectionTester.querylessTestRunner'
(which is documented with 'Other Properties').
-- Added config parameter 'forceUseNamedDriverClass', which ensures that an instance of the class named in
'driverClass' is used as the JDBC driver, rather than simply preloading the named class and then
relying upon DriverManager to determine which driver to use. Thanks to Martin Coene and meinc on SourceForge
for calling attention to this issue.
-- We now support the parentLogger property of DataSources (rather than just throwing an SQLFeatureNotSupportedException).
This should resolve the issue identified by Maarten Coene (thanks!), where JMX clients repeatedly see
Exceptions as they try to read the parentLogger property.
-- [in mchange commons java] MLog levels WARNING and SEVERE were misordered due to bad copy/paste of
the integer value of levels. Many thanks to Martin Coene for discovering and tracking down the problem.
-- [in mchange commons java] An error in MLog logging to log4j caused an IllegalArgumentException
at log level OFF. This problem was identified and fixed by Sebastian Baumhekel
-- [in mchange commons java] Defines a jdk14logging ForwardingLogger that adapts MLog Loggers into
JDK14 (java.util.logging) Loggers
-- Unused resources cleaned up on BasicResourcePool close() are properly added to the formerResources
WeakHashMap, to prevent multiple checkins of those resources from triggering foreign resource warnings.
Thanks again to Brett Wooldridge for calling attention to this issue.
-- Guarded marking of transactions as maybe dirty in Statement proxies, so that they don't trigger NPEs
then Exceptions for illegal operation on a closed Connection for methods like Connection.isClosed.
Many thanks to Brett Wooldridge for calling attention to this issue.
-- Prevent JMX clients from updating properties on underlying pools (and therefore potentially
triggering a wind-down / reconstruction of internal pools) if the new value of the property
is the same as the original value. Thanks to ask-dev on Stack Overflow for calling attention
to this issue.
-- Guard against case where connectionCustomizerClassName is an empty or blank String.
Thanks to ask-dev on Stack Overflow for calling attention to this issue.
-- Fixed typos is docs where testConnectionOnCheckout and testConnectionOnCheckin
erroneously included excess capital letters, testConnectionOnCheckOut and testConnectionOnCheckIn.
Thanks to Usman Mutawakil for finding and reporting this issue.
-- For drivers that support JDBC4 Connection.isValid(), this becomes the default
Connection test if no preferredTestQuery / automaticTestTable is specified.
HOWEVER, for older drivers that do not support Connection.isValid(), the already
slow default test becomes EVEN SLOWER than its already legendary slowness. If
you are using a driver that you know supports Connection.isValid(), you can skip
setting preferredTestQuery or automaticTestTable. But if you are not sure, better
set one of these or endure painfully slow tests.
-- Fixed a rare problem caused when some JDBC drivers are simultaneously autoloaded
on initialization of DriverManager.class and by DriverManagerDataSource.
DriverManagerDataSource's class init now preloads DriverManager.class
-- Fixed a ClassCastException in some codepaths of BasicResourceFactory construction.
Thanks to petrjanata on SourceForge for reporting and tracking down the issue.
-- Modified mlog logging library to permit dynamic refresh if config (again, for
integration with dynamically-reloading appservers).
-- Refactored C3P0Config to permit refreshing of config information on-demand at
runtime, plus imposing overrides. (This is in preperation for integration with
application servers, whose native config files might be used to configure c3p0,
and might change more frequently than a VM or ClassLoader recycle).
-- Improvements to loading of resources for hocon/typesafe-config. We now properly
load and merge multiple resources (if found), and use typesafe conventions for
merging xxx.conf, xxx.json, and files. [mchange-commons-java]
-- Cleaned up reflection of default properties, such that only coercible properties
get Stringified (since Stringifying non-coercible properties render them useless
for setting as bean params), and to eliminate treating C3P0Defaults.getKnownProperties
as itself a default. Many thanks to wuSam on github for exercising the library in an
unusual way and revealing these issues.
-- Made JMX names of C3P0Registry customizable via the config property
'', so that multiple installations/ClassLoaders
can be managed by a single MBean server. Many thanks to Vlad Skarzhevskyy for
calling attention to this issue.
-- Fixed a rare Connection leak in WrapperConnectionPoolDataSource that occurs
when Connection acquisition succeeds but construction of a PooledConnection
to wrap it fails with an Exception. Many thanks to Edward Chan for reporting
the issue.
-- Fixed an error introduced in the 0.9.2 series, which caused MBeans
for PooledDataSources to be unregistered on close() under names different
from that under which they were registered, creating a memory leak.
Many thanks to Vlad Skarzhevskyy for reporting this issue.
-- Updated documentation with user-defined configuration extensions and
information on how to define named configuration, per-user overrides,
and user-defined extensions in all three supported config file formats.
-- Implemented user-defined configuration extensions.
-- Implemented properties file and HOCON definition of named configurations
and per-user overrides.
-- Added privilegeSpawnedThreads config parameter, mostly to detach caller's
AccessControlContext from Threads that provoke the init, to prevent references from
c3p0 Threads to app server (esp tomcat) ClassLoaders that ought to be garbage-
collectable on hot undeploy. Might also be useful for helping c3p0 run from less
trusted code, although that has not come up yet. Thanks to Claudio D'angelo for
calling attention to the issue.
-- Added new config parameter 'contextClassLoaderSource', which can be set to one of
'caller', 'library', or 'none'. By default, the Thread that provokes pool initialization
passes its ContextClassLoader to c3p0-spawned Threads, which makes it impossible to garbage
collect those classloaders in e.g. application server undeploys. Setting this parameter to
'library' (rather than the default 'caller') forces c3p0-spawned Threads to use the
ClassLoader that loaded c3p0 as the contextClassLoader.
-- clone() System.getProperties() when reading config, to avoid potential
ConcurrentModificationExceptions if other Threads are messing with System properties.
Thanks Keith Amling! [mchange-commons-java]
-- Access to EncounterCounter in C3P0ImplUtils ought to have been synchronized. It was
not. Fixed, by placing the counter underneath an sync'ed wrapper. Many thanks to
Keith Amling and Dan Berindei for calling attention to this issue.
-- For reasons that remain obscure, the published build of c3p0-0.9.5-pre2 is missing
JDBC4 methods (many thanks to edysli on github for pointing this out). Apparently
it was accidentally built against JDBC3 libraries. Added a compile-time check
for JDBC4 api in generated code to C3P0ImplUtils.
-- Divided ComboPooledDataSource into final implementation class and an abstract class
which can be extended to derive debugging filters. Created two such filters,
com.mchange.v2.c3p0.debug.AfterCloseLoggingComboPooledDataSource and
-- C3P0PooledConnectionPool destroyResource() calls are now protected by in-use resource
locks to prevent rare deadlocks during simultaneous test / destroy.
-- HOCON resources are now loaded as a single group, so that resolution of substitutions
works across multiple files.
-- Fixed bug in HOCON support where files that include substitutions like ${myScope.varName}
would throw a ConfigException.NotResolved. These substitutions are now resolved. Many thanks
to Stjepan Buljat for calling attention to this issue.
-- Support standard HOCON config files and application.json as well
as application.conf
-- Fixed ClassCastException when (now deprecated!) usesTraditionalReflectiveProxies is set to
true. Many thanks to Konrad Garus for reporting the issue.
-- NewProxyConnectionGenerator has been modified so that Connections only mark transactions
as known-resolved after the putatively resolving operation has successfully completed,
rather than when it is merely attempted.
-- With FINEST/TRACE debug logging, full JVM stack traces are now emitted during any
ThreadPoolAsynchronousRunner APPARENT DEADLOCKS, to help debug any deadlocks within the JVM.
-- Made constructors of delegated proxies package-visible rather than public. Many thanks to
Oliver Zemann for calling attention to this issue.
-- Some fixes and improvements to cfg and log libraries, so that fewer superfluous warnings
get emitted on startup.
-- Modified unwrap / isWrapperFor methods in proxies to be more permissive, to accept class
objects for superinterfaces of the wrapped interface as well as of implementation
-- Added support for SLF4J / logback logging, defined short aliases for adapters to
the various logging libraries, updated documentation with new logging features
-- Added and documented HOCON / typesafe-config libary support, available if (and only if)
the typesafe-config library is included with the application.
-- Refactored mchange-commons-java to eliminate very rare potential deadlocks under simultaneous
ClassLoading of file in cfg and log packages, and defined a facade into the cfg library use of
which should eliminate the possibility of such deadlocks if it is used assiduously.
-- actually documented 'com.mchange.v2.log.jdk14logging.suppressStackWalk'. i said i had earlier
but apparently not
-- set 'com.mchange.v2.log.jdk14logging.suppressStackWalk' to true by default in mchange-commons-java
-- Included 'com.mchange.v2.log.jdk14logging.suppressStackWalk' in c3p0 docs.
-- In mchange-commons: Added a config property for jdk14 logging users:
'com.mchange.v2.log.jdk14logging.suppressStackWalk' allows clients to speed up logging
by suppressing analyzing the stack trace for the caller. (Only the name of the logger will
be given.)
-- Modified DataSource to include an (unsupported) getParentLogger() method for JDK7 builds
-- Handle JDBC4 Connection.isValid properly
-- Lots of updates to get c3p0 to proxy and compile JDBC4
-- Removed some jdk14 support stuff from
-- Dual licensed c3p0, LGPL v.2.1 and EPL 1.0. Modified license files, source headers, etc.
-- Updated documentation re hibernate configuration. (Documentation was several years stale.)
-- Merged John Sumsion's remarkable integration of c3p0's full SourceForge history into the
git repository. (Enormous, huge thanks to John Sumsion!)
-- Added logging (at level FINE, in com.mchange.v2.c3p0.impl.C3P0PooledConnectionPoolManager) when
a pool is initialized for a new authentication within a DataSource.
-- We now allow users to place an XML formatted config file in arbitrary locations under
the classpath (i.e. as ClassLoader resource), including jar file META-INF directories.
If the config property com.mchange.v2.c3p0.cfg.xml begins with "classloader:", the remainder of
the String will be interpreted as a ClassLoader resource path. Thanks to Sam Razialruh for
suggesting this feature.
-- In DataSources.pooledDataSource( DataSource unpooledDataSource, Properties props ), switched
from using keySet().iterator() to the propertyNames() Enumeration to properly capture keys
set in the Properties object's defaults. Thanks to Jisakiel in SourceForge for calling attention
to this issue.
-- Modified DelegatorGenerator [in mchange-commons-java] to delegate reflectively at
runtime via the main delegation interface (by default, and in c3p0's proxies), rather
than via the runtime class, which may not be public. (Added a bit of API to DelegatorGenerator
so that users can specify whether reflective delegation should be via the main
delegation interface, some other expected superclass or interface, or the actual
runtime class.) Many thanks to Kiran Kollipara for pointing out the risk of IllegalAccessExceptions
under the previous implementation.
-- Added hooks [in mchange-commons-java] to com.mchange.v2.codegen.intc.DelegatorGenerator
to permit reflective delegation of methods not declared by the main interface. Added
some unproblematic JDBC4 methods to generated PreparedStatement proxies in order (hopefully)
to workaround hibernate issues. (see and )
-- Added some extra logging (at level FINER [or DEBUG] to NewPooledConnection, in hopes
that it will be helpful in tracking down Exceptions that occur during automatic cleanup
of unclosed Statements and ResultSets.
-- Added some logic that continues to enforce sequential (i.e. nonsimultaneous) use of
PooledConnections in Connection tests, without requiriing Threads to use NewPooledConnection's
lock. Clients of NewPooledConnection should not hold its lock directly, to avoid
potential deadlocks due to Connection eent handlers. [Many thanks to Jan Zuchhold
for calling attention to this problem.]
-- Added a DEBUG STACK TRACE at log level FINEST in BasicResourcePool for tracking the cause
of resource exclusion.
-- Added a file for github.
-- Updated documentation to include dataSourceName as a config parameter, and to describe
its use in generating stable JMX identifers.
-- Made dataSourceName a first-class config parameter, configurable in all the usual ways,
to help support its use for stable JMX names.
-- Included (excellent, very lightweight) patch by Bryan Varner [ ]
that includes the property dataSourceName in the registered name of c3p0
JMX mbeans. As Varner writes, "This allows you to query mbeans by stable name patterns,
allowing for reliable instrumentation via JMX between VM instances." Many thanks
to Bryan Varner.
-- Some fixes of issues with per-user config overrides, verified and
updated docs with respect to per-user configs.
-- Unsynchronized access to identityToken in the various identityTokenized
beans, marking the property volatile instead. This should resolve an
occasional deadlock. Many thanks to Phillip Henry, see
-- Fixed a bug whereunder PoolBackedDataSources threw an Exception
trying to bind per-user overrides that properly should be bound
only to WrapperConnectionPoolDataSource (or to ComboPooledDataSource,
which wraps that). Many thanks to Walter Marcelo Pedrozo Barboza
for calling attention to this issue.
-- Factored out VMID generation into mchange-commons-java.
-- Fixed broken version on Maven dependency to mchange-commons-java.
Version was hardcoded, now uses version number that c3p0 was built
-- Added workaround for bug in Maven 2.2.1 that sometimes leads to corruption
of deployed artifacts. See
-- Eliminated bead RELEASE suffix from use in staging to Maven repository,
which should fix managed dependency problems for Maven users.
-- Required tests of PooledConnections to own the locks associated with those
PooledConnections in order to prevent occasional deadlocks when PooledConnection.close()
coincides with a Connection test. Many thanks to an anonymous SourceForge user for
calling attention to this issue.
-- Added support for deployment to Maven central repository to build.xml,
and noted this version's coordinates in the repository.
-- Added a src directory to the distribution, containing jarred java
sources (including codegenerated sources) intended for debuggers/IDEs.
As a convenience, mchange-commons-java sources are included as well.
-- Fixed issue where idle Connections being tested by the pool
were not properly marked with the statement cache as "in-use".
-- Updated the build file to make it easy to update versions and dates
on source code headers.
-- ResultSets are now automatically closed and their proxies dereferenced
as soon as parent Statements are closed, rather than waiting for the parent
Connection to close as we did before. Thanks to jo-hennig and soronthar for
calling attention to this issue.
-- Modified (in mchange commons library) ThreadPoolAsynchronousRunner to
occasionally (stochastically) call Timer.purge() when maxAdministrativeTaskTime
settings leads to frequent cancellations of TimerTasks. Thanks to Kevin Conaway,
who observed memory issues from the piling up of canceled but unpurged TimerTasks.
-- Modified (in mchange commons library) VersionUtils to be more tolerant
of unusual version Strings. Thanks to martyhu for the suggestion.
-- When a ProxyConnection is marked closed, we now fire a connectionClosed
event even if an Exception occurred during cleanup, to be sure that proxies
detach. We still fire a connectionErrorOccurred prior to connectionClosed,
so that listeners (especially C3P0PooledConnectionPool) can invalidate the
can't-be-cleaned-up PooledConnection.
-- Modified GooGooStatementCache to call clearBatch on PreparedStatement checkin,
so that stale batches from prior use of Statements are never visible to new
users. Many thanks to David Carr for pointing out the problem and the fix.
-- getWarnings and clearWarnings no longer set the txn_known_resolved flag to false.
Thanks to Manuel Darveau for pointing out the unnecessary caution.
-- Gave C3P0PooledConnectionPoolManager's timer a name. Thanks to Matt Kusnierz
for the suggestion.
-- Fixed an infinite loop in ComboPooledDataSource, which occurred following
some Exceptions, which provoked calls to toString() to build the Exception
message, which call itself provoked an Exception, ad infinitum. All potentially
Exception-provoking attribute lookups are now excluded from toString(). Many
thanks to Svante v. Erichsen for calling attention to this problem and
suggesting a fix.
-- Modified the build so that JDK 1.4 compatible versions of c3p0-0.9.2
can be generated easily. (c3p0-0.9.2 will be the last version of c3p0
to support JDK 1.4 builds.)
-- Moved development onto github. See, with
a dependency against
-- Rationalized management of the ConnectionEventListener that checks in
PooledConnections on proxy Connection close(). Now added on check-out and
removed on check-in, rather than being always present but temporarily removed
during administrative tasks.
-- Implemented asynchronous Statement destroyer, hopefully completing the task of
making the statement cache robust to drivers that cannot deal with Statements
being closed underneath Connections still in use (e.g. Oracle, JTDS). Note
that this implementation requires at least one dedicated Thread for statement
destruction (if we used the common Thread pool, there will be deadlocks as
Statement close tasks await not-in-use Connections, while Connection-related
tasks can't be completed because the Thread pool is saturated with Statement
close tasks. Stats about the asynchronous Statement destroyer are available
via the JMX MBean, prefixed statementDestroyer. Setting the new config parameter
c3p0.statementCacheNumDeferredCloseThreads enables the fix.
-- Resolved a potential race condition that could lead to pool freezes, especially
when acquisitions occasionally fail and acquireIncrement is set to 1. [Many
thanks to Brendan Dougherty for carefully describing this issue.]
-- Added a guard to BasicResourcePool.doAcquire() to ensure that the pool has
not been closed or broken before assimilating a task [Many thanks to
Sean Rohead for tracking down this issue and suggesting the fix.]
-- Made SCATTERED_ACQUIRE_TASK default to true. This provides much better overall
performance for Connection acquisition from potentially unreliable sources
-- Modifed to ensure that unresolved transaction settings apply to expired
unreturned Connections. (see config params autoCommitOnClose and
forceIgnoreUnresolvedTransactions) [Thanks to Matthew Lieder for calling
attention to this issue.]
-- Cleaned up Statement cache fix that ensures no statements are closed while
their parent Connection is in use. By default, the fix is not enabled, because
it's extra work that most databases don't need. [Most drivers support Statement
close() while a parent Connection is in use, so the extra work is unnecessary.]
-- Separated traditional build into two libraries, mchange-commons and c3p0.
-- Modified GooGooStatementCache and C3P0PooledConnectionPool to ensure that
1) the Statement cache knows which Connections are currently in use by outside
clients; and 2) the Statement cache refuses to cull Statements belonging to
Connections in use. Theoretically, drivers should support asynchronous statement
close. In practice, some drivers don't do so nicely, or block pending completion
of other (potentially long) operations, leading to APPARENT DEADLOCKS. (Oracle
users in particular have reported deadlocks in which all pool threads are blocked
on Statement close tasks.) [Many, many thanks to Ovidiu Feodorov for a very detailed
account of problems that occur under c3p0 when very long queries are executed under
-- Defined a NullMLogger, and modified Log4jMLog to use that, instead of a broken
log4j MLogger, if c3p0's logging library fails to acquire a non-null Log4j logger.
Thanks to Oli Glimmer for calling attention to this issue. (Strange NPEs due to
apparently null log4j loggers have been encountered before, and previous attempts
to resolve apparently weren't adequate.)
-- Fixed a problem whereby ThreadPoolAsynchronousRunner was not robust to Errors provoked
during task execution. [Thanks to George Khoshy for reporting this issue.]
-- Made VersionUtils more robust to cases where some components of the java.version system
property are non-integral [Thanks to Dirk Weigenand for calling attention to this problem,
and suggesting a solution!]
-- Fixed a variety of latent, hidden bugs discovered by "findbugs". Many thanks to Carsten Heyl
for running this utility and reporting the issues.
-- Added lastAcquisitionFailureDefaultUser to ComboPooledDataSource's TO_STRING_IGNORE_PROPS,
as some moments' Exception doesn't belong in a DataSource's toString(), and as trying to
find this when stringifying close()ed DataSources provoked an infinte recursion (as an Exception
was triggered trying to get the last failure info, which tries to include the stringified
DataSource in its message, again provoking an Exception, ad infinitum.) Many thanks to Santiago
Aguiar for finding and diagnosing this problem.
-- Added better documentation of the com.mchange.v2.c3p0.cfg.xml configuration property.
[Thanks to Legolas Wood for calling attention to this shortcoming.]
-- Calling getHoldability() to find the default Connection holdability provokes an Error
on some DB2 drivers. Added a "careful" check which catches and works around this.
[Thanks to Lari Hotari for finding the problem, and providing a solution!]
-- Refactored NewPooledConnection to not call event multicaster methods while holding
the NewPooledConnection's lock. Holding locks while firing events is an old-fashioned
way of provoking deadlocks, and it turns out c3p0 was not immune. Whoops! Another
embarrassing one. Many thanks to Rhythm Tyagi and Pappu Sharadavani for finding this
-- Marked variable "logger" in Log4jMLogger. Rare null pointer exceptions apparently
result from a stale value of the previously set variable. Did the same for
Log4jMLogger, although no analoguous issue has thus far been reported. [Thanks to
Noa Resare and Denis on sourceforge for calling attention to this issue.]
-- Hid the attribute "properties" from access via JMX, since it often contains password
information. I don't think properties is particularly useful via JMX, and this seems
like a better solution than returning properties with a hidden/modified password
property, since users might break things unintuitively by resetting a property to
a value that, according to the mbean, was the original value. [Thanks to John Sumsion
for calling attention to this issue.]
-- Fixed bug whereby setting any positive unreturnedConnectionTimeout forced c3p0's default
effectivePropertyCycle down to 1 second. [Many thanks to Luke Dang for tracking down
this bug.]
-- Fixed an embarrassing bug whereby c3p0 was reflectively testing for methods (setHoldability
and setReadOnly) whose argument types are primitive int and boolean, by looking for
Integer.class instead of int.class and Boolean.class instead of boolean.class. [Many
thanks to John Kristian for tracking down this subtle, and embarrassing issue very
precisely and suggesting the fix.]
-- Modified logging of uncleared SQLWarnings, so that warnings are emitted at INFO rather
than WARNING, and via a dedicated logger [com.mchange.v2.c3p0.SQLWarnings], so that users
can easily turn this behavior off. [Thanks to Oleksandr Alesinskyy for calling attention
to the annoyingness of unconditionally logging all of this at WARNING.]
-- Fixed a NullPointerException in DoubleWeakHashMap that showed up thanks to the newly
implemented canonicalization of Jdk14MLog instances. [Many thanks to Carlos Cajina for
finding and reporting this bug.]
-- Modified Jdk14MLog to canonicalize named MLogger instances. In doing so, put the getLogger()
method behind a synchronization barrier, which hopefully will resolve an odd visibility issue
bug reported by a user (wherein the underlying jdk Logger appeared occasionally to be null
in threads that might have initialized prior to the logger) [Thanks to Denis on Sourcefore
for reporting this subtle and interesting issue.]
-- Added a lastAcquisitionFailure property to PooledDataSource (and the JMX view thereof).
Thanks to Troy Whittier for the suggestion, and the patch!
-- SQL Warnings present at Connection acquisition, or uncleared upon Connection checkin
are now logged (at WARNING level) and cleared prior to being made available as part
of the pool. This is necessary to sustain the invariant that all pooled Connections
are identical and interchangible; warnings are Connection state that must be managed
and homogenized. [Thanks to Naxe on the hibernate forums for calling attention to this
-- Resolved a rare NullPointerException which occurs when a Statement cache is closed
beneath an active, checked out Connection. A client tries to prepare a Statement,
but the closed Statement cache throws an NPE. Resolved this by having the closed
Statement cache throw a ResourceClosedException, and having c3p0's proxy Statement
recover by generating an uncached PreparedStatement. [Thanks to Sven / ford_perfect
on sourceforge for finding and calling attention to this issue.]
-- Added c3p0 version to C3P0RegistryManagerMBean [Thanks to Sven / ford_perfect
on sourceforge for the suggestion.]
-- Documentation updates, and style-sheet fixes for FireFox.
-- Despite the ugliness, added logic to prepend a VMID to identity
tokens generated for c3p0 DataSources. See the docs (Appendix A,
other configuration properties) for more information. Though there's
little harm in the longer, uglier identityTokens, users who dislike
them can turn them off using the property com.mchange.v2.c3p0.VMID.
-- Wrote ScatteringAcquireTask, which represents a significant change to
how c3p0's underlying resource pool deals with Connections acquisition
failures. c3p0 is designed to retry Connection acquisition up to
"c3p0.acquireRetryAttempts" times (30 by default) with a delay of
"c3p0.acquireRetryDelay" milliseconds (1000 by default) between tries.
Should no Connection be successfully acquired after acquireRetryAttempts,
c3p0 gives up, logs an error, and frees clients waiting to check out
a Connection with an Exception (declaring the pool permanently broken
iff c3p0.breakOnAcquireFailure is set to true).
Connection acquisition is performed asynchronously by c3p0's thread
pool. Under the old implementation, when database Connections could
not be acquired, a single task dispatched to the thread pool would
hog a thread for the full length of the cycle -- 30 seconds by
default -- spending most of that time sleeping between retries.
Under the new implementation, each acquisition attempt is its own
separate task, and if an acquire attempt fails, a timer is used to
schedule a retry adter acquireRetryDelay without hogging any thread.
I think the new implementation is unambiguously superior to the old,
as it permits clients to use smaller thread pools (c3p0.numHelperThreads)
and diminishes the likelihood of apparent thread pool deadlocks.
However, it is too significant a change to introduce this late
in the "mature" c3p0-0.9.1 series, so the old implementation is enabled
by default until the c3p0-0.9.2 development cycle begins.
in a file or as a system property. Users are strongly
encouraged to set this property. You'll see better resource utilization,
and I'll get to hear issue reports if I've screwed anything up in
the implementation.
-- Modified BasicResourcePool so that checkin (and removal) of resources
is always synchronous if the pool is marked broken. [Otherwise the
check-ins to a closed or broken pool may fail, as they try to use an
already close()ed asynchronous task runner.]
-- Modified BasicResourcePool.AcquireTask so that an interrupt always
forces the whole task to end, not just a single acquisition
-- Exposed numThreadsAwaitingCheckout in PooledDataSource interface
and its associated MBean. [Thanks to John Sumsion for suggesting
this property.]
-- Prevented password from being exposed via
DynamicPooledDataSourceManagerMBean. [Thanks to John Sumsion
for calling attention to this oversight!]
-- Fixed an error whereby DataSources.pooledDataSource( ... ) methods
failed to properly set configuration properties defined on
PoolBackedDataSource [e.g. numHelperThreads]. Many thanks to
John Sumsion for noticing the problem and sending the fix!
-- Fixed AbstractConnectionCustomizer to implement (with no-op) methods
the current definition of the ConnectionCustomizer interface. (Failed
to update after modifying an early version of this interface.)
-- Modified NewPooledConnection to only collect stack trace information
on close() when we are debugging at FINEST or DEBUG (log4j) level.
Previously, stack trace information was collected anytime a
c3p0 PooledConnection was close()ed. This was unnecessary, although
the performance cost was probably small to negligible, since
PooledConnections are typically reused many times before they are
close()ed. Stack trace information is still captured when
PooledConnections are destroyed due to an error, rather than an
ordinary call to close().
-- Hid the getLastXXXFailure(user, password) from the operations list of
DynamicPooledDataSourceManagerMBean. The stack traces of failures
are visible via the sampleLastXXXFailureStackTrace(); there's no
need to expose the Throwbales as well via JMX.
-- Added convenience methods getNumPooledDataSources() and getNumPoolsAllDataSources()
to C3P0Registry and the related MBean. [Thanks to lukeda on hibernate
forums for calling attention to the need for a pool count, since
DataSources can wrap multiple pools associated with different
-- Eliminated the "trial Connection acquisition" added to c3p0-0.9.1-pre10
when PooledDataSources construct the pool with their default authentification
information. Effectively we trust that the default auntentification is
good, while we distrust client-specified (or JMX-queried) username/password
combinations, and perform the extra check. This is a compromise, as users
who frequently create and destroy DataSources don't like the extra overhead
of the trial Connection acquisition, but we really want to avoid the initialization
of bad pools when users misspecify authentification information. We've also prevented
mere queries of pool status from triggering the initialization of a
non-default-authentification pool. Only calls to getConnection(user, password)
will provoke the initialization of a non-default-auth pool. Querying
a non-default pool -- e.g. getNumBusyConnections( "steve", "test" ) -- will
throw an Exception if no Conections have ever been requested for ("steve", "test")
rather than initializing a new pool. [Thanks to Jiji Sasidharan for calling
attention to this issue.]
-- Finally rationalized C3P0Registry's Object retention policy. C3P0Registry
retains references to all C3P0 DataSources, to provide a central
point of client access for users, and to ensure that multiple JNDI
lookups within a single VM result in the same DataSource instance.
This is both intuitive, and avoids multiplying Threads and pools.
In previous versions of c3p0, hard references were retained indefinitely.
Closed DataSources hold no non-memory resources, and are small Objects,
but in unusual scenarios wherein very many c3p0 DataSources are created
and destroyed within the lifetime of a single VM, this amounts to a
memory leak. C3P0Registry now retains hard references to DataSources
so long as they have not been close()ed (or DataSources.destroy()ed).
When they have been close()ed, only weak references to c3p0 DataSources
persist, and they are eligible for garbage collection. [Thanks to
Jeremy Grodberg for calling attention to this as a real-world issue.]
-- Removed some of the new pool stat params which distractingly appear in
ComboPooledDataSource's toString() method (and c3p0 pools' config dump).
-- Exposed statement pool stats via PooledDataSource interface and MBean.
Added accessors to the statement cache, pool, pool manager, and
-- Fixed a bug whereby multiple near-simultaneous Connection requests failed
to provoke sufficient pool growth/Connection acquisition. This is a bug
reintroduced in c3p0-0.9.1-pre4, and is a duplicate of an early issue,
resolved under the old pool size management scheme. c3p0 must (and does)
keep track not only of its current size and pending growth, but also the
number of potentially unserved recent Connection requests. c3p0-0.9.1-pre4
thru c3p0-0.9.1-pre9 failed to make use of the incoming request count when
deciding whether and how much to grow the pool. Many thanks to John Kristian
for calling attention to this subtle issue.
-- Exposed information about Connection test failures (failure count of
each type of test, stack trace of last failure and last failure of each
type) via the PooledDataSource interface and its assiciated MBean.
-- Fixed a rare situation where a Connection test on check-out could
be performed by a Thread holding the pool's lock, if an initial
attempt to checkout a Connection failed and a second checkout attempt
is initiated internally. The recursive checkout is now performed
without the pool's lock.
-- Established a consisent policy re logging of Connection test failures.
They are now logged at FINE level (DEBUG under log4j). Previously tests
on checkin and checkout were logged at FINE, but idle tests were logged
at INFO. Is it worth letting this be configurable, as some users may prefer
to see test failures under normal operation?
-- Modified C3P0PooledConnectionPoolManager to perform a trial Connection
acquisition before trying to establish a pool, as pools established
on bad database or authentification information churn indefinitely trying
to acquire minPoolSize Connections, clogging up the thread pool and
sometimes leading to "APPARENT DEADLOCKS" if users touch a pool with
bad authentification information.
-- Defined UnifiedConnectionTester (and AbstractConnectionTester) so that
ConnectionTesters can provide any Exception associated with a failure,
which will show up as the "root cause" of the Exception eventually
thrown to signal the failure.
-- Added logging at WARNING level when a Connection tester reports the
database is invalid, triggering a reset of the Connection pool.
-- Updated the documentation. It is now fully current.
-- Changed the name of the jboss configuration mbean from
com.mchange.v2.c3p0.mbean.C3P0PooledDataSource to com.mchange.v2.c3p0.jboss.C3P0PooledDataSource,
in order to distinguish the jboss configuration stuff from the normal management
beans, which really are quite different. The old com.mchange.v2.c3p0.mbean.C3P0PooledDataSource
still exists, so that existing users don't see anything break, but only the new version will
be updated to support new parameters.
-- Added a means by which users can suppress JMX MBean registration if they
wish. Setting the property
to will
prevent mbean registration. (Advanced users can also use this to install
their own implementations of the ManagementCoordinator interface, though
I doubt many people will find this useful.)
-- Fixed infinitely loop in ComboPooledDataSource.setFactoryClassLocation( ... ) [introduced with
the reorganization of ComboPooledDataSource to inherit from AbstractPoolBackedDataSource (?)]
-- Added some code to warn on common spelling errors of
and c3p0-config.xml (Users have frequently mistaken the digit 0 for
the letter O in c3p0.)
-- Wrote a DynamicMBean for PooledDataSources. It seems to work,
covers both ComboPooledDataSource and "wrapping" PoolBackedDataSources,
and should cover new config parameters and accessors automatically.
-- More refinements to connection test logic. We test proxy Connections
when there's a statement cache and a test query (to save time, since
the test query is likely cached), we test physical connections if we'll
need to do the slow, default test (statement caching doesn't help), or
if there is no statement cache.
-- Removed the stub servlet I'd previously added. Although it jdk1.5+, the
JMX approach is a better one for monitoring and modifying c3p0 pools at
runtime, and I'm going to focus on that for now. (If there's a lot of
clamor for it, I'll add back and flesh out the servlet. It just doesn't
seem like a great use of my time for now.)
-- Modified build to create a separate jdk1.3 compatable jar by stripping
source of lines beginning with "assert". I want to start using assertions,
but will keep them to recognizable, strippable one-liners.
-- Modified DriverManagerDataSource to handle its driverClass parameter
in a less fragile way. Rather than attempting to load the driverClass on
property set, DriverManagerDataSource now lazily tries to load the class
upon its first call to getConnection() after the property is initialized
or updated.
-- Fixed bug whereby jdbcUrl and driverClass were not initialized properly
if set in config files rather than in code. [Thanks to Felipe Dominguez
for calling attention to this problem!]
-- Changed AbstractPoolBackedDataSource from a non-public class in
com.mchange.v2.c3p0 to a public class in com.mchange.v2.c3p0.impl to
workaround fragilities associated with sun bug id #6342411, wherein
reflective invocation of public methods on a non-public class fail,
even when involed via a public base-class.
-- Fixed NullPointerException (introduced on -pre7) on initializing with
a named config.
-- Continuing along the same lines, modified C3P0Registry and
C3P0ManagementCoordinator to unregister (both from C3P0Registry and
associated MBeans on DataSource close. When all PooledDataSources
are unregistered, the C3P0RegistryManager MBean unregisters as well.
-- Modified ActiveManagementCoordinator to check for, and unregister, the
C3P0Registry mbean prior to registration, to prevent conflicts on
undeploy/redepoloy cycle. [Thanks to Ben Groeneveld for calling attention
to this problem, and suggesting the fix.]
-- Modified C3P0PooledConnectionPool to perform Connection tests on the raw
physical Connection rather than the proxy Connection, when possible (i.e.
when we're using a C3P0 PooledConnection implementation that we know how to
get underneath of). This is better because 1) it's slightly faster; 2) it's
significantly faster on failure, as C3P0 PooledConnections test Connections
after a failure with statusOnException(), leading to an unnecessary
double-check; and 3) authors of custom ConnectionTesters can use vendor-specific
API to test Connections if they wish.
-- Defined the ConnectionCustomizer interface as a hook through which clients
can "set up" newly acquired or checked out connections (and clean up prior
to check-in or destruction). Added config param connectionCustomizerClassName.
-- Formerly unused config parameter propertyCycle now controls how frequently
c3p0 checks timestamps against expiration parameters (maxIdleTime,
maxIdleTimeExcessConnections, maxConnectionAge, unreturnedConnectionTimeout)
-- Added config parameters unreturnedConnectionTimeout and
debugUnreturnedConnectionStackTraces, reluctantly but by popular demand.
If unreturnedConnectionTimeout is set, Connections will be automatically
destroyed and removed from the pool if not returned in the given number
of seconds. If debugUnreturnedConnectionStackTraces is also set, stack
traces will be captured on check-out of any Connection, and the stack traces
that were unreturned and cleaned up by the pool will be logged. Any clean-up
of an unreturned Connection is logged at INFO level, in hopes of annoying
people into cleaning up their code and close()-ing their Connections reliably.
-- Added config parameter maxIdleTimeExcessConnections. Connections in excess of
the pool's minimum size will be expired out if they remain idle for longer
than maxIdleTimeExcessConnections. [Thanks to Navin M. and Ben Groeneveld for
the suggestion.]
-- Added configuration parameter maxConnectionAge to limit the absolute age
of Connections (age in seconds from acquisition, rather than in seconds
form last check-in).
-- Put more status accessors in ThreadPoolAsynchronousRunner and exposed
thread pool state via JMX (PooledDataSourceManagerMBean).
-- Preliminary support for management via mbeans/jmx under jdk1.5+ is
implemented. (These mbeans are distinct from the mbean designed for
configuring c3p0 under jboss, and are in the package Currently, pool statistics are instrumented,
as well as pool reset operations, but viewing and modifying the pool
configuration parameters is not supported. (I'll write a dynamic mbean to
capture this stuff; there are too many config properties I'm already
updating by hand in too many places. Later.) [Thanks to Ben Groeneveld for
the suggestion, including some nice sample code!]
-- Added configuration parameter maxAdministrativeTaskTime, which forces
interrupt() to be called on pool administrative tasks (Connection acquisition,
destruction and testing; prepared statement destruction) if these tasks seem
to hang. This is one strategy for dealing with database that hang and lead to
deadlock messages and eventually large memory footprints. It's useful only if
the hanging operations respond to interrupt()s, not thread-monitor related
deadlocks (which hopefully don't ever happen). c3p0 should recover from any
Exceptions provoked by the interrupt() calls. (Hopefully improvements to
GooGooStatementCache have rendered this parameter largely unnecessary by
resolving the most common source of deadlocks.)
-- Major revision to GooGooStatementCache that may eliminate most APPARENT
DEADLOCK messages. c3p0-0.9.0 introduced a complicated implementation
of closeAll(), in order to ensure that Statement and Connection close did
not happen simultaneously. Upon review, there was a significant bug in the
implementation, and even when corrected, the strategy is itself prone to
deadlocks should the ThreadPool become saturated with destroy tasks, which
block awaiting statement remove tasks. Rewrote closeAll() as a simple
synchronous method, rather than simulating synchrony with wait()/notifyAll(),
but with care not to hold the StatementCache's lock during Statement.close()
calls. There is a slight loss of asynchrony, as Statement.close() methods
are executed sequentially rather than simultaneously, but as a practical
matter, since the sequential destruction occurs asynchronously in the
thread pool and is invisible to clients, both client-observed and over-all
performance should not suffer. (It may even be improved, as the 0.9.0.x
strategy was complicated and had some overhead.) Note: This revision should
also eliminate occasional ConcurrentModificationExceptions associated
with the Statement cache.
-- Changed asynchronous tasks implemented as anonymous inner classes to named
inner classes, so that users reviewing active and pending tasks (usually
in trying to make sense of an APPARENT DEADLOCK) have a better idea of
what's going on.
-- Major revision to GooGooStatementCache that may eliminate most APPARENT
DEADLOCK messages. c3p0-0.9.0 introduced a complicated implementation
of closeAll(), in order to ensure that Statement and Connection close did
not happen simultaneously. Upon review, there was a significant bug in the
implementation, and even when corrected, the strategy is itself prone to
deadlocks should the ThreadPool become saturated with destroy tasks, which
block awaiting statement remove tasks. Rewrote closeAll() as a simple
synchronous method, rather than simulating synchrony with wait()/notifyAll(),
but with care not to hold the StatementCache's lock during Statement.close()
calls. There is a slight loss of asynchrony, as Statement.close() methods
are executed sequentially rather than simultaneously, but as a practical
matter, since the sequential destruction occurs asynchronously in the
thread pool and is invisible to clients, both client-observed and over-all
performance should not suffer. (It may even be improved, as the 0.9.0.x
strategy was complicated and had some overhead.)
-- Users still occasionally report seeing "APPARENT DEADLOCK" messages
from ThreadPoolAsynchronousRunner. Previously, most reports had to
do with cached PreparedStatements that hang when closed. Previous
fixes (in 0.9.0) seemed to resolve these problems. Most new reports
have to do with Connection acquisition or Connection close() tasks
hanging indefinitely, with statement cacheing not necessarily involved.
It's unclear whether these reports stem from a c3p0 bug, or rare
situations in which various JDBC driver hangs hang c3p0. (If the
condition is temporary, c3p0 recovers after the "APPARENT DEADLOCK"
warning. In any case, ThreadPoolAsynchronousRunner has been modified
to provide much better debug information, particularly in jdk1.5
environments, where hung Thread stack traces are dumped.
-- Modified the build file and refactored C3P0Config / C3P0ConfigUtils / C3P0ConfigXmlUtils
to make sure that c3p0 still runs under jdk1.3, even though it only builds
under jdk 1.5. JMX support is 1.5 only, and XML config depends on the
availability of standard XML extensions under jdk 1.3. (This is going to get harder,
as I hope to start using assertions. For the moment, assertions are
commented out. I think I'll have ant filter away the assertions for 1.3
builds and leave them in otherwise, based on a build property.
-- Significantly reorganized (hopefully simplified) of C3P0Registry and
ComboPooledDataSource. Much of the complication of all this is supporting
the most annoying, least used, but nevertheless very important feature
of supporting JNDI lookups across JVM boundaries via Serialization or JNDI
-- Reorganized ComboPooledDataSource to inherit from PoolBackedDataSource.
In doing so, changed setConnectionPoolDataSource() of PoolBackedDataSource
to throw a PropertyVetoException. (It is a "constrained property" now.)
The very, very few users who call this method directly may have to update
their code to handle the potential Exception. This change does not affect
compatability for clients that use ComboPooledDataSource or who create
PoolBackedDataSources via the c3p0's factories (e.g. DataSources).
-- Fixed bug whereby config parameter breakAfterAcquireFailure did not take
effect at the ResourcePool level.
-- Fixed bug whereby initialPoolSize parameter failed to take effect at the
ResourcePool level.
-- Fixed a bug whereby Connection tests on checkout held the pool's lock,
preventing other activity during the Connection test. [Thanks to John
Sumsion for calling attention to this subtle problem.]
-- ThreadPoolAsynchronousRunner has seen some minor efficiency improvements
(a task that calls interrupt() long-running tasks on post-deadlock blocked,
replaced threads no longer runs when there are no post deadlock blocked,
replaced threads).
-- Previous versions used a Stringified identityHashCode() as a unique ID
for various C3P0 objects. This was not correct (and bit at least one
user in practice), as identityHashCode()s are not guaranteed unique,
and in practice are not on 64-bit VMs. c3p0 now checks identityHashCodes
to make sure they are not reused, and if seen before appends a count to
keep unique IDs unique. [Many thanks to Prishan Makandurage for calling
attention to this issue.]
-- Fixed fact that statusOnException() in NewPooledConnection() ignored
preferredTestQuery parameter, always using the very slow default
connection test rather than the user-defined test. [Thanks to Andrea
Luciano for calling attention to this issue.]
-- C3P0PooledConnection (part of the "traditionalReflectiveProxies"
codebase that's largely been superceded, but that is still nominally
supported) held a static reference to a ClassLoader, in some cases
preventing the ClassLoader from being garbage collected (for example
when webapp contexts, with context specific Classloaders are
undeployed by some app servers). This was entirely unnecessary, as
the cached reference was used exactly once in the lifecycle of the
class. The reference to the ClassLoader is no longer stored. [Many
thanks to Michael Fortin for very specifically tracking down this
subtle problem!]
-- Added logic to log individual acquisition failures at INFO level
if acquireRetryAttempts <= 0. if num_acq_attempts <= 0, we try to
acquire forever, so the end-of-batch log message below will never
be triggered if there is a persistent problem so in this case,
it's better flag a higher-than-debug-level message for
each failed attempt. [Thanks to Eric Crahen for calling attention
to this issue.]
-- Wrote a stub of a status-monitoring servlet. Not yet documented, or
even usefully functional, but hopefully will be soon.
-- Updated documentation to include new config file format, including
named and per-user config. Docs could still use some work, but the
functionality is now described.
-- For consistency, the parameters user and password are now configurable
via config files and System properties.
-- Modified DefaultC3P0ConfigFinder to properly give greater precedence to
System properties than to XML-defined unspecified user, default config
-- Added WARNING logging of create SQL and original database Exception when
creation of an automaticTestTable fails. [Thanks to Alexander Grivnin.]
-- Added logic so that the Exception assoicated with the last failed acquisition
attempt in a round of attempts is logged along with the failure if the
acquisition attempts. [Thanks to Barthel Steckemetz and Patrick Eger for
calling attention to this issue.]
-- Fixed a problem whereby modifying the config of a ComboPooledDataSource
programmatically, while in use, causes previously checked-out Connections
to be close()ed underneath the user. Changing configs midstream still
causes a complete resetting of the pool (because the pool holds most
config params as immutable to avoid having to synchronized for every
config param read), but old Connections from superceded pools will remain
valid until they are rechecked into the pool. [Thanks to hhu for noticing
this problem!]
-- Fixed a resource leak whereby Connections which could not be reset after
client-initiated close() were properly "excluded" from future inclusion
in the pool, but never checked back into the pool to be destroyed. (Excluded
resources are simply marked for destruction rather than reassimilation on
checkin.) [Many thanks to Levi Purvis for tracking this down!]
-- "test-classpath" was erroneously specified in build.xml to include some
extra jar files, which rendered undetectable (to my tests) the fact
that com.mchange.v1.lang.BooleanUtils was not jarred up into
c3p0-0.9.1-pre5.jar as it needed to be. Very, very embarrrassing. [Thanks
to Goksel Uyulmaz for calling attention to this gaffe.]
-- reorganized synchronization of getConnection() in DriverManagerDataSource,
so that a lock-up in one Connection acquisition attempt does not prevent
other threads from accessing the DataSource. [Thanks to Fabian Gonzalez
for calling attention to this occasional deadlock.]
-- added overrideDefaultUser and overrideDefaultPassword parameters, and
a pooledDataSource( ) factory method which specifies override
authentification info, regardless of the user and password properties
set on the DataSource.
-- Reimplemented resource acquisition in BasicResourcePool so that each
Connection acquisition is a separate asynchronous task, to avoid tying
up the pool with very long, multiple acquisition tasks. Ended up being
a total rewrite of pool size management. Pools should be much more
conservative in acquiring new Connections. They won't grow to maxPoolSize
as easily as before. This is a big change. We'll see how it holds up as
people beat the crap out of it.
-- Added some extra logging of resource destruction, to help debug cases
where the resources are removed from the pool, but appear not to have
been properly cleaned up.
-- Finished implementation of new C3P0Config, added constructor methods to
various DataSource implementations that accept a configName parameter for
named configurations, and modified the DataSources class to use the new
configuration approach instead of PoolConfig wherever possible.
-- Continued implementation of new C3P0Config, including XML parsing
and setting of default values. Still have to test XML config, and
implement named and per-user configs.
-- Changed policy of BasicResourcePool to keep accidentally overacquired
resources so long as the pool is kept below maxPoolSize, rather than
immediately discarding Connections beyond the number we intended to
-- Fixed two issues whereby checking in resources to closed pools (both
Connection and Statement pools) provoked Exceptions. Check-ins should
succeed, even to close()ed pools, with resources silently destroyed if
the pools ae closed. [Thanks to Chris Kessel for finding both of these
-- Modified to a one-statement-cache-per-pool model, rather than a
global-statement-cache-for-all-pools, consistent with the general
philosophy that config parameters are on a per-auth-pool basis.
-- Fixed a deadlock which could occur if a pool is closed at the same time as
one of its Connections is closed. One thread gets the pool's lock, then tries
to get a PooledConnections lock to close it. An ordinary client closes a
Connection, which acquires the PooledConnection's lock, and then tries to
check back in to the pool, requiring the pool's lock. Broke the deadlock by
making the destruction of resources on pool close asynchronous, by a Thread
without the resource pool's lock.. [Many thanks to Roger Westerlund for
sending a full thread-stack dump capturing this deadlock!]
-- Completed transformation to one C3P0PooledConnectionPoolManager per
PoolBackedDataSource model.
[This is a half-assed internal release... it compiles and works, but mostly
it's just a checkpoint as c3p0 undergoes significant transformations.]
-- Deprecated PoolConfig, and began implementation of new C3P0Config approach
to configuration.
-- Partial transformation of C3P0PooledConnectionPoolManager from shared-manager
for several DataSources model to simple one-pool-per-DataSource model enabled
by canonicalization of DataSources.
-- In a subtle but major modification of BasicResourcePool, changed checkin/checkout
behavior from FIFO to LIFO to increase the likelihood that unnecessary Connections
pooled at peak times idle long enough to expire. [Many thanks to Vlad Ilyushchenko
and hontvari on sourceforge for noticing this important issue and suggesting the
-- Added code to mbean C3P0PooledDataSource to automatically create JNDI
subcontexts if a JNDI name is specified for which the full path has not
been created (analogous to mkdir -p in UNIX). [Thanks to David D. Kilzer
for noticing the issue, and for submitting the fix!]
-- Added logic to GooGooStatementCache to ensure that multiple threads do not
try to close the same PreparedStatement, in order to try to resolve longstanding
PreparedStatement.close() freeze issues. [Thanks to cardgames on sourceforge for
noticing that the freezes seem to occur when several threads are trying to close()
just one statement!]
-- Added some checks in DriverManagerDataSource (as well as WrapperConnectionPoolDataSource)
so that a mismatch between JDBC URL and driver does not lead to null Connection
references and downstrem NullPointerExceptions. [Thanks to Richard Maher for
calling attention to this issue.]
-- Undid some unnecessary synchronization in WrapperConnectionPoolDataSource,
which would cause deadlocks when the acquisition of a Connection from the
underlying driver froze for some Thread. (Other threads would then block
trying to access the WrapperConnectionPoolDataSource.) [Thanks to Fabian
Gonzales for calling attention to this issue, and for providing a VM thread
stack dump to help chase it down!]
-- modified BasicMultiPropertiesConfig to never attempt to read the resource "/",
representing (in this case) System properties, as a stream, which under some
classloaders caused ClassCastExceptions. [Thanks to Joost Schouten for calling
attention to this issue.]
-- fixed initial check of the readOnly and typeMap properties of Connections,
so that users don't see disconcerting Exceptions at debug log levels if their drivers don't
support these parameters.
-- Added jar attributes Extension-Name, Specification-Vendor, Specification-Version,
Implementation-Vendor-Id, Implementation-Vendor, and Implementation-Version
[per request of Pascal Grange].
-- Added some extra debugging information for peculiar situation where an IdentityTokenized
is constructed and registered with C3P0Registry, but coalesces to a different Object.
This should never happen, as freshly constructed IdentityTokenizeds should have String
versions of their identity hashcodes as tokens, and upon their registration, no copies
should yet have been created and registered. [Thanks to Jose Rodriguez for noting this
-- added debug-level logging of connection tests and results
-- tightened up means by which the readOnly and typeMap properties of Connections are reset,
so that users don't see disconcerting Exceptions at debug log levels if their drivers don't
support these parameters.
-- fixed problem in BasicResourcePool that left the immediately-close-checked-out-connections option
in the close() method ineffective, meaning Connections checked-out from a pool would
leak unless clients remembered to check them into pool for destruction.
-- fixed bug wherein connection-testing on checking occured in a thread holding the pool's
lock, leading to pool hangs or deadlocks if a connection hangs. [Thanks to Tobias Jenkner
for calling attention to this problem!]
-- modify BasicPropertiesConfig to not throw class cast Exceptions in the rare case that
a Properties object (presumably System properties rather than a file derived props Object)
contains (via the Map/Hastable API) non-String keys or values. [Thanks to Prima Upot for
calling attention to this issue.]
-- fixed a build issue where some classes required for code generation were not necessarily
compiled prior to the code generation step, causing builds to fail. [Thanks to James
Neville for calling attention to this issue.]
-- under some circumstances, the current list of ThreadPoolAsynchronousRunner's deadlock detector
seems to hold onto a lot of memory... presumably this is when long pending task lists build up
for some reason... nevertheless, we now dereference the "current" list in between deadlocker detector
invocations. The "last" list has to be retained between invocations, but the "current" list need
not be. [Thanks to Venkatesh Seetharamaiah for calling attention to this
issue, and for documenting the apparent source of object retention.]
-- modified log4j and jdk14logging MLog implementations to test for library classes upon
classload (MLog class only tests for the presence of the mlog implementations, which
doesn't notice if their dependant classes aren't there -- dependent classes must test
for the presence of the libraries.)
-- Modified GooGooStatementCache to warn as INFO (and with a descriptive message in debug)
of multple PreparedStatement preparation. [Thanks to mxnmatch on hibernate's forum for
calling attention to this issue.]
-- Modified BasicResourcePool to not warn on expected InterruptExceptions that occur if
clients are waiting for a Connection when the pool is closed. [Thanks to Blunted on
hibernate's forum for calling attention to this issue.]
-- Modified thread pool's strategy for dealing with deadlocks to prevent OutOfMemoryErrors
from sudden spawning of multiple threads to clear backlogged tasks. [Thanks to Greg Whalin
for calling attention to this problem.]
-- Fixed bugs in debug reporting of nested Exceptions/Warnings when converting or
rethrowing SQLExceptions [thanks to an anonymous sourceforge poster for finding this]
-- modified BasicResourcePool to shorten the period during which the pool's lock
is held during resource acquisition
-- maxIdleTime is generally enforced every (maxIdleTime / 8) seconds. Added logic to ensure that
even for very long maxIdleTimes, the longest maxIdleTime will ever go unenforced is 15 minutes.
-- Very tentative initial support in build for JUnit tests
-- Fixed documentation typos and added sample config for Tomcat 5.5. [Thanks to David Newcomb
for finding the typos and to Carl F. Hall for the sample config!]
-- ProxyConnections were not properly resetting readOnly, holdability, catalog, and typeMap
properties on Connection close. Fixed. [Thanks to Andy (bjorkmann?) for calling attention
to this bug.]
-- Added a complicated workaround to issues that may occur when a Connection
and its child statements simultaneously try to close. GooGooStatementCache's closeAll(Connection c)
method now wait()s for all Statements to close prior to returning, avoiding such issues (Oracle deadlock,
mysql NullPointerException), though dramatically complicating the code. Grrr. Theoretically, drivers should
be robust to the simultaneous closing of Statements and Connections, but they're not, so
my code gets to be more complicated. [Thanks to Juan Perez, Daniel Edberg, and Damien Evans for calling
attention to this issue.]
-- Proxy classes were overaggressively invalidating themselves in response to a signalled
Connection error. This problem was very similar to the problem fixed in 0.8.5-pre9, except
for rather than the PooledConnection too quickly closing its inner Connection, the proxies
too quickly detached themselves from the pooled Connection, considering themselves broken,
after a Connection error event. Users expect to be able to work with their Connections even
after c3p0 has in its conservatism decided they no are no longer worthy of a place in the
pool. [Thanks to "Geoff Fortytwo" for finding and helping to track down this issue.]
-- Fixed issue that prevented user-defined ConnectionTesters from being used.
-- Fixed NullPointerException on double-close of proxy Statements. [Thanks to Stephen Waud
for finding this bug and tracking it down.]
-- Modified C3P0PooledConnectionPool to generate a more informative Exception (one that
includes the stack trace of the initial problem) when a Connection test fails. [Thanks
to novotny from the hibernate forums for calling attention to this issue.]
` -- PooledDataSources now set SQLState 08001 on SQLExceptions resulting from a determination
by the pool that the database is down or unavailable. [Thanks to ml2709 on the hibernate
forums for calling attention to this issue.]
-- Modified ThreadPoolAsynchronousRunner to enforce its max task time even on one-off
"emergency threads" used to flush the pool of tasks after an apparent deadlock.
Hopefully this will help to eliminate the OutOfMemoryErrors that users occasionally
see if some Database operation starts to fail by blocking indefinitely, and is attempted
many many times. As long as these operations are susceptable to interrupt(), this should
resolve the problem. (If tasks block in a way that is immune to interrupt(), there is
little c3p0 can do to recover the Threads and prevent the number of blocked Threads
from growing until memory runs out.) [Thanks to Jean-Christophe Rousseau and Arup Vidyerthy
for calling attention to this issue.]
-- Wrapped System.getProperties() calls in try / catch blocks that catch and log
the SecurityExceptions, and ignore System properties if access to them is forbidden. configuration should work fine in security-controlled apps, but you
won't be able to configure c3p0 using "java -Dc3p0.xxxx=yyy ..." Since most users use for configuration, I don't think this will be a major problem. In the
future, I may be able to work around the restriction by having ask only for c3p0-specific
System properties, rather than calling the unrestricted System.getProperties() method.
[Thanks to zambak on SourceForge for calling attention to this issue.]
-- Modified generated proxy classes to properly report the type of the closed Object
if used after call to close().
-- Fixed a really embarrassing oversight, wherein cached PreparedStatements
were being physically close()ed prior to their return to the cache! Tests
on psql failed to reveal this, because psql-7.4 Statement.close() does nothing
to invalidate the Statement! [Many thanks to tbayboy on hibernate's forum for
discovering this issue, and to fassev, also on hibernate's forum, for correctly
deducing the cause!]
-- Fixed an embarrassing oversight where operations of proxy Statements
and ResultSets failed to mark transactions as potentially dirty, leading
to the possibility that Connections with unresolved transactional work could
be checked into the pool.
-- Modified c3p0 Statements to no longer permit calls to getConnection()
after Statement close, and to null out the backreference to the
parent Connection. [Thanks to Edward Bridges for pointing out the issue.]
-- NewPooledConnection threw a NullPointerException on close() of
a ResultSet whose creating Statement had already been closed.
Fixed. [Thanks to Edward Bridges for pointing out the problem.]
-- Modified PoolConfig to trim() read-in to avoid
NumberFormatExceptions if there is extra spacing in the file.
[Thanks to johnchan for calling attention to this issue.]
-- Documentation updates and improvements
-- Integrated configurable wrapper logging library, so that
c3p0 can log to log4j or jdk14logging.
-- First pass at documentation of JBoss integration.
-- Better one-per-JVM canonicalization (IdentityTokenized,
IdentityTokenResolvable, C3P0Registry)
-- First implementation of MBean for JBoss compatability
-- Fixed JNDI ref-based DataSources
-- Fixed a really embarrassing bug that made Statement pooling worse than
useless where users have set maxStatements but not maxStatementsPerConnection.
maxStatementsPerConnection -- that is, 0 -- was being used in place of maxStatements
on Statement cache initialization. [Thanks to Gwendal Tanguy for not only,
calling attention to this issue, but tracking down the precise location of
the bug!]
-- Documentation updates and improvements
-- Fixed an issue where failures on reset() of a pooled Connection caused
a Connection error to be signalled, without updating the status of the
PooledConnection to indicate that it is broken. Thanks to Henry Le for
calling attention to this issue.
-- Modified c3p0's behavior on detecting an apparently broken Connection.
Previously c3p0 would very aggressively close any Connections that, after
an Exception, failed to pass a Connection test. This sometimes surprised
and confused users, when Conections they expected to remain open were
closed from underneath them. Under normal circumstances, a Connection
which has failed its Connection test is broken anyway, so closing it
does no harm. But under some circumstances, a Connection may fail its
Connection test, but still be partly functional from the application's
point of view. So, now c3p0 simply marks Connections that fail their
Connection test following an Exception as broken, and excludes them from
future re-entry into the pool, but leaves those Connections open for users
to continue working with them prior to their explicit close() of the
Connection. [Thanks to Julian Legeny for calling attention to this issue.]
-- Modified c3p0's logging behavior to be a little bit less annoying. Rather than
frequently printing duplicate stack traces to be sure root cause Exceptions are
logged as well as converted, rethrown Exceptions, c3p0 now checks the Java
version, and uses the initCause() method to log the root cause in versions 1.4
and above. Also, some Exceptions that were expected and thrown without information
as to the cause now use the initCause() method to provide this in JDK 1.4 and
above. [Thanks to Ruslan Gainutdinov, Anthiny Pereira, Carsten ????, and
Eric Hsieh for calling attention to these issues.]
-- Fixed bug whereby proxy wrappers were sometimes placed around null return values
from native Objects, particularly Statement's getResultSets() method. [Thanks to
Ruslan Gainutdinov for calling attention to this issue.]
-- Fixed bug in NewProxyConnection whereby setTransactionIsolation was called even
if the Connections' isolation level had not been changed from its default. [Thanks
to Levent Aksu for reporting this bug.]
-- Added C3P0ProxyStatement and rawStatementOperations for accessing vendor-specific
Statement API.
-- Fixed bug whereby recently added properties could not be configured properly
via a PoolConfig Object passed to the DataSources factory method. [Thanks to
Julian Legeny for finding this bug.]
-- Undid new behavior whereby c3p0 warns when it detects that a transaction
has not been comitted or rolled-back prior to close and must be automatically
rolled-back. Because c3p0 autorollsback conservatively, whenever it is possible
that there may be any transactional resources held by the Connection being
checked in, users who make read-only queries, then close without commit or
rollback, saw the warning, and became annoyed. Warning when it's "right" is hard,
and would involve paying attention to the substance of user-queries, as well
as the transaction's isolation level. In the future, when c3p0 offers better
control over logging, we'll make the warning optional. Note that this release
only undoes the warning -- the new logic that detects potential transactional work
and only rollsback when there has been some remains. [Thanks to David Graham for
calling attention to this issue.]
-- Synchronized StatementCacheKey.find( ... ) method. Subclasses of StatementCacheKey
have always explicitly relied upon this method being synchronized in the parent class,
but somehow it was not... [Thanks to Manfred Hutt for reporting this bug.]
-- Modified both PooledConnection implementations to reset the transaction
isolation level on PooledConnection.reset()
-- Finalized PooledDataSource api for pool stats and resets.
-- Fixed some bugs in new Proxy classes relating to close(), methods,
especially NullPointerExceptions on duplicate close() operations.
-- Fixed some misleading documentation re: testConnectionOnCheckin and
automaticTestTable vs. preferredTestQuery
-- Changed wording of warning message on purge of idle-test-failing
resource to no longer suggest that some user action is required.
[Thanks to Krishna Kuchibhotla for calling attention to this
-- Modified GooGooStatementCache so that its close() is synchronous,
since when the Statement cache is closed, it's asynchronous
runner has often been closed already underneath it. (Actually
changed as of c3p0-0.8.5-pre7, but forgot to log it.)
-- Added flag (in both new and tradional reflective proxies) to
prevent automatic rollbacks on Connection close()es immediately post-
commit(), rollback(), or setAutoCommit().
-- DatabaseMetaData now properly returns proxy rather than naked
raw Connection. [Thanks to jhoeller on the hibernate forums]
-- In response to an apparent BEA WebLogic specific problem with
DriverManager.getConnection( ... ), DriverManagerDataSource now
caches its driver and uses Driver.connect( ... ) rather than
DriverManager.getConnection() when acquiring Connections. Thanks to
Lars Torunski for calling attention to this issue.
-- Fixed some missing synchroniztion that really should be provided for accessors
and mutators of DataSource bean properties.
-- Added config parameter automaticTestTable, which takes the name of a table
that c3p0 will create and use as the basis for Connection tests. It's easier
to use than preferred test query, because you don't have to ensure the existance
of any tables in the schema prior to using a Pooled data source.
-- Fixed idiotic oversight whereby Connections didn't get their transaction status
resolved and reset on check-in...
-- Modified resource pools to only conditionally support async ResourcePoolEvent
generation. Since c3p0 doesn't use ResourcePoolEvents, we can do without the
superfluous CarefulRunnableQueue, and its associated Thread, which was serving
as an event queue.
-- Fixed bug where asynchronous refurbishment (testing) of resources would be
attempted even on check-in to a closed pool, whose async threads had terminated.
-- c3p0 now supports the following new configuraion parameters: preferredTestQuery,
testConnectionOnCheckin, checkoutTimeout, and maxStatementsPerConnection.
preferredTestQuery should allow c3p0 to test Connections (whether on checkout,
checkin, or after a specified idle time) much more efficiently than with
it's default test [a call to getTables(...) on the Connection's DatabaseMetaData.].
checkoutTimeout allows clients to break out of a getConnection() call after a
a specified time period, rather than waiting (potentially indefinitely) for
a new Connection to be acquired from the database, or a checked-out Connection
to be rechecked in. testConnectionOnCheckin should be self-explanatory.
maxStatementsPerConnections allows users to specify how many statements to cache
on a per-connection base instead of, or in addition to, the global limit maxStatements.
-- Fixed various bugs relating to the referenceMaker and getReference() method of
-- Added TestUtils class to util package (modified from an old C3P0TestUtils
class now removed from the test directory), which now contains facilities
for establishing the identity of the physical Connection beneath a proxy,
so that tests can check to see when they are seeing the same physical
Connections recycled. [Thanks to Julian Legeny for calling attention to this
-- Modified code generators so that inner objects of all NewProxy classes
go to null on object close(). Closed proxies should be hopelessly broken.
-- Added missing parameter "usesTraditionalReflectiveProxies"
to ComboPooledDataSource
-- Fixed bug whereby CarefulRunnableQueue threads linger indefinitely, even
though their close() method has been called. [Thanks to muirwa on the
hibernate forums for calling attention to this issue.]
-- added utility class for Oracle users who wish to access vendor specific
Connection API relating to BLOBs and CLOBs. [Thanks to Dave Smith for calling
attention to this issue.]
-- modified ThreadPoolAsynchronousRunner to provide more information should an
"apparent deadlock" occur, and fixed that class' recovery strategy from such
an event. [Thanks to Damien Evans for calling attention to this issue.]
-- added config parameter usesTraditionalReflectiveProxies, which defaults to false,
and stitched in the 'new' codegenerated, non-reflective proxy implementation for
the default case.
-- Added new config parameters breakAfterAcquireFailure, acquireRetryAttempts, and
acquireRetryDelay to ComboPoolBackedDataSource (forget to modify this class when
I added the params earlier.)
-- Added description of raw connection operations to docs.
-- Defined AuthMaskedProperties object, and used this to prevent usernames
and passwords from being dumped to logs (a security issue). [Thanks to Zac Jacobson
for calling attention to this issue.]
-- Made BasicResourcePool's refurbishment of checked-in resources
asynchronous, and used this to add a test of the resource on check-in.
But since the extra test may slow stuff down substantially, I've disabled it
pending a testConnectionsOnCheckin config parameter.
-- Fixed NullPointerException in C3P0ImplUtils.findAuth(Object o) which
occurred when the Object at issue had write-only properties. [Thanks to
mrfekson for calling attention to this issue.]
-- Added acquireRetryAttempts, acquireRetryDelay, and breakAfterAcquireFailure
as configurable properties. Supporting breakAfterAcquireFailure == false
required substantial changes to BasicResourcePool. [Thanks to Zac Jacobson
for suggesting the behavior for breakAfterAcquireFailure == false.]
-- Added poolOwnerIdentityToken property to PoolBackedDataSource,
which becomes part of the state of C3P0PooledConectionPoolManager,
to avoid possibility that multiple, identically configured pools
that users create separately and consider distinct would be closed
if any one of them were closed. This parameter ensures that for
each user-created DataSource, there will be a distinct
C3P0PooledConectionPoolManager. Multiple copies, as often arise
via deserialization and/or dereferencing JNDI DataSources, will
still share a single pool and configuration.
-- Added reset methods, hard and soft, to PooledDataSource and
its implementations. Also added some extra aggregate pool statistics
methods. [Thanks to Travis Reeder for suggesting the hardReset() behavior.]
-- Fixed bug that PoolBackedDataSource always provided statistics for the
default authentication pool, even when alternate usernames and passwords
were provided.
-- Defined the C3P0ProxyConnection interface, and modified both reflective and unreflective
proxy Connection code to implement it. The interface provides a method through which
advanced users can work with the raw, unproxied database Connection. This was motivated
by some Oracle-specific API that could not be passed through the generic Connections
returned by c3p0 pools. Thanks to Dave Smith for calling attention to this issue.
-- Fixed a NullPointerException that resulted when users called DriverManagerDataSource's
getConnection( username, password ) with either username or password as null. Users
will now see the SQLException provided by their database for bad authentication, which
should make the problem more obvious.
-- Modified all variants of StatementCacheKey, as well as C3P0PooledConnection (proxy Statements)
to support jdbc3.0 API for autogenerated keys and ResultSet holdability. All jdbc 3.0 API
is now supported.
-- Fixed bug in which proxy Statements and ResultSets returned their
naked, unproxied parents from their getConnection() / getStatement()
methods. Thanks to Christian Josefsson for noticing this!
-- Added check to ensure that ResourcePools are not broken prior to
posting asynchronous events. It'd be better to factor the (as yet
unutilized async event stuff into a subclass to simplify
-- Fixed documentation typo that said default numHelperThreads was "false"...
now it's correctly set to "3".
-- Added extra System.err information to instrument the causes of broken
resource pools.
-- finalize() method of BasicResourcePool checks to be sure user has not
already closed the pool before closing, to avoid spurious multiple
close warnings. Thanks to Gavin King.
-- Added (hopefully not too annoying) dump of DataSource configuration
on pool initialization to assist user debugging.
-- Added methods to force destroy all resources used by a PooledDataSource,
even if other DataSource instances are sharing those resources. This is
intended primarily for applications that wish to discard the ClassLoader that
loaded c3p0, regardless of whatever's currently going on.
-- Turned off annoying trace messages when a C3P0PooledConnection closes.
-- Fixed issue in statement cache whereby statements that fail to check-in
properly (that throw an exception in "clearParameters") cause an exception
in removeStatement, because removeStatement sees a statement that appears
neither to be checked-out, nor in the deathmarch for checked-in statements.
Resolved by re-adding truculent statement to the checkedOut set, and then
destroying using removeStatement's codepath for removing and (force-)destroying
checked-out statements. Thanks to Zach Scott for calling attention to
this issue.
-- Updated and HTML-ized documentation, updated READMEs.
-- Includes, but does not yet use, cleaner, nonreflective
reimplementation of C3P0PooledConnection and all of the
JDBC proxy wrappers. (c3p0-0.8.4 will stick with the old,
tested implementation, c3p0-0.8.5 will replace.)
-- Removed temporary debug wrapper around acquired Connections
added in c3p0-0.8.4-test4-3.
-- Fixed race condition in BasicResourcePool close(), whereby
pooled resources were to be destroyed by an asynchronous thread
that itself was shut down by the close() method, so the resource
closures did not necessarily occur. Pooled resources are now
destroyed synchronously in BasicResourcePool.close()
-- Fixed erroneous call to C3P0PooledConnection reset() even when
the PooledConnection is known to be broken.
-- Added temporary debug wrapper that dumps a stack trace on physical
Connection close() to help Adrian track down a stubborn issue.
-- Made sure PooledConnection's that are known to be broken are not reset()
when the ProxyConnection that noticed the break cleans itself up.
-- Fixed a problem in C3P0PooledConnection's ProxyConnectionInvocationHandler
when Connection-invalidating exceptions occurred inside the factored-out
doSilentClose() method rather than in the invoke() method. This could lead to
broken PooledConnections not being noticed and expunged from the pool
prior to recheck-out.
-- Ensured stack-trace of any Connection-invalidating Exceptions are
logged. (Previously only the messages were logged.)
-- Access to ProxyConnections is now synchronized to avoid a possible
race condition where a Connection could close while another method
was in progress.
-- Cleaned up C3P0PooledConnection a bit, got rid of no longer used
reflective code, modified to use new SQL interface filter classes.
-- Reorganized bean generation stuff to sit under com.mchange.v2.codegen
-- Added com.mchange.v2.sql.filter package, and code generation stuff so
that abstract filter classes that implement JDBC interfaces can be
easily regenerated from current versions. (This is preparation for
JDK 1.4.x build support.)
-- Reorganized C3P0PooledConnection and C3P0PooledConnectionPool so that
PooledConnections only fire connectionErrorOccurred events on errors that
signal the Connection is invalid. (Previously we fired the event in all
cases, then tested the validity of the connection in the event handler.)
-- Fixed a bug where SQLExceptions that do not signal invalid Connections
(such as an exception on transaction commit under optimistic concurrency)
caused Connections to be closed, but still returned to the pool, leaving
the pool corrupt. [Thanks to Adrian Petru Dimulescu for discovering and
reporting this subtle problem!]
-- Backed off a binary incompatible change to DataSources API. Users
interested in statistics about running DataSources will need to
cast the result of DataSources.pooledDataSource() to a
com.mchange.v2.c3p0.PooledDataSource object. Other users should
be able to use DataSources without recompilinbg libraries that
depend upon it.
-- Defined com.mchange.v2.c3p0.ComboPooledDataSource, a single, directly-instantiable
JavaBean-style DataSource backed by a c3p0 pool, which makes integration
with various app servers easier (e.g. works with Tomcat), and which may
provide a more straightforward API for users than DataSources and PoolConfig.
-- Switched various classes from using RoundRobinAsynchronousRunner
to a new ThreadPoolAsynchronousRunner.
-- Defined PoolingDataSource interface, which permits clients to access
statistics about the state of underlying pools, and made C3P0 pool-backed
DataSources implement that interface.
-- fixed a subtle bug with respect to ownership of shared resources in which
finalize() methods for discarded DataSources could cause helper threads and
other resources to be cleaned up prematurely, while they are still in use
by other DataSources.
-- major reorganization of code generation for DataSources and the
resulting objects.
-- reorganization of Serializable and Referenceable code of DataSources
for JNDI Binding
-- made PoolConfig defaults (which are set by system properties, a file, or else hardcoded defaults) apply to DataSources
that are directly instantiated rather than created via the
com.mchange.v2.c3p0.DataSources factory class.
-- organized scattered stuff into a self-contained, distributable build
directly, and divided source and binary distributions
-- relicensed library; now available under LGPL.
-- take greater care to abort if a helper thread wakes from wait()
inside of a broken resource pool
-- log more debug information if a pool unexpectedly breaks
-- resolved a deadlock arising from ConnectionEventSupport objects
calling event listener methods while holding the ConnectionEventSupport's
lock. (Thanks to an anonymous Sourceforge bug submitter for tracking this
-- added two configuration parameters to modify how c3p0 deals with
potentially unresolved transactions on transaction close --
autoCommitOnClose and forceIgnoreUnresolvedTransactions. c3p0's
default behavior is as it always was (and as I think it nearly
always should be) -- by default we roll back any uncommitted work
on close. See PoolConfig class for complete documentation.
-- fixed stupid build error that made unzipping c3p0 archives messy
since c3p0-0.8.2. (SORRY!)
-- fixed problem with c3p0 trying endlessly to connect to a database
that is down or not present. c3p0 now tries thirty time, once per
second, then gives up. if there's demand, I'll make the delay and
number of attempts user configurable. Thanks to Alfonso da Silva
for calling attention to this problem.
-- modified RoundRobinAsynchronousRunner to die more gracefully if,
somehow, one of its CarefulAsynchronousRunners somehow dies. Thanks to
Travis Reeder for calling attention to this issue.
-- modified ResourcePools to allow resources to be tested periodically
and asynchronously while they are idle in the pool.
-- added to a new configuration property to PoolConfig and related
classes, idleConnectionTestPeriod, which if set to a greater-than-zero
value will cause c3p0 to periodically test idle, pooled connection,
and remove them from the pool if they are broken. Thanks to
Travis Reeder for calling attention to this issue.
-- removed SQLState 08S01 from the set of SQL states presumed to mean
all connections in the pool have become invalid. Apparently MySQL
uses this SQLState to indicate the timing out of idle, individual
connections. Thanks to Travis Reeder for calling attention to this issue.
-- many (!) bug-fixes, small tweaks and improvements.
-- temporarily sprinkled some debugging output prefixed "c3p0-TRAVIS" to
try to resolve some issues for Travis Reeder
-- Addressed a rare problem that ocurred when application servers or
other external code calls interrupt() on c3p0 helper threads. Thanks
to Travis Reeder for reporting this issue! The endless cascade of
ArrayIndexOutOfBoundsException that would occasionallY result has
been fixed, and c3p0 helper threads now ignore interupts,
which should almost completely prevent such interrupts()
from leaving c3p0 pools in a corrupt state. [In the very, very
unlikely event that multiple, well-timed interrupts() cause
a pool to diagnose itself broken, c3p0 will start throwing clear
exceptions to indicate that -- there shouldn't be any deadlock or
any subtle kind of corruption.]
-- Deprecated PoolBackedDataSourceFactory and DriverManagerDataSourceFactory
in favor of the newer DataSources class.
-- Updated example code and documentation.
-- Moved constants from DataSources to PoolConfig where they belong!
-- Removed use of properties default mechanism in setting up JDBC
properties for DriverManagerDataSource, as some drivers use get()
rather than getProperties(), and the Properties.get() method ignores
defaults. [Thanks go to Michael Jakl <> for both finding
and fixing this problem!!!]
-- Fixed subtle problem whereby the introspective constructor of
C3P0PooledConnectionPoolManager would inadvertently check out an
extra pooled connection, by treating the getPooledConnection() as
a simple property read and invoking it. [Thanks go to Michael Jakl
<> for both finding and fixing this bug!!!]
-- Finished DataSources static factory for, um, DataSources.
-- Added PoolConfig class to encapsulate all configuration
information from factories
-- Set up PoolConfig to pay attention to (in order):
1) explicit, user-defined parameters
2) parameters defined as System properties
3) parameters defined in a Properties file resource, at
resource pasth / in PoolConfig's classloader
4) Hardcoded defaults from the C3P0Defaults class
-- Defined a new pool configuration property, "testConnectionOnCheckout".
If set to true (it defaults to false), each pooled Connection will be
verified at checkout time, before being supplied to clients. THIS SLOWS
CONNECTION CHECKOUT DRAMATICALLY (by an order of magnitude on my machine),
but some folks may want it. The expensive, verified Connections are still
faster to acquire than brand new ones.
-- If a Connection experiences an SQLException with a defined SQL State of
08001, 08007, or 08S01, the entire database is considered to have been
shutdown, and the pool whose connection experienced the problem is reset
-- i.e. all existing Connections are discarded, and new Connections are
acquired for the pool. (Thanks! to Gavin King for suggesting the SQL State
approach to detecting disconnects.)
-- ConnectionTester interface and its default implementation were modified
to support reporting DATABASE_IS_INVALID (rather than simply testing a
disconnected, individual Connection).
-- Fixed some really stupid bugs from pre-7, in particular
the static factory methods of the DataSources class were
not declared static. Oops.
-- Major reorganization of data source implementations. These
are now generated from XML templates, because it was becoming
difficult to keep multiple versions of multiple types of
datasources in sync with one another
-- a new factory / utility class has been added,
Soon the older factories (DriverManagerDataSourceFactory and
PoolBackedDataSourceFactory) will be deprecated in favor of this.
-- Began adding support for wrapper DataSources that lazily bind to
an unpooled DataSource specified via JNDI. (Current wrappers require
an actual instance, not a mere reference in order to construct a
PoolBacked or ConnectionPoolDataSource).
-- Fixed bug that led to ConcurrentModificationException
when Connection.close() iterates through and closes
its unclosed Statements.
-- Fixed bug that caused uncached Statements to sometimes
be closed twice, as they were not properly removed from
the set of Connection-associated statements on user-
initiated Statement.close(), and would thus be closed
again on Connection.close().
-- Reorganized classes to clean up top-level com.mchange.v2.c3p0
package. The gory innards of c3p0 now live in an impl package.
-- Reorganized DataSources so that there are "Base" versions suitable
for DBMS-specific subclassing.
-- DataSources looked up by JNDI now "coalesce" -- i.e., if
two separate calls look-up the same JNDI bound DataSource,
they will both see the same DataSource instance in the local
VM. This is especially important for PoolBackedDataSources, as they
own "helper threads" that help to manage the pool. Now each
logical DataSource shares the same pool of helper threads,
regardless of whether it is looked up once and cached, or
looked up multiple times and at multiple places by an application.
-- Number of helper-threads managing a PoolBackedDataSource is now
-- Fixed BAD memory leak -- cachedStatements were being retained
on close. This brings a performance improvement to
the StatementCache as well.
-- Set-up StatementCacheKey to coalesce, such that its equals
method can look like {return (this == o);}. Unfortunately, I
did not see the kind of performance increase I'd anticipated
from this... in fact, whatever performance changes I did see were
pretty negligable. Performed some other, mostly useless, optimizations
to the process of acquiring StatementCacheKeys. (I can't decide...
I've got 3 usuable, broadly similar versions of StatementCacheKey now.)
-- Fixed bugs having to do with when to create and the daemon status
of Timers
-- Appropriately synchronized access to PoolBackedDataSource. (This
should permit me to pull back on synchronization / marking of
volatile in some dependent classes, as access to these is now
controlled via PoolBackedDataSource.)
-- Made instantion of C3P0PooledConnectionPoolManager
by WrapperPoolBackedDataSource "lazy", so that Timer / Async task
threads are only started up when a Connection is actually requested
from a DataSource. This is useful especially for the instance residing
in the JNDI server VM, whose purpose is likely only to serve as an
instance to be looked up over the network. There's no reason why
it should keep several stalled Threads open.
-- Fixed DefaultConnectionTester, whose logic was bass-ackwards.
-- Factory method for StatementCacheKeys now "coalesce" equivalent
instances to conserve memory and to enable a fast (this == o) test
to usually work in equals methods. [Value test remains for when
identity test fails... maybe later I'll be more comfortable with a
1 instance per value guarantee, and remove the rest of the test.
As the constructor is private, the factory method coalesces, and the
class is not Serializable, an identity test should suffice. But
I'm paranoid.]
-- All asynchronous tasks are now distributed to 3 helper threads per
DataSource rather than queuing on just one. (Culing of expired
Connections is still managed by a single java.util.Timer... I don't
think this will be a problem.)
-- Total rewite of Statement cache!