by Steve Waldman <firstname.lastname@example.org>
© 2013 Machinery For Change, Inc.
This software is made available for use, modification, and redistribution, under the terms of the Lesser GNU Public License, v.2.1 (LGPL) or the Eclipse Public License, v.1.0 (EPL), at your option. You should have received copies of both licenses with this distribution.
Note: Coordinates of this version on the Maven central repository: [groupId: com.mchange, artifactId: c3p0, version: 0.9.5-pre6]
c3p0 was designed to be butt-simple to use. Just put the files lib/c3p0-0.9.5-pre6.jar and lib/mchange-commons-java-0.2.6.3.jar in your application's effective CLASSPATH, then make a DataSource like this:
[Optional] If you want to turn on PreparedStatement pooling, you must also set maxStatements and/or maxStatementsPerConnection (both default to 0):
Do whatever you want with your DataSource, which will be backed by a Connection pool set up with default parameters. You can bind the DataSource to a JNDI name service, or use it directly, as you prefer.
When you are done, you can clean up the DataSource you've created like this:
That's it! The rest is detail.
c3p0 is an easy-to-use library for making traditional JDBC drivers "enterprise-ready" by augmenting them with functionality defined by the jdbc3 spec and the optional extensions to jdbc2. In particular, c3p0 provides several useful services:
The library tries hard to get the details right:
c3p0 hopes to provide DataSource implementations more than suitable for use by high-volume "J2EE enterprise applications". Please provide feedback, bug-fixes, etc.!
c3p0-0.9.5-pre6 requires a level 1.6.x or above Java Runtime Environment.
Place the files lib/c3p0-0.9.5-pre6.jar and lib/mchange-commons-java-0.2.6.3.jar somewhere in your CLASSPATH (or any other place where your application's classloader will find it). That's it!
From a users' perspective, c3p0 simply provides standard jdbc DataSource objects. When creating these DataSources, users can control pooling-related, naming-related, and other properties. (See Appendix A for a comprehensive list of configuration properties.) All pooling is entirely transparent to users once a DataSource has been created.
There are three ways of acquiring c3p0 pool-backed DataSources: 1) directly instantiate and configure a ComboPooledDataSource bean; 2) use the DataSources factory class; or 3) "build your own" pool-backed DataSource by directly instantiating PoolBackedDataSource and setting its ConectionPoolDataSource. Most users will probably find instantiating ComboPooledDataSource to be the most convenient approach. Once instantiated, c3p0 DataSources can be bound to nearly any JNDI-compliant name service.
Regardless of how you create your DataSource, c3p0 will use defaults for any configuration parameters that you do not specify programmatically. c3p0 has built-in, hard-coded defaults, but you can override these by creating placing configuration files as a top-level resource in the same CLASSPATH (or ClassLoader) that loads c3p0's jar file. c3p0 DataSources may be configured via simple java.util.Properties files called c3p0.properties, via more advanced HOCON configuration files (e.g. application.conf, application.json), or in an XML format, c3p0-config.xml. See Configuration below.
Perhaps the most straightforward way to create a c3p0 pooling DataSource is to instantiate an instance of com.mchange.v2.c3p0.ComboPooledDataSource. This is a JavaBean-style class with a public, no-arg constructor, but before you use the DataSource, you'll have to be sure to set at least the property jdbcUrl. You may also want to set user and password, and, if you use an old-style JDBC driver that you will not externally preload, you should set the driverClass.
The defaults of any c3p0 DataSource are determined by configuration you supply, or else revert to hard-coded defaults [see configuration properties]. c3p0 supports named configurations so that you can configure multiple DataSources. If you wish to use a named configuration, construct your com.mchange.v2.c3p0.ComboPooledDataSource with the configuration name as a constructor agument:
Of course, you can still override any configuration properties programmatically, as above.
Alternatively, you can use the static factory class com.mchange.v2.c3p0.DataSources to build unpooled DataSources from traditional JDBC drivers, and to build pooled DataSources from unpooled DataSources:
If you use the DataSources factory class, and you want to programmatically override default configuration parameters, you can supply a map of override properties:
If you are using named configurations, you can specify the configuration that defines the default configuration for your DataSource:
You can wrap any DataSouce using DataSource.pooledDataSource( ... ), usually with no problem whatsoever. DataSources are supposed to indicate the username and password associated by default with Connections via standard properties user and password. Some DataSource implementations do not offer these properties. Usually this is not at all a problem. c3p0 works around this by acquiring "default" Connections from the DataSource if it can't find default authentication information, and a client has not specified the authentification information via getConnection( user, password ).
However, in rare circumstances, non-c3p0 unpooled DataSources provide a user property, but not a password property, or you have access to a DataSource that you wish to wrap behind a pool, but you wish to override its build-in authentification defaults without actually modifying the user or password properties.
c3p0 provides configuation properties overrideDefaultUser and overrideDefaultPassword. If you set these properties, programmatically as above, or via any of c3p0's configuration mechanisms, c3p0 PooledDataSources will ignore the user and password property associated with the underlying DataSource, and use the specified overrides instead.
c3p0 DataSources backed by a pool, which include implementations of ComboPooledDataSource and the objects returned by DataSources.pooledDataSource( ... ), all implement the interface com.mchange.v2.c3p0.PooledDataSource, which makes available a number of methods for querying the status of DataSource Connection pools. Below is sample code that queries a DataSource for its status:
The status querying methods all come in three overloaded forms, such as:
c3p0 maintains separate pools for Connections with distinct authentications. The various methods let you query the status of pools individually, or aggregate statistics for all authentifications for which your DataSource is maintaining pools. Note that pool configuration parmeters such as maxPoolSize are enforced on a per-authentification basis! For example, if you have set maxPoolSize to 20, and if the DataSource is managing connections under two username-password pairs [the default, and one other pair established via a call to getConnection(user, password), you should expect to see as many as 40 Connections from getNumConnectionsAllUsers().
Most applications only acquire default-authenticated Connections from DataSources, and can typically just use the getXXXDefaultUser() to gather Connection statistics.
As well as Connection pool realted statistics, you can retrieve status information about each DataSource's Thread pool. Please see PooledDataSource for a complete list of available operations.
If it's inconvenient or impossible to get a reference to your DataSource via JNDI or some other means, you can find all live c3p0 DataSources using the C3P0Registry class, which includes three static methods to help you out:
The first method will hand you the Set of all live c3p0 PooledDataSources. If you are sure your application only makes one PooledDataSources, or you can distinguish between the DataSources by their configuration properties (inspected via "getters"), the first method may be sufficient. Because this will not always be the case, c3p0 PooledDataSources have a special property called dataSourceName. You can set the dataSourceName property directly when you construct your DataSource, or dataSourceName can be set like any other property in a named or the default config. Otherwise, dataSourceName will default to either 1) the name of your DataSource's configuration, if you constructed it with a named configuration; or 2) a unique (but unpredicatble) name if you are using the default configuration.
There is no guarantee that a dataSourceName will be unique. For example, if two c3p0 DataSources share the same named configuration, and you have not set the dataSourceName programmatically, the two data sources will both share the name of the configuration. To get all of the DataSources with a particular dataSourceName, use pooledDataSourcesByName( ... ). If you've ensured that your DataSource's name is unique (as you will generally want to do, if you intend to use C3P0Registry to find your DataSources), you can use the convenience method pooledDataSourceByName( ... ), which will return your DataSource directly, or null if no DataSource with that name is available. If you use pooledDataSourceByName( ... ) and more than one DataSource shares the name supplied, which one it will return is undefined.
The easy way to clean up after c3p0-created DataSources is to use the static destroy method defined by the class DataSources. Only PooledDataSources need to be cleaned up, but DataSources.destroy( ... ) does no harm if it is called on an unpooled or non-c3p0 DataSource.
Alternatively, c3p0's PooledDataSource interface contains a close() method that you can call when you know you are finished with a DataSource. So, you can cast a c3p0 derived DataSource to a PooledDataSource and close it:
Unreferenced instances of PooledDataSource that are not close()ed by clients close() themselves prior to garbage collection in their finalize() methods. As always, finalization should be considered a backstop and not a prompt or sure approach to resource cleanup.
There is little reason for most programmers to do this, but you can build a PooledDataSource in a step-by-step way by instantiating and configuring an unpooled DriverManagerDataSource, instantiating a WrapperConnectionPoolDataSource and setting the unpooled DataSource as its nestedDataSource property, and then using that to set the connectionPoolDataSource property of a new PoolBackedDataSource.
This sequence of events is primarily interesting if your driver offers an implementation of ConnectionPoolDataSource, and you'd like c3p0 to use that. Rather than using c3p0's WrapperConnectionPoolDataSource, you can create a PoolBackedDataSource and set its connectionPoolDataSource property. Statement pooling, ConnectionCustomizers, and many c3p0-specific properties are unsupported with third party implementations of ConnectionPoolDataSource. (Third-party DataSource implementations can be substituted for c3p0's DriverManagerDataSource with no significant loss of functionality.)
JDBC drivers sometimes define vendor-specific, non-standard API on Connection and Statement implementations. C3P0 wraps these Objects behind a proxies, so you cannot cast C3P0-returned Connections or Statements to the vendor-specific implementation classes. C3P0 does not provide any means of accessing the raw Connections and Statements directly, because C3P0 needs to keep track of Statements and ResultSets created in order to prevent resource leaks and pool corruption.
C3P0 does provide an API that allows you to invoke non-standard methods reflectively on an underlying Connection. To use it, first cast the returned Connection to a C3P0ProxyConnection. Then call the method rawConnectionOperation, supplying the java.lang.reflect.Method object for the non-standard method you wish to call as an argument. The Method you supply will be invoked on the target you provide on the second argument (null for static methods), and using the arguments you supply in the third argument to that function. For the target, and for any of the method arguments, you can supply the special token C3P0ProxyConnection.RAW_CONNECTION, which will be replaced with the underlying vendor-specific Connection object before the Method is invoked.
C3P0ProxyStatement offers an exactly analogous API.
Any Statements (including Prepared and CallableStatements) and ResultSets returned by raw operations will be c3p0-managed, and will be properly cleaned-up on close() of the parent proxy Connection. Users must take care to clean up any non-standard resources returned by a vendor-specific method.
Here's an example of using Oracle-specific API to call a static method on a raw Connection:
C3P0 includes special support for some Oracle-specific methods. See Appendix F.
While c3p0 does not require very much configuration, it is very tweakable. Most of the interesting knobs and dials are represented as JavaBean properties. Following JavaBean conventions, we note that if an Object has a property of type T called foo, it will have methods that look like...
public T getFoo();...or both, depending upon whether the property is read-only, write-only, or read-writable.
public void setFoo(T foo);
There are several ways to modify c3p0 properties: You can directly alter the property values associated with a particular DataSource in your code, or you can configure c3p0 externally, via a simple Java properties file, via an XML configuration file, or via System properties. Configuration files are normally looked up under standard names (c3p0.properties or c3p0-config.xml) at the top level of an application's classpath, but the XML configuration can be placed anywhere in an application's file system or classpath, if the system property com.mchange.v2.c3p0.cfg.xml is set.
DataSources are usually configured before they are used, either during or immediately following their construction. c3p0 does support property modifications midstream, however.
If you obtain a DataSource by instantiating a ComboPooledDataSource, configure it by simply calling appropriate setter methods offered by that class before attempting a call to getConnection(). See the example above.
If you obtain a DataSource by using factory methods of the utility class com.mchange.v2.c3p0.DataSources, and wish to use a non-default configuration, you can supply a Map of property names (beginning with lower-case letters) to property values (either as Strings or "boxed" Java primitives like Integer or Boolean).
All tweakable properties are documented for reference in Appendix A. The most basic and important c3p0 configuration topics are discussed below.
c3p0 Connection pools are very easy to configure via the following basic parameters:
initialPoolSize, minPoolSize, maxPoolSize define the number of Connections that will be pooled. Please ensure that minPoolSize <= maxPoolSize. Unreasonable values of initialPoolSize will be ignored, and minPoolSize will be used instead.
Within the range between minPoolSize and maxPoolSize, the number of Connections in a pool varies according to usage patterns. The number of Connections increases whenever a Connection is requested by a user, no Connections are available, and the pool has not yet reached maxPoolSize in the number of Connections managed. Since Connection acquisition is very slow, it is almost always useful to increase the number of Connections eagerly, in batches, rather than forcing each client to wait for a new Connection to provoke a single acquisition when the load is increasing. acquireIncrement determines how many Connections a c3p0 pool will attempt to acquire when the pool has run out of Connections. (Regardless of acquireIncrement, the pool will never allow maxPoolSize to be exceeded.)
The number of Connections in a pool decreases whenever a pool tests a Connection and finds it to be broken (see Configuring Connection Testing below), or when a Connection is expired by the pool after sitting idle for a period of time, or for being too old (See Managing Pool Size and Connection Age.)
Different applications have different needs with regard to trade-offs between performance, footprint, and reliability. C3P0 offers a wide variety of options for controlling how quickly pools that have grown large under load revert to minPoolSize, and whether "old" Connections in the pool should be proactively replaced to maintain their reliablity.
By default, pools will never expire Connections. If you wish Connections to be expired over time in order to maintain "freshness", set maxIdleTime and/or maxConnectionAge. maxIdleTime defines how many seconds a Connection should be permitted to go unused before being culled from the pool. maxConnectionAge forces the pool to cull any Connections that were acquired from the database more than the set number of seconds in the past.
maxIdleTimeExcessConnections is about minimizing the number of Connections held by c3p0 pools when the pool is not under load. By default, c3p0 pools grow under load, but only shrink if Connections fail a Connection test or are expired away via the parameters described above. Some users want their pools to quickly release unnecessary Connections after a spike in usage that forces a large pool size. You can achieve this by setting maxIdleTimeExcessConnections to a value much shorter than maxIdleTime, forcing Connections beyond your set minimum size to be released if they sit idle for more than a short period of time.
Some general advice about all of these timeout parameters: Slow down! The point of Connection pooling is to bear the cost of acquiring a Connection only once, and then to reuse the Connection many, many times. Most databases support Connections that remain open for hours at a time. There's no need to churn through all your Connections every few seconds or minutes. Setting maxConnectionAge or maxIdleTime to 1800 (30 minutes) is quite aggressive. For most databases, several hours may be more appropriate. You can ensure the reliability of your Connections by testing them, rather than by tossing them. (see Configuring Connection Testing.) The only one of these parameters that should generally be set to a few minutes or less is maxIdleTimeExcessConnections.
c3p0 can be configured to test the Connections that it pools in a variety of ways, to minimize the likelihood that your application will see broken or "stale" Connections. Pooled Connections can go bad for a variety of reasons -- some JDBC drivers intentionally "time-out" long-lasting database Connections; back-end databases or networks sometimes go down "stranding" pooled Connections; and Connections can simply become corrupted over time and use due to resource leaks, driver bugs, or other causes.
c3p0 provides users a great deal of flexibility in testing Connections, via the following configuration parameters:
idleConnectionTestPeriod, testConnectionOnCheckout, and testConnectionOnCheckin control when Connections will be tested. automaticTestTable, connectionTesterClassName, and preferredTestQuery control how they will be tested.
When configuring Connection testing, first try to minimize the cost of each test. By default, Connections are tested by calling the getTables() method on a Connection's associated DatabaseMetaData object. This has the advantage of working with any database, and regardless of the database schema. However, a call to DatabaseMetaData.getTables() is often much slower than a simple database query.
The most convenient way to speed up Connection testing is to define the parameter automaticTestTable. Using the name you provide, c3p0 will create an empty table, and make a simple query against it to test the database. Alternatively, if your database schema is fixed prior to your application's use of the database, you can simply define a test query with the preferredTestQuery parameter. Be careful, however. Setting preferredTestQuery will lead to errors as Connection tests fail if the query target table does not exist in your database table prior to initialization of your DataSource.
Advanced users may define any kind of Connection testing they wish, by implementing a ConnectionTester and supplying the fully qualified name of the class as connectionTesterClassName. If you'd like your custom ConnectionTesters to honor and support the preferredTestQuery and automaticTestTable parameters, implement UnifiedConnectionTester, most conveniently by extending AbstractConnectionTester. See the api docs for more information.
The most reliable time to test Connections is on check-out. But this is also the most costly choice from a client-performance perspective. Most applications should work quite reliably using a combination of idleConnectionTestPeriod and testConnectionOnCheckIn. Both the idle test and the check-in test are performed asynchronously, which can lead to better performance, both perceived and actual.
For some applications, high performance is more important than the risk of an occasional database exception. In its default configuration, c3p0 does no Connection testing at all. Setting a fairly long idleConnectionTestPeriod, and not testing on checkout and check-in at all is an excellent, high-performance approach.
If you don't know what to do, try this:
Use SELECT 1 for your preferredTestQuery, if you are running MySQL or Postgres. For other databases, look for suggestions here. Leave automatedTestTable undefined.
Begin by setting testConnectionOnCheckOut to true and get your application to run correctly and stably. If you are happy with your application's performance, you can stop here! This is the simplest, most reliable form of Connection-testing, but it does have a client-visible performance cost.
If you'd like to improve performance by eliminating Connection testing from clients' code path:
Set testConnectionOnCheckOut to false
Set testConnectionOnCheckIn to true
Set idleConnectionTestPeriod to 30, fire up you application and observe. This is a pretty robust setting, all Connections will tested on check-in and every 30 seconds thereafter while in the pool. Your application should experience broken or stale Connections only very rarely, and the pool should recover from a database shutdown and restart quickly. But there is some overhead associated with all that Connection testing.
If database restarts will be rare so quick recovery is not an issue, consider reducing the frequency of tests by idleConnectionTestPeriod to, say, 300, and see whether clients are troubled by stale or broken Connections. If not, stick with 300, or try an even bigger number. Consider setting testConnectionOnCheckIn back to false to avoid unnecessary tests on checkin. Alternatively, if your application does encounter bad Connections, consider reducing idleConnectionTestPeriod and set testConnectionOnCheckIn back to true. There are no correct or incorrect values for these parameters: you are trading off overhead for reliability in deciding how frequently to test. The exact numbers are not so critical. It's usually easy to find configurations that perform well. It's rarely worth spending time in pursuit of "optimal" values here.
So, when should you stick with simple and reliable (Step 2 above), and when is it worth going for better performance (Step 3)? In general, it depends on how much work clients typically do with Connections once they check them out. If clients usually make complex queries and/or perform multiple operations, adding the extra cost of one fast test per checkout will not much affect performance. But if your application typically checks out a Connection and performs one simple query with it, throwing in an additional test can really slow things down.
That's nice in theory, but often people don't really have a good sense of how much work clients perform on average. The best thing to do is usually to try Step 3, see if it helps (however you measure performance), see if it hurts (is your application troubled by broken Connections? does it recover from database restarts well enough?), and then decide. You can always go back to simple, slow, and robust. Just set testConnectionOnCheckOut to true, testConnectionOnCheckIn to false, and set idleConnectionTestPeriod to 0. But do be sure to set an efficient preferredTestQuery !!!
c3p0 implements transparent PreparedStatement pooling as defined by the JDBC spec. Under some circumstances, statement pooling can dramatically improve application performance. Under other circumstances, the overhead of statement pooling can slightly harm performance. Whether and how much statement pooling will help depends on how much parsing, planning, and optimizing of queries your databases does when the statements are prepared. Databases (and JDBC drivers) vary widely in this respect. It's a good idea to benchmark your application with and without statement pooling to see if and how much it helps.
You configure statement pooling in c3p0 via the following configuration parameters:
maxStatements is JDBC's standard parameter for controlling statement pooling. maxStatements defines the total number PreparedStatements a DataSource will cache. The pool will destroy the least-recently-used PreparedStatement when it hits this limit. This sounds simple, but it's actually a strange approach, because cached statements conceptually belong to individual Connections; they are not global resources. To figure out a size for maxStatements that does not "churn" cached statements, you need to consider the number of frequently used PreparedStatements in your application, and multiply that by the number of Connections you expect in the pool (maxPoolSize in a busy application).
maxStatementsPerConnection is a non-standard configuration parameter that makes a bit more sense conceptually. It defines how many statements each pooled Connection is allowed to own. You can set this to a bit more than the number of PreparedStatements your application frequently uses, to avoid churning.
If either of these parameters are greater than zero, statement pooling will be enabled. If both parameters are greater than zero, both limits will be enforced. If only one is greater than zero, statement pooling will be enabled, but only one limit will be enforced.
If statementCacheNumDeferredCloseThreads is greater than zero, the Statement pool will defer physically close()ing cached Statements until its parent Connection is not in use by any client or internally (in e.g. a test) by the pool itself. For some JDBC drivers (especially Oracle), attempts to close a Statement freeze if the parent Connection is in use. This parameter defaults to 0. Set it to a positive value if you observe "APPARENT DEADLOCKS" realted to Connection close tasks. Almost always, that value should be one: if you need more than one Thread dedicated solely to Statement destruction, you probably should set maxStatements and/or maxStatementsPerConnection to higher values so you don't churn through cached Statements so quickly.
c3p0 DataSources are designed (and configured by default) to recover from temporary database outages, such as those which occur during a database restart or brief loss of network connectivity. You can affect how c3p0 handles errors in acquiring Connections via the following configurable properties:
When a c3p0 DataSource attempts and fails to acquire a Connection, it will retry up to acquireRetryAttempts times, with a delay of acquireRetryDelay between each attempt. If all attempts fail, any clients waiting for Connections from the DataSource will see an Exception, indicating that a Connection could not be acquired. Note that clients do not see any Exception until a full round of attempts fail, which may be some time after the initial Connection attempt. If acquireRetryAttempts is set to 0, c3p0 will attempt to acquire new Connections indefinitely, and calls to getConnection() may block indefinitely waiting for a successful acquisition.
Once a full round of acquisition attempts fails, there are two possible policies. By default, the c3p0 DataSource will remain active, and will try again to acquire Connections in response to future requests for Connections. If you set breakAfterAcquireFailure to true, the DataSource will consider itself broken after a failed round of Connection attempts, and future client requests will fail immediately.
Note that if a database restart occurs, a pool may contain previously acquired but now stale Connections. By default, these stale Connections will only be detected and purged lazily, when an application attempts to use them, and sees an Exception. Setting maxIdleTime or maxConnectionAge can help speed up the replacement of broken Connections. (See Managing ConnectionAge.) If you wish to avoid application Exceptions entirely, you must adopt a connection testing strategy that is likely to detect stale Connections prior to their delivery to clients. (See "Configuring Connection Testing".) Even with active Connection testing (testConnectionOnCheckout set to true, or testConnectionOnCheckin and a short idleConnectionTestPeriod), your application may see occasional Exceptions on database restart, for example if the restart occurs after a Connection to the database has already been checked out.
Application frequently wish to set up Connections in some standard, reusable way immediately after Connection acquisitions. Examples of this include setting-up character encodings, or date and time related behavior, using vendor-specific APIs or non-standard SQL statement executions. Occasionally it is useful to override the default values of standard Connection properties such as transactionIsolation, holdability, or readOnly. c3p0 provides a "hook" interface that you can implement, which gives you the opportunity to modify or track Connections just after they are checked out from the database, immediately just prior to being handed to clients on checkout, just prior to being returned to the pool on check-in, and just prior to final destruction by the pool. The Connections handed to ConnectionCustomizers are raw, physical Connections, with all vendor-specific API accessible. See the API docs for ConnectionCustomizer.
To install a ConnectionCustomizer just implement the interface, make your class accessible to c3p0's ClassLoader, and set the configuration parameter below:
ConnectionCustomizers are required to be immutable classes with public no argument constructors. They shouldn't store any state. For (rare) applications that wish to track the behavior of individual DataSources with ConnectionCustomizers, the lifecycle methods each accept a DataSource-specific "identityToken", which is unique to each PooledDataSource. ConnectionCustomizers can be configured via user-defined configuration extensions.
Below is a sample ConnectionCustomizer. Implementations that do not need to override all four ConnectionCustomizer methods can extend AbstractConnectionCustomizer to inherit no-op implementations of all methods.
For an example ConnectionCustomizer that employs user-defined configuration properties, see below.
Connections checked into a pool cannot have any unresolved transactional work associated with them. If users have set autoCommit to false on a Connection, and c3p0 cannot guarantee that there is no pending transactional work, c3p0 must either rollback() or commit() on check-in (when a user calls close()). The JDBC spec is (unforgivably) silent on the question of whether unresolved work should be committed or rolled back on Connection close. By default, c3p0 rolls back unresolved transactional work when a user calls close().
You can adjust this behavior via the following configuration properties:
If you wish c3p0 to allow unresolved transactional work to commit on checkin, set autoCommitOnClose to true. If you wish c3p0 to leave transaction management to you, and neither commit nor rollback (nor modify the state of Connection autoCommit), you may set forceIgnoreUnresolvedTransactions to true. Setting forceIgnoreUnresolvedTransactions is strongly discouraged, because if clients are not careful to commit or rollback themselves prior to close(), or do not set Connection autoCommit consistently, bizarre unreproduceable behavior and database lockups can occur.
Sometimes client applications are sloppy about close()ing all Connections they check out. Eventually, the pool grows to maxPoolSize, and then runs out of Connections, because of these bad clients.
The right way to address this problem is to fix the client application. c3p0 can help you debug, by letting you know where Connections are checked out that occasionally don't get checked in. In rare and unfortunate situations, development of the client application is closed, and even though it is buggy, you cannot fix it. c3p0 can help you work around the broken application, preventing it from exhausting the pool.
The following parameters can help you debug or workaround broken client applications.
unreturnedConnectionTimeout defines a limit (in seconds) to how long a Connection may remain checked out. If set to a nozero value, unreturned, checked-out Connections that exceed this limit will be summarily destroyed, and then replaced in the pool. Obviously, you must take care to set this parameter to a value large enough that all intended operations on checked out Connections have time to complete. You can use this parameter to merely workaround unreliable client apps that fail to close() Connections.
Much better than working-around is fixing. If, in addition to setting unreturnedConnectionTimeout, you set debugUnreturnedConnectionStackTraces to true, then a stack trace will be captured each time a Connection is checked-out. Whenever an unreturned Connection times out, that stack trace will be printed, revealing where a Connection was checked out that was not checked in promptly. debugUnreturnedConnectionStackTraces is intended to be used only for debugging, as capturing a stack trace can slow down Connection check-out.
c3p0 spawns a variety of Threads (helper threads, java.util.Timer threads), and does so lazily in response to the first client request experienced by a PooledDataSource. By default, the Threads spawned by c3p0 inherit a java.security.AccessControlContext and a contextClassLoader property from this first-calling Thread. If that Thread came from a client that may need to be hot-undeployed, references to these objects may prevent the undeployed application, often partitioned into a ClassLoader, from being garbage collected. (See for example this description of Tomcat memory leaks on redeployment.)
c3p0 provides two configuration parameters that can help with this:
contextClassLoaderSource should be set to one of caller, library, or none. The default (which yields the default behavior described above) is caller. Set this to library to use c3p0's ClassLoader, so that no reference is maintained to a client that may need to be redeployed.
privilegeSpawnedThreads is a boolean, false by default. Set this to true so that c3p0's Threads use the the c3p0 library's AccessControlContext, rather than an AccessControlContext that may be associated with the client application and prevent its garbage collection.
See Appendix A for information about the following configuration properties:
numHelperThreads and maxAdministrativeTaskTime help to configure the behavior of DataSource thread pools. By default, each DataSource has only three associated helper threads. If performance seems to drag under heavy load, or if you observe via JMX or direct inspection of a PooledDataSource, that the number of "pending tasks" is usually greater than zero, try increasing numHelperThreads. maxAdministrativeTaskTime may be useful for users experiencing tasks that hang indefinitely and "APPARENT DEADLOCK" messages. (See Appendix A for more.)
checkoutTimeout limits how long a client will wait for a Connection, if all Connections are checked out and one cannot be supplied immediately. usesTraditionalReflectiveProxies is of little practical use and is now formally deprecated. It permits you to use an old, now superceded implementation of C3P0-generated proxy objects. (C3P0 used to use reflective, dynamic proxies. Now, for enhanced performance, it uses code-generated, nonrefective implementations.) factoryClassLocation can be used to indicate where a URL from which c3p0 classes can be downloaded, if c3p0 DataSources will be retrieved as References from a JNDI DataSource by clients who do not have c3p0 locally installed.
If JMX libraries and a JMX MBeanServer are available in your environment (they are include in JDK 1.5 and above), you can inspect and configure your c3p0 datasources via a JMX administration tool (such as jconsole, bundled with jdk 1.5). You will find that c3p0 registers MBeans under com.mchange.v2.c3p0, one with statistics about the library as a whole (called C3P0Registry), and an MBean for each PooledDataSource you deploy. You can view and modify your DataSource's configuration properties, track the activity of Connection, Statement, and Thread pools, and reset pools and DataSources via the PooledDataSource MBean. (You may wish to view the API docs of PooledDataSource for documentation of the available operations.)
The name under which mbeans for c3p0 PooledDataSources will be registered includes the property dataSourceName. So, if you set this property, you can ensure that semanically equivalent data sources are identifiable across application restarts.
You can also customize the name under which the C3P0Registry will appear under JMX, which is useful if you wish to monitor multiple installations of c3p0 from a single MBean server. As a System property, in c3p0.properties, or in HOCON config, just set the following property (hopefully choosing a better name than 'myRegistryName'):
If you do not want c3p0 to register MBeans with your JMX environment, you can suppress this behavior with the following, set as a System property, in c3p0.properties, or in HOCON config:
c3p0 uses a custom logging library similar to jakarta commons-logging. Log messages can be directed to the popular log4j logging library, to slf4j (with its logback backend), to the standard logging facility introduced with jdk1.4, or to System.err. Nearly all configuration should be done at the level of your preferred logging library. There are a very few configuration options specific to c3p0's logging, and usually the defaults will be fine. Logging-related parameters may be placed in your c3p0.properties file, in HOCON configuration files, in a file called mchange-log.properties at the top-level of your classpath, or they may be defined as System properties. (The logging properties defined below may not be defined in c3p0-config.xml!) See the box below.
c3p0's logging behavior is affected by certain build-time options. If build-option c3p0.debug is set to false, all messages at a logging level below INFO will be suppressed. Build-option c3p0.trace controls how fine-grained c3p0's below INFO level reporting will be. For the moment, distributed c3p0 binaries are compiled with debug set to true and trace set to its maximum level of 10. But binaries may eventually be distributed with debug set to false. (For the moment, the performance impact of the logging level-checks seems very small, and it's most flexible to compile in all the messages, and let your logging library control which are emitted.) When c3p0 starts up, it emits the build-time values of debug and trace, along with the version and build time.
You can define named configurations which augment and override the default configuration that you define. When you instantiate a c3p0 PooledDataSource, whether via the ComboPooledDataSource constructor or via the DataSources factory class, you can supply a configuration name. For example, using ComboPooledDataSource:
Or using the DataSources factory class:
To define named configurations...
In a properties-style config file...
In a HOCON config file...
In an XML config file...
You can define overrides of default or named configurations that apply only to pools of Connections authenticated for a particular user. Not all configuration parameters support per-user overrides. See Appendix A for details.
To define per-user configurations...
In a properties-style config file...
In a HOCON config file...
In an XML config file...
Users can add their own configuration information, usually to customize the behavior of ConnectionCustomizers. User configuration is stored as a Map containing String keys and values, stored under the following configuration parameter:
The extensions Map can be set programatically like any other configuration parameter. However, there is special support for defining keys and values for the extensions Map in configuration files. In a properties-style config file...
In a HOCON config file...
In an XML config file...
To find the extensions defined for a PooledDataSource, you must have access to its identityToken, which is supplied as an argument to all ConnectionCustomizer methods. Given an identityToken, you can use the method C3P0Registry.extensionsForToken(...) to access the extensions Map.
Because extensions are primary designed to be used within ConnectionCustomizer implementations, the AbstractConnectionCustomizer class also defines a protected extensionsForToken(...) method as a convenience.
Here is an example ConnectionCustomizer implementation that makes use of user-defined configuration extensions. It defines an initSql extension, whose value should be a String containing SQL that should be executed when a Connection is checked out from the pool:
Named configurations, per-user overrides, and user-defined configuration extensions can easily be mixed.
In a properties-style config file...
In a HOCON config file...
In an XML config file...
Enhanced performance is the purpose of Connection and Statement pooling, and a major goal of the c3p0 library. For most applications, Connection pooling will provide a significant performance gain, especially if you are acquiring an unpooled Connection for each client access. If you are letting a single, shared Connection serve many clients to avoid Connection acquisition overhead, you may suffer performance issues and problems managing transactions when your Connection is under concurrent load; Connection pooling will enable you to switch to a one Connection-per-client model with little or no cost. If you are writing Enterprise Java Beans, you may be tempted to acquire a Connection once and not return it until the bean is about to be destroyed or passivated. But this can be resource-costly, as dormant pooled beans needlessly hold the Connection's network and database resources. Connection pooling permits beans to only "own" a Connection while they are using it.
But, there are performance costs to c3p0 as well. In order to implement automatic cleanup of unclosed ResultSets and Statements when parent resources are returned to pools, all client-visible Connections, ResultSets, Statements are really wrappers around objects provided by an underlying unpooled DataSource or "traditional" JDBC driver. Thus, there is some extra overhead to all JDBC calls.
Some attention has been paid to minimizing the "wrapper" overhead of c3p0. In my environment, the wrapper overhead amounts from several hundreths to several thousandths of the cost of Connection acquisition, so unless you are making many, many JDBC calls in fast succession, there will be a net gain in performance and resource-utilization efficiency. Significantly, the overhead associated with ResultSet operations (where one might iterate through a table with thousands of records) appears to be negligibly small.
Connections and Statements are pooled on a per-authentication basis. So, if one pool-backed DataSource is used to acquire Connections both for [user=alice, password=secret1] and [user=bob, password=secret2], there will be two distinct pools, and the DataSource might in the worst case manage twice the number of Connections specified by the maxPoolSize property.
This fact is a natural consequence of the definition of the DataSource spec (which allows Connections to be acquired with multiple user authentications), and the requirement that all Connections in a single pool be functionally identical. This "issue" will not be changed or fixed. It's noted here just so you understand what's going on.
The overhead of Statement pooling is too high. For drivers that do not perform significant preprocessing of PreparedStatements, the pooling overhead outweighs any savings. Statement pooling is thus turned off by default. If your driver does preprocess PreparedStatements, especially if it does so via IPC with the RDBMS, you will probably see a significant performance gain by turning Statement pooling on. (Do this by setting the configuration property maxStatements or maxStatementsPerConnection to a value greater than zero.).
Thank you for using c3p0!!!
The following properties can be set directly in code as JavaBeans properties, via a System properties or a c3p0.properties file (with c3p0. prepended to the property name), or in a c3p0-config.xml file. See the section on Configuration above. Click on the property name for a full description.
Normally, c3p0's configuration information is placed in a either a c3p0-config.xml or c3p0.properties file at the top-level of an application's CLASSPATH. However, if you wish to place configuration information elsewhere, you may place c3p0 configuration information (in the XML file format only!) anywhere you'd like in the filesystem visible to your application. Just set the following property to the full, absolute path of the XML config file:
If you set this property to a value beginning with "classloader:", c3p0 will search for an XML config file as a ClassLoader resource, that is, in any location you specify under your classpath, including jar-file META-INF directories.
The following properties affect c3p0's logging behavior. Please see Configuring Logging above for specific information.
The following property controls c3p0's JMX management interface. Plese see Configuring and Managing c3p0 via JMX above for more information.
Is it better to be beautiful or correct? Beginning with c3p0-0.9.1, c3p0 opts somewhat reluctantly for correctness.
Here's the deal. Every c3p0 DataSource is allocated a unique "identity token", which is used to ensure that multiple JNDI lookups of the same PooledDataSource always return the same instance, even if the JNDI name-server stores a Serialized or Referenced instance. Previously, c3p0 was happy for generated IDs to be unique within a single VM (and it didn't even get that quite right, before c3p0-0.9.1). But in theory, one VM might look up two different DataSources, generated by two different VMs, that by unlikely coincidence have the same "identity token", leading to errors as one of the two DataSources sneakily substitutes for the second. Though this hypothetical issue has never been reported in practice, c3p0 resolves it by prepending a VMID to its identity tokens. This makes them long and ugly, but correct.
If you don't like the long and ugly VMID, you can set your own, or you can turn off this solution to a hypothetical non-problem entirely with the following property:
Set it to NONE to turn off the VMID, set it to AUTO to let c3p0 generate a VMID, or provide any other String to set the VMID that will be used directly. The default is AUTO.
c3p0-0.9.1 included a new implementation of asynchronous Connection acquisition that should improve c3p0's performance and resource utilization in cases where database acquisition attempts, for whatever reason, occasionally fail. The new implementation should be significantly better than the "traditional" Connection acquisition strategy, but was added too late in the c3p0-0.9.1 development cycle to be fully tested and enabled by default. Users are encouraged to try the new implementation, both because it is better, and to help iron out any unanticipated problems.
For a full description of the new implementation and the resource bottleneck it is designed to overcome, please see the CHANGELOG entry for c3p0-0.9.1-pre11.
As of c3p0-0.9.2 this feature is enabled by default. To revert to the traditional Connection acquisition behavior, set the following parameter to false.
c3p0 configuration parameters can be set directly in Java code, via a simple Java properties file, via a Typesafe "HOCON" file, via an XML configuration file, or via System properties. Any which way works (the the XML configuration is most powerful, though, as it supports multiple named configurations and per-user overrides. Choose whatever works best for you.
To override the library's built-in defaults, create a file called c3p0.properties and place it at the "root" of your classpath or classloader. For a typical standalone application, that means place the file in a directory named in your CLASSPATH environment variable. For a typical web-application, the file should be placed in WEB-INF/classes. In general, the file must be available as a classloader resource under the name /c3p0.properties, in the classloader that loaded c3p0's jar file. Review the API docs (especilly getResource... methods) of java.lang.Class, java.lang.ClassLoader, and java.util.ResourceBundle if this is unfamiliar.
The format of c3p0.properties should be a normal Java Properties file format, whose keys are c3p0 configurable properties. See Appendix A. for the specifics. An example c3p0.properties file is produced below:
c3p0 does not include Typesafe's library, but if you bundle it with your application, c3p0 will support configuration in this format. You may place c3p0 configuration in the standard /application.conf or /reference.conf files, or you may use a special /c3p0.conf file. (These files must be placed as top-level ClassLoader resources, see above.) /application.json, /application.properties, and c3p0.json are also supported, and substitutions from /c3p0.properties will be resolved. The Typesafe config library can be downloaded from the Maven Central Repository.
Remember, HOCON configuration in only supported if you explicitly bundle the Typesafe config library with your application. The library is not included with c3p0's binary distribution, nor is it downloaded with c3p0 as a transitive dependency!
Here are some examples of setting c3p0 configuration in HOCON:
Note that you must specify ordinary config params explicitly inside a c3p0 scope one way or another, even in a c3p0.conf file. "Dot notation" can be used equivalently to scopes:
Also, the following two specifications are equivalent:
c3p0 properties can also be defined as System properties, using the same "c3p0." prefix for properties specified in a c3p0.properties file.
System properties override settings in c3p0.properties. Please see Precedence of Configuration Settings for more information.
You can use the XML config file for all c3p0 configuration, including configuration of defaults, named configurations, per-user overrides, and configuration extensions.
By default, c3p0 will look for an XML configuration file in its classloader's resource path under the name "/c3p0-config.xml". That means the XML file should be placed in a directly or jar file directly named in your applications CLASSPATH, in WEB-INF/classes, or some similar location.
If you prefer not to bundle your configuration with your code, you can specify an ordinary filesystem location for c3p0's configuration file via the system property com.mchange.v2.c3p0.cfg.xml. (You can also use this property to specify an alternative location in the ClassLoader resource path, e.g. META-INF. See Locating Configuration Information.)
Here is an example c3p0-config.xml file:
c3p0 now permits configuration parameters to be set in a number of different ways and places. Fortunately, there is a clear order of precedence that determines which configuration will "take" in the event of conflicting settings. Conceptually, c3p0 goes down this list from top to bottom, using the first setting it finds.
Most applications will never use per-user or named configurations. For these applications, we present a simplified precedence hierarchy:
For applications that do use named and per-user configurations, here is the complete, normative precedence hierarchy:
Hibernate's C3P0ConnectionProvider renames 7 c3p0 configuration properties, which, if set in your hibernate configuration, will override any configuration you may have set in a c3p0.properties file:
|c3p0-native property name||hibernate configuration key|
|c3p0.initialPoolSize||not available -- uses minimum size|
|c3p0.testConnectionOnCheckout||hibernate.c3p0.validate hibernate 2.x only!|
You can set any c3p0 properties in your hibernate config using the prefix hibernate.c3p0. For example
might be set to help debug Connection leaks.
You can always set c3p0 config in a c3p0.properties or c3p0-config.xml file (see "Overriding c3p0 defaults via c3p0.properties", "Overriding c3p0 defaults via c3p0-config.xml"), but any configuration set in Hibernate config files will override c3p0-native configuation!
You can easily configure Apache's Tomcat web application server to use c3p0 pooled DataSources. Below is a Tomcat 5.0 sample config to get you started. It's a fragment of Tomcat's conf/server.xml file, which should be modified to suit and placed inside a <Context> element.
For Tomcat 5.5, try something like the following (thanks to Carl F. Hall for the sample!):
The rest is standard J2EE stuff: You'll need to declare your DataSource reference in your web.xml file:
And you can access your DataSource from code within your web application like this:
To use c3p0 with JBoss:
Please note: As of c3p0-0.9.1, the class name of the jboss configuration mbean has changed to com.mchange.v2.c3p0.jboss.C3P0PooledDataSource (from com.mchange.v2.c3p0.mbean.C3P0PooledDataSource), in order to distinguish what is really jboss-specific functionality from c3p0's more general JMX support.
The old jboss config mbeans are deprecated, but will still work. However, support for new configuration parameters will only be added under the new name. Updating requires a one-word change to your c3p0-service.xml, change "mbean" to "jboss" where your old file says 'code="com.mchange.v2.c3p0.mbean.C3P0PooledDataSource"'. Just do it!
The Oracle thin JDBC driver provides a non-standard API for creating temporary BLOBs and CLOBs that requires users to call methods on the raw, Oracle-specific Connection implementation. Advanced users might use the raw connection operations described above to access this functionality, but a convenience class is available in a separate jar file (c3p0-oracle-thin-extras-0.9.5-pre6.jar) for easier access to this functionality. Please see the API docs for com.mchange.v2.c3p0.dbms.OracleUtils for details.