The point of this project is to provide a simplified way of accessing databases. It is a wrapper around the JDBC driver, and tries to hide some of the more error-prone, unsafe, and non-portable parts of the standard API. It uses standard Java types for all operations (as opposed to java.sql.*), and acts as a compatibility layer in making every attempt to behave consistently with all supported databases.
The operations supported are those typical of relational databases, and are expressed as SQL. This is NOT an object-relational mapping layer or an attempt to create a new query language.
If you are looking for convenient utilities built on top of this library, try https://github.com/susom/database-goodies.
Connections, prepared statements, and result sets are hidden so there is no opportunity for client code to make a mistake and cause resource leaks. For example, the following code is complete and correct with respect to resources, exceptions, and transactions.
String url = "jdbc:hsqldb:file:hsqldb;shutdown=true";
Builder dbb = DatabaseProvider.fromDriverManager(url);
dbb.transact(db -> {
String s = db.get().toSelect("select s from t where i=?")
.argInteger(5)
.queryOneOrThrow(r -> r.getString());
System.out.println("Result: " + s);
});
This style of using callbacks also fits nicely with asynchronous programming models. Support for Vert.x is included (use DatabaseProviderVertx). There is also a simple Vert.x server example and a more sophisticated example with concurrency.
Annotations are included so you can use the Checker Framework static analysis tool to prove there are no SQL Injection vulnerabilities in your application. Make your automated build fail immediately when a vulnerability is introduced.
// Checker will fail the build for this
db.toSelect("select a from t where b=" + userInput).query(...);
Of course, there are times when you need to dynamically construct SQL, so there is a safe way to do that as well:
Sql sql = new Sql();
sql.append("select a from b where c=?").argInteger(1);
if (d) {
sql.append(" and d=?").argString("foo");
// Note the following would also fail with Checker
//sql.append(" and d=" + userInput);
}
db.toSelect(sql).query(...);
Internal connection pooling is included, leveraging the excellent HikariCP library.
String url = "jdbc:hsqldb:file:hsqldb;shutdown=true";
Config config = Config.from().value("database.url", url).get();
Builder dbb = DatabaseProvider.pooledBuilder(config);
for (...) {
dbb.transact(db -> {
...
});
}
// ... much later, on shutdown
dbb.close();
Due to the normal lifetime of a connection pool, you are obligated to explicitly shutdown the pool (for example, in a JVM shutdown handler) when it is no longer needed.
The various argX()
calls know the type of the object you intend to pass, and can
therefore handle null values correctly. No more errors because you pass a null and the JDBC
driver can't figure out what type it should be.
db.toInsert("insert into foo (bar) values (?)").argLong(maybeNull).insert(1);
You can use traditional positional parameters in the SQL (the '?' character), or you can use named parameters. This can help reduce errors due to counting incorrectly.
db.toUpdate("update foo set bar=?").argLong(23L).update();
db.toUpdate("update foo set bar=:baz").argLong("baz", 23L).update();
You can use both positional and named within the same SQL statement. The positional
parameters must be in the correct order, but the arg*()
calls for the named
parameters can be mixed anywhere among the positional ones.
db.toSelect("select c from t where a=:a and b=?")
.argString("value for b")
.argString(":a", "value for a")
.queryLongOrNull();
All SQLExceptions are wrapped into a DatabaseException that inherits from RuntimeException. This makes code much cleaner because in server programming there is usually no way to recover from an error (it is handled in a generic way by higher level code).
As long as your database columns have enough precision, Date objects will round-trip correctly with millisecond precision. No more fiddling with Timestamp and dealing with millisecond truncation and nanoseconds.
Date now = new Date(); // java.util.Date
db.toInsert("insert into t (pk,d) values (?,?)")
.argInteger(123)
.argDate(now)
.insert(1);
Date sameNow = db.toSelect("select d from t where pk=?")
.argInteger(123)
.queryDateOrNull();
if (now.equals(sameNow)) {
// don't look so surprised...
}
There is also a convenient way to deal with "now", which hides the new Date()
call
within the configurable Options
. This is handy for testing because you can explicitly
control and manipulate the clock.
db.toInsert(...).argDateNowPerApp().insert();
Since every database seems to have a different way of dealing with time, this library also tries to smooth out some of the syntactic (and semantic) differences in using time according to the database server (the operating system time).
db.toInsert(...).argDateNowPerDb().insert();
For Oracle the above code will substitute systimestamp(3)
for the parameter, while for PostgreSQL
the value date_trunc('milliseconds',localtimestamp)
will be used. To make testing easier,
there is also a configuration option to make the above code do exactly the same thing as
argDateNowPerApp()
so you can in effect control the database server time as well as that
of the application server.
There are use cases where a true database date is more appropriate than a timestamp, e.g., for recording date of birth or other fields where time is not really relevant or known. The LocalDate API is stored as a database Date, not a Timestamp, for modern databases that support a true date type.
LocalDate has been implemented and tested using ISO 8601 format YYYY-MM-DD. There should be no time or timezone associated.
Postgres and MS SQLServer have implemented a fully compliant LocalDate that is consistentently returned as java.sql.Date. There are no known issues with these databases.
Oracle does not support a true date type; the Oracle Date is actually implemented as a timestamp. Oracle supports LocalDate by using a time of midnight and a timezone of 0. As a result, there are times when the database type returned have been timestamp.
HSQLDB also does not support a true date type - it's date implementation uses a timestamp of 0 (midnight), but has a bug that sets timezone. For this reason, if a date is stored from one timezone (e.g., system default timezone), and a user attempts to query for that date from another timezone, it may not find the value. For this reason, HSQLDB is not recommended for LocalDate type data if you are working across time zones.
To work around old database implementations of LocalDate as a Timestamp, when a timestamp is
received, the scale attribute is examined. A scale of 0 on a Timestamp is always associated
with a LocalDate across all tested DB implementations; true timestamps always have a non-zero scale.
When a scale of zero is found, it is handled as LocalDate.
More information on specific implementations is available here:
- Derby: https://db.apache.org/derby/papers/JDBCImplementation.html#Derby+SQL+DATE
- HSQLDB: http://www.h2database.com/html/datatypes.html
- known problem:http://www.h2database.com/html/datatypes.html
- Oracle: https://docs.oracle.com/javase/8/docs/api/java/sql/Timestamp.html
Similarly BigDecimal tries to maintain scale in a more intuitive manner (most drivers will revert to "full precision" meaning they pad the scale out to what the database column specifies).
Deal with them explicitly as either String/byte[] or streams. No downcasting or driver-specific APIs, and treat them the same as other parameters.
Tracks important metrics and logs to SLF4J in a way that is cleaner and gives you more control than having database-related logging scattered throughout your code. The logging is customizable so you can choose to see substituted parameters as well.
Get database: 393.282ms(getConn=389.948ms,checkAutoCommit=1.056ms,dbInit=2.273ms)
DDL: 15.658ms(prep=8.017ms,exec=7.619ms,close=0.021ms) create table dbtest (a numeric)
Insert: 71.295ms(prep=65.093ms,exec=6.153ms,close=0.048ms) insert into dbtest (a) values (?) ParamSql: insert into dbtest (a) values (23)
Query: 38.627ms(prep=27.642ms,exec=9.846ms,read=1.013ms,close=0.125ms) select count(1) from dbtest
It is often a good idea to generate primary keys from a sequence, but this is not always easy to do in a clean and efficient way. This library provides a way to do it that will be efficient on advanced databases, and still be able to transparently fall back to multiple database calls when necessary.
Long pk = db.toInsert("insert into t (pk,s) values (?,?)")
.argPkSeq("pk_seq")
.argString("Hi")
.insertReturningPkSeq("pk");
This has a more general form for returning multiple columns. For example, if you inserted a database timestamp and need that value as well to update an object in memory:
db.toInsert("insert into t (pk,d,s) values (?,?,?)")
.argPkSeq("pk_seq")
.argDateNowPerDb()
.argString("Hi")
.insertReturning("t", "pk", rs -> {
...
if (rs.next()) {
... = rs.getLongOrNull(1); // value of pk
... = rs.getDateOrNull(2); // value of d
}
...
}, "d");
Built to make life easier in modern IDEs. Everything you need is accessed from a single interface (Database).
Methods within the library have also been annotated to help IDEs like IntelliJ provide better support. For example, it can warn you about checking nulls, or forgetting to use a return value on a fluent API. Try using the error-prone plugin and/or build tools in your project.
You can define your database schema using a simple Java API and execute the database specific DDL. When defining this schema you use the same basic Java types you use when querying, and appropriate database-specific column types will be chosen such that data will round-trip correctly. This API also smooths over some syntax differences like sequence creation.
// Observe that this will work across the supported databases, with
// specific syntax and SQL types tuned for that database.
new Schema()
.addTable("t")
.addColumn("pk").primaryKey().table()
.addColumn("d").asDate().table()
.addColumn("s").asString(80).table().schema()
.addSequence("pk_seq").schema().execute(db);
Basic example including setup:
String url = "jdbc:hsqldb:file:hsqldb;shutdown=true";
DatabaseProvider.fromDriverManager(url).transact(dbs -> {
Database db = dbs.get();
db.dropTableQuietly("t");
new Schema().addTable("t").addColumn("a").asInteger().schema().execute(db);
db.toInsert("insert into t (a) values (?)").argInteger(32).insert(1);
db.toUpdate("update t set a=:val").argInteger("val", 23).update();
Long rows = db.toSelect("select count(1) from t").queryLongOrNull();
System.out.println("Rows: " + rows);
});
Note the lack of error handling, resource management, and transaction calls. This is not because it is left as an exercise for the reader, but because it is handled automatically.
For a more realistic server-side example, a container will usually manage creation of the Database or Supplier, and business layer code will declare a dependency on this:
public class MyBusiness {
private Supplier<Database> db;
public MyBusiness(Supplier<Database> db) {
this.db = db;
}
public Long doStuff(String data) {
if (isCached(data)) {
// Note how this might never allocate a database connection
return cached(data);
}
return db.get().toSelect("select count(*) from a where b=:data")
.argString("data", data).queryLong();
}
// Note we use only java.util.Date, not java.sql.*
public List<Date> doMoreStuff(Date after) {
return db.get().toSelect("select my_date from a where b > ?")
.argDate(after).queryMany(r -> rs.getDateOrNull("my_date"));
}
}
Of course there are also convenience methods for simple cases like this having only one column in the result:
public List<Date> doMoreStuff(Date after) {
return db.get().toSelect("select my_date from a where b > ?")
.argDate(after).queryDates();
}
The library is available in the public Maven repository:
<dependency>
<groupId>com.github.susom</groupId>
<artifactId>database</artifactId>
<version>4.0</version>
</dependency>
Just add that to your pom.xml, use one of the static builder methods on
com.github.susom.database.DatabaseProvider
(see example above), and enjoy!
To see more examples of how to use the library, take a look at some of the tests:
There are also a variety of samples in the demo directory.
The functionality is currently tested with Oracle, PostgreSQL, HyperSQL (HSQLDB), SQL Server, and Derby. It won't work out of the box with other databases right now, but might in the future. If you want pure Java, use HyperSQL rather than Derby. If you are really into MySQL/MariaDB, take a look at the "mysql" branch, but be warned it is early stages and got a little stuck because of significant feature gaps in that database (e.g. no sequences). There is basic support for read-only usage of BigQuery using the Google-recommended Magnitude Simba JDBC drivers. The BigQuery driver supports some DDL and DML statements, but BigQuery does not support transactions, sequences, constraints, or indexes.
The library is compiled and tested with Java 8, so it won't work with Java 7 and earlier. If you really must use Java 7, grab the latest 1.x release of this library.
No fancy things with results (e.g. scrolling or updating result sets).