This changes dramatically when you need clustering, fail-over, load-balancing and few other buzz-words. There are several use-cases for that:
- single server cannot handle required number of concurrent, long running jobs and the executions need to be split into several machines – but each task must be executed exactly ones
- we cannot afford to run jobs too late – if one server is down, another should run the job on time
- …or less strictly – the job needs to run eventually – even if the one and only server was down for maintenance, delayed jobs need to be run as soon as possible after restart
In all cases above we need some sort of non-transient global storage to keep track which jobs were executed, so that they are run exactly ones by one machine. Relational database as a shared memory works good in this scenario.
So if you think you need to schedule jobs and have some of the requirements above, keep reading. I will show you how to configure Quartz with Spring and fully integrate them. First of all we need a DataSource:
import org.apache.commons.dbcp.BasicDataSource
import com.googlecode.flyway.core.Flyway
import org.jdbcdslog.DataSourceProxy
import org.springframework.jdbc.datasource.{DataSourceTransactionManager, LazyConnectionDataSourceProxy}
import org.h2.Driver
@Configuration
@EnableTransactionManagement
class Persistence {
@Bean
def transactionManager() = new DataSourceTransactionManager(dataSource())
@Bean
@Primary
def dataSource() = {
val proxy = new DataSourceProxy()
proxy.setTargetDSDirect(dbcpDataSource())
new LazyConnectionDataSourceProxy(proxy)
}
@Bean(destroyMethod = "close")
def dbcpDataSource() = {
val dataSource = new BasicDataSource
dataSource.setDriverClassName(classOf[Driver].getName)
dataSource.setUrl("jdbc:h2:mem:quartz-demo;DB_CLOSE_DELAY=-1;DB_CLOSE_ON_EXIT=FALSE;MVCC=TRUE")
dataSource.setUsername("sa")
dataSource.setPassword("")
dataSource.setMaxActive(20)
dataSource.setMaxIdle(20)
dataSource.setMaxWait(10000)
dataSource.setInitialSize(5)
dataSource.setValidationQuery("SELECT 1")
dataSource
}
}
As you might have guessed, Quartz needs some database tables to work with. It does not create them automatically, but SQL scripts for several databases are provided, including H2 which as you can see I am using. I think Flyway is the easiest way to run database scripts on startup:
@Bean(initMethod = "migrate")
def flyway() = {
val fly = new Flyway()
fly.setDataSource(dataSource())
fly
}
BTW in case you haven’t noticed: no, there is no XML in our sample application and yes, we are using Spring.
Let’s move on to Quartz:
@Configuration
class Scheduling {
@Resource
val persistence: Persistence = null
@Bean
@DependsOn(Array("flyway"))
def schedulerFactory() = {
val schedulerFactoryBean = new SchedulerFactoryBean()
schedulerFactoryBean.setDataSource(persistence.dataSource())
schedulerFactoryBean.setTransactionManager(persistence.transactionManager())
schedulerFactoryBean.setConfigLocation(new ClassPathResource("quartz.properties"))
schedulerFactoryBean.setJobFactory(jobFactory())
schedulerFactoryBean.setApplicationContextSchedulerContextKey("applicationContext")
schedulerFactoryBean.setSchedulerContextAsMap(Map().asJava)
schedulerFactoryBean.setWaitForJobsToCompleteOnShutdown(true)
schedulerFactoryBean
}
@Bean
def jobFactory() = new SpringBeanJobFactory
}
It is nice to know you can inject instance of @Configuration annotated classes into another such class for convenience. Except that – nothing fancy. Note that we need @DependsOn(Array(“flyway”)) on Quartz scheduler factory – otherwise the scheduler might start before Flyway fired the migration script with Quartz database tables causing unpleasant errors on startup. The essential bits are SpringBeanJobFactory and schedulerContextAsMap. The special factory makes Spring responsible for creating Job instances. Unfortunately this factory is quite limited which we will see shortly in the following example. First we need a Spring bean and a Quartz job:
@Service
class Printer extends Logging {
def print(msg: String) {
logger.info(msg)
}
}
class PrintMessageJob extends Job with Logging {
@BeanProperty
var printer: Printer = _
@BeanProperty
var msg = ""
def execute(context: JobExecutionContext) {
printer print msg
}
}
First unexpected input is @BeanProperty instead of @Autowired or @Resource. Turns out that Job is not really a Spring bean, even though Spring creates an instance of it. Instead Spring discovers required dependencies using available setters. So where does the msg string come from? Keep going:
val trigger = newTrigger().
withIdentity("Every-few-seconds", "Demo").
withSchedule(
simpleSchedule().
withIntervalInSeconds(4).
repeatForever()
).
build()
val job = newJob(classOf[PrintMessageJob]).
withIdentity("Print-message", "Demo").
usingJobData("msg", "Hello, world!").
build()
scheduler.scheduleJob(job, trigger)
Quartz 2.0 ships with a nice internal DSL for creating jobs and triggers in a readable manner. As you can see I am passing an extra “Hello, world!” parameter to the job. This parameter is stored in so called JobData in the database per job or per trigger. It will be provided to the job when it is executed. This way you can parametrize your jobs. However when executed our job throws NullPointerException… Apparently printer reference was not set and silently ignored. Turns out Spring won’t simply look through all the beans available in the ApplicationContext. The SpringBeanJobFactory only looks into Jobs’ and Triggers’ JobData and into so called scheduler context (already mentioned). If you want to inject any Spring bean into Job you must explicitly place a reference to that bean in schedulerContext:
@Configuration
class Scheduling {
@Resource
val printer: Printer = null
@Bean
def schedulerFactory() = {
val schedulerFactoryBean = new SchedulerFactoryBean()
//...
schedulerFactoryBean.setSchedulerContextAsMap(schedulerContextMap().asJava)
//...
schedulerFactoryBean
}
def schedulerContextMap() =
Map(
"printer" -> printer
)
}
Unfortunately each and every Spring bean you want to inject to job has to be explicitly referenced in schedulerContextMap. Even worse, if you forget about it, Quartz will silently log NPE at runtime. In the future we will write more robust job factory. But for starters we have a working Spring + Quartz application ready for further experiments, sources as always available under my GitHub account.
You might ask yourself way can’t we simply use MethodInvokingJobDetailFactoryBean? Well, first of all because it does not work with persistent job stores. Secondly – because it is unable to pass JobData to Job – so we can no longer distinguish between different job runs. For instance our job printing message would have to always print the same message hard-coded in the class.
BTW if anyone asks you: How many classes does a Java enterprise developer need to print “Hello world!” you can proudly reply: 4 classes, 30 JARs taking 20 MiB of space and a relational database with 10+ tables. Seriously, this is an output of our article here…
Reference: Configuring Quartz with JDBCJobStore in Spring from our JCG partner Tomasz Nurkiewicz at the Java and neighbourhood blog.
