/**
* Creates a new instance given a configuration and a load balancer.
*
* @param config The configuration details for this job scheduler.
* @param balancer The load balancer to use for distributing jobs.
*/
public MeandreJobScheduler(MeandreJobSchedulerConfig config, MeandreLoadBalancer balancer) {
loadBalancer = balancer;
workers = config.getServers();
for (MeandreServerProxy server : workers) {
loadBalancer.addServer(server);
}
}
/** Initializes this instance. */
@PostConstruct
public void init() {
assert config != null : "No configuration was provided to the job scheduler.";
workers = config.getServers();
for (MeandreServerProxy server : workers) {
loadBalancer.addServer(server);
}
}
/**
* Remove a server from the job scheduler.
*
* @param server Server to remove.
*/
public void removeServer(MeandreServerProxy server) {
workersLock.lock();
try {
workers.remove(server);
loadBalancer.removeServer(server);
} finally {
workersLock.unlock();
}
}
/** Attempt to run any queued jobs. */
public void runJobs() {
Session session = null;
JobDao jobDao = daoFactory.getJobDao();
try {
queueLock.lock();
workersLock.lock();
if (jobQueue.size() < 1) {
logger.fine("No queued jobs.");
return;
}
session = jobDao.getSessionFactory().openSession();
jobDao.startManagedSession(session);
while (!jobQueue.isEmpty()) {
logger.fine("Found " + jobQueue.size() + " queued jobs.");
MeandreServerProxy server = loadBalancer.nextAvailableServer();
if (server == null) {
logger.info("All servers are busy. Will try again in " + POLL_PERIOD + " seconds.");
return;
}
logger.fine("Server " + server + " is available for processing.");
Job job = jobQueue.peek();
job.incrementNumTries();
job.setJobStatus(JobStatus.SUBMITTED);
job.setSubmitTimestamp(new Date());
logger.fine("Preparing to update job " + job.getId() + " as submitted.");
Transaction transaction = session.beginTransaction();
transaction.begin();
try {
jobDao.makePersistent(job);
transaction.commit();
logger.fine("Job " + job.getId() + " updated.");
} catch (HibernateException e) {
logger.warning("Data access exception: " + e.getMessage());
rollback(transaction);
return;
} catch (DataAccessException e) {
logger.warning("Data access exception: " + e.getMessage());
rollback(transaction);
return;
} catch (Exception e) {
logger.warning(e.getMessage());
rollback(transaction);
return;
}
try {
logger.fine("Attempting to contact server " + server + " to execute job.");
ExecResponse response = server.executeJob(job);
logger.fine("Execution response received.");
// If the executeJob() method above succeeded, the Meandre
// server will have (most likely) changed the job status to
// "started". If the status changes, the NEMA probe running
// on the Meandre server will have written the new status to
// the same database the flow service uses. Therefore, we
// want to refresh the state of the job here to pick up the
// status change. Otherwise, the old status (submitted) will
// be rewritten to the database, and the "started" state
// will be lost.
session.refresh(job);
logger.fine("Attempting to record job execution response.");
job.setHost(server.getMeandreServerProxyConfig().getHost());
job.setPort(server.getMeandreServerProxyConfig().getPort());
job.setExecPort(response.getPort());
job.setExecutionInstanceId(response.getUri());
transaction = session.beginTransaction();
transaction.begin();
try {
jobDao.makePersistent(job);
transaction.commit();
logger.fine("Job execution response recorded in database.");
jobQueue.remove();
} catch (HibernateException e) {
logger.warning("Data access exception: " + e.getMessage());
rollback(transaction);
return;
} catch (DataAccessException e) {
logger.warning("Data access exception: " + e.getMessage());
rollback(transaction);
return;
} catch (Exception e) {
logger.warning(e.getMessage());
rollback(transaction);
return;
}
} catch (MeandreServerException serverException) {
logger.warning(serverException.getMessage());
job.setSubmitTimestamp(null);
job.setJobStatus(JobStatus.SCHEDULED);
if (job.getNumTries() == MAX_EXECUTION_TRIES) {
logger.info(
"Unsuccessfully tried "
+ MAX_EXECUTION_TRIES
+ " times to execute job "
+ job.getId()
+ ". Will mark job as failed.");
job.setJobStatus(JobStatus.FAILED);
job.setEndTimestamp(new Date());
job.setUpdateTimestamp(new Date());
jobQueue.remove();
}
transaction = session.beginTransaction();
transaction.begin();
try {
jobDao.makePersistent(job);
transaction.commit();
} catch (HibernateException e) {
logger.warning("Data access exception: " + e.getMessage());
rollback(transaction);
return;
} catch (DataAccessException e) {
logger.warning("Data access exception: " + e.getMessage());
rollback(transaction);
return;
} catch (Exception e) {
logger.warning(e.getMessage());
rollback(transaction);
return;
}
}
}
} catch (Exception e) {
e.printStackTrace();
} finally {
workersLock.unlock();
queueLock.unlock();
if (session != null) {
try {
jobDao.endManagedSession();
session.close();
} catch (HibernateException e) {
logger.warning(e.getMessage());
}
}
}
}
