views:

204

answers:

3

Hi,

I am new to Java and junit. I have the following peice of code that I want to test. Would appreciate if you could send your ideas about what's the best way to go about testing it.

Basically, the following code is about electing a leader form a Cluster. The leader holds a lock on the shared cache and services of the leader get resumed and disposed if it somehow looses the lock on the cache.

How can i make sure that a leader/thread still holds the lock on the cache and that another thread cannot get its services resumed while the first is in execution?

public interface ContinuousService {

public void resume();
public void pause();
}


public abstract class ClusterServiceManager {
private volatile boolean leader = false;
private volatile boolean electable = true;
private List<ContinuousService> services;

protected synchronized void onElected() {
    if (!leader) {
        for (ContinuousService service : services) {
            service.resume();
        }
        leader = true;
    }
}

protected synchronized void onDeposed() {
    if (leader) {
        for (ContinuousService service : services) {
            service.pause();
        }
        leader = false;
    }
}

public void setServices(List<ContinuousService> services) {
    this.services = services;
}

@ManagedAttribute
public boolean isElectable() {
    return electable;
}

@ManagedAttribute
public boolean isLeader() {
    return leader;
}



public class TangosolLeaderElector extends ClusterServiceManager implements Runnable {
private static final Logger log = LoggerFactory.getLogger(TangosolLeaderElector.class);
private String election;
private long electionWaitTime= 5000L;

private NamedCache cache;

public void start() {
    log.info("Starting LeaderElector ({})",election);
    Thread t = new Thread(this, "LeaderElector ("+election+")");
    t.setDaemon(true);
    t.start();
}

public void run() {
    // Give the connection a chance to start itself up
    try {
        Thread.sleep(1000);
    } catch (InterruptedException e) {}

    boolean wasElectable = !isElectable();
    while (true) {
        if (isElectable()) {
            if (!wasElectable) {
                log.info("Leadership requested on election: {}",election);
                wasElectable = isElectable();
            }
            boolean elected = false;
            try {
                // Try and get the lock on the LeaderElectorCache for the current election
                if (!cache.lock(election, electionWaitTime)) {
                    // We didn't get the lock. cycle round again. 
                    // This code to ensure we check the electable flag every now & then
                    continue;
                }
                elected = true;
                log.info("Leadership taken on election: {}",election);
                onElected();

                // Wait here until the services fail in some way.
                while (true) {
                    try {
                        Thread.sleep(electionWaitTime);
                    } catch (InterruptedException e) {}
                    if (!cache.lock(election, 0)) {
                        log.warn("Cache lock no longer held for election: {}", election);
                        break;
                    } else if (!isElectable()) {
                        log.warn("Node is no longer electable for election: {}", election);
                        break;
                    }
                    // We're fine - loop round and go back to sleep.
                }
            } catch (Exception e) {
                if (log.isErrorEnabled()) {
                    log.error("Leadership election " + election + " failed (try bfmq logs for details)", e);
                }
            } finally {
                if (elected) {
                    cache.unlock(election);
                    log.info("Leadership resigned on election: {}",election);
                    onDeposed();
                }
                // On deposition, do not try and get re-elected for at least the standard wait time.
                try { Thread.sleep(electionWaitTime); } catch (InterruptedException e) {}
            }
        } else {
            // Not electable - wait a bit and check again.
            if (wasElectable) {
                log.info("Leadership NOT requested on election ({}) - node not electable",election);
                wasElectable = isElectable();
            }
            try {
                Thread.sleep(electionWaitTime);
            } catch (InterruptedException e) {}
        }
    }
}

public void setElection(String election) {
    this.election = election;
}

@ManagedAttribute
public String getElection() {
    return election;
}

public void setNamedCache(NamedCache nc) {
    this.cache = nc;
}
+1  A: 

If you are not too particular about using JUnit, then you can use TestNG framework. They have multi threaded support.

Kartik
+1  A: 

There's an example here that I made use of a while back.

sal
A: 

As an alternative to testing frameworks (or to extra extensions on top of JUnit, if you are using that) is just some plain old code:

  • Create several threads and apply them to this routine.
  • Loop through the threads and test each one until you find out who is the leader.
  • Apply different environmental changes (including the passage of time) to your program's state and redo the test. Is the leader still the leader?
  • Now, force the leader to abdicate (kill that thread or something). Did another thread take over?
Greg Harman