NullPointerException on concurrent requests via async CoapClient#advanced

[eugene-nikolaev]

Time-to-time my tests were failing with the following stacktrace:

32 INFO [CoapEndpoint]: Starting endpoint at 0.0.0.0/0.0.0.0:0 - (org.eclipse.californium.core.network.CoapEndpoint.java:115) start() in thread Thread-6 at (2016-06-15 19:46:40)
    at org.eclipse.californium.core.network.CoapEndpoint.runInProtocolStage(CoapEndpoint.java:692)
    at org.eclipse.californium.core.network.CoapEndpoint.sendRequest(CoapEndpoint.java:390)
    at org.eclipse.californium.core.CoapClient.send(CoapClient.java:897)
    at org.eclipse.californium.core.CoapClient.send(CoapClient.java:879)
    at org.eclipse.californium.core.CoapClient.asynchronous(CoapClient.java:742)
    at org.eclipse.californium.core.CoapClient.advanced(CoapClient.java:663)
    at <SKIPPED>
    at java.lang.Thread.run(Thread.java:745)
32 INFO [EndpointManager]: Created implicit default endpoint 0.0.0.0/0.0.0.0:39102 - (org.eclipse.californium.core.network.EndpointManager.java:98) createDefaultEndpoint() in thread Thread-6 at (2016-06-15 19:46:40)

The code made concurrent asynchronous requests via CoapClient using default endpoint (no endpoint was specified).

After some debug, I discovered a race condition.

EndpointManager:

   public Endpoint getDefaultEndpoint() {
        if (default_endpoint == null) {
            createDefaultEndpoint();
        }
        return default_endpoint;
    }

    private synchronized void createDefaultEndpoint() {
        if (default_endpoint != null) return;

        default_endpoint = new CoapEndpoint();

        try {
            default_endpoint.start();
                        ...

CoapEndpoint:

    public synchronized void start() throws IOException {
        ....

        if (this.executor == null) {
                             ...
            setExecutor(Executors.newSingleThreadScheduledExecutor(
                    new Utils.DaemonThreadFactory("CoapEndpoint-" + connector.getAddress() + '#'))); //$NON-NLS-1$
                 ...

     private void runInProtocolStage(final Runnable task) {
        executor.execute(new Runnable() {
                            ...

So, when two threads invoked CoapClient#advanced(CoapHandler, Request) they called EndpointManager#getDefaultEndpoint somewhere down in the hierarchy.

getDefaultEndpoint invoked synchronized createDefaultEndpoint but as long as getDefaultEndpoint is not synchronized, second thread got default_endpoint != null and proceeded. But the default_endpoint was not started yet so CoapEndpoint#executor was not filled.

It is actually a cause of that NPE.
createDefaultSecureEndpoint is not synchronized too.
I am preparing a PR (actually just adding two 'synchronized') now.
But I am not sure it is possible to make a unit test for this case.