/** @param log Logger. */
private void execute(GridLogger log) {
try {
log.info("Started execute.");
// Countdown shared job latch so that the main thread know that all jobs are
// inside the "execute" routine.
jobLatch.countDown();
log.info("After job latch.");
// Await for the main thread to allow jobs to proceed.
latch.await();
log.info("After latch.");
if (awaitMasterLeaveCallback) {
latch0.await();
log.info("After latch0.");
} else log.info("Latch 0 skipped.");
} catch (InterruptedException e) {
// We do not expect any interruptions here, hence this statement.
fail("Unexpected exception: " + e);
}
}
/** @throws Exception If failed. */
@SuppressWarnings("unchecked")
public void testCancel() throws Exception {
Grid grid = G.grid(getTestGridName());
grid.compute()
.localDeployTask(GridCancelTestTask.class, U.detectClassLoader(GridCancelTestTask.class));
GridComputeTaskFuture<?> fut = grid.compute().execute(GridCancelTestTask.class.getName(), null);
// Wait until jobs begin execution.
boolean await = startSignal.await(WAIT_TIME, TimeUnit.MILLISECONDS);
assert await : "Jobs did not start.";
info("Test task result: " + fut);
assert fut != null;
// Only first job should successfully complete.
Object res = fut.get();
assert (Integer) res == 1;
// Wait for all jobs to finish.
await = stopSignal.await(WAIT_TIME, TimeUnit.MILLISECONDS);
assert await : "Jobs did not stop.";
// One is definitely processed. But there might be some more processed or cancelled or processed
// and cancelled.
// Thus total number should be at least SPLIT_COUNT and at most (SPLIT_COUNT - 1) *2 +1
assert (cancelCnt + processedCnt) >= SPLIT_COUNT
&& (cancelCnt + processedCnt) <= (SPLIT_COUNT - 1) * 2 + 1
: "Invalid cancel count value: " + cancelCnt;
}
/**
* @param expJobs Expected jobs number.
* @param taskStarter Task started.
* @throws Exception If failed.
*/
private void testMasterLeaveAwareCallback(
int expJobs, GridClosure<GridProjection, GridFuture<?>> taskStarter) throws Exception {
jobLatch = new CountDownLatch(expJobs);
invokeLatch = new CountDownLatch(expJobs);
for (int i = 0; i < GRID_CNT; i++) startGrid(i);
int lastGridIdx = GRID_CNT - 1;
GridFuture<?> fut = taskStarter.apply(grid(lastGridIdx).forPredicate(excludeLastPredicate()));
jobLatch.await();
stopGrid(lastGridIdx, true);
latch.countDown();
assert invokeLatch.await(5000, MILLISECONDS);
try {
fut.get();
} catch (GridException e) {
log.debug("Task failed: " + e);
}
}
/**
* Ensure that {@link GridComputeJobMasterLeaveAware} callback is invoked when fails to send
* {@link GridJobExecuteResponse} to master node.
*
* @throws Exception If failed.
*/
public void testCannotSendJobExecuteResponse() throws Exception {
awaitMasterLeaveCallback = false;
// Start grids.
for (int i = 0; i < GRID_CNT; i++) startGrid(i);
int lastGridIdx = GRID_CNT - 1;
grid(lastGridIdx)
.forPredicate(excludeLastPredicate())
.compute()
.execute(new TestTask(GRID_CNT - 1), null);
jobLatch.await();
for (int i = 0; i < lastGridIdx; i++)
((CommunicationSpi) grid(i).configuration().getCommunicationSpi()).waitLatch();
latch.countDown();
// Ensure that all worker nodes has already started job response sending.
for (int i = 0; i < lastGridIdx; i++)
((CommunicationSpi) grid(i).configuration().getCommunicationSpi()).awaitResponse();
// Now we stop master grid.
stopGrid(lastGridIdx, true);
// Release communication SPI wait latches. As master node is stopped, job worker will receive
// and exception.
for (int i = 0; i < lastGridIdx; i++)
((CommunicationSpi) grid(i).configuration().getCommunicationSpi()).releaseWaitLatch();
assert invokeLatch.await(5000, MILLISECONDS);
}
/**
* Ensure that {@link GridComputeJobMasterLeaveAware} callback is invoked on job which is
* initiated by master and is currently running on it.
*
* @throws Exception If failed.
*/
public void testLocalJobOnMaster() throws Exception {
invokeLatch = new CountDownLatch(1);
jobLatch = new CountDownLatch(1);
Grid g = startGrid(0);
g.compute().execute(new TestTask(1), null);
jobLatch.await();
// Count down the latch in a separate thread.
new Thread(
new Runnable() {
@Override
public void run() {
try {
U.sleep(500);
} catch (GridInterruptedException ignore) {
// No-op.
}
latch.countDown();
}
})
.start();
stopGrid(0, true);
latch.countDown();
assert invokeLatch.await(5000, MILLISECONDS);
}
/** @throws Exception If failed. */
public void testTopologyListener() throws Exception {
final Collection<UUID> added = new ArrayList<>(1);
final Collection<UUID> rmvd = new ArrayList<>(1);
final CountDownLatch addedLatch = new CountDownLatch(1);
final CountDownLatch rmvLatch = new CountDownLatch(1);
assertEquals(NODES_CNT, client.compute().refreshTopology(false, false).size());
GridClientTopologyListener lsnr =
new GridClientTopologyListener() {
@Override
public void onNodeAdded(GridClientNode node) {
added.add(node.nodeId());
addedLatch.countDown();
}
@Override
public void onNodeRemoved(GridClientNode node) {
rmvd.add(node.nodeId());
rmvLatch.countDown();
}
};
client.addTopologyListener(lsnr);
try {
Grid g = startGrid(NODES_CNT + 1);
UUID id = g.localNode().id();
assertTrue(addedLatch.await(2 * TOP_REFRESH_FREQ, MILLISECONDS));
assertEquals(1, added.size());
assertEquals(id, F.first(added));
stopGrid(NODES_CNT + 1);
assertTrue(rmvLatch.await(2 * TOP_REFRESH_FREQ, MILLISECONDS));
assertEquals(1, rmvd.size());
assertEquals(id, F.first(rmvd));
} finally {
client.removeTopologyListener(lsnr);
stopGrid(NODES_CNT + 1);
}
}
/** {@inheritDoc} */
@Override
public Serializable execute() {
int arg = this.<Integer>argument(0);
try {
if (log.isInfoEnabled()) log.info("Executing job [job=" + this + ", arg=" + arg + ']');
startSignal.countDown();
try {
if (!startSignal.await(WAIT_TIME, TimeUnit.MILLISECONDS)) fail();
if (arg == 1) {
if (log.isInfoEnabled()) log.info("Job one is proceeding.");
} else Thread.sleep(WAIT_TIME);
} catch (InterruptedException e) {
if (log.isInfoEnabled())
log.info("Job got cancelled [arg=" + arg + ", ses=" + ses + ", e=" + e + ']');
return 0;
}
if (log.isInfoEnabled()) log.info("Completing job: " + ses);
return argument(0);
} finally {
stopSignal.countDown();
processedCnt++;
}
}
/**
* Ensure that {@link GridComputeJobMasterLeaveAware} callback is invoked when master node leaves
* topology normally.
*
* @throws Exception If failed.
*/
public void testMasterStoppedNormally() throws Exception {
// Start grids.
for (int i = 0; i < GRID_CNT; i++) startGrid(i);
int lastGridIdx = GRID_CNT - 1;
grid(lastGridIdx)
.forPredicate(excludeLastPredicate())
.compute()
.execute(new TestTask(GRID_CNT - 1), null);
jobLatch.await();
stopGrid(lastGridIdx, true);
latch.countDown();
assert invokeLatch.await(5000, MILLISECONDS);
}
/**
* Ensure that {@link GridComputeJobMasterLeaveAware} callback is invoked when master node leaves
* topology abruptly (e.g. due to a network failure or immediate node shutdown).
*
* @throws Exception If failed.
*/
public void testMasterStoppedAbruptly() throws Exception {
// Start grids.
for (int i = 0; i < GRID_CNT; i++) startGrid(i);
int lastGridIdx = GRID_CNT - 1;
grid(lastGridIdx)
.forPredicate(excludeLastPredicate())
.compute()
.execute(new TestTask(GRID_CNT - 1), null);
jobLatch.await();
((CommunicationSpi) grid(lastGridIdx).configuration().getCommunicationSpi()).blockMessages();
stopGrid(lastGridIdx, true);
latch.countDown();
assert invokeLatch.await(5000, MILLISECONDS);
}