@Override
public void releaseSlot() {
// try to transition to the CANCELED state. That state marks
// that the releasing is in progress
if (markCancelled()) {
// kill all tasks currently running in this slot
Execution exec = this.executedTask;
if (exec != null && !exec.isFinished()) {
exec.fail(
new Exception(
"The slot in which the task was executed has been released. Probably loss of TaskManager "
+ getInstance()));
}
// release directly (if we are directly allocated),
// otherwise release through the parent shared slot
if (getParent() == null) {
// we have to give back the slot to the owning instance
getInstance().returnAllocatedSlot(this);
} else {
// we have to ask our parent to dispose us
getParent().releaseChild(this);
}
}
}
@Test
public void testRegistrationOfExecutionsFailedExternally() {
try {
final JobVertexID jid1 = new JobVertexID();
final JobVertexID jid2 = new JobVertexID();
JobVertex v1 = new JobVertex("v1", jid1);
JobVertex v2 = new JobVertex("v2", jid2);
Map<ExecutionAttemptID, Execution> executions = setupExecution(v1, 7, v2, 6);
for (Execution e : executions.values()) {
e.fail(null);
}
assertEquals(0, executions.size());
} catch (Exception e) {
e.printStackTrace();
fail(e.getMessage());
}
}
void scheduleOrUpdateConsumers(List<List<ExecutionEdge>> allConsumers) {
final int numConsumers = allConsumers.size();
if (numConsumers > 1) {
fail(
new IllegalStateException(
"Currently, only a single consumer group per partition is supported."));
} else if (numConsumers == 0) {
return;
}
for (ExecutionEdge edge : allConsumers.get(0)) {
final ExecutionVertex consumerVertex = edge.getTarget();
final Execution consumer = consumerVertex.getCurrentExecutionAttempt();
final ExecutionState consumerState = consumer.getState();
final IntermediateResultPartition partition = edge.getSource();
// ----------------------------------------------------------------
// Consumer is created => try to deploy and cache input channel
// descriptors if there is a deployment race
// ----------------------------------------------------------------
if (consumerState == CREATED) {
final Execution partitionExecution = partition.getProducer().getCurrentExecutionAttempt();
consumerVertex.cachePartitionInfo(
PartialInputChannelDeploymentDescriptor.fromEdge(partition, partitionExecution));
// When deploying a consuming task, its task deployment descriptor will contain all
// deployment information available at the respective time. It is possible that some
// of the partitions to be consumed have not been created yet. These are updated
// runtime via the update messages.
//
// TODO The current approach may send many update messages even though the consuming
// task has already been deployed with all necessary information. We have to check
// whether this is a problem and fix it, if it is.
future(
new Callable<Boolean>() {
@Override
public Boolean call() throws Exception {
try {
consumerVertex.scheduleForExecution(
consumerVertex.getExecutionGraph().getScheduler(),
consumerVertex.getExecutionGraph().isQueuedSchedulingAllowed());
} catch (Throwable t) {
fail(
new IllegalStateException(
"Could not schedule consumer " + "vertex " + consumerVertex, t));
}
return true;
}
},
executionContext);
// double check to resolve race conditions
if (consumerVertex.getExecutionState() == RUNNING) {
consumerVertex.sendPartitionInfos();
}
}
// ----------------------------------------------------------------
// Consumer is running => send update message now
// ----------------------------------------------------------------
else {
if (consumerState == RUNNING) {
final SimpleSlot consumerSlot = consumer.getAssignedResource();
if (consumerSlot == null) {
// The consumer has been reset concurrently
continue;
}
final Instance consumerInstance = consumerSlot.getInstance();
final ResultPartitionID partitionId =
new ResultPartitionID(partition.getPartitionId(), attemptId);
final Instance partitionInstance =
partition.getProducer().getCurrentAssignedResource().getInstance();
final ResultPartitionLocation partitionLocation;
if (consumerInstance.equals(partitionInstance)) {
// Consuming task is deployed to the same instance as the partition => local
partitionLocation = ResultPartitionLocation.createLocal();
} else {
// Different instances => remote
final ConnectionID connectionId =
new ConnectionID(
partitionInstance.getInstanceConnectionInfo(),
partition.getIntermediateResult().getConnectionIndex());
partitionLocation = ResultPartitionLocation.createRemote(connectionId);
}
final InputChannelDeploymentDescriptor descriptor =
new InputChannelDeploymentDescriptor(partitionId, partitionLocation);
final UpdatePartitionInfo updateTaskMessage =
new UpdateTaskSinglePartitionInfo(
consumer.getAttemptId(), partition.getIntermediateResult().getId(), descriptor);
sendUpdatePartitionInfoRpcCall(consumerSlot, updateTaskMessage);
}
// ----------------------------------------------------------------
// Consumer is scheduled or deploying => cache input channel
// deployment descriptors and send update message later
// ----------------------------------------------------------------
else if (consumerState == SCHEDULED || consumerState == DEPLOYING) {
final Execution partitionExecution = partition.getProducer().getCurrentExecutionAttempt();
consumerVertex.cachePartitionInfo(
PartialInputChannelDeploymentDescriptor.fromEdge(partition, partitionExecution));
// double check to resolve race conditions
if (consumerVertex.getExecutionState() == RUNNING) {
consumerVertex.sendPartitionInfos();
}
}
}
}
}