use of org.apache.flink.runtime.execution.ExecutionState in project flink by apache.
the class JobManagerTest method testRequestPartitionState.
/**
* Tests responses to partition state requests.
*/
@Test
public void testRequestPartitionState() throws Exception {
new JavaTestKit(system) {
{
new Within(duration("15 seconds")) {
@Override
protected void run() {
// Setup
TestingCluster cluster = null;
try {
cluster = startTestingCluster(2, 1, DEFAULT_AKKA_ASK_TIMEOUT());
final IntermediateDataSetID rid = new IntermediateDataSetID();
// Create a task
final JobVertex sender = new JobVertex("Sender");
sender.setParallelism(1);
// just block
sender.setInvokableClass(BlockingNoOpInvokable.class);
sender.createAndAddResultDataSet(rid, PIPELINED);
final JobGraph jobGraph = new JobGraph("Blocking test job", sender);
final JobID jid = jobGraph.getJobID();
final ActorGateway jobManagerGateway = cluster.getLeaderGateway(TestingUtils.TESTING_DURATION());
// we can set the leader session ID to None because we don't use this gateway to send messages
final ActorGateway testActorGateway = new AkkaActorGateway(getTestActor(), null);
// Submit the job and wait for all vertices to be running
jobManagerGateway.tell(new SubmitJob(jobGraph, ListeningBehaviour.EXECUTION_RESULT), testActorGateway);
expectMsgClass(JobSubmitSuccess.class);
jobManagerGateway.tell(new WaitForAllVerticesToBeRunningOrFinished(jid), testActorGateway);
expectMsgClass(AllVerticesRunning.class);
// This is the mock execution ID of the task requesting the state of the partition
final ExecutionAttemptID receiver = new ExecutionAttemptID();
// Request the execution graph to get the runtime info
jobManagerGateway.tell(new RequestExecutionGraph(jid), testActorGateway);
final ExecutionGraph eg = (ExecutionGraph) expectMsgClass(ExecutionGraphFound.class).executionGraph();
final ExecutionVertex vertex = eg.getJobVertex(sender.getID()).getTaskVertices()[0];
final IntermediateResultPartition partition = vertex.getProducedPartitions().values().iterator().next();
final ResultPartitionID partitionId = new ResultPartitionID(partition.getPartitionId(), vertex.getCurrentExecutionAttempt().getAttemptId());
// - The test ----------------------------------------------------------------------
// 1. All execution states
RequestPartitionProducerState request = new RequestPartitionProducerState(jid, rid, partitionId);
for (ExecutionState state : ExecutionState.values()) {
ExecutionGraphTestUtils.setVertexState(vertex, state);
Future<ExecutionState> futurePartitionState = jobManagerGateway.ask(request, getRemainingTime()).mapTo(ClassTag$.MODULE$.<ExecutionState>apply(ExecutionState.class));
ExecutionState resp = Await.result(futurePartitionState, getRemainingTime());
assertEquals(state, resp);
}
// 2. Non-existing execution
request = new RequestPartitionProducerState(jid, rid, new ResultPartitionID());
Future<?> futurePartitionState = jobManagerGateway.ask(request, getRemainingTime());
try {
Await.result(futurePartitionState, getRemainingTime());
fail("Did not fail with expected RuntimeException");
} catch (RuntimeException e) {
assertEquals(IllegalArgumentException.class, e.getCause().getClass());
}
// 3. Non-existing job
request = new RequestPartitionProducerState(new JobID(), rid, new ResultPartitionID());
futurePartitionState = jobManagerGateway.ask(request, getRemainingTime());
try {
Await.result(futurePartitionState, getRemainingTime());
fail("Did not fail with expected IllegalArgumentException");
} catch (IllegalArgumentException ignored) {
}
} catch (Exception e) {
e.printStackTrace();
fail(e.getMessage());
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}
};
}
};
}
use of org.apache.flink.runtime.execution.ExecutionState in project flink by apache.
the class JobDetailsHandler method createJobDetailsJson.
public static String createJobDetailsJson(AccessExecutionGraph graph, @Nullable MetricFetcher fetcher) throws IOException {
final StringWriter writer = new StringWriter();
final JsonGenerator gen = JsonFactory.jacksonFactory.createGenerator(writer);
final long now = System.currentTimeMillis();
gen.writeStartObject();
// basic info
gen.writeStringField("jid", graph.getJobID().toString());
gen.writeStringField("name", graph.getJobName());
gen.writeBooleanField("isStoppable", graph.isStoppable());
gen.writeStringField("state", graph.getState().name());
// times and duration
final long jobStartTime = graph.getStatusTimestamp(JobStatus.CREATED);
final long jobEndTime = graph.getState().isGloballyTerminalState() ? graph.getStatusTimestamp(graph.getState()) : -1L;
gen.writeNumberField("start-time", jobStartTime);
gen.writeNumberField("end-time", jobEndTime);
gen.writeNumberField("duration", (jobEndTime > 0 ? jobEndTime : now) - jobStartTime);
gen.writeNumberField("now", now);
// timestamps
gen.writeObjectFieldStart("timestamps");
for (JobStatus status : JobStatus.values()) {
gen.writeNumberField(status.name(), graph.getStatusTimestamp(status));
}
gen.writeEndObject();
// job vertices
int[] jobVerticesPerState = new int[ExecutionState.values().length];
gen.writeArrayFieldStart("vertices");
for (AccessExecutionJobVertex ejv : graph.getVerticesTopologically()) {
int[] tasksPerState = new int[ExecutionState.values().length];
long startTime = Long.MAX_VALUE;
long endTime = 0;
boolean allFinished = true;
for (AccessExecutionVertex vertex : ejv.getTaskVertices()) {
final ExecutionState state = vertex.getExecutionState();
tasksPerState[state.ordinal()]++;
// take the earliest start time
long started = vertex.getStateTimestamp(ExecutionState.DEPLOYING);
if (started > 0) {
startTime = Math.min(startTime, started);
}
allFinished &= state.isTerminal();
endTime = Math.max(endTime, vertex.getStateTimestamp(state));
}
long duration;
if (startTime < Long.MAX_VALUE) {
if (allFinished) {
duration = endTime - startTime;
} else {
endTime = -1L;
duration = now - startTime;
}
} else {
startTime = -1L;
endTime = -1L;
duration = -1L;
}
ExecutionState jobVertexState = ExecutionJobVertex.getAggregateJobVertexState(tasksPerState, ejv.getParallelism());
jobVerticesPerState[jobVertexState.ordinal()]++;
gen.writeStartObject();
gen.writeStringField("id", ejv.getJobVertexId().toString());
gen.writeStringField("name", ejv.getName());
gen.writeNumberField("parallelism", ejv.getParallelism());
gen.writeStringField("status", jobVertexState.name());
gen.writeNumberField("start-time", startTime);
gen.writeNumberField("end-time", endTime);
gen.writeNumberField("duration", duration);
gen.writeObjectFieldStart("tasks");
for (ExecutionState state : ExecutionState.values()) {
gen.writeNumberField(state.name(), tasksPerState[state.ordinal()]);
}
gen.writeEndObject();
MutableIOMetrics counts = new MutableIOMetrics();
for (AccessExecutionVertex vertex : ejv.getTaskVertices()) {
counts.addIOMetrics(vertex.getCurrentExecutionAttempt(), fetcher, graph.getJobID().toString(), ejv.getJobVertexId().toString());
}
counts.writeIOMetricsAsJson(gen);
gen.writeEndObject();
}
gen.writeEndArray();
gen.writeObjectFieldStart("status-counts");
for (ExecutionState state : ExecutionState.values()) {
gen.writeNumberField(state.name(), jobVerticesPerState[state.ordinal()]);
}
gen.writeEndObject();
gen.writeFieldName("plan");
gen.writeRawValue(graph.getJsonPlan());
gen.writeEndObject();
gen.close();
return writer.toString();
}
use of org.apache.flink.runtime.execution.ExecutionState in project flink by apache.
the class JobVertexDetailsHandler method createVertexDetailsJson.
public static String createVertexDetailsJson(AccessExecutionJobVertex jobVertex, String jobID, @Nullable MetricFetcher fetcher) throws IOException {
final long now = System.currentTimeMillis();
StringWriter writer = new StringWriter();
JsonGenerator gen = JsonFactory.jacksonFactory.createGenerator(writer);
gen.writeStartObject();
gen.writeStringField("id", jobVertex.getJobVertexId().toString());
gen.writeStringField("name", jobVertex.getName());
gen.writeNumberField("parallelism", jobVertex.getParallelism());
gen.writeNumberField("now", now);
gen.writeArrayFieldStart("subtasks");
int num = 0;
for (AccessExecutionVertex vertex : jobVertex.getTaskVertices()) {
final ExecutionState status = vertex.getExecutionState();
TaskManagerLocation location = vertex.getCurrentAssignedResourceLocation();
String locationString = location == null ? "(unassigned)" : location.getHostname() + ":" + location.dataPort();
long startTime = vertex.getStateTimestamp(ExecutionState.DEPLOYING);
if (startTime == 0) {
startTime = -1;
}
long endTime = status.isTerminal() ? vertex.getStateTimestamp(status) : -1;
long duration = startTime > 0 ? ((endTime > 0 ? endTime : now) - startTime) : -1;
gen.writeStartObject();
gen.writeNumberField("subtask", num);
gen.writeStringField("status", status.name());
gen.writeNumberField("attempt", vertex.getCurrentExecutionAttempt().getAttemptNumber());
gen.writeStringField("host", locationString);
gen.writeNumberField("start-time", startTime);
gen.writeNumberField("end-time", endTime);
gen.writeNumberField("duration", duration);
MutableIOMetrics counts = new MutableIOMetrics();
counts.addIOMetrics(vertex.getCurrentExecutionAttempt(), fetcher, jobID, jobVertex.getJobVertexId().toString());
counts.writeIOMetricsAsJson(gen);
gen.writeEndObject();
num++;
}
gen.writeEndArray();
gen.writeEndObject();
gen.close();
return writer.toString();
}
use of org.apache.flink.runtime.execution.ExecutionState in project flink by apache.
the class JobVertexTaskManagersHandler method createVertexDetailsByTaskManagerJson.
public static String createVertexDetailsByTaskManagerJson(AccessExecutionJobVertex jobVertex, String jobID, @Nullable MetricFetcher fetcher) throws IOException {
StringWriter writer = new StringWriter();
JsonGenerator gen = JsonFactory.jacksonFactory.createGenerator(writer);
// Build a map that groups tasks by TaskManager
Map<String, List<AccessExecutionVertex>> taskManagerVertices = new HashMap<>();
for (AccessExecutionVertex vertex : jobVertex.getTaskVertices()) {
TaskManagerLocation location = vertex.getCurrentAssignedResourceLocation();
String taskManager = location == null ? "(unassigned)" : location.getHostname() + ":" + location.dataPort();
List<AccessExecutionVertex> vertices = taskManagerVertices.get(taskManager);
if (vertices == null) {
vertices = new ArrayList<>();
taskManagerVertices.put(taskManager, vertices);
}
vertices.add(vertex);
}
// Build JSON response
final long now = System.currentTimeMillis();
gen.writeStartObject();
gen.writeStringField("id", jobVertex.getJobVertexId().toString());
gen.writeStringField("name", jobVertex.getName());
gen.writeNumberField("now", now);
gen.writeArrayFieldStart("taskmanagers");
for (Map.Entry<String, List<AccessExecutionVertex>> entry : taskManagerVertices.entrySet()) {
String host = entry.getKey();
List<AccessExecutionVertex> taskVertices = entry.getValue();
int[] tasksPerState = new int[ExecutionState.values().length];
long startTime = Long.MAX_VALUE;
long endTime = 0;
boolean allFinished = true;
MutableIOMetrics counts = new MutableIOMetrics();
for (AccessExecutionVertex vertex : taskVertices) {
final ExecutionState state = vertex.getExecutionState();
tasksPerState[state.ordinal()]++;
// take the earliest start time
long started = vertex.getStateTimestamp(ExecutionState.DEPLOYING);
if (started > 0) {
startTime = Math.min(startTime, started);
}
allFinished &= state.isTerminal();
endTime = Math.max(endTime, vertex.getStateTimestamp(state));
counts.addIOMetrics(vertex.getCurrentExecutionAttempt(), fetcher, jobID, jobVertex.getJobVertexId().toString());
}
long duration;
if (startTime < Long.MAX_VALUE) {
if (allFinished) {
duration = endTime - startTime;
} else {
endTime = -1L;
duration = now - startTime;
}
} else {
startTime = -1L;
endTime = -1L;
duration = -1L;
}
ExecutionState jobVertexState = ExecutionJobVertex.getAggregateJobVertexState(tasksPerState, taskVertices.size());
gen.writeStartObject();
gen.writeStringField("host", host);
gen.writeStringField("status", jobVertexState.name());
gen.writeNumberField("start-time", startTime);
gen.writeNumberField("end-time", endTime);
gen.writeNumberField("duration", duration);
counts.writeIOMetricsAsJson(gen);
gen.writeObjectFieldStart("status-counts");
for (ExecutionState state : ExecutionState.values()) {
gen.writeNumberField(state.name(), tasksPerState[state.ordinal()]);
}
gen.writeEndObject();
gen.writeEndObject();
}
gen.writeEndArray();
gen.writeEndObject();
gen.close();
return writer.toString();
}
use of org.apache.flink.runtime.execution.ExecutionState in project flink by apache.
the class ExecutionVertex method resetForNewExecution.
// --------------------------------------------------------------------------------------------
// Actions
// --------------------------------------------------------------------------------------------
public void resetForNewExecution() {
LOG.debug("Resetting execution vertex {} for new execution.", getSimpleName());
synchronized (priorExecutions) {
Execution execution = currentExecution;
ExecutionState state = execution.getState();
if (state == FINISHED || state == CANCELED || state == FAILED) {
priorExecutions.add(execution);
currentExecution = new Execution(getExecutionGraph().getFutureExecutor(), this, execution.getAttemptNumber() + 1, System.currentTimeMillis(), timeout);
CoLocationGroup grp = jobVertex.getCoLocationGroup();
if (grp != null) {
this.locationConstraint = grp.getLocationConstraint(subTaskIndex);
}
} else {
throw new IllegalStateException("Cannot reset a vertex that is in state " + state);
}
}
}
Aggregations