Search in sources :

Example 36 with SimpleSlot

use of org.apache.flink.runtime.instance.SimpleSlot in project flink by apache.

the class SchedulerSlotSharingTest method allocateSlotWithSharing.

@Test
public void allocateSlotWithSharing() {
    try {
        JobVertexID jid1 = new JobVertexID();
        JobVertexID jid2 = new JobVertexID();
        SlotSharingGroup sharingGroup = new SlotSharingGroup(jid1, jid2);
        Scheduler scheduler = new Scheduler(TestingUtils.directExecutionContext());
        scheduler.newInstanceAvailable(getRandomInstance(2));
        scheduler.newInstanceAvailable(getRandomInstance(2));
        // schedule 4 tasks from the first vertex group
        SimpleSlot s1 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 0, 5), sharingGroup), false).get();
        SimpleSlot s2 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 1, 5), sharingGroup), false).get();
        SimpleSlot s3 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 2, 5), sharingGroup), false).get();
        SimpleSlot s4 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 3, 5), sharingGroup), false).get();
        assertNotNull(s1);
        assertNotNull(s2);
        assertNotNull(s3);
        assertNotNull(s4);
        assertTrue(areAllDistinct(s1, s2, s3, s4));
        // we cannot schedule another task from the first vertex group
        try {
            scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 4, 5), sharingGroup), false).get();
            fail("Scheduler accepted too many tasks at the same time");
        } catch (ExecutionException e) {
            assertTrue(e.getCause() instanceof NoResourceAvailableException);
        } catch (Exception e) {
            fail("Wrong exception.");
        }
        // schedule some tasks from the second ID group
        SimpleSlot s1_2 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 0, 5), sharingGroup), false).get();
        SimpleSlot s2_2 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 1, 5), sharingGroup), false).get();
        SimpleSlot s3_2 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 2, 5), sharingGroup), false).get();
        SimpleSlot s4_2 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 3, 5), sharingGroup), false).get();
        assertNotNull(s1_2);
        assertNotNull(s2_2);
        assertNotNull(s3_2);
        assertNotNull(s4_2);
        // we cannot schedule another task from the second vertex group
        try {
            scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 4, 5), sharingGroup), false).get();
            fail("Scheduler accepted too many tasks at the same time");
        } catch (ExecutionException e) {
            assertTrue(e.getCause() instanceof NoResourceAvailableException);
        } catch (Exception e) {
            fail("Wrong exception.");
        }
        // now, we release some vertices (sub-slots) from the first group.
        // that should allow us to schedule more vertices from the first group
        s1.releaseSlot();
        s4.releaseSlot();
        assertEquals(4, sharingGroup.getTaskAssignment().getNumberOfSlots());
        assertEquals(2, sharingGroup.getTaskAssignment().getNumberOfAvailableSlotsForGroup(jid1));
        assertEquals(0, sharingGroup.getTaskAssignment().getNumberOfAvailableSlotsForGroup(jid2));
        // we can still not schedule anything from the second group of vertices
        try {
            scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 4, 5), sharingGroup), false).get();
            fail("Scheduler accepted too many tasks at the same time");
        } catch (ExecutionException e) {
            assertTrue(e.getCause() instanceof NoResourceAvailableException);
        } catch (Exception e) {
            fail("Wrong exception.");
        }
        // we can schedule something from the first vertex group
        SimpleSlot s5 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 4, 5), sharingGroup), false).get();
        assertNotNull(s5);
        assertEquals(4, sharingGroup.getTaskAssignment().getNumberOfSlots());
        assertEquals(1, sharingGroup.getTaskAssignment().getNumberOfAvailableSlotsForGroup(jid1));
        assertEquals(0, sharingGroup.getTaskAssignment().getNumberOfAvailableSlotsForGroup(jid2));
        // now we release a slot from the second vertex group and schedule another task from that group
        s2_2.releaseSlot();
        SimpleSlot s5_2 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 4, 5), sharingGroup), false).get();
        assertNotNull(s5_2);
        // release all slots
        s2.releaseSlot();
        s3.releaseSlot();
        s5.releaseSlot();
        s1_2.releaseSlot();
        s3_2.releaseSlot();
        s4_2.releaseSlot();
        s5_2.releaseSlot();
        // test that everything is released
        assertEquals(0, sharingGroup.getTaskAssignment().getNumberOfSlots());
        assertEquals(4, scheduler.getNumberOfAvailableSlots());
        // check the scheduler's bookkeeping
        assertEquals(0, scheduler.getNumberOfLocalizedAssignments());
        assertEquals(0, scheduler.getNumberOfNonLocalizedAssignments());
        assertEquals(10, scheduler.getNumberOfUnconstrainedAssignments());
    } catch (Exception e) {
        e.printStackTrace();
        fail(e.getMessage());
    }
}
Also used : JobVertexID(org.apache.flink.runtime.jobgraph.JobVertexID) ExecutionException(java.util.concurrent.ExecutionException) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) ExecutionException(java.util.concurrent.ExecutionException) Test(org.junit.Test)

Example 37 with SimpleSlot

use of org.apache.flink.runtime.instance.SimpleSlot in project flink by apache.

the class SchedulerSlotSharingTest method testLocalizedAssignment3.

/**
	 * Tests that the scheduler can fall back to non-local
	 */
@Test
public void testLocalizedAssignment3() {
    try {
        JobVertexID jid1 = new JobVertexID();
        JobVertexID jid2 = new JobVertexID();
        SlotSharingGroup sharingGroup = new SlotSharingGroup(jid1, jid2);
        Instance i1 = getRandomInstance(2);
        Instance i2 = getRandomInstance(2);
        TaskManagerLocation loc1 = i1.getTaskManagerLocation();
        Scheduler scheduler = new Scheduler(TestingUtils.directExecutionContext());
        scheduler.newInstanceAvailable(i1);
        scheduler.newInstanceAvailable(i2);
        // schedule until the one instance is full
        SimpleSlot s1 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid1, 0, 2, loc1), sharingGroup), false).get();
        SimpleSlot s2 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid1, 1, 2, loc1), sharingGroup), false).get();
        SimpleSlot s3 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid2, 0, 4, loc1), sharingGroup), false).get();
        SimpleSlot s4 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid2, 1, 4, loc1), sharingGroup), false).get();
        // schedule two more with preference of same instance --> need to go to other instance
        SimpleSlot s5 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid2, 3, 4, loc1), sharingGroup), false).get();
        SimpleSlot s6 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid2, 4, 4, loc1), sharingGroup), false).get();
        assertNotNull(s1);
        assertNotNull(s2);
        assertNotNull(s3);
        assertNotNull(s4);
        assertNotNull(s5);
        assertNotNull(s6);
        assertEquals(4, sharingGroup.getTaskAssignment().getNumberOfSlots());
        assertEquals(0, i1.getNumberOfAvailableSlots());
        assertEquals(0, i2.getNumberOfAvailableSlots());
        assertEquals(i1.getTaskManagerID(), s1.getTaskManagerID());
        assertEquals(i1.getTaskManagerID(), s2.getTaskManagerID());
        assertEquals(i1.getTaskManagerID(), s3.getTaskManagerID());
        assertEquals(i1.getTaskManagerID(), s4.getTaskManagerID());
        assertEquals(i2.getTaskManagerID(), s5.getTaskManagerID());
        assertEquals(i2.getTaskManagerID(), s6.getTaskManagerID());
        // check the scheduler's bookkeeping
        assertEquals(4, scheduler.getNumberOfLocalizedAssignments());
        assertEquals(2, scheduler.getNumberOfNonLocalizedAssignments());
        assertEquals(0, scheduler.getNumberOfUnconstrainedAssignments());
    } catch (Exception e) {
        e.printStackTrace();
        fail(e.getMessage());
    }
}
Also used : Instance(org.apache.flink.runtime.instance.Instance) SchedulerTestUtils.getRandomInstance(org.apache.flink.runtime.jobmanager.scheduler.SchedulerTestUtils.getRandomInstance) TaskManagerLocation(org.apache.flink.runtime.taskmanager.TaskManagerLocation) JobVertexID(org.apache.flink.runtime.jobgraph.JobVertexID) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) ExecutionException(java.util.concurrent.ExecutionException) Test(org.junit.Test)

Example 38 with SimpleSlot

use of org.apache.flink.runtime.instance.SimpleSlot in project flink by apache.

the class SchedulerSlotSharingTest method testLocalizedAssignment2.

/**
	 * Tests that the scheduler assigns to new local slots, rather than to existing non-local slots
	 */
@Test
public void testLocalizedAssignment2() {
    try {
        JobVertexID jid1 = new JobVertexID();
        JobVertexID jid2 = new JobVertexID();
        SlotSharingGroup sharingGroup = new SlotSharingGroup(jid1, jid2);
        Instance i1 = getRandomInstance(2);
        Instance i2 = getRandomInstance(2);
        TaskManagerLocation loc1 = i1.getTaskManagerLocation();
        TaskManagerLocation loc2 = i2.getTaskManagerLocation();
        Scheduler scheduler = new Scheduler(TestingUtils.directExecutionContext());
        scheduler.newInstanceAvailable(i1);
        scheduler.newInstanceAvailable(i2);
        // schedule one to each instance
        SimpleSlot s1 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid1, 0, 2, loc1), sharingGroup), false).get();
        SimpleSlot s2 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid1, 1, 2, loc1), sharingGroup), false).get();
        assertNotNull(s1);
        assertNotNull(s2);
        assertEquals(2, sharingGroup.getTaskAssignment().getNumberOfSlots());
        assertEquals(0, i1.getNumberOfAvailableSlots());
        assertEquals(2, i2.getNumberOfAvailableSlots());
        // schedule one from the other group to each instance
        SimpleSlot s3 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid2, 0, 2, loc2), sharingGroup), false).get();
        SimpleSlot s4 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid2, 1, 2, loc2), sharingGroup), false).get();
        assertNotNull(s3);
        assertNotNull(s4);
        assertEquals(4, sharingGroup.getTaskAssignment().getNumberOfSlots());
        assertEquals(0, i1.getNumberOfAvailableSlots());
        assertEquals(0, i2.getNumberOfAvailableSlots());
        // check the scheduler's bookkeeping
        assertEquals(4, scheduler.getNumberOfLocalizedAssignments());
        assertEquals(0, scheduler.getNumberOfNonLocalizedAssignments());
        assertEquals(0, scheduler.getNumberOfUnconstrainedAssignments());
    } catch (Exception e) {
        e.printStackTrace();
        fail(e.getMessage());
    }
}
Also used : Instance(org.apache.flink.runtime.instance.Instance) SchedulerTestUtils.getRandomInstance(org.apache.flink.runtime.jobmanager.scheduler.SchedulerTestUtils.getRandomInstance) TaskManagerLocation(org.apache.flink.runtime.taskmanager.TaskManagerLocation) JobVertexID(org.apache.flink.runtime.jobgraph.JobVertexID) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) ExecutionException(java.util.concurrent.ExecutionException) Test(org.junit.Test)

Example 39 with SimpleSlot

use of org.apache.flink.runtime.instance.SimpleSlot in project flink by apache.

the class ScheduleWithCoLocationHintTest method scheduleMixedCoLocationSlotSharing.

@Test
public void scheduleMixedCoLocationSlotSharing() {
    try {
        JobVertexID jid1 = new JobVertexID();
        JobVertexID jid2 = new JobVertexID();
        JobVertexID jid3 = new JobVertexID();
        JobVertexID jid4 = new JobVertexID();
        Scheduler scheduler = new Scheduler(TestingUtils.directExecutionContext());
        scheduler.newInstanceAvailable(getRandomInstance(1));
        scheduler.newInstanceAvailable(getRandomInstance(1));
        scheduler.newInstanceAvailable(getRandomInstance(1));
        scheduler.newInstanceAvailable(getRandomInstance(1));
        assertEquals(4, scheduler.getNumberOfAvailableSlots());
        CoLocationGroup grp = new CoLocationGroup();
        CoLocationConstraint clc1 = new CoLocationConstraint(grp);
        CoLocationConstraint clc2 = new CoLocationConstraint(grp);
        CoLocationConstraint clc3 = new CoLocationConstraint(grp);
        CoLocationConstraint clc4 = new CoLocationConstraint(grp);
        SlotSharingGroup shareGroup = new SlotSharingGroup();
        // first wave
        scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 0, 4), shareGroup), false);
        scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 2, 4), shareGroup), false);
        scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 1, 4), shareGroup), false);
        scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid1, 3, 4), shareGroup), false);
        // second wave
        SimpleSlot s21 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 0, 4), shareGroup, clc1), false).get();
        SimpleSlot s22 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 2, 4), shareGroup, clc2), false).get();
        SimpleSlot s23 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 1, 4), shareGroup, clc3), false).get();
        SimpleSlot s24 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid2, 3, 4), shareGroup, clc4), false).get();
        // third wave
        SimpleSlot s31 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid3, 1, 4), shareGroup, clc2), false).get();
        SimpleSlot s32 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid3, 2, 4), shareGroup, clc3), false).get();
        SimpleSlot s33 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid3, 3, 4), shareGroup, clc4), false).get();
        SimpleSlot s34 = scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid3, 0, 4), shareGroup, clc1), false).get();
        scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid4, 0, 4), shareGroup), false);
        scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid4, 1, 4), shareGroup), false);
        scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid4, 2, 4), shareGroup), false);
        scheduler.allocateSlot(new ScheduledUnit(getTestVertex(jid4, 3, 4), shareGroup), false);
        assertEquals(s21.getTaskManagerID(), s34.getTaskManagerID());
        assertEquals(s22.getTaskManagerID(), s31.getTaskManagerID());
        assertEquals(s23.getTaskManagerID(), s32.getTaskManagerID());
        assertEquals(s24.getTaskManagerID(), s33.getTaskManagerID());
        assertEquals(4, scheduler.getNumberOfLocalizedAssignments());
        assertEquals(0, scheduler.getNumberOfNonLocalizedAssignments());
        assertEquals(12, scheduler.getNumberOfUnconstrainedAssignments());
    } catch (Exception e) {
        e.printStackTrace();
        fail(e.getMessage());
    }
}
Also used : JobVertexID(org.apache.flink.runtime.jobgraph.JobVertexID) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) ExecutionException(java.util.concurrent.ExecutionException) Test(org.junit.Test)

Example 40 with SimpleSlot

use of org.apache.flink.runtime.instance.SimpleSlot in project flink by apache.

the class ScheduleWithCoLocationHintTest method testSlotReleasedInBetween.

@Test
public void testSlotReleasedInBetween() {
    try {
        JobVertexID jid1 = new JobVertexID();
        JobVertexID jid2 = new JobVertexID();
        Scheduler scheduler = new Scheduler(TestingUtils.directExecutionContext());
        Instance i1 = getRandomInstance(1);
        Instance i2 = getRandomInstance(1);
        TaskManagerLocation loc1 = i1.getTaskManagerLocation();
        TaskManagerLocation loc2 = i2.getTaskManagerLocation();
        scheduler.newInstanceAvailable(i2);
        scheduler.newInstanceAvailable(i1);
        assertEquals(2, scheduler.getNumberOfAvailableSlots());
        SlotSharingGroup sharingGroup = new SlotSharingGroup();
        CoLocationGroup ccg = new CoLocationGroup();
        CoLocationConstraint cc1 = new CoLocationConstraint(ccg);
        CoLocationConstraint cc2 = new CoLocationConstraint(ccg);
        SimpleSlot s1 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid1, 0, 2, loc1), sharingGroup, cc1), false).get();
        SimpleSlot s2 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid1, 1, 2, loc2), sharingGroup, cc2), false).get();
        s1.releaseSlot();
        s2.releaseSlot();
        assertEquals(2, scheduler.getNumberOfAvailableSlots());
        assertEquals(0, sharingGroup.getTaskAssignment().getNumberOfSlots());
        SimpleSlot s3 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid2, 0, 2, loc2), sharingGroup, cc1), false).get();
        SimpleSlot s4 = scheduler.allocateSlot(new ScheduledUnit(getTestVertexWithLocation(jid2, 1, 2, loc1), sharingGroup, cc2), false).get();
        // still preserves the previous instance mapping)
        assertEquals(i1.getTaskManagerID(), s3.getTaskManagerID());
        assertEquals(i2.getTaskManagerID(), s4.getTaskManagerID());
        s3.releaseSlot();
        s4.releaseSlot();
        assertEquals(2, scheduler.getNumberOfAvailableSlots());
        assertEquals(4, scheduler.getNumberOfLocalizedAssignments());
        assertEquals(0, scheduler.getNumberOfNonLocalizedAssignments());
        assertEquals(0, scheduler.getNumberOfUnconstrainedAssignments());
    } catch (Exception e) {
        e.printStackTrace();
        fail(e.getMessage());
    }
}
Also used : Instance(org.apache.flink.runtime.instance.Instance) SchedulerTestUtils.getRandomInstance(org.apache.flink.runtime.jobmanager.scheduler.SchedulerTestUtils.getRandomInstance) TaskManagerLocation(org.apache.flink.runtime.taskmanager.TaskManagerLocation) JobVertexID(org.apache.flink.runtime.jobgraph.JobVertexID) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) ExecutionException(java.util.concurrent.ExecutionException) Test(org.junit.Test)

Aggregations

SimpleSlot (org.apache.flink.runtime.instance.SimpleSlot)78 Test (org.junit.Test)59 JobVertexID (org.apache.flink.runtime.jobgraph.JobVertexID)46 Instance (org.apache.flink.runtime.instance.Instance)38 ExecutionException (java.util.concurrent.ExecutionException)25 ActorTaskManagerGateway (org.apache.flink.runtime.jobmanager.slots.ActorTaskManagerGateway)22 JobID (org.apache.flink.api.common.JobID)20 TaskManagerLocation (org.apache.flink.runtime.taskmanager.TaskManagerLocation)16 TaskManagerGateway (org.apache.flink.runtime.jobmanager.slots.TaskManagerGateway)14 FlinkCompletableFuture (org.apache.flink.runtime.concurrent.impl.FlinkCompletableFuture)13 SchedulerTestUtils.getRandomInstance (org.apache.flink.runtime.jobmanager.scheduler.SchedulerTestUtils.getRandomInstance)13 ExecutionGraphTestUtils.getExecutionVertex (org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.getExecutionVertex)12 ExecutionGraphTestUtils.getInstance (org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.getInstance)12 IOException (java.io.IOException)10 ActorGateway (org.apache.flink.runtime.instance.ActorGateway)8 BaseTestingActorGateway (org.apache.flink.runtime.instance.BaseTestingActorGateway)8 JobVertex (org.apache.flink.runtime.jobgraph.JobVertex)8 Scheduler (org.apache.flink.runtime.jobmanager.scheduler.Scheduler)8 AllocatedSlot (org.apache.flink.runtime.jobmanager.slots.AllocatedSlot)8 Future (org.apache.flink.runtime.concurrent.Future)7