func SetupCluster(t *testing.T, store *store.MemoryStore, watch chan events.Event) *api.Task {
ctx := context.Background()
// Start the global orchestrator.
global := NewGlobalOrchestrator(store)
go func() {
assert.NoError(t, global.Run(ctx))
}()
addService(t, store, service1)
testutils.Expect(t, watch, state.EventCreateService{})
testutils.Expect(t, watch, state.EventCommit{})
addNode(t, store, node1)
testutils.Expect(t, watch, state.EventCreateNode{})
testutils.Expect(t, watch, state.EventCommit{})
// return task creation from orchestrator
return testutils.WatchTaskCreate(t, watch)
}
func TestOrchestratorRestartWindow(t *testing.T) {
t.Parallel()
ctx := context.Background()
s := store.NewMemoryStore(nil)
assert.NotNil(t, s)
defer s.Close()
orchestrator := NewReplicatedOrchestrator(s)
defer orchestrator.Stop()
watch, cancel := state.Watch(s.WatchQueue() /*state.EventCreateTask{}, state.EventUpdateTask{}*/)
defer cancel()
// Create a service with two instances specified before the orchestrator is
// started. This should result in two tasks when the orchestrator
// starts up.
err := s.Update(func(tx store.Tx) error {
j1 := &api.Service{
ID: "id1",
Spec: api.ServiceSpec{
Annotations: api.Annotations{
Name: "name1",
},
Mode: &api.ServiceSpec_Replicated{
Replicated: &api.ReplicatedService{
Replicas: 2,
},
},
Task: api.TaskSpec{
Restart: &api.RestartPolicy{
Condition: api.RestartOnAny,
Delay: gogotypes.DurationProto(100 * time.Millisecond),
MaxAttempts: 1,
Window: gogotypes.DurationProto(500 * time.Millisecond),
},
},
},
}
assert.NoError(t, store.CreateService(tx, j1))
return nil
})
assert.NoError(t, err)
// Start the orchestrator.
go func() {
assert.NoError(t, orchestrator.Run(ctx))
}()
observedTask1 := testutils.WatchTaskCreate(t, watch)
assert.Equal(t, observedTask1.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask1.ServiceAnnotations.Name, "name1")
observedTask2 := testutils.WatchTaskCreate(t, watch)
assert.Equal(t, observedTask2.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask2.ServiceAnnotations.Name, "name1")
// Fail the first task. Confirm that it gets restarted.
updatedTask1 := observedTask1.Copy()
updatedTask1.Status = api.TaskStatus{State: api.TaskStateFailed}
before := time.Now()
err = s.Update(func(tx store.Tx) error {
assert.NoError(t, store.UpdateTask(tx, updatedTask1))
return nil
})
assert.NoError(t, err)
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
observedTask3 := testutils.WatchTaskCreate(t, watch)
testutils.Expect(t, watch, state.EventCommit{})
assert.Equal(t, observedTask3.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask3.DesiredState, api.TaskStateReady)
assert.Equal(t, observedTask3.ServiceAnnotations.Name, "name1")
observedTask4 := testutils.WatchTaskUpdate(t, watch)
after := time.Now()
// At least 100 ms should have elapsed. Only check the lower bound,
// because the system may be slow and it could have taken longer.
if after.Sub(before) < 100*time.Millisecond {
t.Fatal("restart delay should have elapsed")
}
assert.Equal(t, observedTask4.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask4.DesiredState, api.TaskStateRunning)
assert.Equal(t, observedTask4.ServiceAnnotations.Name, "name1")
// Fail the second task. Confirm that it gets restarted.
updatedTask2 := observedTask2.Copy()
updatedTask2.Status = api.TaskStatus{State: api.TaskStateFailed}
err = s.Update(func(tx store.Tx) error {
assert.NoError(t, store.UpdateTask(tx, updatedTask2))
return nil
})
assert.NoError(t, err)
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
observedTask5 := testutils.WatchTaskCreate(t, watch)
testutils.Expect(t, watch, state.EventCommit{})
assert.Equal(t, observedTask5.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask5.DesiredState, api.TaskStateReady)
assert.Equal(t, observedTask5.ServiceAnnotations.Name, "name1")
observedTask6 := testutils.WatchTaskUpdate(t, watch) // task gets started after a delay
testutils.Expect(t, watch, state.EventCommit{})
assert.Equal(t, observedTask6.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask6.DesiredState, api.TaskStateRunning)
assert.Equal(t, observedTask6.ServiceAnnotations.Name, "name1")
// Fail the first instance again. It should not be restarted.
updatedTask1 = observedTask3.Copy()
updatedTask1.Status = api.TaskStatus{State: api.TaskStateFailed}
err = s.Update(func(tx store.Tx) error {
assert.NoError(t, store.UpdateTask(tx, updatedTask1))
return nil
})
assert.NoError(t, err)
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
select {
case <-watch:
t.Fatal("got unexpected event")
case <-time.After(200 * time.Millisecond):
}
time.Sleep(time.Second)
// Fail the second instance again. It should get restarted because
// enough time has elapsed since the last restarts.
updatedTask2 = observedTask5.Copy()
updatedTask2.Status = api.TaskStatus{State: api.TaskStateFailed}
before = time.Now()
err = s.Update(func(tx store.Tx) error {
assert.NoError(t, store.UpdateTask(tx, updatedTask2))
return nil
})
assert.NoError(t, err)
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
observedTask7 := testutils.WatchTaskCreate(t, watch)
testutils.Expect(t, watch, state.EventCommit{})
assert.Equal(t, observedTask7.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask7.DesiredState, api.TaskStateReady)
observedTask8 := testutils.WatchTaskUpdate(t, watch)
after = time.Now()
// At least 100 ms should have elapsed. Only check the lower bound,
// because the system may be slow and it could have taken longer.
if after.Sub(before) < 100*time.Millisecond {
t.Fatal("restart delay should have elapsed")
}
assert.Equal(t, observedTask8.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask8.DesiredState, api.TaskStateRunning)
assert.Equal(t, observedTask8.ServiceAnnotations.Name, "name1")
}
func TestOrchestratorRestartDelay(t *testing.T) {
ctx := context.Background()
s := store.NewMemoryStore(nil)
assert.NotNil(t, s)
defer s.Close()
orchestrator := NewReplicatedOrchestrator(s)
defer orchestrator.Stop()
watch, cancel := state.Watch(s.WatchQueue() /*state.EventCreateTask{}, state.EventUpdateTask{}*/)
defer cancel()
// Create a service with two instances specified before the orchestrator is
// started. This should result in two tasks when the orchestrator
// starts up.
err := s.Update(func(tx store.Tx) error {
j1 := &api.Service{
ID: "id1",
Spec: api.ServiceSpec{
Annotations: api.Annotations{
Name: "name1",
},
Task: api.TaskSpec{
Runtime: &api.TaskSpec_Container{
Container: &api.ContainerSpec{},
},
Restart: &api.RestartPolicy{
Condition: api.RestartOnAny,
Delay: gogotypes.DurationProto(100 * time.Millisecond),
},
},
Mode: &api.ServiceSpec_Replicated{
Replicated: &api.ReplicatedService{
Replicas: 2,
},
},
},
}
assert.NoError(t, store.CreateService(tx, j1))
return nil
})
assert.NoError(t, err)
// Start the orchestrator.
go func() {
assert.NoError(t, orchestrator.Run(ctx))
}()
observedTask1 := testutils.WatchTaskCreate(t, watch)
assert.Equal(t, observedTask1.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask1.ServiceAnnotations.Name, "name1")
observedTask2 := testutils.WatchTaskCreate(t, watch)
assert.Equal(t, observedTask2.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask2.ServiceAnnotations.Name, "name1")
// Fail the first task. Confirm that it gets restarted.
updatedTask1 := observedTask1.Copy()
updatedTask1.Status = api.TaskStatus{State: api.TaskStateFailed}
before := time.Now()
err = s.Update(func(tx store.Tx) error {
assert.NoError(t, store.UpdateTask(tx, updatedTask1))
return nil
})
assert.NoError(t, err)
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
observedTask3 := testutils.WatchTaskCreate(t, watch)
testutils.Expect(t, watch, state.EventCommit{})
assert.Equal(t, observedTask3.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask3.DesiredState, api.TaskStateReady)
assert.Equal(t, observedTask3.ServiceAnnotations.Name, "name1")
observedTask4 := testutils.WatchTaskUpdate(t, watch)
after := time.Now()
// At least 100 ms should have elapsed. Only check the lower bound,
// because the system may be slow and it could have taken longer.
if after.Sub(before) < 100*time.Millisecond {
t.Fatalf("restart delay should have elapsed. Got: %v", after.Sub(before))
}
assert.Equal(t, observedTask4.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask4.DesiredState, api.TaskStateRunning)
assert.Equal(t, observedTask4.ServiceAnnotations.Name, "name1")
}
func TestOrchestratorRestartOnNone(t *testing.T) {
ctx := context.Background()
s := store.NewMemoryStore(nil)
assert.NotNil(t, s)
defer s.Close()
orchestrator := NewReplicatedOrchestrator(s)
defer orchestrator.Stop()
watch, cancel := state.Watch(s.WatchQueue() /*state.EventCreateTask{}, state.EventUpdateTask{}*/)
defer cancel()
// Create a service with two instances specified before the orchestrator is
// started. This should result in two tasks when the orchestrator
// starts up.
err := s.Update(func(tx store.Tx) error {
j1 := &api.Service{
ID: "id1",
Spec: api.ServiceSpec{
Annotations: api.Annotations{
Name: "name1",
},
Task: api.TaskSpec{
Runtime: &api.TaskSpec_Container{
Container: &api.ContainerSpec{},
},
Restart: &api.RestartPolicy{
Condition: api.RestartOnNone,
},
},
Mode: &api.ServiceSpec_Replicated{
Replicated: &api.ReplicatedService{
Replicas: 2,
},
},
},
}
assert.NoError(t, store.CreateService(tx, j1))
return nil
})
assert.NoError(t, err)
// Start the orchestrator.
go func() {
assert.NoError(t, orchestrator.Run(ctx))
}()
observedTask1 := testutils.WatchTaskCreate(t, watch)
assert.Equal(t, observedTask1.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask1.ServiceAnnotations.Name, "name1")
observedTask2 := testutils.WatchTaskCreate(t, watch)
assert.Equal(t, observedTask2.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask2.ServiceAnnotations.Name, "name1")
// Fail the first task. Confirm that it does not get restarted.
updatedTask1 := observedTask1.Copy()
updatedTask1.Status.State = api.TaskStateFailed
err = s.Update(func(tx store.Tx) error {
assert.NoError(t, store.UpdateTask(tx, updatedTask1))
return nil
})
assert.NoError(t, err)
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
select {
case <-watch:
t.Fatal("got unexpected event")
case <-time.After(100 * time.Millisecond):
}
// Mark the second task as completed. Confirm that it does not get restarted.
updatedTask2 := observedTask2.Copy()
updatedTask2.Status = api.TaskStatus{State: api.TaskStateCompleted}
err = s.Update(func(tx store.Tx) error {
assert.NoError(t, store.UpdateTask(tx, updatedTask2))
return nil
})
assert.NoError(t, err)
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
select {
case <-watch:
t.Fatal("got unexpected event")
case <-time.After(100 * time.Millisecond):
}
}
func TestTaskHistory(t *testing.T) {
ctx := context.Background()
s := store.NewMemoryStore(nil)
assert.NotNil(t, s)
defer s.Close()
assert.NoError(t, s.Update(func(tx store.Tx) error {
store.CreateCluster(tx, &api.Cluster{
ID: identity.NewID(),
Spec: api.ClusterSpec{
Annotations: api.Annotations{
Name: store.DefaultClusterName,
},
Orchestration: api.OrchestrationConfig{
TaskHistoryRetentionLimit: 2,
},
},
})
return nil
}))
taskReaper := taskreaper.New(s)
defer taskReaper.Stop()
orchestrator := NewReplicatedOrchestrator(s)
defer orchestrator.Stop()
watch, cancel := state.Watch(s.WatchQueue() /*state.EventCreateTask{}, state.EventUpdateTask{}*/)
defer cancel()
// Create a service with two instances specified before the orchestrator is
// started. This should result in two tasks when the orchestrator
// starts up.
err := s.Update(func(tx store.Tx) error {
j1 := &api.Service{
ID: "id1",
Spec: api.ServiceSpec{
Annotations: api.Annotations{
Name: "name1",
},
Mode: &api.ServiceSpec_Replicated{
Replicated: &api.ReplicatedService{
Replicas: 2,
},
},
Task: api.TaskSpec{
Restart: &api.RestartPolicy{
Condition: api.RestartOnAny,
Delay: ptypes.DurationProto(0),
},
},
},
}
assert.NoError(t, store.CreateService(tx, j1))
return nil
})
assert.NoError(t, err)
// Start the orchestrator.
go func() {
assert.NoError(t, orchestrator.Run(ctx))
}()
go taskReaper.Run()
observedTask1 := testutils.WatchTaskCreate(t, watch)
assert.Equal(t, observedTask1.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask1.ServiceAnnotations.Name, "name1")
observedTask2 := testutils.WatchTaskCreate(t, watch)
assert.Equal(t, observedTask2.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask2.ServiceAnnotations.Name, "name1")
// Fail both tasks. They should both get restarted.
updatedTask1 := observedTask1.Copy()
updatedTask1.Status.State = api.TaskStateFailed
updatedTask1.ServiceAnnotations = api.Annotations{Name: "original"}
updatedTask2 := observedTask2.Copy()
updatedTask2.Status.State = api.TaskStateFailed
updatedTask2.ServiceAnnotations = api.Annotations{Name: "original"}
err = s.Update(func(tx store.Tx) error {
assert.NoError(t, store.UpdateTask(tx, updatedTask1))
assert.NoError(t, store.UpdateTask(tx, updatedTask2))
return nil
})
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventUpdateTask{})
testutils.Expect(t, watch, state.EventCommit{})
testutils.Expect(t, watch, state.EventUpdateTask{})
observedTask3 := testutils.WatchTaskCreate(t, watch)
assert.Equal(t, observedTask3.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask3.ServiceAnnotations.Name, "name1")
testutils.Expect(t, watch, state.EventUpdateTask{})
observedTask4 := testutils.WatchTaskCreate(t, watch)
assert.Equal(t, observedTask4.Status.State, api.TaskStateNew)
assert.Equal(t, observedTask4.ServiceAnnotations.Name, "name1")
// Fail these replacement tasks. Since TaskHistory is set to 2, this
// should cause the oldest tasks for each instance to get deleted.
updatedTask3 := observedTask3.Copy()
updatedTask3.Status.State = api.TaskStateFailed
updatedTask4 := observedTask4.Copy()
updatedTask4.Status.State = api.TaskStateFailed
err = s.Update(func(tx store.Tx) error {
assert.NoError(t, store.UpdateTask(tx, updatedTask3))
assert.NoError(t, store.UpdateTask(tx, updatedTask4))
return nil
})
deletedTask1 := testutils.WatchTaskDelete(t, watch)
deletedTask2 := testutils.WatchTaskDelete(t, watch)
assert.Equal(t, api.TaskStateFailed, deletedTask1.Status.State)
assert.Equal(t, "original", deletedTask1.ServiceAnnotations.Name)
assert.Equal(t, api.TaskStateFailed, deletedTask2.Status.State)
assert.Equal(t, "original", deletedTask2.ServiceAnnotations.Name)
var foundTasks []*api.Task
s.View(func(tx store.ReadTx) {
foundTasks, err = store.FindTasks(tx, store.All)
})
assert.NoError(t, err)
assert.Len(t, foundTasks, 4)
}