// StatusUpdate takes care of updating the status
func (s *eremeticScheduler) StatusUpdate(driver sched.SchedulerDriver, status *mesos.TaskStatus) {
id := status.TaskId.GetValue()
log.Debugf("Received task status [%s] for task [%s]", status.State.String(), id)
task, err := database.ReadTask(id)
if err != nil {
log.Debugf("Error reading task from database: %s", err)
}
if task.ID == "" {
task = types.EremeticTask{
ID: id,
SlaveId: status.SlaveId.GetValue(),
}
}
if !task.IsRunning() && *status.State == mesos.TaskState_TASK_RUNNING {
TasksRunning.Inc()
}
if types.IsTerminal(status.State) {
TasksTerminated.With(prometheus.Labels{"status": status.State.String()}).Inc()
if task.WasRunning() {
TasksRunning.Dec()
}
}
task.UpdateStatus(types.Status{
Status: status.State.String(),
Time: time.Now().Unix(),
})
if *status.State == mesos.TaskState_TASK_FAILED && !task.WasRunning() {
if task.Retry >= maxRetries {
log.Warnf("giving up on %s after %d retry attempts", id, task.Retry)
} else {
log.Infof("task %s was never running. re-scheduling", id)
task.UpdateStatus(types.Status{
Status: mesos.TaskState_TASK_STAGING.String(),
Time: time.Now().Unix(),
})
task.Retry++
go func() {
QueueSize.Inc()
s.tasks <- id
}()
}
}
if types.IsTerminal(status.State) {
handler.NotifyCallback(&task)
}
database.PutTask(&task)
}
// GetTaskInfo returns information about the given task.
func GetTaskInfo(scheduler types.Scheduler) http.HandlerFunc {
return func(w http.ResponseWriter, r *http.Request) {
vars := mux.Vars(r)
id := vars["taskId"]
log.Debugf("Fetching task for id: %s", id)
task, _ := database.ReadTask(id)
if strings.Contains(r.Header.Get("Accept"), "text/html") {
renderHTML(w, r, task, id)
} else {
if reflect.DeepEqual(task, (types.EremeticTask{})) {
writeJSON(http.StatusNotFound, nil, w)
return
}
writeJSON(http.StatusOK, task, w)
}
}
}
// ResourceOffers handles the Resource Offers
func (s *eremeticScheduler) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) {
log.Tracef("Received %d resource offers", len(offers))
var offer *mesos.Offer
loop:
for len(offers) > 0 {
select {
case <-s.shutdown:
log.Info("Shutting down: declining offers")
break loop
case tid := <-s.tasks:
log.Debugf("Trying to find offer to launch %s with", tid)
t, _ := database.ReadTask(tid)
offer, offers = matchOffer(t, offers)
if offer == nil {
log.Warnf("Could not find a matching offer for %s", tid)
TasksDelayed.Inc()
go func() { s.tasks <- tid }()
break loop
}
log.Debugf("Preparing to launch task %s with offer %s", tid, offer.Id.GetValue())
t, task := s.newTask(t, offer)
database.PutTask(&t)
driver.LaunchTasks([]*mesos.OfferID{offer.Id}, []*mesos.TaskInfo{task}, defaultFilter)
TasksLaunched.Inc()
QueueSize.Dec()
continue
default:
break loop
}
}
log.Trace("No tasks to launch. Declining offers.")
for _, offer := range offers {
driver.DeclineOffer(offer.Id, defaultFilter)
}
}
func ReconcileTasks(driver sched.SchedulerDriver) *Reconcile {
cancel := make(chan struct{})
done := make(chan struct{})
go func() {
var (
c uint
delay int
)
tasks, err := database.ListNonTerminalTasks()
if err != nil {
log.Errorf("Failed to list non-terminal tasks: %s", err)
close(done)
return
}
log.Infof("Trying to reconcile with %d task(s)", len(tasks))
start := time.Now()
for len(tasks) > 0 {
select {
case <-cancel:
log.Info("Cancelling reconciliation job")
close(done)
return
case <-time.After(time.Duration(delay) * time.Second):
// Filter tasks that has received a status update
ntasks := []*types.EremeticTask{}
for _, t := range tasks {
nt, err := database.ReadTask(t.ID)
if err != nil {
log.Warnf("Task %s not found in database", t.ID)
continue
}
if nt.LastUpdated().Before(start) {
ntasks = append(ntasks, &nt)
}
}
tasks = ntasks
// Send reconciliation request
if len(tasks) > 0 {
var statuses []*mesos.TaskStatus
for _, t := range tasks {
statuses = append(statuses, &mesos.TaskStatus{
State: mesos.TaskState_TASK_STAGING.Enum(),
TaskId: &mesos.TaskID{Value: proto.String(t.ID)},
SlaveId: &mesos.SlaveID{Value: proto.String(t.SlaveId)},
})
}
log.Debugf("Sending reconciliation request #%d", c)
driver.ReconcileTasks(statuses)
}
if delay < maxReconciliationDelay {
delay = 10 << c
if delay >= maxReconciliationDelay {
delay = maxReconciliationDelay
}
}
c += 1
}
}
log.Info("Reconciliation done")
close(done)
}()
return &Reconcile{
cancel: cancel,
done: done,
}
}
func TestReconcile(t *testing.T) {
dir, _ := os.Getwd()
database.NewDB(fmt.Sprintf("%s/../db/test.db", dir))
database.Clean()
defer database.Close()
maxReconciliationDelay = 1
Convey("ReconcileTasks", t, func() {
Convey("Finishes when there are no tasks", func() {
driver := NewMockScheduler()
r := ReconcileTasks(driver)
select {
case <-r.done:
}
So(driver.AssertNotCalled(t, "ReconcileTasks"), ShouldBeTrue)
})
Convey("Sends reconcile request", func() {
driver := NewMockScheduler()
driver.On("ReconcileTasks").Run(func(mock.Arguments) {
t, err := database.ReadTask("1234")
if err != nil {
panic("mock error")
}
t.UpdateStatus(types.Status{
Status: mesos.TaskState_TASK_RUNNING.String(),
Time: time.Now().Unix() + 1,
})
database.PutTask(&t)
}).Once()
database.PutTask(&types.EremeticTask{
ID: "1234",
Status: []types.Status{
types.Status{
Status: mesos.TaskState_TASK_STAGING.String(),
Time: time.Now().Unix(),
},
},
})
r := ReconcileTasks(driver)
select {
case <-r.done:
}
So(driver.AssertCalled(t, "ReconcileTasks"), ShouldBeTrue)
})
Convey("Cancel reconciliation", func() {
driver := NewMockScheduler()
database.PutTask(&types.EremeticTask{
ID: "1234",
Status: []types.Status{
types.Status{
Status: mesos.TaskState_TASK_STAGING.String(),
Time: time.Now().Unix(),
},
},
})
r := ReconcileTasks(driver)
r.Cancel()
select {
case <-r.done:
}
So(driver.AssertNotCalled(t, "ReconcileTasks"), ShouldBeTrue)
})
})
}
func TestScheduler(t *testing.T) {
dir, _ := os.Getwd()
database.NewDB(fmt.Sprintf("%s/../db/test.db", dir))
database.Clean()
defer database.Close()
Convey("eremeticScheduler", t, func() {
s := eremeticScheduler{}
id := "eremetic-task.9999"
database.PutTask(&types.EremeticTask{ID: id})
Convey("newTask", func() {
task := types.EremeticTask{
ID: "eremetic-task.1234",
}
offer := mesos.Offer{
FrameworkId: &mesos.FrameworkID{
Value: proto.String("framework-id"),
},
SlaveId: &mesos.SlaveID{
Value: proto.String("slave-id"),
},
Hostname: proto.String("hostname"),
}
taskData, mesosTask := s.newTask(task, &offer)
So(mesosTask.GetTaskId().GetValue(), ShouldEqual, task.ID)
So(taskData.SlaveId, ShouldEqual, "slave-id")
})
Convey("createEremeticScheduler", func() {
s := createEremeticScheduler()
So(s.tasksCreated, ShouldEqual, 0)
})
Convey("API", func() {
Convey("Registered", func() {
driver := NewMockScheduler()
driver.On("ReconcileTasks").Return("ok").Once()
fID := mesos.FrameworkID{Value: proto.String("1234")}
mInfo := mesos.MasterInfo{}
s.Registered(driver, &fID, &mInfo)
So(driver.AssertCalled(t, "ReconcileTasks"), ShouldBeTrue)
})
Convey("Reregistered", func() {
driver := NewMockScheduler()
driver.On("ReconcileTasks").Return("ok").Once()
database.Clean()
s.Reregistered(driver, &mesos.MasterInfo{})
So(driver.AssertCalled(t, "ReconcileTasks"), ShouldBeTrue)
})
Convey("Disconnected", func() {
s.Disconnected(nil)
})
Convey("ResourceOffers", func() {
driver := NewMockScheduler()
var offers []*mesos.Offer
Convey("No offers", func() {
s.ResourceOffers(driver, offers)
So(driver.AssertNotCalled(t, "DeclineOffer"), ShouldBeTrue)
So(driver.AssertNotCalled(t, "LaunchTasks"), ShouldBeTrue)
})
Convey("No tasks", func() {
offers = append(offers, &mesos.Offer{Id: &mesos.OfferID{Value: proto.String("1234")}})
driver.On("DeclineOffer").Return("declined").Once()
s.ResourceOffers(driver, offers)
So(driver.AssertCalled(t, "DeclineOffer"), ShouldBeTrue)
So(driver.AssertNotCalled(t, "LaunchTasks"), ShouldBeTrue)
})
})
Convey("StatusUpdate", func() {
Convey("Running then failing", func() {
s.StatusUpdate(nil, &mesos.TaskStatus{
TaskId: &mesos.TaskID{
Value: proto.String(id),
},
State: mesos.TaskState_TASK_RUNNING.Enum(),
})
task, _ := database.ReadTask(id)
So(len(task.Status), ShouldEqual, 1)
So(task.Status[0].Status, ShouldEqual, mesos.TaskState_TASK_RUNNING.String())
s.StatusUpdate(nil, &mesos.TaskStatus{
TaskId: &mesos.TaskID{
Value: proto.String(id),
},
State: mesos.TaskState_TASK_FAILED.Enum(),
})
task, _ = database.ReadTask(id)
So(len(task.Status), ShouldEqual, 2)
So(task.Status[0].Status, ShouldEqual, mesos.TaskState_TASK_RUNNING.String())
So(task.Status[1].Status, ShouldEqual, mesos.TaskState_TASK_FAILED.String())
})
Convey("Failing immediatly", func() {
s.tasks = make(chan string, 100)
s.StatusUpdate(nil, &mesos.TaskStatus{
TaskId: &mesos.TaskID{
Value: proto.String(id),
},
State: mesos.TaskState_TASK_FAILED.Enum(),
})
task, _ := database.ReadTask(id)
So(len(task.Status), ShouldEqual, 2)
So(task.Status[0].Status, ShouldEqual, mesos.TaskState_TASK_FAILED.String())
So(task.Status[1].Status, ShouldEqual, mesos.TaskState_TASK_STAGING.String())
select {
case c := <-s.tasks:
So(c, ShouldEqual, id)
}
})
})
Convey("FrameworkMessage", func() {
driver := NewMockScheduler()
message := `{"message": "this is a message"}`
Convey("From Eremetic", func() {
source := "eremetic-executor"
executor := mesos.ExecutorID{
Value: proto.String(source),
}
s.FrameworkMessage(driver, &executor, &mesos.SlaveID{}, message)
})
Convey("From an unknown source", func() {
source := "other-source"
executor := mesos.ExecutorID{
Value: proto.String(source),
}
s.FrameworkMessage(driver, &executor, &mesos.SlaveID{}, message)
})
Convey("A bad json", func() {
source := "eremetic-executor"
executor := mesos.ExecutorID{
Value: proto.String(source),
}
s.FrameworkMessage(driver, &executor, &mesos.SlaveID{}, "not a json")
})
})
Convey("OfferRescinded", func() {
s.OfferRescinded(nil, &mesos.OfferID{})
})
Convey("SlaveLost", func() {
s.SlaveLost(nil, &mesos.SlaveID{})
})
Convey("ExecutorLost", func() {
s.ExecutorLost(nil, &mesos.ExecutorID{}, &mesos.SlaveID{}, 2)
})
Convey("Error", func() {
s.Error(nil, "Error")
})
})
})
Convey("ScheduleTask", t, func() {
Convey("Given a valid Request", func() {
scheduler := &eremeticScheduler{
tasks: make(chan string, 100),
}
request := types.Request{
TaskCPUs: 0.5,
TaskMem: 22.0,
DockerImage: "busybox",
Command: "echo hello",
}
Convey("It should put a task id on the channel", func() {
taskID, err := scheduler.ScheduleTask(request)
So(err, ShouldBeNil)
select {
case c := <-scheduler.tasks:
So(c, ShouldEqual, taskID)
task, _ := database.ReadTask(taskID)
So(task.TaskCPUs, ShouldEqual, request.TaskCPUs)
So(task.TaskMem, ShouldEqual, request.TaskMem)
So(task.Command, ShouldEqual, request.Command)
So(task.User, ShouldEqual, "root")
So(task.Environment, ShouldBeEmpty)
So(task.Image, ShouldEqual, request.DockerImage)
So(task.ID, ShouldStartWith, "eremetic-task.")
}
})
})
})
}