From 97002ea3535090f041e32e0b0847c912915b5323 Mon Sep 17 00:00:00 2001 From: Renan DelValle Date: Thu, 13 Oct 2016 17:50:54 -0400 Subject: [PATCH] Adding First Fit Watts Only --- schedulers/firstfitwattsonly.go | 236 ++++++++++++++++++++++++++++++++ 1 file changed, 236 insertions(+) create mode 100644 schedulers/firstfitwattsonly.go diff --git a/schedulers/firstfitwattsonly.go b/schedulers/firstfitwattsonly.go new file mode 100644 index 0000000..b19ce3b --- /dev/null +++ b/schedulers/firstfitwattsonly.go @@ -0,0 +1,236 @@ +package schedulers + +import ( + "bitbucket.org/bingcloud/electron/def" + "fmt" + "github.com/golang/protobuf/proto" + mesos "github.com/mesos/mesos-go/mesosproto" + "github.com/mesos/mesos-go/mesosutil" + sched "github.com/mesos/mesos-go/scheduler" + "log" + "strings" + "time" +) + +// Decides if to take an offer or not +func (*FirstFitWattsOnly) takeOffer(offer *mesos.Offer, task def.Task) bool { + + _, _, watts := OfferAgg(offer) + + //TODO: Insert watts calculation here instead of taking them as a parameter + + if watts >= task.Watts { + return true + } + + return false +} + +type FirstFitWattsOnly struct { + tasksCreated int + tasksRunning int + tasks []def.Task + metrics map[string]def.Metric + running map[string]map[string]bool + ignoreWatts bool + + // First set of PCP values are garbage values, signal to logger to start recording when we're + // about to schedule a new task + RecordPCP bool + + // This channel is closed when the program receives an interrupt, + // signalling that the program should shut down. + Shutdown chan struct{} + // This channel is closed after shutdown is closed, and only when all + // outstanding tasks have been cleaned up + Done chan struct{} + + // Controls when to shutdown pcp logging + PCPLog chan struct{} +} + +// New electron scheduler +func NewFirstFitWattsOnly(tasks []def.Task, ignoreWatts bool) *FirstFitWattsOnly { + + s := &FirstFitWattsOnly{ + tasks: tasks, + ignoreWatts: ignoreWatts, + Shutdown: make(chan struct{}), + Done: make(chan struct{}), + PCPLog: make(chan struct{}), + running: make(map[string]map[string]bool), + RecordPCP: false, + } + return s +} + +func (s *FirstFitWattsOnly) newTask(offer *mesos.Offer, task def.Task) *mesos.TaskInfo { + taskName := fmt.Sprintf("%s-%d", task.Name, *task.Instances) + s.tasksCreated++ + + if !s.RecordPCP { + // Turn on logging + s.RecordPCP = true + time.Sleep(1 * time.Second) // Make sure we're recording by the time the first task starts + } + + // If this is our first time running into this Agent + if _, ok := s.running[offer.GetSlaveId().GoString()]; !ok { + s.running[offer.GetSlaveId().GoString()] = make(map[string]bool) + } + + // Add task to list of tasks running on node + s.running[offer.GetSlaveId().GoString()][taskName] = true + + resources := []*mesos.Resource{ + mesosutil.NewScalarResource("watts", task.Watts), + } + + return &mesos.TaskInfo{ + Name: proto.String(taskName), + TaskId: &mesos.TaskID{ + Value: proto.String("electron-" + taskName), + }, + SlaveId: offer.SlaveId, + Resources: resources, + Command: &mesos.CommandInfo{ + Value: proto.String(task.CMD), + }, + Container: &mesos.ContainerInfo{ + Type: mesos.ContainerInfo_DOCKER.Enum(), + Docker: &mesos.ContainerInfo_DockerInfo{ + Image: proto.String(task.Image), + Network: mesos.ContainerInfo_DockerInfo_BRIDGE.Enum(), // Run everything isolated + }, + }, + } +} + +func (s *FirstFitWattsOnly) Registered( + _ sched.SchedulerDriver, + frameworkID *mesos.FrameworkID, + masterInfo *mesos.MasterInfo) { + log.Printf("Framework %s registered with master %s", frameworkID, masterInfo) +} + +func (s *FirstFitWattsOnly) Reregistered(_ sched.SchedulerDriver, masterInfo *mesos.MasterInfo) { + log.Printf("Framework re-registered with master %s", masterInfo) +} + +func (s *FirstFitWattsOnly) Disconnected(sched.SchedulerDriver) { + log.Println("Framework disconnected with master") +} + +func (s *FirstFitWattsOnly) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) { + log.Printf("Received %d resource offers", len(offers)) + + for _, offer := range offers { + select { + case <-s.Shutdown: + log.Println("Done scheduling tasks: declining offer on [", offer.GetHostname(), "]") + driver.DeclineOffer(offer.Id, longFilter) + + log.Println("Number of tasks still running: ", s.tasksRunning) + continue + default: + } + + tasks := []*mesos.TaskInfo{} + + // First fit strategy + + taken := false + for i, task := range s.tasks { + + // Check host if it exists + if task.Host != "" { + // Don't take offer if it doesn't match our task's host requirement + if !strings.HasPrefix(*offer.Hostname, task.Host) { + continue + } + } + + // Decision to take the offer or not + if s.takeOffer(offer, task) { + + log.Println("Co-Located with: ") + coLocated(s.running[offer.GetSlaveId().GoString()]) + + tasks = append(tasks, s.newTask(offer, task)) + + log.Printf("Starting %s on [%s]\n", task.Name, offer.GetHostname()) + driver.LaunchTasks([]*mesos.OfferID{offer.Id}, tasks, defaultFilter) + + taken = true + + fmt.Println("Inst: ", *task.Instances) + *task.Instances-- + + if *task.Instances <= 0 { + // All instances of task have been scheduled, remove it + s.tasks[i] = s.tasks[len(s.tasks)-1] + s.tasks = s.tasks[:len(s.tasks)-1] + + if len(s.tasks) <= 0 { + log.Println("Done scheduling all tasks") + close(s.Shutdown) + } + } + break // Offer taken, move on + } + } + + // If there was no match for the task + if !taken { + fmt.Println("There is not enough resources to launch a task:") + cpus, mem, watts := OfferAgg(offer) + + log.Printf("\n", cpus, mem, watts) + driver.DeclineOffer(offer.Id, defaultFilter) + } + + } +} + +func (s *FirstFitWattsOnly) StatusUpdate(driver sched.SchedulerDriver, status *mesos.TaskStatus) { + log.Printf("Received task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value) + + if *status.State == mesos.TaskState_TASK_RUNNING { + s.tasksRunning++ + } else if IsTerminal(status.State) { + delete(s.running[status.GetSlaveId().GoString()], *status.TaskId.Value) + s.tasksRunning-- + if s.tasksRunning == 0 { + select { + case <-s.Shutdown: + close(s.Done) + default: + } + } + } + log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value) +} + +func (s *FirstFitWattsOnly) FrameworkMessage( + driver sched.SchedulerDriver, + executorID *mesos.ExecutorID, + slaveID *mesos.SlaveID, + message string) { + + log.Println("Getting a framework message: ", message) + log.Printf("Received a framework message from some unknown source: %s", *executorID.Value) +} + +func (s *FirstFitWattsOnly) OfferRescinded(_ sched.SchedulerDriver, offerID *mesos.OfferID) { + log.Printf("Offer %s rescinded", offerID) +} +func (s *FirstFitWattsOnly) SlaveLost(_ sched.SchedulerDriver, slaveID *mesos.SlaveID) { + log.Printf("Slave %s lost", slaveID) +} +func (s *FirstFitWattsOnly) ExecutorLost(_ sched.SchedulerDriver, executorID *mesos.ExecutorID, slaveID *mesos.SlaveID, status int) { + log.Printf("Executor %s on slave %s was lost", executorID, slaveID) +} + +func (s *FirstFitWattsOnly) Error(_ sched.SchedulerDriver, err string) { + log.Printf("Receiving an error: %s", err) +}