spdf/elektron
Archived
1
0
Fork 0
Code Issues 4 Pull requests 1 Projects Releases Packages Wiki Activity

Merged in ffClassMapWatts (pull request #7)

Browse source 
FirstFit with ClassMapWatts. Retrofitting schedulers to use base.go and also to log scheduling trace.
This commit is contained in:
Renan DelValle 2017-01-24 00:31:42 +00:00
commit 08e874c77a
17 changed files with 765 additions and 333 deletions
Download patch file Download diff file Expand all files Collapse all files

4
README.md
View file

@ -11,7 +11,9 @@ To Do:
* Write test code for each scheduler (This should be after the design change) * Write test code for each scheduler (This should be after the design change)
* Some of the constants in constants/constants.go can vary based on the environment. * Some of the constants in constants/constants.go can vary based on the environment.
Possible to setup the constants at runtime based on the environment? Possible to setup the constants at runtime based on the environment?
* Retrofit schedulers for scheduling tracing * Log fix for declining offer -- different reason when insufficient resources as compared to when there are no
longer any tasks to schedule.
* Have a centralised logFile that can be filtered by identifier. All electron logs should go into this file.
**Requires [Performance Co-Pilot](http://pcp.io/) tool pmdumptext to be installed on the **Requires [Performance Co-Pilot](http://pcp.io/) tool pmdumptext to be installed on the

2
pcp/loganddynamiccap.go
View file

@ -57,7 +57,7 @@ func meanCluster(history *ring.Ring) float64 {
return (total / count) return (total / count)
} }
func StartLogAndDynamicCap(quit chan struct{}, logging *bool, prefix string, hiThreshold, loThreshold float64) { func StartPCPLogAndExtremaDynamicCap(quit chan struct{}, logging *bool, prefix string, hiThreshold, loThreshold float64) {
const pcpCommand string = "pmdumptext -m -l -f '' -t 1.0 -d , -c config" const pcpCommand string = "pmdumptext -m -l -f '' -t 1.0 -d , -c config"
cmd := exec.Command("sh", "-c", pcpCommand) cmd := exec.Command("sh", "-c", pcpCommand)
cmd.SysProcAttr = &syscall.SysProcAttr{Setpgid: true} cmd.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}

4
scheduler.go
View file

@ -58,7 +58,7 @@ func main() {
startTime := time.Now().Format("20060102150405") startTime := time.Now().Format("20060102150405")
logPrefix := *pcplogPrefix + "_" + startTime logPrefix := *pcplogPrefix + "_" + startTime
scheduler := schedulers.NewBPSWClassMapWatts(tasks, *ignoreWatts, logPrefix) scheduler := schedulers.NewFirstFitSortedWattsReducedWAR(tasks, *ignoreWatts, logPrefix)
driver, err := sched.NewMesosSchedulerDriver(sched.DriverConfig{ driver, err := sched.NewMesosSchedulerDriver(sched.DriverConfig{
Master: *master, Master: *master,
Framework: &mesos.FrameworkInfo{ Framework: &mesos.FrameworkInfo{
@ -73,7 +73,7 @@ func main() {
} }
//go pcp.Start(scheduler.PCPLog, &scheduler.RecordPCP, logPrefix) //go pcp.Start(scheduler.PCPLog, &scheduler.RecordPCP, logPrefix)
go pcp.StartLogAndDynamicCap(scheduler.PCPLog, &scheduler.RecordPCP, logPrefix, *hiThreshold, *loThreshold) go pcp.StartPCPLogAndExtremaDynamicCap(scheduler.PCPLog, &scheduler.RecordPCP, logPrefix, *hiThreshold, *loThreshold)
time.Sleep(1 * time.Second) // Take a second between starting PCP log and continuing time.Sleep(1 * time.Second) // Take a second between starting PCP log and continuing
// Attempt to handle signint to not leave pmdumptext running // Attempt to handle signint to not leave pmdumptext running

122
schedulers/pistoncapper.go → schedulers/binpackedpistoncapping.go
View file

@ -12,6 +12,7 @@ import (
sched "github.com/mesos/mesos-go/scheduler" sched "github.com/mesos/mesos-go/scheduler"
"log" "log"
"math" "math"
"os"
"strings" "strings"
"sync" "sync"
"time" "time"
@ -23,7 +24,8 @@ import (
This is basically extending the BinPacking algorithm to also cap each node at a different values, This is basically extending the BinPacking algorithm to also cap each node at a different values,
corresponding to the load on that node. corresponding to the load on that node.
*/ */
type PistonCapper struct { type BinPackedPistonCapper struct {
base // Type embedded to inherit common functions
tasksCreated int tasksCreated int
tasksRunning int tasksRunning int
tasks []def.Task tasks []def.Task
@ -49,11 +51,19 @@ type PistonCapper struct {
// Controls when to shutdown pcp logging. // Controls when to shutdown pcp logging.
PCPLog chan struct{} PCPLog chan struct{}
schedTrace *log.Logger
} }
// New electron scheduler. // New electron scheduler.
func NewPistonCapper(tasks []def.Task, ignoreWatts bool) *PistonCapper { func NewBinPackedPistonCapper(tasks []def.Task, ignoreWatts bool, schedTracePrefix string) *BinPackedPistonCapper {
s := &PistonCapper{
logFile, err := os.Create("./" + schedTracePrefix + "_schedTrace.log")
if err != nil {
log.Fatal(err)
}
s := &BinPackedPistonCapper{
tasks: tasks, tasks: tasks,
ignoreWatts: ignoreWatts, ignoreWatts: ignoreWatts,
Shutdown: make(chan struct{}), Shutdown: make(chan struct{}),
@ -65,12 +75,13 @@ func NewPistonCapper(tasks []def.Task, ignoreWatts bool) *PistonCapper {
RecordPCP: false, RecordPCP: false,
ticker: time.NewTicker(5 * time.Second), ticker: time.NewTicker(5 * time.Second),
isCapping: false, isCapping: false,
schedTrace: log.New(logFile, "", log.LstdFlags),
} }
return s return s
} }
// check whether task fits the offer or not. // check whether task fits the offer or not.
func (s *PistonCapper) takeOffer(offerWatts float64, offerCPU float64, offerRAM float64, func (s *BinPackedPistonCapper) takeOffer(offerWatts float64, offerCPU float64, offerRAM float64,
totalWatts float64, totalCPU float64, totalRAM float64, task def.Task) bool { totalWatts float64, totalCPU float64, totalRAM float64, task def.Task) bool {
if (s.ignoreWatts || (offerWatts >= (totalWatts + task.Watts))) && if (s.ignoreWatts || (offerWatts >= (totalWatts + task.Watts))) &&
(offerCPU >= (totalCPU + task.CPU)) && (offerCPU >= (totalCPU + task.CPU)) &&
@ -82,9 +93,9 @@ func (s *PistonCapper) takeOffer(offerWatts float64, offerCPU float64, offerRAM
} }
// mutex // mutex
var mutex sync.Mutex var bpPistonMutex sync.Mutex
func (s *PistonCapper) newTask(offer *mesos.Offer, task def.Task) *mesos.TaskInfo { func (s *BinPackedPistonCapper) newTask(offer *mesos.Offer, task def.Task) *mesos.TaskInfo {
taskName := fmt.Sprintf("%s-%d", task.Name, *task.Instances) taskName := fmt.Sprintf("%s-%d", task.Name, *task.Instances)
s.tasksCreated++ s.tasksCreated++
@ -140,45 +151,39 @@ func (s *PistonCapper) newTask(offer *mesos.Offer, task def.Task) *mesos.TaskInf
} }
} }
func (s *PistonCapper) Registered( func (s *BinPackedPistonCapper) Disconnected(sched.SchedulerDriver) {
_ sched.SchedulerDriver, // Need to stop the capping process
frameworkID *mesos.FrameworkID, s.ticker.Stop()
masterInfo *mesos.MasterInfo) { bpPistonMutex.Lock()
log.Printf("Framework %s registered with master %s", frameworkID, masterInfo) s.isCapping = false
} bpPistonMutex.Unlock()
func (s *PistonCapper) Reregistered(_ sched.SchedulerDriver, masterInfo *mesos.MasterInfo) {
log.Printf("Framework re-registered with master %s", masterInfo)
}
func (s *PistonCapper) Disconnected(sched.SchedulerDriver) {
log.Println("Framework disconnected with master") log.Println("Framework disconnected with master")
} }
// go routine to cap the each node in the cluster at regular intervals of time. // go routine to cap the each node in the cluster at regular intervals of time.
var capValues = make(map[string]float64) var bpPistonCapValues = make(map[string]float64)
// Storing the previous cap value for each host so as to not repeatedly cap the nodes to the same value. (reduces overhead) // Storing the previous cap value for each host so as to not repeatedly cap the nodes to the same value. (reduces overhead)
var previousRoundedCapValues = make(map[string]int) var bpPistonPreviousRoundedCapValues = make(map[string]int)
func (s *PistonCapper) startCapping() { func (s *BinPackedPistonCapper) startCapping() {
go func() { go func() {
for { for {
select { select {
case <-s.ticker.C: case <-s.ticker.C:
// Need to cap each node // Need to cap each node
mutex.Lock() bpPistonMutex.Lock()
for host, capValue := range capValues { for host, capValue := range bpPistonCapValues {
roundedCapValue := int(math.Floor(capValue + 0.5)) roundedCapValue := int(math.Floor(capValue + 0.5))
// has the cap value changed // has the cap value changed
if prevRoundedCap, ok := previousRoundedCapValues[host]; ok { if prevRoundedCap, ok := bpPistonPreviousRoundedCapValues[host]; ok {
if prevRoundedCap != roundedCapValue { if prevRoundedCap != roundedCapValue {
if err := rapl.Cap(host, "rapl", roundedCapValue); err != nil { if err := rapl.Cap(host, "rapl", roundedCapValue); err != nil {
log.Println(err) log.Println(err)
} else { } else {
log.Printf("Capped [%s] at %d", host, int(math.Floor(capValue+0.5))) log.Printf("Capped [%s] at %d", host, int(math.Floor(capValue+0.5)))
} }
previousRoundedCapValues[host] = roundedCapValue bpPistonPreviousRoundedCapValues[host] = roundedCapValue
} }
} else { } else {
if err := rapl.Cap(host, "rapl", roundedCapValue); err != nil { if err := rapl.Cap(host, "rapl", roundedCapValue); err != nil {
@ -186,27 +191,27 @@ func (s *PistonCapper) startCapping() {
} else { } else {
log.Printf("Capped [%s] at %d", host, int(math.Floor(capValue+0.5))) log.Printf("Capped [%s] at %d", host, int(math.Floor(capValue+0.5)))
} }
previousRoundedCapValues[host] = roundedCapValue bpPistonPreviousRoundedCapValues[host] = roundedCapValue
} }
} }
mutex.Unlock() bpPistonMutex.Unlock()
} }
} }
}() }()
} }
// Stop the capping // Stop the capping
func (s *PistonCapper) stopCapping() { func (s *BinPackedPistonCapper) stopCapping() {
if s.isCapping { if s.isCapping {
log.Println("Stopping the capping.") log.Println("Stopping the capping.")
s.ticker.Stop() s.ticker.Stop()
mutex.Lock() bpPistonMutex.Lock()
s.isCapping = false s.isCapping = false
mutex.Unlock() bpPistonMutex.Unlock()
} }
} }
func (s *PistonCapper) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) { func (s *BinPackedPistonCapper) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) {
log.Printf("Received %d resource offers", len(offers)) log.Printf("Received %d resource offers", len(offers))
// retrieving the total power for each host in the offers // retrieving the total power for each host in the offers
@ -249,7 +254,8 @@ func (s *PistonCapper) ResourceOffers(driver sched.SchedulerDriver, offers []*me
// Store the partialLoad for host corresponding to this offer. // Store the partialLoad for host corresponding to this offer.
// Once we can't fit any more tasks, we update capValue for this host with partialLoad and then launch the fit tasks. // Once we can't fit any more tasks, we update capValue for this host with partialLoad and then launch the fit tasks.
partialLoad := 0.0 partialLoad := 0.0
for i, task := range s.tasks { for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Check host if it exists // Check host if it exists
if task.Host != "" { if task.Host != "" {
// Don't take offer if it doens't match our task's host requirement. // Don't take offer if it doens't match our task's host requirement.
@ -274,9 +280,11 @@ func (s *PistonCapper) ResourceOffers(driver sched.SchedulerDriver, offers []*me
totalRAM += task.RAM totalRAM += task.RAM
log.Println("Co-Located with: ") log.Println("Co-Located with: ")
coLocated(s.running[offer.GetSlaveId().GoString()]) coLocated(s.running[offer.GetSlaveId().GoString()])
fitTasks = append(fitTasks, s.newTask(offer, task)) taskToSchedule := s.newTask(offer, task)
fitTasks = append(fitTasks, taskToSchedule)
log.Println("Inst: ", *task.Instances) log.Println("Inst: ", *task.Instances)
s.schedTrace.Print(offer.GetHostname() + ":" + taskToSchedule.GetTaskId().GetValue())
*task.Instances-- *task.Instances--
// updating the cap value for offer.Hostname // updating the cap value for offer.Hostname
partialLoad += ((task.Watts * constants.CapMargin) / s.totalPower[*offer.Hostname]) * 100 partialLoad += ((task.Watts * constants.CapMargin) / s.totalPower[*offer.Hostname]) * 100
@ -297,9 +305,9 @@ func (s *PistonCapper) ResourceOffers(driver sched.SchedulerDriver, offers []*me
if taken { if taken {
// Updating the cap value for offer.Hostname // Updating the cap value for offer.Hostname
mutex.Lock() bpPistonMutex.Lock()
capValues[*offer.Hostname] += partialLoad bpPistonCapValues[*offer.Hostname] += partialLoad
mutex.Unlock() bpPistonMutex.Unlock()
log.Printf("Starting on [%s]\n", offer.GetHostname()) log.Printf("Starting on [%s]\n", offer.GetHostname())
driver.LaunchTasks([]*mesos.OfferID{offer.Id}, fitTasks, defaultFilter) driver.LaunchTasks([]*mesos.OfferID{offer.Id}, fitTasks, defaultFilter)
} else { } else {
@ -314,7 +322,7 @@ func (s *PistonCapper) ResourceOffers(driver sched.SchedulerDriver, offers []*me
} }
// Remove finished task from the taskMonitor // Remove finished task from the taskMonitor
func (s *PistonCapper) deleteFromTaskMonitor(finishedTaskID string) (def.Task, string, error) { func (s *BinPackedPistonCapper) deleteFromTaskMonitor(finishedTaskID string) (def.Task, string, error) {
hostOfFinishedTask := "" hostOfFinishedTask := ""
indexOfFinishedTask := -1 indexOfFinishedTask := -1
found := false found := false
@ -345,13 +353,13 @@ func (s *PistonCapper) deleteFromTaskMonitor(finishedTaskID string) (def.Task, s
return finishedTask, hostOfFinishedTask, nil return finishedTask, hostOfFinishedTask, nil
} }
func (s *PistonCapper) StatusUpdate(driver sched.SchedulerDriver, status *mesos.TaskStatus) { func (s *BinPackedPistonCapper) StatusUpdate(driver sched.SchedulerDriver, status *mesos.TaskStatus) {
log.Printf("Received task status [%s] for task [%s]\n", NameFor(status.State), *status.TaskId.Value) log.Printf("Received task status [%s] for task [%s]\n", NameFor(status.State), *status.TaskId.Value)
if *status.State == mesos.TaskState_TASK_RUNNING { if *status.State == mesos.TaskState_TASK_RUNNING {
mutex.Lock() bpPistonMutex.Lock()
s.tasksRunning++ s.tasksRunning++
mutex.Unlock() bpPistonMutex.Unlock()
} else if IsTerminal(status.State) { } else if IsTerminal(status.State) {
delete(s.running[status.GetSlaveId().GoString()], *status.TaskId.Value) delete(s.running[status.GetSlaveId().GoString()], *status.TaskId.Value)
// Deleting the task from the taskMonitor // Deleting the task from the taskMonitor
@ -361,14 +369,14 @@ func (s *PistonCapper) StatusUpdate(driver sched.SchedulerDriver, status *mesos.
} }
// Need to update the cap values for host of the finishedTask // Need to update the cap values for host of the finishedTask
mutex.Lock() bpPistonMutex.Lock()
capValues[hostOfFinishedTask] -= ((finishedTask.Watts * constants.CapMargin) / s.totalPower[hostOfFinishedTask]) * 100 bpPistonCapValues[hostOfFinishedTask] -= ((finishedTask.Watts * constants.CapMargin) / s.totalPower[hostOfFinishedTask]) * 100
// Checking to see if the cap value has become 0, in which case we uncap the host. // Checking to see if the cap value has become 0, in which case we uncap the host.
if int(math.Floor(capValues[hostOfFinishedTask]+0.5)) == 0 { if int(math.Floor(bpPistonCapValues[hostOfFinishedTask]+0.5)) == 0 {
capValues[hostOfFinishedTask] = 100 bpPistonCapValues[hostOfFinishedTask] = 100
} }
s.tasksRunning-- s.tasksRunning--
mutex.Unlock() bpPistonMutex.Unlock()
if s.tasksRunning == 0 { if s.tasksRunning == 0 {
select { select {
@ -381,27 +389,3 @@ func (s *PistonCapper) StatusUpdate(driver sched.SchedulerDriver, status *mesos.
} }
log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value) log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value)
} }
func (s *PistonCapper) 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 *PistonCapper) OfferRescinded(_ sched.SchedulerDriver, offerID *mesos.OfferID) {
log.Printf("Offer %s rescinded", offerID)
}
func (s *PistonCapper) SlaveLost(_ sched.SchedulerDriver, slaveID *mesos.SlaveID) {
log.Printf("Slave %s lost", slaveID)
}
func (s *PistonCapper) ExecutorLost(_ sched.SchedulerDriver, executorID *mesos.ExecutorID, slaveID *mesos.SlaveID, status int) {
log.Printf("Executor %s on slave %s was lost", executorID, slaveID)
}
func (s *PistonCapper) Error(_ sched.SchedulerDriver, err string) {
log.Printf("Receiving an error: %s", err)
}

58
schedulers/binpacksortedwatts.go
View file

@ -8,6 +8,7 @@ import (
"github.com/mesos/mesos-go/mesosutil" "github.com/mesos/mesos-go/mesosutil"
sched "github.com/mesos/mesos-go/scheduler" sched "github.com/mesos/mesos-go/scheduler"
"log" "log"
"os"
"sort" "sort"
"strings" "strings"
"time" "time"
@ -28,6 +29,7 @@ func (*BinPackSortedWatts) takeOffer(offer *mesos.Offer, task def.Task) bool {
} }
type BinPackSortedWatts struct { type BinPackSortedWatts struct {
base // Type embedded to inherit common functions
tasksCreated int tasksCreated int
tasksRunning int tasksRunning int
tasks []def.Task tasks []def.Task
@ -48,12 +50,19 @@ type BinPackSortedWatts struct {
// Controls when to shutdown pcp logging // Controls when to shutdown pcp logging
PCPLog chan struct{} PCPLog chan struct{}
schedTrace *log.Logger
} }
// New electron scheduler // New electron scheduler
func NewBinPackSortedWatts(tasks []def.Task, ignoreWatts bool) *BinPackSortedWatts { func NewBinPackSortedWatts(tasks []def.Task, ignoreWatts bool, schedTracePrefix string) *BinPackSortedWatts {
sort.Sort(def.WattsSorter(tasks)) sort.Sort(def.WattsSorter(tasks))
logFile, err := os.Create("./" + schedTracePrefix + "_schedTrace.log")
if err != nil {
log.Fatal(err)
}
s := &BinPackSortedWatts{ s := &BinPackSortedWatts{
tasks: tasks, tasks: tasks,
ignoreWatts: ignoreWatts, ignoreWatts: ignoreWatts,
@ -62,6 +71,7 @@ func NewBinPackSortedWatts(tasks []def.Task, ignoreWatts bool) *BinPackSortedWat
PCPLog: make(chan struct{}), PCPLog: make(chan struct{}),
running: make(map[string]map[string]bool), running: make(map[string]map[string]bool),
RecordPCP: false, RecordPCP: false,
schedTrace: log.New(logFile, "", log.LstdFlags),
} }
return s return s
} }
@ -113,21 +123,6 @@ func (s *BinPackSortedWatts) newTask(offer *mesos.Offer, task def.Task) *mesos.T
} }
} }
func (s *BinPackSortedWatts) Registered(
_ sched.SchedulerDriver,
frameworkID *mesos.FrameworkID,
masterInfo *mesos.MasterInfo) {
log.Printf("Framework %s registered with master %s", frameworkID, masterInfo)
}
func (s *BinPackSortedWatts) Reregistered(_ sched.SchedulerDriver, masterInfo *mesos.MasterInfo) {
log.Printf("Framework re-registered with master %s", masterInfo)
}
func (s *BinPackSortedWatts) Disconnected(sched.SchedulerDriver) {
log.Println("Framework disconnected with master")
}
func (s *BinPackSortedWatts) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) { func (s *BinPackSortedWatts) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) {
log.Printf("Received %d resource offers", len(offers)) log.Printf("Received %d resource offers", len(offers))
@ -150,7 +145,8 @@ func (s *BinPackSortedWatts) ResourceOffers(driver sched.SchedulerDriver, offers
totalWatts := 0.0 totalWatts := 0.0
totalCPU := 0.0 totalCPU := 0.0
totalRAM := 0.0 totalRAM := 0.0
for i, task := range s.tasks { for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Check host if it exists // Check host if it exists
if task.Host != "" { if task.Host != "" {
@ -172,9 +168,11 @@ func (s *BinPackSortedWatts) ResourceOffers(driver sched.SchedulerDriver, offers
totalRAM += task.RAM totalRAM += task.RAM
log.Println("Co-Located with: ") log.Println("Co-Located with: ")
coLocated(s.running[offer.GetSlaveId().GoString()]) coLocated(s.running[offer.GetSlaveId().GoString()])
tasks = append(tasks, s.newTask(offer, task)) taskToSchedule := s.newTask(offer, task)
tasks = append(tasks, taskToSchedule)
fmt.Println("Inst: ", *task.Instances) fmt.Println("Inst: ", *task.Instances)
s.schedTrace.Print(offer.GetHostname() + ":" + taskToSchedule.GetTaskId().GetValue())
*task.Instances-- *task.Instances--
if *task.Instances <= 0 { if *task.Instances <= 0 {
@ -225,27 +223,3 @@ func (s *BinPackSortedWatts) StatusUpdate(driver sched.SchedulerDriver, status *
} }
log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value) log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value)
} }
func (s *BinPackSortedWatts) 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 *BinPackSortedWatts) OfferRescinded(_ sched.SchedulerDriver, offerID *mesos.OfferID) {
log.Printf("Offer %s rescinded", offerID)
}
func (s *BinPackSortedWatts) SlaveLost(_ sched.SchedulerDriver, slaveID *mesos.SlaveID) {
log.Printf("Slave %s lost", slaveID)
}
func (s *BinPackSortedWatts) ExecutorLost(_ sched.SchedulerDriver, executorID *mesos.ExecutorID, slaveID *mesos.SlaveID, status int) {
log.Printf("Executor %s on slave %s was lost", executorID, slaveID)
}
func (s *BinPackSortedWatts) Error(_ sched.SchedulerDriver, err string) {
log.Printf("Receiving an error: %s", err)
}

4
schedulers/bpMaxMin.go
View file

@ -124,7 +124,6 @@ func (s *BPMaxMinWatts) newTask(offer *mesos.Offer, task def.Task) *mesos.TaskIn
} }
} }
// Determine if the remaining space inside of the offer is enough for this // Determine if the remaining space inside of the offer is enough for this
// the task we need to create. If it is, create a TaskInfo and return it. // the task we need to create. If it is, create a TaskInfo and return it.
func (s *BPMaxMinWatts) CheckFit(i int, func (s *BPMaxMinWatts) CheckFit(i int,
@ -194,7 +193,7 @@ func (s *BPMaxMinWatts) ResourceOffers(driver sched.SchedulerDriver, offers []*m
// Attempt to schedule a single instance of the heaviest workload available first // Attempt to schedule a single instance of the heaviest workload available first
// Start from the back until one fits // Start from the back until one fits
for i:= len(s.tasks)-1; i >= 0; i-- { for i := len(s.tasks) - 1; i >= 0; i-- {
task := s.tasks[i] task := s.tasks[i]
// Check host if it exists // Check host if it exists
@ -239,7 +238,6 @@ func (s *BPMaxMinWatts) ResourceOffers(driver sched.SchedulerDriver, offers []*m
} }
} }
if offerTaken { if offerTaken {
log.Printf("Starting on [%s]\n", offer.GetHostname()) log.Printf("Starting on [%s]\n", offer.GetHostname())
driver.LaunchTasks([]*mesos.OfferID{offer.Id}, tasks, defaultFilter) driver.LaunchTasks([]*mesos.OfferID{offer.Id}, tasks, defaultFilter)

2
schedulers/bpMaxMinProacCC.go
View file

@ -155,7 +155,7 @@ func (s *BPMaxMinProacCC) newTask(offer *mesos.Offer, task def.Task) *mesos.Task
} }
// go routine to cap the entire cluster in regular intervals of time. // go routine to cap the entire cluster in regular intervals of time.
var bpMaxMinProacCCCapValue = 0.0 // initial value to indicate that we haven't capped the cluster yet. var bpMaxMinProacCCCapValue = 0.0 // initial value to indicate that we haven't capped the cluster yet.
var bpMaxMinProacCCNewCapValue = 0.0 // newly computed cap value var bpMaxMinProacCCNewCapValue = 0.0 // newly computed cap value
func (s *BPMaxMinProacCC) startCapping() { func (s *BPMaxMinProacCC) startCapping() {
go func() { go func() {

3
schedulers/bpswClassMapWatts.go
View file

@ -145,7 +145,8 @@ func (s *BPSWClassMapWatts) ResourceOffers(driver sched.SchedulerDriver, offers
totalWatts := 0.0 totalWatts := 0.0
totalCPU := 0.0 totalCPU := 0.0
totalRAM := 0.0 totalRAM := 0.0
for i, task := range s.tasks { for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Check host if it exists // Check host if it exists
if task.Host != "" { if task.Host != "" {

8
schedulers/bpswClassMapWattsPistonCapping.go
View file

@ -179,7 +179,7 @@ func (s *BPSWClassMapWattsPistonCapping) startCapping() {
if err := rapl.Cap(host, "rapl", roundedCapValue); err != nil { if err := rapl.Cap(host, "rapl", roundedCapValue); err != nil {
log.Println(err) log.Println(err)
} else { } else {
log.Printf("Capped [%s] at %d", host, int(math.Floor(capValue))) log.Printf("Capped [%s] at %d", host, roundedCapValue)
} }
bpswClassMapWattsPistonPreviousRoundedCapValues[host] = roundedCapValue bpswClassMapWattsPistonPreviousRoundedCapValues[host] = roundedCapValue
} }
@ -187,7 +187,7 @@ func (s *BPSWClassMapWattsPistonCapping) startCapping() {
if err := rapl.Cap(host, "rapl", roundedCapValue); err != nil { if err := rapl.Cap(host, "rapl", roundedCapValue); err != nil {
log.Println(err) log.Println(err)
} else { } else {
log.Printf("Capped [%s] at %d", host, int(math.Floor(capValue+0.5))) log.Printf("Capped [%s] at %d", host, roundedCapValue)
} }
bpswClassMapWattsPistonPreviousRoundedCapValues[host] = roundedCapValue bpswClassMapWattsPistonPreviousRoundedCapValues[host] = roundedCapValue
} }
@ -247,8 +247,8 @@ func (s *BPSWClassMapWattsPistonCapping) ResourceOffers(driver sched.SchedulerDr
// Store the partialLoad for host corresponding to this offer // Store the partialLoad for host corresponding to this offer
// Once we can't fit any more tasks, we update the capValue for this host with partialLoad and then launch the fitted tasks. // Once we can't fit any more tasks, we update the capValue for this host with partialLoad and then launch the fitted tasks.
partialLoad := 0.0 partialLoad := 0.0
for i, task := range s.tasks { for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Check host if it exists // Check host if it exists
if task.Host != "" { if task.Host != "" {
// Don't take offer if it doesn't match our task's host requirement // Don't take offer if it doesn't match our task's host requirement

11
schedulers/bpswClassMapWattsProacCC.go
View file

@ -165,7 +165,7 @@ func (s *BPSWClassMapWattsProacCC) Disconnected(sched.SchedulerDriver) {
} }
// go routine to cap the entire cluster in regular intervals of time. // go routine to cap the entire cluster in regular intervals of time.
var bpswClassMapWattsProacCCCapValue = 0.0 // initial value to indicate that we haven't capped the cluster yet. var bpswClassMapWattsProacCCCapValue = 0.0 // initial value to indicate that we haven't capped the cluster yet.
var bpswClassMapWattsProacCCNewCapValue = 0.0 // newly computed cap value var bpswClassMapWattsProacCCNewCapValue = 0.0 // newly computed cap value
func (s *BPSWClassMapWattsProacCC) startCapping() { func (s *BPSWClassMapWattsProacCC) startCapping() {
go func() { go func() {
@ -208,11 +208,11 @@ func (s *BPSWClassMapWattsProacCC) startRecapping() {
if s.isRecapping && bpswClassMapWattsProacCCRecapValue > 0.0 { if s.isRecapping && bpswClassMapWattsProacCCRecapValue > 0.0 {
for _, host := range constants.Hosts { for _, host := range constants.Hosts {
// Rounding capValue to the nearest int // Rounding capValue to the nearest int
if err := rapl.Cap(host, "rapl", int(math.Floor(bpswClassMapWattsProacCCRecapValue +0.5))); err != nil { if err := rapl.Cap(host, "rapl", int(math.Floor(bpswClassMapWattsProacCCRecapValue+0.5))); err != nil {
log.Println(err) log.Println(err)
} }
} }
log.Printf("Recapping the cluster to %d", int(math.Floor(bpswClassMapWattsProacCCRecapValue +0.5))) log.Printf("Recapping the cluster to %d", int(math.Floor(bpswClassMapWattsProacCCRecapValue+0.5)))
} }
// Setting recapping to false // Setting recapping to false
s.isRecapping = false s.isRecapping = false
@ -282,7 +282,8 @@ func (s *BPSWClassMapWattsProacCC) ResourceOffers(driver sched.SchedulerDriver,
totalWatts := 0.0 totalWatts := 0.0
totalCPU := 0.0 totalCPU := 0.0
totalRAM := 0.0 totalRAM := 0.0
for i, task := range s.tasks { for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Check host if it exists // Check host if it exists
if task.Host != "" { if task.Host != "" {
// Don't take offer it it doesn't match our task's host requirement. // Don't take offer it it doesn't match our task's host requirement.
@ -382,7 +383,7 @@ func (s *BPSWClassMapWattsProacCC) StatusUpdate(driver sched.SchedulerDriver, st
tempCap, err := s.capper.CleverRecap(s.totalPower, s.taskMonitor, *status.TaskId.Value) tempCap, err := s.capper.CleverRecap(s.totalPower, s.taskMonitor, *status.TaskId.Value)
if err == nil { if err == nil {
// If new determined cap value is different from the current recap value, then we need to recap // If new determined cap value is different from the current recap value, then we need to recap
if int(math.Floor(tempCap+0.5)) != int(math.Floor(bpswClassMapWattsProacCCRecapValue +0.5)) { if int(math.Floor(tempCap+0.5)) != int(math.Floor(bpswClassMapWattsProacCCRecapValue+0.5)) {
bpswClassMapWattsProacCCRecapValue = tempCap bpswClassMapWattsProacCCRecapValue = tempCap
bpswClassMapWattsProacCCMutex.Lock() bpswClassMapWattsProacCCMutex.Lock()
s.isRecapping = true s.isRecapping = true

58
schedulers/firstfit.go
View file

@ -8,6 +8,7 @@ import (
"github.com/mesos/mesos-go/mesosutil" "github.com/mesos/mesos-go/mesosutil"
sched "github.com/mesos/mesos-go/scheduler" sched "github.com/mesos/mesos-go/scheduler"
"log" "log"
"os"
"strings" "strings"
"time" "time"
) )
@ -28,6 +29,7 @@ func (s *FirstFit) takeOffer(offer *mesos.Offer, task def.Task) bool {
// electronScheduler implements the Scheduler interface // electronScheduler implements the Scheduler interface
type FirstFit struct { type FirstFit struct {
base // Type embedded to inherit common functions
tasksCreated int tasksCreated int
tasksRunning int tasksRunning int
tasks []def.Task tasks []def.Task
@ -48,10 +50,17 @@ type FirstFit struct {
// Controls when to shutdown pcp logging // Controls when to shutdown pcp logging
PCPLog chan struct{} PCPLog chan struct{}
schedTrace *log.Logger
} }
// New electron scheduler // New electron scheduler
func NewFirstFit(tasks []def.Task, ignoreWatts bool) *FirstFit { func NewFirstFit(tasks []def.Task, ignoreWatts bool, schedTracePrefix string) *FirstFit {
logFile, err := os.Create("./" + schedTracePrefix + "_schedTrace.log")
if err != nil {
log.Fatal(err)
}
s := &FirstFit{ s := &FirstFit{
tasks: tasks, tasks: tasks,
@ -61,6 +70,7 @@ func NewFirstFit(tasks []def.Task, ignoreWatts bool) *FirstFit {
PCPLog: make(chan struct{}), PCPLog: make(chan struct{}),
running: make(map[string]map[string]bool), running: make(map[string]map[string]bool),
RecordPCP: false, RecordPCP: false,
schedTrace: log.New(logFile, "", log.LstdFlags),
} }
return s return s
} }
@ -112,21 +122,6 @@ func (s *FirstFit) newTask(offer *mesos.Offer, task def.Task) *mesos.TaskInfo {
} }
} }
func (s *FirstFit) Registered(
_ sched.SchedulerDriver,
frameworkID *mesos.FrameworkID,
masterInfo *mesos.MasterInfo) {
log.Printf("Framework %s registered with master %s", frameworkID, masterInfo)
}
func (s *FirstFit) Reregistered(_ sched.SchedulerDriver, masterInfo *mesos.MasterInfo) {
log.Printf("Framework re-registered with master %s", masterInfo)
}
func (s *FirstFit) Disconnected(sched.SchedulerDriver) {
log.Println("Framework disconnected with master")
}
func (s *FirstFit) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) { func (s *FirstFit) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) {
log.Printf("Received %d resource offers", len(offers)) log.Printf("Received %d resource offers", len(offers))
@ -146,7 +141,8 @@ func (s *FirstFit) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.
// First fit strategy // First fit strategy
taken := false taken := false
for i, task := range s.tasks { for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Check host if it exists // Check host if it exists
if task.Host != "" { if task.Host != "" {
@ -162,7 +158,8 @@ func (s *FirstFit) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.
log.Println("Co-Located with: ") log.Println("Co-Located with: ")
coLocated(s.running[offer.GetSlaveId().GoString()]) coLocated(s.running[offer.GetSlaveId().GoString()])
tasks = append(tasks, s.newTask(offer, task)) taskToSchedule := s.newTask(offer, task)
tasks = append(tasks, taskToSchedule)
log.Printf("Starting %s on [%s]\n", task.Name, offer.GetHostname()) log.Printf("Starting %s on [%s]\n", task.Name, offer.GetHostname())
driver.LaunchTasks([]*mesos.OfferID{offer.Id}, tasks, defaultFilter) driver.LaunchTasks([]*mesos.OfferID{offer.Id}, tasks, defaultFilter)
@ -170,6 +167,7 @@ func (s *FirstFit) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.
taken = true taken = true
fmt.Println("Inst: ", *task.Instances) fmt.Println("Inst: ", *task.Instances)
s.schedTrace.Print(offer.GetHostname() + ":" + taskToSchedule.GetTaskId().GetValue())
*task.Instances-- *task.Instances--
if *task.Instances <= 0 { if *task.Instances <= 0 {
@ -216,27 +214,3 @@ func (s *FirstFit) StatusUpdate(driver sched.SchedulerDriver, status *mesos.Task
} }
log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value) log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value)
} }
func (s *FirstFit) 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 *FirstFit) OfferRescinded(_ sched.SchedulerDriver, offerID *mesos.OfferID) {
log.Printf("Offer %s rescinded", offerID)
}
func (s *FirstFit) SlaveLost(_ sched.SchedulerDriver, slaveID *mesos.SlaveID) {
log.Printf("Slave %s lost", slaveID)
}
func (s *FirstFit) ExecutorLost(_ sched.SchedulerDriver, executorID *mesos.ExecutorID, slaveID *mesos.SlaveID, status int) {
log.Printf("Executor %s on slave %s was lost", executorID, slaveID)
}
func (s *FirstFit) Error(_ sched.SchedulerDriver, err string) {
log.Printf("Receiving an error: %s", err)
}

205
schedulers/firstfitSortedWattsClassMapWatts.go Normal file
View file

@ -0,0 +1,205 @@
package schedulers
import (
"bitbucket.org/sunybingcloud/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"
"os"
"sort"
"strings"
"time"
)
// electron scheduler implements the Scheduler interface
type FirstFitSortedWattsClassMapWatts struct {
base // Type embedded to inherit common features.
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{}
schedTrace *log.Logger
}
// New electorn scheduler
func NewFirstFitSortedWattsClassMapWatts(tasks []def.Task, ignoreWatts bool, schedTracePrefix string) *FirstFitSortedWattsClassMapWatts {
sort.Sort(def.WattsSorter(tasks))
logFile, err := os.Create("./" + schedTracePrefix + "_schedTrace.log")
if err != nil {
log.Fatal(err)
}
s := &FirstFitSortedWattsClassMapWatts{
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,
schedTrace: log.New(logFile, "", log.LstdFlags),
}
return s
}
func (s *FirstFitSortedWattsClassMapWatts) newTask(offer *mesos.Offer, task def.Task, newTaskClass string) *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("cpus", task.CPU),
mesosutil.NewScalarResource("mem", task.RAM),
}
if !s.ignoreWatts {
resources = append(resources, mesosutil.NewScalarResource("watts", task.ClassToWatts[newTaskClass]))
}
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 *FirstFitSortedWattsClassMapWatts) 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:
}
offerCPU, offerRAM, offerWatts := OfferAgg(offer)
// First fit strategy
taken := false
for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Check host if it exists
if task.Host != "" {
// Don't take offer if it doens't match our task's host requirement.
if !strings.HasPrefix(*offer.Hostname, task.Host) {
continue
}
}
// retrieving the node class from the offer
var nodeClass string
for _, attr := range offer.GetAttributes() {
if attr.GetName() == "class" {
nodeClass = attr.GetText().GetValue()
}
}
// Decision to take the offer or not
if (s.ignoreWatts || (offerWatts >= task.ClassToWatts[nodeClass])) &&
(offerCPU >= task.CPU) && (offerRAM >= task.RAM) {
log.Println("Co-Located with: ")
coLocated(s.running[offer.GetSlaveId().GoString()])
taskToSchedule := s.newTask(offer, task, nodeClass)
s.schedTrace.Print(offer.GetHostname() + ":" + taskToSchedule.GetTaskId().GetValue())
log.Printf("Starting %s on [%s]\n", task.Name, offer.GetHostname())
driver.LaunchTasks([]*mesos.OfferID{offer.Id}, []*mesos.TaskInfo{taskToSchedule}, 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 = append(s.tasks[:i], s.tasks[i+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("<CPU: %f, RAM: %f, Watts: %f>\n", cpus, mem, watts)
driver.DeclineOffer(offer.Id, defaultFilter)
}
}
}
func (s *FirstFitSortedWattsClassMapWatts) 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)
}

389
schedulers/firstfitSortedWattsClassMapWattsProacCC.go Normal file
View file

@ -0,0 +1,389 @@
package schedulers
import (
"bitbucket.org/sunybingcloud/electron/constants"
"bitbucket.org/sunybingcloud/electron/def"
"bitbucket.org/sunybingcloud/electron/pcp"
"bitbucket.org/sunybingcloud/electron/rapl"
"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"
"math"
"os"
"sort"
"strings"
"sync"
"time"
)
// electron scheduler implements the Scheduler interface
type FirstFitSortedWattsClassMapWattsProacCC struct {
base // Type embedded to inherit common features.
tasksCreated int
tasksRunning int
tasks []def.Task
metrics map[string]def.Metric
running map[string]map[string]bool
taskMonitor map[string][]def.Task
availablePower map[string]float64
totalPower map[string]float64
ignoreWatts bool
capper *pcp.ClusterwideCapper
ticker *time.Ticker
recapTicker *time.Ticker
isCapping bool // indicate whether we are currently performing cluster-wide capping.
isRecapping bool // indicate whether we are currently performing cluster-wide recapping.
// 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{}
schedTrace *log.Logger
}
// New electron scheduler
func NewFirstFitSortedWattsClassMapWattsProacCC(tasks []def.Task, ignoreWatts bool, schedTracePrefix string) *FirstFitSortedWattsClassMapWattsProacCC {
sort.Sort(def.WattsSorter(tasks))
logFile, err := os.Create("./" + schedTracePrefix + "_schedTrace.log")
if err != nil {
log.Fatal(err)
}
s := &FirstFitSortedWattsClassMapWattsProacCC{
tasks: tasks,
ignoreWatts: ignoreWatts,
Shutdown: make(chan struct{}),
Done: make(chan struct{}),
PCPLog: make(chan struct{}),
running: make(map[string]map[string]bool),
taskMonitor: make(map[string][]def.Task),
availablePower: make(map[string]float64),
totalPower: make(map[string]float64),
RecordPCP: false,
capper: pcp.GetClusterwideCapperInstance(),
ticker: time.NewTicker(10 * time.Second),
recapTicker: time.NewTicker(20 * time.Second),
isCapping: false,
isRecapping: false,
schedTrace: log.New(logFile, "", log.LstdFlags),
}
return s
}
// mutex
var ffswClassMapWattsProacCCMutex sync.Mutex
func (s *FirstFitSortedWattsClassMapWattsProacCC) newTask(offer *mesos.Offer, task def.Task, newTaskClass string) *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)
}
// Setting the task ID to the task. This is done so that we can consider each task to be different,
// even though they have the same parameters.
task.SetTaskID(*proto.String("electron-" + taskName))
// Add task to the list of tasks running on the node.
s.running[offer.GetSlaveId().GoString()][taskName] = true
if len(s.taskMonitor[*offer.Hostname]) == 0 {
s.taskMonitor[*offer.Hostname] = []def.Task{task}
} else {
s.taskMonitor[*offer.Hostname] = append(s.taskMonitor[*offer.Hostname], task)
}
resources := []*mesos.Resource{
mesosutil.NewScalarResource("cpus", task.CPU),
mesosutil.NewScalarResource("mem", task.RAM),
}
if !s.ignoreWatts {
resources = append(resources, mesosutil.NewScalarResource("watts", task.ClassToWatts[newTaskClass]))
}
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 *FirstFitSortedWattsClassMapWattsProacCC) Disconnected(sched.SchedulerDriver) {
// Need to stop the capping process
s.ticker.Stop()
s.recapTicker.Stop()
ffswClassMapWattsProacCCMutex.Lock()
s.isCapping = false
ffswClassMapWattsProacCCMutex.Unlock()
log.Println("Framework disconnected with master")
}
// go routine to cap the entire cluster in regular intervals of time
var ffswClassMapWattsProacCCCapValue = 0.0 // initial value to indicate that we haven't capped the cluster yet.
var ffswClassMapWattsProacCCNewCapValue = 0.0 // newly computed cap value
func (s *FirstFitSortedWattsClassMapWattsProacCC) startCapping() {
go func() {
for {
select {
case <-s.ticker.C:
// Need to cap the cluster only if new cap value different from the old cap value.
// This way we don't unnecessarily cap the cluster.
ffswClassMapWattsProacCCMutex.Lock()
if s.isCapping {
if int(math.Floor(ffswClassMapWattsProacCCNewCapValue+0.5)) != int(math.Floor(ffswClassMapWattsProacCCCapValue+0.5)) {
// updating cap value
ffswClassMapWattsProacCCCapValue = ffswClassMapWattsProacCCNewCapValue
if ffswClassMapWattsProacCCCapValue > 0.0 {
for _, host := range constants.Hosts {
// Rounding cap value to the nearest int
if err := rapl.Cap(host, "rapl", int(math.Floor(ffswClassMapWattsProacCCCapValue+0.5))); err != nil {
log.Println(err)
}
}
log.Printf("Capped the cluster to %d", int(math.Floor(ffswClassMapWattsProacCCCapValue+0.5)))
}
}
}
ffswClassMapWattsProacCCMutex.Unlock()
}
}
}()
}
// go routine to recap the entire cluster in regular intervals of time.
var ffswClassMapWattsProacCCRecapValue = 0.0 // The cluster-wide cap value when recapping.
func (s *FirstFitSortedWattsClassMapWattsProacCC) startRecapping() {
go func() {
for {
select {
case <-s.recapTicker.C:
ffswClassMapWattsProacCCMutex.Lock()
// If stopped performing cluster wide capping, then we need to recap
if s.isRecapping && ffswClassMapWattsProacCCRecapValue > 0.0 {
for _, host := range constants.Hosts {
// Rounding the cap value to the nearest int
if err := rapl.Cap(host, "rapl", int(math.Floor(ffswClassMapWattsProacCCRecapValue+0.5))); err != nil {
log.Println(err)
}
}
log.Printf("Recapping the cluster to %d", int(math.Floor(ffswClassMapWattsProacCCRecapValue+0.5)))
}
// Setting recapping to false
s.isRecapping = false
ffswClassMapWattsProacCCMutex.Unlock()
}
}
}()
}
// Stop the cluster wide capping
func (s *FirstFitSortedWattsClassMapWattsProacCC) stopCapping() {
if s.isCapping {
log.Println("Stopping the cluster-wide capping.")
s.ticker.Stop()
ffswClassMapWattsProacCCMutex.Lock()
s.isCapping = false
s.isRecapping = true
ffswClassMapWattsProacCCMutex.Unlock()
}
}
// Stop the cluster wide recapping
func (s *FirstFitSortedWattsClassMapWattsProacCC) stopRecapping() {
// If not capping, then definitely recapping.
if !s.isCapping && s.isRecapping {
log.Println("Stopping the cluster-wide re-capping.")
s.recapTicker.Stop()
ffswClassMapWattsProacCCMutex.Lock()
s.isRecapping = false
ffswClassMapWattsProacCCMutex.Unlock()
}
}
func (s *FirstFitSortedWattsClassMapWattsProacCC) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) {
log.Printf("Received %d resource offers", len(offers))
// retrieving the available power for all the hosts in the offers.
for _, offer := range offers {
_, _, offerWatts := OfferAgg(offer)
s.availablePower[*offer.Hostname] = offerWatts
// setting total power if the first time
if _, ok := s.totalPower[*offer.Hostname]; !ok {
s.totalPower[*offer.Hostname] = offerWatts
}
}
for host, tpower := range s.totalPower {
log.Printf("TotalPower[%s] = %f", host, tpower)
}
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:
}
offerCPU, offerRAM, offerWatts := OfferAgg(offer)
// First fit strategy
taken := false
for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Check host if it exists
if task.Host != "" {
// Don't take offer if it doens't match our task's host requirement.
if !strings.HasPrefix(*offer.Hostname, task.Host) {
continue
}
}
// Retrieving the node class from the offer
var nodeClass string
for _, attr := range offer.GetAttributes() {
if attr.GetName() == "class" {
nodeClass = attr.GetText().GetValue()
}
}
// Decision to take the offer or not
if (s.ignoreWatts || (offerWatts >= task.ClassToWatts[nodeClass])) &&
(offerCPU >= task.CPU) && (offerRAM >= task.RAM) {
// Capping the cluster if haven't yet started
if !s.isCapping {
ffswClassMapWattsProacCCMutex.Lock()
s.isCapping = true
ffswClassMapWattsProacCCMutex.Unlock()
s.startCapping()
}
fmt.Println("Watts being used: ", task.ClassToWatts[nodeClass])
tempCap, err := s.capper.FCFSDeterminedCap(s.totalPower, &task)
if err == nil {
ffswClassMapWattsProacCCMutex.Lock()
ffswClassMapWattsProacCCNewCapValue = tempCap
ffswClassMapWattsProacCCMutex.Unlock()
} else {
log.Println("Failed to determine new cluster-wide cap: ")
log.Println(err)
}
log.Println("Co-Located with: ")
coLocated(s.running[offer.GetSlaveId().GoString()])
taskToSchedule := s.newTask(offer, task, nodeClass)
s.schedTrace.Print(offer.GetHostname() + ":" + taskToSchedule.GetTaskId().GetValue())
log.Printf("Starting %s on [%s]\n", task.Name, offer.GetHostname())
driver.LaunchTasks([]*mesos.OfferID{offer.Id}, []*mesos.TaskInfo{taskToSchedule}, 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 = append(s.tasks[:i], s.tasks[i+1:]...)
if len(s.tasks) == 0 {
log.Println("Done scheduling all tasks")
// Need to stop the cluster-wide capping as there aren't any more tasks to schedule
s.stopCapping()
s.startRecapping() // Load changes after every task finishes and hence, we need to change the capping of the cluster
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("<CPU: %f, RAM: %f, Watts: %f>\n", cpus, mem, watts)
driver.DeclineOffer(offer.Id, defaultFilter)
}
}
}
func (s *FirstFitSortedWattsClassMapWattsProacCC) 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)
// Need to remove the task from the window
s.capper.TaskFinished(*status.TaskId.Value)
// Determining the new cluster wide recap value
//tempCap, err := s.capper.NaiveRecap(s.totalPower, s.taskMonitor, *status.TaskId.Value)
tempCap, err := s.capper.CleverRecap(s.totalPower, s.taskMonitor, *status.TaskId.Value)
if err == nil {
// If new determined cap value is different from the current recap value, then we need to recap
if int(math.Floor(tempCap+0.5)) != int(math.Floor(ffswClassMapWattsProacCCRecapValue+0.5)) {
ffswClassMapWattsProacCCRecapValue = tempCap
ffswClassMapWattsProacCCMutex.Lock()
s.isRecapping = true
ffswClassMapWattsProacCCMutex.Unlock()
log.Printf("Determined re-cap value: %f\n", ffswClassMapWattsProacCCRecapValue)
} else {
ffswClassMapWattsProacCCMutex.Lock()
s.isRecapping = false
ffswClassMapWattsProacCCMutex.Unlock()
}
} else {
log.Println(err)
}
s.tasksRunning--
if s.tasksRunning == 0 {
select {
case <-s.Shutdown:
// Need to stop the cluster-wide recapping
s.stopRecapping()
close(s.Done)
default:
}
}
}
log.Printf("DONE: Task status [%s] for task[%s]", NameFor(status.State), *status.TaskId.Value)
}

58
schedulers/firstfitsortedwatts.go
View file

@ -8,6 +8,7 @@ import (
"github.com/mesos/mesos-go/mesosutil" "github.com/mesos/mesos-go/mesosutil"
sched "github.com/mesos/mesos-go/scheduler" sched "github.com/mesos/mesos-go/scheduler"
"log" "log"
"os"
"sort" "sort"
"strings" "strings"
"time" "time"
@ -29,6 +30,7 @@ func (s *FirstFitSortedWatts) takeOffer(offer *mesos.Offer, task def.Task) bool
// electronScheduler implements the Scheduler interface // electronScheduler implements the Scheduler interface
type FirstFitSortedWatts struct { type FirstFitSortedWatts struct {
base // Type embedded to inherit common functions
tasksCreated int tasksCreated int
tasksRunning int tasksRunning int
tasks []def.Task tasks []def.Task
@ -49,13 +51,20 @@ type FirstFitSortedWatts struct {
// Controls when to shutdown pcp logging // Controls when to shutdown pcp logging
PCPLog chan struct{} PCPLog chan struct{}
schedTrace *log.Logger
} }
// New electron scheduler // New electron scheduler
func NewFirstFitSortedWatts(tasks []def.Task, ignoreWatts bool) *FirstFitSortedWatts { func NewFirstFitSortedWatts(tasks []def.Task, ignoreWatts bool, schedTracePrefix string) *FirstFitSortedWatts {
sort.Sort(def.WattsSorter(tasks)) sort.Sort(def.WattsSorter(tasks))
logFile, err := os.Create("./" + schedTracePrefix + "_schedTrace.log")
if err != nil {
log.Fatal(err)
}
s := &FirstFitSortedWatts{ s := &FirstFitSortedWatts{
tasks: tasks, tasks: tasks,
ignoreWatts: ignoreWatts, ignoreWatts: ignoreWatts,
@ -64,6 +73,7 @@ func NewFirstFitSortedWatts(tasks []def.Task, ignoreWatts bool) *FirstFitSortedW
PCPLog: make(chan struct{}), PCPLog: make(chan struct{}),
running: make(map[string]map[string]bool), running: make(map[string]map[string]bool),
RecordPCP: false, RecordPCP: false,
schedTrace: log.New(logFile, "", log.LstdFlags),
} }
return s return s
} }
@ -115,21 +125,6 @@ func (s *FirstFitSortedWatts) newTask(offer *mesos.Offer, task def.Task) *mesos.
} }
} }
func (s *FirstFitSortedWatts) Registered(
_ sched.SchedulerDriver,
frameworkID *mesos.FrameworkID,
masterInfo *mesos.MasterInfo) {
log.Printf("Framework %s registered with master %s", frameworkID, masterInfo)
}
func (s *FirstFitSortedWatts) Reregistered(_ sched.SchedulerDriver, masterInfo *mesos.MasterInfo) {
log.Printf("Framework re-registered with master %s", masterInfo)
}
func (s *FirstFitSortedWatts) Disconnected(sched.SchedulerDriver) {
log.Println("Framework disconnected with master")
}
func (s *FirstFitSortedWatts) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) { func (s *FirstFitSortedWatts) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) {
log.Printf("Received %d resource offers", len(offers)) log.Printf("Received %d resource offers", len(offers))
@ -149,7 +144,8 @@ func (s *FirstFitSortedWatts) ResourceOffers(driver sched.SchedulerDriver, offer
// First fit strategy // First fit strategy
taken := false taken := false
for i, task := range s.tasks { for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Check host if it exists // Check host if it exists
if task.Host != "" { if task.Host != "" {
@ -165,7 +161,8 @@ func (s *FirstFitSortedWatts) ResourceOffers(driver sched.SchedulerDriver, offer
log.Println("Co-Located with: ") log.Println("Co-Located with: ")
coLocated(s.running[offer.GetSlaveId().GoString()]) coLocated(s.running[offer.GetSlaveId().GoString()])
tasks = append(tasks, s.newTask(offer, task)) taskToSchedule := s.newTask(offer, task)
tasks = append(tasks, taskToSchedule)
log.Printf("Starting %s on [%s]\n", task.Name, offer.GetHostname()) log.Printf("Starting %s on [%s]\n", task.Name, offer.GetHostname())
driver.LaunchTasks([]*mesos.OfferID{offer.Id}, tasks, defaultFilter) driver.LaunchTasks([]*mesos.OfferID{offer.Id}, tasks, defaultFilter)
@ -173,6 +170,7 @@ func (s *FirstFitSortedWatts) ResourceOffers(driver sched.SchedulerDriver, offer
taken = true taken = true
fmt.Println("Inst: ", *task.Instances) fmt.Println("Inst: ", *task.Instances)
s.schedTrace.Print(offer.GetHostname() + ":" + taskToSchedule.GetTaskId().GetValue())
*task.Instances-- *task.Instances--
if *task.Instances <= 0 { if *task.Instances <= 0 {
@ -218,27 +216,3 @@ func (s *FirstFitSortedWatts) StatusUpdate(driver sched.SchedulerDriver, status
} }
log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value) log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value)
} }
func (s *FirstFitSortedWatts) 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 *FirstFitSortedWatts) OfferRescinded(_ sched.SchedulerDriver, offerID *mesos.OfferID) {
log.Printf("Offer %s rescinded", offerID)
}
func (s *FirstFitSortedWatts) SlaveLost(_ sched.SchedulerDriver, slaveID *mesos.SlaveID) {
log.Printf("Slave %s lost", slaveID)
}
func (s *FirstFitSortedWatts) ExecutorLost(_ sched.SchedulerDriver, executorID *mesos.ExecutorID, slaveID *mesos.SlaveID, status int) {
log.Printf("Executor %s on slave %s was lost", executorID, slaveID)
}
func (s *FirstFitSortedWatts) Error(_ sched.SchedulerDriver, err string) {
log.Printf("Receiving an error: %s", err)
}

58
schedulers/firstfitwattsonly.go
View file

@ -8,6 +8,7 @@ import (
"github.com/mesos/mesos-go/mesosutil" "github.com/mesos/mesos-go/mesosutil"
sched "github.com/mesos/mesos-go/scheduler" sched "github.com/mesos/mesos-go/scheduler"
"log" "log"
"os"
"strings" "strings"
"time" "time"
) )
@ -27,6 +28,7 @@ func (*FirstFitWattsOnly) takeOffer(offer *mesos.Offer, task def.Task) bool {
} }
type FirstFitWattsOnly struct { type FirstFitWattsOnly struct {
base // Type embedded to inherit common functions
tasksCreated int tasksCreated int
tasksRunning int tasksRunning int
tasks []def.Task tasks []def.Task
@ -47,10 +49,17 @@ type FirstFitWattsOnly struct {
// Controls when to shutdown pcp logging // Controls when to shutdown pcp logging
PCPLog chan struct{} PCPLog chan struct{}
schedTrace *log.Logger
} }
// New electron scheduler // New electron scheduler
func NewFirstFitWattsOnly(tasks []def.Task, ignoreWatts bool) *FirstFitWattsOnly { func NewFirstFitWattsOnly(tasks []def.Task, ignoreWatts bool, schedTracePrefix string) *FirstFitWattsOnly {
logFile, err := os.Create("./" + schedTracePrefix + "_schedTrace.log")
if err != nil {
log.Fatal(err)
}
s := &FirstFitWattsOnly{ s := &FirstFitWattsOnly{
tasks: tasks, tasks: tasks,
@ -60,6 +69,7 @@ func NewFirstFitWattsOnly(tasks []def.Task, ignoreWatts bool) *FirstFitWattsOnly
PCPLog: make(chan struct{}), PCPLog: make(chan struct{}),
running: make(map[string]map[string]bool), running: make(map[string]map[string]bool),
RecordPCP: false, RecordPCP: false,
schedTrace: log.New(logFile, "", log.LstdFlags),
} }
return s return s
} }
@ -106,21 +116,6 @@ func (s *FirstFitWattsOnly) newTask(offer *mesos.Offer, task def.Task) *mesos.Ta
} }
} }
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) { func (s *FirstFitWattsOnly) ResourceOffers(driver sched.SchedulerDriver, offers []*mesos.Offer) {
log.Printf("Received %d resource offers", len(offers)) log.Printf("Received %d resource offers", len(offers))
@ -140,7 +135,8 @@ func (s *FirstFitWattsOnly) ResourceOffers(driver sched.SchedulerDriver, offers
// First fit strategy // First fit strategy
taken := false taken := false
for i, task := range s.tasks { for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Check host if it exists // Check host if it exists
if task.Host != "" { if task.Host != "" {
@ -156,7 +152,8 @@ func (s *FirstFitWattsOnly) ResourceOffers(driver sched.SchedulerDriver, offers
log.Println("Co-Located with: ") log.Println("Co-Located with: ")
coLocated(s.running[offer.GetSlaveId().GoString()]) coLocated(s.running[offer.GetSlaveId().GoString()])
tasks = append(tasks, s.newTask(offer, task)) taskToSchedule := s.newTask(offer, task)
tasks = append(tasks, taskToSchedule)
log.Printf("Starting %s on [%s]\n", task.Name, offer.GetHostname()) log.Printf("Starting %s on [%s]\n", task.Name, offer.GetHostname())
driver.LaunchTasks([]*mesos.OfferID{offer.Id}, tasks, defaultFilter) driver.LaunchTasks([]*mesos.OfferID{offer.Id}, tasks, defaultFilter)
@ -164,6 +161,7 @@ func (s *FirstFitWattsOnly) ResourceOffers(driver sched.SchedulerDriver, offers
taken = true taken = true
fmt.Println("Inst: ", *task.Instances) fmt.Println("Inst: ", *task.Instances)
s.schedTrace.Print(offer.GetHostname() + ":" + taskToSchedule.GetTaskId().GetValue())
*task.Instances-- *task.Instances--
if *task.Instances <= 0 { if *task.Instances <= 0 {
@ -210,27 +208,3 @@ func (s *FirstFitWattsOnly) StatusUpdate(driver sched.SchedulerDriver, status *m
} }
log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value) 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)
}

56
schedulers/proactiveclusterwidecappingfcfs.go
View file

@ -12,6 +12,7 @@ import (
sched "github.com/mesos/mesos-go/scheduler" sched "github.com/mesos/mesos-go/scheduler"
"log" "log"
"math" "math"
"os"
"strings" "strings"
"sync" "sync"
"time" "time"
@ -29,6 +30,7 @@ func (_ *ProactiveClusterwideCapFCFS) takeOffer(offer *mesos.Offer, task def.Tas
// electronScheduler implements the Scheduler interface. // electronScheduler implements the Scheduler interface.
type ProactiveClusterwideCapFCFS struct { type ProactiveClusterwideCapFCFS struct {
base // Type embedded to inherit common functions
tasksCreated int tasksCreated int
tasksRunning int tasksRunning int
tasks []def.Task tasks []def.Task
@ -58,10 +60,18 @@ type ProactiveClusterwideCapFCFS struct {
// Controls when to shutdown pcp logging. // Controls when to shutdown pcp logging.
PCPLog chan struct{} PCPLog chan struct{}
schedTrace *log.Logger
} }
// New electron scheduler. // New electron scheduler.
func NewProactiveClusterwideCapFCFS(tasks []def.Task, ignoreWatts bool) *ProactiveClusterwideCapFCFS { func NewProactiveClusterwideCapFCFS(tasks []def.Task, ignoreWatts bool, schedTracePrefix string) *ProactiveClusterwideCapFCFS {
logFile, err := os.Create("./" + schedTracePrefix + "_schedTrace.log")
if err != nil {
log.Fatal(err)
}
s := &ProactiveClusterwideCapFCFS{ s := &ProactiveClusterwideCapFCFS{
tasks: tasks, tasks: tasks,
ignoreWatts: ignoreWatts, ignoreWatts: ignoreWatts,
@ -78,6 +88,7 @@ func NewProactiveClusterwideCapFCFS(tasks []def.Task, ignoreWatts bool) *Proacti
recapTicker: time.NewTicker(20 * time.Second), recapTicker: time.NewTicker(20 * time.Second),
isCapping: false, isCapping: false,
isRecapping: false, isRecapping: false,
schedTrace: log.New(logFile, "", log.LstdFlags),
} }
return s return s
} }
@ -140,17 +151,6 @@ func (s *ProactiveClusterwideCapFCFS) newTask(offer *mesos.Offer, task def.Task)
} }
} }
func (s *ProactiveClusterwideCapFCFS) Registered(
_ sched.SchedulerDriver,
frameworkID *mesos.FrameworkID,
masterInfo *mesos.MasterInfo) {
log.Printf("Framework %s registered with master %s", frameworkID, masterInfo)
}
func (s *ProactiveClusterwideCapFCFS) Reregistered(_ sched.SchedulerDriver, masterInfo *mesos.MasterInfo) {
log.Printf("Framework re-registered with master %s", masterInfo)
}
func (s *ProactiveClusterwideCapFCFS) Disconnected(sched.SchedulerDriver) { func (s *ProactiveClusterwideCapFCFS) Disconnected(sched.SchedulerDriver) {
// Need to stop the capping process. // Need to stop the capping process.
s.ticker.Stop() s.ticker.Stop()
@ -275,7 +275,8 @@ func (s *ProactiveClusterwideCapFCFS) ResourceOffers(driver sched.SchedulerDrive
*/ */
taken := false taken := false
for i, task := range s.tasks { for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Don't take offer if it doesn't match our task's host requirement. // Don't take offer if it doesn't match our task's host requirement.
if !strings.HasPrefix(*offer.Hostname, task.Host) { if !strings.HasPrefix(*offer.Hostname, task.Host) {
continue continue
@ -302,9 +303,11 @@ func (s *ProactiveClusterwideCapFCFS) ResourceOffers(driver sched.SchedulerDrive
log.Println(err) log.Println(err)
} }
log.Printf("Starting on [%s]\n", offer.GetHostname()) log.Printf("Starting on [%s]\n", offer.GetHostname())
toSchedule := []*mesos.TaskInfo{s.newTask(offer, task)} taskToSchedule := s.newTask(offer, task)
toSchedule := []*mesos.TaskInfo{taskToSchedule}
driver.LaunchTasks([]*mesos.OfferID{offer.Id}, toSchedule, defaultFilter) driver.LaunchTasks([]*mesos.OfferID{offer.Id}, toSchedule, defaultFilter)
log.Printf("Inst: %d", *task.Instances) log.Printf("Inst: %d", *task.Instances)
s.schedTrace.Print(offer.GetHostname() + ":" + taskToSchedule.GetTaskId().GetValue())
*task.Instances-- *task.Instances--
if *task.Instances <= 0 { if *task.Instances <= 0 {
// All instances of the task have been scheduled. Need to remove it from the list of tasks to schedule. // All instances of the task have been scheduled. Need to remove it from the list of tasks to schedule.
@ -383,28 +386,3 @@ func (s *ProactiveClusterwideCapFCFS) StatusUpdate(driver sched.SchedulerDriver,
} }
log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value) log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value)
} }
func (s *ProactiveClusterwideCapFCFS) 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 *ProactiveClusterwideCapFCFS) OfferRescinded(_ sched.SchedulerDriver, offerID *mesos.OfferID) {
log.Printf("Offer %s rescinded", offerID)
}
func (s *ProactiveClusterwideCapFCFS) SlaveLost(_ sched.SchedulerDriver, slaveID *mesos.SlaveID) {
log.Printf("Slave %s lost", slaveID)
}
func (s *ProactiveClusterwideCapFCFS) ExecutorLost(_ sched.SchedulerDriver, executorID *mesos.ExecutorID, slaveID *mesos.SlaveID, status int) {
log.Printf("Executor %s on slave %s was lost", executorID, slaveID)
}
func (s *ProactiveClusterwideCapFCFS) Error(_ sched.SchedulerDriver, err string) {
log.Printf("Receiving an error: %s", err)
}

56
schedulers/proactiveclusterwidecappingranked.go
View file

@ -22,6 +22,7 @@ import (
sched "github.com/mesos/mesos-go/scheduler" sched "github.com/mesos/mesos-go/scheduler"
"log" "log"
"math" "math"
"os"
"sort" "sort"
"strings" "strings"
"sync" "sync"
@ -40,6 +41,7 @@ func (_ *ProactiveClusterwideCapRanked) takeOffer(offer *mesos.Offer, task def.T
// electronScheduler implements the Scheduler interface // electronScheduler implements the Scheduler interface
type ProactiveClusterwideCapRanked struct { type ProactiveClusterwideCapRanked struct {
base // Type embedded to inherit common functions
tasksCreated int tasksCreated int
tasksRunning int tasksRunning int
tasks []def.Task tasks []def.Task
@ -69,10 +71,18 @@ type ProactiveClusterwideCapRanked struct {
// Controls when to shutdown pcp logging. // Controls when to shutdown pcp logging.
PCPLog chan struct{} PCPLog chan struct{}
schedTrace *log.Logger
} }
// New electron scheduler. // New electron scheduler.
func NewProactiveClusterwideCapRanked(tasks []def.Task, ignoreWatts bool) *ProactiveClusterwideCapRanked { func NewProactiveClusterwideCapRanked(tasks []def.Task, ignoreWatts bool, schedTracePrefix string) *ProactiveClusterwideCapRanked {
logFile, err := os.Create("./" + schedTracePrefix + "_schedTrace.log")
if err != nil {
log.Fatal(err)
}
s := &ProactiveClusterwideCapRanked{ s := &ProactiveClusterwideCapRanked{
tasks: tasks, tasks: tasks,
ignoreWatts: ignoreWatts, ignoreWatts: ignoreWatts,
@ -89,6 +99,7 @@ func NewProactiveClusterwideCapRanked(tasks []def.Task, ignoreWatts bool) *Proac
recapTicker: time.NewTicker(20 * time.Second), recapTicker: time.NewTicker(20 * time.Second),
isCapping: false, isCapping: false,
isRecapping: false, isRecapping: false,
schedTrace: log.New(logFile, "", log.LstdFlags),
} }
return s return s
} }
@ -151,17 +162,6 @@ func (s *ProactiveClusterwideCapRanked) newTask(offer *mesos.Offer, task def.Tas
} }
} }
func (s *ProactiveClusterwideCapRanked) Registered(
_ sched.SchedulerDriver,
frameworkID *mesos.FrameworkID,
masterInfo *mesos.MasterInfo) {
log.Printf("Framework %s registered with master %s", frameworkID, masterInfo)
}
func (s *ProactiveClusterwideCapRanked) Reregistered(_ sched.SchedulerDriver, masterInfo *mesos.MasterInfo) {
log.Printf("Framework re-registered with master %s", masterInfo)
}
func (s *ProactiveClusterwideCapRanked) Disconnected(sched.SchedulerDriver) { func (s *ProactiveClusterwideCapRanked) Disconnected(sched.SchedulerDriver) {
// Need to stop the capping process. // Need to stop the capping process.
s.ticker.Stop() s.ticker.Stop()
@ -299,7 +299,8 @@ func (s *ProactiveClusterwideCapRanked) ResourceOffers(driver sched.SchedulerDri
*/ */
taken := false taken := false
for i, task := range s.tasks { for i := 0; i < len(s.tasks); i++ {
task := s.tasks[i]
// Don't take offer if it doesn't match our task's host requirement. // Don't take offer if it doesn't match our task's host requirement.
if !strings.HasPrefix(*offer.Hostname, task.Host) { if !strings.HasPrefix(*offer.Hostname, task.Host) {
continue continue
@ -325,9 +326,11 @@ func (s *ProactiveClusterwideCapRanked) ResourceOffers(driver sched.SchedulerDri
log.Println("Failed to determine the new cluster wide cap: ", err) log.Println("Failed to determine the new cluster wide cap: ", err)
} }
log.Printf("Starting on [%s]\n", offer.GetHostname()) log.Printf("Starting on [%s]\n", offer.GetHostname())
to_schedule := []*mesos.TaskInfo{s.newTask(offer, task)} taskToSchedule := s.newTask(offer, task)
to_schedule := []*mesos.TaskInfo{taskToSchedule}
driver.LaunchTasks([]*mesos.OfferID{offer.Id}, to_schedule, defaultFilter) driver.LaunchTasks([]*mesos.OfferID{offer.Id}, to_schedule, defaultFilter)
log.Printf("Inst: %d", *task.Instances) log.Printf("Inst: %d", *task.Instances)
s.schedTrace.Print(offer.GetHostname() + ":" + taskToSchedule.GetTaskId().GetValue())
*task.Instances-- *task.Instances--
if *task.Instances <= 0 { if *task.Instances <= 0 {
// All instances of the task have been scheduled. Need to remove it from the list of tasks to schedule. // All instances of the task have been scheduled. Need to remove it from the list of tasks to schedule.
@ -407,28 +410,3 @@ func (s *ProactiveClusterwideCapRanked) StatusUpdate(driver sched.SchedulerDrive
} }
log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value) log.Printf("DONE: Task status [%s] for task [%s]", NameFor(status.State), *status.TaskId.Value)
} }
func (s *ProactiveClusterwideCapRanked) 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 *ProactiveClusterwideCapRanked) OfferRescinded(_ sched.SchedulerDriver, offerID *mesos.OfferID) {
log.Printf("Offer %s rescinded", offerID)
}
func (s *ProactiveClusterwideCapRanked) SlaveLost(_ sched.SchedulerDriver, slaveID *mesos.SlaveID) {
log.Printf("Slave %s lost", slaveID)
}
func (s *ProactiveClusterwideCapRanked) ExecutorLost(_ sched.SchedulerDriver, executorID *mesos.ExecutorID, slaveID *mesos.SlaveID, status int) {
log.Printf("Executor %s on slave %s was lost", executorID, slaveID)
}
func (s *ProactiveClusterwideCapRanked) Error(_ sched.SchedulerDriver, err string) {
log.Printf("Receiving an error: %s", err)
}