YAO/YAO-scheduler
1
0
Fork 0
mirror of https://github.com/newnius/YAO-scheduler.git synced 2025-12-15 08:16:43 +00:00
Code Issues Releases Wiki Activity

update

Browse Source
This commit is contained in:
Newnius
2020-04-30 21:22:21 +08:00
parent 6632a4f199
commit 014592fa43
6 changed files with 47 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/job_manager.go
View File

@@ -33,7 +33,7 @@ func (jm *JobManager) start() {
if jm.killedFlag {
break
}
resource = jm.scheduler.AcquireResource(jm.job, jm.job.Tasks[i])
resource = jm.scheduler.AcquireResource(jm.job, jm.job.Tasks[i], jm.resources)
if len(resource.Status) > 0 {
break
}

20
src/resource_pool.go
View File

@@ -40,7 +40,11 @@ type ResourcePool struct {
}
func (pool *ResourcePool) GPUModelToPower(model string) int {
mapper := map[string]int{"k40": 1, "K80": 2, "P100": 3}
mapper := map[string]int{
"K40": 1, "Tesla K40": 1,
"K80": 2, "Tesla K80": 2,
"P100": 3, "Tesla P100": 3,
}
if power, err := mapper[model]; !err {
return power
}
@@ -297,3 +301,17 @@ func (pool *ResourcePool) detach(GPU string, jobName string) {
func (pool *ResourcePool) getBindings() map[string]map[string]int {
return pool.bindings
}
func (pool *ResourcePool) pickNode(nodes []*NodeStatus) *NodeStatus {
/* shuffle */
r := rand.New(rand.NewSource(time.Now().Unix()))
for n := len(nodes); n > 0; n-- {
randIndex := r.Intn(n)
nodes[n-1], nodes[randIndex] = nodes[randIndex], nodes[n-1]
}
/* sort */
return nodes[0]
}

2
src/scheduler.go
View File

@@ -7,7 +7,7 @@ type Scheduler interface {
UpdateProgress(jobName string, state State)
AcquireResource(Job, Task) NodeStatus
AcquireResource(Job, Task, []NodeStatus) NodeStatus
ReleaseResource(Job, NodeStatus)

2
src/scheduler_FCFS.go
View File

@@ -87,7 +87,7 @@ func (scheduler *SchedulerFCFS) Schedule(job Job) {
job.Status = Created
}
func (scheduler *SchedulerFCFS) AcquireResource(job Job, task Task) NodeStatus {
func (scheduler *SchedulerFCFS) AcquireResource(job Job, task Task, nodes []NodeStatus) NodeStatus {
poolID := rand.Intn(pool.poolsCount)
pool.poolsMu[poolID].Lock()
defer pool.poolsMu[poolID].Unlock()

22
src/scheduler_fair.go
View File

@@ -35,6 +35,7 @@ type SchedulerFair struct {
enablePreScheduleRatio float64
UsingGPU int
UsingGPUMu sync.Mutex
}
type FairJobSorter []Job
@@ -193,7 +194,7 @@ func (scheduler *SchedulerFair) Schedule(job Job) {
job.Status = Created
}
func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
func (scheduler *SchedulerFair) AcquireResource(job Job, task Task, nodes []NodeStatus) NodeStatus {
poolID := rand.Intn(pool.poolsCount)
res := NodeStatus{}
@@ -202,7 +203,7 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
allocationType := 0
availableGPUs := map[string][]GPUStatus{}
var candidates []NodeStatus
var candidates []*NodeStatus
/* first, choose sharable GPUs */
if scheduler.enableShare && (pool.TotalGPU != 0 && float64(scheduler.UsingGPU)/float64(pool.TotalGPU) >= scheduler.enableShareRatio) {
@@ -236,7 +237,7 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
}
}
if len(available) >= task.NumberGPU {
candidates = append(candidates, node)
candidates = append(candidates, &node)
if len(candidates) >= 8 {
break
}
@@ -266,7 +267,7 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
}
}
if len(available) >= task.NumberGPU {
candidates = append(candidates, node)
candidates = append(candidates, &node)
availableGPUs[node.ClientID] = available
if len(candidates) >= 8 {
break
@@ -316,7 +317,7 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
}
}
if len(available) >= task.NumberGPU {
candidates = append(candidates, node)
candidates = append(candidates, &node)
availableGPUs[node.ClientID] = available
if len(candidates) >= 8 {
break
@@ -338,7 +339,7 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
/* assign */
if len(candidates) > 0 {
node := candidates[0]
node := pool.pickNode(candidates)
res.ClientID = node.ClientID
res.ClientHost = node.ClientHost
res.Status = availableGPUs[node.ClientID][0:task.NumberGPU]
@@ -354,10 +355,16 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
}
}
if allocationType == 2 {
scheduler.UsingGPUMu.Lock()
scheduler.UsingGPU += task.NumberGPU
scheduler.UsingGPUMu.Unlock()
log.Info(res.Status, " is using")
}
}
for i := range locks {
pool.poolsMu[i].Unlock()
}
@@ -399,7 +406,10 @@ func (scheduler *SchedulerFair) ReleaseResource(job Job, agent NodeStatus) {
node.Status[j].MemoryAllocated = 0
}
if node.Status[j].MemoryAllocated == 0 {
scheduler.UsingGPUMu.Lock()
scheduler.UsingGPU--
scheduler.UsingGPUMu.Unlock()
log.Info(node.Status[j].UUID, " is released")
}
log.Info(node.Status[j].MemoryAllocated)
}

2
src/scheduler_priority.go
View File

@@ -111,7 +111,7 @@ func (scheduler *SchedulerPriority) Schedule(job Job) {
job.Status = Created
}
func (scheduler *SchedulerPriority) AcquireResource(job Job, task Task) NodeStatus {
func (scheduler *SchedulerPriority) AcquireResource(job Job, task Task, nodes []NodeStatus) NodeStatus {
poolID := rand.Intn(pool.poolsCount)
pool.poolsMu[poolID].Lock()
defer pool.poolsMu[poolID].Unlock()