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 14:39:10 +08:00
parent 38ca417a2d
commit b24f623966
1 changed files with 7 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
src/scheduler_fair.go
View File

@@ -188,7 +188,6 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
var candidates []NodeStatus var candidates []NodeStatus
/* first round, find vacant gpu */ /* first round, find vacant gpu */
for i := 0; i < pool.poolsCount; i++ { for i := 0; i < pool.poolsCount; i++ {
log.Info("lock,", (i+poolID)%pool.poolsCount)
pool.poolsMu[(i+poolID)%pool.poolsCount].Lock() pool.poolsMu[(i+poolID)%pool.poolsCount].Lock()
locks[(i+poolID)%pool.poolsCount] = pool.poolsMu[(i+poolID)%pool.poolsCount] locks[(i+poolID)%pool.poolsCount] = pool.poolsMu[(i+poolID)%pool.poolsCount]
for _, node := range pool.pools[(i+poolID)%pool.poolsCount] { for _, node := range pool.pools[(i+poolID)%pool.poolsCount] {
@@ -199,13 +198,7 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
} }
} }
if len(available) >= task.NumberGPU { if len(available) >= task.NumberGPU {
tmp := NodeStatus{} candidates = append(candidates, node)
tmp.ClientID = node.ClientID
tmp.ClientHost = node.ClientHost
tmp.Status = available
tmp.NumCPU = node.NumCPU
tmp.MemTotal = node.MemAvailable
candidates = append(candidates, tmp)
if len(candidates) >= 8 { if len(candidates) >= 8 {
break break
} }
@@ -223,10 +216,10 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
log.Info("dasdsa") log.Info("dasdsa")
if util, valid := InstanceOfOptimizer().predictUtilGPU(job.Name); valid { if util, valid := InstanceOfOptimizer().predictUtilGPU(job.Name); valid {
for i := poolID; i < pool.poolsCount; i++ { for i := 0; i < pool.poolsCount; i++ {
if _, err := locks[i]; err { if _, err := locks[(i+poolID)%pool.poolsCount]; err {
pool.poolsMu[i].Lock() pool.poolsMu[(i+poolID)%pool.poolsCount].Lock()
locks[i] = pool.poolsMu[i] locks[(i+poolID)%pool.poolsCount] = pool.poolsMu[(i+poolID)%pool.poolsCount]
} }
for _, node := range pool.pools[i] { for _, node := range pool.pools[i] {
@@ -248,13 +241,7 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
} }
} }
if len(available) >= task.NumberGPU { if len(available) >= task.NumberGPU {
tmp := NodeStatus{} candidates = append(candidates, node)
tmp.ClientID = node.ClientID
tmp.ClientHost = node.ClientHost
tmp.Status = available
tmp.NumCPU = node.NumCPU
tmp.MemTotal = node.MemAvailable
candidates = append(candidates, tmp)
if len(candidates) >= 8 { if len(candidates) >= 8 {
break break
} }
@@ -287,9 +274,7 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
} }
} }
for i, _ := range locks { for i := range locks {
log.Info("unlock ", i)
//lock.Unlock()
pool.poolsMu[i].Unlock() pool.poolsMu[i].Unlock()
} }
go func(res NodeStatus) { go func(res NodeStatus) {