YAO/YAO-scheduler
1
0
Fork 0
mirror of https://github.com/newnius/YAO-scheduler.git synced 2025-06-07 22:31:55 +00:00
Code Issues Releases Wiki Activity

update fair

Browse Source
This commit is contained in:
Newnius 2020-05-31 18:55:07 +08:00
parent b99e66a4ca
commit 41242024b6
2 changed files with 14 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/job_manager.go
View File

@ -189,7 +189,9 @@ func (jm *JobManager) checkStatus(status []TaskStatus) {
} }
//InstanceJobHistoryLogger().submitTaskStatus(jm.job.Name, status[i]) //InstanceJobHistoryLogger().submitTaskStatus(jm.job.Name, status[i])
} else { } else {
jm.resourcesMu.Lock()
if jm.resources[i].ClientID == "_released_" { if jm.resources[i].ClientID == "_released_" {
jm.resourcesMu.Unlock()
continue continue
} }
log.Info(jm.job.Name, "-", i, " ", status[i].Status) log.Info(jm.job.Name, "-", i, " ", status[i].Status)
@ -200,7 +202,6 @@ func (jm *JobManager) checkStatus(status []TaskStatus) {
jm.scheduler.UpdateProgress(jm.job, Failed) jm.scheduler.UpdateProgress(jm.job, Failed)
} }
jm.resourcesMu.Lock()
if jm.resources[i].ClientID != "_released_" { if jm.resources[i].ClientID != "_released_" {
jm.scheduler.ReleaseResource(jm.job, jm.resources[i]) jm.scheduler.ReleaseResource(jm.job, jm.resources[i])
log.Info("return resource ", jm.resources[i].ClientID) log.Info("return resource ", jm.resources[i].ClientID)

14
src/scheduler_fair.go
View File

@ -128,6 +128,16 @@ func (scheduler *SchedulerFair) Start() {
/* phase 2: borrow */ /* phase 2: borrow */
if bestQueue == "" && scheduler.enableBorrow { if bestQueue == "" && scheduler.enableBorrow {
quotas := map[string]ResourceCount{}
for queue, quota := range scheduler.queuesQuota {
quotas[queue] = ResourceCount{NumberGPU: quota.NumberGPU}
}
for _, IOUs := range scheduler.IOUs {
for queue, IOU := range IOUs {
quota := quotas[queue]
quota.NumberGPU += IOU.NumberGPU
}
}
/* firstly, check if quota sum can run a job */ /* firstly, check if quota sum can run a job */
totalGPU := 0 totalGPU := 0
for _, quota := range scheduler.queuesQuota { for _, quota := range scheduler.queuesQuota {
@ -151,7 +161,7 @@ func (scheduler *SchedulerFair) Start() {
continue continue
} }
needGPU := numberGPUtmp*1000 - scheduler.queuesQuota[queue].NumberGPU needGPU := numberGPUtmp*1000 - quotas[queue].NumberGPU
/* the less, the better */ /* the less, the better */
if bestQueue == "" || needGPU < minRequestGPU { if bestQueue == "" || needGPU < minRequestGPU {
bestQueue = queue bestQueue = queue
@ -167,7 +177,7 @@ func (scheduler *SchedulerFair) Start() {
log.Info(bestQueue, ": ", "total=", totalGPU, " still need ", minRequestGPU) log.Info(bestQueue, ": ", "total=", totalGPU, " still need ", minRequestGPU)
for { for {
/* if all satisfied, break */ /* if all satisfied, break */
if minRequestGPU == 0 { if minRequestGPU <= 0 {
break break
} }
least := math.MaxInt32 least := math.MaxInt32