1
0
mirror of https://github.com/newnius/YAO-scheduler.git synced 2025-06-06 22:01:55 +00:00

update fair

This commit is contained in:
Newnius 2020-05-28 11:23:27 +08:00
parent d16df29b50
commit c044d43490
3 changed files with 10 additions and 7 deletions

View File

@ -197,17 +197,20 @@ func (jm *JobManager) checkStatus(status []TaskStatus) {
jm.scheduler.UpdateProgress(jm.job, Failed)
}
jm.resourcesMu.Lock()
if jm.resources[i].ClientID != "_released_" {
jm.scheduler.ReleaseResource(jm.job, jm.resources[i])
log.Info("return resource ", jm.resources[i].ClientID)
jm.resources[i].ClientID = "_released_"
}
for _, t := range jm.resources[i].Status {
InstanceOfResourcePool().detach(t.UUID, jm.job)
}
for _, t := range jm.resources[i].Status {
InstanceOfResourcePool().detach(t.UUID, jm.job)
}
InstanceJobHistoryLogger().submitTaskStatus(jm.job.Name, status[i])
jm.resourcesMu.Unlock()
InstanceJobHistoryLogger().submitTaskStatus(jm.job.Name, status[i])
}
}
}
if flagRunning && onlyPS && !jm.killFlag {

View File

@ -908,7 +908,7 @@ func (pool *ResourcePool) releaseResource(job Job, agent NodeStatus) {
node.Status[j].MemoryAllocated -= gpu.MemoryTotal
if node.Status[j].MemoryAllocated < 0 {
// in case of error
log.Warn(node.ClientID, "More Memory Allocated")
log.Warn(node.ClientID, " More Memory Allocated")
node.Status[j].MemoryAllocated = 0
}
if node.Status[j].MemoryAllocated == 0 {

View File

@ -320,7 +320,7 @@ func (scheduler *SchedulerFair) UpdateQuota() {
pool := InstanceOfResourcePool()
available := pool.TotalGPU - usingGPU - allocatedGPU
if available == 0 {
if available <= 0 {
return
}
log.Info("Can allocate ", available)