mirror of
https://github.com/newnius/YAO-scheduler.git
synced 2025-06-07 14:21:55 +00:00
update
This commit is contained in:
parent
426f9a7765
commit
6a4526a2c9
@ -403,8 +403,7 @@ func (jm *JobManager) stop(force bool) MsgStop {
|
||||
if res.Code != 0 {
|
||||
log.Warn(res.Error)
|
||||
}
|
||||
log.Info(task.HostName, " is killed:", task.Id)
|
||||
|
||||
log.Info(jm.job.Name, ":", task.HostName, " is killed:", task.Id)
|
||||
}(taskStatus)
|
||||
}
|
||||
|
||||
|
@ -20,4 +20,7 @@ type TaskStatus struct {
|
||||
Mem float64 `json:"mem"`
|
||||
BwRX float64 `json:"bw_rx"`
|
||||
BWTx float64 `json:"bw_tx"`
|
||||
UtilGPU int `json:"gpu_util"`
|
||||
UtilMemGPU int `json:"gpu_mem_util"`
|
||||
MemGPU int `json:"gpu_mem"`
|
||||
}
|
||||
|
@ -61,12 +61,16 @@ func (optimizer *Optimizer) feedStats(job Job, role string, stats [][]TaskStatus
|
||||
var Mems []float64
|
||||
var BwRxs []float64
|
||||
var BwTxs []float64
|
||||
var UtilGPUs []float64
|
||||
var MemGPUs []float64
|
||||
for _, stat := range stats {
|
||||
for _, task := range stat {
|
||||
UtilsCPU = append(UtilsCPU, task.UtilCPU)
|
||||
Mems = append(Mems, task.Mem)
|
||||
BwRxs = append(BwRxs, task.BwRX)
|
||||
BwTxs = append(BwTxs, task.BWTx)
|
||||
UtilGPUs = append(UtilGPUs, float64(task.UtilGPU))
|
||||
MemGPUs = append(MemGPUs, float64(task.MemGPU))
|
||||
}
|
||||
}
|
||||
tmp := map[string]float64{
|
||||
@ -76,6 +80,9 @@ func (optimizer *Optimizer) feedStats(job Job, role string, stats [][]TaskStatus
|
||||
"mem": optimizer.max(Mems),
|
||||
"bw_rx": optimizer.mean(BwRxs),
|
||||
"bw_tx": optimizer.mean(BwTxs),
|
||||
"gpu_util": optimizer.mean(UtilGPUs),
|
||||
"gpu_util_std": optimizer.std(UtilGPUs),
|
||||
"gpu_mem": optimizer.max(MemGPUs),
|
||||
}
|
||||
labels, _ := json.Marshal(tmp)
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user