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
2020-05-27 14:15:43 +08:00
parent ab0c1882cc
commit 5d8693eb5b
3 changed files with 7 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/evaluator.go
View File

@@ -66,11 +66,6 @@ func (eva *Evaluator) add(node NodeStatus, task Task) {
eva.totalWorker++ eva.totalWorker++
} }
if task.IsPS {
//eva.costLoad += 1
} else {
//eva.costLoad += 0.5
}
numberGPU := 1 numberGPU := 1
for _, gpu := range node.Status { for _, gpu := range node.Status {
if gpu.MemoryAllocated != 0 { if gpu.MemoryAllocated != 0 {
@@ -104,11 +99,6 @@ func (eva *Evaluator) remove(node NodeStatus, task Task) {
eva.totalWorker-- eva.totalWorker--
} }
if task.IsPS {
//eva.costLoad -= 1
} else {
//eva.costLoad -= 0.5
}
numberGPU := 1 numberGPU := 1
for _, gpu := range node.Status { for _, gpu := range node.Status {
if gpu.MemoryAllocated != 0 { if gpu.MemoryAllocated != 0 {

2
src/job_manager.go
View File

@@ -80,7 +80,7 @@ func (jm *JobManager) start() {
v.Set("mem_limit", strconv.Itoa(jm.job.Tasks[index].Memory)+"m") v.Set("mem_limit", strconv.Itoa(jm.job.Tasks[index].Memory)+"m")
v.Set("cpu_limit", strconv.Itoa(jm.job.Tasks[index].NumberCPU)) v.Set("cpu_limit", strconv.Itoa(jm.job.Tasks[index].NumberCPU))
v.Set("network", jm.network) v.Set("network", jm.network)
v.Set("should_wait", "1") v.Set("should_wait", "0")
v.Set("output_dir", "/tmp/") v.Set("output_dir", "/tmp/")
v.Set("hdfs_address", "http://192.168.100.104:50070/") v.Set("hdfs_address", "http://192.168.100.104:50070/")
v.Set("hdfs_dir", "/user/yao/output/"+jm.job.Name) v.Set("hdfs_dir", "/user/yao/output/"+jm.job.Name)

7
src/resource_pool.go
View File

@@ -896,7 +896,12 @@ func (pool *ResourcePool) releaseResource(job Job, agent NodeStatus) {
seg.Lock.Lock() seg.Lock.Lock()
defer seg.Lock.Unlock() defer seg.Lock.Unlock()
node := seg.Nodes[agent.ClientID] node, ok := seg.Nodes[agent.ClientID]
if !ok {
/* in case node is offline */
/* TODO, update usingTotalGPU correctly */
return
}
for _, gpu := range agent.Status { for _, gpu := range agent.Status {
for j := range node.Status { for j := range node.Status {
if gpu.UUID == node.Status[j].UUID { if gpu.UUID == node.Status[j].UUID {