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

minify logs

Browse Source
This commit is contained in:
Newnius
2020-04-13 19:41:28 +08:00
parent 3f215b0e1d
commit 03b1929383
3 changed files with 13 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/job_manager.go
View File

@@ -19,7 +19,7 @@ type JobManager struct {
} }
func (jm *JobManager) start() { func (jm *JobManager) start() {
log.Info("start job ", jm.job.Name, time.Now()) log.Debug("start job ", jm.job.Name, time.Now())
jm.jobStatus = JobStatus{Name: jm.job.Name, tasks: map[string]TaskStatus{}} jm.jobStatus = JobStatus{Name: jm.job.Name, tasks: map[string]TaskStatus{}}
network := jm.scheduler.AcquireNetwork() network := jm.scheduler.AcquireNetwork()
@@ -39,7 +39,7 @@ func (jm *JobManager) start() {
} }
time.Sleep(time.Second * 1) time.Sleep(time.Second * 1)
} }
log.Info("Receive resource", resource) log.Debug("Receive resource", resource)
jm.resources = append(jm.resources, resource) jm.resources = append(jm.resources, resource)
for _, t := range resource.Status { for _, t := range resource.Status {
@@ -49,7 +49,7 @@ func (jm *JobManager) start() {
} }
jm.scheduler.UpdateProgress(jm.job.Name, Running) jm.scheduler.UpdateProgress(jm.job.Name, Running)
log.Info("ready to run job ", jm.job.Name, time.Now()) log.Debug("ready to run job ", jm.job.Name, time.Now())
/* bring up containers */ /* bring up containers */
for i := range jm.job.Tasks { for i := range jm.job.Tasks {
@@ -123,7 +123,7 @@ func (jm *JobManager) start() {
/* return resource */ /* return resource */
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.Debug("return resource ", jm.resources[i].ClientID)
for _, t := range jm.resources[i].Status { for _, t := range jm.resources[i].Status {
jm.scheduler.Detach(t.UUID, jm.job.Name) jm.scheduler.Detach(t.UUID, jm.job.Name)

2
src/resource_pool.go
View File

@@ -174,7 +174,7 @@ func (pool *ResourcePool) acquireNetwork() string {
pool.networkMu.Lock() pool.networkMu.Lock()
defer pool.networkMu.Unlock() defer pool.networkMu.Unlock()
var network string var network string
log.Info(pool.networksFree) log.Debug(pool.networksFree)
if len(pool.networksFree) == 0 { if len(pool.networksFree) == 0 {
for { for {
for { for {

14
src/scheduler_fair.go
View File

@@ -71,6 +71,7 @@ func (scheduler *SchedulerFair) Start() {
jm.start() jm.start()
}() }()
} else { } else {
log.Info("No more jobs to schedule", time.Now())
scheduler.scheduling.Unlock() scheduler.scheduling.Unlock()
go func() { go func() {
scheduler.UpdateNextQueue() scheduler.UpdateNextQueue()
@@ -206,14 +207,15 @@ func (scheduler *SchedulerFair) AcquireResource(job Job, task Task) NodeStatus {
func (scheduler *SchedulerFair) ReleaseResource(job Job, agent NodeStatus) { func (scheduler *SchedulerFair) ReleaseResource(job Job, agent NodeStatus) {
pool.mu.Lock() pool.mu.Lock()
defer pool.mu.Unlock() defer pool.mu.Unlock()
nodes := pool.nodes[agent.ClientID] node := pool.nodes[agent.ClientID]
for _, gpu := range agent.Status { for _, gpu := range agent.Status {
for j := range nodes.Status { for j := range node.Status {
if gpu.UUID == nodes.Status[j].UUID { if gpu.UUID == node.Status[j].UUID {
nodes.Status[j].MemoryAllocated -= gpu.MemoryTotal node.Status[j].MemoryAllocated -= gpu.MemoryTotal
if nodes.Status[j].MemoryAllocated < 0 { if node.Status[j].MemoryAllocated < 0 {
// in case of error // in case of error
nodes.Status[j].MemoryAllocated = 0 log.Warn(node.ClientID, "More Memory Allocated")
node.Status[j].MemoryAllocated = 0
} }
} }
} }