@@ -1105,7 +1105,6 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1105
1105
klog .V (3 ).Infof ("[ScheduleNext] Cannot pop QueueJob from qjqueue! err=%#v" , retryErr )
1106
1106
return err
1107
1107
}
1108
- klog .V (3 ).Infof ("[ScheduleNext] activeQ.Pop_afterPriorityUpdate %s/%s *Delay=%.6f seconds RemainingLength=%d" , qj .Namespace , qj .Name , time .Now ().Sub (qj .Status .ControllerFirstTimestamp .Time ).Seconds (), qjm .qjqueue .Length ())
1109
1108
klog .V (4 ).Infof ("[ScheduleNext] activeQ.Pop_afterPriorityUpdate %s/%s *Delay=%.6f seconds RemainingLength=%d Version=%s Status=%+v" , qj .Namespace , qj .Name , time .Now ().Sub (qj .Status .ControllerFirstTimestamp .Time ).Seconds (), qjm .qjqueue .Length (), qj .ResourceVersion , qj .Status )
1110
1109
apiCacheAWJob , retryErr := qjm .getAppWrapper (qj .Namespace , qj .Name , "[ScheduleNext] -- after dynamic priority pop" )
1111
1110
if retryErr != nil {
@@ -1116,7 +1115,6 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1116
1115
return err
1117
1116
}
1118
1117
if apiCacheAWJob .Status .CanRun {
1119
- klog .V (3 ).Infof ("[ScheduleNext] AppWrapper job: %s/%s from API is already scheduled. Ignoring request" , qj .Namespace , qj .Name )
1120
1118
klog .V (4 ).Infof ("[ScheduleNext] AppWrapper job: %s/%s from API is already scheduled. Ignoring request: Status=%+v" , qj .Namespace , qj .Name , qj .Status )
1121
1119
return nil
1122
1120
}
@@ -1142,10 +1140,8 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1142
1140
1143
1141
klog .V (4 ).Infof ("[ScheduleNext] after Pop qjqLength=%d qj %s/%s Version=%s activeQ=%t Unsched=%t Status=%v" , qjm .qjqueue .Length (), qj .Namespace , qj .Name , qj .ResourceVersion , qjm .qjqueue .IfExistActiveQ (qj ), qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .Status )
1144
1142
if qjm .isDispatcher {
1145
- klog .V (3 ).Infof ("[ScheduleNext] [Dispatcher Mode] Attempting to dispatch next appwrapper: '%s/%s" , qj .Namespace , qj .Name )
1146
1143
klog .V (4 ).Infof ("[ScheduleNext] [Dispatcher Mode] Attempting to dispatch next appwrapper: '%s/%s Status=%v" , qj .Namespace , qj .Name , qj .Status )
1147
1144
} else {
1148
- klog .V (3 ).Infof ("[ScheduleNext] [Agent Mode] Attempting to dispatch next appwrapper: '%s/%s'" , qj .Namespace , qj .Name )
1149
1145
klog .V (4 ).Infof ("[ScheduleNext] [Agent Mode] Attempting to dispatch next appwrapper: '%s/%s' Status=%v" , qj .Namespace , qj .Name , qj .Status )
1150
1146
}
1151
1147
@@ -1283,16 +1279,14 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1283
1279
klog .Info ("%s %s %s" , quotaFits , preemptAWs , msg )
1284
1280
1285
1281
if quotaFits {
1286
- klog .V (3 ).Infof ("[ScheduleNext] [Agent mode] quota evaluation successful for app wrapper '%s/%s'" , qj .Namespace , qj .Name )
1287
1282
klog .V (4 ).Infof ("[ScheduleNext] [Agent mode] quota evaluation successful for app wrapper '%s/%s' activeQ=%t Unsched=%t Version=%s Status=%+v" ,
1288
1283
qj .Namespace , qj .Name , time .Now ().Sub (HOLStartTime ), qjm .qjqueue .IfExistActiveQ (qj ), qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .ResourceVersion , qj .Status )
1289
1284
// Set any jobs that are marked for preemption
1290
1285
qjm .preemptAWJobs (ctx , preemptAWs )
1291
1286
} else { // Not enough free quota to dispatch appwrapper
1292
1287
dispatchFailedMessage = "Insufficient quota and/or resources to dispatch AppWrapper."
1293
1288
dispatchFailedReason = "quota limit exceeded"
1294
- klog .V (3 ).Infof ("[ScheduleNext] [Agent Mode] Blocking dispatch for app wrapper '%s/%s' due to quota limits, msg=%s" , qj .Namespace , qj .Name , msg )
1295
- klog .V (4 ).Infof ("[ScheduleNext] [Agent Mode] Blocking dispatch for app wrapper '%s/%s' due to quota limits, activeQ=%t Unsched=%t Version=%s Status=%+v msg=%s" ,
1289
+ klog .Infof ("[ScheduleNext] [Agent Mode] Blocking dispatch for app wrapper '%s/%s' due to quota limits, activeQ=%t Unsched=%t Version=%s Status=%+v msg=%s" ,
1296
1290
qj .Namespace , qj .Name , time .Now ().Sub (HOLStartTime ), qjm .qjqueue .IfExistActiveQ (qj ), qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .ResourceVersion , qj .Status , msg )
1297
1291
// Call update etcd here to retrigger AW execution for failed quota
1298
1292
// TODO: quota management tests fail if this is converted into go-routine, need to inspect why?
@@ -1311,15 +1305,12 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1311
1305
1312
1306
if aggqj .LessEqual (resources ) { // Check if enough resources to dispatch
1313
1307
fits = true
1314
- klog .V (3 ).Infof ("[ScheduleNext] [Agent Mode] available resource successful check for '%s/%s'" , qj .Namespace , qj .Name )
1315
1308
klog .V (4 ).Infof ("[ScheduleNext] [Agent Mode] available resource successful check for '%s/%s' at %s activeQ=%t Unsched=%t Version=%s Status=%+v." ,
1316
1309
qj .Namespace , qj .Name , time .Now ().Sub (HOLStartTime ), qjm .qjqueue .IfExistActiveQ (qj ), qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .ResourceVersion , qj .Status )
1317
1310
} else { // Not enough free resources to dispatch HOL
1318
1311
fits = false
1319
1312
dispatchFailedMessage = "Insufficient resources to dispatch AppWrapper."
1320
- klog .V (3 ).Infof ("[ScheduleNext] [Agent Mode] Failed to dispatch app wrapper '%s/%s' due to insufficient resources" ,
1321
- qj .Namespace , qj .Name )
1322
- klog .V (4 ).Infof ("[ScheduleNext] [Agent Mode] Failed to dispatch app wrapper '%s/%s' due to insufficient resources, activeQ=%t Unsched=%t Version=%s Status=%+v" ,
1313
+ klog .Infof ("[ScheduleNext] [Agent Mode] Failed to dispatch app wrapper '%s/%s' due to insufficient resources, activeQ=%t Unsched=%t Version=%s Status=%+v" ,
1323
1314
qj .Namespace , qj .Name , qjm .qjqueue .IfExistActiveQ (qj ),
1324
1315
qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .ResourceVersion , qj .Status )
1325
1316
// TODO: Remove forwarded logic as a big AW will never be forwarded
@@ -1383,9 +1374,7 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1383
1374
fowardingLoopCount += 1
1384
1375
}
1385
1376
if ! forwarded { // start thread to backoff
1386
- klog .V (3 ).Infof ("[ScheduleNext] [Agent Mode] backing off app wrapper '%s/%s' after waiting for %s" ,
1387
- qj .Namespace , qj .Name , time .Now ().Sub (HOLStartTime ))
1388
- klog .V (4 ).Infof ("[ScheduleNext] [Agent Mode] backing off app wrapper '%s/%s' after waiting for %s activeQ=%t Unsched=%t Version=%s Status=%+v" ,
1377
+ klog .Infof ("[ScheduleNext] [Agent Mode] backing off app wrapper '%s/%s' after waiting for %s activeQ=%t Unsched=%t Version=%s Status=%+v" ,
1389
1378
qj .Namespace , qj .Name , time .Now ().Sub (HOLStartTime ), qjm .qjqueue .IfExistActiveQ (qj ), qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .ResourceVersion , qj .Status )
1390
1379
if qjm .quotaManager != nil && quotaFits {
1391
1380
qjm .quotaManager .Release (qj )
@@ -1524,8 +1513,6 @@ func (qjm *XController) backoff(ctx context.Context, q *arbv1.AppWrapper, reason
1524
1513
klog .Errorf ("[backoff] Failed to update status for %s/%s. Continuing with possible stale object without updating conditions. err=%s" , q .Namespace , q .Name , err )
1525
1514
}
1526
1515
qjm .qjqueue .AddUnschedulableIfNotPresent (q )
1527
- klog .V (3 ).Infof ("[backoff] %s/%s move to unschedulableQ before sleep for %d seconds. Unsched=%t" , q .Namespace , q .Name ,
1528
- qjm .config .BackoffTimeOrDefault (defaultBackoffTime ), qjm .qjqueue .IfExistUnschedulableQ (q ))
1529
1516
klog .V (4 ).Infof ("[backoff] %s/%s move to unschedulableQ before sleep for %d seconds. activeQ=%t Unsched=%t Version=%s Status=%+v" , q .Namespace , q .Name ,
1530
1517
qjm .config .BackoffTimeOrDefault (defaultBackoffTime ), qjm .qjqueue .IfExistActiveQ (q ), qjm .qjqueue .IfExistUnschedulableQ (q ), q .ResourceVersion , q .Status )
1531
1518
time .Sleep (time .Duration (qjm .config .BackoffTimeOrDefault (defaultBackoffTime )) * time .Second )
@@ -1634,8 +1621,6 @@ func (qjm *XController) UpdateQueueJobs(newjob *arbv1.AppWrapper) {
1634
1621
qjm .eventQueue .Delete (updateQj )
1635
1622
qjm .qjqueue .Delete (updateQj )
1636
1623
}
1637
- klog .V (3 ).Infof ("[UpdateQueueJobs] Done getting completion status for app wrapper '%s/%s' Status.CanRun=%t Status.State=%s" , newjob .Namespace , newjob .Name ,
1638
- newjob .Status .CanRun , newjob .Status .State )
1639
1624
klog .V (4 ).Infof ("[UpdateQueueJobs] Done getting completion status for app wrapper '%s/%s' Version=%s Status.CanRun=%t Status.State=%s, pod counts [Pending: %d, Running: %d, Succeded: %d, Failed %d]" , newjob .Namespace , newjob .Name , newjob .ResourceVersion ,
1640
1625
newjob .Status .CanRun , newjob .Status .State , newjob .Status .Pending , newjob .Status .Running , newjob .Status .Succeeded , newjob .Status .Failed )
1641
1626
}
@@ -1886,7 +1871,7 @@ func (cc *XController) agentEventQueueWorker() {
1886
1871
1887
1872
return nil
1888
1873
}
1889
- klog .V (3 ).Infof ("[Controller: Dispatcher Mode] XQJ Status Update from AGENT: Name:%s, Namespace:%s, Status: %+v\n " , queuejob .Name , queuejob .Namespace , queuejob .Status )
1874
+ klog .V (4 ).Infof ("[Controller: Dispatcher Mode] XQJ Status Update from AGENT: Name:%s, Namespace:%s, Status: %+v\n " , queuejob .Name , queuejob .Namespace , queuejob .Status )
1890
1875
1891
1876
// sync AppWrapper
1892
1877
if err := cc .updateQueueJobStatus (ctx , queuejob ); err != nil {
@@ -2207,7 +2192,6 @@ func (cc *XController) manageQueueJob(ctx context.Context, qj *arbv1.AppWrapper,
2207
2192
}
2208
2193
return nil
2209
2194
} else if qj .Status .CanRun && qj .Status .State == arbv1 .AppWrapperStateActive {
2210
- klog .V (3 ).Infof ("[manageQueueJob] Getting completion status for app wrapper '%s/%s' Status.CanRun=%t Status.State=%s" , qj .Namespace , qj .Name , qj .Status .CanRun , qj .Status .State )
2211
2195
klog .V (4 ).Infof ("[manageQueueJob] Getting completion status for app wrapper '%s/%s' Version=%s Status.CanRun=%t Status.State=%s, pod counts [Pending: %d, Running: %d, Succeded: %d, Failed %d]" , qj .Namespace , qj .Name , qj .ResourceVersion ,
2212
2196
qj .Status .CanRun , qj .Status .State , qj .Status .Pending , qj .Status .Running , qj .Status .Succeeded , qj .Status .Failed )
2213
2197
0 commit comments