@@ -1058,7 +1058,6 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1058
1058
klog .V (3 ).Infof ("[ScheduleNext] Cannot pop QueueJob from qjqueue! err=%#v" , retryErr )
1059
1059
return err
1060
1060
}
1061
- 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 ())
1062
1061
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 )
1063
1062
apiCacheAWJob , retryErr := qjm .getAppWrapper (qj .Namespace , qj .Name , "[ScheduleNext] -- after dynamic priority pop" )
1064
1063
if retryErr != nil {
@@ -1069,7 +1068,6 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1069
1068
return err
1070
1069
}
1071
1070
if apiCacheAWJob .Status .CanRun {
1072
- klog .V (3 ).Infof ("[ScheduleNext] AppWrapper job: %s/%s from API is already scheduled. Ignoring request" , qj .Namespace , qj .Name )
1073
1071
klog .V (4 ).Infof ("[ScheduleNext] AppWrapper job: %s/%s from API is already scheduled. Ignoring request: Status=%+v" , qj .Namespace , qj .Name , qj .Status )
1074
1072
return nil
1075
1073
}
@@ -1095,10 +1093,8 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1095
1093
1096
1094
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 )
1097
1095
if qjm .isDispatcher {
1098
- klog .V (3 ).Infof ("[ScheduleNext] [Dispatcher Mode] Attempting to dispatch next appwrapper: '%s/%s" , qj .Namespace , qj .Name )
1099
1096
klog .V (4 ).Infof ("[ScheduleNext] [Dispatcher Mode] Attempting to dispatch next appwrapper: '%s/%s Status=%v" , qj .Namespace , qj .Name , qj .Status )
1100
1097
} else {
1101
- klog .V (3 ).Infof ("[ScheduleNext] [Agent Mode] Attempting to dispatch next appwrapper: '%s/%s'" , qj .Namespace , qj .Name )
1102
1098
klog .V (4 ).Infof ("[ScheduleNext] [Agent Mode] Attempting to dispatch next appwrapper: '%s/%s' Status=%v" , qj .Namespace , qj .Name , qj .Status )
1103
1099
}
1104
1100
@@ -1236,16 +1232,14 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1236
1232
klog .Info ("%s %s %s" , quotaFits , preemptAWs , msg )
1237
1233
1238
1234
if quotaFits {
1239
- klog .V (3 ).Infof ("[ScheduleNext] [Agent mode] quota evaluation successful for app wrapper '%s/%s'" , qj .Namespace , qj .Name )
1240
1235
klog .V (4 ).Infof ("[ScheduleNext] [Agent mode] quota evaluation successful for app wrapper '%s/%s' activeQ=%t Unsched=%t Version=%s Status=%+v" ,
1241
1236
qj .Namespace , qj .Name , time .Now ().Sub (HOLStartTime ), qjm .qjqueue .IfExistActiveQ (qj ), qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .ResourceVersion , qj .Status )
1242
1237
// Set any jobs that are marked for preemption
1243
1238
qjm .preemptAWJobs (ctx , preemptAWs )
1244
1239
} else { // Not enough free quota to dispatch appwrapper
1245
1240
dispatchFailedMessage = "Insufficient quota and/or resources to dispatch AppWrapper."
1246
1241
dispatchFailedReason = "quota limit exceeded"
1247
- 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 )
1248
- 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" ,
1242
+ 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" ,
1249
1243
qj .Namespace , qj .Name , time .Now ().Sub (HOLStartTime ), qjm .qjqueue .IfExistActiveQ (qj ), qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .ResourceVersion , qj .Status , msg )
1250
1244
// Call update etcd here to retrigger AW execution for failed quota
1251
1245
// TODO: quota management tests fail if this is converted into go-routine, need to inspect why?
@@ -1264,15 +1258,12 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1264
1258
1265
1259
if aggqj .LessEqual (resources ) { // Check if enough resources to dispatch
1266
1260
fits = true
1267
- klog .V (3 ).Infof ("[ScheduleNext] [Agent Mode] available resource successful check for '%s/%s'" , qj .Namespace , qj .Name )
1268
1261
klog .V (4 ).Infof ("[ScheduleNext] [Agent Mode] available resource successful check for '%s/%s' at %s activeQ=%t Unsched=%t Version=%s Status=%+v." ,
1269
1262
qj .Namespace , qj .Name , time .Now ().Sub (HOLStartTime ), qjm .qjqueue .IfExistActiveQ (qj ), qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .ResourceVersion , qj .Status )
1270
1263
} else { // Not enough free resources to dispatch HOL
1271
1264
fits = false
1272
1265
dispatchFailedMessage = "Insufficient resources to dispatch AppWrapper."
1273
- klog .V (3 ).Infof ("[ScheduleNext] [Agent Mode] Failed to dispatch app wrapper '%s/%s' due to insufficient resources" ,
1274
- qj .Namespace , qj .Name )
1275
- 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" ,
1266
+ klog .Infof ("[ScheduleNext] [Agent Mode] Failed to dispatch app wrapper '%s/%s' due to insufficient resources, activeQ=%t Unsched=%t Version=%s Status=%+v" ,
1276
1267
qj .Namespace , qj .Name , qjm .qjqueue .IfExistActiveQ (qj ),
1277
1268
qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .ResourceVersion , qj .Status )
1278
1269
// TODO: Remove forwarded logic as a big AW will never be forwarded
@@ -1336,9 +1327,7 @@ func (qjm *XController) ScheduleNext(qj *arbv1.AppWrapper) {
1336
1327
fowardingLoopCount += 1
1337
1328
}
1338
1329
if ! forwarded { // start thread to backoff
1339
- klog .V (3 ).Infof ("[ScheduleNext] [Agent Mode] backing off app wrapper '%s/%s' after waiting for %s" ,
1340
- qj .Namespace , qj .Name , time .Now ().Sub (HOLStartTime ))
1341
- 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" ,
1330
+ klog .Infof ("[ScheduleNext] [Agent Mode] backing off app wrapper '%s/%s' after waiting for %s activeQ=%t Unsched=%t Version=%s Status=%+v" ,
1342
1331
qj .Namespace , qj .Name , time .Now ().Sub (HOLStartTime ), qjm .qjqueue .IfExistActiveQ (qj ), qjm .qjqueue .IfExistUnschedulableQ (qj ), qj .ResourceVersion , qj .Status )
1343
1332
if qjm .quotaManager != nil && quotaFits {
1344
1333
qjm .quotaManager .Release (qj )
@@ -1477,8 +1466,6 @@ func (qjm *XController) backoff(ctx context.Context, q *arbv1.AppWrapper, reason
1477
1466
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 )
1478
1467
}
1479
1468
qjm .qjqueue .AddUnschedulableIfNotPresent (q )
1480
- klog .V (3 ).Infof ("[backoff] %s/%s move to unschedulableQ before sleep for %d seconds. Unsched=%t" , q .Namespace , q .Name ,
1481
- qjm .config .BackoffTimeOrDefault (defaultBackoffTime ), qjm .qjqueue .IfExistUnschedulableQ (q ))
1482
1469
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 ,
1483
1470
qjm .config .BackoffTimeOrDefault (defaultBackoffTime ), qjm .qjqueue .IfExistActiveQ (q ), qjm .qjqueue .IfExistUnschedulableQ (q ), q .ResourceVersion , q .Status )
1484
1471
time .Sleep (time .Duration (qjm .config .BackoffTimeOrDefault (defaultBackoffTime )) * time .Second )
@@ -1587,8 +1574,6 @@ func (qjm *XController) UpdateQueueJobs(newjob *arbv1.AppWrapper) {
1587
1574
qjm .eventQueue .Delete (updateQj )
1588
1575
qjm .qjqueue .Delete (updateQj )
1589
1576
}
1590
- klog .V (3 ).Infof ("[UpdateQueueJobs] Done getting completion status for app wrapper '%s/%s' Status.CanRun=%t Status.State=%s" , newjob .Namespace , newjob .Name ,
1591
- newjob .Status .CanRun , newjob .Status .State )
1592
1577
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 ,
1593
1578
newjob .Status .CanRun , newjob .Status .State , newjob .Status .Pending , newjob .Status .Running , newjob .Status .Succeeded , newjob .Status .Failed )
1594
1579
}
@@ -1839,7 +1824,7 @@ func (cc *XController) agentEventQueueWorker() {
1839
1824
1840
1825
return nil
1841
1826
}
1842
- 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 )
1827
+ 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 )
1843
1828
1844
1829
// sync AppWrapper
1845
1830
if err := cc .updateQueueJobStatus (ctx , queuejob ); err != nil {
@@ -2160,7 +2145,6 @@ func (cc *XController) manageQueueJob(ctx context.Context, qj *arbv1.AppWrapper,
2160
2145
}
2161
2146
return nil
2162
2147
} else if qj .Status .CanRun && qj .Status .State == arbv1 .AppWrapperStateActive {
2163
- 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 )
2164
2148
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 ,
2165
2149
qj .Status .CanRun , qj .Status .State , qj .Status .Pending , qj .Status .Running , qj .Status .Succeeded , qj .Status .Failed )
2166
2150
0 commit comments