Revert "Limit number of Pods counted in a single Job sync"

This reverts commit 7d9cb88fed.

pkg/controller/job/job_controller.go

@@ -880,8 +880,8 @@ func (jm *Controller) removeTrackingFinalizersFromAllPods(pods []*v1.Pod) error
 //    or the job was removed.
 // 3. Increment job counters for pods that no longer have a finalizer.
 // 4. Add Complete condition if satisfied with current counters.
-// It does this up to a limited number of Pods so that the size of .status
-// doesn't grow too much and this sync doesn't starve other Jobs.
+// It does this in a controlled way such that the size of .status doesn't grow
+// too much.
 func (jm *Controller) trackJobStatusAndRemoveFinalizers(job *batch.Job, pods []*v1.Pod, succeededIndexes orderedIntervals, uncounted uncountedTerminatedPods, finishedCond *batch.JobCondition, needsFlush bool) error {
 	isIndexed := isIndexedJob(job)
 	var podsToRemoveFinalizer []*v1.Pod
@@ -936,15 +936,17 @@ func (jm *Controller) trackJobStatusAndRemoveFinalizers(job *batch.Job, pods []*
 			}
 		}
 		if len(newSucceededIndexes)+len(uncountedStatus.Succeeded)+len(uncountedStatus.Failed) >= maxUncountedPods {
-			// The controller added enough Pods already to .status.uncountedTerminatedPods
-			// We stop counting pods and removing finalizers here to:
-			// 1. Ensure that the UIDs representation are under 20 KB.
-			// 2. Cap the number of finalizer removals so that syncing of big Jobs
-			//    doesn't starve smaller ones.
-			//
-			// The job will be synced again because the Job status and Pod updates
-			// will put the Job back to the work queue.
-			break
+			if len(newSucceededIndexes) > 0 {
+				succeededIndexes = succeededIndexes.withOrderedIndexes(newSucceededIndexes)
+				job.Status.Succeeded = int32(succeededIndexes.total())
+				job.Status.CompletedIndexes = succeededIndexes.String()
+			}
+			var err error
+			if needsFlush, err = jm.flushUncountedAndRemoveFinalizers(job, podsToRemoveFinalizer, uidsWithFinalizer, needsFlush); err != nil {
+				return err
+			}
+			podsToRemoveFinalizer = nil
+			newSucceededIndexes = nil
 		}
 	}
 	if len(newSucceededIndexes) > 0 {

pkg/controller/job/job_controller_test.go

@@ -1463,7 +1463,7 @@ func TestTrackJobStatusAndRemoveFinalizers(t *testing.T) {
 				pods = append(pods, buildPod().uid("b").phase(v1.PodFailed).trackingFinalizer().Pod)
 				return pods
 			}(),
-			wantRmFinalizers: 499,
+			wantRmFinalizers: 501,
 			wantStatusUpdates: []batch.JobStatus{
 				{
 					UncountedTerminatedPods: &batch.UncountedTerminatedPods{
@@ -1480,11 +1480,17 @@ func TestTrackJobStatusAndRemoveFinalizers(t *testing.T) {
 				},
 				{
 					UncountedTerminatedPods: &batch.UncountedTerminatedPods{
-						Failed: []types.UID{"b"},
+						Succeeded: []types.UID{"499"},
+						Failed:    []types.UID{"b"},
 					},
 					Succeeded: 499,
 					Failed:    1,
 				},
+				{
+					UncountedTerminatedPods: &batch.UncountedTerminatedPods{},
+					Succeeded:               500,
+					Failed:                  2,
+				},
 			},
 		},
 		"too many indexed finished": {
@@ -1501,13 +1507,18 @@ func TestTrackJobStatusAndRemoveFinalizers(t *testing.T) {
 				}
 				return pods
 			}(),
-			wantRmFinalizers: 500,
+			wantRmFinalizers: 501,
 			wantStatusUpdates: []batch.JobStatus{
 				{
 					UncountedTerminatedPods: &batch.UncountedTerminatedPods{},
 					CompletedIndexes:        "0-499",
 					Succeeded:               500,
 				},
+				{
+					CompletedIndexes:        "0-500",
+					UncountedTerminatedPods: &batch.UncountedTerminatedPods{},
+					Succeeded:               501,
+				},
 			},
 		},
 	}