//
	// We also don't care if the GC starts again after this for the same reasons.
	releasem(mp)
	semrelease(&worldsema)

	// Snapshot allm and freem.
	//
	// Snapshotting after the generation counter update is sufficient.
	// Because an m must be on either allm or sched.freem if it has an active trace
	// buffer, new threads added to allm after this point must necessarily observe

[[sub:java_library_known_issues_windows_performance]]
=== Significant build performance drop on Windows for huge multi-projects

Another side effect of the snapshotting of individual class files, only affecting Windows systems, is that the performance can significantly drop when processing a very large amount of class files on the compile classpath.

// Should be a small cidr due to how it is stored in etcd.
// bigger cidr (specially those offered by IPv6) will add no value
// and significantly increase snapshotting time.
// NOTE: This is identical to validation performed in the apiserver.
func ValidateServiceSubnetSize(subnetStr string, fldPath *field.Path) field.ErrorList {
	allErrs := field.ErrorList{}

	defer trace.LogIfLong(100 * time.Millisecond)
	if err := sched.Cache.UpdateSnapshot(klog.FromContext(ctx), sched.nodeInfoSnapshot); err != nil {
		return result, err
	}
	trace.Step("Snapshotting scheduler cache and node infos done")

	if sched.nodeInfoSnapshot.NumNodes() == 0 {
		return result, ErrNoNodesAvailable
	}

	feasibleNodes, diagnosis, err := sched.findNodesThatFitPod(ctx, fwk, state, pod)

For the 1.8 release, SIG Storage extended the Kubernetes storage API. In addition to providing simple
volume availability, the API now enables volume resizing and snapshotting. These features are in alpha.
The SIG also focused on providing more control over storage: the ability to set requests and