github/descheduler
1
0
Fork 0
mirror of https://github.com/kubernetes-sigs/descheduler.git synced 2026-05-10 19:26:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(descheduler): reset prometheus usage client at each extension point

Browse Source 
Profile creation was moved outside the descheduling cycle in b214c147,
but reconcileInClusterSAToken() still runs only in runFnc(), after
newDescheduler() returns. This leaves the prometheus client nil when
LowNodeUtilization's New() runs, causing "prometheus client not
initialized" at startup.

Avoid failing at plugin creation time if the prometheus
client is not yet available. Instead, usageClientForMetrics() is now
called at the start of every extension point via a resetUsageClient()
helper, so each descheduling cycle picks up the latest client regardless
of when the SA token is reconciled or rotated.

Fixes: https://github.com/kubernetes-sigs/descheduler/issues/1840

Signed-off-by: Simone Tiraboschi <stirabos@redhat.com>
This commit is contained in:
Simone Tiraboschi
2026-04-28 18:52:08 +02:00
parent fa8ae489ae
commit e56144c7a2
2 changed files with 105 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
pkg/framework/plugins/nodeutilization/lownodeutilization.go
View File

@@ -103,18 +103,13 @@ func NewLowNodeUtilization(
}
// this plugins supports different ways of collecting usage data. each
// different way provides its own "usageClient". here we make sure we
// have the correct one or an error is triggered. XXX MetricsServer is
// deprecated, removed once dropped.
// different way provides its own "usageClient". the metrics-based client
// is reset at every extension point so we always get the latest prometheus
// client (whose SA token can be rotated after plugin creation).
// XXX MetricsServer is deprecated, removed once dropped.
var usageClient usageClient = newRequestedUsageClient(
extendedResourceNames, handle.GetPodsAssignedToNodeFunc(),
)
if metrics != nil {
usageClient, err = usageClientForMetrics(args, handle, extendedResourceNames)
if err != nil {
return nil, err
}
}
return &LowNodeUtilization{
logger: logger,
@@ -134,12 +129,33 @@ func (l *LowNodeUtilization) Name() string {
return LowNodeUtilizationPluginName
}
// resetUsageClient refreshes the usageClient field from the current handle
// state. It must be called at the start of every extension point so that a
// rotated prometheus SA token is picked up without restarting the process.
func (l *LowNodeUtilization) resetUsageClient() error {
if l.args.MetricsUtilization == nil {
return nil
}
client, err := usageClientForMetrics(l.args, l.handle, l.extendedResourceNames)
if err != nil {
return err
}
l.usageClient = client
return nil
}
// Balance holds the main logic of the plugin. It evicts pods from over
// utilized nodes to under utilized nodes. The goal here is to evenly
// distribute pods across nodes.
func (l *LowNodeUtilization) Balance(ctx context.Context, nodes []*v1.Node) *frameworktypes.Status {
logger := klog.FromContext(klog.NewContext(ctx, l.logger)).WithValues("ExtensionPoint", frameworktypes.BalanceExtensionPoint)
if err := l.resetUsageClient(); err != nil {
return &frameworktypes.Status{
Err: fmt.Errorf("error initializing usage client: %v", err),
}
}
if err := l.usageClient.sync(ctx, nodes); err != nil {
return &frameworktypes.Status{
Err: fmt.Errorf("error getting node usage: %v", err),