This is an automated email from the ASF dual-hosted git repository.
wilfred-s pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/yunikorn-k8shim.git
The following commit(s) were added to refs/heads/master by this push:
new 0baa5ea3 [YUNIKORN-3371] DRA resource-slice tracker goroutine leaks in
tests (#1087)
0baa5ea3 is described below
commit 0baa5ea33387f31413dfb9b4e81ca8b79276f831
Author: PoiBlackTea <[email protected]>
AuthorDate: Thu Sep 10 21:02:29 2026 +1000
[YUNIKORN-3371] DRA resource-slice tracker goroutine leaks in tests (#1087)
- Save resourceSliceTracker in Context and add Context.Stop() to shut down
the tracker
- Call Context.Stop() in KubernetesShim.Stop()
- Disable EnableDeviceTaintRules in support.SharedDRAManager() to prevent
goroutine leaks in mock tests
- Add TestContextStop unit test
Signed-off-by: PoiBlackTea <[email protected]>
Closes: #1087
Signed-off-by: Wilfred Spiegelenburg <[email protected]>
---
pkg/cache/context.go | 32 +++++++++++++++++++++++---------
pkg/cache/context_test.go | 8 ++++++++
pkg/plugin/support/mock_helpers.go | 7 ++++++-
pkg/shim/scheduler.go | 4 ++++
4 files changed, 41 insertions(+), 10 deletions(-)
diff --git a/pkg/cache/context.go b/pkg/cache/context.go
index e0878250..6e7e2fec 100644
--- a/pkg/cache/context.go
+++ b/pkg/cache/context.go
@@ -72,15 +72,16 @@ var (
// context maintains scheduling state, like apps and apps' tasks.
type Context struct {
- applications map[string]*Application // apps
- schedulerCache *schedulercache.SchedulerCache // external cache
- apiProvider client.APIProvider // apis to interact with
api-server, scheduler-core, etc
- predManager predicates.PredicateManager // K8s predicates
- namespace string // yunikorn namespace
- configMaps []*v1.ConfigMap // cached yunikorn
configmaps
- lock *locking.RWMutex // lock - used not only
for context data but also to ensure that multiple event types are not executed
concurrently
- txnID atomic.Uint64 // transaction ID counter
- klogger klog.Logger
+ applications map[string]*Application // apps
+ schedulerCache *schedulercache.SchedulerCache // external cache
+ apiProvider client.APIProvider // apis to interact
with api-server, scheduler-core, etc
+ predManager predicates.PredicateManager // K8s predicates
+ namespace string // yunikorn
namespace
+ configMaps []*v1.ConfigMap // cached yunikorn
configmaps
+ lock *locking.RWMutex // lock - used not
only for context data but also to ensure that multiple event types are not
executed concurrently
+ txnID atomic.Uint64 // transaction ID
counter
+ klogger klog.Logger
+ resourceSliceTracker *tracker.Tracker
}
// NewContext create a new context for the scheduler using a default (empty)
configuration
@@ -127,6 +128,7 @@ func NewContextWithBootstrapConfigMaps(apis
client.APIProvider, bootstrapConfigM
log.Log(log.ShimClient).Error("unable to create the
resource slice tracker", zap.Error(err))
return nil
}
+ ctx.resourceSliceTracker = resourceSliceTracker
sharedDRAManager =
dynamicresources.NewDRAManager(context.TODO(), resourceClaimCache,
resourceSliceTracker, informerFactory)
}
@@ -138,6 +140,18 @@ func NewContextWithBootstrapConfigMaps(apis
client.APIProvider, bootstrapConfigM
return ctx
}
+// Stop ends all background activity managed by Context.
+func (ctx *Context) Stop() {
+ ctx.lock.Lock()
+ t := ctx.resourceSliceTracker
+ ctx.resourceSliceTracker = nil
+ ctx.lock.Unlock()
+
+ if t != nil {
+ t.Stop()
+ }
+}
+
func (ctx *Context) AddSchedulingEventHandlers() error {
err := ctx.apiProvider.AddEventHandler(&client.ResourceEventHandlers{
Type: client.ConfigMapInformerHandlers,
diff --git a/pkg/cache/context_test.go b/pkg/cache/context_test.go
index 07a534d2..787b76c8 100644
--- a/pkg/cache/context_test.go
+++ b/pkg/cache/context_test.go
@@ -2770,3 +2770,11 @@ func TestTerminatedOrphanedForeignPodNotAdopted(t
*testing.T) {
})
}
}
+
+func TestContextStop(t *testing.T) {
+ context := NewContext(client.NewMockedAPIProvider(false))
+ assert.Assert(t, context.resourceSliceTracker != nil)
+
+ context.Stop()
+ assert.Assert(t, context.resourceSliceTracker == nil)
+}
diff --git a/pkg/plugin/support/mock_helpers.go
b/pkg/plugin/support/mock_helpers.go
index 0d74eb3a..4ed16545 100644
--- a/pkg/plugin/support/mock_helpers.go
+++ b/pkg/plugin/support/mock_helpers.go
@@ -50,8 +50,13 @@ func SharedDRAManager() *dynamicresources.DefaultDRAManager {
if
feature.DefaultFeatureGate.Enabled(features.DynamicResourceAllocation) {
resourceClaimInformer :=
InformerFactory(ClientSet()).Resource().V1().ResourceClaims().Informer()
resourceClaimCache :=
assumecache.NewAssumeCache(klog.NewKlogr(), resourceClaimInformer,
"ResourceClaim", "", nil)
+ // Disable device taint rules in mock helper to prevent
goroutine leaks in tests.
+ // When EnableDeviceTaintRules is true, tracker.StartTracker
spawns a background goroutine waiting
+ // for informers to sync, which leaks because unit test
informers never sync. Existing unit tests
+ // do not exercise device taints or DRA resources. If future
tests specifically require device taint
+ // tracking, this should be re-enabled along with a proper
shutdown lifecycle.
resourceSliceTracker, err :=
tracker.StartTracker(context.TODO(), tracker.Options{
- EnableDeviceTaintRules:
feature.DefaultFeatureGate.Enabled(features.DRADeviceTaints),
+ EnableDeviceTaintRules: false,
SliceInformer:
InformerFactory(ClientSet()).Resource().V1().ResourceSlices(),
ClassInformer:
InformerFactory(ClientSet()).Resource().V1().DeviceClasses(),
TaintInformer:
InformerFactory(ClientSet()).Resource().V1beta2().DeviceTaintRules()})
diff --git a/pkg/shim/scheduler.go b/pkg/shim/scheduler.go
index 3d55070a..9a640250 100644
--- a/pkg/shim/scheduler.go
+++ b/pkg/shim/scheduler.go
@@ -252,6 +252,10 @@ func (ss *KubernetesShim) Stop() {
ss.phManager.Stop()
// stop the dispatcher
dispatcher.Stop()
+ // stop the context
+ if ss.context != nil {
+ ss.context.Stop()
+ }
})
if !stopped {
log.Log(log.ShimScheduler).Info("scheduler is already stopped")
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
