Fix: only schedule node drain canaries on nodes with osds

Signed-off-by: Rohan CJ <rohantmp@gmail.com>
This commit is contained in:
Rohan CJ committed 2019-09-06 09:41:56 +05:30
1 parent 6fcd2c21fa
commit e1bd6595d7
2 files changed
+68 -5

No files matched your search

+47 -1
View File
@@ -20,10 +20,13 @@ import (
"reflect"
"time"
"github.com/rook/rook/pkg/operator/ceph/cluster/osd"
"github.com/rook/rook/pkg/operator/ceph/disruption/controllerconfig"
"github.com/rook/rook/pkg/operator/k8sutil"
appsv1 "k8s.io/api/apps/v1"
corev1 "k8s.io/api/core/v1"
"k8s.io/apimachinery/pkg/types"
"sigs.k8s.io/controller-runtime/pkg/controller"
"sigs.k8s.io/controller-runtime/pkg/event"
"sigs.k8s.io/controller-runtime/pkg/handler"
@@ -67,7 +70,7 @@ func Add(mgr manager.Manager, context *controllerconfig.Context) error {
if err != nil {
return err
}
// Watch for changes to the node canary deployments
err = c.Watch(&source.Kind{Type: &appsv1.Deployment{}}, &handler.EnqueueRequestForOwner{
OwnerType: &corev1.Node{},
IsController: true,
@@ -76,5 +79,48 @@ func Add(mgr manager.Manager, context *controllerconfig.Context) error {
return err
}
// Watch for changes to the osd pod nodename and enqueue thier nodes
err = c.Watch(
&source.Kind{Type: &corev1.Pod{}},
&handler.EnqueueRequestsFromMapFunc{
ToRequests: handler.ToRequestsFunc(func(obj handler.MapObject) []reconcile.Request {
pod, ok := obj.Object.(*corev1.Pod)
if !ok {
return []reconcile.Request{}
}
labels := pod.GetLabels()
if appName, ok := labels[k8sutil.AppAttr]; !ok || appName != osd.AppName {
return []reconcile.Request{}
}
nodeName := pod.Spec.NodeName
if len(nodeName) < 1 {
return []reconcile.Request{}
}
req := reconcile.Request{NamespacedName: types.NamespacedName{Name: nodeName}}
return []reconcile.Request{req}
}),
},
predicate.Funcs{
UpdateFunc: func(event event.UpdateEvent) bool {
oldPod, ok := event.ObjectOld.(*corev1.Pod)
if !ok {
return false
}
newPod, ok := event.ObjectNew.(*corev1.Pod)
if !ok {
return false
}
// only enqueue if the nodename has changed
if oldPod.Spec.NodeName == newPod.Spec.NodeName {
return false
}
return true
},
},
)
if err != nil {
return err
}
return nil
}
@@ -22,6 +22,7 @@ import (
"github.com/coreos/pkg/capnslog"
"github.com/rook/rook/pkg/operator/ceph/cluster/osd"
"github.com/rook/rook/pkg/operator/ceph/disruption/controllerconfig"
"github.com/rook/rook/pkg/operator/k8sutil"
@@ -82,6 +83,18 @@ func (r *ReconcileNode) reconcile(request reconcile.Request) (reconcile.Result,
return reconcile.Result{}, fmt.Errorf("Label key %s does not exist on node %s", corev1.LabelHostname, request.NamespacedName)
}
osdPodList := &corev1.PodList{}
err = r.client.List(context.TODO(), osdPodList, client.MatchingLabels{k8sutil.AppAttr: osd.AppName})
if err != nil {
return reconcile.Result{}, fmt.Errorf("could not list the osd pods: %+v", err)
}
occupiedByOSD := false
for _, osdPod := range osdPodList.Items {
if osdPod.Spec.NodeName == request.Name {
occupiedByOSD = true
break
}
}
// Create or Update the deployment default/foo
deploy := &appsv1.Deployment{
ObjectMeta: metav1.ObjectMeta{
@@ -123,11 +136,15 @@ func (r *ReconcileNode) reconcile(request reconcile.Request) (reconcile.Result,
return nil
}
op, err := controllerutil.CreateOrUpdate(context.TODO(), r.client, deploy, mutateFunc)
if err != nil {
return reconcile.Result{}, fmt.Errorf("Node reconcile failed on op: %s : %+v", op, err)
if occupiedByOSD {
op, err := controllerutil.CreateOrUpdate(context.TODO(), r.client, deploy, mutateFunc)
if err != nil {
return reconcile.Result{}, fmt.Errorf("Node reconcile failed on op: %s : %+v", op, err)
}
logger.Debugf("deployment successfully reconciled for node %s. operation: %s", request.Name, op)
} else {
logger.Debugf("not watching for drains on node %s as there are no osds running there.", request.Name)
}
logger.Debugf("Deployment successfully reconciled. operation: %s", op)
return reconcile.Result{}, nil
}