Files
flock/pkg/agent/nodecondition.go
T

67 lines
1.9 KiB
Go
Raw Normal View History

package agent
import (
"context"
"encoding/json"
"fmt"
"log/slog"
"time"
corev1 "k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/types"
"k8s.io/client-go/kubernetes"
"k8s.io/client-go/rest"
)
// keepNetworkAvailable maintains a NetworkUnavailable=False condition on
// the node's status. Calico-node sets this False while it owns CNI; on
// shutdown it sets it to True with reason CalicoIsDown, which adds the
// node.kubernetes.io/network-unavailable taint and blocks new scheduling.
// Once flock-agent is in charge, we own the condition.
//
// Re-applies every minute — heartbeat-style — so a stale condition from a
// previous CNI is overwritten without an explicit transition.
func keepNetworkAvailable(ctx context.Context, cfg *rest.Config, node string, logger *slog.Logger) {
cs, err := kubernetes.NewForConfig(cfg)
if err != nil {
logger.Warn("network-condition: kubernetes client", "err", err)
return
}
apply := func() {
now := metav1.Now()
patch := map[string]interface{}{
"status": map[string]interface{}{
"conditions": []corev1.NodeCondition{{
Type: corev1.NodeNetworkUnavailable,
Status: corev1.ConditionFalse,
Reason: "FlockReady",
Message: "flock-agent owns CNI on this node",
LastHeartbeatTime: now,
LastTransitionTime: now,
}},
},
}
body, _ := json.Marshal(patch)
_, err := cs.CoreV1().Nodes().Patch(ctx, node, types.MergePatchType, body, metav1.PatchOptions{}, "status")
if err != nil {
logger.Warn("network-condition: patch failed", "err", err)
return
}
}
apply()
t := time.NewTicker(60 * time.Second)
defer t.Stop()
for {
select {
case <-ctx.Done():
return
case <-t.C:
apply()
}
}
}
// silence unused-import warnings on non-Linux builds where this is unused.
var _ = fmt.Sprintf