Merge pull request planetlabs#66 from planetlabs/koobz/no-drain

Add skip-drain flag.

README.md

@@ -38,6 +38,14 @@ Flags:
       --drain-buffer=10m0s       Minimum time between starting each drain. Nodes are always cordoned immediately.
       --node-label=KEY=VALUE ...
                                  Only nodes with this label will be eligible for cordoning and draining. May be specified multiple times.
+      --namespace="kube-system"  Namespace used to create leader election lock object.
+      --leader-election-lease-duration=15s
+                                 Lease duration for leader election.
+      --leader-election-renew-deadline=10s
+                                 Leader election renew deadline.
+      --leader-election-retry-period=2s
+                                 Leader election retry period.
+      --skip-drain               Whether to skip draining nodes after cordoning.
       --evict-daemonset-pods     Evict pods that were created by an extant DaemonSet.
       --evict-emptydir-pods      Evict pods with local storage, i.e. with emptyDir volumes.
       --evict-unreplicated-pods  Evict pods that were not created by a replication controller.
@@ -46,7 +54,6 @@ Flags:
 
 Args:
   <node-conditions>  Nodes for which any of these conditions are true will be cordoned and drained.
-
 ```
 
 ## Considerations
@@ -149,4 +156,11 @@ is marked as `Failed`. If you want to reschedule a drain tentative on that node,
 
 ```
 kubectl annotate node {node-name} draino/drain-retry=true
-```
+```
+## Modes
+
+### Dry Run
+Draino can be run in dry run mode using the `--dry-run` flag.
+
+### Cordon Only
+Draino can also optionally be run in a mode where the nodes are only cordoned, and not drained. This can be achieved by using the `--skip-drain` flag.

cmd/draino/draino.go

@@ -64,6 +64,7 @@ func main() {
 		leaderElectionRenewDeadline = app.Flag("leader-election-renew-deadline", "Leader election renew deadline.").Default(DefaultLeaderElectionRenewDeadline.String()).Duration()
 		leaderElectionRetryPeriod   = app.Flag("leader-election-retry-period", "Leader election retry period.").Default(DefaultLeaderElectionRetryPeriod.String()).Duration()
 
+		skipDrain             = app.Flag("skip-drain", "Whether to skip draining nodes after cordoning.").Default("false").Bool()
 		evictDaemonSetPods    = app.Flag("evict-daemonset-pods", "Evict pods that were created by an extant DaemonSet.").Bool()
 		evictLocalStoragePods = app.Flag("evict-emptydir-pods", "Evict pods with local storage, i.e. with emptyDir volumes.").Bool()
 		evictUnreplicatedPods = app.Flag("evict-unreplicated-pods", "Evict pods that were not created by a replication controller.").Bool()
@@ -142,7 +143,10 @@ func main() {
 		kubernetes.NewAPICordonDrainer(cs,
 			kubernetes.MaxGracePeriod(*maxGracePeriod),
 			kubernetes.EvictionHeadroom(*evictionHeadroom),
-			kubernetes.WithPodFilter(kubernetes.NewPodFilters(pf...))),
+			kubernetes.WithSkipDrain(*skipDrain),
+			kubernetes.WithPodFilter(kubernetes.NewPodFilters(pf...)),
+			kubernetes.WithAPICordonDrainerLogger(log),
+		),
 		kubernetes.NewEventRecorder(cs),
 		kubernetes.WithLogger(log),
 		kubernetes.WithDrainBuffer(*drainBuffer))

internal/kubernetes/drainer.go

@@ -21,6 +21,7 @@ import (
 	"time"
 
 	"github.com/pkg/errors"
+	"go.uber.org/zap"
 	core "k8s.io/api/core/v1"
 	policy "k8s.io/api/policy/v1beta1"
 	apierrors "k8s.io/apimachinery/pkg/api/errors"
@@ -38,6 +39,7 @@ const (
 	kindDaemonSet = "DaemonSet"
 
 	ConditionDrainedScheduled = "DrainScheduled"
+	DefaultSkipDrain          = false
 )
 
 type errTimeout struct{}
@@ -93,11 +95,13 @@ func (d *NoopCordonDrainer) MarkDrain(n *core.Node, when, finish time.Time, fail
 // APICordonDrainer drains Kubernetes nodes via the Kubernetes API.
 type APICordonDrainer struct {
 	c kubernetes.Interface
+	l *zap.Logger
 
 	filter PodFilterFunc
 
 	maxGracePeriod   time.Duration
 	evictionHeadroom time.Duration
+	skipDrain        bool
 }
 
 // SuppliedCondition defines the condition will be watched.
@@ -135,14 +139,31 @@ func WithPodFilter(f PodFilterFunc) APICordonDrainerOption {
 	}
 }
 
+// WithDrain determines if we're actually going to drain nodes
+func WithSkipDrain(b bool) APICordonDrainerOption {
+	return func(d *APICordonDrainer) {
+		d.skipDrain = b
+	}
+}
+
+// WithAPICordonDrainerLogger configures a APICordonDrainer to use the supplied
+// logger.
+func WithAPICordonDrainerLogger(l *zap.Logger) APICordonDrainerOption {
+	return func(d *APICordonDrainer) {
+		d.l = l
+	}
+}
+
 // NewAPICordonDrainer returns a CordonDrainer that cordons and drains nodes via
 // the Kubernetes API.
 func NewAPICordonDrainer(c kubernetes.Interface, ao ...APICordonDrainerOption) *APICordonDrainer {
 	d := &APICordonDrainer{
 		c:                c,
+		l:                zap.NewNop(),
 		filter:           NewPodFilters(),
 		maxGracePeriod:   DefaultMaxGracePeriod,
 		evictionHeadroom: DefaultEvictionOverhead,
+		skipDrain:        DefaultSkipDrain,
 	}
 	for _, o := range ao {
 		o(d)
@@ -234,6 +255,13 @@ func IsMarkedForDrain(n *core.Node) bool {
 
 // Drain the supplied node. Evicts the node of all but mirror and DaemonSet pods.
 func (d *APICordonDrainer) Drain(n *core.Node) error {
+
+	// Do nothing if draining is not enabled.
+	if d.skipDrain {
+		d.l.Debug("Skipping drain because draining is disabled")
+		return nil
+	}
+
 	pods, err := d.getPods(n.GetName())
 	if err != nil {
 		return errors.Wrapf(err, "cannot get pods for node %s", n.GetName())

internal/kubernetes/podfilters_test.go

@@ -198,7 +198,7 @@ func TestPodFilters(t *testing.T) {
 			name: "NoPodAnnotations",
 			pod: core.Pod{
 				ObjectMeta: meta.ObjectMeta{
-					Name:        podName,
+					Name: podName,
 				},
 			},
 			filter:       UnprotectedPodFilter("Protect"),
@@ -208,7 +208,7 @@ func TestPodFilters(t *testing.T) {
 			name: "NoPodAnnotationsWithEmptyUserValue",
 			pod: core.Pod{
 				ObjectMeta: meta.ObjectMeta{
-					Name:        podName,
+					Name: podName,
 				},
 			},
 			filter:       UnprotectedPodFilter("Protect="),