Add no-drain flag and wait for workers ready after upgrade

Signed-off-by: Jussi Nummelin <jnummelin@mirantis.com>
k0sproject · Feb 24, 2021 · 2116ac4 · 2116ac4
1 parent 7e6a116
commit 2116ac4
Show file tree

Hide file tree

Showing 3 changed files with 31 additions and 18 deletions.
diff --git a/cmd/apply.go b/cmd/apply.go
@@ -23,6 +23,10 @@ var applyCommand = &cli.Command{
 			Name:  "no-wait",
 			Usage: "Do not wait for worker nodes to join",
 		},
+		&cli.BoolFlag{
+			Name:  "no-drain",
+			Usage: "Do not drain worker nodes when upgrading",
+		},
 		debugFlag,
 		traceFlag,
 		analyticsFlag,
@@ -46,6 +50,7 @@ var applyCommand = &cli.Command{
 		}
 
 		phase.NoWait = ctx.Bool("no-wait")
+
 		manager := phase.Manager{Config: &c}
 
 		manager.AddPhase(
@@ -63,7 +68,9 @@ var applyCommand = &cli.Command{
 			&phase.InstallControllers{},
 			&phase.InstallWorkers{},
 			&phase.UpgradeControllers{},
-			&phase.UpgradeWorkers{},
+			&phase.UpgradeWorkers{
+				NoDrain: ctx.Bool("no-drain"),
+			},
 			&phase.Disconnect{},
 		)
 

diff --git a/phase/gather_k0s_facts.go b/phase/gather_k0s_facts.go
@@ -84,17 +84,11 @@ func (p *GatherK0sFacts) investigateK0s(h *cluster.Host) error {
 		return nil
 	}
 
-<<<<<<< HEAD
 	switch status.Role {
 	case "server":
 		status.Role = "controller"
 	case "server+worker":
 		status.Role = "controller+worker"
-=======
-	// Legacy role change
-	if status.Role == "server" {
-		status.Role = "controller"
->>>>>>> 489e103... Initial upgrade with server->controller migration
 	}
 
 	if status.Role != h.Role {
@@ -121,10 +115,6 @@ func (p *GatherK0sFacts) investigateK0s(h *cluster.Host) error {
 
 func (p *GatherK0sFacts) needsUpgrade(h *cluster.Host) bool {
 	c, err := semver.NewConstraint(fmt.Sprintf("< %s", p.Config.Spec.K0s.Version))
-	if err != nil {
-		log.Warnf("%s: failed to parse version constraint: %s", h, err.Error())
-		return false
-	}
 	current, err := semver.NewVersion(h.Metadata.K0sRunningVersion)
 	if err != nil {
 		log.Warnf("%s: failed to parse version info: %s", h, err.Error())

diff --git a/phase/upgrade_workers.go b/phase/upgrade_workers.go
@@ -14,6 +14,8 @@ import (
 type UpgradeWorkers struct {
 	GenericPhase
 
+	NoDrain bool
+
 	hosts  cluster.Hosts
 	leader *cluster.Host
 }
@@ -72,11 +74,14 @@ func (p *UpgradeWorkers) Run() error {
 
 func (p *UpgradeWorkers) upgradeWorker(h *cluster.Host) error {
 	log.Infof("%s: upgrade starting", h)
-	log.Debugf("%s: draining...", h)
-	if err := p.leader.DrainNode(h); err != nil {
-		return err
+
+	if !p.NoDrain {
+		log.Debugf("%s: draining...", h)
+		if err := p.leader.DrainNode(h); err != nil {
+			return err
+		}
+		log.Debugf("%s: draining complete", h)
 	}
-	log.Debugf("%s: draining complete", h)
 
 	log.Debugf("%s: Update and restart service", h)
 	if err := h.Configurer.StopService(h, h.K0sServiceName()); err != nil {
@@ -88,9 +93,20 @@ func (p *UpgradeWorkers) upgradeWorker(h *cluster.Host) error {
 	if err := h.Configurer.StartService(h, h.K0sServiceName()); err != nil {
 		return err
 	}
-	log.Debugf("%s: marking node schedulable again", h)
-	if err := p.leader.UncordonNode(h); err != nil {
-		return err
+	if !p.NoDrain {
+		log.Debugf("%s: marking node schedulable again", h)
+		if err := p.leader.UncordonNode(h); err != nil {
+			return err
+		}
+	}
+	if NoWait {
+		log.Debugf("%s: not waiting because --no-wait given", h)
+	} else {
+		log.Infof("%s: waiting for node to become ready again", h)
+		if err := p.Config.Spec.K0sLeader().WaitKubeNodeReady(h); err != nil {
+			return err
+		}
+		h.Metadata.Ready = true
 	}
 	log.Infof("%s: upgrade successful", h)
 	return nil