From 77a75cdb34c5cb972605c76d9aeb151b243a051d Mon Sep 17 00:00:00 2001
From: YZ775 <yuzuki-mimura@cybozu.co.jp>
Date: Tue, 26 Mar 2024 02:28:23 +0000
Subject: [PATCH] change priorty of reboot queue cancel

---
 op/reboot_decide.go     | 17 ++++++++++++++++-
 server/control.go       |  2 ++
 server/strategy.go      |  6 ++++++
 server/strategy_test.go |  9 +++++++--
 4 files changed, 31 insertions(+), 3 deletions(-)

diff --git a/op/reboot_decide.go b/op/reboot_decide.go
index 14e0bb7a..f1d7706b 100644
--- a/op/reboot_decide.go
+++ b/op/reboot_decide.go
@@ -245,7 +245,6 @@ func CheckRebootDequeue(ctx context.Context, c *cke.Cluster, rqEntries []*cke.Re
 	for _, entry := range rqEntries {
 		switch {
 		case !entry.ClusterMember(c):
-		case entry.Status == cke.RebootStatusCancelled:
 		case entry.Status == cke.RebootStatusRebooting && rebootCompleted(ctx, c, entry):
 		default:
 			continue
@@ -257,6 +256,22 @@ func CheckRebootDequeue(ctx context.Context, c *cke.Cluster, rqEntries []*cke.Re
 	return dequeued
 }
 
+func CheckRebootCanceled(ctx context.Context, c *cke.Cluster, rqEntries []*cke.RebootQueueEntry) []*cke.RebootQueueEntry {
+	dequeued := []*cke.RebootQueueEntry{}
+
+	for _, entry := range rqEntries {
+		switch {
+		case entry.Status == cke.RebootStatusCancelled:
+		default:
+			continue
+		}
+
+		dequeued = append(dequeued, entry)
+	}
+
+	return dequeued
+}
+
 func rebootCompleted(ctx context.Context, c *cke.Cluster, entry *cke.RebootQueueEntry) bool {
 	if c.Reboot.CommandTimeoutSeconds != nil && *c.Reboot.CommandTimeoutSeconds != 0 {
 		var cancel context.CancelFunc
diff --git a/server/control.go b/server/control.go
index 0125ca1e..e1ac24d9 100644
--- a/server/control.go
+++ b/server/control.go
@@ -346,6 +346,7 @@ func (c Controller) runOnce(ctx context.Context, leaderKey string, tick <-chan t
 	newlyDrained := op.ChooseDrainedNodes(cluster, apiServers, rqEntries)
 	drainCompleted, drainTimedout, _ := op.CheckDrainCompletion(ctx, inf, nf.HealthyAPIServer(), cluster, rqEntries)
 	rebootDequeued := op.CheckRebootDequeue(ctx, cluster, rqEntries)
+	RebootCanceled := op.CheckRebootCanceled(ctx, cluster, rqEntries)
 
 	ops, phase := DecideOps(cluster, status, constraints, rcs, DecideOpsRebootArgs{
 		RQEntries:      rqEntries,
@@ -353,6 +354,7 @@ func (c Controller) runOnce(ctx context.Context, leaderKey string, tick <-chan t
 		DrainCompleted: drainCompleted,
 		DrainTimedout:  drainTimedout,
 		RebootDequeued: rebootDequeued,
+		RebootCanceled: RebootCanceled,
 	}, c.config)
 
 	st := &cke.ServerStatus{
diff --git a/server/strategy.go b/server/strategy.go
index 9bd38970..f5bef21e 100644
--- a/server/strategy.go
+++ b/server/strategy.go
@@ -22,6 +22,7 @@ type DecideOpsRebootArgs struct {
 	DrainCompleted []*cke.RebootQueueEntry
 	DrainTimedout  []*cke.RebootQueueEntry
 	RebootDequeued []*cke.RebootQueueEntry
+	RebootCanceled []*cke.RebootQueueEntry
 }
 
 // DecideOps returns the next operations to do and the operation phase.
@@ -883,6 +884,11 @@ func rebootOps(c *cke.Cluster, constraints *cke.Constraints, rebootArgs DecideOp
 		return nil, false
 	}
 
+	if len(rebootArgs.RebootCanceled) > 0 {
+		phaseReboot = true
+		ops = append(ops, op.RebootDequeueOp(rebootArgs.RebootCanceled))
+		return ops, phaseReboot
+	}
 	if len(rebootArgs.NewlyDrained) > 0 {
 		phaseReboot = true
 		sshCheckNodes := make([]*cke.Node, 0, len(nf.cluster.Nodes))
diff --git a/server/strategy_test.go b/server/strategy_test.go
index 77615847..52ac7f51 100644
--- a/server/strategy_test.go
+++ b/server/strategy_test.go
@@ -658,6 +658,11 @@ func (d testData) withRebootDequeued(entries []*cke.RebootQueueEntry) testData {
 	return d
 }
 
+func (d testData) withRebootCanceled(entries []*cke.RebootQueueEntry) testData {
+	d.RebootArgs.RebootCanceled = entries
+	return d
+}
+
 func (d testData) withDisableProxy() testData {
 	d.Cluster.Options.Proxy.Disable = true
 	return d
@@ -1274,7 +1279,7 @@ func TestDecideOps(t *testing.T) {
 					Node:   nodeNames[2],
 					Status: cke.RebootStatusCancelled,
 				},
-			}).withRebootDequeued([]*cke.RebootQueueEntry{
+			}).withRebootCanceled([]*cke.RebootQueueEntry{
 				{
 					Index:  1,
 					Node:   nodeNames[2],
@@ -2700,7 +2705,7 @@ func TestDecideOps(t *testing.T) {
 					Node:   nodeNames[4],
 					Status: cke.RebootStatusCancelled,
 				},
-			}).withRebootDequeued([]*cke.RebootQueueEntry{
+			}).withRebootCanceled([]*cke.RebootQueueEntry{
 				{
 					Index:  1,
 					Node:   nodeNames[4],