fix reboot_test.go and collectEntries

dctest/reboot_test.go

@@ -3,6 +3,7 @@ package dctest
 import (
 	"encoding/json"
 	"fmt"
+	"slices"
 	"strings"
 	"time"
 
@@ -229,6 +230,24 @@ func testNecoRebooterRebootGracefully() {
 			return nil
 		}).Should(Succeed())
 
+		By("getting available racks")
+		stdout, _, err := execAt(bootServers[0], "kubectl", "get", "nodes", "-o", "json")
+		Expect(err).NotTo(HaveOccurred())
+		nodes := corev1.NodeList{}
+		err = json.Unmarshal(stdout, &nodes)
+		Expect(err).NotTo(HaveOccurred())
+		racks := []string{}
+		for _, node := range nodes.Items {
+			rack := node.Labels["topology.kubernetes.io/zone"]
+			if len(rack) == 0 {
+				continue
+			}
+			racks = append(racks, rack)
+		}
+		slices.Sort(racks)
+		racks = slices.Compact(racks)
+		Expect(len(racks)).To(BeNumerically(">", 1))
+
 		By("changing neco-rebooter config")
 		config := necorebooter.Config{
 			RebootTimes: []necorebooter.RebootTimes{
@@ -286,7 +305,7 @@ func testNecoRebooterRebootGracefully() {
 					},
 				},
 				NodeSelector: map[string]string{
-					"topology.kubernetes.io/zone": "rack1",
+					"topology.kubernetes.io/zone": racks[0],
 				},
 			},
 		}
@@ -317,18 +336,14 @@ func testNecoRebooterRebootGracefully() {
 		Expect(err).NotTo(HaveOccurred())
 
 		By("creating a pod")
-		stdout, stderr, err := execAtWithInput(bootServers[0], podYaml, "kubectl", "apply", "-f", "-")
-		if err != nil {
-			Fail(fmt.Sprintf("stdout: %s, stderr: %s, err: %v", stdout, stderr, err))
-		}
+		_, _, err = execAtWithInput(bootServers[0], podYaml, "kubectl", "apply", "-f", "-")
+		Expect(err).NotTo(HaveOccurred())
 		By("creating a pdb")
-		stdout, stderr, err = execAtWithInput(bootServers[0], pdbYaml, "kubectl", "apply", "-f", "-")
-		if err != nil {
-			Fail(fmt.Sprintf("stdout: %s, stderr: %s, err: %v", stdout, stderr, err))
-		}
+		_, _, err = execAtWithInput(bootServers[0], pdbYaml, "kubectl", "apply", "-f", "-")
+		Expect(err).NotTo(HaveOccurred())
 
-		By("adding rack1 nodes to reboot-list")
-		execSafeAt(bootServers[0], "sh", "-c", "yes | neco rebooter reboot-worker --rack=1")
+		By(fmt.Sprintf("adding %s nodes to reboot-list", racks[0]))
+		execSafeAt(bootServers[0], "sh", "-c", fmt.Sprintf("yes | neco rebooter reboot-worker --rack=%s", racks[0][4:]))
 
 		By("enable rebooting")
 		execSafeAt(bootServers[0], "ckecli", "rq", "enable")
@@ -346,16 +361,16 @@ func testNecoRebooterRebootGracefully() {
 			return nil
 		}).Should(Succeed())
 
-		By("adding rack2 nodes to reboot-list")
-		execSafeAt(bootServers[0], "sh", "-c", "yes | neco rebooter reboot-worker --rack=2")
+		By(fmt.Sprintf("adding %s nodes to reboot-list", racks[1]))
+		execSafeAt(bootServers[0], "sh", "-c", fmt.Sprintf("yes | neco rebooter reboot-worker --rack=%s", racks[1][4:]))
 
-		By("waiting for skipping rack1 and moving to rack2")
+		By(fmt.Sprintf("waiting for skipping %s and moving to %s", racks[0], racks[1]))
 		Eventually(func() error {
 			stdout, stderr, err := execAt(bootServers[0], "neco", "rebooter", "show-processing-group")
 			if err != nil {
 				return fmt.Errorf("stdout: %s, stderr: %s, err: %v", stdout, stderr, err)
 			}
-			if string(stdout) != "rack2\n" {
+			if strings.Contains(string(stdout), racks[1]) {
 				return fmt.Errorf("reboot-queue is not processed")
 			}
 			return nil

pkg/neco-rebooter/controller.go

@@ -218,6 +218,9 @@ func (c *Controller) collectEntries(rebootListEntries []*neco.RebootListEntry, r
 		}
 	}
 	for _, rqEntry := range rebootQueueEntries {
+		if rqEntry.Status == cke.RebootStatusCancelled {
+			continue
+		}
 		rlEntry := findRebootListEntryFromRebootQueueEntry(rebootListEntries, *rqEntry)
 		if rlEntry == nil {
 			orphanedEntry = append(orphanedEntry, EntrySet{nil, rqEntry})

pkg/neco-rebooter/controller_test.go

@@ -554,6 +554,10 @@ groupLabelKey: topology.kubernetes.io/zone
 			Node:   "node7",
 			Status: cke.RebootStatusQueued,
 		},
+		{ // orphaned but cancelled node
+			Node:   "node8",
+			Status: cke.RebootStatusCancelled,
+		},
 	}
 	collection := c.collectEntries(rebootListEntries, rebootQueueEntry, "group1")
 	if len(collection.CancelledEntry) != 1 {