Merge pull request #590 from cybozu-go/add-rebooting-metrics

Add "cke_node_reboot_status" metrics
cybozu-go · Jan 17, 2023 · 06ada29 · 06ada29
2 parents c6aa43d + 4a41341
commit 06ada29
Show file tree

Hide file tree

Showing 9 changed files with 167 additions and 17 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -5,6 +5,10 @@ This project employs a versioning scheme described in [RELEASE.md](RELEASE.md#ve
 
 ## [Unreleased]
 
+### Added
+
+- Add `cke_node_reboot_status` metrics [#590](https://github.com/cybozu-go/cke/pull/590)
+
 ## [1.24.0]
 
 ### Changed
@@ -42,7 +46,7 @@ This project employs a versioning scheme described in [RELEASE.md](RELEASE.md#ve
 
 ## Ancient changes
 
-- See [release-1.23/CHANGELOG.md](https://github.com/cybozu-go/cke/blob/release-1.23/CHANGELOG.md) for changes in CKE 1.2.
+- See [release-1.23/CHANGELOG.md](https://github.com/cybozu-go/cke/blob/release-1.23/CHANGELOG.md) for changes in CKE 1.23.
 - See [release-1.22/CHANGELOG.md](https://github.com/cybozu-go/cke/blob/release-1.22/CHANGELOG.md) for changes in CKE 1.22.
 - See [release-1.21/CHANGELOG.md](https://github.com/cybozu-go/cke/blob/release-1.21/CHANGELOG.md) for changes in CKE 1.21.
 - See [release-1.20/CHANGELOG.md](https://github.com/cybozu-go/cke/blob/release-1.20/CHANGELOG.md) for changes in CKE 1.20.

diff --git a/docs/metrics.md b/docs/metrics.md
@@ -3,17 +3,18 @@ Metrics
 
 CKE exposes the following metrics with the Prometheus format at `/metrics` REST API endpoint.  All these metrics are prefixed with `cke_`
 
-| Name                                  | Description                                                                | Type  | Labels   |
-| ------------------------------------- | -------------------------------------------------------------------------- | ----- | -------- |
-| leader                                | True (=1) if this server is the leader of CKE.                             | Gauge |          |
-| operation_phase                       | 1 if CKE is operating in the phase specified by the `phase` label.         | Gauge | `phase`  |
-| operation_phase_timestamp_seconds     | The Unix timestamp when `operation_phase` was last updated.                | Gauge |          |
-| reboot_queue_entries                  | The number of reboot queue entries remaining.                              | Gauge |          |
-| reboot_queue_items                    | The number reboot queue entries remaining per status.                      | Gauge | `status` |
-| sabakan_integration_successful        | True (=1) if sabakan-integration satisfies constraints.                    | Gauge |          |
-| sabakan_integration_timestamp_seconds | The Unix timestamp when `sabakan_integration_successful` was last updated. | Gauge |          |
-| sabakan_workers                       | The number of worker nodes for each role.                                  | Gauge | `role`   |
-| sabakan_unused_machines               | The number of unused machines.                                             | Gauge |          |
+| Name                                  | Description                                                                | Type  | Labels           |
+| ------------------------------------- | -------------------------------------------------------------------------- | ----- | ---------------- |
+| leader                                | True (=1) if this server is the leader of CKE.                             | Gauge |                  |
+| node_reboot_status                    | The reboot status of a node.                                               | Gauge | `node`, `status` |
+| operation_phase                       | 1 if CKE is operating in the phase specified by the `phase` label.         | Gauge | `phase`          |
+| operation_phase_timestamp_seconds     | The Unix timestamp when `operation_phase` was last updated.                | Gauge |                  |
+| reboot_queue_entries                  | The number of reboot queue entries remaining.                              | Gauge |                  |
+| reboot_queue_items                    | The number reboot queue entries remaining per status.                      | Gauge | `status`         |
+| sabakan_integration_successful        | True (=1) if sabakan-integration satisfies constraints.                    | Gauge |                  |
+| sabakan_integration_timestamp_seconds | The Unix timestamp when `sabakan_integration_successful` was last updated. | Gauge |                  |
+| sabakan_workers                       | The number of worker nodes for each role.                                  | Gauge | `role`           |
+| sabakan_unused_machines               | The number of unused machines.                                             | Gauge |                  |
 
 All metrics but `leader` are available only when the server is the leader of CKE.
 `sabakan_*` metrics are available only when [Sabakan integration](sabakan-integration.md) is enabled.

diff --git a/metrics/collector.go b/metrics/collector.go
@@ -55,7 +55,7 @@ func NewCollector(client *v3.Client) prometheus.Collector {
 				isAvailable: isOperationPhaseAvailable,
 			},
 			"reboot": {
-				collectors:  []prometheus.Collector{rebootQueueEntries, rebootQueueItems},
+				collectors:  []prometheus.Collector{rebootQueueEntries, rebootQueueItems, nodeRebootStatus},
 				isAvailable: isRebootAvailable,
 			},
 			"sabakan_integration": {

diff --git a/metrics/metrics.go b/metrics/metrics.go
@@ -50,6 +50,14 @@ var rebootQueueItems = prometheus.NewGaugeVec(
 	[]string{"status"},
 )
 
+var nodeRebootStatus = prometheus.NewGaugeVec(
+	prometheus.GaugeOpts{
+		Namespace: namespace,
+		Name:      "node_reboot_status",
+		Help:      "The reboot status of a node.",
+	}, []string{"node", "status"},
+)
+
 var sabakanIntegrationSuccessful = prometheus.NewGauge(
 	prometheus.GaugeOpts{
 		Namespace: namespace,

diff --git a/metrics/updater.go b/metrics/updater.go
@@ -53,6 +53,22 @@ func UpdateRebootQueueItems(counts map[string]int) {
 	}
 }
 
+// UpdateNodeRebootStatus updates "node_reboot_status".
+func UpdateNodeRebootStatus(nodeStatus map[string]map[string]bool) {
+	for node, statuses := range nodeStatus {
+		for status, matches := range statuses {
+			value := float64(0)
+			if matches {
+				value = 1
+			}
+			nodeRebootStatus.With(map[string]string{
+				"node":   node,
+				"status": status,
+			}).Set(value)
+		}
+	}
+}
+
 func isRebootAvailable(_ context.Context, _ storage) (bool, error) {
 	return isLeader, nil
 }

diff --git a/metrics/updater_test.go b/metrics/updater_test.go
@@ -79,6 +79,7 @@ func TestMetricsUpdater(t *testing.T) {
 	t.Run("UpdateOperationPhase", testUpdateOperationPhase)
 	t.Run("UpdateRebootQueueEntries", testUpdateRebootQueueEntries)
 	t.Run("UpdateRebootQueueItems", testUpdateRebootQueueItems)
+	t.Run("UpdateNodeRebootStatus", testUpdateNodeRebootStatus)
 	t.Run("UpdateSabakanIntegration", testUpdateSabakanIntegration)
 }
 
@@ -372,6 +373,61 @@ func testUpdateRebootQueueItems(t *testing.T) {
 	}
 }
 
+func testUpdateNodeRebootStatus(t *testing.T) {
+	input := map[string]map[string]bool{
+		"node1": {
+			"queued":    false,
+			"draining":  false,
+			"rebooting": true,
+			"cancelled": false,
+		},
+		"node2": {
+			"queued":    false,
+			"draining":  false,
+			"rebooting": false,
+			"cancelled": false,
+		},
+	}
+	expected := input
+
+	ctx := context.Background()
+	defer ctx.Done()
+
+	collector, _ := newTestCollector()
+	handler := GetHandler(collector)
+
+	UpdateNodeRebootStatus(input)
+
+	w := httptest.NewRecorder()
+	req := httptest.NewRequest("GET", "/metrics", nil)
+	handler.ServeHTTP(w, req)
+
+	metricsFamily, err := parseMetrics(w.Result())
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	actual := make(map[string]map[string]bool)
+	for _, mf := range metricsFamily {
+		if *mf.Name != "cke_node_reboot_status" {
+			continue
+		}
+		for _, m := range mf.Metric {
+			labels := labelToMap(m.Label)
+			node := labels["node"]
+			status := labels["status"]
+			if _, ok := actual[node]; !ok {
+				actual[node] = make(map[string]bool)
+			}
+			actual[node][status] = *m.Gauge.Value != 0
+		}
+	}
+
+	if !cmp.Equal(actual, expected) {
+		t.Errorf("unexpected map was build from cke_node_reboot_status.  expected: %v, actual: %v", expected, actual)
+	}
+}
+
 func testUpdateSabakanIntegration(t *testing.T) {
 	testCases := []updateSabakanIntegrationTestCase{
 		{

diff --git a/op/reboot.go b/op/reboot.go
@@ -520,9 +520,16 @@ func (c rebootRecalcMetricsCommand) Run(ctx context.Context, inf cke.Infrastruct
 	if err != nil {
 		return err
 	}
+	cluster, err := inf.Storage().GetCluster(ctx)
+	if err != nil {
+		return err
+	}
+
 	metrics.UpdateRebootQueueEntries(len(rqEntries))
 	itemCounts := cke.CountRebootQueueEntries(rqEntries)
 	metrics.UpdateRebootQueueItems(itemCounts)
+	nodeStatus := cke.BuildNodeRebootStatus(cluster.Nodes, rqEntries)
+	metrics.UpdateNodeRebootStatus(nodeStatus)
 
 	return nil
 }

diff --git a/reboot.go b/reboot.go
@@ -15,6 +15,8 @@ const (
 	RebootStatusCancelled = RebootStatus("cancelled")
 )
 
+var rebootStatuses = []RebootStatus{RebootStatusQueued, RebootStatusDraining, RebootStatusRebooting, RebootStatusCancelled}
+
 // RebootQueueEntry represents a queue entry of reboot operation
 type RebootQueueEntry struct {
 	Index              int64        `json:"index,string"`
@@ -58,14 +60,40 @@ func DedupRebootQueueEntries(entries []*RebootQueueEntry) []*RebootQueueEntry {
 
 func CountRebootQueueEntries(entries []*RebootQueueEntry) map[string]int {
 	ret := map[string]int{}
-	ret[string(RebootStatusQueued)] = 0
-	ret[string(RebootStatusDraining)] = 0
-	ret[string(RebootStatusRebooting)] = 0
-	ret[string(RebootStatusCancelled)] = 0
+	for _, status := range rebootStatuses {
+		// initialize explicitly to provide list of possible statuses
+		ret[string(status)] = 0
+	}
 
 	for _, entry := range entries {
 		ret[string(entry.Status)]++
 	}
 
 	return ret
 }
+
+func BuildNodeRebootStatus(nodes []*Node, entries []*RebootQueueEntry) map[string]map[string]bool {
+	ret := make(map[string]map[string]bool)
+	addr2name := make(map[string]string)
+
+	for _, node := range nodes {
+		name := node.Nodename()
+		ret[name] = make(map[string]bool)
+		for _, status := range rebootStatuses {
+			// initialize explicitly to provide list of possible statuses
+			ret[name][string(status)] = false
+		}
+		addr2name[node.Address] = name
+	}
+
+	for _, entry := range entries {
+		name, ok := addr2name[entry.Node]
+		if !ok {
+			// removed from K8s cluster after queued
+			continue
+		}
+		ret[name][string(entry.Status)] = true
+	}
+
+	return ret
+}
diff --git a/reboot_test.go b/reboot_test.go
@@ -45,3 +45,33 @@ func TestCountRebootQueueEntries(t *testing.T) {
 		t.Errorf("expected: %v, actual: %v", expected, actual)
 	}
 }
+
+func TestBuildNodeRebootStatus(t *testing.T) {
+	inputNodes := []*Node{
+		{Hostname: "node1", Address: "1.1.1.1"},
+		{Hostname: "node2", Address: "2.2.2.2"},
+	}
+	inputEntries := []*RebootQueueEntry{
+		{Node: "1.1.1.1", Status: RebootStatusRebooting},
+		{Node: "3.3.3.3", Status: RebootStatusCancelled},
+	}
+	expected := map[string]map[string]bool{
+		"node1": {
+			"queued":    false,
+			"draining":  false,
+			"rebooting": true,
+			"cancelled": false,
+		},
+		"node2": {
+			"queued":    false,
+			"draining":  false,
+			"rebooting": false,
+			"cancelled": false,
+		},
+	}
+	actual := BuildNodeRebootStatus(inputNodes, inputEntries)
+
+	if !cmp.Equal(actual, expected) {
+		t.Errorf("expected: %v, actual: %v", expected, actual)
+	}
+}