|
| 1 | +package kubeletversionskewcontroller |
| 2 | + |
| 3 | +import ( |
| 4 | + "context" |
| 5 | + "fmt" |
| 6 | + "regexp" |
| 7 | + "sort" |
| 8 | + "strings" |
| 9 | + |
| 10 | + "github.com/blang/semver" |
| 11 | + operatorv1 "github.com/openshift/api/operator/v1" |
| 12 | + "github.com/openshift/library-go/pkg/controller/factory" |
| 13 | + "github.com/openshift/library-go/pkg/operator/events" |
| 14 | + "github.com/openshift/library-go/pkg/operator/management" |
| 15 | + "github.com/openshift/library-go/pkg/operator/status" |
| 16 | + "github.com/openshift/library-go/pkg/operator/v1helpers" |
| 17 | + corev1 "k8s.io/api/core/v1" |
| 18 | + "k8s.io/apimachinery/pkg/labels" |
| 19 | + "k8s.io/apimachinery/pkg/util/runtime" |
| 20 | + corev1listers "k8s.io/client-go/listers/core/v1" |
| 21 | +) |
| 22 | + |
| 23 | +const ( |
| 24 | + KubeletMinorVersionUpgradeableConditionType = "KubeletMinorVersionUpgradeable" |
| 25 | + |
| 26 | + KubeletVersionUnknownReason = "KubeletVersionUnknown" |
| 27 | + KubeletMinorVersionSyncedReason = "KubeletMinorVersionsSynced" |
| 28 | + KubeletMinorVersionSupportedNextUpgradeReason = "KubeletMinorVersionSupportedNextUpgrade" |
| 29 | + KubeletMinorVersionUnsupportedNextUpgradeReason = "KubeletMinorVersionUnsupportedNextUpgrade" |
| 30 | + KubeletMinorVersionUnsupportedReason = "KubeletMinorVersionUnsupported" |
| 31 | + KubeletMinorVersionAheadReason = "KubeletMinorVersionAhead" |
| 32 | +) |
| 33 | + |
| 34 | +// KubeletVersionSkewController sets Upgradeable=False if the kubelet |
| 35 | +// version on a node prevents upgrading to a supported OpenShift version. |
| 36 | +// |
| 37 | +// For odd OpenShift minor versions, kubelet versions 0 or 1 minor |
| 38 | +// versions behind the API server version are supported. |
| 39 | +// |
| 40 | +// For even OpenShift minor versions, kubelet versions 0, 1, or 2 |
| 41 | +// minor versions behind the API server version are supported. |
| 42 | +type KubeletVersionSkewController interface { |
| 43 | + factory.Controller |
| 44 | +} |
| 45 | + |
| 46 | +func NewKubeletVersionSkewController( |
| 47 | + operatorClient v1helpers.OperatorClient, |
| 48 | + kubeInformersForNamespaces v1helpers.KubeInformersForNamespaces, |
| 49 | + recorder events.Recorder, |
| 50 | +) *kubeletVersionSkewController { |
| 51 | + openShiftVersion := semver.MustParse(status.VersionForOperatorFromEnv()) |
| 52 | + nextOpenShiftVersion := semver.Version{Major: openShiftVersion.Major, Minor: openShiftVersion.Minor + 1} |
| 53 | + c := &kubeletVersionSkewController{ |
| 54 | + operatorClient: operatorClient, |
| 55 | + nodeLister: kubeInformersForNamespaces.InformersFor("").Core().V1().Nodes().Lister(), |
| 56 | + apiServerVersion: semver.MustParse(status.VersionForOperandFromEnv()), |
| 57 | + minSupportedSkew: minSupportedKubeletSkewForOpenShiftVersion(openShiftVersion), |
| 58 | + minSupportedSkewNextVersion: minSupportedKubeletSkewForOpenShiftVersion(nextOpenShiftVersion), |
| 59 | + } |
| 60 | + c.Controller = factory.New(). |
| 61 | + WithSync(c.sync). |
| 62 | + WithInformers(kubeInformersForNamespaces.InformersFor("").Core().V1().Nodes().Informer()). |
| 63 | + ToController("KubeletVersionSkewController", recorder.WithComponentSuffix("kubelet-version-skew-controller")) |
| 64 | + return c |
| 65 | +} |
| 66 | + |
| 67 | +func minSupportedKubeletSkewForOpenShiftVersion(v semver.Version) int { |
| 68 | + switch v.Minor % 2 { |
| 69 | + case 0: // even OpenShift versions |
| 70 | + return -2 |
| 71 | + case 1: // odd OpenShift versions |
| 72 | + return -1 |
| 73 | + default: |
| 74 | + panic("should not happen") |
| 75 | + } |
| 76 | +} |
| 77 | + |
| 78 | +type kubeletVersionSkewController struct { |
| 79 | + factory.Controller |
| 80 | + operatorClient v1helpers.OperatorClient |
| 81 | + nodeLister corev1listers.NodeLister |
| 82 | + apiServerVersion semver.Version |
| 83 | + minSupportedSkew int |
| 84 | + minSupportedSkewNextVersion int |
| 85 | +} |
| 86 | + |
| 87 | +func (c *kubeletVersionSkewController) sync(_ context.Context, _ factory.SyncContext) error { |
| 88 | + operatorSpec, _, _, err := c.operatorClient.GetOperatorState() |
| 89 | + if err != nil { |
| 90 | + return err |
| 91 | + } |
| 92 | + if !management.IsOperatorManaged(operatorSpec.ManagementState) { |
| 93 | + return nil |
| 94 | + } |
| 95 | + |
| 96 | + nodes, err := c.nodeLister.List(labels.Everything()) |
| 97 | + if err != nil { |
| 98 | + return err |
| 99 | + } |
| 100 | + sort.Sort(byName(nodes)) |
| 101 | + |
| 102 | + var errors nodeKubeletInfos |
| 103 | + var skewedUnsupported nodeKubeletInfos |
| 104 | + var skewedLimit nodeKubeletInfos |
| 105 | + var skewedButOK nodeKubeletInfos |
| 106 | + var synced nodeKubeletInfos |
| 107 | + var unsupported nodeKubeletInfos |
| 108 | + |
| 109 | + // for each node, check kubelet version |
| 110 | + for _, node := range nodes { |
| 111 | + kubeletVersion, err := nodeKubeletVersion(node) |
| 112 | + if err != nil { |
| 113 | + runtime.HandleError(fmt.Errorf("unable to determine kubelet version on node %s: %w", node.Name, err)) |
| 114 | + errors = append(errors, nodeKubeletInfo{node: node.Name, err: err}) |
| 115 | + continue |
| 116 | + } |
| 117 | + skew := int(kubeletVersion.Minor - c.apiServerVersion.Minor) |
| 118 | + // Assume that an OpenShift minor version upgrade also bumps to the next kube minor version. Revisit |
| 119 | + // this in the future if an OpenShift minor version upgrade ever skips or repeats a kube minor version. |
| 120 | + skewNextVersion := skew - 1 |
| 121 | + switch { |
| 122 | + case skew == 0: |
| 123 | + // synced |
| 124 | + synced = append(synced, nodeKubeletInfo{node: node.Name, version: &kubeletVersion}) |
| 125 | + case skew < c.minSupportedSkew: |
| 126 | + // already in an unsupported state |
| 127 | + skewedUnsupported = append(skewedUnsupported, nodeKubeletInfo{node: node.Name, version: &kubeletVersion}) |
| 128 | + case skewNextVersion < c.minSupportedSkewNextVersion: |
| 129 | + // upgrading to next minor version of API server would result in an unsupported config |
| 130 | + skewedLimit = append(skewedLimit, nodeKubeletInfo{node: node.Name, version: &kubeletVersion}) |
| 131 | + case skew < 0: |
| 132 | + // behind, but upgrading to next minor version of API server is supported |
| 133 | + skewedButOK = append(skewedButOK, nodeKubeletInfo{node: node.Name, version: &kubeletVersion}) |
| 134 | + default: |
| 135 | + // kubelet version newer than api server version. possibly in the middle of a rollback. |
| 136 | + unsupported = append(unsupported, nodeKubeletInfo{node: node.Name, version: &kubeletVersion}) |
| 137 | + } |
| 138 | + } |
| 139 | + |
| 140 | + condition := operatorv1.OperatorCondition{Type: KubeletMinorVersionUpgradeableConditionType} |
| 141 | + // use the most "severe" reason to set the condition status |
| 142 | + switch { |
| 143 | + case len(skewedUnsupported) > 0: |
| 144 | + condition.Reason = KubeletMinorVersionUnsupportedReason |
| 145 | + condition.Status = operatorv1.ConditionFalse |
| 146 | + switch len(skewedUnsupported) { |
| 147 | + case 1: |
| 148 | + condition.Message = fmt.Sprintf("Unsupported kubelet minor version (%v) on node %s is too far behind the target API server version (%v).", skewedUnsupported.version(), skewedUnsupported.nodes(), c.apiServerVersion) |
| 149 | + case 2, 3: |
| 150 | + condition.Message = fmt.Sprintf("Unsupported kubelet minor versions on nodes %s are too far behind the target API server version (%v).", skewedUnsupported.nodes(), c.apiServerVersion) |
| 151 | + default: |
| 152 | + condition.Message = fmt.Sprintf("Unsupported kubelet minor versions on %d nodes are too far behind the target API server version (%v).", len(skewedUnsupported), c.apiServerVersion) |
| 153 | + } |
| 154 | + case len(unsupported) > 0: |
| 155 | + condition.Reason = KubeletMinorVersionAheadReason |
| 156 | + condition.Status = operatorv1.ConditionUnknown |
| 157 | + switch len(unsupported) { |
| 158 | + case 1: |
| 159 | + condition.Message = fmt.Sprintf("Unsupported kubelet minor version (%v) on node %s is ahead of the target API server version (%v).", unsupported.version(), unsupported.nodes(), c.apiServerVersion) |
| 160 | + case 2, 3: |
| 161 | + condition.Message = fmt.Sprintf("Unsupported kubelet minor versions on nodes %s are ahead of the target API server version (%v).", unsupported.nodes(), c.apiServerVersion) |
| 162 | + default: |
| 163 | + condition.Message = fmt.Sprintf("Unsupported kubelet minor versions on %d nodes are ahead of the target API server version (%v).", len(unsupported), c.apiServerVersion) |
| 164 | + } |
| 165 | + case len(errors) > 0: |
| 166 | + condition.Reason = KubeletVersionUnknownReason |
| 167 | + condition.Status = operatorv1.ConditionUnknown |
| 168 | + switch len(errors) { |
| 169 | + case 1: |
| 170 | + condition.Message = fmt.Sprintf("Unable to determine the kubelet version on node %s: %v", errors.nodes(), errors.error()) |
| 171 | + case 2, 3: |
| 172 | + condition.Message = fmt.Sprintf("Unable to determine the kubelet version on nodes %s.", errors.nodes()) |
| 173 | + default: |
| 174 | + condition.Message = fmt.Sprintf("Unable to determine the kubelet version on %d nodes.", len(errors)) |
| 175 | + } |
| 176 | + case len(skewedLimit) > 0: |
| 177 | + condition.Reason = KubeletMinorVersionUnsupportedNextUpgradeReason |
| 178 | + condition.Status = operatorv1.ConditionFalse |
| 179 | + switch len(skewedLimit) { |
| 180 | + case 1: |
| 181 | + condition.Message = fmt.Sprintf("Kubelet minor version (%v) on node %s will not be supported in the next OpenShift minor version upgrade.", skewedLimit.version(), skewedLimit.nodes()) |
| 182 | + case 2, 3: |
| 183 | + condition.Message = fmt.Sprintf("Kubelet minor versions on nodes %s will not be supported in the next OpenShift minor version upgrade.", skewedLimit.nodes()) |
| 184 | + default: |
| 185 | + condition.Message = fmt.Sprintf("Kubelet minor versions on %d nodes will not be supported in the next OpenShift minor version upgrade.", len(skewedLimit)) |
| 186 | + } |
| 187 | + case len(skewedButOK) > 0: |
| 188 | + condition.Reason = KubeletMinorVersionSupportedNextUpgradeReason |
| 189 | + condition.Status = operatorv1.ConditionTrue |
| 190 | + switch len(skewedButOK) { |
| 191 | + case 1: |
| 192 | + condition.Message = fmt.Sprintf("Kubelet minor version (%v) on node %s is behind the expected API server version; nevertheless, it will continue to be supported in the next OpenShift minor version upgrade.", skewedButOK.version(), skewedButOK.nodes()) |
| 193 | + case 2, 3: |
| 194 | + condition.Message = fmt.Sprintf("Kubelet minor versions on nodes %s are behind the expected API server version; nevertheless, they will continue to be supported in the next OpenShift minor version upgrade.", skewedButOK.nodes()) |
| 195 | + default: |
| 196 | + condition.Message = fmt.Sprintf("Kubelet minor versions on %d nodes are behind the expected API server version; nevertheless, they will continue to be supported in the next OpenShift minor version upgrade.", len(skewedButOK)) |
| 197 | + } |
| 198 | + default: |
| 199 | + condition.Reason = KubeletMinorVersionSyncedReason |
| 200 | + condition.Status = operatorv1.ConditionTrue |
| 201 | + condition.Message = "Kubelet and API server minor versions are synced." |
| 202 | + } |
| 203 | + |
| 204 | + _, _, err = v1helpers.UpdateStatus(c.operatorClient, v1helpers.UpdateConditionFn(condition)) |
| 205 | + return err |
| 206 | +} |
| 207 | + |
| 208 | +type nodeKubeletInfo struct { |
| 209 | + node string |
| 210 | + version *semver.Version |
| 211 | + err error |
| 212 | +} |
| 213 | + |
| 214 | +type nodeKubeletInfos []nodeKubeletInfo |
| 215 | + |
| 216 | +func (n nodeKubeletInfos) nodes() string { |
| 217 | + var s []string |
| 218 | + for _, i := range n { |
| 219 | + s = append(s, i.node) |
| 220 | + } |
| 221 | + switch len(s) { |
| 222 | + case 0, 1: |
| 223 | + case 2: |
| 224 | + return strings.Join(s, " and ") |
| 225 | + default: |
| 226 | + s[len(s)-1] = "and " + s[len(s)-1] |
| 227 | + } |
| 228 | + return strings.Join(s, ", ") |
| 229 | +} |
| 230 | + |
| 231 | +func (n nodeKubeletInfos) error() error { |
| 232 | + if len(n) > 0 { |
| 233 | + return n[0].err |
| 234 | + } |
| 235 | + return nil |
| 236 | +} |
| 237 | + |
| 238 | +func (n nodeKubeletInfos) version() *semver.Version { |
| 239 | + if len(n) > 0 { |
| 240 | + return n[0].version |
| 241 | + } |
| 242 | + return nil |
| 243 | +} |
| 244 | + |
| 245 | +func nodeKubeletVersion(node *corev1.Node) (semver.Version, error) { |
| 246 | + return semver.Parse(strings.TrimPrefix(node.Status.NodeInfo.KubeletVersion, "v")) |
| 247 | +} |
| 248 | + |
| 249 | +var byNodeRegexp = regexp.MustCompile(`node [^ ]*`) |
| 250 | + |
| 251 | +type byName []*corev1.Node |
| 252 | + |
| 253 | +func (n byName) Len() int { return len(n) } |
| 254 | +func (n byName) Swap(i, j int) { n[i], n[j] = n[j], n[i] } |
| 255 | +func (n byName) Less(i, j int) bool { return strings.Compare(n[i].Name, n[j].Name) < 0 } |
0 commit comments