37 Star 411 Fork 76

GVPrancher/rancher

加入 Gitee
与超过 1200万 开发者一起发现、参与优秀开源项目,私有仓库也完全免费 :)
免费加入
文件
克隆/下载
operatorHandler.go 7.72 KB
一键复制 编辑 原始数据 按行查看 历史
package monitoring
import (
"fmt"
"reflect"
"strings"
"github.com/pkg/errors"
"github.com/rancher/rancher/pkg/monitoring"
"github.com/rancher/rancher/pkg/settings"
mgmtv3 "github.com/rancher/types/apis/management.cattle.io/v3"
projectv3 "github.com/rancher/types/apis/project.cattle.io/v3"
"github.com/sirupsen/logrus"
apierrors "k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/runtime"
)
type operatorHandler struct {
clusterName string
cattleClusterClient mgmtv3.ClusterInterface
app *appHandler
}
func (h *operatorHandler) syncCluster(key string, obj *mgmtv3.Cluster) (runtime.Object, error) {
if obj == nil || obj.DeletionTimestamp != nil || obj.Name != h.clusterName {
return obj, nil
}
if !mgmtv3.ClusterConditionAgentDeployed.IsTrue(obj) {
return obj, nil
}
var newCluster *mgmtv3.Cluster
var err error
//should deploy
if obj.Spec.EnableClusterAlerting || obj.Spec.EnableClusterMonitoring {
newObj, err := mgmtv3.ClusterConditionPrometheusOperatorDeployed.Do(obj, func() (runtime.Object, error) {
cpy := obj.DeepCopy()
return cpy, deploySystemMonitor(cpy, h.app)
})
if err != nil {
logrus.Warnf("deploy prometheus operator error, %v", err)
}
newCluster = newObj.(*mgmtv3.Cluster)
} else { // should withdraw
newCluster = obj.DeepCopy()
if err = withdrawSystemMonitor(newCluster, h.app); err != nil {
logrus.Warnf("withdraw prometheus operator error, %v", err)
}
}
if newCluster != nil && !reflect.DeepEqual(newCluster, obj) {
if newCluster, err = h.cattleClusterClient.Update(newCluster); err != nil {
return nil, err
}
return newCluster, nil
}
return obj, nil
}
func (h *operatorHandler) syncProject(key string, project *mgmtv3.Project) (runtime.Object, error) {
if project == nil || project.DeletionTimestamp != nil || project.Spec.ClusterName != h.clusterName {
return project, nil
}
clusterID := project.Spec.ClusterName
cluster, err := h.cattleClusterClient.Get(clusterID, metav1.GetOptions{})
if err != nil {
return nil, errors.Wrapf(err, "failed to find Cluster %s", clusterID)
}
if !mgmtv3.ClusterConditionAgentDeployed.IsTrue(cluster) {
return project, nil
}
var newCluster *mgmtv3.Cluster
//should deploy
if cluster.Spec.EnableClusterAlerting || project.Spec.EnableProjectMonitoring {
newObj, err := mgmtv3.ClusterConditionPrometheusOperatorDeployed.Do(cluster, func() (runtime.Object, error) {
cpy := cluster.DeepCopy()
return cpy, deploySystemMonitor(cpy, h.app)
})
if err != nil {
logrus.Warnf("deploy prometheus operator error, %v", err)
}
newCluster = newObj.(*mgmtv3.Cluster)
} else { // should withdraw
newCluster = cluster.DeepCopy()
if err = withdrawSystemMonitor(newCluster, h.app); err != nil {
logrus.Warnf("withdraw prometheus operator error, %v", err)
}
}
if newCluster != nil && !reflect.DeepEqual(newCluster, cluster) {
if _, err = h.cattleClusterClient.Update(newCluster); err != nil {
return nil, err
}
}
return project, nil
}
func withdrawSystemMonitor(cluster *mgmtv3.Cluster, app *appHandler) error {
isAlertingDisabling := mgmtv3.ClusterConditionAlertingEnabled.IsFalse(cluster) ||
mgmtv3.ClusterConditionAlertingEnabled.GetStatus(cluster) == ""
isClusterMonitoringDisabling := mgmtv3.ClusterConditionMonitoringEnabled.IsFalse(cluster) ||
mgmtv3.ClusterConditionMonitoringEnabled.GetStatus(cluster) == ""
//status false and empty should withdraw. when status unknown, it means the deployment has error while deploying apps
isOperatorDeploying := !mgmtv3.ClusterConditionPrometheusOperatorDeployed.IsFalse(cluster)
areAllOwnedProjectMonitoringDisabling, err := allOwnedProjectsMonitoringDisabling(app.cattleProjectClient)
if err != nil {
mgmtv3.ClusterConditionPrometheusOperatorDeployed.Unknown(cluster)
mgmtv3.ClusterConditionPrometheusOperatorDeployed.ReasonAndMessageFromError(cluster, errors.Wrap(err, "failed to list owned projects of cluster"))
return err
}
if areAllOwnedProjectMonitoringDisabling && isAlertingDisabling && isClusterMonitoringDisabling && isOperatorDeploying {
appName, appTargetNamespace := monitoring.SystemMonitoringInfo()
if err := monitoring.WithdrawApp(app.cattleAppClient, monitoring.OwnedAppListOptions(cluster.Name, appName, appTargetNamespace)); err != nil {
mgmtv3.ClusterConditionPrometheusOperatorDeployed.Unknown(cluster)
mgmtv3.ClusterConditionPrometheusOperatorDeployed.ReasonAndMessageFromError(cluster, errors.Wrap(err, "failed to withdraw prometheus operator app"))
return err
}
mgmtv3.ClusterConditionPrometheusOperatorDeployed.False(cluster)
mgmtv3.ClusterConditionPrometheusOperatorDeployed.Reason(cluster, "")
mgmtv3.ClusterConditionPrometheusOperatorDeployed.Message(cluster, "")
}
return nil
}
func allOwnedProjectsMonitoringDisabling(projectClient mgmtv3.ProjectInterface) (bool, error) {
ownedProjectList, err := projectClient.List(metav1.ListOptions{})
if err != nil {
return false, err
}
for _, ownedProject := range ownedProjectList.Items {
if ownedProject.Spec.EnableProjectMonitoring {
return false, nil
}
}
return true, nil
}
func deploySystemMonitor(cluster *mgmtv3.Cluster, app *appHandler) (backErr error) {
appName, appTargetNamespace := monitoring.SystemMonitoringInfo()
appCatalogID := settings.SystemMonitoringCatalogID.Get()
appDeployProjectID, err := monitoring.GetSystemProjectID(app.cattleProjectClient)
if err != nil {
return errors.Wrap(err, "failed to get System Project ID")
}
appProjectName, err := monitoring.EnsureAppProjectName(app.agentNamespaceClient, appDeployProjectID, cluster.Name, appTargetNamespace)
if err != nil {
return errors.Wrap(err, "failed to ensure monitoring project name")
}
appAnswers := map[string]string{
"enabled": "true",
"apiGroup": monitoring.APIVersion.Group,
"nameOverride": "prometheus-operator",
}
mustAppAnswers := map[string]string{
"operator.apiGroup": monitoring.APIVersion.Group,
"operator.nameOverride": "prometheus-operator",
}
// take operator answers from overwrite answers
for ansKey, ansVal := range monitoring.GetOverwroteAppAnswers(cluster.Annotations) {
if strings.HasPrefix(ansKey, "operator.") {
appAnswers[ansKey] = ansVal
}
}
// cannot overwrite mustAppAnswers
for mustKey, mustVal := range mustAppAnswers {
appAnswers[mustKey] = mustVal
}
creator, err := app.systemAccountManager.GetSystemUser(cluster.Name)
if err != nil {
return err
}
annotations := map[string]string{
"cluster.cattle.io/addon": appName,
creatorIDAnno: creator.Name,
}
targetApp := &projectv3.App{
ObjectMeta: metav1.ObjectMeta{
Annotations: annotations,
Labels: monitoring.OwnedLabels(appName, appTargetNamespace, appProjectName, monitoring.SystemLevel),
Name: appName,
Namespace: appDeployProjectID,
},
Spec: projectv3.AppSpec{
Answers: appAnswers,
Description: "Prometheus Operator for Rancher Monitoring",
ExternalID: appCatalogID,
ProjectName: appProjectName,
TargetNamespace: appTargetNamespace,
},
}
// redeploy operator App forcibly if cannot find the workload
var forceRedeploy bool
appWorkload, err := app.agentDeploymentClient.GetNamespaced(appTargetNamespace, fmt.Sprintf("prometheus-operator-%s", appName), metav1.GetOptions{})
if err != nil && !apierrors.IsNotFound(err) {
return errors.Wrapf(err, "failed to get deployment %s/prometheus-operator-%s", appTargetNamespace, appName)
}
if appWorkload == nil || appWorkload.Name == "" || appWorkload.DeletionTimestamp != nil {
forceRedeploy = true
}
_, err = monitoring.DeployApp(app.cattleAppClient, appDeployProjectID, targetApp, forceRedeploy)
if err != nil {
return errors.Wrap(err, "failed to ensure prometheus operator app")
}
return nil
}
Loading...
马建仓 AI 助手
尝试更多
代码解读
代码找茬
代码优化
Go
1
https://gitee.com/rancher/rancher.git
git@gitee.com:rancher/rancher.git
rancher
rancher
rancher
v2.2.5-rc9

搜索帮助

0d507c66 1850385 C8b1a773 1850385