marker/shuffle-cracked
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: push before i try to do replica based scaling

Browse Source
This commit is contained in:
Aditya
2024-07-05 17:52:30 +05:30
parent 8d9409e184
commit 178c03e255
1 changed files with 274 additions and 327 deletions
Download Patch File Download Diff File Expand all files Collapse all files
functions/onprem/orborus/orborus.go
+126 -179
View File
@@ -49,13 +49,11 @@ import (
//"github.com/mackerelio/go-osstat/memory"
//"github.com/shirou/gopsutil/cpu"
corev1 "k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
appsv1 "k8s.io/api/apps/v1"
corev1 "k8s.io/api/core/v1"
rbacv1 "k8s.io/api/rbac/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/intstr"
// int32Ptr
)
// Starts jobs in bulk, so this could be increased
@@ -77,7 +75,6 @@ var isKubernetes = os.Getenv("IS_KUBERNETES")
var kubernetesNamespace = os.Getenv("KUBERNETES_NAMESPACE")
var maxCPUPercent = 90
// var baseimagename = "docker.pkg.github.com/shuffle/shuffle"
// var baseimagename = "ghcr.io/frikky"
// var baseimagename = "shuffle/shuffle"
@@ -233,11 +230,13 @@ func cleanupExistingNodes(ctx context.Context) error {
}
}
log.Printf("[INFO] Cleaned up all pods, services and deployments in namespace %s", kubernetesNamespace)
log.Printf("[INFO] Cleaned up all pods and services in namespace %s. Waiting 10 seconds for cleanup to reflect", kubernetesNamespace)
time.Sleep(10 * time.Second)
return nil
}
serviceListOptions := types.ServiceListOptions{}
services, err := dockercli.ServiceList(
context.Background(),
@@ -687,7 +686,6 @@ func buildEnvVars(envMap map[string]string) []corev1.EnvVar {
return envVars
}
func handleBackendImageDownload(ctx context.Context, images string) error {
// Should use docker to:
// 1. Pull the image & tag it
@@ -702,8 +700,7 @@ func handleBackendImageDownload(ctx context.Context, images string) error {
log.Printf("[DEBUG] Should remove existing image (s): %s", images)
// Remove the image
removeOptions := image.RemoveOptions{
}
removeOptions := image.RemoveOptions{}
for _, image := range strings.Split(images, ",") {
image = strings.TrimSpace(image)
@@ -723,6 +720,117 @@ func handleBackendImageDownload(ctx context.Context, images string) error {
return nil
}
func fixk8sRoles() {
clientset, _, err := shuffle.GetKubernetesClient()
if err != nil {
log.Printf("[ERROR] Error getting kubernetes client: %s", err)
os.Exit(1)
}
kubernetesNamespace := "default"
// Check if namespace exist as variable. If so, make it
if len(os.Getenv("KUBERNETES_NAMESPACE")) > 0 {
kubernetesNamespace = os.Getenv("KUBERNETES_NAMESPACE")
}
// fix roles
// check if "service-creator" role is assigned to the service account "default"
// roleBindingNames := []string{"service-creator-binding", "pod-creator-binding", "deployment-creator-binding"}
serviceAccountName := "default"
roleBindingName := "creator-all"
resourceTypes := []string{"services", "pods", "deployments"}
// Check if the RoleBinding exists
roleBinding, err := clientset.RbacV1().RoleBindings(kubernetesNamespace).Get(context.TODO(), roleBindingName, metav1.GetOptions{})
if err != nil {
log.Printf("[WARNING] Failed to get RoleBinding %s: %s", roleBindingName, err)
// create role and rolebinding
role := &rbacv1.Role{
ObjectMeta: metav1.ObjectMeta{
Name: roleBindingName,
},
Rules: []rbacv1.PolicyRule{
{
APIGroups: []string{""},
Resources: resourceTypes,
Verbs: []string{"create", "list"},
},
},
}
ctx := context.TODO()
_, err := clientset.RbacV1().Roles(kubernetesNamespace).Create(ctx, role, metav1.CreateOptions{})
if err != nil {
log.Printf("[ERROR] Failed to create Role %s: %s", roleBindingName, err)
if !strings.Contains(fmt.Sprintf("%s", err), "already exists") {
log.Printf("[INFO] role %s already exists", roleBindingName)
}
}
roleBinding := &rbacv1.RoleBinding{
ObjectMeta: metav1.ObjectMeta{
Name: roleBindingName,
},
Subjects: []rbacv1.Subject{
{
Kind: "ServiceAccount",
Name: serviceAccountName,
Namespace: kubernetesNamespace,
},
},
RoleRef: rbacv1.RoleRef{
Kind: "Role",
Name: roleBindingName,
},
}
_, err = clientset.RbacV1().RoleBindings(kubernetesNamespace).Create(ctx, roleBinding, metav1.CreateOptions{})
if err != nil {
log.Printf("[ERROR] Failed to create RoleBinding %s: %s", roleBindingName, err)
if strings.Contains(fmt.Sprintf("%s", err), "already exists") {
log.Printf("[INFO] rolebinding %s already exists", roleBindingName)
}
}
log.Printf("[INFO] Created Role %s and RoleBinding %s", roleBindingName, roleBindingName)
} else {
log.Printf("[INFO] RoleBinding %s exists", roleBindingName)
}
// Check if the RoleBinding is assigned to the service account
var found bool
for _, subject := range roleBinding.Subjects {
if subject.Kind == "ServiceAccount" && subject.Name == serviceAccountName {
found = true
break
}
}
if !found {
log.Printf("[WARNING] Service account %s is not assigned to RoleBinding %s\n", serviceAccountName, roleBindingName)
// assign the service account to the rolebinding
roleBinding.Subjects = append(roleBinding.Subjects, rbacv1.Subject{
Kind: "ServiceAccount",
Name: serviceAccountName,
Namespace: kubernetesNamespace,
})
ctx := context.TODO()
_, err := clientset.RbacV1().RoleBindings(kubernetesNamespace).Update(ctx, roleBinding, metav1.UpdateOptions{})
if err != nil {
log.Printf("[ERROR](ns - %s) Failed to update RoleBinding %s: %s", kubernetesNamespace, roleBindingName, err)
if !strings.Contains(fmt.Sprintf("%s", err), "already exists") {
log.Printf("[INFO] rolebinding %s already exists", roleBindingName)
}
}
}
}
func deployK8sWorker(image string, identifier string, env []string) error {
env = append(env, fmt.Sprintf("IS_KUBERNETES=true"))
env = append(env, fmt.Sprintf("KUBERNETES_NAMESPACE=%s", os.Getenv("KUBERNETES_NAMESPACE")))
@@ -759,7 +867,6 @@ func deployK8sWorker(image string, identifier string, env []string) error {
// use k8s downward API to find it if we are in a pod
}
// Check if namespace exist as variable. If so, make it
if len(os.Getenv("KUBERNETES_NAMESPACE")) > 0 && !namespacemade {
kubernetesNamespace = os.Getenv("KUBERNETES_NAMESPACE")
@@ -785,6 +892,7 @@ func deployK8sWorker(image string, identifier string, env []string) error {
env = append(env, fmt.Sprintf("BASE_URL=%s", baseUrl))
env = append(env, fmt.Sprintf("SHUFFLE_SWARM_CONFIG=%s", swarmConfig))
env = append(env, fmt.Sprintf("WORKER_HOSTNAME=%s", "shuffle-workers"))
if len(kubernetesNamespace) == 0 {
foundNamespace, err := shuffle.GetKubernetesNamespace()
@@ -844,7 +952,6 @@ func deployK8sWorker(image string, identifier string, env []string) error {
}
}
// While testing:
// kubectl delete pods --all --all-namespaces; kubectl delete services --all --all-namespaces
// pod := &corev1.Pod{
@@ -874,7 +981,6 @@ func deployK8sWorker(image string, identifier string, env []string) error {
// log.Printf("[ERROR] Failed listing pods: %s", err)
// }
// createdPod, err := clientset.CoreV1().Pods(kubernetesNamespace).Create(context.Background(), pod, metav1.CreateOptions{})
// if err != nil {
// //log.Printf("[ERROR] Failed creating pod: %v", err)
@@ -909,28 +1015,11 @@ func deployK8sWorker(image string, identifier string, env []string) error {
// return err
// }
// return nil
// experimenting with k8s deployments to enable autoscaling.
// Spec: corev1.PodSpec{
// RestartPolicy: "Never",
// // DNSPolicy: "Default",
// DNSPolicy: corev1.DNSClusterFirst,
// // NodeSelector: map[string]string{
// // "node": "master",
// // },
// Containers: []corev1.Container{
// containerAttachment,
// },
// },
deployment := &appsv1.Deployment{
ObjectMeta: metav1.ObjectMeta{
Name: identifier,
},
Spec: appsv1.DeploymentSpec{
Replicas: int32Ptr(1),
Selector: &metav1.LabelSelector{
MatchLabels: containerLabels,
},
@@ -939,13 +1028,10 @@ func deployK8sWorker(image string, identifier string, env []string) error {
Labels: containerLabels,
},
Spec: corev1.PodSpec{
DNSPolicy: corev1.DNSClusterFirst,
// NodeSelector: map[string]string{
// "node": "master",
// },
Containers: []corev1.Container{
containerAttachment,
},
DNSPolicy: corev1.DNSClusterFirst,
},
},
},
@@ -957,27 +1043,7 @@ func deployK8sWorker(image string, identifier string, env []string) error {
return err
}
// // kubectl expose pod shuffle-workers --type=LoadBalancer --port=33333
// service := &corev1.Service{
// ObjectMeta: metav1.ObjectMeta{
// Name: identifier,
// },
// Spec: corev1.ServiceSpec{
// Selector: map[string]string{
// "container": "shuffle-workers",
// },
// Ports: []corev1.ServicePort{
// {
// Protocol: "TCP",
// Port: 33333,
// TargetPort: intstr.FromInt(33333),
// },
// },
// Type: corev1.ServiceTypeLoadBalancer,
// },
// }
// kubectl expose deployment shuffle-workers --type=NodePort --port=33333
// kubectl expose deployment shuffle-workers --type=NodePort --port=33333 --target-port=33333
service := &corev1.Service{
ObjectMeta: metav1.ObjectMeta{
Name: identifier,
@@ -1004,8 +1070,6 @@ func deployK8sWorker(image string, identifier string, env []string) error {
return nil
}
func int32Ptr(i int32) *int32 { return &i }
func deployWorker(image string, identifier string, env []string, executionRequest shuffle.ExecutionRequest) error {
if len(os.Getenv("REGISTRY_URL")) > 0 && os.Getenv("REGISTRY_URL") != "" {
env = append(env, fmt.Sprintf("REGISTRY_URL=%s", os.Getenv("REGISTRY_URL")))
@@ -1407,7 +1471,6 @@ func getOrborusStats(ctx context.Context) shuffle.OrborusStats {
newStats.MaxMemory = int(pers.MemTotal)
}
// Get list of all running containers
containers, err := dockercli.ContainerList(ctx, container.ListOptions{})
@@ -1522,10 +1585,6 @@ func getOrborusStats(ctx context.Context) shuffle.OrborusStats {
return newStats
}
func sendRemoveRequest(client *http.Client, toBeRemoved shuffle.ExecutionRequestWrapper, baseUrl, environment, auth, org string, sleepTime int) error {
confirmUrl := fmt.Sprintf("%s/api/v1/workflows/queue/confirm", baseUrl)
@@ -1604,112 +1663,7 @@ func main() {
}
if isKubernetes == "true" {
clientset, _, err := shuffle.GetKubernetesClient()
if err != nil {
log.Printf("[ERROR] Error getting kubernetes client: %s", err)
os.Exit(1)
}
kubernetesNamespace := "default"
// Check if namespace exist as variable. If so, make it
if len(os.Getenv("KUBERNETES_NAMESPACE")) > 0 && !namespacemade {
kubernetesNamespace = os.Getenv("KUBERNETES_NAMESPACE")
}
// fix roles
// check if "service-creator" role is assigned to the service account "default"
roleBindingName := "service-creator-binding"
serviceAccountName := "default"
// Check if the RoleBinding exists
roleBinding, err := clientset.RbacV1().RoleBindings(kubernetesNamespace).Get(context.TODO(), roleBindingName, metav1.GetOptions{})
if err != nil {
log.Printf("[WARNING] Failed to get RoleBinding %s: %s", roleBindingName, err)
// create role and rolebinding
role := &rbacv1.Role{
ObjectMeta: metav1.ObjectMeta{
Name: roleBindingName,
},
Rules: []rbacv1.PolicyRule{
{
APIGroups: []string{""},
Resources: []string{"services"},
Verbs: []string{"create"},
},
},
}
ctx := context.TODO()
_, err := clientset.RbacV1().Roles(kubernetesNamespace).Create(ctx, role, metav1.CreateOptions{})
if err != nil {
log.Printf("[ERROR] Failed to create Role %s: %s", roleBindingName, err)
if !strings.Contains(fmt.Sprintf("%s", err), "already exists") {
log.Printf("[INFO] role %s already exists", roleBindingName)
}
}
roleBinding := &rbacv1.RoleBinding{
ObjectMeta: metav1.ObjectMeta{
Name: roleBindingName,
},
Subjects: []rbacv1.Subject{
{
Kind: "ServiceAccount",
Name: serviceAccountName,
Namespace: kubernetesNamespace,
},
},
RoleRef: rbacv1.RoleRef{
Kind: "Role",
Name: roleBindingName,
},
}
_, err = clientset.RbacV1().RoleBindings(kubernetesNamespace).Create(ctx, roleBinding, metav1.CreateOptions{})
if err != nil {
log.Printf("[ERROR] Failed to create RoleBinding %s: %s", roleBindingName, err)
if !strings.Contains(fmt.Sprintf("%s", err), "already exists") {
log.Printf("[INFO] rolebinding %s already exists", roleBindingName)
}
}
log.Printf("[INFO] Created Role %s and RoleBinding %s", roleBindingName, roleBindingName)
} else {
log.Printf("[INFO] RoleBinding %s exists", roleBindingName)
}
// Check if the RoleBinding is assigned to the service account
var found bool
for _, subject := range roleBinding.Subjects {
if subject.Kind == "ServiceAccount" && subject.Name == serviceAccountName {
found = true
break
}
}
if !found {
log.Printf("[WARNING] Service account %s is not assigned to RoleBinding %s\n", serviceAccountName, roleBindingName)
// assign the service account to the rolebinding
roleBinding.Subjects = append(roleBinding.Subjects, rbacv1.Subject{
Kind: "ServiceAccount",
Name: serviceAccountName,
Namespace: kubernetesNamespace,
})
ctx := context.TODO()
_, err := clientset.RbacV1().RoleBindings(kubernetesNamespace).Update(ctx, roleBinding, metav1.UpdateOptions{})
if err != nil {
log.Printf("[ERROR] Failed to update RoleBinding %s: %s", roleBindingName, err)
if !strings.Contains(fmt.Sprintf("%s", err), "already exists") {
log.Printf("[INFO] rolebinding %s already exists", roleBindingName)
}
}
}
fixk8sRoles()
}
startupDelay := os.Getenv("SHUFFLE_ORBORUS_STARTUP_DELAY")
@@ -1987,7 +1941,6 @@ func main() {
continue
}
if hasStarted && len(executionRequests.Data) > 0 {
//log.Printf("[INFO] Body: %s", string(body))
// Type string `json:"type"`
@@ -2065,7 +2018,7 @@ func main() {
log.Printf("[WARNING] Throttle - Cutting down requests from %d to %d (MAX: %d, CUR: %d)", len(executionRequests.Data), allowed, maxConcurrency, executionCount)
executionRequests.Data = executionRequests.Data[0:allowed]
}
} else if (swarmControlMode && (swarmConfig == "run" || swarmConfig == "swarm")) {
} else if swarmControlMode && (swarmConfig == "run" || swarmConfig == "swarm") {
if len(executionRequests.Data) > 50 {
executionRequests.Data = executionRequests.Data[0:50]
}
@@ -2214,7 +2167,6 @@ func main() {
}
}
// func deployPipeline(image, identifier, command string) error {
// if isKubernetes == "true" {
// return errors.New("Kubernetes not implemented")
@@ -2239,7 +2191,6 @@ func main() {
// envVariables := []string{
// }
// // Add volume binds for storage
// // Want read/write with full access for the container
// //sourceFolder := "/Users/frikky/git/shuffle/shuffle-database"
@@ -2278,7 +2229,6 @@ func main() {
// "shuffle": "shuffle",
// }
// cont, err := dockercli.ContainerCreate(
// ctx,
// config,
@@ -2364,8 +2314,6 @@ func main() {
// return nil
// }
// Tenzir command samples
// docker pull ghcr.io/dominiklohmann/tenzir-arm64:latest
// docker tag ghcr.io/dominiklohmann/tenzir-arm64:latest tenzir/tenzir:latest
@@ -2376,11 +2324,11 @@ func handlePipeline(incRequest shuffle.ExecutionRequest) error {
if tenzirUrl == "" {
tenzirUrl = "http://localhost:5160"
log.Printf("[WARNING] SHUFFLE_TENZIR_URL not set, falling back to default URL: %s",tenzirUrl)
log.Printf("[WARNING] SHUFFLE_TENZIR_URL not set, falling back to default URL: %s", tenzirUrl)
}
err := deployTenzirNode()
if err != nil{
if err != nil {
log.Printf("[ERROR] failed to deploy the pipeline, reason: %s", err)
return err
}
@@ -2541,7 +2489,7 @@ func deployTenzirNode() error {
func checkTenzirNode() error {
retries := 20
retryInterval := 3 * time.Second
url := fmt.Sprintf("%s/api/v0/ping",tenzirUrl)
url := fmt.Sprintf("%s/api/v0/ping", tenzirUrl)
forwardMethod := "POST"
client := http.Client{}
@@ -2928,7 +2876,6 @@ func getRunningWorkers(ctx context.Context, workerTimeout int) int {
if isKubernetes == "true" {
log.Printf("[INFO] Getting running workers in kubernetes")
thresholdTime := time.Now().Add(time.Duration(-workerTimeout) * time.Second)
clientset, _, err := shuffle.GetKubernetesClient()