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
+257 -310
View File
@@ -49,13 +49,11 @@ import (
//"github.com/mackerelio/go-osstat/memory" //"github.com/mackerelio/go-osstat/memory"
//"github.com/shirou/gopsutil/cpu" //"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" appsv1 "k8s.io/api/apps/v1"
corev1 "k8s.io/api/core/v1"
rbacv1 "k8s.io/api/rbac/v1" rbacv1 "k8s.io/api/rbac/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/intstr" "k8s.io/apimachinery/pkg/util/intstr"
// int32Ptr
) )
// Starts jobs in bulk, so this could be increased // 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 kubernetesNamespace = os.Getenv("KUBERNETES_NAMESPACE")
var maxCPUPercent = 90 var maxCPUPercent = 90
// var baseimagename = "docker.pkg.github.com/shuffle/shuffle" // var baseimagename = "docker.pkg.github.com/shuffle/shuffle"
// var baseimagename = "ghcr.io/frikky" // var baseimagename = "ghcr.io/frikky"
// var baseimagename = "shuffle/shuffle" // var baseimagename = "shuffle/shuffle"
@@ -106,7 +103,7 @@ var memcached = os.Getenv("SHUFFLE_MEMCACHED")
var tenzirUrl = os.Getenv("SHUFFLE_TENZIR_URL") var tenzirUrl = os.Getenv("SHUFFLE_TENZIR_URL")
var executionIds = []string{} var executionIds = []string{}
var namespacemade = false // For K8s var namespacemade = false // For K8s
var dockercli *dockerclient.Client var dockercli *dockerclient.Client
var containerId string var containerId string
@@ -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 return nil
} }
serviceListOptions := types.ServiceListOptions{} serviceListOptions := types.ServiceListOptions{}
services, err := dockercli.ServiceList( services, err := dockercli.ServiceList(
context.Background(), context.Background(),
@@ -687,7 +686,6 @@ func buildEnvVars(envMap map[string]string) []corev1.EnvVar {
return envVars return envVars
} }
func handleBackendImageDownload(ctx context.Context, images string) error { func handleBackendImageDownload(ctx context.Context, images string) error {
// Should use docker to: // Should use docker to:
// 1. Pull the image & tag it // 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) log.Printf("[DEBUG] Should remove existing image (s): %s", images)
// Remove the image // Remove the image
removeOptions := image.RemoveOptions{ removeOptions := image.RemoveOptions{}
}
for _, image := range strings.Split(images, ",") { for _, image := range strings.Split(images, ",") {
image = strings.TrimSpace(image) image = strings.TrimSpace(image)
@@ -723,6 +720,117 @@ func handleBackendImageDownload(ctx context.Context, images string) error {
return nil 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 { func deployK8sWorker(image string, identifier string, env []string) error {
env = append(env, fmt.Sprintf("IS_KUBERNETES=true")) env = append(env, fmt.Sprintf("IS_KUBERNETES=true"))
env = append(env, fmt.Sprintf("KUBERNETES_NAMESPACE=%s", os.Getenv("KUBERNETES_NAMESPACE"))) 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 // use k8s downward API to find it if we are in a pod
} }
// Check if namespace exist as variable. If so, make it // Check if namespace exist as variable. If so, make it
if len(os.Getenv("KUBERNETES_NAMESPACE")) > 0 && !namespacemade { if len(os.Getenv("KUBERNETES_NAMESPACE")) > 0 && !namespacemade {
kubernetesNamespace = os.Getenv("KUBERNETES_NAMESPACE") 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("BASE_URL=%s", baseUrl))
env = append(env, fmt.Sprintf("SHUFFLE_SWARM_CONFIG=%s", swarmConfig)) env = append(env, fmt.Sprintf("SHUFFLE_SWARM_CONFIG=%s", swarmConfig))
env = append(env, fmt.Sprintf("WORKER_HOSTNAME=%s", "shuffle-workers"))
if len(kubernetesNamespace) == 0 { if len(kubernetesNamespace) == 0 {
foundNamespace, err := shuffle.GetKubernetesNamespace() foundNamespace, err := shuffle.GetKubernetesNamespace()
@@ -844,7 +952,6 @@ func deployK8sWorker(image string, identifier string, env []string) error {
} }
} }
// While testing: // While testing:
// kubectl delete pods --all --all-namespaces; kubectl delete services --all --all-namespaces // kubectl delete pods --all --all-namespaces; kubectl delete services --all --all-namespaces
// pod := &corev1.Pod{ // pod := &corev1.Pod{
@@ -874,7 +981,6 @@ func deployK8sWorker(image string, identifier string, env []string) error {
// log.Printf("[ERROR] Failed listing pods: %s", err) // log.Printf("[ERROR] Failed listing pods: %s", err)
// } // }
// createdPod, err := clientset.CoreV1().Pods(kubernetesNamespace).Create(context.Background(), pod, metav1.CreateOptions{}) // createdPod, err := clientset.CoreV1().Pods(kubernetesNamespace).Create(context.Background(), pod, metav1.CreateOptions{})
// if err != nil { // if err != nil {
// //log.Printf("[ERROR] Failed creating pod: %v", err) // //log.Printf("[ERROR] Failed creating pod: %v", err)
@@ -909,28 +1015,11 @@ func deployK8sWorker(image string, identifier string, env []string) error {
// return err // 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{ deployment := &appsv1.Deployment{
ObjectMeta: metav1.ObjectMeta{ ObjectMeta: metav1.ObjectMeta{
Name: identifier, Name: identifier,
}, },
Spec: appsv1.DeploymentSpec{ Spec: appsv1.DeploymentSpec{
Replicas: int32Ptr(1),
Selector: &metav1.LabelSelector{ Selector: &metav1.LabelSelector{
MatchLabels: containerLabels, MatchLabels: containerLabels,
}, },
@@ -939,13 +1028,10 @@ func deployK8sWorker(image string, identifier string, env []string) error {
Labels: containerLabels, Labels: containerLabels,
}, },
Spec: corev1.PodSpec{ Spec: corev1.PodSpec{
DNSPolicy: corev1.DNSClusterFirst,
// NodeSelector: map[string]string{
// "node": "master",
// },
Containers: []corev1.Container{ Containers: []corev1.Container{
containerAttachment, containerAttachment,
}, },
DNSPolicy: corev1.DNSClusterFirst,
}, },
}, },
}, },
@@ -957,27 +1043,7 @@ func deployK8sWorker(image string, identifier string, env []string) error {
return err return err
} }
// // kubectl expose pod shuffle-workers --type=LoadBalancer --port=33333 // kubectl expose deployment shuffle-workers --type=NodePort --port=33333 --target-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
service := &corev1.Service{ service := &corev1.Service{
ObjectMeta: metav1.ObjectMeta{ ObjectMeta: metav1.ObjectMeta{
Name: identifier, Name: identifier,
@@ -986,8 +1052,8 @@ func deployK8sWorker(image string, identifier string, env []string) error {
Selector: containerLabels, Selector: containerLabels,
Ports: []corev1.ServicePort{ Ports: []corev1.ServicePort{
{ {
Protocol: "TCP", Protocol: "TCP",
Port: 33333, Port: 33333,
TargetPort: intstr.FromInt(33333), TargetPort: intstr.FromInt(33333),
}, },
}, },
@@ -1004,8 +1070,6 @@ func deployK8sWorker(image string, identifier string, env []string) error {
return nil return nil
} }
func int32Ptr(i int32) *int32 { return &i }
func deployWorker(image string, identifier string, env []string, executionRequest shuffle.ExecutionRequest) error { func deployWorker(image string, identifier string, env []string, executionRequest shuffle.ExecutionRequest) error {
if len(os.Getenv("REGISTRY_URL")) > 0 && os.Getenv("REGISTRY_URL") != "" { if len(os.Getenv("REGISTRY_URL")) > 0 && os.Getenv("REGISTRY_URL") != "" {
env = append(env, fmt.Sprintf("REGISTRY_URL=%s", os.Getenv("REGISTRY_URL"))) env = append(env, fmt.Sprintf("REGISTRY_URL=%s", os.Getenv("REGISTRY_URL")))
@@ -1407,8 +1471,7 @@ func getOrborusStats(ctx context.Context) shuffle.OrborusStats {
newStats.MaxMemory = int(pers.MemTotal) newStats.MaxMemory = int(pers.MemTotal)
} }
// Get list of all running containers
// Get list of all running containers
containers, err := dockercli.ContainerList(ctx, container.ListOptions{}) containers, err := dockercli.ContainerList(ctx, container.ListOptions{})
if err != nil { if err != nil {
@@ -1522,10 +1585,6 @@ func getOrborusStats(ctx context.Context) shuffle.OrborusStats {
return newStats return newStats
} }
func sendRemoveRequest(client *http.Client, toBeRemoved shuffle.ExecutionRequestWrapper, baseUrl, environment, auth, org string, sleepTime int) error { 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) confirmUrl := fmt.Sprintf("%s/api/v1/workflows/queue/confirm", baseUrl)
@@ -1604,112 +1663,7 @@ func main() {
} }
if isKubernetes == "true" { if isKubernetes == "true" {
clientset, _, err := shuffle.GetKubernetesClient() fixk8sRoles()
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)
}
}
}
} }
startupDelay := os.Getenv("SHUFFLE_ORBORUS_STARTUP_DELAY") startupDelay := os.Getenv("SHUFFLE_ORBORUS_STARTUP_DELAY")
@@ -1987,7 +1941,6 @@ func main() {
continue continue
} }
if hasStarted && len(executionRequests.Data) > 0 { if hasStarted && len(executionRequests.Data) > 0 {
//log.Printf("[INFO] Body: %s", string(body)) //log.Printf("[INFO] Body: %s", string(body))
// Type string `json:"type"` // 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) 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] 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 { if len(executionRequests.Data) > 50 {
executionRequests.Data = executionRequests.Data[0:50] executionRequests.Data = executionRequests.Data[0:50]
} }
@@ -2214,7 +2167,6 @@ func main() {
} }
} }
// func deployPipeline(image, identifier, command string) error { // func deployPipeline(image, identifier, command string) error {
// if isKubernetes == "true" { // if isKubernetes == "true" {
// return errors.New("Kubernetes not implemented") // return errors.New("Kubernetes not implemented")
@@ -2239,7 +2191,6 @@ func main() {
// envVariables := []string{ // envVariables := []string{
// } // }
// // Add volume binds for storage // // Add volume binds for storage
// // Want read/write with full access for the container // // Want read/write with full access for the container
// //sourceFolder := "/Users/frikky/git/shuffle/shuffle-database" // //sourceFolder := "/Users/frikky/git/shuffle/shuffle-database"
@@ -2278,7 +2229,6 @@ func main() {
// "shuffle": "shuffle", // "shuffle": "shuffle",
// } // }
// cont, err := dockercli.ContainerCreate( // cont, err := dockercli.ContainerCreate(
// ctx, // ctx,
// config, // config,
@@ -2364,8 +2314,6 @@ func main() {
// return nil // return nil
// } // }
// Tenzir command samples // Tenzir command samples
// docker pull ghcr.io/dominiklohmann/tenzir-arm64:latest // docker pull ghcr.io/dominiklohmann/tenzir-arm64:latest
// docker tag ghcr.io/dominiklohmann/tenzir-arm64:latest tenzir/tenzir:latest // docker tag ghcr.io/dominiklohmann/tenzir-arm64:latest tenzir/tenzir:latest
@@ -2376,11 +2324,11 @@ func handlePipeline(incRequest shuffle.ExecutionRequest) error {
if tenzirUrl == "" { if tenzirUrl == "" {
tenzirUrl = "http://localhost:5160" 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() err := deployTenzirNode()
if err != nil{ if err != nil {
log.Printf("[ERROR] failed to deploy the pipeline, reason: %s", err) log.Printf("[ERROR] failed to deploy the pipeline, reason: %s", err)
return err return err
} }
@@ -2430,7 +2378,7 @@ func handlePipeline(incRequest shuffle.ExecutionRequest) error {
log.Printf("[INFO] successfully stopped the Pipeline: %s", pipelineId) log.Printf("[INFO] successfully stopped the Pipeline: %s", pipelineId)
} }
} else if incRequest.Type == "PIPELINE_START" { } else if incRequest.Type == "PIPELINE_START" {
log.Printf("[INFO] Should start the pipeline %#v", identifier) log.Printf("[INFO] Should start the pipeline %#v", identifier)
pipelineId, err := searchPipeline(identifier) pipelineId, err := searchPipeline(identifier)
if err != nil { if err != nil {
@@ -2459,157 +2407,157 @@ func handlePipeline(incRequest shuffle.ExecutionRequest) error {
} }
func deployTenzirNode() error { func deployTenzirNode() error {
if isKubernetes == "true" { if isKubernetes == "true" {
return errors.New("kubernetes not implemented") return errors.New("kubernetes not implemented")
} }
ctx := context.Background() ctx := context.Background()
cacheKey := "tenzir-key" cacheKey := "tenzir-key"
imageName := "tenzir/tenzir:latest" imageName := "tenzir/tenzir:latest"
containerName := "tenzir-node" containerName := "tenzir-node"
containerStartOptions := container.StartOptions{} containerStartOptions := container.StartOptions{}
_, err := shuffle.GetCache(ctx, cacheKey) _, err := shuffle.GetCache(ctx, cacheKey)
if err == nil { if err == nil {
return nil return nil
} }
containerInfo, err := dockercli.ContainerInspect(ctx, containerName) containerInfo, err := dockercli.ContainerInspect(ctx, containerName)
if err != nil { if err != nil {
if dockerclient.IsErrNotFound(err) { if dockerclient.IsErrNotFound(err) {
// Check if image exists // Check if image exists
_, _, err := dockercli.ImageInspectWithRaw(ctx, imageName) _, _, err := dockercli.ImageInspectWithRaw(ctx, imageName)
if dockerclient.IsErrNotFound(err) { if dockerclient.IsErrNotFound(err) {
log.Printf("[DEBUG] pulling image %s", imageName) log.Printf("[DEBUG] pulling image %s", imageName)
pullOptions := image.PullOptions{} pullOptions := image.PullOptions{}
out, err := dockercli.ImagePull(ctx, imageName, pullOptions) out, err := dockercli.ImagePull(ctx, imageName, pullOptions)
if err != nil { if err != nil {
log.Printf("[ERROR] Failed to pull the Tenzir image: %s", err) log.Printf("[ERROR] Failed to pull the Tenzir image: %s", err)
return err return err
} }
defer out.Close() defer out.Close()
io.Copy(io.Discard, out) io.Copy(io.Discard, out)
} else if err != nil { } else if err != nil {
return err return err
} }
err = createAndStartTenzirNode(ctx, containerName, imageName, containerStartOptions) err = createAndStartTenzirNode(ctx, containerName, imageName, containerStartOptions)
if err != nil { if err != nil {
return err return err
} }
} else { } else {
return err return err
} }
} else { } else {
if !containerInfo.State.Running { if !containerInfo.State.Running {
log.Printf("[DEBUG] Tenzir Node exists but is not running") log.Printf("[DEBUG] Tenzir Node exists but is not running")
err := dockercli.ContainerStart(ctx, containerName, containerStartOptions) err := dockercli.ContainerStart(ctx, containerName, containerStartOptions)
if err != nil { if err != nil {
log.Printf("[ERROR] Failed to start Tenzir Node container: %v", err) log.Printf("[ERROR] Failed to start Tenzir Node container: %v", err)
return err return err
} }
log.Printf("[INFO] Waiting for Tenzir to become available ...") log.Printf("[INFO] Waiting for Tenzir to become available ...")
err = checkTenzirNode() err = checkTenzirNode()
if err != nil { if err != nil {
return err return err
} }
} }
} }
tenzirStatus := struct { tenzirStatus := struct {
ContainerStatus string `json:"container_status"` ContainerStatus string `json:"container_status"`
}{ }{
ContainerStatus: "running", ContainerStatus: "running",
} }
cacheData, err := json.Marshal(tenzirStatus) cacheData, err := json.Marshal(tenzirStatus)
if err != nil { if err != nil {
log.Printf("[WARNING] Failed marshalling execution: %s", err) log.Printf("[WARNING] Failed marshalling execution: %s", err)
} }
err = shuffle.SetCache(ctx, cacheKey, cacheData, 1) err = shuffle.SetCache(ctx, cacheKey, cacheData, 1)
if err != nil { if err != nil {
log.Printf("[WARNING] Failed updating cache for tenzir: %s", err) log.Printf("[WARNING] Failed updating cache for tenzir: %s", err)
} }
return nil return nil
} }
func checkTenzirNode() error { func checkTenzirNode() error {
retries := 20 retries := 20
retryInterval := 3 * time.Second retryInterval := 3 * time.Second
url := fmt.Sprintf("%s/api/v0/ping",tenzirUrl) url := fmt.Sprintf("%s/api/v0/ping", tenzirUrl)
forwardMethod := "POST" forwardMethod := "POST"
client := http.Client{} client := http.Client{}
req, err := http.NewRequest(forwardMethod, url, nil) req, err := http.NewRequest(forwardMethod, url, nil)
if err != nil { if err != nil {
log.Printf("[ERROR] Failed to create HTTP request: %s", err) log.Printf("[ERROR] Failed to create HTTP request: %s", err)
return err return err
} }
for i := 0; i < retries; i++ { for i := 0; i < retries; i++ {
resp, err := client.Do(req) resp, err := client.Do(req)
if err == nil && resp.StatusCode == http.StatusOK { if err == nil && resp.StatusCode == http.StatusOK {
return nil return nil
} }
time.Sleep(retryInterval) time.Sleep(retryInterval)
} }
return fmt.Errorf("tenzir node is not available") return fmt.Errorf("tenzir node is not available")
} }
func createAndStartTenzirNode(ctx context.Context, containerName, imageName string, containerStartOptions container.StartOptions) error { func createAndStartTenzirNode(ctx context.Context, containerName, imageName string, containerStartOptions container.StartOptions) error {
healthconfig := &container.HealthConfig{ healthconfig := &container.HealthConfig{
Test: []string{"tenzir --connection-timeout=30s --connection-retry-delay=1s 'api /ping'"}, Test: []string{"tenzir --connection-timeout=30s --connection-retry-delay=1s 'api /ping'"},
Interval: 30 * time.Second, Interval: 30 * time.Second,
Retries: 1, Retries: 1,
} }
config := &container.Config{ config := &container.Config{
Cmd: []string{"--commands=web server --mode=dev --bind=0.0.0.0"}, Cmd: []string{"--commands=web server --mode=dev --bind=0.0.0.0"},
Image: imageName, Image: imageName,
Healthcheck: healthconfig, Healthcheck: healthconfig,
ExposedPorts: nat.PortSet{"5160/tcp": struct{}{}}, ExposedPorts: nat.PortSet{"5160/tcp": struct{}{}},
Entrypoint: []string{containerName}, Entrypoint: []string{containerName},
} }
hostConfig := &container.HostConfig{ hostConfig := &container.HostConfig{
PortBindings: nat.PortMap{ PortBindings: nat.PortMap{
"5160/tcp": []nat.PortBinding{{HostPort: "5160"}}, "5160/tcp": []nat.PortBinding{{HostPort: "5160"}},
}, },
Mounts: []mount.Mount{ Mounts: []mount.Mount{
{ {
Type: mount.TypeVolume, Type: mount.TypeVolume,
Source: containerName, Source: containerName,
Target: "/var/lib/tenzir/", Target: "/var/lib/tenzir/",
}, },
}, },
VolumeDriver: "local", VolumeDriver: "local",
} }
_, err := dockercli.ContainerCreate(ctx, config, hostConfig, nil, nil, containerName) _, err := dockercli.ContainerCreate(ctx, config, hostConfig, nil, nil, containerName)
if err != nil { if err != nil {
return err return err
} }
err = dockercli.ContainerStart(ctx, containerName, containerStartOptions) err = dockercli.ContainerStart(ctx, containerName, containerStartOptions)
if err != nil { if err != nil {
log.Printf("[ERROR] Failed to start Tenzir Node container: %v", err) log.Printf("[ERROR] Failed to start Tenzir Node container: %v", err)
return err return err
} }
log.Printf("[INFO] Tenzir Node container started successfully") log.Printf("[INFO] Tenzir Node container started successfully")
log.Printf("[INFO] Waiting for Tenzir to become available ...") log.Printf("[INFO] Waiting for Tenzir to become available ...")
err = checkTenzirNode() err = checkTenzirNode()
if err != nil { if err != nil {
return err return err
} }
log.Printf("[INFO] Successfully deployed Tenzir Node !") log.Printf("[INFO] Successfully deployed Tenzir Node !")
return nil return nil
} }
func createPipeline(command, identifier string) (string, error) { func createPipeline(command, identifier string) (string, error) {
@@ -2617,7 +2565,7 @@ func createPipeline(command, identifier string) (string, error) {
toBeDeleted := false toBeDeleted := false
pipelineId, err := searchPipeline(identifier) pipelineId, err := searchPipeline(identifier)
url := fmt.Sprintf("%s/api/v0/pipeline/create", tenzirUrl) url := fmt.Sprintf("%s/api/v0/pipeline/create", tenzirUrl)
forwardMethod := "POST" forwardMethod := "POST"
if err != nil { if err != nil {
@@ -2630,7 +2578,7 @@ func createPipeline(command, identifier string) (string, error) {
log.Printf("[INFO] an existing pipeline found with ID: %s. it will be deleted", pipelineId) log.Printf("[INFO] an existing pipeline found with ID: %s. it will be deleted", pipelineId)
toBeDeleted = true toBeDeleted = true
} }
if strings.Contains(command, "shuffler.io") { if strings.Contains(command, "shuffler.io") {
} else { } else {
var scheme string var scheme string
@@ -2723,7 +2671,7 @@ func createPipeline(command, identifier string) (string, error) {
func updatePipelineState(pipelineId, action string) (string, error) { func updatePipelineState(pipelineId, action string) (string, error) {
url := fmt.Sprintf("%s/api/v0/pipeline/update", tenzirUrl) url := fmt.Sprintf("%s/api/v0/pipeline/update", tenzirUrl)
forwardMethod := "POST" forwardMethod := "POST"
requestBody := map[string]interface{}{ requestBody := map[string]interface{}{
@@ -2792,7 +2740,7 @@ func deletePipeline(pipelineId string) error {
"id": pipelineId, "id": pipelineId,
} }
url := fmt.Sprintf("%s/api/v0/pipeline/delete", tenzirUrl) url := fmt.Sprintf("%s/api/v0/pipeline/delete", tenzirUrl)
forwardMethod := "POST" forwardMethod := "POST"
requestBodyJSON, err := json.Marshal(requestBody) requestBodyJSON, err := json.Marshal(requestBody)
@@ -2840,7 +2788,7 @@ func searchPipeline(identifier string) (string, error) {
var reqBody []byte var reqBody []byte
url := fmt.Sprintf("%s/api/v0/pipeline/list", tenzirUrl) url := fmt.Sprintf("%s/api/v0/pipeline/list", tenzirUrl)
resp, err := http.Post(url, "application/json", bytes.NewBuffer(reqBody)) resp, err := http.Post(url, "application/json", bytes.NewBuffer(reqBody))
if err != nil { if err != nil {
@@ -2925,10 +2873,9 @@ func searchPipeline(identifier string) (string, error) {
func getRunningWorkers(ctx context.Context, workerTimeout int) int { func getRunningWorkers(ctx context.Context, workerTimeout int) int {
//log.Printf("[DEBUG] Getting running workers with API version %s", dockerApiVersion) //log.Printf("[DEBUG] Getting running workers with API version %s", dockerApiVersion)
counter := 0 counter := 0
if isKubernetes == "true" { if isKubernetes == "true" {
log.Printf("[INFO] Getting running workers in kubernetes") log.Printf("[INFO] Getting running workers in kubernetes")
thresholdTime := time.Now().Add(time.Duration(-workerTimeout) * time.Second) thresholdTime := time.Now().Add(time.Duration(-workerTimeout) * time.Second)
clientset, _, err := shuffle.GetKubernetesClient() clientset, _, err := shuffle.GetKubernetesClient()