Refactor CI-Lib Implementation to Support Litmus 3.0

README.md

@@ -35,6 +35,101 @@ Litmus supports CI plugin for the following CI platforms:
   </tr>
 </table>
 
+## Environment Variables
+
+Chaos CI Lib uses standardized environment variables to configure environments, infrastructure, and probes. Below is the comprehensive list of supported environment variables.
+
+### Litmus SDK & Authentication
+
+| Variable | Description | Default | Example |
+|----------|-------------|---------|---------|
+| `LITMUS_ENDPOINT` | Litmus server endpoint URL | `""` | `https://chaos.example.com` |
+| `LITMUS_USERNAME` | Username for Litmus authentication | `""` | `admin` |
+| `LITMUS_PASSWORD` | Password for Litmus authentication | `""` | `litmus` |
+| `LITMUS_PROJECT_ID` | ID of the Litmus project to use | `""` | `project-123` |
+
+### Environment Management Variables
+
+| Variable | Description | Default | Example |
+|----------|-------------|---------|---------|
+| `CREATE_ENV` | Whether to create a new environment | `true` | `false` |
+| `USE_EXISTING_ENV` | Whether to use an existing environment | `false` | `true` |
+| `EXISTING_ENV_ID` | ID of existing environment (required if `USE_EXISTING_ENV=true`) | `""` | `env-123456` |
+| `ENV_NAME` | Name for the new environment | `chaos-ci-env` | `my-k8s-env` |
+| `ENV_TYPE` | Type of environment to create | `NON_PROD` | `PROD` |
+| `ENV_DESCRIPTION` | Description of the environment | `CI Test Environment` | `Production Test Environment` |
+
+### Infrastructure Management Variables
+
+| Variable | Description | Default | Example |
+|----------|-------------|---------|---------|
+| `INSTALL_INFRA` | Whether to install infrastructure | `true` | `false` |
+| `USE_EXISTING_INFRA` | Whether to use existing infrastructure | `false` | `true` |
+| `EXISTING_INFRA_ID` | ID of existing infrastructure (required if `USE_EXISTING_INFRA=true`) | `""` | `infra-123456` |
+| `INFRA_NAME` | Name for the infrastructure | `ci-infra-{expName}` | `my-k8s-infra` |
+| `INFRA_NAMESPACE` | Kubernetes namespace for infrastructure | `litmus` | `chaos-testing` |
+| `INFRA_SCOPE` | Scope of infrastructure | `namespace` | `cluster` |
+| `INFRA_SERVICE_ACCOUNT` | Service account for infrastructure | `litmus` | `chaos-runner` |
+| `INFRA_DESCRIPTION` | Description of infrastructure | `CI Test Infrastructure` | `Production Test Infra` |
+| `INFRA_PLATFORM_NAME` | Platform name | `others` | `gcp` |
+| `INFRA_NS_EXISTS` | Whether namespace already exists | `false` | `true` |
+| `INFRA_SA_EXISTS` | Whether service account already exists | `false` | `true` |
+| `INFRA_SKIP_SSL` | Whether to skip SSL verification | `false` | `true` |
+| `INFRA_NODE_SELECTOR` | Node selector for infrastructure | `""` | `disk=ssd` |
+| `INFRA_TOLERATIONS` | Tolerations for infrastructure | `""` | `key=value:NoSchedule` |
+
+### Probe Management Variables
+
+| Variable | Description | Default | Example |
+|----------|-------------|---------|---------|
+| `LITMUS_CREATE_PROBE` | Whether to create a probe | `false` | `true` |
+| `LITMUS_PROBE_NAME` | Name of the probe | `http-probe` | `http-status-check` |
+| `LITMUS_PROBE_TYPE` | Type of probe | `httpProbe` | `httpProbe` |
+| `LITMUS_PROBE_MODE` | Mode of the probe | `SOT` | `Continuous` |
+| `LITMUS_PROBE_URL` | URL for HTTP probe | `http://localhost:8080/health` | `http://app:8080/health` |
+| `LITMUS_PROBE_TIMEOUT` | Timeout for probe | `30s` | `5s` |
+| `LITMUS_PROBE_INTERVAL` | Interval for probe | `10s` | `5s` |
+| `LITMUS_PROBE_ATTEMPTS` | Number of attempts for probe | `1` | `3` |
+| `LITMUS_PROBE_RESPONSE_CODE` | Expected HTTP response code | `200` | `200` |
+
+### Example Usage
+
+To create a new environment and infrastructure:
+```bash
+# Authentication
+export LITMUS_ENDPOINT="https://chaos.example.com"
+export LITMUS_USERNAME="admin"
+export LITMUS_PASSWORD="litmus"
+export LITMUS_PROJECT_ID="project-123"
+
+# Environment setup
+export CREATE_ENV="true"
+export ENV_NAME="test-environment"
+export ENV_TYPE="NON_PROD"
+
+# Infrastructure setup
+export INSTALL_INFRA="true"
+export INFRA_NAME="test-infra"
+export INFRA_NAMESPACE="chaos-testing"
+export INFRA_SCOPE="namespace"
+
+# Optional probe setup
+export LITMUS_CREATE_PROBE="true"
+export LITMUS_PROBE_NAME="http-status-check"
+export LITMUS_PROBE_TYPE="httpProbe"
+export LITMUS_PROBE_URL="http://app:8080/health"
+export LITMUS_PROBE_RESPONSE_CODE="200"
+```
+
+To use existing environment and infrastructure:
+```bash
+# Set environment variables for existing resources
+export USE_EXISTING_ENV="true"
+export EXISTING_ENV_ID="env-123456"
+export USE_EXISTING_INFRA="true"
+export EXISTING_INFRA_ID="infra-789012"
+```
+
 ## How to get started?
 
 Refer the [LitmusChaos Docs](https://docs.litmuschaos.io) and [Experiment Docs](https://litmuschaos.github.io/litmus/experiments/categories/contents/)

experiments/container-kill_test.go

@@ -1,17 +1,21 @@
 package experiments
 
 import (
+	"fmt"
 	"testing"
+	"time"
 
 	"github.com/litmuschaos/chaos-ci-lib/pkg"
 	"github.com/litmuschaos/chaos-ci-lib/pkg/environment"
-	engine "github.com/litmuschaos/chaos-ci-lib/pkg/generic/container-kill/lib"
-	"github.com/litmuschaos/chaos-ci-lib/pkg/log"
+	"github.com/litmuschaos/chaos-ci-lib/pkg/infrastructure"
 	"github.com/litmuschaos/chaos-ci-lib/pkg/types"
-	"github.com/litmuschaos/chaos-operator/pkg/apis/litmuschaos/v1alpha1"
+	"github.com/litmuschaos/chaos-ci-lib/pkg/workflow"
+	"github.com/litmuschaos/litmus-go-sdk/pkg/sdk"
+	models "github.com/litmuschaos/litmus/chaoscenter/graphql/server/graph/model"
 	. "github.com/onsi/ginkgo"
 	. "github.com/onsi/gomega"
 	_ "k8s.io/client-go/plugin/pkg/client/auth/gcp"
+	"k8s.io/klog"
 )
 
 func TestContainerKill(t *testing.T) {
@@ -22,62 +26,144 @@ func TestContainerKill(t *testing.T) {
 //BDD for running container-kill experiment
 var _ = Describe("BDD of running container-kill experiment", func() {
 
-	Context("Check for container-kill experiment", func() {
+	Context("Check for container-kill experiment via SDK", func() {
+		// Define variables accessible to It and AfterEach
+		var (
+			experimentsDetails types.ExperimentDetails
+			sdkClient          sdk.Client
+			err                error
+		)
 
-		It("Should check for the pod delete experiment", func() {
-
-			experimentsDetails := types.ExperimentDetails{}
-			clients := environment.ClientSets{}
-			chaosEngine := v1alpha1.ChaosEngine{}
-
-			//Getting kubeConfig and Generate ClientSets
-			By("[PreChaos]: Getting kubeconfig and generate clientset")
-			err := clients.GenerateClientSetFromKubeConfig()
-			Expect(err).To(BeNil(), "Unable to Get the kubeconfig, due to {%v}", err)
+		BeforeEach(func() {
+			experimentsDetails = types.ExperimentDetails{}
+			err = nil
 
 			//Fetching all the default ENV
 			By("[PreChaos]: Fetching all default ENVs")
-			log.Infof("[PreReq]: Getting the ENVs for the %v experiment", experimentsDetails.ExperimentName)
+			klog.Infof("[PreReq]: Getting the ENVs for the %v experiment", experimentsDetails.ExperimentName)
 			environment.GetENV(&experimentsDetails, "container-kill", "container-kill-engine")
 
-			// Install RBAC for experiment Execution
-			By("[Prepare]: Prepare and install RBAC")
-			err = pkg.InstallRbac(&experimentsDetails, experimentsDetails.ChaosNamespace)
-			Expect(err).To(BeNil(), "fail to install rbac for the experiment, due to {%v}", err)
-
-			// Install ChaosEngine for experiment Execution
-			By("[Prepare]: Prepare and install ChaosEngine")
-			err = engine.InstallContainerKillEngine(&experimentsDetails, &chaosEngine, clients)
-			Expect(err).To(BeNil(), "fail to install chaosengine, due to {%v}", err)
-
-			//Checking runner pod running state
-			By("[Status]: Runner pod running status check")
-			err = pkg.RunnerPodStatus(&experimentsDetails, chaosEngine.Namespace, clients)
-			if err != nil && chaosEngine.Namespace != experimentsDetails.AppNS {
-				err = pkg.RunnerPodStatus(&experimentsDetails, experimentsDetails.AppNS, clients)
+			// Initialize SDK client
+			By("[PreChaos]: Initializing SDK client")
+			sdkClient, err = environment.GenerateClientSetFromSDK()
+			Expect(err).To(BeNil(), "Unable to generate Litmus SDK client, due to {%v}", err)
+
+			// Setup infrastructure 
+			By("[PreChaos]: Setting up infrastructure")
+			err = infrastructure.SetupInfrastructure(&experimentsDetails, sdkClient)
+			Expect(err).To(BeNil(), "Failed to setup infrastructure, due to {%v}", err)
+
+			// Validate that infrastructure ID is properly set
+			Expect(experimentsDetails.ConnectedInfraID).NotTo(BeEmpty(), "Setup failed: ConnectedInfraID is empty after connection attempt.")
+
+			// Setup probe if configured to do so
+			if experimentsDetails.CreateProbe {
+				By("[PreChaos]: Setting up probe")
+				err = workflow.CreateProbe(&experimentsDetails, sdkClient, experimentsDetails.LitmusProjectID)
+				Expect(err).To(BeNil(), "Failed to create probe, due to {%v}", err)
+				// Validate that probe was created successfully
+				Expect(experimentsDetails.CreatedProbeID).NotTo(BeEmpty(), "Probe creation failed: CreatedProbeID is empty")
 			}
-			Expect(err).To(BeNil(), "Runner pod status check failed, due to {%v}", err)
-
-			//Chaos pod running status check
-			err = pkg.ChaosPodStatus(&experimentsDetails, clients)
-			Expect(err).To(BeNil(), "Chaos pod status check failed, due to {%v}", err)
-
-			//Waiting for chaos pod to get completed
-			//And Print the logs of the chaos pod
-			By("[Status]: Wait for chaos pod completion and then print logs")
-			err = pkg.ChaosPodLogs(&experimentsDetails, clients)
-			Expect(err).To(BeNil(), "Fail to get the experiment chaos pod logs, due to {%v}", err)
-
-			//Checking the chaosresult verdict
-			By("[Verdict]: Checking the chaosresult verdict")
-			err = pkg.ChaosResultVerdict(&experimentsDetails, clients)
-			Expect(err).To(BeNil(), "ChasoResult Verdict check failed, due to {%v}", err)
-
-			//Checking chaosengine verdict
-			By("Checking the Verdict of Chaos Engine")
-			err = pkg.ChaosEngineVerdict(&experimentsDetails, clients)
-			Expect(err).To(BeNil(), "ChaosEngine Verdict check failed, due to {%v}", err)
+		})
 
+		It("Should run the container kill experiment via SDK", func() {
+
+			// Ensure pre-checks passed from BeforeEach
+			Expect(err).To(BeNil(), "Error during BeforeEach setup: %v", err)
+			klog.Info("Executing V3 SDK Path for Experiment")
+
+
+            // 1. Construct Experiment Request
+            By("[SDK Prepare]: Constructing Chaos Experiment Request")
+            experimentName := pkg.GenerateUniqueExperimentName("container-kill")
+            experimentsDetails.ExperimentName = experimentName
+            experimentID := pkg.GenerateExperimentID()
+            experimentRequest, errConstruct := workflow.ConstructContainerKillExperimentRequest(&experimentsDetails, experimentID, experimentName)
+            Expect(errConstruct).To(BeNil(), "Failed to construct experiment request: %v", errConstruct)
+
+			// 2. Create and Run Experiment via SDK
+			By("[SDK Prepare]: Creating and Running Chaos Experiment")
+			createResponse, err := sdkClient.Experiments().Create(experimentsDetails.LitmusProjectID, *experimentRequest)
+            Expect(err).To(BeNil(), "Failed to create experiment via SDK: %v", err)
+			klog.Infof("Created experiment: %s", createResponse)
+
+			// 3. Get the experiment run ID
+            By("[SDK Query]: Polling for experiment run to become available")
+			var experimentRunID string
+			maxRetries := 10
+			found := false
+
+			for i := 0; i < maxRetries; i++ {
+				time.Sleep(3 * time.Second)
+
+				listExperimentRunsReq := models.ListExperimentRunRequest{
+					ExperimentIDs: []*string{&experimentID},
+				}
+
+				runsList, err := sdkClient.Experiments().ListRuns(listExperimentRunsReq)
+				if err != nil {
+					klog.Warningf("Error fetching experiment runs: %v", err)
+					continue
+				}
+
+				klog.Infof("Attempt %d: Found %d experiment runs", i+1, 
+					len(runsList.ExperimentRuns))
+
+				if len(runsList.ExperimentRuns) > 0 {
+					experimentRunID = runsList.ExperimentRuns[0].ExperimentRunID
+					klog.Infof("Found experiment run ID: %s", experimentRunID)
+					found = true
+					break
+				}
+
+				klog.Infof("Retrying after delay...")
+			}
+
+			Expect(found).To(BeTrue(), "No experiment runs found for experiment after %d retries", maxRetries)
+
+			// 4. Poll for Experiment Run Status
+			By("[SDK Status]: Polling for Experiment Run Status")
+			var finalPhase string
+			var pollError error
+			timeout := time.After(time.Duration(experimentsDetails.ExperimentTimeout) * time.Minute)
+			ticker := time.NewTicker(time.Duration(experimentsDetails.ExperimentPollingInterval) * time.Second)
+			defer ticker.Stop()
+
+		pollLoop:
+			for {
+				select {
+				case <-timeout:
+					pollError = fmt.Errorf("timed out waiting for experiment run %s to complete after %d minutes", experimentRunID, experimentsDetails.ExperimentTimeout)
+					klog.Error(pollError)
+					break pollLoop
+				case <-ticker.C:
+					phase, errStatus := sdkClient.Experiments().GetRunPhase(experimentRunID)
+					if errStatus != nil {
+						klog.Errorf("Error fetching experiment run status for %s: %v", experimentRunID, errStatus)
+						continue
+					}
+					klog.Infof("Experiment Run %s current phase: %s", experimentRunID, phase)
+					finalPhases := []string{"Completed", "Completed_With_Error", "Failed", "Error", "Stopped", "Skipped", "Aborted", "Timeout", "Terminated"}
+					if pkg.ContainsString(finalPhases, phase) {
+						finalPhase = phase
+						klog.Infof("Experiment Run %s reached final phase: %s", experimentRunID, phase)
+						break pollLoop
+					}
+				}
+			}
+
+			// 5. Post Validation / Verdict Check
+			By("[SDK Verdict]: Checking Experiment Run Verdict")
+			Expect(pollError).To(BeNil())
+			Expect(finalPhase).NotTo(BeEmpty(), "Final phase should not be empty after polling")
+			Expect(finalPhase).To(Equal("Completed"), fmt.Sprintf("Experiment Run phase should be Completed, but got %s", finalPhase))
+		})
+		// Cleanup using AfterEach
+		AfterEach(func() {
+			// Disconnect infrastructure using the new module
+			By("[CleanUp]: Cleaning up infrastructure")
+			errDisconnect := infrastructure.DisconnectInfrastructure(&experimentsDetails, sdkClient)
+			Expect(errDisconnect).To(BeNil(), "Failed to clean up infrastructure, due to {%v}", errDisconnect)
 		})
 	})
-})
+})