tests: consolidate integration tests on a shared framework and the rlts scenarios

Extract the generic pieces of the snapshot-agent integration harness into
tests/integration/harness — in-cluster client setup, node selection
(TEST_NODE / free-GPU pick), pod lifecycle, exec (with stdin) / HTTP /
VRAM helpers, and attach-by-label for chart-deployed pods. The
snapshot-agent suite keeps its specifics (agent deployment, engine specs,
agentctl call path) and composes the shared Cluster.

Move the orchestrator E2E scenario drivers (RunSingleRLJobScenario,
RunQueuedRLJobsScenario, FakeRLJob, PodFactory, Logger) from
test/e2e/acceleratororchestrator to pkg/accelerator-orchestrator/scenarios
so one implementation serves the rlts CLI, the in-process go tests (which
stay in test/e2e with their fakes), and the integration suite.

The suite gains the orchestrator phase: run.sh installs the official
accelerator-orchestrator chart, labels TEST_NODE with the
samplers/trainers groups, and TestOrchestrator runs the shared scenarios
against it with a new pause pod template (no GPU work). The chart gains a
snapshotAgentPort value; the phase points it at an unused port because a
real agent correctly reports jobs without GPU activity as idle-not-loaded
(stalling lock handoffs) — the agentless mode is what
'rlts test orchestrator' targets on orchestrator-only deployments, and
agent-driven C/R scenarios need real GPU workloads (future work).

run.sh is the single integration entrypoint: --phase all (default) =
standalone + k8s + orchestrator, and --build now also covers the
orchestrator image.

Signed-off-by: Aishu Kamal <aishuk@google.com>

Author: aishukamal

cmd/rlts/cmd/test_e2e.go

@@ -27,7 +27,7 @@ import (
 	"k8s.io/client-go/tools/clientcmd"
 
 	pb "github.com/llm-d-incubation/llm-d-rl-time-slicing/pkg/accelerator-orchestrator/api/v1alpha1"
-	"github.com/llm-d-incubation/llm-d-rl-time-slicing/test/e2e/acceleratororchestrator"
+	"github.com/llm-d-incubation/llm-d-rl-time-slicing/pkg/accelerator-orchestrator/scenarios"
 )
 
 var testCmd = &cobra.Command{
@@ -163,7 +163,7 @@ var orchestratorTestCmd = &cobra.Command{
 
 		// Scenario 1: Single RL Job
 		fmt.Println("--- Running Scenario: Single RL Job ---")
-		err = acceleratororchestrator.RunSingleRLJobScenario(ctx, clientset, client, cliLogger, samplerTemplateKey, trainerTemplateKey)
+		err = scenarios.RunSingleRLJobScenario(ctx, clientset, client, cliLogger, samplerTemplateKey, trainerTemplateKey)
 		scenario1Passed := err == nil
 		if !scenario1Passed {
 			fmt.Printf("[FAIL] Single RL Job Scenario failed: %v\n\n", err)
@@ -173,7 +173,7 @@ var orchestratorTestCmd = &cobra.Command{
 
 		// Scenario 2: Queued RL Jobs
 		fmt.Println("--- Running Scenario: Queued RL Jobs ---")
-		err = acceleratororchestrator.RunQueuedRLJobsScenario(ctx, clientset, client, cliLogger, samplerTemplateKey, trainerTemplateKey)
+		err = scenarios.RunQueuedRLJobsScenario(ctx, clientset, client, cliLogger, samplerTemplateKey, trainerTemplateKey)
 		scenario2Passed := err == nil
 		if !scenario2Passed {
 			fmt.Printf("[FAIL] Queued RL Jobs Scenario failed: %v\n\n", err)
@@ -220,7 +220,7 @@ func init() {
 		"Name of the Kubernetes PodTemplate to use for trainer pods (blank for default pause pod)")
 }
 
-// cliLogger implements acceleratororchestrator.Logger interface to print to stdout.
+// cliLogger implements scenarios.Logger interface to print to stdout.
 type cliLogger struct{}
 
 func (c *cliLogger) Log(args ...interface{}) {

deploy/acceleratororchestrator/templates/deployment.yaml

@@ -20,6 +20,8 @@ spec:
         - name: {{ .Chart.Name }}
           image: "{{ .Values.image.repository }}:{{ .Values.image.tag | default .Chart.AppVersion }}"
           imagePullPolicy: {{ .Values.image.pullPolicy }}
+          args:
+            - "-snapshot-agent-port={{ .Values.snapshotAgentPort }}"
           ports:
             - name: grpc
               containerPort: 50051

deploy/acceleratororchestrator/values.yaml

@@ -9,6 +9,10 @@ service:
   type: ClusterIP
   port: 50051
 
+# Port the orchestrator uses to reach the snapshot agents on group nodes
+# (hostNetwork). Must match the snapshot-agent chart's `port` value.
+snapshotAgentPort: 9001
+
 resources: {}
   # limits:
   #   cpu: 100m

…e/acceleratororchestrator/fake_rl_job.go …or-orchestrator/scenarios/fake_rl_job.gotest/e2e/acceleratororchestrator/fake_rl_job.go renamed to pkg/accelerator-orchestrator/scenarios/fake_rl_job.go test/e2e/acceleratororchestrator/fake_rl_job.go renamed to pkg/accelerator-orchestrator/scenarios/fake_rl_job.go

@@ -1,4 +1,18 @@
-package acceleratororchestrator
+// Copyright 2026 The llm-d Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package scenarios
 
 import (
 	"context"

…e/acceleratororchestrator/pod_factory.go …or-orchestrator/scenarios/pod_factory.gotest/e2e/acceleratororchestrator/pod_factory.go renamed to pkg/accelerator-orchestrator/scenarios/pod_factory.go test/e2e/acceleratororchestrator/pod_factory.go renamed to pkg/accelerator-orchestrator/scenarios/pod_factory.go

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-package acceleratororchestrator
+package scenarios
 
 import (
 	"sync"
@@ -62,6 +62,20 @@ func NewPodFactory() *PodFactory {
 			},
 		},
 	})
+	// "pause" runs no GPU work at all: for exercising the lock/scheduling
+	// protocol on clusters where the GPUs must stay untouched.
+	factory.Register("pause", &corev1.Pod{
+		Spec: corev1.PodSpec{
+			TerminationGracePeriodSeconds: &gracePeriodSec,
+			Containers: []corev1.Container{
+				{
+					Name:            "pause",
+					Image:           "registry.k8s.io/pause:3.10",
+					ImagePullPolicy: corev1.PullIfNotPresent,
+				},
+			},
+		},
+	})
 	factory.Register("vllm", &corev1.Pod{
 		Spec: corev1.PodSpec{
 			Containers: []corev1.Container{

…e2e/acceleratororchestrator/scenarios.go …ator-orchestrator/scenarios/scenarios.gotest/e2e/acceleratororchestrator/scenarios.go renamed to pkg/accelerator-orchestrator/scenarios/scenarios.go test/e2e/acceleratororchestrator/scenarios.go renamed to pkg/accelerator-orchestrator/scenarios/scenarios.go

@@ -12,7 +12,12 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-package acceleratororchestrator
+// Package scenarios contains the accelerator-orchestrator E2E scenario
+// drivers. It is the single implementation shared by the rlts CLI
+// (`rlts test orchestrator`) and the Go test suites: scenarios take a gRPC
+// client and a Kubernetes clientset, report progress through the Logger
+// interface, and return errors rather than printing or exiting.
+package scenarios
 
 import (
 	"context"

test/e2e/acceleratororchestrator/e2e_test.go

@@ -11,6 +11,7 @@ import (
 	pb "github.com/llm-d-incubation/llm-d-rl-time-slicing/pkg/accelerator-orchestrator/api/v1alpha1"
 	"github.com/llm-d-incubation/llm-d-rl-time-slicing/pkg/accelerator-orchestrator/controller"
 	"github.com/llm-d-incubation/llm-d-rl-time-slicing/pkg/accelerator-orchestrator/infrastructure"
+	"github.com/llm-d-incubation/llm-d-rl-time-slicing/pkg/accelerator-orchestrator/scenarios"
 	"github.com/llm-d-incubation/llm-d-rl-time-slicing/pkg/accelerator-orchestrator/server"
 	"github.com/llm-d-incubation/llm-d-rl-time-slicing/pkg/accelerator-orchestrator/store"
 	google_grpc "google.golang.org/grpc"
@@ -152,7 +153,7 @@ func TestE2E_SingleRLJob(t *testing.T) {
 	t.Log("Store initialized with samplers and trainers groups")
 
 	// Run Scenario
-	if err := RunSingleRLJobScenario(ctx, clientset, client, t, "", ""); err != nil {
+	if err := scenarios.RunSingleRLJobScenario(ctx, clientset, client, t, "", ""); err != nil {
 		t.Fatalf("Scenario failed: %v", err)
 	}
 }
@@ -283,7 +284,7 @@ func TestE2E_QueuedRLJobs(t *testing.T) {
 	t.Log("Store initialized with samplers and trainers groups")
 
 	// Run Scenario
-	if err := RunQueuedRLJobsScenario(ctx, clientset, client, t, "", ""); err != nil {
+	if err := scenarios.RunQueuedRLJobsScenario(ctx, clientset, client, t, "", ""); err != nil {
 		t.Fatalf("Scenario failed: %v", err)
 	}
 }