[inference] Various helpers

projects/caliper/engine/parameter_matrix.py

@@ -237,8 +237,4 @@ def create_legend_name(
 
     legend_name = ", ".join(param_pairs)
 
-    # Truncate if too long
-    if len(legend_name) > max_length:
-        legend_name = legend_name[: max_length - 3] + "..."
-
     return legend_name

projects/guidellm/toolbox/run_guidellm_benchmark/main.py

@@ -120,19 +120,12 @@ def _best_effort_delete(description: str, *oc_args: str) -> None:
 
 @task
 def create_guidellm_resources_task(args, ctx):
-    """Create the GuideLLM benchmark PVC and job"""
+    """Create the GuideLLM benchmark job and PVC with job as owner"""
 
     # Ensure src directory exists
     (args.artifact_dir / "src").mkdir(parents=True, exist_ok=True)
 
-    oc_apply(
-        args.artifact_dir / "src" / "guidellm-pvc.yaml",
-        render_guidellm_pvc_from_parts(
-            namespace=ctx.target_namespace,
-            name=ctx.benchmark_name,
-            pvc_size=args.pvc_size,
-        ),
-    )
+    # Create the job first
     oc_apply(
         args.artifact_dir / "src" / "guidellm-job.yaml",
         render_guidellm_job_from_parts(
@@ -144,10 +137,35 @@ def create_guidellm_resources_task(args, ctx):
             hf_token_secret=args.hf_token_secret,
         ),
     )
-    return f"GuideLLM benchmark {ctx.benchmark_name} created"
+
+    # Get the job metadata for owner reference
+    job_data = oc_get_json("job", name=ctx.benchmark_name, namespace=ctx.target_namespace)
+
+    # Create owner reference from job metadata
+    owner_reference = {
+        "apiVersion": "batch/v1",
+        "kind": "Job",
+        "name": job_data["metadata"]["name"],
+        "uid": job_data["metadata"]["uid"],
+        "controller": True,
+        "blockOwnerDeletion": True,
+    }
+
+    # Create the PVC with job as owner
+    oc_apply(
+        args.artifact_dir / "src" / "guidellm-pvc.yaml",
+        render_guidellm_pvc_from_parts(
+            namespace=ctx.target_namespace,
+            name=ctx.benchmark_name,
+            pvc_size=args.pvc_size,
+            owner_reference=owner_reference,
+        ),
+    )
+
+    return f"GuideLLM benchmark {ctx.benchmark_name} created with job as PVC owner"
 
 
-@retry(attempts=180, delay=10, backoff=1.0)
+@retry(attempts=1080, delay=10, backoff=1.0)
 @task
 def wait_guidellm_benchmark_task(args, ctx):
     """Wait for the GuideLLM benchmark job to complete"""

projects/guidellm/toolbox/run_guidellm_benchmark/utils.py

@@ -146,13 +146,16 @@ def _build_multi_run_script(*, endpoint_url: str, runs: list[GuideLLMRun]) -> st
     return "\n".join(lines)
 
 
-def render_guidellm_pvc_from_parts(*, namespace: str, name: str, pvc_size: str) -> dict[str, Any]:
+def render_guidellm_pvc_from_parts(
+    *, namespace: str, name: str, pvc_size: str, owner_reference: dict[str, Any] | None = None
+) -> dict[str, Any]:
     """Render a GuideLL-M PVC manifest from individual components.
 
     Args:
         namespace: Target namespace
         name: Name of the benchmark job and PVC
         pvc_size: Size of the PVC
+        owner_reference: Optional owner reference to set (e.g., for job ownership)
 
     Returns:
         PVC manifest as dict
@@ -165,7 +168,13 @@ def render_guidellm_pvc_from_parts(*, namespace: str, name: str, pvc_size: str)
             "pvc_size": pvc_size,
         },
     )
-    return yaml.safe_load(rendered_yaml)
+    manifest = yaml.safe_load(rendered_yaml)
+
+    # Add owner reference if provided
+    if owner_reference:
+        manifest["metadata"]["ownerReferences"] = [owner_reference]
+
+    return manifest
 
 
 def render_guidellm_job_from_parts(

projects/guidellm/toolbox/run_smoke_request/main.py

@@ -247,6 +247,7 @@ def capture_smoke_pod_debug_info(args, ctx):
             "-o",
             "yaml",
             check=False,
+            stdout_dest=artifacts_dir / f"{ctx.pod_name}.yaml",
         )
 
         # Capture pod description
@@ -283,6 +284,7 @@ def cleanup_smoke_pod(args, ctx):
         "-n",
         args.namespace,
         "--ignore-not-found=true",
+        "--wait=false",
         check=False,
     )
     return f"Cleaned up smoke pod {ctx.pod_name}"

projects/kserve/toolbox/capture_llmisvc_state/main.py

@@ -26,6 +26,7 @@ def setup_directories(args, context):
     """Create the artifacts directory"""
 
     shell.mkdir("artifacts")
+    shell.mkdir("artifacts/logs")
     return "Artifacts directory created"
 
 
@@ -121,7 +122,7 @@ def capture_namespace_pods(args, context):
     """Capture all pods in the namespace with wide output"""
     shell.run(
         f"oc get pods -owide -n {context.target_namespace}",
-        stdout_dest=args.artifact_dir / "artifacts/namespace.pods.status",
+        stdout_dest=args.artifact_dir / "artifacts/namespace.pods.status.txt",
         check=False,
     )
     return "Namespace pods status captured"
@@ -132,7 +133,7 @@ def capture_namespace_services(args, context):
     """Capture all services in the namespace"""
     shell.run(
         f"oc get svc -n {context.target_namespace}",
-        stdout_dest=args.artifact_dir / "artifacts/namespace.services.status",
+        stdout_dest=args.artifact_dir / "artifacts/namespace.services.status.txt",
         check=False,
     )
     return "Namespace services captured"
@@ -173,20 +174,19 @@ def capture_pod_logs(args, context):
     if not pod_names or not result.stdout.strip():
         return "No pods found to capture logs"
 
-    log_file = args.artifact_dir / "artifacts/llminferenceservice.pods.logs"
+    logs_dir = args.artifact_dir / "artifacts/logs"
+    captured_count = 0
 
-    with open(log_file, "w") as handle:
-        for pod_name in pod_names:
-            handle.write(f"=== Logs for pod: {pod_name} ===\n")
-            log_result = shell.run(
-                f"oc logs {pod_name} -n {context.target_namespace} --all-containers=true",
-                check=False,
-                log_stdout=False,
-            )
-            handle.write(log_result.stdout)
-            handle.write("\n")
+    for pod_name in pod_names:
+        log_file = logs_dir / f"{pod_name}.log"
+        shell.run(
+            f"oc logs {pod_name} -n {context.target_namespace} --all-containers=true",
+            stdout_dest=log_file,
+            check=False,
+        )
+        captured_count += 1
 
-    return f"Pod logs captured for {len(pod_names)} pods"
+    return f"Pod logs captured for {captured_count} pods in dedicated files"
 
 
 @task
@@ -195,26 +195,26 @@ def capture_pod_previous_logs(args, context):
     result = shell.run(
         f'oc get pods -l "app.kubernetes.io/name={args.llmisvc_name}" -n {context.target_namespace} -o jsonpath="{{.items[*].metadata.name}}"',
         check=False,
+        log_stdout=False,
     )
 
     pod_names = result.stdout.strip().split()
     if not pod_names or not result.stdout.strip():
         return "No pods found to capture previous logs"
 
-    log_file = args.artifact_dir / "artifacts/llminferenceservice.pods.previous.logs"
+    logs_dir = args.artifact_dir / "artifacts/logs"
+    captured_count = 0
 
-    with open(log_file, "w") as handle:
-        for pod_name in pod_names:
-            handle.write(f"=== Previous logs for pod: {pod_name} ===\n")
-            log_result = shell.run(
-                f"oc logs {pod_name} -n {context.target_namespace} --previous --all-containers=true",
-                check=False,
-                log_stdout=False,
-            )
-            handle.write(log_result.stdout)
-            handle.write("\n")
+    for pod_name in pod_names:
+        log_file = logs_dir / f"{pod_name}.previous.log"
+        shell.run(
+            f"oc logs {pod_name} -n {context.target_namespace} --previous --all-containers=true",
+            stdout_dest=log_file,
+            check=False,
+        )
+        captured_count += 1
 
-    return f"Pod previous logs captured for {len(pod_names)} pods"
+    return f"Pod previous logs captured for {captured_count} pods in dedicated files"
 
 
 @task

projects/kserve/toolbox/deploy_llmisvc/__init__.py

@@ -0,0 +1,5 @@
+from __future__ import annotations
+
+from projects.kserve.toolbox.deploy_llmisvc.main import try_resolve_endpoint_url
+
+__all__ = ["try_resolve_endpoint_url"]