[Setup/Standup] fix: Switch to pykube-ng in an attempt to restore CI/CD (llm-d#328)

* [Setup/Standup] fix: Switch to `pykube-ng` in an attempt to restore CI/CD

Also, temporarily disable skip schema validation for step 9

---------

Signed-off-by: maugustosilva <maugusto.silva@gmail.com>

Author: maugustosilva

…eriments/precise_prefix_cache_aware.yaml …eriments/precise-prefix-cache-aware.yamlexperiments/precise_prefix_cache_aware.yaml renamed to experiments/precise-prefix-cache-aware.yaml experiments/precise_prefix_cache_aware.yaml renamed to experiments/precise-prefix-cache-aware.yaml

@@ -825,7 +825,7 @@ def check_affinity():
         return False
 
 
-def add_annotations(varname):
+def add_annotations(varname: str) -> str:
     """
     Generate pod annotations YAML.
     Equivalent to the bash add_annotations function.
@@ -834,6 +834,7 @@ def add_annotations(varname):
     if not annotations:
         return ""
 
+    #FIXME (This should be extracted "ev" dictionary)
     # Determine indentation based on environment type
     standalone_active = int(os.environ.get("LLMDBENCH_CONTROL_ENVIRONMENT_TYPE_STANDALONE_ACTIVE", 0))
     modelservice_active = int(os.environ.get("LLMDBENCH_CONTROL_ENVIRONMENT_TYPE_MODELSERVICE_ACTIVE", 0))
@@ -1172,93 +1173,3 @@ def check_affinity():
     except Exception as e:
         announce(f"❌ Error connecting to Kubernetes: {e}")
         return False
-
-
-def add_annotations():
-    """
-    Generate pod annotations YAML.
-    Equivalent to the bash add_annotations function.
-    """
-    annotations = os.environ.get("LLMDBENCH_VLLM_COMMON_ANNOTATIONS", "")
-    if not annotations:
-        return ""
-
-    # Determine indentation based on environment type
-    standalone_active = int(os.environ.get("LLMDBENCH_CONTROL_ENVIRONMENT_TYPE_STANDALONE_ACTIVE", 0))
-    modelservice_active = int(os.environ.get("LLMDBENCH_CONTROL_ENVIRONMENT_TYPE_MODELSERVICE_ACTIVE", 0))
-
-    if standalone_active == 1:
-        indent = "        "  # 8 spaces
-    elif modelservice_active == 1:
-        indent = "      "    # 6 spaces
-    else:
-        indent = "        "  # default 8 spaces
-
-    # Parse annotations (comma-separated key:value pairs)
-    annotation_lines = []
-    for entry in annotations.split(","):
-        if ":" in entry:
-            key, value = entry.split(":", 1)
-            annotation_lines.append(f"{indent}{key.strip()}: {value.strip()}")
-
-    return "\n".join(annotation_lines)
-
-
-def add_command_line_options(args_string):
-    """
-    Generate command line options for container args.
-    Equivalent to the bash add_command_line_options function.
-    """
-    current_step = os.environ.get("LLMDBENCH_CURRENT_STEP", "")
-
-    if current_step == "06":
-        # For step 06 (standalone), format as YAML list item
-        if args_string:
-            return f"        - |\n          {args_string}"
-        else:
-            return "        - |"
-    elif current_step == "09":
-        # For step 09 (modelservice), different formatting
-        if args_string:
-            return f"      {args_string}"
-        else:
-            return ""
-    else:
-        return args_string or ""
-
-
-def add_additional_env_to_yaml(env_vars_string):
-    """
-    Generate additional environment variables YAML.
-    Equivalent to the bash add_additional_env_to_yaml function.
-    """
-    if not env_vars_string:
-        return ""
-
-    # Determine indentation based on environment type
-    standalone_active = int(os.environ.get("LLMDBENCH_CONTROL_ENVIRONMENT_TYPE_STANDALONE_ACTIVE", 0))
-    modelservice_active = int(os.environ.get("LLMDBENCH_CONTROL_ENVIRONMENT_TYPE_MODELSERVICE_ACTIVE", 0))
-
-    if standalone_active == 1:
-        name_indent = "        "  # 8 spaces
-        value_indent = "          "  # 10 spaces
-    elif modelservice_active == 1:
-        name_indent = "      "    # 6 spaces
-        value_indent = "        "  # 8 spaces
-    else:
-        name_indent = "        "  # default 8 spaces
-        value_indent = "          "  # default 10 spaces
-
-    # Parse environment variables (comma-separated list)
-    env_lines = []
-    for envvar in env_vars_string.split(","):
-        envvar = envvar.strip()
-        if envvar:
-            # Remove LLMDBENCH_VLLM_STANDALONE_ prefix if present
-            clean_name = envvar.replace("LLMDBENCH_VLLM_STANDALONE_", "")
-            env_value = os.environ.get(envvar, "")
-
-            env_lines.append(f"{name_indent}- name: {clean_name}")
-            env_lines.append(f"{value_indent}value: \"{env_value}\"")
-
-    return "\n".join(env_lines)

setup/functions.py

@@ -135,7 +135,7 @@ if ! command -v pip3 &> /dev/null; then
     echo "pip3 installed successfully."
 fi
 
-python_deps="kubernetes pykube kubernetes-asyncio GitPython requests PyYAML"
+python_deps="kubernetes pykube-ng kubernetes-asyncio GitPython requests PyYAML"
 
 for dep in $python_deps; do
     # use pip3 show to check if the package is already installed

setup/install_deps.sh

@@ -250,7 +250,7 @@ def generate_deployment_yaml(ev, model, model_label):
     affinity_key, affinity_value = ev["vllm_common_affinity"].split(":", 1)
 
     # Generate command line options
-    args = add_command_line_options(ev.get("vllm_standalone_args", ""))
+    args = add_command_line_options(ev["vllm_standalone_args"])
 
     # Generate additional environment variables
     additional_env = add_additional_env_to_yaml(ev.get("vllm_common_envvars_to_yaml", ""))

setup/steps/06_deploy_vllm_standalone_models.py

@@ -246,7 +246,7 @@ EOF
     rm -f $LLMDBENCH_CONTROL_WORK_DIR/setup/helm/${LLMDBENCH_VLLM_MODELSERVICE_RELEASE}/${MODEL_NUM}/ms-rules.yaml
 
     announce "🚀 Installing helm chart \"ms-${LLMDBENCH_VLLM_MODELSERVICE_RELEASE}\" via helmfile..."
-    llmdbench_execute_cmd "helmfile --namespace ${LLMDBENCH_VLLM_COMMON_NAMESPACE} --kubeconfig ${LLMDBENCH_CONTROL_WORK_DIR}/environment/context.ctx --selector name=${LLMDBENCH_DEPLOY_CURRENT_MODEL_ID_LABEL}-ms apply -f $LLMDBENCH_CONTROL_WORK_DIR/setup/helm/${LLMDBENCH_VLLM_MODELSERVICE_RELEASE}/helmfile-${MODEL_NUM}.yaml --skip-diff-on-install" ${LLMDBENCH_CONTROL_DRY_RUN} ${LLMDBENCH_CONTROL_VERBOSE}
+    llmdbench_execute_cmd "helmfile --namespace ${LLMDBENCH_VLLM_COMMON_NAMESPACE} --kubeconfig ${LLMDBENCH_CONTROL_WORK_DIR}/environment/context.ctx --selector name=${LLMDBENCH_DEPLOY_CURRENT_MODEL_ID_LABEL}-ms apply -f $LLMDBENCH_CONTROL_WORK_DIR/setup/helm/${LLMDBENCH_VLLM_MODELSERVICE_RELEASE}/helmfile-${MODEL_NUM}.yaml --skip-diff-on-install --skip-schema-validation" ${LLMDBENCH_CONTROL_DRY_RUN} ${LLMDBENCH_CONTROL_VERBOSE}
     announce "✅ ${LLMDBENCH_VLLM_COMMON_NAMESPACE}-${LLMDBENCH_DEPLOY_CURRENT_MODEL_ID_LABEL}-ms helm chart deployed successfully"
 
     if [[ $LLMDBENCH_VLLM_MODELSERVICE_DECODE_REPLICAS -gt 0 ]]; then