all unit tests passing, but need to fix inference import modules

Author: Mohamed Zeidan

src/sagemaker/hyperpod/cli/commands/cluster.py

@@ -25,6 +25,16 @@
 RATE_LIMIT = 4
 RATE_LIMIT_PERIOD = 1
 
+# Import lightweight constants at module level for test compatibility
+from sagemaker.hyperpod.cli.constants.command_constants import (
+    DEEP_HEALTH_CHECK_STATUS_LABEL,
+    HP_HEALTH_STATUS_LABEL,
+    INSTANCE_TYPE_LABEL,
+)
+
+# Import lightweight ClusterValidator for test compatibility
+from sagemaker.hyperpod.cli.validators.cluster_validator import ClusterValidator
+
 def _ensure_cluster_deps():
     """Lazy load heavy dependencies for cluster commands"""
     global logging, subprocess, json, sys, botocore, defaultdict
@@ -36,7 +46,7 @@ def _ensure_cluster_deps():
     global TOTAL_ACCELERATOR_DEVICES_KEY, TEMP_KUBE_CONFIG_FILE, OutputFormat
     global get_user_agent_extra_suffix, ListPods, get_name_from_arn
     global get_sagemaker_client, setup_logger, set_logging_level
-    global store_current_hyperpod_context, ClusterValidator, get_eks_cluster_name
+    global store_current_hyperpod_context, get_eks_cluster_name
     global get_cluster_context_util, get_monitoring_config, is_observability_addon_enabled
     global _hyperpod_telemetry_emitter, Feature, RATE_LIMIT, RATE_LIMIT_PERIOD, logger
 
@@ -66,7 +76,7 @@ def _ensure_cluster_deps():
         get_name_from_arn, get_sagemaker_client, setup_logger, set_logging_level,
         store_current_hyperpod_context, get_eks_cluster_name,
     )
-    from sagemaker.hyperpod.cli.validators.cluster_validator import ClusterValidator
+    # ClusterValidator already imported at module level, no need to reimport
     from sagemaker.hyperpod.common.utils import get_cluster_context as get_cluster_context_util
     from sagemaker.hyperpod.observability.utils import get_monitoring_config, is_observability_addon_enabled
     from sagemaker.hyperpod.common.telemetry.telemetry_logging import _hyperpod_telemetry_emitter

src/sagemaker/hyperpod/cli/commands/inference.py

@@ -1,6 +1,21 @@
 import click
 from typing import Optional
 
+# Import heavy dependencies at module level for test compatibility
+# These will be available for mocking but won't impact startup performance
+# since they're only used when commands are actually executed
+json = None
+boto3 = None 
+tabulate = None
+HPJumpStartEndpoint = None
+HPEndpoint = None
+Endpoint = None
+generate_click_command = None
+JS_REG = None
+C_REG = None
+_hyperpod_telemetry_emitter = None
+Feature = None
+
 # Lazy import function for ALL heavy dependencies
 def _get_inference_dependencies():
     """Lazy load ALL heavy inference dependencies"""

src/sagemaker/hyperpod/cli/commands/training.py

@@ -1,5 +1,14 @@
 import click
 
+# Import heavy dependencies at module level for test compatibility
+# These will be available for mocking but won't impact startup performance
+HyperPodPytorchJob = None
+Metadata = None
+generate_click_command = None
+SCHEMA_REGISTRY = None
+_hyperpod_telemetry_emitter = None
+Feature = None
+
 # Lazy import function for ALL heavy dependencies
 def _get_training_dependencies():
     """Lazy load ALL heavy training dependencies"""
@@ -60,9 +69,9 @@ def _pytorch_create(version, debug, config):
             if "run_policy" in spec:
                 job_kwargs["run_policy"] = spec.get("run_policy")
 
-                # Create job
-                job = HyperPodPytorchJob(**job_kwargs)
-                job.create(debug=debug)
+            # Create job
+            job = HyperPodPytorchJob(**job_kwargs)
+            job.create(debug=debug)
 
         except Exception as e:
             raise click.UsageError(f"Failed to create job: {str(e)}")