Merge Pub/q1 dev into staging

common/library/module_utils/input_validation/common_utils/data_validation.py

@@ -58,6 +58,13 @@ def schema(config):
             error_bucket.append("input data reading failed.")
             return error_bucket
 
+        # Normalize case-sensitive fields for omnia_config.yml
+        if "omnia_config" in input_file_path:
+            if "slurm_cluster" in input_data:
+                for cluster in input_data["slurm_cluster"]:
+                    if "node_discovery_mode" in cluster and isinstance(cluster["node_discovery_mode"], str):
+                        cluster["node_discovery_mode"] = cluster["node_discovery_mode"].lower()
+
         # Load schema
         with open(schema_file_path, "r", encoding="utf-8") as schema_file:
             j_schema = json.load(schema_file)

common/library/module_utils/input_validation/common_utils/en_us_validation_msg.py

@@ -166,6 +166,10 @@
 LDMS_REQUIRES_SLURM_MSG = (
     "requires Slurm package 'slurm_custom' to be present in the 'softwares' list in software_config.json."
 )
+USER_REPO_NAME_PREFIX_FAIL_MSG = (
+    "Repository name '{repo_name}' in {repo_key} must start with '{expected_prefix}'. "
+    "Please update the name to '{expected_prefix}{repo_name}'."
+)
 
 # omnia_config.yml
 INVALID_PASSWORD_MSG = ("Provided password is invalid. Password must meet the specified "

common/library/module_utils/input_validation/schema/omnia_config.json

@@ -23,6 +23,50 @@
             "type": "boolean", 
             "description": "Variable indicates whether a specific configuration file path under config_sources should be used as-is without merging" 
           },
+          "node_discovery_mode": {
+            "type": "string",
+            "enum": ["homogeneous", "heterogeneous"],
+            "description": "Node hardware discovery mode. 'homogeneous' for group-based discovery, 'heterogeneous' for individual node discovery. Default: heterogeneous"
+          },
+          "node_hardware_defaults": {
+            "type": "object",
+            "description": "Hardware specifications for homogeneous node groups. Key is group name (grp0-grp100), value is hardware specs.",
+            "patternProperties": {
+              "^grp([0-9]|[1-9][0-9]|100)$": {
+                "type": "object",
+                "properties": {
+                  "sockets": {
+                    "type": "integer",
+                    "minimum": 1,
+                    "description": "Number of CPU sockets per node"
+                  },
+                  "cores_per_socket": {
+                    "type": "integer",
+                    "minimum": 1,
+                    "description": "Number of CPU cores per socket"
+                  },
+                  "threads_per_core": {
+                    "type": "integer",
+                    "minimum": 1,
+                    "description": "Number of CPU threads per core"
+                  },
+                  "real_memory": {
+                    "type": "integer",
+                    "minimum": 1,
+                    "description": "Memory in MB (exact value to use in Slurm)"
+                  },
+                  "gres": {
+                    "type": "string",
+                    "pattern": "^gpu:[0-9]+$",
+                    "description": "GPU resources in format 'gpu:N' (optional)"
+                  }
+                },
+                "required": ["sockets", "cores_per_socket", "threads_per_core", "real_memory"],
+                "additionalProperties": false
+              }
+            },
+            "additionalProperties": false
+          },
           "config_sources": {
             "type": "object",
             "description": "Config can be a file path or inline mapping",

common/library/module_utils/input_validation/validation_flows/common_validation.py

@@ -1110,7 +1110,30 @@ def validate_omnia_config(
                     "slurm NFS not provided",
                     f"NFS name {', '.join(diff_set)} required for slurm is not defined in {storage_config}"
                     ))
-
+
+        # Validate node_hardware_defaults requires node_discovery_mode=homogeneous
+        for clst in data.get('slurm_cluster', []):
+            node_hardware_defaults = clst.get('node_hardware_defaults')
+            node_discovery_mode = clst.get('node_discovery_mode')
+
+            # Normalize mode to lowercase for case-insensitive comparison
+            if node_discovery_mode and isinstance(node_discovery_mode, str):
+                node_discovery_mode = node_discovery_mode.lower()
+
+            if node_hardware_defaults and len(node_hardware_defaults) > 0:
+                if not node_discovery_mode or node_discovery_mode != 'homogeneous':
+                    group_names = list(node_hardware_defaults.keys())
+                    errors.append(
+                        create_error_msg(
+                            input_file_path,
+                            "slurm_cluster configuration inconsistency",
+                            f"'node_hardware_defaults' is specified for groups {group_names}, but 'node_discovery_mode' is not set to 'homogeneous'. "
+                            f"Current mode: {node_discovery_mode if node_discovery_mode else 'not set (defaults to heterogeneous)'}. "
+                            f"Either set 'node_discovery_mode: \"homogeneous\"' to use the hardware specifications, "
+                            f"or remove 'node_hardware_defaults' to use heterogeneous discovery."
+                        ))
+
+        cnfg_src = [clst.get('config_sources', {}) for clst in data.get('slurm_cluster')]
         skip_conf_validation = os.path.exists("/opt/omnia/input/.skip_slurm_conf_validation")
         cnfg_src = [clst.get('config_sources', {}) for clst in data.get('slurm_cluster')]
         skip_merge_list = [clst.get('skip_merge', False) for clst in data.get('slurm_cluster')]

common/library/module_utils/input_validation/validation_flows/local_repo_validation.py

@@ -20,6 +20,7 @@
 import re
 from ansible.module_utils.input_validation.common_utils import validation_utils
 from ansible.module_utils.input_validation.common_utils import config
+from ansible.module_utils.input_validation.common_utils import en_us_validation_msg
 from ansible.module_utils.local_repo.software_utils import load_yaml, load_json
 
 file_names = config.files
@@ -128,6 +129,27 @@ def validate_local_repo_config(input_file_path, data,
             if key_path and not os.path.exists(key_path):
                 errors.append(create_error_msg(local_repo_yml, "user_registry", 
                                              f"Key file not found: {key_path}"))
+
+    # Validate user_repo_url name prefixes
+    user_repo_prefix_map = {
+        "user_repo_url_x86_64": "x86_64_",
+        "user_repo_url_aarch64": "aarch64_",
+    }
+    for repo_key, expected_prefix in user_repo_prefix_map.items():
+        user_repos = data.get(repo_key)
+        if user_repos:
+            for repo in user_repos:
+                repo_name = repo.get("name", "")
+                if repo_name and not repo_name.startswith(expected_prefix):
+                    errors.append(create_error_msg(
+                        local_repo_yml, repo_key,
+                        en_us_validation_msg.USER_REPO_NAME_PREFIX_FAIL_MSG.format(
+                            repo_name=repo_name,
+                            repo_key=repo_key,
+                            expected_prefix=expected_prefix
+                        )
+                    ))
+
     repo_names = {}
     sub_result = check_subscription_status(logger)
     logger.info(f"validate_local_repo_config: Subscription status: {sub_result}")

common/library/module_utils/local_repo/config.py

@@ -120,10 +120,10 @@
 }
 
 CLI_FILE_PATH = "/root/.config/pulp/cli.toml"
-POST_TIMEOUT = 3600  # seconds
-TAR_POLL_VAL = 45    # minutes
-FILE_POLL_VAL = 1    # minutes
-ISO_POLL_VAL = 45    # minutes
+TAR_TIMEOUT_MIN = 45    # minutes
+FILE_TIMEOUT_MIN = 1    # minutes
+ISO_TIMEOUT_MIN = 45    # minutes
+TASK_POLL_INTERVAL = 10  # seconds
 FILE_URI = "/pulp/api/v3/content/file/files/"
 PULP_SSL_CA_CERT = "/etc/pki/ca-trust/source/anchors/pulp_webserver.crt"
 # ----------------------------
@@ -160,7 +160,7 @@
     "get_repo_version": "pulp container repository show --href %s",
     "list_tags_by_version": "pulp show --href /pulp/api/v3/content/container/tags/?repository_version=%s",
     "rename_repository": "pulp container repository update --name %s --new-name %s",
-    "orphan_cleanup": "pulp orphan cleanup",
+    "orphan_cleanup": "pulp orphan cleanup --protection-time 0",
     "container_distribution_show": "pulp container distribution show --name %s | jq .repository",
     "show_repository_version": "pulp container repository show --href %s | jq .latest_version_href",
     "list_image_tags": "pulp show --href /pulp/api/v3/content/container/tags/?repository_version=%s"

common/library/module_utils/local_repo/download_common.py

@@ -1,4 +1,4 @@
-# Copyright 2025 Dell Inc. or its subsidiaries. All Rights Reserved.
+# Copyright 2026 Dell Inc. or its subsidiaries. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -37,10 +37,10 @@
     pulp_file_commands,
     pulp_rpm_commands,
     CLI_FILE_PATH,
-    POST_TIMEOUT,
-    ISO_POLL_VAL,
-    TAR_POLL_VAL,
-    FILE_POLL_VAL,
+    ISO_TIMEOUT_MIN,
+    TAR_TIMEOUT_MIN,
+    FILE_TIMEOUT_MIN,
+    TASK_POLL_INTERVAL,
     FILE_URI,
     PULP_SSL_CA_CERT
 )
@@ -201,15 +201,15 @@ def wait_for_task(task_href, base_url, username, password, logger, timeout=3600,
     logger.error("Timeout waiting for task to complete")
     return False
 
-def handle_file_upload(repository_name, relative_path, file_url, poll_interval, logger):
+def handle_file_upload(repository_name, relative_path, file_url, timeout_minutes, logger):
     """
     Ensure repository exists, then POST a file to Pulp and wait for the task to complete.
 
     Args:
         repository_name (str): Name of the repository.
         relative_path (str): Relative path for the file in the repository.
         file_url (str): URL of the file to upload.
-        poll_interval: Polling time
+        timeout_minutes (int): Maximum time in minutes to wait for task completion.
         logger (logging.Logger): Logger instance.
 
     Returns:
@@ -262,31 +262,32 @@ def handle_file_upload(repository_name, relative_path, file_url, poll_interval,
         return "Failed"
 
     # Wait for task completion
-    logger.info(f"Waiting for task {task_href} to complete...")
+    timeout_seconds = timeout_minutes * 60
+    logger.info(f"Waiting for task {task_href} to complete (timeout: {timeout_minutes} min)...")
     task_result = wait_for_task(task_href, base_url, config["username"], passcode,
-                               logger, timeout=POST_TIMEOUT, interval=poll_interval)
+                               logger, timeout=timeout_seconds, interval=TASK_POLL_INTERVAL)
     if task_result:
         logger.info(f"File successfully uploaded to repository '{repository_name}'.")
         return "Success"
     else:
         logger.error(f"Task {task_href} failed or timed out. File upload to repository '{repository_name}' failed.")
         return "Failed"
 
-def handle_post_request(repository_name, relative_path, base_path, file_url, poll_interval,logger):
+def handle_post_request(repository_name, relative_path, base_path, file_url, timeout_minutes,logger):
     """
     Handles the full Pulp upload and distribution process for a given repository and file.
     Args:
         repository_name (str): Name of the Pulp repository.
         relative_path (str): Path where the file should be stored inside the repository.
         base_path (str): The base path for the distribution.
         file_url (str): URL of the file to be uploaded.
-        poll_interval: Interval for polling
+        timeout_minutes (int): Maximum time in minutes to wait for upload task completion.
         logger (logging.Logger): Logger for logging messages and errors.
 
     Returns:
         str: "Success" if the operation completes successfully, "Failed" otherwise.
     """
-    result = handle_file_upload(repository_name, relative_path, file_url, poll_interval,logger)
+    result = handle_file_upload(repository_name, relative_path, file_url, timeout_minutes,logger)
     if result =="Success":
         distribution_name = repository_name
         logger.info("Creating publication...")
@@ -483,7 +484,7 @@ def process_manifest(file,repo_store_path, status_file_path, cluster_os_type, cl
         relative_path = output_file
         base_path = manifest_directory.strip("/")
         status = handle_post_request(repository_name, relative_path,
-                 base_path, url, FILE_POLL_VAL, logger)
+                 base_path, url, FILE_TIMEOUT_MIN, logger)
     except Exception as e:
         logger.error(f"Error processing manifest: {e}")
         status= "Failed"
@@ -775,7 +776,7 @@ def process_tarball(package, repo_store_path, status_file_path, version_variable
             if url:
                 try:
                     status = handle_post_request(repository_name, relative_path,
-                             base_path, url, TAR_POLL_VAL,logger)
+                             base_path, url, TAR_TIMEOUT_MIN,logger)
                 except Exception as e:
                     logger.error(f"Error processing tarball: {e}")
                     status = "Failed"
@@ -882,7 +883,7 @@ def process_iso(package, repo_store_path, status_file_path,
                 # non-zero for failure)
                 subprocess.run(['wget', '-q', '--spider', '--tries=1', url], check=True)
                 status = handle_post_request(repository_name, relative_path,
-                         base_path, url, ISO_POLL_VAL,logger)
+                         base_path, url, ISO_TIMEOUT_MIN,logger)
         except subprocess.CalledProcessError as e:
             logger.error(f"Error executing iso commands: {e}")
             status = "Failed"
@@ -1176,4 +1177,4 @@ def process_rpm_file(package, repo_store_path, status_file_path, cluster_os_type
         # Write the status to the file
         write_status_to_file(status_file_path, package_name, package_type, status, logger, file_lock, repo_name)
         logger.info("#" * 30 + f" {process_rpm_file.__name__} end " + "#" * 30)
-        return status
+        return status

common/library/modules/parallel_tasks.py

@@ -106,7 +106,7 @@ def update_status_csv(csv_dir, software, overall_status,slogger):
 
     # Transform the new status.
     transformed_status = re.sub(r'failure', 'failed', overall_status.lower())
-    transformed_status = re.sub(r'partial', 'failed', overall_status.lower())
+    transformed_status = re.sub(r'timeout', 'failed', transformed_status)
 
     # Update or add the entry for each given software.
     if isinstance(software, list):