utils/nvme: Enhanced get_ns_status to support ns topology fallback

The 'nvme show-topology /dev/<controller>' form fails on some nvme-cli
builds, returning {"error": "Invalid device name"} instead of a valid
JSON topology. Additionally, in multi-path subsystems where two
controllers (e.g. nvme0 and nvme3) share a namespace (e.g. nvme3n1),
the namespace block device is named after only one of the controllers,
making a namespace name constructed from controller_name unreliable.

Enhance get_ns_status with a two-stage approach:

Primary path (unchanged behaviour):
  nvme show-topology /dev/<controller_name> -o json
  Parses the original JSON structure where paths["Name"] matches
  the controller name. Returns immediately on success.

Fallback path (new):
  nvme show-topology -o json  (whole-system, no device argument)
  Triggered when the primary path returns an error payload (dict
  instead of list), raises JSONDecodeError, or yields no match.
  Locates the correct path by matching NSID and Controller.Name
  inside the Paths[].Controller[] array, which is unambiguous
  across multi-controller subsystems and avoids any namespace
  device name construction.

Both paths return [State, ANAState] preserving the existing API.

Signed-off-by: Maram Srimannarayana Murthy <msmurthy@linux.vnet.ibm.com>

Author: maramsmurthy

avocado/utils/nvme.py

@@ -652,24 +652,67 @@ def create_namespaces(controller_name, ns_count, shared_ns=False):
 
 def get_ns_status(controller_name, ns_id):
     """
-    Returns the status of namespaces on the specified controller
+    Returns the status of namespaces on the specified controller.
 
-    :param controller_name: name of the controller like nvme0
-    :param ns_id: ID of namespace for which we need the status
+    Tries ``nvme show-topology /dev/<controller_name>`` first.  If that
+    command returns an error payload (observed with some nvme-cli builds
+    where passing a controller node is rejected) or yields no match, the
+    function falls back to ``nvme show-topology -o json`` (whole-system)
+    and locates the path by matching ``NSID`` and ``Controller.Name``.
+    This correctly handles multi-path subsystems where the namespace block
+    device is named after a peer controller (e.g. ``nvme0`` and ``nvme3``
+    both serving ``nvme3n1``).
 
-    :rtype: list
+    :param controller_name: name of the controller, e.g. ``nvme0``
+    :param ns_id: NSID of the namespace whose status is required (int)
+
+    :rtype: list  -- ``[State, ANAState]`` for the matching path, or ``[]``
     """
     stat = []
+
     cmd = f"nvme show-topology /dev/{controller_name} -o json"
     data = process.run(cmd, ignore_status=True, sudo=True, shell=True).stdout_text
-    json_data = json.loads(data)
-    for data in json_data:
-        for subsystem in data["Subsystems"]:
-            for namespace in subsystem["Namespaces"]:
-                nsid = namespace["NSID"]
-                for paths in namespace["Paths"]:
-                    if nsid == ns_id and paths["Name"] == controller_name:
-                        stat.extend([paths["State"], paths["ANAState"]])
+    try:
+        json_data = json.loads(data)
+    except json.JSONDecodeError:
+        json_data = None
+
+    if isinstance(json_data, list):
+        for entry in json_data:
+            for subsystem in entry.get("Subsystems", []):
+                for namespace in subsystem.get("Namespaces", []):
+                    nsid = namespace.get("NSID")
+                    for path in namespace.get("Paths", []):
+                        if nsid == ns_id and path.get("Name") == controller_name:
+                            stat.extend([path["State"], path["ANAState"]])
+        if stat:
+            return stat
+
+    LOGGER.debug(
+        "get_ns_status: primary show-topology for %s failed or returned no "
+        "match; retrying with whole-system show-topology",
+        controller_name,
+    )
+    cmd = "nvme show-topology -o json"
+    data = process.run(cmd, ignore_status=True, sudo=True, shell=True).stdout_text
+    try:
+        json_data = json.loads(data)
+    except json.JSONDecodeError:
+        LOGGER.warning(
+            "get_ns_status: could not parse whole-system show-topology output"
+        )
+        return stat
+
+    if isinstance(json_data, list):
+        for entry in json_data:
+            for subsystem in entry.get("Subsystems", []):
+                for namespace in subsystem.get("Namespaces", []):
+                    if namespace.get("NSID") != ns_id:
+                        continue
+                    for path in namespace.get("Paths", []):
+                        for controller in path.get("Controller", []):
+                            if controller.get("Name") == controller_name:
+                                stat.extend([controller["State"], path["ANAState"]])
     return stat
 
 

docs/source/releases/next.rst

@@ -23,6 +23,12 @@ Bug Fixes
 
 * The Podman spawner now resolves the default ``podman`` binary through
   ``PATH``, supporting installations outside ``/usr/bin``.
+* :func:`avocado.utils.nvme.get_ns_status` now handles nvme-cli builds
+  that reject a controller node argument to ``nvme show-topology``, and
+  correctly resolves namespace status for peer controllers in multi-path
+  subsystems (e.g. ``nvme0`` and ``nvme3`` both serving ``nvme3n1``).
+  A whole-system ``nvme show-topology -o json`` fallback is used when
+  the controller-addressed form fails or yields no match.
 
 Internal changes
 ================