Fix NVMe remote disk detection with model-based identification

[umfranci]

The current code uses a controller-based heuristic to identify remote disks, which is unreliable on certain VM configurations.

Changes

• In nvmecli.py

  • get_devices(): Add support for new nvme-cli JSON schema (Subsystems → Controllers → Namespaces) alongside the legacy flat schema.
  • get_device_models(): New method returning a device-path-to-model mapping. Uses try/except for graceful fallback on malformed JSON.

• In nvme.py

  • _remove_nvme_remote_disks(): Use model-based detection (checking for "NVMe Accelerator" in model name) as primary method. Includes prefix matching for controller paths (/dev/nvme0) against namespace keys (/dev/nvme0n1). Falls back to legacy controller-based heuristic if nvme-cli data is unavailable.

• In features.py

  • _is_remote_data_disk(): Use model-based detection as primary method with legacy fallback.

Notes

• No behavior change on VMs with SCSI disk controllers.
• No behavior change on VMs where nvme-cli is unavailable (fallback path is preserved).

[Copilot]

Pull request overview

This PR improves NVMe remote disk detection on Azure by switching from a controller-based heuristic to primarily using nvme-cli’s reported device model (e.g., identifying remote disks via “NVMe Accelerator”), while retaining a legacy fallback when nvme-cli data isn’t available.

Changes:

• Extend Nvmecli.get_devices() parsing to support both legacy and newer nvme-cli JSON schemas, including deriving missing namespace IDs from DevicePath.
• Add Nvmecli.get_device_models() to map NVMe namespace device paths to model names.
• Update NVMe remote-disk filtering in Nvme and Azure Disk feature logic to use model-based detection first with legacy fallback.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
lisa/tools/nvmecli.py	Adds namespace-id derivation for older nvme-cli JSON and introduces model mapping across legacy/new schemas.
lisa/features/nvme.py	Uses nvme-cli model mapping to filter out remote (“Accelerator”) NVMe devices with fallback to prior heuristic.
lisa/sut_orchestrator/azure/features.py	Updates Azure remote data disk detection to prefer model-based identification with legacy fallback.

---

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

This PR improves Azure NVMe remote disk detection by switching from a controller-based heuristic to model-based identification (using nvme-cli “ModelNumber”), and extends nvme-cli parsing to support both legacy and newer JSON schemas.

Changes:

• Extend Nvmecli.get_devices() to tolerate missing NameSpace in legacy nvme-cli JSON by deriving it from DevicePath, and keep supporting the newer Subsystems→Controllers→Namespaces schema.
• Add Nvmecli.get_device_models() to map /dev/nvme* namespace paths to model names from nvme-cli JSON (legacy + new schema).
• Update NVMe remote disk filtering to prefer model-based detection (with legacy fallback) in both the NVMe feature and Azure disk feature logic.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
lisa/tools/nvmecli.py	Adds namespace-id derivation for legacy JSON missing NameSpace, and introduces model extraction (get_device_models) across both nvme-cli schemas.
lisa/features/nvme.py	Updates remote NVMe disk filtering to primarily use model-based detection with safe controller-prefix matching fallback behavior unchanged.
lisa/sut_orchestrator/azure/features.py	Updates Azure remote data disk classification to prefer model-based detection, falling back to the legacy OS-controller heuristic.

**Key Test Cases:**
verify_nvme_basic|verify_nvme_disk_controller_type|verify_hot_add_disk_serial

**Impacted LISA Features:**
Nvme, Disk

**Tested Azure Marketplace Images:**
- canonical 0001-com-ubuntu-server-jammy 22_04-lts latest
- redhat rhel 8_10 latest

---

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

This PR improves Azure NVMe remote-disk detection by switching from a controller-based heuristic to model-based identification, and updates nvme-cli parsing to support both legacy and newer JSON schemas.

Changes:

• Extend Nvmecli.get_devices() to parse both legacy “flat” and newer “Subsystems → Controllers → Namespaces” JSON schemas (and derive missing namespace IDs from DevicePath when needed).
• Add Nvmecli.get_device_models() and use it as the primary signal for identifying Azure remote NVMe disks (model contains “NVMe Accelerator”), with legacy heuristic fallback.
• Update Azure disk feature logic to use model-based remote-disk detection first, preserving legacy fallback behavior.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
lisa/tools/nvmecli.py	Adds new-schema parsing and introduces get_device_models() for device-path → model mapping.
lisa/features/nvme.py	Uses model-based detection to filter out remote NVMe devices from local NVMe enumeration, with legacy fallback.
lisa/sut_orchestrator/azure/features.py	Updates remote data disk detection to prefer model-based classification, with legacy fallback.

Key Test Cases:
verify_nvme_disk_controller_type|verify_hot_add_disk_serial

Impacted LISA Features:
Nvme, Disk

Tested Azure Marketplace Images:

• canonical 0001-com-ubuntu-server-jammy 22_04-lts latest
• redhat rhel 9_5 latest

[kanchansenlaskar]

Idea is not just return a disk list but also to modify the original disk_list by removing nvme remote disks. Any particular reason why a new list was returned without modifying disk_list? If no, please make that necessary change and then check the call hierarchy and test it with all the tests making use of this modified method.

[umfranci]

Updated the code to modify the original list - making it consistent with the existing code

[kanchansenlaskar]

While this is fine, please use like this instead, to make it less ambiguous:
ns_match = re.search(r"nvme\d+n(\d+)$", device_path)

[umfranci]

Updated the code to use the suggested pattern matching

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

[Copilot]

get_device_models() assumes json.loads(...)["Devices"] is iterable of dict-like objects. If Devices exists but is not a list of dicts (e.g., malformed output where Devices is a dict/str/null), the subsequent for nvme_device in nvme_devices will raise (e.g., AttributeError on nvme_device.get) instead of gracefully returning {}. Consider validating the parsed structure (top-level is dict and Devices is a list) before iterating, and return {} when it isn't.