Use the developer-provided openop_tfsdk tree as the refactor baseline.

Use the proposed public action vocabulary and keep the Kubernetes surface limited to implemented actions.

TF-SDK loggers now come from structlog, so they follow the host
process's structlog configuration (e.g. SRM), with no extra setup.

Original-Author: Igli <kociaj4@gmail.com>
Reviewed-Commit: 68263cd13627df876c46c0b5da26c6694a9dbeaf
Publication-Reason: ETSI requires the author to be a project member.

Both stay in the code but are unsupported in this release, as discussed
in the 2026-10-08 meeting: there is no access to test them.

Original-Author: Igli <kociaj4@gmail.com>
Reviewed-Commit: 861cb6e126159d4518f1b39a9e1d44942d31ac5d
Publication-Reason: ETSI requires the author to be a project member.

CI installs isort 5.13.2 from requirements.txt, while pre-commit runs
6.0.1. With the black profile, 6.x keeps an import exploded when it ends
in a trailing comma and 5.13 joins it back onto one line, so files the
hook had formatted failed CI's check (9 of them on this branch). Pin
6.0.1 in requirements.txt so the hook and CI agree.

kubernetes/client.py also had `from requests import Response` inside
the first-party block; it moves up to the third-party imports, where
both versions put it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

- Open5GCore looks up the QoS profile in the case it validated it in (T5).
- QoD sends the device IPv6 address itself as ueIpv6Addr (T11).
- Open5GS and Open5GCore reject an unsupported profile with an error the
  adapter reports as ValidationError (T11).
- QoD uses the control path's notification_url as notificationDestination (T7).
- A QoD device without an IPv4 or IPv6 address is rejected with
  ValidationError before the NEF is called (T8).

Original-Author: Igli <kociaj4@gmail.com>
Reviewed-Commit: 6a1ae35edff047c58599a735b19996551eb8ecd3

- The adapter declares the container runtime kind. A container deployment
  unit runs its image from a small chart shipped with the adapter: a
  Deployment, plus a Service for the ports of its compute intent (NodePort
  when a port is external, ClusterIP otherwise).
- Terminating a release helm does not know raises ResourceNotFound.
- A missing helm binary raises DomainExecutionFailed instead of
  FileNotFoundError.
- helm upgrade uses --atomic, like helm install.

Original-Author: Igli <kociaj4@gmail.com>
Reviewed-Commit: b3977281a0167852c3a25940969fcc9ffd2264ba

- QoD activate reports active once the NEF created the subscription, as
  traffic influence does (TF-SDK-1).
- QoD without a notification destination is rejected with ValidationError
  instead of sending "None" to the NEF (TF-SDK-2). The NEF then no longer
  stores "None", which also made reading the session back fail.
- Open5GCore no longer overwrites the UE IPv4 with a hardcoded address
  (TF-SDK-3).
- Validation errors keep their reason: the client's message, or the field
  and rule of a pydantic error (TF-SDK-6).
- The OAI client no longer logs the UE's IP and external id with a
  non-ASCII arrow; on a non-UTF-8 console that log line failed the
  location request.

Original-Author: Igli <kociaj4@gmail.com>
Reviewed-Commit: 149fdb0e5a2fcbc11ec2bebab8cae207c26e0ac5

- A location query raises SubscriberNotFound, UnsupportedIdentifier,
  TargetNotLocatable, MaxAgeUnsatisfiable or MaxSurfaceUnsatisfiable, the
  names SRM maps to CAMARA 404 and 422 (T1).
- A subscriber the NEF does not know is SubscriberNotFound whether it was
  looked up by IPv4 or by network access identifier (T2).
- The OAI client no longer reports a cell-only answer as a circle at (0, 0);
  the adapter raises TargetNotLocatable (T3).
- DeviceTarget carries ipv4_private and ipv4_port (srm.params/v1). The OAI
  location lookup and the Open5GS/Open5GCore QoD request use the private
  address, the one the core knows the UE by; without it nothing changes (T4).
- A location older than max_age_seconds or larger than max_surface_sqm is
  refused with MaxAgeUnsatisfiable or MaxSurfaceUnsatisfiable (T5).
- The location fix time is the NEF's ueLocationTimestamp when present (I1).

Original-Author: Igli <kociaj4@gmail.com>
Reviewed-Commit: 908134ee3b13b61302772ea99c7e4a2ced92c2a3

Reviewed-Commit: 80548ca4fbfb7133c6fa7e46aa11708668c018f8

fix(ci): check imports with the same isort as pre-commit

See merge request !202

The adapter used the deployment unit ref as the release name. Charts name
their Services after the release, and a Service name must be a DNS-1035
label: lowercase, starting with a letter. SRM passes instance UUIDs (about
10 in 16 start with a digit) and instance names with '_', so those installs
failed with a Service "... is invalid" that the logs reported only as
helm_command_failed.

release_name() lowercases the ref, turns anything but letters, digits and
'-' into '-', puts "r-" in front when it doesn't start with a letter, and
keeps to helm's 53 characters. A ref that is already valid keeps its name,
and terminate, update and status follow the external id activate returns,
so existing releases are unaffected. helm's "is invalid" errors now log as
invalid_kubernetes_object.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

fix(kubernetes): derive Helm names accepted by Kubernetes

See merge request !201

The project is now the OpenOP TF SDK. The README, the contribution guide,
the O-RAN file headers and the package metadata use that name, and the
issue tracker link points to the work items of the ETSI repository.

The legacy kubernetes client keeps its old names until it is removed.

Pin Black 26.10.1 and Flake8 7.4.1 with their required dependencies. Apply the formatter-only migration with identical Python ASTs. Verify hooks and 137 offline SDK tests on Python 3.10 through 3.13.

SRM resolves the traffic influence application server from the endpoints a
workload reports at deploy, but the adapter returned none. After helm
install, a container with external ports now reports one endpoint per
port: the node port Kubernetes assigned, read with kubectl, and the host
configured as endpoint_host in the control path.

Without endpoint_host, or when kubectl is missing or fails, the result has
no endpoints as before and the deploy still succeeds.

Use per-adapter kubeconfig and context for Helm lifecycle operations and Service observation. Capture compatibility defaults at construction, isolate client and subprocess configuration, and reject unresolved or unusable explicit connections. Verify simultaneous control paths, failure isolation and service-account execution.

feat: Compute resource (required) support for container type

See merge request !206

fix: configure OAI QoD profiles per control path

See merge request !207

Return Helm workload NodePort endpoints for Release 2.0.0

See merge request !208

Prepare TF-SDK for release 2.0.0

See merge request !205

Prepare TF-SDK 2.0.0rc4 candidate

See merge request !210

Simplify release approval and collect CI reports

See merge request !211

Automate SDK package publication

See merge request !212

feat: adding logs to warn the reason for empty endpoints

See merge request !213

fix: rename of appInstanceId field to appServerIp to correctly name what it is

See merge request !214

Prepare TF-SDK 2.0.0 and publish releases after main merges

See merge request !215

Run dependency audits automatically

See merge request !216

Release TF-SDK 2.0.0

See merge request !209