chore(deps): move the kpubdata pin to the 0.9 line

CHANGELOG.md

@@ -39,6 +39,7 @@
 
 ### Changed
 
+- The kpubdata pin moves to the 0.9 line (`kpubdata>=0.9.0,<0.10`, #990, #1002, kpubdata#780). 0.9.0 is the first release with `Client(env_keys=False)`: on 0.8.0 `serve` refused to start in a multi-user deployment because the operator's keys could not be kept out of a request, and now it starts. Seven imports from kpubdata's private modules — `find_spec`, `discover_specs`, `KPubDataConfig` and `SENSITIVE_PARAM_KEYS` — use the public names, and their allowlist entries are gone. **What changes for a deployment:** kpubdata 0.9.0 declares terms for more datasets, and Builder acts on them — `datago.air_quality` is now `forbidden` (no publish, and no rows through query, preview, exports or downloads), and the 16 specs whose `allowed` was never confirmed are `unknown` (no public publish). kpubdata's own breaking changes apply too: HTTP 401 is `AuthError`, a total count of zero is `0`, NODATA is an empty result.
 - The automated deployment to the `app-01` VM is removed. `.github/workflows/deploy.yml` ran on every push to `main` that touched the service, built and pushed an image, then failed at its SSH step — the deployment it was written for is not being used. The workflow and `ops/deploy/app-01-rollout.sh`, which only it ran, are deleted. `docker.yml` still publishes the images to GHCR, and `docker-compose.prod.app.yml` is unchanged for anyone who runs it by hand. `docs/deploy.md`, `docs/CREDENTIAL_SURFACE.md` and ADR 0017 say so.
 - A pull request builds the documentation site in strict mode. `mkdocs build --strict` ran only in `docs.yml`, on a push to `main`, so the link #1026 fixed was found after it had merged and the deploy had failed. `ci.yml` has a `Docs build (strict)` job now and the `CI gate` waits for it; `docs.yml` still does the deploy. `tests/unit/test_ci_docs_gate.py` holds that the job exists, that the gate needs it and that it runs the command the deploy runs.
 - The contract declares four things the service already sent (#994, API contract 1.80.0, description only). `X-Provider-Key` is a header parameter of the five operations that call a provider (`previewBuild`, `createBuild`, `submitBuild`, `getProviderStatus`, `testProviderConnection`) — it was mentioned only in `info.description`, so a client's drift test had nothing to check it against. The 429 `auth_throttled` and the overload 503 are shared responses any operation can answer (`components.responses.AuthThrottled`, `ServerOverloaded`), and `X-Request-ID` is `components.headers.RequestId`. `tests/unit/test_contract_shared_declarations.py` compares each with what the code sends. Generating the route table from the route adapter, the remaining part of #994, is not done.

CONTRIBUTING.md

@@ -102,11 +102,11 @@ CI(`publish-dataset.yml`)와 패키지 배포 환경에서는 `--no-sources` 플
 uv sync --extra dev --extra publish --no-sources
 ```
 
-`--no-sources`는 `[tool.uv.sources]`를 무시하고, `pyproject.toml`의 `dependencies`에 명시된 **PyPI 릴리스 핀**(`kpubdata>=0.8.0,<0.9`)을 직접 설치합니다. 형제 디렉터리가 없어도 작동합니다.
+`--no-sources`는 `[tool.uv.sources]`를 무시하고, `pyproject.toml`의 `dependencies`에 명시된 **PyPI 릴리스 핀**(`kpubdata>=0.9.0,<0.10`)을 직접 설치합니다. 형제 디렉터리가 없어도 작동합니다.
 
-#### 핀 범위(`>=0.8.0,<0.9`)를 이렇게 설정한 이유
+#### 핀 범위(`>=0.9.0,<0.10`)를 이렇게 설정한 이유
 
-`kpubdata-builder`는 `kpubdata`의 공개 API만 씁니다(kpubdata ADR 0007, `scripts/check_kpubdata_imports.py`). kpubdata 0.8.0은 코드 열의 앞자리 0을 보존해 `str`로 돌려주는 breaking 릴리스이며(kpubdata#613), #702·#688·#689가 기다리던 라이선스·열 메타데이터 필드를 더합니다. 0.9 이상은 검증 전이라 허용하지 않습니다 (관련 이슈: #213). CI 의 `kpubdata <버전>` 잡이 범위 안의 모든 릴리스로 스위트를 돌립니다(#832).
+`kpubdata-builder`는 `kpubdata`의 공개 API만 씁니다(kpubdata ADR 0007, `scripts/check_kpubdata_imports.py`). kpubdata 0.9.0은 `Client(env_keys=False)` 가 들어간 첫 릴리스입니다 — 다중 사용자 모드가 운영자 키를 요청에서 떼어 놓는 데 필요합니다(#990, kpubdata#780). breaking 릴리스이기도 합니다: 모르는 `Client` 인자를 거부하고, HTTP 401 이 `AuthError` 가 되고, 건수 0 을 `None` 이 아니라 `0` 으로 보고하고, NODATA 가 모든 경로에서 빈 결과입니다. 그래서 0.9 라인에 고정합니다.
 
 핀의 정본은 `pyproject.toml`의 `dependencies`입니다. 이 표와 어긋나면 `pyproject.toml`이 맞습니다.
 
@@ -115,7 +115,7 @@ uv sync --extra dev --extra publish --no-sources
 | 환경 | 명령 | `kpubdata` 소스 |
 | :--- | :--- | :--- |
 | 로컬 개발 | `uv sync --extra dev` | `../kpubdata` (editable, 형제 디렉터리 필요) |
-| CI / 배포 | `uv sync ... --no-sources` | PyPI (`kpubdata>=0.8.0,<0.9`) |
+| CI / 배포 | `uv sync ... --no-sources` | PyPI (`kpubdata>=0.9.0,<0.10`) |
 
 ## 3. 브랜치 전략과 협업 규칙
 

pyproject.toml

@@ -28,11 +28,12 @@ classifiers = [
 ]
 dependencies = [
   # Builder depends on the kpubdata public API only (ADR 0007 in kpubdata,
-  # scripts/check_kpubdata_imports.py). 0.8.0 is breaking: code columns keep their
-  # leading zeros and come back as str (kpubdata#613), and it adds the licence and
-  # column-metadata fields #702, #688 and #689 wait on. Pin to the 0.8 line; compat
-  # policy: #213.
-  "kpubdata>=0.8.0,<0.9",
+  # scripts/check_kpubdata_imports.py). 0.9.0 is the first release with
+  # `Client(env_keys=False)`, which the multi-user mode needs to keep the operator's
+  # keys out of a request (#990, kpubdata#780), and it is breaking: unknown `Client`
+  # arguments are refused, HTTP 401 is `AuthError`, a total count of zero is `0`, and
+  # NODATA is an empty result on every path. Pin to the 0.9 line; compat policy: #213.
+  "kpubdata>=0.9.0,<0.10",
   "cryptography>=50,<51",
   # The tabular engine the DuckDB migration moves to (ADR 0021, #866). 1.2.0 is the first
   # release with every sandbox setting the query profile needs — allowed_paths,

scripts/check_kpubdata_imports.py

@@ -62,31 +62,9 @@
     "and transport; kpubdata has no public verify API (kpubdata#667). Signatures are "
     "pinned by tests/unit/test_kpubdata_internal_surface.py."
 )
-_SPEC_LOOKUP_REASON = (
-    "Looks up a SpecDefinition by id; the public Client returns DatasetRef, not the "
-    "spec model verify needs (kpubdata#667)."
-)
-_CONFIG_REASON = (
-    "Reads the operator's provider key the way kpubdata resolves it from the "
-    "environment; Client does not expose key resolution publicly (kpubdata#667)."
-)
 
 #: (path, module, name) -> reason. ``name`` is "" for ``import module``.
 ALLOWLIST: dict[tuple[str, str, str], str] = {
-    (
-        "src/kpubdata_builder/logging_redaction.py",
-        "kpubdata.transport._sensitive",
-        "SENSITIVE_PARAM_KEYS",
-    ): (
-        "The one canonical list of credential parameter names; Builder keeps no copy "
-        "so the two cannot drift. Needs a public export (kpubdata#667)."
-    ),
-    ("src/kpubdata_builder/service/providers.py", "kpubdata.config", "KPubDataConfig"): (
-        _CONFIG_REASON
-    ),
-    ("src/kpubdata_builder/verify/runner.py", "kpubdata.config", "KPubDataConfig"): (
-        _VERIFY_REASON
-    ),
     ("src/kpubdata_builder/verify/runner.py", "kpubdata.core.executor", "SpecExecutor"): (
         _VERIFY_REASON
     ),
@@ -112,14 +90,6 @@
     ("src/kpubdata_builder/verify/runner.py", "kpubdata.transport.http", "HttpTransport"): (
         _VERIFY_REASON
     ),
-    ("src/kpubdata_builder/agent/monitor.py", "kpubdata.core.spec", "find_spec"): (
-        _SPEC_LOOKUP_REASON
-    ),
-    ("src/kpubdata_builder/agent/pipeline.py", "kpubdata.core.spec", "find_spec"): (
-        _SPEC_LOOKUP_REASON
-    ),
-    ("src/kpubdata_builder/cli.py", "kpubdata.core.spec", "find_spec"): _SPEC_LOOKUP_REASON,
-    ("src/kpubdata_builder/cli.py", "kpubdata.core.spec", "discover_specs"): (_SPEC_LOOKUP_REASON),
 }
 
 

src/kpubdata_builder/agent/monitor.py

@@ -97,7 +97,7 @@ def check_approval(dataset_id: str, *, api_key: str | None = None) -> str:
     Returns:
         Status string: HEALTHY, NEEDS_APPLICATION, INVALID_KEY, etc.
     """
-    from kpubdata.core.spec import find_spec
+    from kpubdata import find_spec
 
     from kpubdata_builder.verify import verify_dataset
 

src/kpubdata_builder/agent/pipeline.py

@@ -81,7 +81,7 @@ def run_pipeline(
         [
             sys.executable,
             "-c",
-            "import sys; from kpubdata.core.spec import find_spec; "
+            "import sys; from kpubdata import find_spec; "
             "sys.exit(0 if find_spec(sys.argv[1]) is not None else 1)",
             dataset_id,
         ],

src/kpubdata_builder/cli.py

@@ -474,7 +474,7 @@ def build_parser() -> argparse.ArgumentParser:
 #: Why a client that must not use the operator's keys cannot be built (#990).
 ENV_KEYS_UNSUPPORTED = (
     "the installed kpubdata cannot keep the environment's provider keys out of a client "
-    "(Client has no env_keys option; it was added after 0.8.0), so a request without its "
+    "(Client has no env_keys option; it came with 0.9.0), so a request without its "
     "own key would run on the operator's. Install a kpubdata release that has it, or "
     "turn KPUBDATA_BUILDER_REQUIRE_OWN_PROVIDER_CREDENTIAL and multi-user mode off."
 )
@@ -485,7 +485,8 @@ def client_keeps_environment_keys_out() -> bool:
 
     kpubdata 0.8.0 has no such option and accepts any keyword without an error, so
     passing ``env_keys=False`` to it does nothing and says nothing. Asking the signature
-    is the only way to know the option is honoured.
+    is the only way to know the option is honoured. Builder's pin is the 0.9 line, which
+    has it; the check stays for an environment that installed something older anyway.
     """
     import inspect
 
@@ -1347,7 +1348,7 @@ def _run_verify(
         page_size: Records per test request.
         hashes_path: Optional path to previous schema hashes YAML.
     """
-    from kpubdata.core.spec import discover_specs, find_spec
+    from kpubdata import discover_specs, find_spec
 
     from .verify import runner as _verify_runner
 

src/kpubdata_builder/logging_redaction.py

@@ -28,7 +28,7 @@
 from collections.abc import Iterable
 from urllib.parse import quote, quote_plus
 
-from kpubdata.transport._sensitive import SENSITIVE_PARAM_KEYS
+from kpubdata import SENSITIVE_PARAM_KEYS
 
 REDACTED = "[REDACTED]"
 

src/kpubdata_builder/service/providers.py

@@ -11,8 +11,7 @@
 from datetime import datetime, timezone
 from typing import Literal, Protocol, cast
 
-from kpubdata import Client, Operation
-from kpubdata.config import KPubDataConfig
+from kpubdata import Client, KPubDataConfig, Operation
 from kpubdata.core.models import DatasetRef
 from kpubdata.exceptions import (
     AuthError,

src/kpubdata_builder/verify/runner.py

@@ -119,7 +119,7 @@ def verify_dataset(
     # ``tests/unit/test_kpubdata_internal_surface.py`` pins this list,
     # so upgrade breaks in CI, not at runtime. When starting to use new internal symbols,
     # add them to that list too.
-    from kpubdata.config import KPubDataConfig
+    from kpubdata import KPubDataConfig
     from kpubdata.core.executor import (
         SpecExecutor,
         check_payload_error,

tests/unit/test_verify_cli.py

@@ -12,7 +12,7 @@
 # All tests in this module patch the verify.runner module which is imported
 # lazily inside _run_verify via ``from .verify import runner as _verify_runner``.
 _RUNNER_MOD = "kpubdata_builder.verify.runner"
-_SPEC_MOD = "kpubdata.core.spec"
+_SPEC_MOD = "kpubdata"
 
 
 class TestVerifyParser: