ap-client 0.8.0.dev0__tar.gz → 0.8.0.dev3__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (36) hide show
  1. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/PKG-INFO +3 -3
  2. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/dataset_commands.py +41 -344
  3. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/instance_commands.py +17 -35
  4. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/irepo_sdk.py +0 -57
  5. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/repo_read_guard.py +26 -14
  6. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/pyproject.toml +3 -3
  7. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/__init__.py +0 -0
  8. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/akpool_commands.py +0 -0
  9. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/api.py +0 -0
  10. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/cli.py +0 -0
  11. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/config.py +0 -0
  12. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/dataset_name.py +0 -0
  13. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/dataset_register.py +0 -0
  14. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/dfs_commands.py +0 -0
  15. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/exporter.py +0 -0
  16. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/fs_commands.py +0 -0
  17. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/fs_storage.py +0 -0
  18. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/fs_transfer.py +0 -0
  19. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/instance_options.py +0 -0
  20. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/managed_ak.py +0 -0
  21. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/manifest_view.py +0 -0
  22. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/ossutil_hash.py +0 -0
  23. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/ossutil_io.py +0 -0
  24. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/profile_cascade_all.py +0 -0
  25. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/profile_cascade_cli.py +0 -0
  26. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/profile_cascade_display.py +0 -0
  27. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/profile_cascade_state.py +0 -0
  28. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/profile_commands.py +0 -0
  29. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/profile_diagnostics.py +0 -0
  30. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/profile_display.py +0 -0
  31. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/profile_references.py +0 -0
  32. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/share_api.py +0 -0
  33. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/split_publish.py +0 -0
  34. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/submission_metadata.py +0 -0
  35. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/tbb.py +0 -0
  36. {ap_client-0.8.0.dev0 → ap_client-0.8.0.dev3}/ap_client/waiter.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: ap-client
3
- Version: 0.8.0.dev0
3
+ Version: 0.8.0.dev3
4
4
  Summary: Agent Platform API Client & CLI
5
5
  Requires-Python: >=3.10
6
6
  Requires-Dist: irepo-artifact==0.2.6
@@ -10,9 +10,9 @@ Requires-Dist: rich>=13.0.0
10
10
  Requires-Dist: typer>=0.9.0
11
11
  Requires-Dist: websockets>=13.0
12
12
  Provides-Extra: all
13
- Requires-Dist: instance-repo[oss]<2.0.0,>=1.4.0; extra == 'all'
13
+ Requires-Dist: instance-repo[oss]<2.0.0,>=1.4.6; extra == 'all'
14
14
  Provides-Extra: dataset
15
- Requires-Dist: instance-repo[oss]<2.0.0,>=1.4.0; extra == 'dataset'
15
+ Requires-Dist: instance-repo[oss]<2.0.0,>=1.4.6; extra == 'dataset'
16
16
  Description-Content-Type: text/markdown
17
17
 
18
18
  A lightweight Python SDK and command line interface for Agent Platform. It provides helpers for configuring API access and managing templates, datasets, jobs, and groups.
@@ -634,27 +634,19 @@ def _fetch_version(
634
634
  _SPLIT_PAGE_SIZE = 200
635
635
 
636
636
 
637
- def _unversioned_splits(
638
- dataset: str, *, stage: str, environment: Optional[str] = None
639
- ) -> list[dict]:
640
- """读**无版本空间**的 split(split_first)——走 SDK ``repo.splits``(recordsvc)。
641
-
642
- 为什么不能像有版本时那样读 ``versions.get`` + 内嵌 ``splits[]``:``GET/PATCH
643
- /datasets/versions/detail`` **必须**带非空 version(apiserver 400 92005
644
- "version is required",SDK 也在客户端前置拒绝),无版本空间表达不出来。真正承载它的是
645
- recordsvc 的 ``/datasets/splits``(``version_scope=unversioned``)。
646
- """
637
+ def _split_list(dataset: str, version: str, environment: Optional[str]) -> list[dict]:
647
638
  from .irepo_sdk import _reason
648
639
 
649
640
  env = _environment(environment)
650
641
  repo = _dataset_repo()
642
+ scope = {"version": version} if version else {"version_scope": "unversioned"}
651
643
  try:
652
644
  records: list[dict] = []
653
645
  page = 1
654
646
  while True:
655
647
  result = repo.splits.list(
656
648
  dataset,
657
- version_scope="unversioned",
649
+ **scope,
658
650
  page_size=_SPLIT_PAGE_SIZE,
659
651
  **({"page": page} if page > 1 else {}),
660
652
  **_read_env_kwargs(repo, repo.splits, "list", env),
@@ -662,70 +654,32 @@ def _unversioned_splits(
662
654
  items = getattr(result, "items", None) or []
663
655
  for item in items:
664
656
  record = _model_dict(item)
665
- if not _version_value(record.get("version")):
657
+ if _version_value(record.get("version")) == version:
666
658
  records.append(record)
667
659
  if not items or getattr(result, "has_more", False) is not True:
668
- break
660
+ return records
669
661
  page += 1
670
- except Exception as exc: # noqa: BLE001 - 统一走设计 §5.1 的映射
671
- raise _fail(
672
- f"dataset {dataset} (no version space): {stage} failed: {_reason(exc)}",
673
- f"Check the dataset name and permission: ap dataset get {dataset}",
674
- ) from exc
675
- return records
676
-
677
-
678
- def _unversioned_split(
679
- dataset: str, split: str, *, environment: Optional[str] = None
680
- ) -> Optional[dict]:
681
- """取无版本空间里的单个 split 记录;不存在返回 ``None``(不当作错误)。"""
682
- from .irepo_sdk import _reason
683
-
684
- env = _environment(environment)
685
- repo = _dataset_repo()
686
- try:
687
- record = repo.splits.get(dataset, split, **_read_env_kwargs(repo, repo.splits, "get", env))
688
662
  except Exception as exc: # noqa: BLE001
689
663
  raise _fail(
690
- f"dataset {dataset} (no version space) split {split}: query failed: {_reason(exc)}",
664
+ f"dataset {dataset} version {_version_label(version)}: split query failed: {_reason(exc)}",
691
665
  f"Check the dataset name and permission: ap dataset get {dataset}",
692
666
  ) from exc
693
- detail = _model_dict(record) if record is not None else None
694
- return detail if detail is not None and not _version_value(detail.get("version")) else None
695
667
 
696
668
 
697
- def _versioned_split(
698
- dataset: str,
699
- version: str,
700
- split: str,
701
- *,
702
- environment: Optional[str] = None,
703
- oss_fallback: bool = True,
669
+ def _split_get(
670
+ dataset: str, version: str, split: str, environment: Optional[str]
704
671
  ) -> Optional[dict]:
705
- """取指定 version 里的单个 split 记录;不存在返回 ``None``(不当作错误)。
706
-
707
- 与 ``_unversioned_split`` 同源,走 recordsvc 的 ``splits.get``(SDK 自带
708
- version_first 内嵌投影的 legacy 回退)。``version`` 为空表示无版本空间
709
- (split_first):SDK 不拼 version 参数,记录过滤条件退化为"version 为空",
710
- 与 ``_unversioned_split`` 等价。``split get`` 保持默认 OSS 发现回退;
711
- ``split create`` 的幂等检查须传 ``oss_fallback=False``:dataset 刚创建或
712
- 尚未上传时 OSS prefix 不存在,``_discover_split`` 会把"split 不存在"误报为
713
- 查询失败。禁用动作经 ``oss_fallback_kwargs(force=True)`` 走 SDK 兼容阶梯,
714
- 不支持该开关的旧版客户端不会收到这个参数。
715
- """
716
672
  from .irepo_sdk import _reason
717
673
 
718
674
  env = _environment(environment)
719
675
  repo = _dataset_repo()
720
- read_kwargs = _read_env_kwargs(repo, repo.splits, "get", env)
721
- if not oss_fallback:
722
- # 写路径预检强制关 OSS 发现,但必须走 guard 的 SDK 兼容阶梯:旧版
723
- # (<=1.3.1) metadata-only SplitsClient 的 get 既无 fallback_to_oss 参数
724
- # 也无 _content,阶梯返回 {}(旧版没有 OSS 发现,404→None 即"不存在");
725
- # 盲塞参数会 TypeError(线上 keluo 环境反馈)。
726
- read_kwargs.update(oss_fallback_kwargs(repo, repo.splits, "get", env, force=True))
727
676
  try:
728
- record = repo.splits.get(dataset, split, version=version, **read_kwargs)
677
+ record = repo.splits.get(
678
+ dataset,
679
+ split,
680
+ version=version,
681
+ **_read_env_kwargs(repo, repo.splits, "get", env),
682
+ )
729
683
  except Exception as exc: # noqa: BLE001
730
684
  raise _fail(
731
685
  f"dataset {dataset} version {_version_label(version)} split {split}: "
@@ -733,56 +687,12 @@ def _versioned_split(
733
687
  f"Check the dataset name and permission: ap dataset get {dataset}",
734
688
  ) from exc
735
689
  detail = _model_dict(record) if record is not None else None
736
- return (
737
- detail if detail is not None and _version_value(detail.get("version")) == version else None
738
- )
739
-
740
-
741
- def _versioned_splits(
742
- dataset: str, version: str, *, stage: str, environment: Optional[str] = None
743
- ) -> list[dict]:
744
- """读**指定 version**的 split——走 SDK ``repo.splits``(recordsvc)。
745
-
746
- 与 ``_unversioned_splits`` 同源:version 文档不内嵌 ``splits[]``(服务端只给
747
- ``split_count`` 聚合,split 记录由 recordsvc 的 ``/datasets/splits`` 承载),
748
- SDK ``splits.list(version=...)`` 是唯一正确读法,且自带存量 version_first
749
- 内嵌投影的 legacy 回退。
750
- """
751
- from .irepo_sdk import _reason
752
-
753
- env = _environment(environment)
754
- repo = _dataset_repo()
755
- try:
756
- records: list[dict] = []
757
- page = 1
758
- while True:
759
- result = repo.splits.list(
760
- dataset,
761
- version=version,
762
- page_size=_SPLIT_PAGE_SIZE,
763
- **({"page": page} if page > 1 else {}),
764
- **_read_env_kwargs(repo, repo.splits, "list", env),
765
- )
766
- items = getattr(result, "items", None) or []
767
- for item in items:
768
- record = _model_dict(item)
769
- # 与 ``_versioned_split`` 同口径:只收 version 匹配的记录(服务端
770
- # exact scope 保证;legacy 投影与 OSS 发现也会带 version)。
771
- if _version_value(record.get("version")) == version:
772
- records.append(record)
773
- if not items or getattr(result, "has_more", False) is not True:
774
- break
775
- page += 1
776
- except Exception as exc: # noqa: BLE001 - 统一走设计 §5.1 的映射
777
- raise _fail(
778
- f"dataset {dataset} version {version}: {stage} failed: {_reason(exc)}",
779
- f"Check the version name: ap dataset version get {version} --dataset {dataset}",
780
- ) from exc
781
- return records
690
+ if detail is None or _version_value(detail.get("version")) != version:
691
+ return None
692
+ return detail
782
693
 
783
694
 
784
695
  def _split_records(detail: dict) -> list[dict]:
785
- """version 文档内嵌的 ``splits[]``(设计 §4.3:split 没有独立资源)。"""
786
696
  splits = detail.get("splits")
787
697
  if not isinstance(splits, list):
788
698
  return []
@@ -795,25 +705,6 @@ def _split_records(detail: dict) -> list[dict]:
795
705
  return records
796
706
 
797
707
 
798
- def _split_names_of(records: list[dict]) -> list[str]:
799
- """recordsvc split 记录里的名字(去重)。"""
800
- names: list[str] = []
801
- for record in records:
802
- name = str(record.get("name") or "").strip()
803
- if name and name not in names:
804
- names.append(name)
805
- return names
806
-
807
-
808
- def _split_names(detail: dict) -> list[str]:
809
- names: list[str] = []
810
- for record in _split_records(detail):
811
- name = str(record.get("name") or "").strip()
812
- if name and name not in names:
813
- names.append(name)
814
- return names
815
-
816
-
817
708
  def _dataset_repo(*, request_timeout: float | None = None) -> "Repo":
818
709
  """dataset 写路径统一走 InstanceRepo SDK;未装 ``dataset`` extra 时退出 2。"""
819
710
  from .irepo_sdk import SdkMissing, load_repo
@@ -825,53 +716,6 @@ def _dataset_repo(*, request_timeout: float | None = None) -> "Repo":
825
716
  raise typer.Exit(2) from exc
826
717
 
827
718
 
828
- def _patch_version(
829
- dataset: str,
830
- version: str,
831
- body: dict,
832
- *,
833
- stage: str,
834
- fallback_hint: str = "",
835
- environment: Optional[str] = None,
836
- metadata_model: str = "",
837
- ) -> None:
838
- """split 写操作统一走 SDK;body 为 splits/split_run_types 单键。
839
-
840
- ``environment`` 必须与读路径同口径(缺省 ``online``):SDK 的 ``versions.update``
841
- 会按 profile 派生(dev 集群→``pre``),而版本在 ``online``,于是"能查到却 PATCH 404"。
842
- 见 :func:`ap_client.irepo_sdk.update_version_detail`。
843
- """
844
- from .irepo_sdk import _reason, update_version_detail
845
-
846
- repo = _dataset_repo()
847
- try:
848
- update_version_detail(
849
- repo,
850
- dataset,
851
- version,
852
- body,
853
- environment=_environment(environment),
854
- metadata_model=metadata_model,
855
- )
856
- except Exception as exc: # noqa: BLE001 - 统一走设计 §5.1 的映射
857
- code = getattr(exc, "biz_code", None)
858
- if code == 94018:
859
- hint = f"Writer role required: ap dataset access grant {dataset} --user <you> --role writer"
860
- elif code == 94016:
861
- hint = (
862
- f"Ask an admin to run ap dataset access grant {dataset} --user <you> --role writer"
863
- )
864
- else:
865
- shown = version if version else '""'
866
- hint = fallback_hint or (
867
- f"Check the version status: ap dataset version get {shown} --dataset {dataset}"
868
- )
869
- raise _fail(
870
- f"dataset {dataset} version {_version_label(version)}: {stage} failed: {_reason(exc)}",
871
- hint,
872
- ) from exc
873
-
874
-
875
719
  # ---- 行投影 ---------------------------------------------------------------
876
720
 
877
721
 
@@ -1935,14 +1779,8 @@ def dataset_split_list(
1935
1779
  resolved_version = _version_value(version)
1936
1780
  resolved_run_type = _check_choice(run_type, _RUN_TYPES, "--run-type") if run_type else ""
1937
1781
 
1938
- if resolved_version:
1939
- records = _versioned_splits(
1940
- dataset, resolved_version, stage="split query", environment=environment
1941
- )
1942
- rows = [_split_row(record, {}) for record in records]
1943
- else:
1944
- records = _unversioned_splits(dataset, stage="split query", environment=environment)
1945
- rows = [_split_row(record, {}) for record in records]
1782
+ records = _split_list(dataset, resolved_version, environment)
1783
+ rows = [_split_row(record, {}) for record in records]
1946
1784
  if resolved_run_type:
1947
1785
  rows = [row for row in rows if str(row["run_type"]).lower() == resolved_run_type]
1948
1786
  _emit(
@@ -1964,48 +1802,19 @@ def dataset_split_get(
1964
1802
  """Show one split of a version, or of the version-less (split_first) space."""
1965
1803
  output_format = _fmt(output_format)
1966
1804
  resolved_version = _version_value(version)
1967
-
1968
- if resolved_version:
1969
- record = _versioned_split(dataset, resolved_version, split, environment=environment)
1970
- if record is not None:
1971
- payload = dict(record)
1972
- payload.setdefault("dataset", dataset)
1973
- payload.setdefault("version", resolved_version)
1974
- payload.setdefault("registered", True)
1975
- _emit_detail(payload, output_format)
1976
- return
1977
- known = (
1978
- ", ".join(
1979
- _split_names_of(
1980
- _versioned_splits(
1981
- dataset, resolved_version, stage="split query", environment=environment
1982
- )
1983
- )
1984
- )
1985
- or "<none>"
1986
- )
1987
- else:
1988
- record = _unversioned_split(dataset, split, environment=environment)
1989
- if record is not None:
1990
- payload = dict(record)
1991
- payload.setdefault("dataset", dataset)
1992
- payload.setdefault("version", "")
1993
- payload.setdefault("registered", True)
1994
- _emit_detail(payload, output_format)
1995
- return
1996
- known = (
1997
- ", ".join(
1998
- _split_names_of(
1999
- _unversioned_splits(dataset, stage="split query", environment=environment)
2000
- )
2001
- )
2002
- or "<none>"
2003
- )
1805
+ record = _split_get(dataset, resolved_version, split, environment)
1806
+ if record is not None:
1807
+ payload = dict(record)
1808
+ payload.setdefault("dataset", dataset)
1809
+ payload.setdefault("version", resolved_version)
1810
+ payload.setdefault("registered", True)
1811
+ _emit_detail(payload, output_format)
1812
+ return
2004
1813
 
2005
1814
  raise _fail(
2006
1815
  f"dataset {dataset} version {_version_label(resolved_version)} split {split}: query failed: not found",
2007
- f"Existing splits in this {'version' if resolved_version else 'dataset (no version space)'}: {known}"
2008
- + ("" if resolved_version else f"; {_NO_VERSION_NOTE}"),
1816
+ f"List existing splits: ap dataset split list --dataset {dataset}"
1817
+ + (f" --version {resolved_version}" if resolved_version else f"; {_NO_VERSION_NOTE}"),
2009
1818
  )
2010
1819
 
2011
1820
 
@@ -2026,29 +1835,6 @@ def dataset_split_create(
2026
1835
  resolved_run_type = _check_choice(run_type, _RUN_TYPES, "--run-type") if run_type else ""
2027
1836
  env = _environment(environment)
2028
1837
 
2029
- # 有/无版本创建都走独立 Split API(POST /datasets/splits,version 可空):服务端
2030
- # 按 version 是否为空写 dataset/split 或 dataset/version/split,SDK 把空 version
2031
- # 转成不带 version 的请求,因此不需要分支。无版本路径直接落 recordsvc 的
2032
- # unversioned split record,不要求 OSS 上先有实例文件(空 split 壳也能建);
2033
- # ingest 只是另一条产出记录的路径。旧 Version PATCH(splits 全量替换 /
2034
- # split_run_types)已被服务端拒绝("split-first Version group updates are
2035
- # not supported")。
2036
- # 幂等预检禁用 OSS 发现回退:dataset 刚创建或尚未上传时 OSS prefix 不存在,
2037
- # _discover_split 会把"split 不存在"误报为查询失败("not found in OSS");
2038
- # recordsvc 404 → legacy 回退 → None 即判定"不存在 → 走 create"。有/无版本
2039
- # 共用同一预检,空 version 的过滤条件退化为"version 为空"。
2040
- existing = _versioned_split(
2041
- dataset, resolved_version, split, environment=environment, oss_fallback=False
2042
- )
2043
- if existing is not None:
2044
- # 幂等:已存在同名 split(设计 §4.3)。
2045
- payload = dict(existing)
2046
- payload.setdefault("dataset", dataset)
2047
- payload.setdefault("version", resolved_version)
2048
- payload["created"] = False
2049
- _emit_detail(payload, output_format)
2050
- return
2051
-
2052
1838
  from .instance_commands import _sdk_read_kwargs
2053
1839
  from .irepo_sdk import _reason
2054
1840
 
@@ -2068,7 +1854,6 @@ def dataset_split_create(
2068
1854
  payload.setdefault("dataset", dataset)
2069
1855
  payload.setdefault("version", resolved_version)
2070
1856
  payload.setdefault("split", split)
2071
- payload["created"] = True
2072
1857
  _emit_detail(payload, output_format)
2073
1858
 
2074
1859
 
@@ -2083,7 +1868,7 @@ def dataset_split_update(
2083
1868
  environment: Optional[str] = typer.Option(None, "--environment", help=_ENVIRONMENT_HELP),
2084
1869
  output_format: str = typer.Option(None, "--format", help=_FORMAT_HELP),
2085
1870
  ):
2086
- """Retag a split's run type (PATCH version.split_run_types, or the split record)."""
1871
+ """Retag a split's run type through the SDK."""
2087
1872
 
2088
1873
  output_format = _fmt(output_format)
2089
1874
  resolved_version = _version_value(version)
@@ -2094,92 +1879,7 @@ def dataset_split_update(
2094
1879
  exit_code=2,
2095
1880
  )
2096
1881
  resolved_run_type = _check_choice(run_type, _RUN_TYPES, "--run-type")
2097
-
2098
- if not resolved_version:
2099
- # 无版本空间改 run_type 走 recordsvc 的 split 记录(PATCH /datasets/splits/detail):
2100
- # 写端点仅限 online——SDK 1.1.2 起派生/显式 pre 都会前置 SchemaError 拒绝,
2101
- # 不再出现"存在性检查读 pre、写入默默落 online"的读写环境撕裂。
2102
- env = _environment(environment)
2103
- if env == "pre":
2104
- raise _fail(
2105
- f"Failed to update split {split} of dataset {dataset}: split records can only be changed in the online environment",
2106
- "PATCH on version-less split records only supports online; drop --environment pre, "
2107
- "or use a version-scoped run_type (--version <version>)",
2108
- exit_code=2,
2109
- )
2110
- if _unversioned_split(dataset, split, environment=environment) is None:
2111
- known = (
2112
- ", ".join(
2113
- _split_names_of(
2114
- _unversioned_splits(
2115
- dataset, stage="split update precheck", environment=environment
2116
- )
2117
- )
2118
- )
2119
- or "<none>"
2120
- )
2121
- raise _fail(
2122
- f"Failed to update split {split} of dataset {dataset} (no version space): not found",
2123
- f"Existing splits in this dataset's version-less space: {known}; {_NO_VERSION_NOTE}",
2124
- )
2125
- from .instance_commands import _sdk_read_kwargs
2126
- from .irepo_sdk import _reason
2127
-
2128
- repo = _dataset_repo()
2129
- # 写请求与上面的存在性预检查同口径路由环境(缺省 online);auto 由
2130
- # _sdk_read_kwargs 省略下发,跟随集群绑定。旧 SDK 不接受 environment 时
2131
- # 按显式值报升级提示,不静默忽略。
2132
- update_kwargs = _sdk_read_kwargs(repo.splits, "update", environment=env, metadata_model="")
2133
- try:
2134
- repo.splits.update(dataset, split, run_type=resolved_run_type, **update_kwargs)
2135
- except Exception as exc: # noqa: BLE001 - 统一走设计 §5.1 的映射
2136
- raise _fail(
2137
- f"Failed to update split {split} of dataset {dataset} (no version space): {_reason(exc)}",
2138
- "Write permission on this dataset is required; check the split name and the online environment, then retry",
2139
- ) from exc
2140
- _emit_detail(
2141
- {
2142
- "dataset": dataset,
2143
- "version": "",
2144
- "split": split,
2145
- "run_type": resolved_run_type,
2146
- "updated": True,
2147
- },
2148
- output_format,
2149
- )
2150
- return
2151
-
2152
- # 版本化更新走独立 Split API(PATCH /datasets/splits/detail),预检用精确
2153
- # split 身份(splits.get with version)——旧实现查 Version 内嵌 splits 再
2154
- # PATCH Version,与独立 Split 记录不匹配,导致"已存在却报 not found"。
2155
1882
  env = _environment(environment)
2156
- if env == "pre":
2157
- raise _fail(
2158
- f"Failed to update split {split} of dataset {dataset}: split records can only be changed in the online environment",
2159
- "PATCH on split records only supports online; drop --environment pre",
2160
- exit_code=2,
2161
- )
2162
- record = _versioned_split(dataset, resolved_version, split, environment=environment)
2163
- if record is None:
2164
- known = (
2165
- ", ".join(
2166
- _split_names_of(
2167
- _versioned_splits(
2168
- dataset,
2169
- resolved_version,
2170
- stage="split update precheck",
2171
- environment=environment,
2172
- )
2173
- )
2174
- )
2175
- or "<none>"
2176
- )
2177
- raise _fail(
2178
- f"dataset {dataset} version {_version_label(resolved_version)} split {split}: "
2179
- f"update failed: not found",
2180
- f"Create it first: ap dataset split create {split} --dataset {dataset} "
2181
- f"--version {resolved_version} (existing: {known})",
2182
- )
2183
1883
 
2184
1884
  from .instance_commands import _sdk_read_kwargs
2185
1885
  from .irepo_sdk import _reason
@@ -2187,21 +1887,18 @@ def dataset_split_update(
2187
1887
  repo = _dataset_repo()
2188
1888
  update_kwargs = _sdk_read_kwargs(repo.splits, "update", environment=env, metadata_model="")
2189
1889
  try:
2190
- repo.splits.update(
1890
+ record = repo.splits.update(
2191
1891
  dataset, split, version=resolved_version, run_type=resolved_run_type, **update_kwargs
2192
1892
  )
2193
- except Exception as exc: # noqa: BLE001 - 统一走设计 §5.1 的映射
1893
+ except Exception as exc: # noqa: BLE001
2194
1894
  raise _fail(
2195
- f"Failed to update split {split} of dataset {dataset} version {resolved_version}: {_reason(exc)}",
2196
- "Write permission on this dataset is required; check the split name and the online environment, then retry",
1895
+ f"Failed to update split {split} of dataset {dataset} version {_version_label(resolved_version)}: {_reason(exc)}",
1896
+ "Write permission on this dataset is required; check the split name and environment, then retry",
2197
1897
  ) from exc
2198
- _emit_detail(
2199
- {
2200
- "dataset": dataset,
2201
- "version": resolved_version,
2202
- "split": split,
2203
- "run_type": resolved_run_type,
2204
- "updated": True,
2205
- },
2206
- output_format,
2207
- )
1898
+ payload = _model_dict(record)
1899
+ payload.setdefault("dataset", dataset)
1900
+ payload.setdefault("version", resolved_version)
1901
+ payload.setdefault("split", split)
1902
+ payload.setdefault("run_type", resolved_run_type)
1903
+ payload["updated"] = True
1904
+ _emit_detail(payload, output_format)
@@ -243,37 +243,11 @@ def _describe(exc: BaseException) -> str:
243
243
 
244
244
 
245
245
  def _sdk_read_kwargs(sdk_obj: Any, method: str, *, environment: str, metadata_model: str) -> dict:
246
- """按 SDK 能力下发 environment / metadata_model,不允许静默忽略显式值。"""
247
- fn = getattr(sdk_obj, method, None)
248
- accepted: set[str] = set()
249
- if fn is not None:
250
- try:
251
- params = inspect.signature(fn).parameters
252
- except (TypeError, ValueError): # pragma: no cover - 打桩异常形态
253
- params = {}
254
- for name, param in params.items():
255
- if param.kind is inspect.Parameter.VAR_KEYWORD:
256
- accepted.update(("environment", "metadata_model"))
257
- elif name in ("environment", "metadata_model"):
258
- accepted.add(name)
259
-
260
246
  kwargs: dict[str, str] = {}
261
- dropped: list[str] = []
262
- for key, value in (("environment", environment), ("metadata_model", metadata_model)):
263
- # ``auto`` 是 CLI 层取值:跟随集群绑定,不下发(服务端契约只有 online/pre)。
264
- if key == "environment" and value == "auto":
265
- continue
266
- # 空值不下发:CLI 不替 SDK 决定 metadata_model(见 dataset_commands._metadata_model)。
267
- if key in accepted and value:
268
- kwargs[key] = value
269
- elif value:
270
- dropped.append(key)
271
- if dropped:
272
- raise _fail(
273
- f"The installed instance-repo does not support per-call {'/'.join(dropped)}",
274
- "Upgrade to instance-repo 1.1.4 or newer and retry",
275
- exit_code=2,
276
- )
247
+ if environment and environment != "auto":
248
+ kwargs["environment"] = environment
249
+ if metadata_model:
250
+ kwargs["metadata_model"] = metadata_model
277
251
  return kwargs
278
252
 
279
253
 
@@ -982,6 +956,7 @@ def instance_pull(
982
956
  dataset: str = typer.Option(..., "--dataset", help="Dataset name"),
983
957
  split: str = typer.Option(..., "--split", help="Split name (required for remote operations)"),
984
958
  version: Optional[str] = typer.Option(None, "--version", help=_VERSION_HELP),
959
+ environment: Optional[str] = typer.Option(None, "--environment", help=_ENVIRONMENT_HELP),
985
960
  output: Optional[str] = typer.Option(
986
961
  None, "--output", "-o", help="Output directory (default: current directory)"
987
962
  ),
@@ -997,6 +972,7 @@ def instance_pull(
997
972
  from .irepo_sdk import sdk_error_exit
998
973
 
999
974
  output_format = _fmt(output_format)
975
+ env = _environment(environment)
1000
976
  resolved_path = _path_value(path)
1001
977
  resolved = _version_value(version)
1002
978
  root = Path(output or ".")
@@ -1008,9 +984,17 @@ def instance_pull(
1008
984
  )
1009
985
 
1010
986
  repo = _load_repo_or_exit()
1011
- pull_options = {}
987
+ from .repo_read_guard import pull_client
988
+
989
+ try:
990
+ client = pull_client(repo.instances, env)
991
+ pull_options = oss_fallback_kwargs(repo, client, "pull", env)
992
+ except (TypeError, ValueError, RuntimeError) as exc:
993
+ raise _fail(
994
+ str(exc), "Upgrade Dataset support: pip install --upgrade 'ap-client[dataset]'", 2
995
+ ) from exc
1012
996
  if resolved_path is not None:
1013
- pull_options = _enhanced_options(repo.instances, "pull", {"path": resolved_path})
997
+ pull_options.update(_enhanced_options(client, "pull", {"path": resolved_path}))
1014
998
  try:
1015
999
  private_endpoint_reachable = _probe_pull_private_endpoint(repo, dataset)
1016
1000
  except Exception as exc: # noqa: BLE001
@@ -1027,9 +1011,7 @@ def instance_pull(
1027
1011
  staging = Path(tempfile.mkdtemp(prefix=f"ap-pull-{instance}-"))
1028
1012
  try:
1029
1013
  try:
1030
- written = repo.instances.pull(
1031
- dataset, resolved, instance, staging, split, **pull_options
1032
- )
1014
+ written = client.pull(dataset, resolved, instance, staging, split, **pull_options)
1033
1015
  except Exception as exc: # noqa: BLE001
1034
1016
  sdk_error_exit(exc, what=f"instance {instance}", stage="pull")
1035
1017
  staged = _place_under_target(Path(str(written)), staging / instance)
@@ -59,7 +59,6 @@ __all__ = [
59
59
  "is_already_satisfied",
60
60
  "load_repo",
61
61
  "sdk_error_exit",
62
- "update_version_detail",
63
62
  ]
64
63
 
65
64
  #: 数据集元数据的环境枚举(服务端 ``dataset environment``)。
@@ -1239,62 +1238,6 @@ def _is_verbose() -> bool:
1239
1238
  return _parse_bool(os.environ.get("AP_VERBOSE"))
1240
1239
 
1241
1240
 
1242
- #: 集群绑定存储 → 环境侧(dev 三个变体都是 pre 侧),与 dataset_commands._STORAGE_SIDES 同口径。
1243
- _STORAGE_SIDES = {
1244
- "dev": "pre",
1245
- "internal-dev": "pre",
1246
- "external-dev": "pre",
1247
- "prod": "online",
1248
- }
1249
-
1250
-
1251
- def _derive_environment(repo: Any) -> str:
1252
- """``auto`` 的实际取值:按 repo profile 的集群存储绑定派生,未知回退 online。"""
1253
- profile = getattr(repo, "profile", None)
1254
- storage_env = str(getattr(profile, "storage_env", "") or "").strip().lower()
1255
- return _STORAGE_SIDES.get(storage_env, "online")
1256
-
1257
-
1258
- def update_version_detail(
1259
- repo: Any,
1260
- dataset: str,
1261
- version: str,
1262
- body: dict,
1263
- *,
1264
- environment: str,
1265
- metadata_model: str = "",
1266
- ) -> None:
1267
- """``PATCH /apis/v1/datasets/versions/detail`` with explicit routing selectors."""
1268
- if environment == "auto":
1269
- # auto 是 CLI 层取值(跟随集群绑定),服务端契约只有 online/pre。直连
1270
- # transport 的兼容分支必须显式带上派生值——只删 query 会落到服务端
1271
- # 缺省 online,在 dev 集群上造成读写撕裂;SDK 分支传派生值与 SDK 自行
1272
- # 按 profile 派生等价,两条路径口径一致。
1273
- environment = _derive_environment(repo)
1274
- update = repo.versions.update
1275
- try:
1276
- accepted = inspect.signature(update).parameters
1277
- except (TypeError, ValueError): # pragma: no cover - 内建/装饰过的可调用
1278
- accepted = {}
1279
- if "environment" in accepted and "metadata_model" in accepted:
1280
- kwargs = dict(body)
1281
- if environment:
1282
- kwargs["environment"] = environment
1283
- if metadata_model:
1284
- kwargs["metadata_model"] = metadata_model
1285
- update(dataset, version, **kwargs)
1286
- return
1287
-
1288
- from urllib.parse import quote
1289
-
1290
- query = f"dataset_name={quote(dataset, safe='')}&version={quote(version, safe='')}"
1291
- if environment:
1292
- query += f"&environment={quote(environment, safe='')}"
1293
- if metadata_model:
1294
- query += f"&metadata_model={quote(metadata_model, safe='')}"
1295
- repo.transport.patch(f"/apis/v1/datasets/versions/detail?{query}", dict(body))
1296
-
1297
-
1298
1241
  def _reason(exc: Exception) -> str:
1299
1242
  message = getattr(exc, "message", None)
1300
1243
  text = str(message or "").strip() or str(exc).strip()
@@ -3,6 +3,8 @@
3
3
  from __future__ import annotations
4
4
 
5
5
  import inspect
6
+ from copy import copy
7
+ from functools import partial
6
8
  from typing import Any
7
9
 
8
10
  STORAGE_SIDES = {
@@ -13,34 +15,44 @@ STORAGE_SIDES = {
13
15
  }
14
16
 
15
17
 
16
- def oss_fallback_kwargs(
17
- repo: Any, client: Any, method: str, environment: str, *, force: bool = False
18
- ) -> dict:
18
+ def pull_client(client: Any, environment: str) -> Any:
19
+ """Scope SDK 1.4.x pull metadata reads without changing its storage binding.
20
+
21
+ pull has no environment argument, although its registry supports one.
22
+ Copy both objects so other operations on the Repo keep their original context.
23
+ """
24
+ if environment == "auto":
25
+ return client
26
+ registry = getattr(client, "_reg", None)
27
+ get = getattr(registry, "get", None)
28
+ if not callable(get) or "environment" not in inspect.signature(get).parameters:
29
+ raise ValueError(
30
+ "The installed instance-repo cannot scope pull metadata reads; "
31
+ "upgrade 'ap-client[dataset]'"
32
+ )
33
+ scoped = copy(client)
34
+ scoped._reg = copy(registry)
35
+ scoped._reg.get = partial(get, environment=environment)
36
+ return scoped
37
+
38
+
39
+ def oss_fallback_kwargs(repo: Any, client: Any, method: str, environment: str) -> dict:
19
40
  """Allow API reads across environments; disable unsafe storage fallback.
20
41
 
21
42
  Missing/unknown storage provenance is insufficient for explicit online/pre
22
43
  discovery. auto follows the bound store. Bind missing SDK discovery provenance.
23
-
24
- ``force`` hard-disables discovery even when ``environment`` is ``auto`` or
25
- matches the bound storage — write-path prechecks (split-create idempotency)
26
- must read a recordsvc miss as ``None`` instead of letting OSS discovery turn
27
- a missing prefix into "query failed". It reuses the same SDK compatibility
28
- ladder: older metadata-only clients (no ``_content``) have no discovery to
29
- disable and get ``{}``; a content-bearing client without the switch fails
30
- closed with an upgrade hint instead of blindly passing an unsupported
31
- ``fallback_to_oss`` argument (``TypeError`` on instance-repo <=1.3.1).
32
44
  """
33
45
  profile = getattr(repo, "profile", None)
34
46
  # instance-repo 1.3.5.dev2+ reads _p in Dataset discovery without binding it
35
47
  # during construction. Supply the actual Repo profile before that check.
36
48
  if callable(getattr(client, "_discovery_env_match", None)) and not hasattr(client, "_p"):
37
49
  client._p = profile
38
- if environment == "auto" and not force:
50
+ if environment == "auto":
39
51
  return {}
40
52
  storage = getattr(profile, "storage_env", "")
41
53
  storage = storage.strip().lower() if isinstance(storage, str) else ""
42
54
  side = STORAGE_SIDES.get(storage, "")
43
- if side and side == environment and not force:
55
+ if side and side == environment:
44
56
  return {}
45
57
 
46
58
  try:
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
4
4
 
5
5
  [project]
6
6
  name = "ap-client"
7
- version = "0.8.0.dev0"
7
+ version = "0.8.0.dev3"
8
8
  description = "Agent Platform API Client & CLI"
9
9
  readme = { text = "A lightweight Python SDK and command line interface for Agent Platform. It provides helpers for configuring API access and managing templates, datasets, jobs, and groups.", content-type = "text/markdown" }
10
10
  requires-python = ">=3.10"
@@ -19,10 +19,10 @@ dependencies = [
19
19
 
20
20
  [project.optional-dependencies]
21
21
  dataset = [
22
- "instance-repo[oss]>=1.4.0,<2.0.0",
22
+ "instance-repo[oss]>=1.4.6,<2.0.0",
23
23
  ]
24
24
  all = [
25
- "instance-repo[oss]>=1.4.0,<2.0.0",
25
+ "instance-repo[oss]>=1.4.6,<2.0.0",
26
26
  ]
27
27
 
28
28
  [project.scripts]