instance-repo 1.0.9.dev2__tar.gz → 1.0.9.dev4__tar.gz

{instance_repo-1.0.9.dev2 → instance_repo-1.0.9.dev4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: instance-repo
-Version: 1.0.9.dev2
+Version: 1.0.9.dev4
 Summary: InstanceRepo SDK — 评测 Instance 统一存储客户端（SDK-only；控制面由 apiserver 承载）
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown

{instance_repo-1.0.9.dev2 → instance_repo-1.0.9.dev4}/instance_repo/__init__.py

@@ -18,7 +18,7 @@ from .errors import (
     TransportError,
 )
 
-__version__ = "1.0.9.dev2"
+__version__ = "1.0.9.dev4"
 
 __all__ = [
     "Repo", "load_instances",

{instance_repo-1.0.9.dev2 → instance_repo-1.0.9.dev4}/instance_repo/cli.py

@@ -312,16 +312,16 @@ def build_parser() -> argparse.ArgumentParser:
                      help="get 用：覆盖 profile 派生的 environment（pre/online）")
 
     # ── 1.0.10 控制面管理子命令 ─────────────────────────────────────────────────
-    sp = sub.add_parser("split", help="split 记录管理（list/get/set-run-type）")
-    sp.add_argument("action", choices=["list", "get", "set-run-type"])
+    sp = sub.add_parser("split", help="split 记录管理（list/get/create/set-run-type）")
+    sp.add_argument("action", choices=["list", "get", "create", "set-run-type"])
     sp.add_argument("ref", nargs="?",
-                    help="list: {L1}/{L2}；get/set-run-type: {L1}/{L2}/{split}"
-                         "（或用 --dataset/--split）")
+                    help="list: {L1}/{L2}；get/set-run-type: {L1}/{L2}/{split}；"
+                         "create: {L1}/{L2}/{split}（或用 --dataset/--split）")
     sp.add_argument("--dataset", default=None, help="{L1}/{L2}")
     sp.add_argument("--split", default=None,
                     help="目标 split 名（get/set-run-type 用）")
     sp.add_argument("--version", default=None,
-                    help="list 按 version 精确过滤；get 可选定位版本")
+                    help="list 按 version 精确过滤；get 可选定位版本；create 必填")
     sp.add_argument("--scope", choices=["unversioned", "exact", "all"], default=None,
                     help="list 跨版本策略：缺省无 version → all（SDK 取全量哲学），"
                          "有 version 省略（服务端只接受 exact）")
@@ -329,9 +329,13 @@ def build_parser() -> argparse.ArgumentParser:
                     help="list 按状态过滤")
     sp.add_argument("--run-type", choices=["train", "eval", ""], default=None,
                     dest="run_type",
-                    help="set-run-type 用：train/eval 或空串清除")
+                    help="set-run-type/create 用：train/eval 或空串清除")
+    sp.add_argument("--storage-path", default=None, dest="storage_path",
+                    help="create 用：split 壳的 storage_path（缺省自动计算）")
     sp.add_argument("--page", type=int, default=1)
     sp.add_argument("--page-size", type=int, default=50, dest="page_size")
+    sp.add_argument("--environment", default=None,
+                    help="create 覆盖 profile 派生的 environment（pre/online）")
 
     ds = sub.add_parser("dataset",
                         help="数据集/基准/工作流管理（list/get/update/access/"
@@ -416,7 +420,7 @@ def _resolve_split_ref(args) -> tuple[str, str]:
     if args.ref:
         segs = _split_ref(args.ref, 3)
         return f"{segs[0]}/{segs[1]}", segs[2]
-    raise SystemExit("split get/set-run-type 需要 {L1}/{L2}/{split} 或 "
+    raise SystemExit("split get/create/set-run-type 需要 {L1}/{L2}/{split} 或 "
                      "--dataset/--split")
 
 
@@ -870,19 +874,17 @@ def main(argv=None, *, repo: Repo | None = None, out=None) -> int:
             # 1.0.10：取版本完整详情（splits 明细 + manifest + 时间戳）。
             # split_first 分支时服务端聚合 legacy 投影。
             dv = r.versions.get(dataset, version, environment=args.environment)
-            print(f"version\t{dv.get('version','')}\tstatus={dv.get('status','')}"
-                  f"\tstorage_path={dv.get('storage_path','')}", file=out)
-            print(f"published_at={dv.get('published_at') or ''}"
-                  f"\tlast_ingested_at={dv.get('last_ingested_at') or ''}",
-                  file=out)
-            print(f"dataset_version_id={dv.get('dataset_version_id','')}",
-                  file=out)
-            splits = dv.get("splits") or []
-            for s in splits:
-                print(f"  split\t{s.get('name','')}\tstatus={s.get('status','')}"
-                      f"\toss_prefix={s.get('oss_prefix','')}"
+            # versions.get 返回 DatasetVersion dataclass（字段访问，非 dict 取值）。
+            print(f"version\t{dv.version}\tstatus={dv.status}"
+                  f"\tstorage_path={dv.storage_path or ''}", file=out)
+            print(f"published_at={dv.published_at or ''}"
+                  f"\tlast_ingested_at={dv.last_ingested_at or ''}", file=out)
+            print(f"dataset_version_id={dv.dataset_version_id or ''}", file=out)
+            for s in dv.splits:
+                print(f"  split\t{s.get('name', '')}\tstatus={s.get('status', '')}"
+                      f"\toss_prefix={s.get('oss_prefix', '')}"
                       f"\tinstance_count={s.get('instance_count') or 0}", file=out)
-            manifest = dv.get("manifest") or {}
+            manifest = dv.manifest or {}
             print(f"manifest\tinstance_count={manifest.get('instance_count') or 0}"
                   f"\tfile_count={manifest.get('file_count') or 0}"
                   f"\tsize_bytes={manifest.get('size_bytes') or 0}", file=out)
@@ -919,6 +921,22 @@ def main(argv=None, *, repo: Repo | None = None, out=None) -> int:
                       file=out)
             print(f"page={res.page}/{_page_count(res)} total={res.total}", file=out)
             return 0
+        if args.action == "create":
+            dataset, split_name = _resolve_split_ref(args)
+            if not (args.version or "").strip():
+                raise SystemExit(
+                    "split create 需要 --version（apiserver CreateVersion 硬约束）")
+            sp = r.splits.create(dataset, split_name,
+                                 version=args.version,
+                                 run_type=args.run_type,
+                                 storage_path=args.storage_path,
+                                 environment=args.environment)
+            src = "projected" if sp.projected else "record"
+            print(f"created split\t{sp.name}\tversion={sp.version or ''}"
+                  f"\tstatus={sp.status}\trun_type={sp.run_type or ''}"
+                  f"\t[{src}]", file=out)
+            print(f"storage_path={sp.storage_path}", file=out)
+            return 0
         # get / set-run-type：需要定位到具体 split
         dataset, split_name = _resolve_split_ref(args)
         if args.action == "get":

{instance_repo-1.0.9.dev2 → instance_repo-1.0.9.dev4}/instance_repo/clients/splits.py

@@ -222,13 +222,68 @@ class SplitsClient:
 
     # ── 写 ──
 
+    def create(self, dataset: str, split: str, *,
+               version: str, run_type: str | None = None,
+               storage_path: str | None = None,
+               environment: str | None = None) -> DatasetSplit:
+        """创建 split 记录——组合 versions.create(splits=[split]) 的便利封装。
+
+        apiserver **没有** split create 端点（``POST /datasets/splits`` 不存在）：
+        split 记录由 ingest 扫描 OSS 或 versions.create/update(splits=[...])
+        间接产生（pre 集合 find-or-upsert ensure 真实 split record）。本方法是
+        "先建空 split 壳"的官方路径升级为 SDK 直接 API：
+
+        1. ``versions.create(dataset, version, splits=[split])``（version **必填**
+           ——apiserver CreateVersion 硬约束；storage_path 缺省自动计算）；
+        2. 可选 ``run_type``：create 后经 versions.update(split_run_types) 打标；
+        3. 读回 ``splits.get``（同 environment）返回记录。
+
+        参数:
+          dataset      : dataset 名（必填）。
+          split        : split 名（必填）。
+          version      : version（必填——CreateVersion 硬约束）。
+          run_type     : 可选，train/eval（create 后打标）。
+          storage_path : 可选，缺省按 profile 自动计算（与 versions.create 同源）。
+          environment  : 可选，缺省由 profile 派生（pre/online）。
+
+        返回 :class:`~instance_repo.models.DatasetSplit`（读回；pre 集合 ensure
+        的 record 或 version 内嵌投影）。
+        """
+        from .versions import VersionsClient
+        if not (dataset or "").strip():
+            raise SchemaError("dataset is required for splits.create")
+        if not (split or "").strip():
+            raise SchemaError("split is required for splits.create")
+        if not (version or "").strip():
+            raise SchemaError(
+                "version is required for splits.create (apiserver CreateVersion "
+                "requires version; split records are created via versions.create)")
+        rt = (run_type or "").strip()
+        if rt and rt not in RUN_TYPE_VALUES:
+            raise SchemaError(
+                f"invalid run_type {rt!r}; must be one of "
+                f"{', '.join(RUN_TYPE_VALUES)} or None")
+        env = self._env(environment)
+        vc = VersionsClient(self._t, self._p)
+        vc.create(dataset, version, splits=[split],
+                  storage_path=storage_path, environment=env or None)
+        if rt:
+            # run_type 打标与 create 互斥字段：经 split_run_types 单独打（apiserver
+            # UpdateVersion 的 run_type/split_run_types/splits 两两互斥）。
+            vc.update(dataset, version, split_run_types={split: rt})
+        return self.get(dataset, split, version=version,
+                        environment=env or None)
+
     def update(self, dataset: str, split: str | None = None, *,
                version: str | None = None, run_type: str = "",
-               dataset_split_id: str | None = None) -> DatasetSplit:
+               dataset_split_id: str | None = None,
+               environment: str | None = None) -> DatasetSplit:
         """更新 split 记录（``PATCH /datasets/splits/detail``）。
 
         服务端 body 仅接受 ``run_type``（**必填**："train"/"eval"，空串=清除）；
         仅 online 环境（split 记录 pre 环境的写经 versions.update 走集合路由）。
+        ``environment`` 缺省由 profile 派生（同 get/list）；profile 派生为 pre
+        而目标 split 记录在 online 集合时，须显式传 ``environment="online"``。
 
         寻址二选一：``dataset_split_id``（优先，split-first 精确记录 id）或
         ``dataset``+``split``（+可选 ``version``）name 三元组。
@@ -248,6 +303,8 @@ class SplitsClient:
             raise SchemaError(
                 f"invalid run_type {rt!r}; must be one of "
                 f"{', '.join(RUN_TYPE_VALUES)} (empty string to clear)")
+        if (env := self._env(environment)):
+            q += f"&environment={quote(env, safe='')}"
         data = self._t.patch(f"{self._BASE}/detail?{q}", {"run_type": rt})
         return DatasetSplit.from_dict(data) if isinstance(data, dict) else DatasetSplit()
 

{instance_repo-1.0.9.dev2 → instance_repo-1.0.9.dev4}/instance_repo.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: instance-repo
-Version: 1.0.9.dev2
+Version: 1.0.9.dev4
 Summary: InstanceRepo SDK — 评测 Instance 统一存储客户端（SDK-only；控制面由 apiserver 承载）
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown

{instance_repo-1.0.9.dev2 → instance_repo-1.0.9.dev4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "instance-repo"
-version = "1.0.9.dev2"
+version = "1.0.9.dev4"
 description = "InstanceRepo SDK — 评测 Instance 统一存储客户端（SDK-only；控制面由 apiserver 承载）"
 requires-python = ">=3.10"
 readme = "README.md"