instance-repo 1.3.5.dev3__tar.gz → 1.3.5.dev5__tar.gz

{instance_repo-1.3.5.dev3 → instance_repo-1.3.5.dev5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: instance-repo
-Version: 1.3.5.dev3
+Version: 1.3.5.dev5
 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-1.3.5.dev3 → instance_repo-1.3.5.dev5}/instance_repo/__init__.py

@@ -20,7 +20,7 @@ from .errors import (
     TransportError,
 )
 
-__version__ = "1.3.5.dev3"
+__version__ = "1.3.5.dev5"
 
 __all__ = [
     "Repo", "load_instances",

{instance_repo-1.3.5.dev3 → instance_repo-1.3.5.dev5}/instance_repo/clients/instances.py

@@ -175,8 +175,8 @@ class InstancesClient:
             result = self._reg.get(
                 dataset, version, instance_id, split, **overrides)
         except TransportError as e:
-            # 路径 1: dataset 不存在(92001) → OSS 回退
-            if (e.biz_code == 92001
+            # 路径 1: dataset 不存在(92001) 或 split 不存在(92023) → OSS 回退
+            if (e.biz_code in (92001, 92023)
                     and fallback_to_oss and self._content is not None
                     and self._discovery_env_match(environment)):
                 return self._discover_instance(
@@ -204,7 +204,7 @@ class InstancesClient:
         try:
             return self._reg.query(dataset, version, split, **overrides)
         except TransportError as e:
-            if (e.biz_code == 92001
+            if (e.biz_code in (92001, 92023)
                     and fallback_to_oss and self._content is not None
                     and self._discovery_env_match(environment)):
                 return self._discover_instances(dataset, version, split)
@@ -283,7 +283,7 @@ class InstancesClient:
             data, pagination = self._t.request_with_meta(
                 "GET", f"{self._BASE}?{query}")
         except TransportError as e:
-            if (e.biz_code == 92001
+            if (e.biz_code in (92001, 92023)
                     and fallback_to_oss and self._content is not None
                     and self._discovery_env_match(environment)):
                 return self._discover_instances_paged(
@@ -463,17 +463,18 @@ class InstancesClient:
                 ("dataset_name", dataset.strip()),
                 ("split", split),
                 ("prefix", full_prefix),
-                ("page_size", str(int(page_size)))]
-            if (version or "").strip():
-                params.append(("version", version.strip()))
+                ("page_size", str(int(page_size))),
+                ("version", (version or "").strip())]
             if (dataset_split_id or "").strip():
                 params.append(("dataset_split_id", dataset_split_id.strip()))
             if tk:
                 params.append(("token", tk))
             if env:
                 params.append(("environment", env))
-            if mm:
-                params.append(("metadata_model", mm))
+            # metadata_model 必须始终发送——apiserver 据此决定走 split_first
+            # 还是 version_first 路径。缺失时 apiserver 默认 version_first，
+            # 会对 split_first 数据集错误要求 version 非空（92005）。
+            params.append(("metadata_model", mm or "split_first"))
             qs = "&".join(f"{k}={quote(v, safe='')}" for k, v in params)
             data = self._t.get(f"/apis/v1/datasets/objects?{qs}")
             if not isinstance(data, dict):
@@ -769,9 +770,11 @@ class InstancesClient:
             put = self._content.put(instance_id, dataset, version, split, tgz,
                                     meta.to_canonical_json(), overwrite=overwrite,
                                     layout=lay)
+            _tgz_size = tgz.stat().st_size
         skipped = bool(put.get("skipped"))
         _upload_elapsed = _time.monotonic() - _t0 - _pack_elapsed
-        _tgz_size = tgz.stat().st_size if not skipped else 0
+        if skipped:
+            _tgz_size = 0
         logger.info("push %s: pack=%.1fs (%d bytes) upload=%.1fs skipped=%s",
                     instance_id, _pack_elapsed, _tgz_size,
                     _upload_elapsed, skipped)
@@ -1184,9 +1187,12 @@ class InstancesClient:
             try:
                 meta = self._reg.get(dataset, version, instance_id, split)
             except TransportError as e:
-                if (e.biz_code == 92001 and fallback_to_oss
+                if (e.biz_code in (92001, 92023) and fallback_to_oss
                         and self._content is not None):
-                    # dataset 未注册 → OSS 回退下载（私网 + 系统调用方语义）
+                    # 92001: dataset 未注册
+                    # 92023: dataset 已注册但 split 在当前 environment 不存在
+                    #        （如数据在 online 但 SDK 查了 pre）
+                    # 两种情况均回退到 OSS 直接下载（私网 + 系统调用方语义）
                     return self._pull_unregistered(
                         dataset, version, instance_id, dest_dir, split,
                         path=path)
@@ -1444,7 +1450,7 @@ class InstancesClient:
             return []
         # path 非空时，LIST 直接用子前缀缩小范围
         list_prefix = asset_prefix + path_prefix if path_prefix else asset_prefix
-        keys = list_fn(list_prefix)
+        keys = list_fn(list_prefix, list_only=True, dataset=dataset)
         if not keys:
             return []
         for key in keys:

{instance_repo-1.3.5.dev3 → instance_repo-1.3.5.dev5}/instance_repo/store/oss.py

@@ -629,7 +629,7 @@ class OssContentStore(BaseContentStore):
         if oss_prefix and not oss_prefix.endswith("/"):
             oss_prefix += "/"
 
-        keys = self.list_objects(oss_prefix)
+        keys = self.list_objects(oss_prefix, list_only=True, dataset=dataset)
 
         def _download_one(key: str) -> Path:
             rel = key[len(oss_prefix):]

{instance_repo-1.3.5.dev3 → instance_repo-1.3.5.dev5}/instance_repo.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: instance-repo
-Version: 1.3.5.dev3
+Version: 1.3.5.dev5
 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-1.3.5.dev3 → instance_repo-1.3.5.dev5}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "instance-repo"
-version = "1.3.5.dev3"
+version = "1.3.5.dev5"
 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"