aipmodel 0.2.69__tar.gz → 0.2.71__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.
- {aipmodel-0.2.69 → aipmodel-0.2.71}/PKG-INFO +1 -1
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel/CephS3Manager.py +33 -7
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel/__init__.py +1 -1
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel/model_registry.py +64 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel.egg-info/PKG-INFO +1 -1
- {aipmodel-0.2.69 → aipmodel-0.2.71}/setup.py +1 -1
- {aipmodel-0.2.69 → aipmodel-0.2.71}/MANIFEST.in +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/README.md +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel/acl_manager.py +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel/exceptions.py +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel/template.py +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel/update_checker.py +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel.egg-info/SOURCES.txt +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel.egg-info/dependency_links.txt +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel.egg-info/requires.txt +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/aipmodel.egg-info/top_level.txt +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/requirements.txt +0 -0
- {aipmodel-0.2.69 → aipmodel-0.2.71}/setup.cfg +0 -0
|
@@ -81,17 +81,32 @@ class CephS3Manager:
|
|
|
81
81
|
print(f"[SDK_INFO] Bucket: {self.bucket_name}")
|
|
82
82
|
print("[SDK_DEBUG] Creating S3 client...")
|
|
83
83
|
|
|
84
|
+
_cfg = {
|
|
85
|
+
"retries": {
|
|
86
|
+
"max_attempts": int(os.environ.get("S3_BOTO_MAX_ATTEMPTS", 10)),
|
|
87
|
+
"mode": "adaptive",
|
|
88
|
+
}
|
|
89
|
+
}
|
|
90
|
+
try:
|
|
91
|
+
# boto3 >= 1.36 defaults to CRC32 checksums, which MinIO/Ceph RGW
|
|
92
|
+
# reject on DeleteObjects ("MissingContentMD5"). "when_required"
|
|
93
|
+
# restores the classic Content-MD5 behavior those backends expect.
|
|
94
|
+
client_config = Config(
|
|
95
|
+
request_checksum_calculation="when_required",
|
|
96
|
+
response_checksum_validation="when_required",
|
|
97
|
+
**_cfg,
|
|
98
|
+
)
|
|
99
|
+
except TypeError:
|
|
100
|
+
# older botocore (< 1.36) doesn't know the checksum params — and
|
|
101
|
+
# doesn't need them (it already sends Content-MD5).
|
|
102
|
+
client_config = Config(**_cfg)
|
|
103
|
+
|
|
84
104
|
self.s3 = boto3.client(
|
|
85
105
|
"s3",
|
|
86
106
|
endpoint_url=CEPH_ENDPOINT_URL,
|
|
87
107
|
aws_access_key_id=CEPH_ADMIN_ACCESS_KEY,
|
|
88
108
|
aws_secret_access_key=CEPH_ADMIN_SECRET_KEY,
|
|
89
|
-
config=
|
|
90
|
-
retries={
|
|
91
|
-
"max_attempts": int(os.environ.get("S3_BOTO_MAX_ATTEMPTS", 10)),
|
|
92
|
-
"mode": "adaptive",
|
|
93
|
-
}
|
|
94
|
-
),
|
|
109
|
+
config=client_config,
|
|
95
110
|
)
|
|
96
111
|
|
|
97
112
|
if self.verbose:
|
|
@@ -832,7 +847,18 @@ class CephS3Manager:
|
|
|
832
847
|
|
|
833
848
|
for i in range(0, len(objs), 1000):
|
|
834
849
|
chunk = objs[i:i+1000]
|
|
835
|
-
|
|
850
|
+
try:
|
|
851
|
+
self.s3.delete_objects(Bucket=self.bucket_name, Delete={"Objects": chunk})
|
|
852
|
+
except ClientError as e:
|
|
853
|
+
# boto3 >= 1.36 sends CRC32 checksums that MinIO/Ceph RGW
|
|
854
|
+
# reject on the batch API ("MissingContentMD5"). Fall back
|
|
855
|
+
# to per-object deletes, which need no checksum header.
|
|
856
|
+
if e.response.get("Error", {}).get("Code") != "MissingContentMD5":
|
|
857
|
+
raise
|
|
858
|
+
if self.verbose:
|
|
859
|
+
print("[SDK_WARN] Batch delete rejected (MissingContentMD5); falling back to per-object deletes")
|
|
860
|
+
for obj in chunk:
|
|
861
|
+
self.s3.delete_object(Bucket=self.bucket_name, Key=obj["Key"])
|
|
836
862
|
|
|
837
863
|
if self.verbose:
|
|
838
864
|
print("[SDK_SUCCESS] Folder deleted")
|
|
@@ -1772,6 +1772,70 @@ class MLOpsManager:
|
|
|
1772
1772
|
pass
|
|
1773
1773
|
return None
|
|
1774
1774
|
|
|
1775
|
+
def remove_missing_version(self, model_name, version):
|
|
1776
|
+
"""
|
|
1777
|
+
Sync-repair for a ZOMBIE version: the registry references `version` but
|
|
1778
|
+
its S3 files are gone (e.g. deleted by hand in Ceph). Removes the version
|
|
1779
|
+
from the registry so registry and storage agree again. Refuses to run if
|
|
1780
|
+
the files actually still exist (use delete_model for a real delete).
|
|
1781
|
+
"""
|
|
1782
|
+
report = self.verify_model_integrity(model_name=model_name)
|
|
1783
|
+
entry = next((v for v in report["versions"] if v["version"] == version), None)
|
|
1784
|
+
if entry is None:
|
|
1785
|
+
raise ModelNotFoundError(f"Version '{version}' not found for model '{model_name}'")
|
|
1786
|
+
if entry["files_exist"]:
|
|
1787
|
+
raise ValueError(
|
|
1788
|
+
f"Files for version '{version}' of '{model_name}' still exist in storage — "
|
|
1789
|
+
f"this is not a zombie. Use delete_model to actually delete it."
|
|
1790
|
+
)
|
|
1791
|
+
# delete_model is metadata-first and its S3 delete is a no-op for an
|
|
1792
|
+
# already-empty prefix, so it is exactly the sync we need.
|
|
1793
|
+
self.delete_model(model_name=model_name, version=version)
|
|
1794
|
+
print(f"[SDK_SUCCESS] Synced: removed zombie version '{version}' of '{model_name}' from registry.")
|
|
1795
|
+
|
|
1796
|
+
def purge_orphan_folders(self, model_name=None, model_id=None):
|
|
1797
|
+
"""
|
|
1798
|
+
Sync-repair for ORPHAN folders inside a registered model: S3 folders
|
|
1799
|
+
under the model that no registry version references (leftovers from an
|
|
1800
|
+
interrupted operation). Deletes those folders. Returns the list purged.
|
|
1801
|
+
"""
|
|
1802
|
+
report = self.verify_model_integrity(model_name=model_name, model_id=model_id)
|
|
1803
|
+
mid = report["model_id"]
|
|
1804
|
+
data = self.models.get_by_id(mid)
|
|
1805
|
+
is_public = self._extract_metadata_value(data.get("metadata", {}), "public", "false").lower() == "true"
|
|
1806
|
+
mgr = self._get_ceph_manager(is_public)
|
|
1807
|
+
purged = []
|
|
1808
|
+
for folder in report["orphan_folders"]:
|
|
1809
|
+
prefix = f"_models/{mid}/{folder}/"
|
|
1810
|
+
try:
|
|
1811
|
+
mgr.delete_folder(prefix)
|
|
1812
|
+
purged.append(folder)
|
|
1813
|
+
except Exception as e:
|
|
1814
|
+
print(f"[SDK_WARN] Could not purge orphan folder {prefix}: {e}")
|
|
1815
|
+
print(f"[SDK_SUCCESS] Purged {len(purged)} orphan folder(s) for model '{report['model_name']}'.")
|
|
1816
|
+
return purged
|
|
1817
|
+
|
|
1818
|
+
def purge_orphan_model(self, model_id):
|
|
1819
|
+
"""
|
|
1820
|
+
Sync-repair for an ORPHAN model: an S3 model folder the registry does
|
|
1821
|
+
not know about (e.g. the registry entry was deleted while storage
|
|
1822
|
+
survived). Deletes the whole `_models/{model_id}/` folder. Refuses to
|
|
1823
|
+
run if the id IS known to the registry (that would be a real delete —
|
|
1824
|
+
use delete_model).
|
|
1825
|
+
"""
|
|
1826
|
+
registry_ids = {m.get("id") for m in (self.models.get_all(self.project_id) or [])}
|
|
1827
|
+
if model_id in registry_ids:
|
|
1828
|
+
raise ValueError(
|
|
1829
|
+
f"Model id '{model_id}' exists in the registry — not an orphan. "
|
|
1830
|
+
f"Use delete_model to delete a registered model."
|
|
1831
|
+
)
|
|
1832
|
+
mgr = self._get_ceph_manager(False)
|
|
1833
|
+
base = f"_models/{model_id}/"
|
|
1834
|
+
if not mgr.check_if_exists(base):
|
|
1835
|
+
raise ModelNotFoundError(f"No storage folder found at {base}")
|
|
1836
|
+
mgr.delete_folder(base)
|
|
1837
|
+
print(f"[SDK_SUCCESS] Purged orphan storage folder {base}.")
|
|
1838
|
+
|
|
1775
1839
|
def recover_model_from_storage(self, model_id):
|
|
1776
1840
|
"""
|
|
1777
1841
|
Rebuild a registry (Mlops-Core) entry for a model that exists in S3 but
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|