Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

ModelBuilder.build() ignores the configured session when it reloads a model

Aperta Adatta ai principianti
  • Qualcosa non va?
#6,459 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 2 giorni

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
2/5
Tempo stimato
1-3 ore
Idoneità per principianti
72/100
Tipo di issue
Bug
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
aws, python
Ambito
cloud, machine-learning

Direzione di ricerca

Il bug si trova nel return finale di ModelBuilder.build() in sagemaker/serve/model_builder.py, dove Model.get() viene chiamato senza la sessione configurata. Leggi come viene impostato self.sagemaker_session in quella classe, quindi verifica se Model.get() accetta un argomento di sessione. Aggiungi un test unitario che usi sessioni boto predefinita e di destinazione diverse e che verifichi che Model.get() riceva la sessione di destinazione dopo CreateModel. È completato quando questo test passa e i test esistenti di ModelBuilder continuano a passare.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

PySDK Version

  • PySDK V2 (2.x)
  • PySDK V3 (3.x)

Describe the bug

ModelBuilder.build() creates a SageMaker model with its configured
sagemaker_session. It then reloads the model with Model.get() without
passing that session.

If the configured session and default boto session use different AWS accounts,
the create request succeeds in the configured account. The subsequent describe
request uses the default account and fails because the model does not exist
there.

The relevant call in sagemaker/serve/model_builder.py is:

return Model.get(model_name=self.model_name, region=self.region)

To reproduce

Configure two AWS CLI profiles that use different AWS accounts. The target
profile must have SageMaker permissions and permission to pass the execution
role.

Set these environment variables:

export TARGET_PROFILE="<profile-for-target-account>"
export DEFAULT_PROFILE="<profile-for-different-account>"
export TARGET_ROLE_ARN="arn:aws:iam::<target-account-id>:role/<sagemaker-execution-role>"
export TARGET_BUCKET="<existing-bucket-in-target-account>"
export IMAGE_URI="<inference-image-uri-accessible-from-target-account>"
export AWS_REGION="us-east-2"

Run this script:

import os
import uuid

import boto3
from botocore.exceptions import ClientError
from sagemaker.core.helper.session_helper import Session
from sagemaker.core.resources import Model
from sagemaker.core.shapes import ContainerDefinition
from sagemaker.serve import ModelBuilder

region = os.environ["AWS_REGION"]
target_profile = os.environ["TARGET_PROFILE"]
default_profile = os.environ["DEFAULT_PROFILE"]
role_arn = os.environ["TARGET_ROLE_ARN"]
bucket = os.environ["TARGET_BUCKET"]
image_uri = os.environ["IMAGE_URI"]
model_name = f"modelbuilder-session-repro-{uuid.uuid4().hex[:8]}"

# Make the process default session use a different account.
boto3.setup_default_session(profile_name=default_profile, region_name=region)

target_boto_session = boto3.Session(
    profile_name=target_profile,
    region_name=region,
)
target_sagemaker_session = Session(
    boto_session=target_boto_session,
    default_bucket=bucket,
)

target_account = target_boto_session.client("sts").get_caller_identity()["Account"]
default_account = boto3.client("sts", region_name=region).get_caller_identity()["Account"]
print(f"Target account: {target_account}")
print(f"Default account: {default_account}")

source_model = Model(
    model_name=f"{model_name}-source",
    primary_container=ContainerDefinition(image=image_uri),
    execution_role_arn=role_arn,
)
builder = ModelBuilder(
    model=[source_model],
    role_arn=role_arn,
    sagemaker_session=target_sagemaker_session,
    instance_type="ml.m5.large",
)

try:
    builder.build(model_name=model_name)
finally:
    # The create request succeeds before build() raises.
    try:
        target_sagemaker_session.sagemaker_client.delete_model(
            ModelName=model_name,
        )
    except ClientError as error:
        if error.response["Error"]["Code"] != "ValidationException":
            raise

The script raises an error similar to:

botocore.exceptions.ClientError: An error occurred (ValidationException)
when calling the DescribeModel operation:
Could not find model "modelbuilder-session-repro-1234abcd".

Before cleanup, DescribeModel succeeds through target_sagemaker_session.
The model is present in the target account.

Expected behavior

ModelBuilder.build() must use the configured session for all operations.
The method must return the created Model resource from the target account.

The reload call can pass the boto session:

return Model.get(
    model_name=self.model_name,
    session=self.sagemaker_session.boto_session,
    region=self.region,
)

Suggested patch:

diff --git a/src/sagemaker/serve/model_builder.py b/src/sagemaker/serve/model_builder.py
--- a/src/sagemaker/serve/model_builder.py
+++ b/src/sagemaker/serve/model_builder.py
@@
-        return Model.get(model_name=self.model_name, region=self.region)
+        return Model.get(
+            model_name=self.model_name,
+            session=self.sagemaker_session.boto_session,
+            region=self.region,
+        )

The regression test must configure different default and target boto sessions.
It must make sure that Model.get() receives the target boto session after
CreateModel.

Screenshots or logs

Target account: <target-account-id>
Default account: <different-account-id>
Creating model with name: modelbuilder-session-repro-1234abcd
Model has been created: modelbuilder-session-repro-1234abcd
ValidationException: Could not find model
"modelbuilder-session-repro-1234abcd".

CloudTrail shows CreateModel in the target account. The failing
DescribeModel request uses the default account.

System information

  • SageMaker Python SDK version: 3.24.0
  • Framework name (eg. PyTorch) or algorithm (eg. KMeans): Not framework-specific
  • Framework version: N/A
  • Python version: 3.12.0
  • CPU or GPU: CPU, Apple Silicon (arm64)
  • Custom Docker image (Y/N): N

Additional context

The failure is hidden when the default boto session uses the same account as
the configured SageMaker session.

This also affects multi-container inference pipelines. CreateModel completes
in the target account, but endpoint creation never starts because the model
reload fails first.

Lingua principale
Python
Stelle
2.3k
Fork
1.3k
Merge medio
3g 7h
PR unite (30g)
86

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di aws/sagemaker-python-sdk

Tutte le issue di aws/sagemaker-python-sdk

Issue simili

Altre issue su Python

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.