Implementation:Bentoml BentoML YataiClient

Knowledge Sources	Bentoml_BentoML
Domains	Cloud, Model Management, Bento Management, Remote Storage
Last Updated	2026-02-13 15:00 GMT

Overview

The YataiClient class provides push and pull operations for synchronizing both BentoML Bentos and Models between local stores and a remote Yatai server.

Description

YataiClient is the legacy client for interacting with the Yatai model and bento management server. It implements four main operations:

• push_bento(): Pushes a Bento (including its dependent models) to the Yatai server. Models are pushed concurrently using a thread pool before the Bento archive itself is uploaded.
• pull_bento(): Pulls a Bento from the Yatai server, downloading dependent models concurrently and extracting the Bento tar archive into the local store.
• push_model(): Pushes a single model to the Yatai server using proxy, presigned URL, or multipart upload strategies.
• pull_model(): Pulls a single model from the Yatai server with support for version resolution (including "latest"), presigned URL downloads, and local caching.

All upload methods support three transmission strategies: proxy (upload through the Yatai API server), presigned_url (direct upload to object storage), and multipart (chunked upload for large files). Progress is tracked through a Spinner with rich progress bars. Thread-safe I/O progress tracking is ensured via mutex locks.

Usage

Used internally by BentoML when users interact with the Yatai server (the legacy centralized model/bento management platform). This is the backend for bentoml push and bentoml pull CLI commands when a Yatai server context is configured.

Code Reference

Source Location

• Repository: Bentoml_BentoML
• File: src/bentoml/_internal/cloud/yatai.py
• Lines: 1-977

Signature

class YataiClient:
    def __init__(self) -> None: ...

    def push_bento(self, bento: Bento, *, force: bool = False,
                   threads: int = 10, context: str | None = None) -> None: ...

    def pull_bento(self, tag: str | Tag, *, force: bool = False,
                   context: str | None = None,
                   bento_store: BentoStore = ...) -> Bento: ...

    def push_model(self, model: Model, *, force: bool = False,
                   threads: int = 10, context: str | None = None) -> None: ...

    def pull_model(self, tag: str | Tag, *, force: bool = False,
                   context: str | None = None,
                   model_store: ModelStore = ...,
                   query: str | None = None) -> Model: ...

Import

from bentoml._internal.cloud.yatai import YataiClient

I/O Contract

Inputs

push_bento()

Name	Type	Required	Description
bento	Bento	Yes	The Bento to push to Yatai, including its dependent models
force	bool	No	Force push even if the Bento already exists on Yatai (default: False)
threads	int	No	Number of threads for multipart upload (default: 10)
context	str or None	No	The Yatai context/endpoint configuration name

pull_bento()

Name	Type	Required	Description
tag	str or Tag	Yes	The tag of the Bento to pull
force	bool	No	Force pull even if the Bento exists locally (default: False)
context	str or None	No	The Yatai context/endpoint configuration name
bento_store	BentoStore	No	The local bento store (injected via DI)

push_model()

Name	Type	Required	Description
model	Model	Yes	The Model to push to Yatai
force	bool	No	Force push even if the model already exists (default: False)
threads	int	No	Number of threads for multipart upload (default: 10)
context	str or None	No	The Yatai context/endpoint configuration name

pull_model()

Name	Type	Required	Description
tag	str or Tag	Yes	The tag of the model to pull
force	bool	No	Force pull even if the model exists locally (default: False)
context	str or None	No	The Yatai context/endpoint configuration name
model_store	ModelStore	No	The local model store (injected via DI)
query	str or None	No	Query string for filtering the latest model version

Outputs

Method	Return Type	Description
push_bento()	None	Uploads the Bento and its models; raises BentoMLException on failure
pull_bento()	Bento	The pulled Bento instance saved to the local store
push_model()	None	Uploads the model; raises BentoMLException on failure
pull_model()	Model	The pulled Model instance saved to the local store

Usage Examples

from bentoml._internal.cloud.yatai import YataiClient
import bentoml

# Create client instance
yatai_client = YataiClient()

# Push a bento to Yatai
bento = bentoml.bentos.get("my_service:latest")
yatai_client.push_bento(bento, force=False, context="default")

# Pull a bento from Yatai
pulled_bento = yatai_client.pull_bento("my_service:v1", context="default")

# Push a model to Yatai
model = bentoml.models.get("my_model:latest")
yatai_client.push_model(model, force=False, context="default")

# Pull a model from Yatai
pulled_model = yatai_client.pull_model("my_model:latest", context="default")

Related Pages

• Implementation:Bentoml_BentoML_Cloud_ModelAPI
• Implementation:Bentoml_BentoML_Configuration_Containers