Skip to content

armgit / dify

Public Imported from GitHub
0

Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.

2.8 KB
Raw Blame
import base64
import hashlib
from collections.abc import Generator
from typing import override

from baidubce.auth.bce_credentials import BceCredentials
from baidubce.bce_client_configuration import BceClientConfiguration
from baidubce.services.bos.bos_client import BosClient

from configs import dify_config
from extensions.storage.base_storage import BaseStorage


class BaiduObsStorage(BaseStorage):
    """Implementation for Baidu OBS storage."""

    def __init__(self):
        super().__init__()
        self.bucket_name = dify_config.BAIDU_OBS_BUCKET_NAME
        client_config = BceClientConfiguration(
            credentials=BceCredentials(
                access_key_id=dify_config.BAIDU_OBS_ACCESS_KEY,
                secret_access_key=dify_config.BAIDU_OBS_SECRET_KEY,
            ),
            endpoint=dify_config.BAIDU_OBS_ENDPOINT,
        )

        self.client = BosClient(config=client_config)

    @override
    def save(self, filename, data):
        md5 = hashlib.md5()
        md5.update(data)
        content_md5 = base64.standard_b64encode(md5.digest())
        self.client.put_object(
            bucket_name=self.bucket_name, key=filename, data=data, content_length=len(data), content_md5=content_md5
        )

    @override
    def load_once(self, filename: str) -> bytes:
        response = self.client.get_object(bucket_name=self.bucket_name, key=filename)
        data: bytes = response.data.read()
        return data

    @override
    def load_stream(self, filename: str) -> Generator:
        response = self.client.get_object(bucket_name=self.bucket_name, key=filename).data
        while chunk := response.read(4096):
            yield chunk

    @override
    def download(self, filename, target_filepath):
        self.client.get_object_to_file(bucket_name=self.bucket_name, key=filename, file_name=target_filepath)

    @override
    def exists(self, filename):
        res = self.client.get_object_meta_data(bucket_name=self.bucket_name, key=filename)
        return res is not None

    @override
    def delete(self, filename: str):
        self.client.delete_object(bucket_name=self.bucket_name, key=filename)

    @override
    def scan(self, path: str, files: bool = True, directories: bool = False) -> list[str]:
        keys: list[str] = []
        marker = ""
        while True:
            page = self.client.list_objects(bucket_name=self.bucket_name, prefix=path, marker=marker, max_keys=1000)
            keys.extend(obj.key for obj in page.contents)
            if not page.is_truncated:
                break
            if not page.next_marker or page.next_marker == marker:
                raise RuntimeError("BOS object listing did not advance its pagination marker")
            marker = page.next_marker
        return self._filter_scan_keys(keys, files, directories)