Multimodal dataset management
MaxCompute supports storing unstructured data, such as images, videos, audio files, PDFs, and Office documents, in tables by using the Blob data type. The client provides features like a file browser, multi-format previews, upload and import capabilities, bulk downloads, and local mounting through WebDAV to help you manage and use multimodal datasets.
Blob multimodal storage

Prerequisites
You have created and successfully connected to a MaxCompute data source.
Enable Blob multimodal storage
Go to Settings > Blob Storage.
Enable the File Browser toggle.
After this feature is enabled:
The Blob Multimodal Storage entry appears in the Multimodal group in the sidebar. Click it to open the file browser.
The read-only WebDAV service starts simultaneously.
You can configure the following settings in this section:
Parameter | Description | Default |
Blob file cache time | The duration (in hours) for storing files in the local disk cache. Set to 0 to disable the cache. | 12 hours |
Blob cache size | The maximum disk space that the cache can use. | 2 GB |
Clear cache | Manually clears all cached Blob files and thumbnails. | — |
Create a multimodal table
A multimodal table is a MaxCompute table with one or more Blob columns for storing unstructured data. The table structure consists of the following components:
Blob columns: Store the binary content of files. Each multimodal table must have at least one Blob column.
Companion metadata columns: Store information such as file name (STRING), file type (STRING), and file size (BIGINT) for queries and filtering.
Tables with a primary key: You can set the file name or a unique identifier as the primary key. This allows you to automatically overwrite existing records when you re-import a file with the same name, enabling data deduplication.
Tables without a primary key: Each import appends new rows. This is suitable for scenarios where deduplication is not required.
Multimodal storage tables are named with the fs_ prefix.
Upload files
In the file browser, you can upload files to a multimodal table in the following ways:
Click to upload: Click the Upload button in the toolbar and select local files.
Drag and drop: Drag files directly into the file browser area.
File browser
Access: Sidebar > Blob Multimodal Storage.
The file browser displays a directory structure of the fs_* file storage tables in the current project. It supports the following features:
Grid View / List View: Switch between file display modes.
Breadcrumb navigation: Click any level in the path to navigate quickly.
View and preview: Click a file to open a preview tab and view its content.
Download: Click the download button for a file to save it locally. You can also select multiple files for a bulk download.
WebDAV mount
The client starts a local, read-only WebDAV service, allowing you to browse MaxCompute files directly from your operating system's file manager.
Connect on macOS:
In Finder, press ⌘K (Go > Connect to Server).
Enter the address:
http://127.0.0.1:8899/dav/(you can copy this address from the settings page).Click Connect. You can now browse and download files as if they were in a local folder.
Limitations:
Only read-only operations, such as browsing and downloading, are supported.
Access is limited to the local machine (127.0.0.1).
Closing the application stops the WebDAV service and terminates the connection.
General import
Upload local files to Blob
Access: Toolbar > Import > Blob Import tab.
Use the Blob Import view to bulk import local files into a Blob column of a MaxCompute table.
Drag and drop or select local files. Any binary file is supported, including images, PDFs, compressed archives, and audio or video files.
Select an import mode:
New Table: Configure the table name and columns. The client automatically generates the DDL for table creation, including a Blob column and companion metadata columns.
Existing Table: Select a target table and map the file content to an existing Blob column.
After confirming the field mapping, click Start Import.
The following is a typical DDL structure for a new table:
CREATE TABLE my_images (
file_name STRING COMMENT 'File name',
file_type STRING COMMENT 'MIME type',
file_size BIGINT COMMENT 'File size in bytes',
content BLOB COMMENT 'File content'
);Manage multimodal datasets with SQL
Query Blob data with SQL
You can retrieve and filter multimodal data using SQL to query tables that contain Blob columns. For details on using the Blob data type in SQL, see Blob Data Type and Multimodal Storage.
Preview and download query results
If a SQL query result contains Blob columns, the results panel provides the following features:
Thumbnail display: Blob cells automatically show a file type icon or an image thumbnail.
Cell download: Click the download button in a cell to save a single file.
Column bulk download: Select multiple files in a Blob column to download them in bulk to a local directory.
Blob/Binary preview tab
Click the preview button in a Blob cell in the query results or a file in the file browser to open a dedicated preview tab. Supported preview formats include:
Category | Supported types |
Image | JPEG, PNG, GIF, WebP, BMP, TIFF |
Video | MP4, WebM, MPEG |
Audio | WAV, MP3, OGG, FLAC |
Document | PDF, Word (docx), Excel (xlsx), PPT (pptx) |
Text | JSON, XML, CSV, TSV, Markdown, plain text |
Binary | HEX (hexadecimal) view |
The preview tab allows you to switch between preview modes, including Auto-detect, Image, Video, Audio, Document, Text, and HEX. This is useful if the file type is unknown or you need to view the content in a different format.