Skip to main content
File storage is a network-shared storage that can be mounted to different instances within the same region

Advantages

1 Shared across instances with multi-point read/write capabilities, unaffected by instance release.2 High data reliability due to multiple redundant copies on the storage backend (Local Data Disks have no redundancy and are more prone to disk failures).

Disadvantages

Generally lower IO performance.

​
Recommendation

  • Store your important data and code in file storage for shared access and high reliability. For training data requiring high IO performance, copy it to the local data disk of the instance to achieve better IO performance. This approach balances convenience, safety, and performance.

​
Initialization

Enter the Console -> File Storage menu, and initialize the file storage for the corresponding region. After creating an instance in that region and starting it, the file storage will be mounted to the /root/autodl-fs directory of the instance. 
/root/autodl-fs

​
Not Mounted

  1. Check whether file storage is supported for instances in the region.
  2. Check whether the file storage interface has been initialized for the region.
  3. If the instance was started before the initialization of file storage, you need to restart the instance.
  4. Check the mount status in Account → Settings → File Storage. If the status is “Unmounted”, remount the file storage and then restart the instance.
  5. For instances created by a sub-account, verify that the sub-account has file storage permissions in Account → Sub-accounts → Configure Permissions → File Storage.
  6. If the file storage still cannot be mounted, contact customer support for assistance.

​
Cache Cleanup

If a web upload is interrupted or files are deleted in JupyterLab, cache files or trash may be generated, occupying file storage space.

​
Solution 1: From Console

File Storage → More Details → Clean Upload Cache

​
Solution 2: From Terminal

Navigate to /root/autodl-fs, use ls -alh to view hidden files, then permanently delete them with rm.

​
Disk Full Error

A “disk full” error may unexpectedly occur even when there is sufficient storage space, if the number of files (inodes) exceeds 200,000.
File storage has an inode limit of 200,000. Since a large number of small files significantly impacts read/write performance, we recommend compressing files before storing them in file storage, and keeping small files on the data disk instead. You can monitor your inode usage at any time in: File Storage → More Details.

​
Billing

For details on data disk billing, please refer to the pricing.

​
Data Cleanup

If the account has not logged in for three consecutive months or has an outstanding fee of more than $10.00, the data in the file storage will be cleared.