High-performance distributed file storage with snapshot and efficient deletion features for CKS workloads
This page describes Distributed File Storage, its features, performance considerations, and billing model. Use it to decide whether Distributed File Storage fits your CKS workload before you provision volumes.Distributed File Storage delivers high write speeds, high IOPS for datapath-related operations, and higher read speeds in cases where block size and queue depth are low. Distributed File Storage is ideal for operations that require frequent synchronization between Pods, and it minimizes performance bottlenecks caused by contention.
Scope of Distributed File Storage
Storage access across or is not supported. You can only access a Distributed File Storage volume from within the cluster where it’s deployed.
All data stored in Distributed File Storage volumes is automatically encrypted at rest using AES-XTS 256-bit encryption with FIPS 140-3 validated encryption keys. Encryption is enabled by default with no configuration required.VAST automatically generates and manages encryption keys and applies them at the cluster level. Keys aren’t accessible to users or operators.Distributed File Storage doesn’t support encryption for data in transit between Pods and the storage backend.
Every Distributed File Storage volume features a folder called .snapshot, which contains a snapshot in time of the volume. This folder is accessible in any directory on the volume.Snapshots of Distributed File Storage volumes occur every 6 hours and persist for 72 hours.
The trash directory allows asynchronous deletion of files and folders in bulk. Serial rm -rf on large directories is a slow and possibly blocking operation. Moving files marked for removal to the trash folder (.vast_trash) is much faster. For more information, see Delete files with VAST trash.
Distributed File Storage accepts POSIX-valid filenames, including names that contain characters such as : and *. Applications that generate ISO 8601 timestamps (for example, 2026-04-21T14:30:00.log) or other filenames with special characters work without modification.The / character is reserved as the path separator, and filenames can’t be empty.
Avoid creating flat directories with many entries in a single directory. The recommended maximum is 500,000 to 1,000,000 entries per directory. Exceeding this limit can cause performance degradation and unpredictable behavior depending on the storage backend. It also makes operations such as listing or iterating over files in that directory slower.If your workload naturally produces many files, organize them into a hierarchy of subdirectories to keep entry counts within this range.
This section explains how CoreWeave calculates charges for Distributed File Storage so you can estimate costs for your provisioned volumes.Unless stated otherwise in a customer contract, DFS is billed based on the total provisioned capacity of all customer Persistent Volume Claims (PVCs).DFS is billed based on hourly total provisioned capacity. To calculate the hourly rate, DFS measures the total provisioned capacity of all customer PVCs every 30 seconds and then averages those capacity measurements across the hour.
