dvccache

dvccache refers to the data cache used by DVC (Data Version Control). It is a content-addressable storage system that stores data artifacts and pipeline outputs to speed up operations, reduce redundant downloads or computations, and enable efficient sharing of data across projects and environments.

The dvccache is typically located in a project’s .dvc/cache directory by default, but its location can be configured through DVC_CACHE_DIR or related settings. The cache is designed to be persistent and shareable across workspaces; multiple projects can point to the same cache location if they are configured to do so. Cache contents are identified by a hash of the data, usually a SHA-256 digest, ensuring that identical content is stored only once. To prevent directory bloat, the cache uses a two-level layout where the first two characters of the hash form a subdirectory, and the remaining characters form the file name.

When data is added to a DVC project (for example, with dvc add), DVC computes the content hash and stores the corresponding file in the dvccache if it is not already present. The working directory then contains lightweight pointers or links to the cached content, rather than duplicating the data. DVC can retrieve data from the cache or from remote caches (for example, S3, GCS, Azure, SSH, or HDFS) as needed. Cache behavior, including linking method (symlink, hardlink, reflink, or copy) and remote cache configuration, can be customized to suit the operating system and performance goals.

Cache management commands such as dvc gc prune unused cache entries help control disk usage. The dvccache is central to DVC’s ability to reproduce experiments, share datasets, and optimize data-intensive workflows, while requiring careful monitoring of storage requirements and cache health.