Edit on GitHub

Internal Directories and Files

Once initialized in a project, DVC populates its installation directory (.dvc/) with the internal directories and files needed for DVC operation.

Not to be confused with .dvc files.

  • .dvc/config: This is the default DVC configuration file. It can be edited by hand or with the dvc config command.

  • .dvc/config.local: This is an optional Git-ignored configuration file, that will overwrite options in .dvc/config. This is useful when you need to specify sensitive values (secrets) which should not reach the Git repo (credentials, private locations, etc). This config file can also be edited by hand or with dvc config --local.

  • .dvc/cache: Default location of the cache directory. The cache stores the project data in a special structure. The data files and directories in the workspace will only contain links to the data files in the cache (refer to Large Dataset Optimization. See dvc config cache for related configuration options, including changing its location.

    Note that DVC includes the cache directory in .gitignore during initialization. No data tracked by DVC should ever be pushed to the Git repository, only the DVC files that are needed to download or reproduce that data.

  • .dvc/cache/runs: Default location of the run cache.

  • .dvc/plots: Directory for plot templates

  • .dvc/tmp: Directory for miscellaneous temporary files

  • .dvc/tmp/updater: This file is used to store the latest available version of DVC. It's used to remind the user to upgrade when the installed version is behind.

  • .dvc/tmp/updater.lock: Lock file for .dvc/tmp/updater

  • .dvc/tmp/lock: Lock file for the entire DVC project

  • .dvc/tmp/rwlock: JSON file that contains read and write locks for specific dependencies and outputs, to allow safely running multiple DVC commands in parallel

  • .dvc/tmp/exps: This directory will contain workspace copies used for temporary or queued experiments.

Structure of the cache directory

The DVC cache is a content-addressable storage (by default in .dvc/cache), which adds a layer of indirection between code and data.

There are two ways in which the data is cached, depending on whether it's a single file, or a directory (which may contain multiple files).

Note files are renamed, reorganized, and directory trees are flattened in the cache, which always has exactly one depth level with 2-character directories (based on hashes of the data contents, as explained next).


DVC calculates the file hash, a 32 characters long string (usually MD5). The first two characters are used to name the directory inside the cache, and the rest become the file name of the cached file. For example, if a data file has a hash value of ec1d2935f811b77cc49b031b999cbf17, its path in the cache will be .dvc/cache/files/md5/ec/1d2935f811b77cc49b031b999cbf17.

Note that file hashes are calculated from file contents only. 2 or more files with different names but the same contents can exist in the workspace and be tracked by DVC, but only one copy is stored in the cache. This helps avoid data duplication.


Let's imagine adding a directory with 2 images:

$ tree data/images/
โ”œโ”€โ”€ cat.jpeg
โ””โ”€โ”€ index.jpeg

$ dvc add data/images

The resulting cache dir looks like this:

โ””โ”€โ”€ files
    โ””โ”€โ”€ md5
        โ”œโ”€โ”€ 40
        โ”‚   โ””โ”€โ”€ 2e97968614f583ece3b35555971f64
        โ”œโ”€โ”€ 6f
        โ”‚   โ””โ”€โ”€ db5336fce0dbfd669f83065f107551.dir
        โ””โ”€โ”€ de
            โ””โ”€โ”€ 7371b0119f4f75f9de703c7c3bac16

The files in the directory are cached normally. The directory itself gets a similar entry, with the .dir extension. It contains the mapping of files inside (as a JSON array), identified by their hash values:

$ cat .dvc/cache/files/md5/6f/db5336fce0dbfd669f83065f107551.dir
[{"md5": "de7371b0119f4f75f9de703c7c3bac16", "relpath": "cat.jpeg"},
{"md5": "402e97968614f583ece3b35555971f64", "relpath": "index.jpeg"}]

That's how DVC knows that the other two cached files belong in the directory.

Run cache

dvc exp run and dvc repro by default populate and reutilize a log of stages that have been run in the project. It is found in the runs/ directory inside the cache (or remote storage).

Runs are identified as combinations of exact dependency contents (or parameter values), and the literal command(s) to execute. These combinations are represented by special hashes that translate to the file paths inside the run cache dir:

$ tree .dvc/cache/runs
โ””โ”€โ”€ 86
    โ””โ”€โ”€ 8632e1555283d6e23ec808c9ee1fadc30630c888d5c08695333609ef341508bf
        โ””โ”€โ”€ e98a34c44fa6b564ef211e76fb3b265bc67f19e5de2e255217d3900d8f...

The files themselves are backups of the dvc.lock file that resulted from that run.

The run's outputs are stored and retrieved from the regular cache.

dvc push and dvc pull (and dvc fetch) can download and upload the run cache to remote storage using the --run-cache flag, for sharing and/or as a back up.

The run cache assumes that stage commands are deterministic (see Avoiding unexpected behavior).

Site cache dir

Directory used for storing temporary files.

Default location:

macOSLinux (typical*)Windows

This location may be overridden with dvc config core.site_cache_dir.

Not to be confused with DVC cache.

  • index: Directory for remote index files that are used for optimizing dvc push, dvc pull, dvc fetch and dvc status -c operations.

  • hashes: This directory is used for optimization. It contains a SQLite state database that stores hash values for files tracked in a DVC project. It also saves the corresponding timestamps and inodes to avoid unnecessary file hash computations.

  • links: This directory is used to clean up your workspace when calling dvc checkout. It contains a SQLite state database that stores a list of file links created by DVC (from cache to workspace).


๐Ÿ› Found an issue? Let us know! Or fix it:

Edit on GitHub

โ“ Have a question? Join our chat, we will help you:

Discord Chat