Edit on GitHub

fetch

Download files or directories from remote storage to the cache.

Synopsis

usage: dvc fetch [-h] [-q | -v] [-j <number>] [-r <name>] [-a] [-T]
                 [--all-commits] [-d] [-R] [--run-cache | --no-run-cache]
                 [--max-size <bytes>] [--type {metrics,plots}]
                 [targets [targets ...]]

positional arguments:
  targets       Limit command scope to these tracked files/directories,
                .dvc files, or stage names.

Description

Downloads tracked files and directories from a dvc remote into the cache (without placing them in the workspace like dvc pull). This makes the tracked data available for linking (or copying) into the workspace (see dvc checkout).

Note that dvc pull includes fetching.

Tracked files                Commands
---------------- ---------------------------------

remote storage
     +
     |         +------------+
     | - - - - | dvc fetch  | ++
     v         +------------+   +   +----------+
project's cache                  ++ | dvc pull |
     +         +------------+   +   +----------+
     | - - - - |dvc checkout| ++
     |         +------------+
     v
 workspace

Here are some scenarios in which dvc fetch is useful, instead of pulling:

  • After checking out a fresh copy of a DVC repository, to get DVC-tracked data from multiple project branches or tags into your machine.
  • To use comparison commands across different Git commits, for example dvc metrics show with its --all-branches option, or dvc plots diff.
  • If you want to avoid linking files from the cache, or keep the workspace clean for any other reason.

Without arguments, it downloads all files and directories referenced in the current workspace (found in dvc.yaml and .dvc files) that are missing from the workspace. Any targets given to this command limit what to fetch. It accepts paths to tracked files or directories (including paths inside tracked directories), .dvc files, and stage names (found in dvc.yaml).

The --all-branches, --all-tags, and --all-commits options enable fetching files/dirs referenced in multiple Git commits.

The dvc remote used is determined in order, based on

  1. the remote fields in the dvc.yaml or .dvc files.
  2. the value passed to the --remote option via CLI.
  3. the value of the core.remote config option (see dvc remote default).

Options

  • -r <name>, --remote <name> - name of the dvc remote to fetch from (see dvc remote list).

  • -d, --with-deps - only meaningful when specifying targets. This determines files to download by resolving all dependencies of the targets: DVC searches backward from the targets in the corresponding pipelines. This will not fetch files referenced in later stages than the targets.

  • -R, --recursive - determines the files to fetch by searching each target directory and its subdirectories for dvc.yaml and .dvc files to inspect. If there are no directories among the targets, this option has no effect.

  • --run-cache, --no-run-cache - whether to download all available history of stage runs from the remote repository. See the same option in dvc push. Default is โ€œโ€”no-run-cache`.

  • -j <number>, --jobs <number> - parallelism level for DVC to download data from remote storage. The default value is 4 * cpu_count(). Note that the default value can be set using the jobs config option with dvc remote modify. Using more jobs may speed up the operation.

  • -a, --all-branches - fetch cache for all Git branches, as well as for the workspace. This means DVC may download files needed to reproduce different versions of a .dvc file, not just the ones currently in the workspace. Note that this can be combined with -T below, for example using the -aT flags.

  • -T, --all-tags - fetch cache for all Git tags, as well as for the workspace. Note that this can be combined with -a above, for example using the -aT flags.

  • -A, --all-commits - fetch cache for all Git commits, as well as for the workspace. This downloads tracked data for the entire commit history of the project.

  • --max-size <bytes> - fetch data files/directories that are each below specified size (bytes). Note that the size is determined by a corresponding size field in the .dvc/dvc.lock file. Which means that even if some files or subdirectories are smaller inside a DVC-tracked directory, the whole directory is still skipped.

  • --type <type> - fetch data files/directories that are of a particular type. Currently only metrics and plots are supported.

  • -h, --help - prints the usage/help message, and exit.

  • -q, --quiet - do not write anything to standard output. Exit with 0 if no problems arise, otherwise 1.

  • -v, --verbose - displays detailed tracing information.

Examples

Let's employ a simple workspace with some data, code, ML models, pipeline stages, such as the DVC project created for the Get Started. Then we can see what dvc fetch does in different scenarios.

Start by cloning our example repo if you don't already have it:

$ git clone https://github.com/iterative/example-get-started
$ cd example-get-started

The workspace looks like this:

.
โ”œโ”€โ”€ data
โ”‚   โ””โ”€โ”€ data.xml.dvc
โ”œโ”€โ”€ dvc.lock
โ”œโ”€โ”€ dvc.yaml
โ”œโ”€โ”€ params.yaml
โ”œโ”€โ”€ prc.json
โ”œโ”€โ”€ scores.json
โ””โ”€โ”€ src
    โ””โ”€โ”€ <code files here>

This project comes with a predefined HTTP remote storage. We can now just run dvc fetch to download the most recent model.pkl, data.xml, and other DVC-tracked files into our local cache.

$ dvc status --cloud
...
	deleted:            data/features/train.pkl
	deleted:            model.pkl

$ dvc fetch

$ tree .dvc/cache/files/md5
.dvc/cache/files/md5
โ”œโ”€โ”€ 20
โ”‚   โ””โ”€โ”€ b786b6e6f80e2b3fcf17827ad18597.dir
โ”œโ”€โ”€ c8
โ”‚    โ”œโ”€โ”€ d307aa005d6974a8525550956d5fb3
โ”‚    โ””โ”€โ”€ ...
...

dvc status --cloud compares the cache contents against the default remote. Refer to dvc status.

Note that the .dvc/cache directory was created and populated.

All the data needed in this version of the project is now in your cache: File names 20b786b... and c8d307a... correspond to the data/features/ directory and model.pkl file, respectively.

To link these files to the workspace:

$ dvc checkout

Example: Specific files or directories

If you tried the previous example, delete the .dvc/cache directory first (e.g. rm -Rf .dvc/cache) to follow this one.

dvc fetch only downloads the tracked data corresponding to any given targets:

$ dvc fetch prepare

$ tree .dvc/cache/files/md5
.dvc/cache/files/md5
โ”œโ”€โ”€ 20
โ”‚   โ””โ”€โ”€ b786b6e6f80e2b3fcf17827ad18597.dir
โ”œโ”€โ”€ 32
โ”‚   โ””โ”€โ”€ b715ef0d71ff4c9e61f55b09c15e75
โ””โ”€โ”€ 6f
    โ””โ”€โ”€ 597d341ceb7d8fbbe88859a892ef81

Cache entries for the data/prepared directory (output of the prepare target), as well as the actual test.tsv and train.tsv files, were downloaded. Their hash values are shown above.

Note that you can fetch data within directories tracked. For example, the featurize stage has the entire data/features directory as output, but we can just get this:

$ dvc fetch data/features/test.pkl

If you check again .dvc/cache, you'll see a couple more files were downloaded: the cache entries for the data/features directory, and data/features/test.pkl itself.

Example: With dependencies

After following the previous example (Specific stages), only the files associated with the prepare stage have been fetched. Several dependencies/outputs of other pipeline stages are still missing from the cache:

$ dvc status -c
...
    deleted:            data/features/test.pkl
    deleted:            data/features/train.pkl
    deleted:            model.pkl

One could do a simple dvc fetch to get all the data, but what if you only want to retrieve the data up to our third stage, train? We can use the --with-deps (or -d) option:

$ dvc fetch --with-deps train

$ tree .dvc/cache/files/md5
.dvc/cache/files/md5
โ”œโ”€โ”€ 20
โ”‚   โ””โ”€โ”€ b786b6e6f80e2b3fcf17827ad18597.dir
โ”œโ”€โ”€ c8
โ”‚   โ”œโ”€โ”€ 43577f9da31eab5ddd3a2cf1465f9b
โ”‚   โ””โ”€โ”€ d307aa005d6974a8525550956d5fb3
โ”œโ”€โ”€ 32
โ”‚   โ””โ”€โ”€ b715ef0d71ff4c9e61f55b09c15e75
โ”œโ”€โ”€ 54
โ”‚   โ””โ”€โ”€ c0f3ef1f379563e0b9ba4accae6807
โ”œโ”€โ”€ 6f
โ”‚   โ””โ”€โ”€ 597d341ceb7d8fbbe88859a892ef81
โ”œโ”€โ”€ a1
โ”‚   โ””โ”€โ”€ 414b22382ffbb76a153ab1f0d69241.dir
โ””โ”€โ”€ a3
    โ””โ”€โ”€ 04afb96060aad90176268345e10355

Fetching using --with-deps starts with the target stage (train) and searches backwards through its pipeline for data to download into the project's cache. All the data for the second and third stages (featurize and train) has now been downloaded to the cache. We could now use dvc checkout to get the data files needed to reproduce this pipeline up to the third stage into the workspace (with dvc repro train).

Note that in this example project, the last stage evaluate doesn't add any more data files than those form previous stages, so at this point all of the data for this pipeline is cached and dvc status -c would output Cache and remote 'storage' are in sync.

Content

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