Skip to content

Publishing packages to your own channel#

Rattler-build comes with an intuitive publish subcommand, that will publish a package to a channel.

You can either point to an already built package, or to a recipe and publish them into your channel. The channel can be either on prefix.dev, anaconda.org, an S3 bucket, a local filesystem folder (or network mount), or a Quetz or Artifactory instance.

Publishing a package is a convenience short hand for:

  1. Building the package from a recipe.yaml
  2. Optionally, automatically increment the build number by passing in --build-number=+1 to set a relative value, or --build-number=12 to set an absolute value for all packages being built.
  3. Uploading the package to a channel
  4. Running rattler-index for a S3 bucket or a filesystem channel to produce repodata.json.

To publish a package you can use:

rattler-build publish ./my-recipe.yaml --to https://prefix.dev/my-channel

rattler-build publish ./output/linux-64/my-package-0.1.2-h123_0.conda --to s3://my-bucket

rattler-build publish ./some/recipe.yaml --to artifactory://my-secret.company.com/package-channel

# on prefix.dev you can also automatically add a sigstore attestation
rattler-build publish ./my-recipe.yaml --to https://prefix.dev/my-channel --generate-attestation

The following schema is used:

  • prefix.dev: https://prefix.dev/<channel-name>
  • anaconda.org: https://anaconda.org/<owner>/<label (optional)> (e.g. https://anaconda.org/foobar)
  • S3: s3://bucket-name (note: we read the standard S3 configuration / environment variables for region, authentication, etc.)
  • Filesystem: file:///path/to/channel
  • Quetz: quetz://server.my-company.com/<channel>
  • Articatory: artifactory://server.my-company.com/<channel>

Options#

The --to option selects the channel to publish the package into. This channel will also be used as the highest priority channel in the list of channels. Other channels can be added using the usual -c conda-forge -c bioconda ... syntax or configured using a config.toml file.

When using publish with a recipe, you can use the same options as when normally building packages.

Bumping the build number#

Sometimes you want to package the same package again, but rebuild it with the latest dependencies. There are two ways to achieve this:

  • By using a variant file, and updating it at certain times
  • By bumping the build number and re-publishing the package again

The publish command makes it easy to "bump the buildnumber" either by setting an absolute build number for all packages the recipe builds (e.g. --build-number=12) or a relative bump for all packages (e.g. --build-number=+1 to add 1 to the highest build number found in the publish channel). When bumping by a relative amount, we download the repodata and determine for each subdir/package combination that you are building the highest build number.

When the recipe does not specify a build number, the build number is automatically bumped on publish to the next available build number.

If the recipe does specify a build number, you have to manually trigger an override using the --build-number CLI flag. Alternatively, you can use the --force upload option on S3, your local filesystem, Anaconda and prefix channels to forcibly replace the previous build. Please note that this is heavily discouraged as lockfiles will get out of date and the old build is irrevocably deleted.

Authentication#

Rattler-build uses the same authentication as other tools in the prefix family. It's easiest to login using the auth subcommand: rattler-build auth login. Note: if you are already logged in with pixi, you are also logged in with rattler-build - they share credentials.

Otherwise you can also use the same options as with upload, and supply tokens as environment variables or CLI arguments.

Indexing S3 and Filesystem channels#

Since S3 and Filesystem channels don't know anything about "indexing" (producing repodata.json), rattler-build will internally use rattler-index to run the indexing step after a successful upload. This will ensure that the repodata in the channel is up to date and users can start downloading the new packages.