Conda Release Infrastructure

This doc is for developers who need to understand or modify the conda release infrastructure specified in conda-release.md.

Key Files

Path	Purpose
conda/conda-recipes/<package>/meta.yaml	Conda recipe for each package (or bundle). Defines build, dependencies, and tests.
conda/conda-recipes/conda_env.yml	Stores AZURESDK_CONDA_VERSION — the version string used for all conda packages in a release (e.g., 2025.12.01).
conda/conda-releaselogs/<package>.md	Per-package (or bundle) release log listing which PyPI package versions are included in each conda release.
conda/update_conda_files.py	The main automation script. Updates versions, creates recipes, and generates release logs.
conda/conda_helper_functions.py	Shared helpers for CSV parsing, bundle detection, PyPI queries, etc.
eng/pipelines/conda-update-pipeline.yml	Pipeline definition for the update/PR workflow.
eng/pipelines/templates/stages/conda-sdk-client.yml	Pipeline definition for building and releasing conda packages. Contains the CondaArtifacts list and per-package release toggles.
eng/pipelines/templates/stages/archetype-conda-release.yml	Release stage template — uploads .conda artifacts to Anaconda.
eng/pipelines/templates/steps/build-conda-artifacts.yml	Build step template — installs tooling and invokes sdk_build_conda.

How update_conda_files.py Works

The script is the core of the update pipeline. When invoked (with an optional --release-date MM.DD argument), it performs these steps in order:

1. Bumps AZURESDK_CONDA_VERSION in conda_env.yml to the target release date (or auto-increments by 3 months).

2. Parses the Python package CSV to get the latest GA versions and release dates for all Azure SDK packages.

3. Categorizes packages into:

   • Outdated — already in conda but with a newer GA version available.
   • New — GA packages not yet included in conda.
   • Data plane vs. management plane (management plane packages are all bundled together).

4. Updates conda-sdk-client.yml:

   • Bumps version fields for outdated packages in the CondaArtifacts list.
   • Adds new CondaArtifacts entries and release toggle parameters for new data plane packages.
   • Appends new management plane packages to the azure-mgmt bundle's checkout list (the list of package paths the pipeline checks out and includes in the bundle).

5. Creates meta.yaml recipes for new data plane packages (under conda/conda-recipes/<package>/). For bundled packages (determined by [tool.azure-sdk-conda] in pyproject.toml), a single recipe covers the entire bundle.

6. Updates azure-mgmt/meta.yaml import tests for new management plane packages.

7. Updates/creates release logs in conda/conda-releaselogs/.

8. Prints a report of any packages that couldn't be automatically processed and may need manual attention.

Package Bundles

Some packages are grouped into a single conda package (a "bundle"). For example, azure-storage bundles azure-storage-blob, azure-storage-queue, etc. Bundle membership is determined by the [tool.azure-sdk-conda] section in each package's pyproject.toml.

Bundled packages share:

• A single meta.yaml recipe (with combined dependencies and imports).
• A single release log file.
• A single release toggle in conda-sdk-client.yml.

Service teams are responsible for updating this metadata, and it is enforced in CI.

Build Pipeline (conda-sdk-client.yml)

Build stage:

See conda-builds.md.

Release stage (archetype-conda-release.yml):

• Requires deployment approval (environment: package-publish).
• Uploads all .conda files from the conda/noarch artifact to Anaconda using anaconda-client, under the Microsoft user.
• Requires Anaconda API token with the “Allow all operations on Conda repositories” and "Allow write access to the API" scopes (see Microsoft Open-Source Python docs)

Versioning

All conda packages in a release share the same version string: YYYY.MM.DD (e.g., 2025.12.01). This is set in conda/conda-recipes/conda_env.yml and referenced by every meta.yaml via environ.get('AZURESDK_CONDA_VERSION').

Building Locally

See conda-builds.md for instructions on building conda packages locally using sdk_build_conda.

Adding a New Package to Conda Manually

If the automation doesn't handle a new package correctly, the manual steps are:

1. Add a release toggle parameter in conda-sdk-client.yml.

2. Add a CondaArtifacts entry with name, service, common_root, in_batch, and checkout fields.

3. Create a meta.yaml under conda/conda-recipes/<package>/ (use an existing recipe as a template).

4. Create a release log under conda/conda-releaselogs/<package>.md.

5. For bundles, ensure each member package has [tool.azure-sdk-conda] in its pyproject.toml with the bundle name:

   [tool.azure-sdk-conda]
   in_bundle = true
   bundle_name = "azure-storage"  # The name of the conda bundle this package belongs to