Essential Terms and Concepts¶

archive¶

A packed environment layer for distribution and deployment. Contains either a fully built base runtime environment layer, or else a built layered environment that depends on the other layers specified in its metadata.

environment¶

build environment¶

deployed environment¶

layered environment¶

A base runtime environment (built from a base runtime layer definition), or a layered virtual environment (built from an application or framework layer definition). May be a build environment, or a deployed environment. Deployed environments may be created either directly (via local export), or indirectly (via archive creation and unpacking). Exported environments and archive deployments contain slightly different metadata (since there are no archive details in the exported environment metadata).

export¶

local export¶

Locally publishing an environment on the same machine, skipping the archive-and-unpack step, and automatically running the post-installation step. Primarily intended to speed up development iteration cycles when testing stack builds and application layer launch modules, but may also be used to export environments that normally use symlinks to target filesystems which don’t support them (such as USB keys). (Note that transferring USB keys between systems is still likely to run into problems related to absolute paths no longer being correct, as Windows drive letters and POSIX mount points are highly likely to differ across machines).

layer¶

application layer¶

base runtime layer¶

framework layer¶

A definition of a set of Python requirements which will be pinned for building and publication as a single consolidated archive. Layer definitions are categorised as follows:

base runtime layers: these layers specify a base Python runtime which is used as a foundation for one or more environment stacks. Any requirements specified as part of a base runtime layer are installed directly into the base runtime (there is no virtual environment defined).
framework layers: these layers primarily contain large dependencies (such as PyTorch) which should not be published multiple times, even when they are used by multiple applications. Applications are constrained to use the versions of any packages installed in the framework layers they depend on. Each framework layer depends on a specific runtime layer.
application layers: these layers specify the actual deployed Python applications which embedding applications will invoke. Applications depend on one or more framework layers

stack¶

environment stack¶

An application layer with its supporting framework and base runtime layers.

stack specification¶

A venvstacks.toml file that defines one or more environment stacks.