Date: 2026-08-30
Proposed for evaluation. Docker is not required.
The pipeline needs Python geospatial packages, native GDAL tools, large rasters and point clouds, LINZ credentials and long builds. Most work is on macOS; the guides also cover Ubuntu and Windows.
Containers could pin dependency versions and help with batch jobs. They also add cost: large bind mounts, file ownership, disk performance, Apple Silicon support, debugging and credential handling.
The local .venv workflow stays the baseline. Add no Docker, Compose or
container runtime until an evaluation shows a real benefit.
Try a container proof of concept if:
The proof of concept must build one area, pass the existing checks, read
host data without copying it into the image, keep LINZ_API_KEY out of the
image and repo, and measure setup time, runtime, memory and disk against the
native workflow on macOS and Ubuntu.
Setup stays simple and local. Reproducibility rests on the documented Python environment. Any adoption must update the setup guides, record the measured trade-offs, and keep a path for developers who do not use containers.