We know worldwide apt mirrors can contribute to measurement error when interpreting package downloads as a proxy for community growth, but another considerable source of error may be the Official Docker Hub Library images for ROS. These images are built once for each supported architecture by CI, then pushed to the registry where they can be pulled millions of times by users or other production/CI environments; for which each image can pre-bundle hundreds of ROS debian packages.
Docker Hub provides an API for such statistics which I linked to in a table from a previous post:
However, Docker Hub does not aggregate such statistics into a queryable time series format, suitable for extrapolating trends and usage patterns. One can endeavor to record and track such statistics via an external cron job or via scheduled sampling:
Better yet, I’ve recently discovered an opensource project that has been collecting and archiving these statistics for popular repos for the past year or so, rending pull counts as convenient line charts:
Some more examples:
- https://dockerstats.com/hubs/docker/library/ros
- https://dockerstats.com/hubs/docker/amd64/ros
- https://dockerstats.com/hubs/docker/arm32v7/ros
- https://dockerstats.com/hubs/docker/arm64v8/ros
- https://dockerstats.com/hubs/docker/osrf/ros
For reference as a scale/growth comparison, here are some more popular/niche repos:
- https://dockerstats.com/hubs/docker/library/ubuntu
- maxes out 32-bit integer, oops: Docker Hub Max 32-bit Integer Reached · Issue #2003 · docker/hub-feedback · GitHub
- https://dockerstats.com/hubs/docker/library/gazebo
- only pre-bundles gazebo sim packages
Perhaps we could link/scrape this source as well and include these statistics in future metrics reports.