Docker image size is blowing up

[francesco-ballarin]

When comparing the last few releases in
https://console.cloud.google.com/gcr/images/kaggle-images/GLOBAL/python
it is quite noticeable that recently there has been a sudden increase in the docker image size, e.g. from virtual size 6.2 GB of v125 to virtual size of 8.3 GB of v129, i.e. a 34% increase.

When uncompressed, docker reports that v125 results in a 19 GB container, while v129 results in a 23 GB container, i.e. a 21% increase.

I noticed this while running (the heaviest of) my workflows on GitHub Actions, randomly getting errors like

System.IO.IOException: No space left on device : '/home/runner/runners/2.302.1/_diag/Worker_20230304-204952-utc.log'

While surely the amount of disk space that GitHub offers is not your responsibility, you may still want to check if there is any way to keep the size of the docker image under control, and if possible keep it from growing at this rate any further.

[djherbis]

@francesco-ballarin Thanks for reporting, its mostly likely a consequence of the massive number of packages we install.

I've filed an internal bug for me to review the size, evaluate which packages can be removed etc.