Rework image updates to be layer-friendly
Apparently, we now have users who use our images to compile locally as seen in #4 (closed). For this, our images are poorly optimized. The major reason is that we always use new installations essentially forcing users to pull a whole new Docker layer which is rather large.
So instead of performing a full installation every week I could imagine the following:
- During the manual TL year switch (is currently manual, so can stay this way), perform a new installation once and push it with appropriate tags.
- During weekly scheduled updates, pull the latest image and then run
This could have several benefits:
- Each update is a separate layer. So a user who pulled our image three weeks ago for building locally only needs to pull 3 small update layers instead of a whole new installation layer (probably order of MiBs instead of GiBs).
- Currently, the installation is error-prone. Several CI tasks are failing at least bi-weekly just because installing 4000 packages at once is still unstable.
Currently, I have no capacities to work on this, so this is
status:help wanted. Please discuss here before opening MRs.