gitlab-ci cache could be more used
Summary by Bruno
Package caching:
-
Linux debian pkg cache (using docker, restricted to pipeline level) (NOT A REAL CACHE) -
Linux flatpak pkg cache (using flatpak-builder git downloads, at project level, but it's broken on image or runner) -
Windows/Crossroad pkg cache (using prefix, restricted to pipeline level) (NOT A REAL CACHE) -
Windows/MSYS2-x64 pkg cache (using pacman, at runner level) -
Windows/MSYS2-ARM pkg cache (using pacman, at runner level)
Build caching (all at project level):
-
e96b751e: Linux debian ccache (enabled, but the good number of runners requires a distributed cache) -
e96494d5: Linux flatpak ccache (enabled, but it's broken on image or runner) -
f40ecca6: Windows/Crossroad ccache (enabled, but the good number of runners requires a distributed cache) -
e96b751e: Windows/MSYS2-x64 ccache -
e96b751e: Windows/MSYS2-ARM ccache
The cache does not seem to work find in gitlab-ci. Cache is mostly used for package management in our script (preventing re-download of packages). Unfortunately it means that (1) builds are longer than they could since packages are always re-downloaded instead of re-used (2) sometimes they could fail if a network error occurs (download fails).
I did some tests but I am still unsure if and what we do wrong, and I'm tired of testing. Who can help? @Salamandar, you seemed to know a bit about Gitlab's CI system? Maybe @nielsdg too?
Anyone?
Edited by Bruno