re-enable kaniko cache #96
Conversation
h5py is now only for Python 3 as python3-h5py. python2-pylint is now pylint.
|
Rerunning the pipeline a few times produced a corrupted centos:7 image (https://gitlab.icp.uni-stuttgart.de/espressomd/docker/-/jobs/120796), so we'll keep this issue WIP until an appropriate kaniko update has been released or I manage to track down myself why caching is broken. |
|
probably related GoogleContainerTools/kaniko#575 |
|
bwcloud seems to have trouble with the exotic arches |
|
This happens quite frequently with QEMU images. Just retry the job until it works, the worst case I had was 3 retries. This is already documented in #97. |
|
I have set up a nightly job for https://gitlab.icp.uni-stuttgart.de/espressomd/docker/commit/60d7eb33b4c74f90ea74bb5b0f966a8140b96a65/pipelines and we shall see whether the cache now works reliably. If it does for a few weeks, we can merge this pull request. |
|
the related PR GoogleContainerTools/kaniko#715 is still open... Why should the situation have changed? |
|
They have closed a bunch of cache-related issues in recent weeks and I don't think that that pull request is really related. I already tried a few builds yesterday and didn't find cache errors. We shall hope and see. |
|
probably GoogleContainerTools/kaniko#639 but it's quite old? Let's see |
|
Caching seems to be reliable now. Two weeks of successful rebuilds: https://gitlab.icp.uni-stuttgart.de/espressomd/docker/commit/60d7eb33b4c74f90ea74bb5b0f966a8140b96a65/pipelines. Please go ahead and merge this pull request. It speeds up the average pipeline by a factor of three. |
reverts #93
It‘s possible that the caching issue was resolved by splitting the build and deploy stages. To find out whether that is the case, I‘ve created this pull request.