We’ve been using an action runner successfully to build and push a docker image to AWS. Today the runner has started failing on the docker build step. However, when building the image locally we have no issues. We haven’t changed the action at all, and the build this first starting failing on had no meaningful changes to the docker image.
On investigation it seems that on the github hosted runners the pip step is no longer grabbing wheels (pip will grab wheels if they are available, and fail back to source packages which are then built) and failing to build source packages. We can reproduce this locally if we force the local build to use source packages, which isn’t ideal as the build time becomes excessive.
Our guess is there’s some local cache / DNS that’s broken on Github side that means the wheels can no longer be found remotely? Are there any other people experiencing this and have suggestions on how we can fix our end?