Any time venv_rebuild=true is used the wheel build directory on the repo server is emptied out. The constraints and requirements text files for the currently running role will be deleted from the repo server and this forces the wheels to be rebuilt for that, and only that role.
If the playbook calls several roles, the next role that runs will also delete the entire wheel cache contents and also the constraints/requirements files for that next role.
This means that every role run by a playbook will individually succeed when venv_rebuild=true, but at any one time the repo server wheel cache only contains the wheels needed for that role.
Subsequent playbooks run without venv_rebuild=true will find the wheel cache in an inconsistent state with the requirements/constraints files on the repo server, and the required wheels will not be built nor be available for download to the installation target. Pip will fall back to trying to build the wheels locally on the target host, and as we have done considerable cleanup of target vs. devel packages it is almost certain that the build will fail due to missing headers of some form.
The behaviour of python_venv_build when venv_build=true seems incorrect and can lead to a situation where it is not possible to run further playbooks correctly.
Looks like a legit fix: https:/