Currently some jobs are pushing in parallel their tag which makes fail some other
concurrent job. This allows to retry the failing push instruction without failing the
overall job.