Re "Bypassing pipeline jobs by canceling the pipeline and manually running later jobs"

We merged the https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/27503 to solve the https://gitlab.com/gitlab-org/gitlab-ce/issues/53064.

However, this solution brings usability problems. As, now if I cancel the job mid-stage (legitimately) I cannot retry it.

We always followed the principle that every user action is retriable, including cancel. As cancel donates as failure to the system.

It seems that solution should be rather being aware when canceled job cannot be retried, because most of them can. One would be that the cancel behavior of for pipeline is wrong, as we should cancel only running jobs, not every job. This means that jobs that did not yet run should likely stay created. As created jobs cannot be manually run.