Originally reported from here:
To provide a user with an actionable response, we should be clear as to the reason a job failed. In this case, it was a pipelines error of running out of disk space (hopefully those 150 minutes weren't counted!).
From the pipelines team, we need:
1) reason a job failed (our fault vs their fault)
If it is pipelines fault:
2a) let's not charge the user (!)
2b) attempt to retry since it's there is only one action a user would want: re-run the build until the reason it stopped is in user land