Preserving a failed build layer for debugging

[peterwaller-arm]

Some builds of mine have layers doing compilation of large which can take a very long time. On failure, I would like to be able to drop into the container in order to debug it, rather than starting from scratch each time, or manually doing it with docker run.

Is this possible currently? I'm expecting to see a --keep-failed option or similar, but I did not find it.

Note that I'm using RUN --mount, and I need those cache mounts in place as well, so ideally the whole container should be preserved (e.g. tmpfs mounts). The image itself would be a start, but not sufficient to solve the whole problem.

[AkihiroSuda]

duplicate: #1053
(My proposal: #1053 (comment) Workaround: #1053 (comment))

[AkihiroSuda]

Let me close this to prevent the discussion from diverging.