I've changed the code in autobuilder2:

https://git.yoctoproject.org/yocto-autobuilder2/commit/?id=d809e29a23d5614d8873c3f5f1fb7293da96cdd0

which increases the space limit from 100G to 200G. I also added
monitoring to /tmp since that caused a lot of build issues in a recent
failure.

This will help but I think we need to improve the janitor to clean up
failed builds. I'm thinking of something like a queuing system so any
build failure gets put into the queue and anything older gets deleted
as soon as the system is under disk pressure?

This would solve the issue that:

* builds the rerun on the same worker destroy failed data immediately
* older builds can build up and clog systems

It would also make it easy to put a policy in place:

* to immediately delete reproducible fails (for exmaple)
* to delete builds when space drops to a threshold

Help with this would be very welcome.

Ideally we'd have a way via a UI to mark failed builds as not wanted
but I'm not sure how to do right now.