We need to improve our automated image QA, for example checking that /run is mounted and read/write even if / is read-only. Assigning to Jussi for further exploratory research and scoping.
I've looked at the infrastructure and it seems everything needed is there already: this bug is just about defining the things to test for and writing the tests -- this seems fairly straight forward so far. I expect there to be two types of tests: 1. file based rootfs tests: See that a specific files exists on rootfs and have the content and permissions we expect 2. runtime tests: same but for files that are only available at runtime Type 1 could be tested with just the rootfs temporarily mounted somewhere, while type 2 requires starting qemu: need to check if having two separate types is useful or if just doing everything over qemu works fine. I've not really scoped out the specific tests yet, but some quick ideas: - is "/" read-only when expected? (maybe check a few directories under root?) - is "/run" mounted? - is "/run" read-write? - is "/var/volatile" read-write ? - permissions sanity checks for key directories/files? Related: do we make sure that packages don't install things to /var/run or /run ? I know I saw people making this mistake in fedora and debian packages a few years ago...
quick question, is this a bug that will require QA to run these tests regularly? I assume yes so I'd add them to the loop. Pending confirmation.
(In reply to comment #2) > quick question, is this a bug that will require QA to run these tests > regularly? I assume yes so I'd add them to the loop. Pending confirmation. These would be part of the sanity tests that run on the autobuilder. QA work should not be required.
Reassigning Jussi's bugs.
We need to define what these tests should be, then it can go back to a NEWCOMER.