On 06.03.19 08:56, Steve Langasek wrote: > On Sat, Mar 02, 2019 at 11:52:05AM +0100, Christian Kastner wrote: > >> Side note: saving away the artifacts in $AUTOPKGTEST_ARTIFACTS >> apparently didn't work on Ubuntu's side; the artifacts tarball only >> contains the usual log files. > > I think this is because your autopkgtest is 'set -e', so the cp command is > never reached when the tests fail, making it decidedly less useful.
Indeed; thanks for pointer. >> Short version: The test defines the needs-root restriction; are you >> absolutely positive that this is satisfied in your test environment? >> Because the results seem to point to the contrary. > > I suspect the issue here is that the containers are non-privileged (i.e. > using user namespaces), so even though the tests are run as "root" in the > container, they don't have /real/ root on the kernel and therefore the > maxbytes limit applies instead of the root_maxbytes limit. > > The definition of the needs-root restriction is certainly ambiguous wrt it > needs to be uid 0 on the root user namespace. And in any case, this isn't > something that Ubuntu is going to be able to change anytime soon - and it's > not uncommon for containers to be unprivileged. So I would suggest that > this test might be more suitable for the machine-isolation test set. Yes, moving the test make sense. In hindsight, it should have always been machine-isolation anyway, as the possible (temporary) change of root_maxbytes functionally changes the system, and therefore could have side effects on unrelated processes. Regards, Christian
