On Fri, Nov 27, 2015 at 04:11:52PM +0000, Simon McVittie wrote: > On 26/11/15 16:11, Antonio Terceiro wrote: > > On Thu, Nov 26, 2015 at 12:48:13PM +0000, Simon McVittie wrote: > >> Is there a way I can reproduce the ci.debian.net environment on a local > >> virtual machine or cluster of virtual machines, preferably jessie amd64? > >> https://ci.debian.net/doc/file.MAINTAINERS.html currently still > >> documents the schroot-based setup. > > > > basically > > > > $ sudo adt-build-lxc debian sid > > $ adt-run dbus --- lxc --sudo adt-sid-$ARCH > > I tried following those instructions, but as well as needing to upgrade > cgmanager to the version in sid (?), adt-build-lxc seems to require > non-default lxc configuration. The default in /etc/lxc/default.conf is > "lxc.network.type = empty" which means it isn't going to get far with > apt, and I have no idea which of the other options is preferred. > > With lxc.network.type = macvlan I was able to get the D-Bus tests to > run, but I get a lot more failures than ci.debian.net did, so I must > still be doing something wrong. Note that D-Bus is multi-user IPC, so > its tests use TCP, abstract Unix sockets, filesystem-backed Unix > sockets, multiple uids, and the test user's home directory; this is > perhaps exercising more of the LXC stack than the average package. > > With adt-virt-qemu, everything passes until we get to the tests that run > as root, which hang (that's this bug); that's more like what I expected. > > It would be great if you could go all the way through the lxc-based > setup in a VM or on a dedicated machine, from installation (or > vmdebootstrap) with a specified suite and package-set to successfully > running an autopkgtest, and document how. Alternatively, if > ci.debian.net workers have some sort of scripting (Puppet?) that I can > study, please point me towards that and I might be able to > reverse-engineer instructions. > > Based on work I've done for $DAYJOB with CI gating acceptance, I think > it's really important for random developers to be able to reproduce the > test environment. schroot was quite good for that, and the combination > of qemu and vmdebootstrap is really straightforward (assuming you have > lots of RAM, which I do), but lxc is not something I've worked with > before, so I'm a bit lost without documentation.
Sure, absolutely. I have updated the instructions and reproduced running test runs starting from an empty VM following them: https://ci.debian.net/doc/file.MAINTAINERS.html It would be great if you could give it a try. > The patch that I applied upstream for this particular bug is working on > Ubuntu's armhf workers, which appear to be LXC-based... so if it works > in my adt-virt-qemu, I'm going to upload it anyway, and ask you to > reverse the blacklisting even though I can't actually test it in LXC. > Hopefully that will work. if the problem also happens under qemu, then the fix you have should do the trick under lxc as well. -- Antonio Terceiro <[email protected]>
signature.asc
Description: PGP signature
