Fixes flake: 'dial tcp: lookup cdn03.quay.io: no such host'
Okay, doesn't actually fix as in _fix_, just fix as in "sweep it
under the rug". The actual bug is in systemd-resolved, or in the
quay.io/cloudflare.net DNS nameservers, or in the weird specific
setup for cdn03 (it's a CNAME, compared to cdn01/02 which are A).
Maybe a combination of all of the above. I don't care; I just
want the flakes gone. I realize that this makes our testing
environment different from default Fedora, and am okay with
that because I suspect many Fedora users disable systemd-resolved
as SOP.
Signed-off-by: Ed Santiago <santiago@redhat.com>
A small handful of bugs have blocked us from running podman-remote
in gating tests. Those are all fixed; it's time to enable testing.
Signed-off-by: Ed Santiago <santiago@redhat.com>
There's a nasty bug in the CI infrastructure: reboots
hang forever[1]. Sometimes they work, but only 25% of
the time -- the rest of the time, gating tests fail.
[1] https://redhat.service-now.com/surl.do?n=PNT0808530
So, let's just disable the reboot and all cgroups v1 tests.
For now, I've disabled with a quick edit that should allow
us to reenable if/when the reboot bug is fixed.
TODO: reevaluate in three months (2020-08). If the reboot
bug is not fixed by then, just scrap the entire cgroups
refactoring and go back to plain standard-test-basic.
Signed-off-by: Ed Santiago <santiago@redhat.com>
Looks like the code I copied from standard-test-roles
has changed; let's reincorporate it.
In particular, there is now an annoying step that pulls
logs and artifacts from remote-host:/tmp/artifacts .
Well, that doesn't exist in our case because /tmp may
not survive reboots. Workaround: after completion of
each test, push our local (master) artifacts to remote.
And, fix a typo in results.yml
Signed-off-by: Ed Santiago <santiago@redhat.com>
Gating tests are still failing. Actually, the BATS portion is
all passing, it just seems to be the end part (infrastructure)
that fails. Let's try creating a results.yml file.
Signed-off-by: Ed Santiago <santiago@redhat.com>
- instead of hacky mkdir /run/user/<fedora id>,
use loginctl enable-linger and hope that it
has the same effect.
reason: the mkdir doesn't seem to be 100% effective.
There are cases in which the first rootless podman
invocation fails with:
Error: could not get runtime: cannot mkdir /run/user/1000/libpod: mkdir /run/user/1000/libpod: no such file or directory
Signed-off-by: Ed Santiago <santiago@redhat.com>
forward: rhbz1740664 (permissions on /dev/net/tun) appears to
be fixed; let's remove the 'modprobe tun' workaround.
back: rootless tests are failing in CI, in a way that I
can't reproduce via 1minutetip. One of the errors is:
cannot mkdir /run/user/1000/libpod: mkdir /run/user/1000/libpod: no such file or directory
...so let's try, after the mkdir, chowning it to fedora.
Signed-off-by: Ed Santiago <santiago@redhat.com>
Systemd bug 1740664 causes /dev/net/tun to be mode 600,
causing slirp4netns to fail on rootless podman.
Suggested workaround is to 'modprobe tun'. Let's
give that a try in the interests of making the
gating tests pass.
Signed-off-by: Ed Santiago <santiago@redhat.com>
Define a new podman-tests subpackage, containing a set of
tests intended to be run as gates. As of this initial checkin,
this consists of a set of BATS tests in /usr/share/podman .
Include a configuration for running under Fedora CI
Signed-off-by: Ed Santiago <santiago@redhat.com>
This is a placeholder only. The idea is that, once the ginkgo
tests are approved and merged and packaged, you add that
package to test_podman.yml:required_packages, then rewrite
the .sh script so it invokes those tests and writes PASS/FAIL
to test.log as appropriate.
Until then, this can serve as a rudimentary gate.
Signed-off-by: Ed Santiago <santiago@redhat.com>
