VM Base images are used as a starting point for runtime VM images.
The in-use VM base images should never be pruned, which is an
operation that potentially occurs periodically from automation
running on the master branch of the libpod repo.
However the only place which updates timestamps (blocking pruning)
of base images, occurs during runtime VM image building. Therefor,
if images are not regularly rebuilt, it's possible their base images
go stale and are pruned.
Changes:
* Add freshly-produced base images (old ones got pruned)
* Wrap the timestamp update script to include base image names
in the update list.
Notes:
* Regularly updating base image timestamps only needs to happen
on the libpod repo's meta task, since all base images live there.
* Using a wrapper is needed to maintain compatibility with multiple
versions of the imgts container image used by other repos / branchs.
Signed-off-by: Chris Evich <cevich@redhat.com>
Cirrus-CI only sets `$CIRRUS_BASE_SHA` when testing PRs since the
destination and it's state is easy to discover. However, when
post-merge and/or tag-push testing, the previous state is not
easily discoverable (changes have already merged). The
`upload_release_archive` script incorrectly assumed this
variable was always set, causing a constant stream of post-merge
testing failures.
Tweak the `is_release()` function to properly handle an empty
`$CIRRUS_BASE_SHA` whether or not `$CIRRUS_TAG` is also set. Also
update the unit-tests to check for this. Also account for a
corner case where hack/get_ci_vm.sh is running on a VM w/o git.
Signed-off-by: Chris Evich <cevich@redhat.com>
Prior to this commit, every push to master had it's builds packaged and
uploaded to google storage. This is a waste, since potential users
are only ever concerned about tagged releases.
Unfortunately because the release process involves humans with
potentially multiple human and automation steps happening in parallel,
it's easy for automation to not detect a tagged release, or trigger on
development|pre-release tags.
Fix this in `upload_release_archive.sh` using a new unit-tested
function `is_release()`. This acts as the definitive authority
on whether or not a specific commit rage or `$CIRRUS_TAG` value
constitutes something worthy of upload.
Signed-off-by: Chris Evich <cevich@redhat.com>
When executing 'make remotesystem' testing, a varlink process is started
up but it's stdio is dumped due to the production of excessive data.
However, this also means if the process has a problem, any errors will
not be accessible.
Instead, grab only the last 100 lines and direct them into a file. Also
update automation's log collection to retrieve this file when the
`$REMOTE_CLIENT` env. var. is `true`.
Signed-off-by: Chris Evich <cevich@redhat.com>
This is needed because older versions of podman (1.5.1) do not
automatically install the new conmon package.
Also, include removal of `/usr/libexec/podman/conmon` when preparing to
install and test podman built from source.
Signed-off-by: Chris Evich <cevich@redhat.com>
Adds an output to all testing tasks which lists the names/versions of
critical/essential packages present on the VM.
Signed-off-by: Chris Evich <cevich@redhat.com>
This is fairly standard thing to have on a user's workstation, supported
by podman. When installed in a VM image, then it's useful for debugging
with `hack/get_ci_vm.sh` at the cost of a minor increase in disk-space.
Signed-off-by: Chris Evich <cevich@redhat.com>
Instead of unconditionally pulling the x86 binary, clone the repository
and build the binary to make it independent of the architecture.
Fixes: #2699
Signed-off-by: Valentin Rothberg <rothberg@redhat.com>
Add the latest Ubuntu version into the testing matrix
and image-build workflow. This is also needed to support
other containers projects which share use of VM images
from this one.
Update package lists to include needs for contianers/storage
use of images.
Signed-off-by: Chris Evich <cevich@redhat.com>
* Fix one disused and two missing required env. vars.
* Slightly optomize processing of commit-author names
* Fix problem of printing duplicate author names when there are multiple
commits.
* Fix bot's IRC connection timeout too short.
* Add a single retry of IRC connection after 5-second delay.
Signed-off-by: Chris Evich <cevich@redhat.com>
Fixed a typo.
Also script was grabbing quotes and other non-email-address junk
while looping. Filter before and after to make sure we get 'em all.
Signed-off-by: Chris Evich <cevich@redhat.com>
When constructing VM cache-images, the latest/greatest podman package is
installed to ensure all necessary dependencies are met. Prior to
testing source-built binaries, most of of the packaged files are removed.
However, if the `io.podman` service or socket is enabled/running, it
could cause the packaged podman and varlink binaries to be both resident
and cached. Since this condition would cause very difficult to diagnose
behaviors, add preventative measures to ensure these services are absent
prior to removing packaged podman files.
Signed-off-by: Chris Evich <cevich@redhat.com>
This is mostly used with Systemd, which really wants to manage
CGroups itself when managing containers via unit file.
Signed-off-by: Matthew Heon <matthew.heon@pm.me>
Also add fixes to help prevent 'fatal: Invalid revision range' error.
Should obtaining all authors from the range still fail, only grab the
HEAD commit author as a fallback.
Signed-off-by: Chris Evich <cevich@redhat.com>
Rather than spamming the podman channel with impersonal success
messages referring to PR numbers, mention the author by nick name
and include the PR title and link.
Also avoid needless logging of all bot-script interactions with
IRC when there is no error detected.
Signed-off-by: Chris Evich <cevich@redhat.com>
The initial implementation was far more complicated than necessary.
Strip out the complexities in favor of a simpler and more direct
approach.
Signed-off-by: Chris Evich <cevich@redhat.com>
Specifically pertaining to executing tests in google cloud, there are
default, pre-allocated class-a subnetworks for each region (data-center).
Each includes a gateway using a `.1` LSB and all are routable from other
regions in google cloud via these gateways.
Because the default CNI configuration also utilizes class-a subnetworks,
this creates the possibility for IPv4 address-space clashes. Since the
default regional cloud subnets are pre-defined/known, preventing clashes
can be accomplished by seeding these subnets in a dummy CNI configuration.
The default behavior of podman is to grab the highest priority CNI
configuration. Name the dummy config. appropriate so it always loads
last. Also name the bridge itself with an obvious name `do-not-use`,
such that any related testing errors should be easier to debug.
Also:
* Minor cleanup of `install_test_configs()`
* Move install_test_configs in `setup_environment.sh` to after possible
run of `remove_packaged_podman_files()` because that also strips out
`/etc/cni/net.d/87-podman-bridge.conflist`.
Signed-off-by: Chris Evich <cevich@redhat.com>
