There's another location which uses the "manifests directory" target
directly, but isn't run when testing a PR: the release workflow.
Signed-off-by: Michael Bridgen <michael@weave.works>
This fixes the case where you create a HelmRelease with `--export` and
the `install: {}` field being there, adding no value to the manifest.
Signed-off-by: Max Jonas Werner <mail@makk.es>
Using the directory cmd/flux/manifests as a prerequisite causes a
problem: if the script that creates the files within fails, the next
invocation of make will see the directory and assume it
succeeded. Since the executable expects certain files to be present,
but they are not explicit prerequisites of the recipe for building the
binary, this results in a successful build but a broken `flux`
executable.
Instead, depend on a file that's explicitly updated when the script
has succeeded, and which itself depends on the inputs.
A couple of the CI workflows run
make cmd/flux/manifests
before doing other things, presumably as a way to avoid running the
whole test suite in a CI pipeline for some purpose other than testing,
so these needed changing as well.
Signed-off-by: Michael Bridgen <michael@weave.works>
Allow specifying the version when building the CLI binary with Make. This is useful for projects that distribute their own Flux CLI binary.
Signed-off-by: Stefan Prodan <stefan.prodan@gmail.com>
When a user provided the `--ca-file` flag to the `bootstrap` command,
the given CA file wasn't taken into account for cloning the repository
locally. It was just passed along to the CR that is created so Flux
can make use of it when cloning the repository in-cluster.
However, users may not want to add a custom CA to their local host's
trust chain and may expect the `--ca-file` flag to be respected also
for cloning the repository locally. This is what this commit
accomplishes.
closes#1775
Signed-off-by: Max Jonas Werner <mail@makk.es>
The create source tests are more interesting than the existing tests as they
create objects then wit for the flux source reconciler to complete. The tests
simulate this with a background goroutine that waits for an object to be
created then uses a test specific function to update it.
The tests set a timeout so that if there is a failure they timeout somewhat
quickly rather than hanging for a longer period of time.
Signed-off-by: Allen Porter <allen@thebends.org>
- remove install instruction as they get out of sync with the docs website
- make the get started guide the first link under docs
Signed-off-by: Stefan Prodan <stefan.prodan@gmail.com>
Speed up unit tests by using a shared envTest. This requires each
test to use its own namespace to avoid clobbering objects for
other tests. Tests previously took around 8 seconds each, and now
the initial test takes 2 seconds with follow up tests taking less
than a second each.
Also update existing tests that use a fixed namespace to use a
generated namespace.
Share gold file template function with yaml files.
Remove the testClusterMode, and instead rely on MainTest to do
the appropriate test setup and rootArgs flag setup. Move the
rootArg flag setup out of NewTestEnvKubeManager to avoid
side effects.
A follow up change can be to push the individual setups
from NewTestEnvKubeManager() into their respective TestMain since
the harness share little code.
Signed-off-by: Allen Porter <allen@thebends.org>
Replace the 4 arguments to cmdTestCase with a function that
can let tests run arbitrary logic if it is more complex than
what is provided by the test harness. Move the existing logic
into functions that the test can use for common assertions on
golden files and golden values.
These changes were pulled out of PR #1696 to make a smaller review.
Signed-off-by: Allen Porter <allen@thebends.org>
Remove use of the fake client, and replace with a real client connected to the
testEnv.
This required fixes to the yaml files as the testEnv has stricter verifcation
of objects. This also meant it was not possible to test a GitRepository with
a missing artifact since that is not a valid state.
The tests are slower than before, taking around 7-10 seconds each because the
testEnv is setup and destroyed for every test. These will be sped up in a
follow up PR.
Signed-off-by: Allen Porter <allen@thebends.org>