Remove build tags which are already implied by the name of the file.
Ensures build tags are used consistently
Signed-off-by: Derek McGowan <derek@mcg.dev>
Signed-off-by: Jayme Howard <g.prime@gmail.com>
Run `go mod tidy`
Signed-off-by: Jayme Howard <g.prime@gmail.com>
Follow correct procedure by running `make vendor`
Signed-off-by: Jayme Howard <g.prime@gmail.com>
In containerd 1.5.x, we introduced support for go modules by adding a
go.mod file in the root directory. This go.mod lists all the things
needed across the whole code base (with the exception of
integration/client which has its own go.mod). So when projects that
need to make calls to containerd API will pull in some code from
containerd/containerd, the `go mod` commands will add all the things
listed in the root go.mod to the projects go.mod file. This causes
some problems as the list of things needed to make a simple API call
is enormous. in effect, making a API call will pull everything that a
typical server needs as well as the root go.mod is all encompassing.
In general if we had smaller things folks could use, that will make it
easier by reducing the number of things that will end up in a consumers
go.mod file.
Now coming to a specific problem, the root containerd go.mod has various
k8s.io/* modules listed. Also kubernetes depends on containerd indirectly
via both moby/moby (working with docker maintainers seperately) and via
google/cadvisor. So when the kubernetes maintainers try to use latest
1.5.x containerd, they will see the kubernetes go.mod ending up depending
on the older version of kubernetes!
So if we can expose just the minimum things needed to make a client API
call then projects like cadvisor can adopt that instead of pulling in
the entire go.mod from containerd. Looking at the existing code in
cadvisor the minimum things needed would be the api/ directory from
containerd. Please see proof of concept here:
github.com/google/cadvisor/pull/2908
To enable that, in this PR, we add a go.mod file in api/ directory. we
split the Protobuild.yaml into two, one for just the things in api/
directory and the rest in the root directory. We adjust various targets
to build things correctly using `protobuild` and also ensure that we
end up with the same generated code as before as well. To ensure we
better take care of the various go.mod/go.sum files, we update the
existing `make vendor` and also add a new `make verify-vendor` that one
can run locally as well in the CI.
Ideally, we would have a `containerd/client` either as a standalone repo
or within `containerd/containerd` as a separate go module. but we will
start here to experiment with a standalone api go module first.
Also there are various follow ups we can do, for example @thaJeztah has
identified two tasks we could do after this PR lands:
github.com/containerd/containerd/pull/5716#discussion_r668821396
Signed-off-by: Davanum Srinivas <davanum@gmail.com>
systemd uses SIGRTMIN+n signals, but containerd didn't support the signals
since Go's sys/unix doesn't support them.
This change introduces SIGRTMIN+n handling by utilizing moby/sys/signal.
Fixes#5402.
https://www.freedesktop.org/software/systemd/man/systemd.html#Signals
Signed-off-by: Kazuyoshi Kato <katokazu@amazon.com>
For Windows, the container image's OS version must closely match the host's OS version.
For this reason, we need to add the --os-version annotation in image manifest lists,
so the Windows nodes can pull the appropriate image from the list.
Previously, the docker manifest CLI did not have the capability to set the --os-version,
it, but it has been introduced in docker 20.10.0.
We're also adding busybox.exe in the image, so we can run Linux commands inside the
container, so the tests will be simpler.
When building Windows images, a docker buildx builder needs to be created and used. When
building Windows images with docker buildx, the flag --output=type=registry is required,
otherwise it cannot be referenced on a Linux node.
Signed-off-by: Claudiu Belu <cbelu@cloudbasesolutions.com>
If a container failed to start due to a bad command, the container could not be
recreated with a proper command in its stead. Adds a test that verifies this scenario.
Signed-off-by: Claudiu Belu <cbelu@cloudbasesolutions.com>
A previous commit introduced EnsureImageExists, which ensures that a particular image
already exists. It also deduplicates the image pulling code. Some tests missed this
update.
Signed-off-by: Claudiu Belu <cbelu@cloudbasesolutions.com>
Most of the tests creating and using Pod Sandboxes in the same way. We can create
a common function that will do that for us, so we can have less duplicated code,
and make it easier to add new tests in the future.
Signed-off-by: Claudiu Belu <cbelu@cloudbasesolutions.com>
There was a known issue regarding how the symlink files mounted as
volumes were being handled on Windows. This commit adds tests that
will check against those issue to ensure there won't be any
regressions.
Signed-off-by: Claudiu Belu <cbelu@cloudbasesolutions.com>
Currently, the cri-integration tests do not work on Windows due to various reasons.
One of the reasons is because all the tests are using Linux-specific images.
Previous commits refactored the image pulling / usage in the cri-integration tests,
making it easier to update, and easier to configure a custom registry to pull images
with Windows support.
For Windows runs, custom registries can be created, which will also contain Windows
images, and the cri-integration tests can be configured to use those registries by
specifying the "--repo-list" argument, a YAML file which will contain an alternative
mapping of the default registries. This is similar to how E2E tests are handled for
Windows runs in Kubernetes.
Some of the tests are Skipped, as they do not pass yet on Windows.
Windows does not collect inodes used stats, thus, the tests that were expecting non-zero
inodes stats were failing.
Signed-off-by: Claudiu Belu <cbelu@cloudbasesolutions.com>
This forces vendoring to only take dependencies of this repository to
be taken into account, effectively cutting the circular dependency (for
the vendored code), and to prevent depending on transitive dependencies
coming from older versions of containerd.
go mod does not allow using the main module as a local "replace" rule using
a path; see golang/go#45492 and golang/go#34417, so instead, an empty module
is used.
One change observed is that older versions containerd depended on an older
version of imgcrypt that had an "indirect" dependency on more current versions
of gopkg.in/yaml.v2 and prometheus/procfs.
For those, a temporary "indirect" dependency was added, until prometheus/client_golang
and kubernetes are updated.
from go mod graph (before):
github.com/containerd/imgcrypt@v1.0.4-0.20210301171431-0ae5c75f59ba gopkg.in/yaml.v2@v2.4.0
github.com/containerd/imgcrypt@v1.0.4-0.20210301171431-0ae5c75f59ba github.com/prometheus/procfs@v0.6.0
For some reason, some older versions of containerd are still taken into account,
causing satori/go.uuid to be added as "indirect" dependency, likely because some
modules have this dependency in their go.sum. This should likely disappear once
those plugins are updated to contain a current version of containerd.
git grep 'github.com/satori/go.uuid'
vendor/github.com/Microsoft/hcsshim/go.sum:github.com/satori/go.uuid v1.2.0/go.mod h1:dA0hQrYB0VpLJoorglMZABFdXlWrHn1NEOzdhQKdks0=
vendor/github.com/containerd/aufs/go.sum:github.com/satori/go.uuid v1.2.0/go.mod h1:dA0hQrYB0VpLJoorglMZABFdXlWrHn1NEOzdhQKdks0=
vendor/github.com/containerd/imgcrypt/go.sum:github.com/satori/go.uuid v1.2.0/go.mod h1:dA0hQrYB0VpLJoorglMZABFdXlWrHn1NEOzdhQKdks0=
vendor/github.com/containerd/nri/go.sum:github.com/satori/go.uuid v1.2.0/go.mod h1:dA0hQrYB0VpLJoorglMZABFdXlWrHn1NEOzdhQKdks0=
vendor/github.com/containerd/zfs/go.sum:github.com/satori/go.uuid v1.2.0/go.mod h1:dA0hQrYB0VpLJoorglMZABFdXlWrHn1NEOzdhQKdks0=
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
We can check that the reported containers stats are not lower than it should
by using the resource-consumer image. This image contains stress / testlimit.exe
which can consume a specified amount of memory.
Signed-off-by: Claudiu Belu <cbelu@cloudbasesolutions.com>
Most of the tests are pulling and deleting the same test images, which
can be quite inefficient, especially on Windows nodes, where the images
are larger than the Linux ones (a nanoserver Container image is ~250MB
in size). We can instead pull them only once, and reuse them. This will
reduce the test run time on Windows considerably.
Additionally, some of the test images are currently hosted on dockerhub
(busybox image), which has introduced image ratelimiting in November 2020,
which means that test runners could potentially hit that limit faster with
the current implementation. This will reduce that risk.
Some tests are specifically deleting images, so we always have to ensure
that they are pulled.
Signed-off-by: Claudiu Belu <cbelu@cloudbasesolutions.com>
Currently, the cri-integration tests do not work on Windows due to various reasons.
One of the reasons is because all the tests are using Linux-specific images. This
commit refactors the image pulling / usage in the cri-integration tests, making it
easier to update, and easier to configure the a custom registry to pull those images
from.
For Windows runs, custom registries can be created, which will also contain Windows
images, and the cri-integration tests can be configured to use those registries by
specifying the "--image-list" argument, a TOML file which will contain an alternative
mapping of the default images.
Signed-off-by: Claudiu Belu <cbelu@cloudbasesolutions.com>
- Add a multi-arch image with linux/amd64 and linux/arm64 (limit to just
what we are running in CI)
- Bump versions to 2.0 as 1.0 is the single-arch image
- Use `docker buildx` instead of just docker, so we don't need to build
manifests by hand
- busybox now does not have `nogroup`, since the test needs it, switch
over from busybox to ubuntu for just the volume-ownership image
Signed-off-by: Davanum Srinivas <davanum@gmail.com>
full diff: https://github.com/google/uuid/compare/v1.1.2...v1.2.0
I did not switch to us the `uuid.NewString()` function; we are not currently using
`uuid.New()`, probably because it can `panic` in case an invalid UUID is produced;
in stead we are using `uuid.NewRandom()`, which is equivalent, but returns an
error instead.
Changes:
- v1.2.0: Add NewString()
This release introduces the NewString() function which is the equivalent of
uuid.New().String().
- v1.1.5: Syntactic cleanup
There are no code changes. A missing period was add to a godoc comment and the
linter was told to not complain that the results of hash.Write() are ignored
(the function cannot fail)
- v1.1.4: Further error optimizations
Do not allocate memory for errors (it is only one word)
- v1.1.3: Optimize error reporting
Optimize length of time it takes to discover an input is bad by no longer using
fmt.Errorf, which is quite slow. It now uses a custom error type that formats
the string when the Error method is called rather than when generating the error.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
