Extract streaming code into dedicated staging modules while keeping stable
compatibility APIs for external client-go consumers.
This commit:
- adds `k8s.io/cri-streaming` for CRI exec/attach/portforward server code
- adds `k8s.io/streaming` as the canonical home for shared transport
primitives (`httpstream`, `spdy`, `wsstream`, runtime helpers)
- switches in-tree transport consumers to `k8s.io/streaming`
- removes in-tree kubelet CRI streaming package
- preserves NO_PROXY/no_proxy CIDR handling in extracted SPDY proxier logic
- adds deprecated `k8s.io/apimachinery/pkg/util/httpstream` compatibility
wrappers (`httpstream`, `spdy`, `wsstream`) backed by `k8s.io/streaming`
- restores exported client-go SPDY/portforward API signatures to
apimachinery `httpstream` types for downstream compatibility
- adds streaming-native client-go adapters/constructors so in-tree callers
can use `k8s.io/streaming` without changing external compatibility APIs
- deduplicates SPDY-over-websocket dial negotiation shared by compat and
streaming tunneling dialers
- logs dropped unknown stream types in `RemoveStreams` adapter fallbacks to
improve compatibility-path debuggability
- adds integration coverage for the streaming-upgrader-to-client-go-compat
adapter path against a real cri-streaming exec endpoint
- clarifies kubectl streaming import aliasing to avoid `httpstream` package
ambiguity
- updates tests, import restrictions, publishing metadata, and vendor/module
metadata for the new staging modules
Signed-off-by: Davanum Srinivas <davanum@gmail.com>
This simplifies specifying the minimum required kubelet (a bit shorter overall,
harder to introduce typos because the string is shorter) and enables usage of
`--sem-ver-filter`.
Jobs that filter by label or text continue to work as before.
As a proof-of-concept, the DRA tests are labeled using the new helper
and are tested with canary jobs which use --sem-ver-filter. The normal
jobs which rely on the label still work.
If it failed to find an IP for every node, it returned a partial list
along with an error. However, (a) this should never happen since every
scheduleable node should have at least one usable IP, and (b) it
didn't matter since all of its callers just treat any error as being
fatal anyway. So just remove the special case.
This returned a randomly-selected IP from across all nodes, preferring
ExternalIP but falling back to an InternalIP. But it was only called
from one place, which was connecting from inside the cluster and thus
wanted an InternalIP anyway.
jig.GetEndpointNodesWithIP was only used by one test, and that test
didn't even want the data in the form jig.GetEndpointNodesWithIP
provided it in. So just inline the code and generate the data
correctly the first time, and remove GetEndpointNodesWithIP (along
with GetEndpointNodes, which was already unused).
This was supposed to have been removed shortly after it was added (the
functionality was moved to kubernetes/perf-tests, where it has since
been improved).
- Moved sample device plugin constants and helper code to the
test/e2e/node/framework, so that both deviceplugin and DRA tests can
use it without creating e2e -> e2e_node dependency.
- Moved SampleDevsAmount constant from the
test/e2e_node/device_plugin_test.go
there are recent changes in k8s e2e test for image pull test to switch
to use fake registry server,
https://github.com/kubernetes/kubernetes/pull/133272https://github.com/kubernetes/kubernetes/pull/134453
Unfortunately, this does not take into account of windows node. so the
corresonding test on windows node strat to break
The try to address this issue, by
(1): update the agnhost windows image to include the fake registery server
functionality as well
(2): update the (common) pull image test to include windows node, will require
(a): deploy the agnhost faker register server as a HPC pod
(b): update the comanndline/options for the windows container specific
Emitting "error dialing" on each retry spams the test log output even when the
tests succeed.
Using Eventually also has other advantages (better failure message, support for
Gingko progress reports when a test is stuck).
This leverages ktesting as wrapper around Ginkgo and testing.T to make all
helper code that is needed to deploy a DRA driver available to Go unit
tests and thus integration tests.
How to proceed with unifying helper code for integration and E2E testing is
open. This is just a minimal first step in that direction. Ideally, such
code should be in separate packages where usage of Ginkgo, e2e/framework
and gomega.Expect/Eventually/Consistently are forbidden.
While at it, the builder gets extended to make cleanup optional.
This will be needed for upgrade/downgrade testing with sub-tests.
(cherry picked from commit 7c7b1e1018)
It turned out that ginkgo.GinkgoT() wasn't as cheap as it should have been (fix
coming in Ginkgo 2.27.5). When instantiated once for each framework.Framework
instance during init by all workers at the same time, the resulting spike in
overall memory usage within the container caused OOM killing of workers in Prow
jobs like ci-kubernetes-e2e-gci-gce with very tight memory limits.
Even with the upcoming fix in Ginkgo it makes sense to set the TB field only
while it really is needed, i.e. while a test runs. This is conceptually similar
to setting and unsetting the test namespace. It may help to flush out incorrect
usage of TB outside of tests.
This makes it possible to call helper packages which expect a TContext from E2E
tests.
The implementation uses GinkgoT as TB and supports registering cleanup
callbacks which expect a context. These callbacks then run with a context that
comes from ginkgo.DeferCleanup, just as if they had called that directly.
(cherry picked from commit 47b613eded)
Refactoring the DRA upgrade/downgrade testing such that it runs as Go test
depended on supporting ktesting in the E2E framework. That change worked during
presubmit testing, but broke some periodic jobs. Therefore the relevant commits
from https://github.com/kubernetes/kubernetes/pull/135664/commits get reverted:
c47ad64820 DRA e2e+integration: test ResourceSlice controller
047682908d ktesting: replace Begin/End with TContext.Step
de47714879 DRA upgrade/downgrade: rewrite as Go unit test
7c7b1e1018 DRA e2e: make driver deployment possible in Go unit tests
65ef31973c DRA upgrade/downgrade: split out individual test steps
47b613eded e2e framework: support creating TContext
The last one is what must have caused the problem, but the other commits depend
on it.
This leverages ktesting as wrapper around Ginkgo and testing.T to make all
helper code that is needed to deploy a DRA driver available to Go unit
tests and thus integration tests.
How to proceed with unifying helper code for integration and E2E testing is
open. This is just a minimal first step in that direction. Ideally, such
code should be in separate packages where usage of Ginkgo, e2e/framework
and gomega.Expect/Eventually/Consistently are forbidden.
While at it, the builder gets extended to make cleanup optional.
This will be needed for upgrade/downgrade testing with sub-tests.
Previously it was necessary to use the Ginkgo wrappers when
using any of the custom arguments like WithSlow(). Now the
hook within Ginkgo for modifying arguments is used such that
e.g. the original ginkgo.It also works.
This makes it possible to call helper packages which expect a TContext from E2E
tests.
The implementation uses GinkgoT as TB and supports registering cleanup
callbacks which expect a context. These callbacks then run with a context that
comes from ginkgo.DeferCleanup, just as if they had called that directly.
Example:
I1208 16:01:05.852628 243 upgradedowngrade_test.go:239] get source code version: bring up v1.34: cluster is running, use KUBECONFIG=/var/run/kubernetes/admin.kubeconfig to access it
I1208 16:01:05.869679 243 reflector.go:446] "Caches populated" type="*v1.ServiceAccount" reflector="k8s.io/client-go/tools/watch/informerwatcher.go:162"
The first line is printed via framework.Logf, which is meant to emulate the
format used by the klog text logger in the second line. The difference is that
klog formats the pid with 7 characters, padding on the left with spaces.
Consistency trumps brevity here, so let's format exactly as in klog.
This fixes some issues found in Kubernetes (data race in ginkgo CLI, gomega
formatting) and helps with diagnosing OOM killing in CI jobs (exit status of
processes).
The modified gomega formatting shows up in some of the output tests for the E2E
framework. They get updated accordingly.
