Current test assumes that test pod is deleted when the test
namespace is deleted. However, namespace deletion is an asynchronous
operation. The pod may still be running and allocating hugepages
resources when next test case creates another pod that requests
the same hugepages resources. This can cause kubelet to fail the test
pod with this kind of error:
OutOfhugepages-2Mi: Node didn't have enough resource: hugepages-2Mi
requested: 6291456, used: 6291456, capacity: 10485760
Explicitly deleting test pod should fix this issue.
Previously the code used to delete pods serially.
In this patch we factor out code to do that in parallel,
using goroutines.
This shaves some time in the e2e tm test run with no intended
changes in behaviour.
Signed-off-by: Francesco Romani <fromani@redhat.com>
Add feature gate to disable the GetAllocatableResources API.
The feature gate isd alpha stage, disabled by default.
Add e2e test to demonstrate the behaviour with feature gate disabled.
Signed-off-by: Francesco Romani <fromani@redhat.com>
Document why teardownSRIOVPod has to wait for all the containers
to be gone before to end, and why is important.
Additionally, change the code to wait for all the containers to be gone,
not just the first. This is both a little cleaner and a little safer,
even though it seems the current code caused no issues so far.
Signed-off-by: Francesco Romani <fromani@redhat.com>
speedup the cleanup after testcases deleting pods in separate
goroutines.
The post-test cleanup stage must be done carefully since pod require
exclusive allocation - so pods must take all the steps to properly
cleanup the tests to avoid to pollute the environment, but
this has a negative effect on test duration (take longer).
Hence, we add safe speedups like doing pod deletions in parallel.
Signed-off-by: Francesco Romani <fromani@redhat.com>
Add e2e tests for the new GetAllocatableResources API.
The tests are added in the `podresources_test` suite
created previously in this series.
Signed-off-by: Francesco Romani <fromani@redhat.com>
Updates comment on building dependencies step in the local node test
runner to reflect the binaries that are actually produced.
Signed-off-by: hasheddan <georgedanielmangum@gmail.com>
- Test was failing due to using `sleep infinity` inside the busybox
container which was going into a crash loop. `sleep infinity` isn't
supported by the sleep version in busybox, so replace it with a `while
true; sleep loop`.
- Replace usage of dbus message emitting from gdbus to dbus-send. The
test was failing on ubuntu which doesn't have gdbus installed.
dbus-send is installed on COS and Ubuntu, so use it instead.
- Replace check of pod phase with the test util function `PodRunningReady`
which checks both phase as well as pod ready condition.
- Add some more verbose logging to ease future debugging.
The `apparmor_parser` binary is not really required for a system to run
AppArmor from a Kubernetes perspective. How to apply the profile is more
in the responsibility of lower level runtimes like CRI-O and containerd,
which may do the binary check on their own.
This synchronizes the current libcontainer implementation with the
vendored Kubernetes source code and allows distributions to use
AppArmor, even when they do not have the parser available in
`/sbin/apparmor_parser`.
Signed-off-by: Sascha Grunert <mail@saschagrunert.de>
The server service monitors the kubelet service and restart it
once the service is down, to avoid kubelet double restarting
we will stop the kubelet service and wait until the kubelet will be
restarted and the node will be ready.
Signed-off-by: Artyom Lukianov <alukiano@redhat.com>
