When supporting rolling updates, we cannot use the same fixed socket paths for
old and new pod. With the revised API, the caller no longer specifies the full
socket paths, only directories. The logic about how to name sockets then can be
in the helper.
While at it, avoid passing a context to the gRPC helper code when
all that the helper code needs is a logger. That leads to confusion
about whether cancellation has an effect.
add missing metric about uncore / L3 / Last-Level cache alignment,
plus its e2e tests.
Exposing uncore alignment requires a bit of refactoring in the static
policy implementation because, differently from full PCPUs alignment
and NUMA alignment, can't be easily and safely inferred by construction.
The main reason for this is that uncore cache alignment is preferred,
not mandatory, thus the cpu allocator can legally use cross-uncore
allocation. Because of that, the final cpuset union step can
create a final cpuset which is not uncore-aligned even though all
its parts are uncore-aligned.
The safest way seems thus to run just a final uncore-alignment check
once the final cpuset is computed.
Signed-off-by: Francesco Romani <fromani@redhat.com>
Graduate the feature to beta, by:
- Allowing `subPath`/`subPathExpr` for image volumes
- Modifying the CRI to pass down the (resolved) sub path
- Adding metrics which are outlined in the KEP
Signed-off-by: Sascha Grunert <sgrunert@redhat.com>
ImageGCNoEviction fails when tests run by kubetest2 as the test depends
on the prepulled test images (framework.TestContext.PrepullImages), but
kubetest2 --prepull-images command line option is set to false by
default.
Prepulling images explicitly for the only test that uses them
should fix the issue.
add (admittedly pretty crude) CPU allocatable check.
A more incisive refactoring is needed, but we need
to unbreak CI first, so this seems the minimal decently clean test.
Signed-off-by: Francesco Romani <fromani@redhat.com>
When proc mount is set to default, it should mask /proc.
The DefaultProcMount test was setting "hostUsers: false" which means to
create a user namespaces. This was not causing issues before, because
user namespaces was disabled by default and therefore the field was
completely ignored. Now that userns is enabled by default, the test is
failing as the runtime doesn't always have userns support.
One option would be to filter for runtimes that do have userns support.
But the default case (/proc is masked) for sure we want to test it
without userns support, as it will be applied to all pods.
To that end, we add a param "hostUsers bool" to testProcMount that will
enable it or not. Then, both test cases that call this function set it
accordingly: the default case sets it to true (no user namespace), and
the unmasked case with a privileged pod sets it to false (use a user
namespace), to verify the /proc mount is unmasked in this case.
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
Our CI machines happen to have 1 fully allocatable CPU for test workloads.
This is really, really the minimal amount. But still should be sufficient for the tests to run
the tests; the CFS quota pod, however, does create a series of pods (at time of writing, 6)
and does the cleanup only at the very end the end. This means pods
requiring resources accumulate on the CI machine node.
The fix implemented here is to just clean up after each subcase.
Doing so the cpu test footprint is equal to the higher requirement (say, 1000 millicores) vs
the sum of all the subcases requirements.
Doing like this doesn't change the test behavior, and make it possible
to run it on very barebones machines.
