The node-kubelet-flaky e2e job that runs the the
`Node Performance Testing [Serial] [Slow] [Flaky]` e2e tests have been
flaking because of inconsistencies on the cpu manager checkpoint file.
This seems to be caused because the checkpoint file is deleted (which is
what needs to happen in order to change the CPU manager policy which is
used for these e2e tests) right after the e2e tests asserts that a pod
does not exist anymore.
However, after a pod is deleted, the CPU manager may still be cleaning
up the resources used by the pod which may result in the checkpoint file
being created.
Whenever this happened, the kubelet would panic if we then try to
subsequently change the CPU manager policy to "static" from "none" or
vice versa (this is done 4 times in these tests).
Signed-off-by: alejandrox1 <alarcj137@gmail.com>
e2e_node tests trigger OOM events on COS versions > 73-11636-0-0
possibly because of this change in the COS v.73-11636-0-0:
Made containerd run as a standalone systemd service
OOM killer usually kills cadvisor and e2e_node.test processes
causing node-kubelet-benchmark failures.
Decreasing amount of pods from 105 to 90 frees enough memory for
the test to succeed.
Lowering the amount of cpu allocated to this workload will set the
resources allocated to be similar to the other npb and tf workload in
this tests.
This will also allow to run all three workloads in a n1-standard-12 gcp
instance - which has 16 cpus and 60 GB.
Signed-off-by: alejandrox1 <alarcj137@gmail.com>
commit 43c56eb403 introduced a change
where CPUAccounting, CPUAccounting and TasksAccounting are enabled for
the systemd service.
It causes a regression on RHEL 7.8 where systemd-run doesn't allow to
set TasksAccounting.
Since Delegate= already enables all the controllers, it is superfluous
to specify them.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
'docker pull' is a time consuming operation. It makes sense to check
if image exists locally before pulling it from a registry.
Checked if image exists by running 'docker inspect'. Only pull if
image doesn't exist.
and they made an invalid dependency to sub e2e framework from the core framework.
So we can use e2epod.WaitTimeoutForPodReadyInNamespace to remove invalid dependency.
The main purpose of this pr is to handle the framework core package dependency subpackage pod.
When kubelet is restarted, it will now remove the resources for huge
page sizes no longer supported. This is required when:
- node disables huge pages
- changing the default huge page size in older versions of linux
(because it will then only support the newly set default).
- Software updates that change what sizes are supported (eg. by changing
boot parameters).
Most of these could have been refactored automatically but it wouldn't
have been uglier. The unsophisticated tooling left lots of unnecessary
struct -> pointer -> struct transitions.
This is gross but because NewDeleteOptions is used by various parts of
storage that still pass around pointers, the return type can't be
changed without significant refactoring within the apiserver. I think
this would be good to cleanup, but I want to minimize apiserver side
changes as much as possible in the client signature refactor.
