If no nodes have NodeExternalIP addresses, then clearly none of the
services are exposed externally, and the test should succeed.
Seen in OpenShift CI.
This reverts commit 0c2cee5676e64976f9e767f40c4c4750a8eeb11f.
The health check containers are not required for any test, but we want
to run them anyway to ensure that they cause no unexpected issues.
The purpose of the pod created by `createBalancedPodForNodes()` is to ensure
that all nodes have equal resource requests (as seen by the scheduler). This
prevents the default scheduling behavior (which attempts to balance resource requests)
from interfering with e2e's which test other priorities/score plugins.
Because the scheduler only worries about requests, specifying `Limits` in this pod
is unnecessary. In fact, if the calculated "balancing" limit is too low, it can cause
the balancing pod to never start due to OOMKill errors, leading to flakes and failures.
In one mock test, the snapshotter needs permission to read
secrets. That was disabled in the RBAC files of recent releases. We
need to patch it back in during deployment.
They are not needed for any of the tests and in practice apparently
caused enough overhead that even unrelated tests timed out. For
example, in the pull-kubernetes-e2e-kind test, 43 out of 5771 tests
failed, including tests from sig-node, sig-cli, sig-api-machinery,
sig-network.
Mirroring the various YAML files by hand is tedious. The new
update-hostpath.sh does all the necessary steps automatically.
The result is now a bit more consistent with the upstream repos in the
sense that the original file names and paths for the RBAC YAML files
are used.
The csi-hostpath-testing.yaml is included for the sake of
completeness, but not used during E2E testing.
The new hostpath driver release is v1.6.2, which adds the
external-health-monitor for the first time.
We were avoiding the scheduled using the pod.Spec.NodeName directly,
however, once we switched to using the node selector, the no_snat
e2e test started to fail because was trying to schedule pods on
nodes with taints, hence, failing the test.
based on this comment in
ea07644522/test/e2e/framework/pod/node_selection.go (L96-L101)
// pod.Spec.NodeName should not be set directly because
// it will bypass the scheduler, potentially causing
// kubelet to Fail the pod immediately if it's out of
// resources. Instead, we want the pod to remain
// pending in the scheduler until the node has resources
// freed up.
To run the tests in a single node cluster, create two pods consuming 2/5 of the extended resource instead of one consuming 2/3.
The low priority pod will be consuming 2/5 of the extended resource instead so in case there's only a single node,
a high priority pod consuming 2/5 of the extended resource can be still scheduled. Thus, making sure only the low priority pod
gets preempted once the preemptor pod consuming 2/5 of the extended resource gets scheduled while keeping the high priority pod untouched.
