The "should have correct firewall rules for e2e cluster" test is GCE
specific, and likely specific to the kube-up configuration.
However, the second half of the test is a generic behaviour based test
that verifies that ports are not reachable.
We can split this into two tests, with an eye to running the generic
test in more places.
This helps avoid some rare instances of corrupt cert files
that cause Kubelet to crash-loop after node reboots, e.g.
if Kubelet opens the file during the shutdown but is unable
to write it.
All of the After* hooks are called, whether the operation is dry-run or
not. Create and Upate both have BeginCreate() hooks which know about
dry-run but delete doesn't, and can't (because of graceful deletion and
finalizers, the Delete() method may return and the object is actually
deleted later).
This adds an argument to AfterDelete(), and the others will follow.
This commit also adds tests for AfterDelete being called in the right
places.
These hooks return a "cleanup" func which is called when the top-level
operation completes, with an indicator of which result.
This is to enable much simpler handling of allocations in Service's REST
implementation, in particular.
Some discussion in https://github.com/kubernetes/kubernetes/pull/95967
This also adds tests for the almost totally untested Decorator,
AfterCreate, and AfterUpdate hooks.
The ``.container-$OS-$ARCH`` make subaction is creating files with the same name, and ``clean`` is meant to delete them. However, the ``clean``'s rm regex is not quite correct.
webhook.WithExponentialBackoff returns an error, and the priority is:
- A: if the last invocation of the webhook function returned an error
that error should be returned, otherwise
- B: the error associated with the context if it has been canceled or
it has expired, or the ErrWaitTimeout returned by the wait package
once all retries have been exhausted.
caller should check the error returned by webhook.WithExponentialBackoff
to handle both A and B. Currently, we only handle A.
The ESIPP tests are using a function to poll an HTTP endpoint.
This function failed the framework if the request to the http endpoint
timed out, causing a panic that ginkgo couldn´t recover.
Also, this function was used inside a pollImmediate loop, so it should
return the error instead of fail.
This reverts commit 0ef7f27fc1.
The info is not enough to debug the problems, there are simply no
conntrack entries but there is no clue about it.
Another problem is that it dumps the conntrack entries from all
nodes, that is more than 40 mins in a scale test job with 5000 nodes.
