Writes to policy resources don't instantaneously take effect in admission. ValidatingAdmissionPolicy
integration tests determine that the policies under test have taken effect by adding a sentinel
policy rule and polling until that rule is applied to a request.
If the marker resource names are the same for each test case in a series of test cases, then
observing a policy's effect on a marker request only indicates that _any_ test policy is in effect,
but it's not necessarily the policy the current test case is waiting for. For example:
1. Test 1 creates a policy and binding.
2. The policy and binding are observed by the admission plugin and take effect.
3. Test 1 observes that a policy is in effect via marker requests.
4. Test 1 exercises the behavior under test and successfully deletes the policy and binding it
created.
5. Test 2 creates a policy and binding.
6. Test 2 observes that a policy is in effect via marker requests, but the policy in effect is still
the one created by Test 1.
7. Test 2 exercises the behavior under test, which fails because it was evaluated against Test 1's
policy.
Generating a per-policy name for the marker resource in each test resolves the timing issue. In the
example, step (6) will not proceed until the admission plugin has observed the policy and binding
created in (5).
Once received job deletion event, it cleans the backoff records for that
job before enqueueing this job so that we can avoid a race condition
that the syncJob() may incorrect use stale backoff records for a newly created
job with same key.
Co-authored-by: Michal Wozniak <michalwozniak@google.com>
Automated cherry pick of #131742: [sig-scheduling] SchedulerPreemption [Serial] validates various priority Pods preempt expectedly with the async preemption: replace finalizers with preStop hook and TerminationGracePeriodSeconds
If someone gains the ability to create static pods, they might try to use that
ability to run code which gets access to the resources associated with some
existing claim which was previously allocated for some other pod. Such an
attempt already fails because the claim status tracks which pods are allowed to
use the claim, the static pod is not in that list, the node is not authorized
to add it, and the kubelet checks that list before starting the pod in
195803cde5/pkg/kubelet/cm/dra/manager.go (L218-L222).
Even if the pod were started, DRA drivers typically manage node-local resources
which can already be accessed via such an attack without involving DRA. DRA
drivers which manage non-node-local resources have to consider access by a
compromised node as part of their threat model.
Nonetheless, it is better to not accept static pods which reference
ResourceClaims or ResourceClaimTemplates in the first place because there
is no valid use case for it.
This is done at different levels for defense in depth:
- configuration validation in the kubelet
- admission checking of node restrictions
- API validation
Co-authored-by: Jordan Liggitt <liggitt@google.com>
Code changes by Jordan, with one small change (resourceClaims -> resourceclaims).
Unit tests by Patrick.
This makes it clear the error comes due to a user namespace
configuration. Otherwise the error returned looks too generic and is not
clear.
Before this PR, the error was:
Warning FailedCreatePodSandBox 1s kubelet Failed to create pod sandbox: the handler "" is not known
Now it is:
Warning FailedCreatePodSandBox 1s kubelet Failed to create pod sandbox: runtime does not support user namespaces
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
When using an old runtime like containerd 1.7, this message is not
implemented and what we get here is an empty non-nil slice. Let's check
the len of the slice instead.
While we are there, let's just return false and no error. In the
following commits we will wrap the error and we didn't find any more
info to add here.
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
