- bump init backoff to Duration=30ms, Factor=8 (Steps=6) to yield ~140s total
- prevent kubelet restarts when DNS is blackholed and NSS must fall back to myhostname
- keep CSI/CSINode initialization alive long enough to complete in ARO DNS-failure scenarios
The code actually calls os.Remove(), not rmdir(). The error message
should accurately reflect the operation being performed.
os.Remove() can remove both files and directories, while rmdir()
only removes directories
Signed-off-by: Humble Devassy Chirammal <humble.devassy@gmail.com>
Certain failures during SetupDevice and MapPodDevice are not treated as
transient in the csi raw block plugin implementation, while they are in
the file mode plugin. This can lead to certain failures causing volumes
to be marked as unmounted incorrectly.
This patch brings the block plugin up to parity with the fs one by
marking the equivalent calls as transient. This mostly covers API server
and some csi driver calls.
CSI drivers can be installed while the node is still initializing since the daemonsets usually tolerate all taints. As a result, we also need to make sure the outdated CSINode object has been removed before installing a new CSI driver.
* Reject pod when attachment limit is exceeded
Signed-off-by: Eddie Torres <torredil@amazon.com>
* Record admission rejection
Signed-off-by: Eddie Torres <torredil@amazon.com>
* Fix pull-kubernetes-linter-hints
Signed-off-by: Eddie Torres <torredil@amazon.com>
* Fix AD Controller unit test failure
Signed-off-by: Eddie Torres <torredil@amazon.com>
* Consolidate error handling logic in WaitForAttachAndMount
Signed-off-by: Eddie Torres <torredil@amazon.com>
* Improve error context
Signed-off-by: Eddie Torres <torredil@amazon.com>
* Update admissionRejectionReasons to include VolumeAttachmentLimitExceededReason
Signed-off-by: Eddie Torres <torredil@amazon.com>
* Update status message
Signed-off-by: Eddie Torres <torredil@amazon.com>
* Add TestWaitForAttachAndMountVolumeAttachLimitExceededError unit test
Signed-off-by: Eddie Torres <torredil@amazon.com>
* Add e2e test
Signed-off-by: Eddie Torres <torredil@amazon.com>
* Fix pull-kubernetes-linter-hints
Signed-off-by: Eddie Torres <torredil@amazon.com>
---------
Signed-off-by: Eddie Torres <torredil@amazon.com>
When a CSI plugin attempts to register on a node, the node checks if the CSINode object exists. If it does but the ownerReference of the CSINode does not match the UID of the current node, possibly because the node object was recreated with the same name, we end up in a race condition where the CSINode object will be updated but subsequently deleted by the GC controller.
In this situation, the CSINode object will be gone and won't be recreated unless the CSI plugin or the kubelet are restarted.
This commit fixes this race by checking that the CSINode object belong to the current node during initialization.
If it doesn't, it means that the CSINode object is left over from a previous node and it must be removed first. Once removed, registration can progress as usual.
The new k8s.io/utils/ptr package provides generic wrapper functions,
which can be used instead of type-specific pointer wrapper functions.
This replaces the latter with the former, and migrates other uses of
the deprecated pointer package to ptr in affacted files.
See kubernetes/utils#283 for details.
Signed-off-by: Lan Liang <gcslyp@gmail.com>
The key difference is that the kubelet must remember all plugin instances
because it could always happen that the new instance dies and leaves only the
old one running.
The endpoints of each instance must be different. Registering a plugin with the
same endpoint as some other instance is not supported and triggers an error,
which should get reported as "not registered" to the plugin. This should only
happen when the kubelet missed some unregistration event and re-registers the
same instance again. The recovery in this case is for the plugin to shut down,
remove its socket, which should get observed by kubelet, and then try again
after a restart.
The "// import <path>" comment has been superseded by Go modules.
We don't have to remove them, but doing so has some advantages:
- They are used inconsistently, which is confusing.
- We can then also remove the (currently broken) hack/update-vanity-imports.sh.
- Last but not least, it would be a first step towards avoiding the k8s.io domain.
This commit was generated with
sed -i -e 's;^package \(.*\) // import.*;package \1;' $(git grep -l '^package.*// import' | grep -v 'vendor/')
Everything was included, except for
package labels // import k8s.io/kubernetes/pkg/util/labels
because that package is marked as "read-only".
