Releases: kaloom/kubernetes-podagent
v0.2.1
v0.2.0
Enhancement to remove blockage in case of network attachment failure
- Remove blockage in case of network attachment failure
- Added persistent storage to store config
- Added cleanup in case of pod deletion
- Migrated to go mod for go pkg dependencies
Add PodagentSkip flag to networks annotation items
Adds podagentSkip flag to prevent podagent from configuring the network.
Fix inconsistent state when podagent is down
This fixes some cases (but not all) where there can be an state between what the annotations describes and what CNI have actually been run.
Event worker Queue
Add a worker queue to release the kubernetes event queue as soon as possible
Support retry attempts on add and delete networks on CNI failure
When a pod is moved, it is possible that the crio process is not ready to accept network request changes. in these situation, we log the problem and retry up to 5 times to process the change. This might block the event processor queue but we don't want to process any more request until the crio process is ready.
Fix the cleanup of network interfaces on pod deletion
When the podagent get called to dynamically attach a network interface to a running pod than the pod get deleted, kactus was getting confused and was not deleting all network interfaces in a pod as it should before the linux network namespace get destroyed. This was caused by the libcni caching the results of the cni-plugin calls under /var/lib/cni/results and loading the cached result of the cni-plugin invoked by the podagent instead of the one related to eth0, use separate cache file to avoid this problem.
support for cri-o container runtime
- update to vendor repos
- added support for cri-o container runtime
Initial release
update kubernetes-common Signed-off-by: Abdallah Chatila <[email protected]>