-
Notifications
You must be signed in to change notification settings - Fork 4.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Disable Masquerading for Cilium ENI mode #16132
Conversation
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here.
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
d9e2af2
to
d4047a6
Compare
/test pull-kops-e2e-cni-cilium-eni |
1 similar comment
/test pull-kops-e2e-cni-cilium-eni |
Signed-off-by: Peter Rifel <pgrifel@gmail.com>
8f8eddd
to
9630751
Compare
9630751
to
0377987
Compare
@rifelpet: The following tests failed, say
Full PR test history. Your PR dashboard. Please help us cut down on flakes by linking to an open issue when you hit one in your PR. Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. I understand the commands that are listed here. |
PR needs rebase. Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. |
This reverts #14694. Enabling masquerading was needed when upgrading from cilium 1.10 to 1.11, but it wasn't understood why this was needed, and it didnt make sense.
Since the cilium 1.14 upgrade, cilium-eni grid tests have started failing (example). Reproing this locally I've confirmed that the tests are failing because pods with hostNetwork=false are timing out reaching pods on other nodes, including the apiserver:
Failed to watch *v1.Node: failed to list *v1.Node: Get "[https://100.64.0.1:443/api/v1/nodes?limit=500&resourceVersion=0](https://100.64.0.1/api/v1/nodes?limit=500&resourceVersion=0)": dial tcp 100.64.0.1:443: i/o timeout
Notice that all failing tests in the grid jobs involve network communication with other pods or kube-apiserver.
Troubleshooting confirmed that the request is leaving the source instance on the instance's default ENI rather than the pod IP's ENI (this is by design of cilium masquerading) and the traffic is dropped. TBH I'm not sure how enabling masquerading would have worked unless there was a bug in cilium 1.11 that has been fixed in 1.14.
With masquerading disabled, almost all of the failing tests pass. The only remaining failures are unrelated to pod networking.