helm: improve k8sServiceHost automatic lookup function #41291
Conversation
|
Commit edd681e does not match "(?m)^Signed-off-by:". Please follow instructions provided in https://docs.cilium.io/en/stable/contributing/development/contributing_guide/#developer-s-certificate-of-origin |
edd681e to
ba407eb
Compare
|
Commit 37dff97 does not match "(?m)^Signed-off-by:". Please follow instructions provided in https://docs.cilium.io/en/stable/contributing/development/contributing_guide/#developer-s-certificate-of-origin |
37dff97 to
f61ea1b
Compare
|
Commit 37dff97 does not match "(?m)^Signed-off-by:". Please follow instructions provided in https://docs.cilium.io/en/stable/contributing/development/contributing_guide/#developer-s-certificate-of-origin |
|
Commit 3312483 does not match "(?m)^Signed-off-by:". Please follow instructions provided in https://docs.cilium.io/en/stable/contributing/development/contributing_guide/#developer-s-certificate-of-origin |
3312483 to
f61ea1b
Compare
f61ea1b to
9127bc3
Compare
In some edge-case scenarios, the chart gets installed before the cluster info ConfigMap has been created or populated. In such cases cilium gets deployed with an erroneous value ("auto") as the k8s service host and fails to start. The helm install operation won't fail immediately, but it will timeout eventually and get rolled back. This makes remediation difficult, specially with large install timeouts, which are often used on larger clusters.
With this change, if the k8s host auto discovery is requested in the chart values (by setting "auto" in `.Values.k8sServiceHost`), the chart will immediately fail to get installed if it can't find the cluster info ConfigMap, and the installer will be forced to retry the chart installation.
Signed-off-by: iuri aranda <iuri@giantswarm.io>
9127bc3 to
f826c41
Compare
sure, done |
|
/test |
|
I see some failed checks, although I doubt it's because of the changes. Do we need to re-run the CI? |
|
I've restarted them, thanks for the heads-up! I believe only organization members can trigger/restart CI. |
Please ensure your pull request adheres to the following guidelines:
description and a
Fixes: #XXXline if the commit addresses a particularGitHub issue.
Fixes: <commit-id>tag, thenplease add the commit author[s] as reviewer[s] to this issue.
In some edge-case scenarios, the chart gets installed before the cluster info ConfigMap has been created or populated. In such cases cilium gets deployed with an erroneous value ("auto") as the k8s service host and fails to start. The helm install operation won't fail immediately, but it will timeout eventually and get rolled back. This makes remediation difficult, specially with large install timeouts, which are often used on larger clusters.
With this change, if the k8s host auto discovery is requested in the chart values (by setting "auto" in
.Values.k8sServiceHost), the chart will immediately fail to get installed if it can't find the cluster info ConfigMap, and the installer will be forced to retry the chart installation.