fluent
bug: With stakater reloader restart option, one replica is getting restarted
Describe the issue
Hi @cw-Guo & Team, Deployed fluent-bit using fluent-operator with v3.3.0. I am using stakater reloader as below to restart the fluent-bit pods in case of any configurational changes:
Restart fluent-bit daemonset using stakater relaoder if there are any configurational changes.
fluent-bit
annotations:
secret.reloader.stakater.com/reload: "fluent-bit-config"
This is restarting only one replica of fluent-bit daemonset though the "kubectl rollout restart daemonsets fluent-bit -n namespace-dummy" restarting all the replicas. Is there anything that I need to work on using stakater reloader? Can you please please help me out here. Thanks in advance!
Similar issue ref : https://github.com/fluent/fluent-operator/issues/1479
To Reproduce
Deployed fluent-bit using fluent-operator with v3.3.0. I am using stakater reloader to restart the fluent-bit pods in case of any configurational changes. This actually restarting only one replica of fluent-bit daemonset
Expected behavior
With stakater reload option, should actually restart all the replicas.
Your Environment
- Fluent Operator version:v3.3.0
- Container Runtime:
- Operating system:
- Kernel version:
How did you install fluent operator?
No response
Additional context
No response
Hi @btalakola that's depends on how the tool stakater reloader implement the rollout.
Steps to resolve this issue:
- understand how stakater reloader implement the rollout, I would assume that they use an annotation to implement that.
- refer to this https://github.com/fluent/fluent-operator/pull/1375 to retain that annotation too.
@cw-Guo Thanks for the response.
- For the Point#1: Stakater reloader implmenets the rollout restart in the following manner : https://github.com/stakater/Reloader/blob/master/internal/pkg/callbacks/rolling_upgrade.go#L382-L387
// UpdateDaemonSet performs rolling upgrade on daemonSet func UpdateDaemonSet(clients kube.Clients, namespace string, resource runtime.Object) error { daemonSet := resource.(*appsv1.DaemonSet) _, err := clients.KubernetesClient.AppsV1().DaemonSets(namespace).Update(context.TODO(), daemonSet, meta_v1.UpdateOptions{FieldManager: "Reloader"}) return err }
2. For the Point#2 : With PR-1375, fluent operator "kubectl rollout restart daemonsets fluent-bit -n ns-dummy " issue is fixed but not the Reloader use case. We have Reloader uses for other deamonsets where the restart of the pods is working fine in case configmap/secret changes but it is not working fine for fluent operator. Can you provide any suggestions to make it to work as expected?
Thanks in advance!
