Change in "fluentd.logs.containers.*" does not roll fluentd pods

[beyondbill]

Versions

Chart: v1.3.1
Helm: v3.3.0
Kubernetes: v1.18.2

Reproduction

On an existing installation, run

helm upgrade --namespace sumologic --set fluentd.logs.containers.xxx=yyy  sumologic sumologic/sumologic

where xxx and yyy are any valid values and yyy is different than the previous value of fluentd.logs.containers.xxx

Expectation

Statefulset sumologic-sumologic-fluentd-logs should roll all its pods to take the new setting fluentd.logs.containers.xxx=yyy into effect.

Actual result

Pods in statefulset sumologic-sumologic-fluentd-logs remain unchanged. The new setting fluentd.logs.containers.xxx=yyy won't take into effect.

Analysis

Helm takes the new setting fluentd.logs.containers.xxx=yyy and update the key-value pair in configMap sumologic-sumologic-fluentd-logs. Although statefulset sumologic-sumologic-fluentd-logs has a reference to configMap sumologic-sumologic-fluentd-logs, the contents of the statefulset template does not change. Therefore the statefulset sees no change and does not roll its pods.

I have not yet come up with any fix for this. Will update the issue when I have any idea.

[beyondbill]

@perk-sumo @sumo-drosiek Thanks for the quick response! The checksum annotation solution is brilliant! Looking forward to the release!

[sumo-drosiek]

No problem 🚀

The idea comes from official helm documentation 📖
https://helm.sh/docs/howto/charts_tips_and_tricks/#automatically-roll-deployments