helm安装GraphScope如何控制etcd的nodeSelector和容器资源限制

[xiaowuyy]

charts/graphscope/README.md中有etcd容器的资源限制配置说明，但是使用后无法生效，查看源码也没有找到相关的逻辑

[ajavanma]

I traced this against the current charts/graphscope chart (version 0.31.0, commit d8581c18). Your post doesn't specify the installed chart version, so the details below are scoped to that source.

There is a mismatch between the README and the implementation: the README still shows etcd.resources, but the chart's coordinator configuration template does not read or forward etcd.resources or an etcd nodeSelector. Adding those keys to a values file therefore has no effect in this version.

The reason is that etcd is delegated to Vineyard. The Kubernetes launcher invokes vineyardctl.inject() with the Vineyard sidecar CPU/memory settings, but no etcd resource or scheduling settings. Its separate Vineyard deployment path likewise passes etcd_replicas=1, without those settings. Changing vineyard.resources controls the Vineyard container, not etcd.

For the integrated path, this needs plumbing through both the chart/coordinator configuration and the Vineyard deployment logic; there isn't a working --set etcd... switch in this chart. Another supported integration point is an existing Vineyard Deployment in the same namespace via vineyard.deployment; then you manage its etcd workload's resources and scheduling in that deployment's manifests. The existing-deployment lookup shows that path.

Check the installed chart version with helm list -n <namespace> before applying this diagnosis to an older release. This is a source review; I haven't deployed the workaround on a cluster.