[receiver/vcenter] vcenter.cluster.name Resource Attribute Does Not Make Sense for Datastore Resource

[StefanKurek]

Component(s)

receiver/vcenter

What happened?

Description

Datastores do not “belong” to Clusters (especially not a single Cluster). Datastores exist on the Datacenter level. But Hosts & Clusters (more specifically the Hosts within Clusters) make use of them.

Currently, there is odd behavior where Datastore metrics are repeated for each cluster (although the metric datapoint values are all the same, just with a different cluster attribute on the containing Resource). Datastore metrics are also repeated for standalone Hosts, but no matter how many of these Hosts exist, there will only be one set of metrics (because the containing Resource will not have a cluster attribute).

Steps to Reproduce

Collect against any vCenter environment with Datastores and some mix of multiple Clusters/Standalone Hosts.

Expected Result

A single set of metrics would come back for each Datastore under a Resource that represents that Datastore.

Actual Result

For each Datastore, Datastore Resources are created for each Cluster with repeated metrics. An additional Datastore Resource will also be created for any amount of standalone Hosts with repeated metrics.

Collector version

v1.5.0/v0.98.0

Environment information

No response

OpenTelemetry Collector configuration

extensions:
  basicauth/prom:
    client_auth:
      username: [PROMUSER]
      password: [PROMPASS]

exporters:
  prometheusremotewrite:
    endpoint: [PROMENDPOINT]
    auth:
      authenticator: basicauth/prom
    resource_to_telemetry_conversion:
      enabled: true # Convert resource attributes to metric labels

processors:
  batch:
    # https://github.com/open-telemetry/opentelemetry-collector/tree/main/processor/batchprocessor

receivers:
  vcenter:
    endpoint: https://[VCENTERHOST]
    username: [VCENTERUSER]
    password: [VCENTERPASS]
    tls:
      insecure: true
    collection_interval: 1m
    initial_delay: 1s

service:
  extensions: [basicauth/prom]
  pipelines:
    metrics:
      receivers: [vcenter]
      processors: [batch]
      exporters: [prometheusremotewrite]

Log output

No response

Additional context

No response

[github-actions]

Pinging code owners:

• receiver/vcenter: @djaglowski @schmikei

%s See Adding Labels via Comments if you do not have permissions to add labels yourself.

[StefanKurek]

@djaglowski I think there are a couple of options.

1. Remove the vcenter.cluster.name attribute completely from the Datastore resources
2. Add a new vcenter.cluster.names attribute to Datastore resources which would contain a list of all clusters that make use of this Datastore (for relational information). We could also choose to do the same thing with a vcenter.hosts.names attribute.

Neither of these seems ideal, but I think (# 1) would be especially unfortunate as we would lose out on this information.

[schmikei]

I think between these two solutions, I'd have a preference of Remove the vcenter.cluster.name attribute completely from the Datastore resources

I'm of the opinion that with #32531 being in I have less of a concern around uniquely identifying datastores. And that this probably more accurately defines a datastore.

[djaglowski]

I agree removing the attribute makes the most sense here. Since it'll be a breaking change, we should first add a warning to be released in the next version, then remove in the following version.