Resource Labels Pod ID and Namespace ID are using the Name not the ID properties

[twistedpair]

The following values are incorrectly populated by MonitoredResourceUtil

• resoure.labels.pod_id
• resoure.labels.namespace_id

The values being used are the name value of the respective entities, not the ID properties.

This is inconsistent with the Kubernetes API, and with other appenders used within GKE (e.g. the Fluentd appender), which properly populate the pod_name and namespace_name labels with the metadata.name property.

Expected Behavior

• resoure.labels.pod_id <- pod.metadata.uid
• resoure.labels.namespace_id <- namespace.metadata.uid
• resoure.labels.pod_name <- pod.metadata.name
• resoure.labels.namespace_name <- namespace.metadata.name

Impact
The present bug means that you must search in cloud logging with extra factually incorrect predicates like

resource.labels.pod_name="my-pod-name" OR resource.labels.pod_id="my-pod-name"

and means certain Stackdriver log based metrics and filters are not possible due to this inaccuracy.

Since other GKE tools like the built in Fluentd appender properly set resource.labels.pod_name, example Fluentd output below:

Environment details

1. Specify the API: Cloud Logging, e.g. used with the Google Logback Cloud Appender
2. OS type and version: all
3. Java version: all
4. logging version(s): all

Steps to reproduce

1. Emit Java log line
2. View results in Cloud Logging

External references such as API reference guides

• Kubernetes Object Metadata Spec

Any additional information below

While changing the pod_id value to the correct entity.metadata.uid property would possibly impact library users, adding the pod_name and namespace_name labels, with the correct values, would allow Cloud Logging users to properly search by the name labels across all logging sources.

[suraj-qlogic]

@twistedpair,Thanks for filling issue.For a Kubernetes API container instance k8s_container is used as monitored resource type.
I have already submitted PR for this feature which is under review.Please follow here java-logging-206 for progress/work related to your request.

[twistedpair]

Thanks, @suraj-qlogic . Glad to see your PR merged.

Looks like namespace_id will still be the wrong value though, but that seems to be a bug in the underlying GCE metadata server. I'll update this issue.

[minherz]

@twistedpair should renaming pod_id to pod_name and namespace_id to namespace_name be sufficient to resolve the problem? from the report it looks like the current label naming is misleading and does not match the logging that is captured from STDOUT by Logging agent. however, i would like to confirm it before applying the change.

[minherz]

I see that the code uses pod_id and namespace_id labels with the old resource type name: container. The k8s_container resource type is populated with the expected labels (pod_name and namespace_name). There is PR #641 that handles the change of the resource type name. After the merge, the expected set of label names will be populate.