v0.4.3-gke.0

Changes in v0.4.3-gke.0:

• Fix a bug in provisioning memory in collectors and allow CPU limit configuration.
• Enforce project_id, location, and cluster labels to always use the values provided through Prometheus external labels (e.g. self-deployed) or operator configuration (managed collection). Setting any reserved labels outside of that will cause them to be relabeled with the prefix “exported_”.
• Add reasonable yet high defaults for resource allocation for all components. This should allow GMP to be deployed on clusters with mandatory resource limit requirements.
• Add instructions for using HPA with prometheus-adapter.
• Added example deployment and scraping CRDs for kube-state-metrics.
• Add best practices and reference diagrams to documentation.
• Add SecCompProfile to obey k8s security requirements.
• Fix a bug to allow the k8s cluster autoscaler to evict GMP pods when removing nodes.

Users who deploy managed collection using kubectl should reapply the manifests in the documentation. Users who deploy the service using gcloud or the GKE UI will be upgraded on a rolling basis, with 1.24 already upgraded, 1.23 scheduled for this week, and older clusters after that. This release has no impact on self-deployed collection users.

Please note that due to a rollout issue, most gcloud deployments never were upgraded to 0.4.1. Therefore, for gcloud users this release also includes the changes from 0.4.1. Note that this release may result in your auto-added “container” label disappearing. To fix this, use a named port instead of a numeric port.

Independently of this release, querying Cloud Monitoring metrics using PromQL is now available for all users.