# GKE Atlas Kubernetes Operator > Source-anchored research for GKE Atlas Kubernetes Operator. Run status: COMPLETED. Parent: [MongoDB Atlas on GCP](https://llms-explorer.com/tree/mongodb-atlas-on-gcp/) · 4 facets · 11 facts · page: https://llms-explorer.com/tree/gke-atlas-kubernetes-operator/ ## How it works - GKE network mode matters for Atlas peering: MongoDB documents VPC-native connectivity and excludes routes-based GKE peering because Atlas does not accept custom routes. Google documents --enable-ip-alias when creating a VPC-native cluster. Plan the GKE network before declaring Atlas peering resources. (medium confidence) — [source](https://www.mongodb.com/docs/atlas/manage-vpc-peering.md) - A July 25, 2025 GKE Autopilot report describes intermittent UnknownHostException and server-selection timeouts despite VPC peering and successful Java-driver and mongosh connections. The thread does not establish an operator defect or a confirmed DNS root cause; resource reconciliation alone is insufficient evidence of application connectivity. (low confidence) — [source](https://www.mongodb.com/community/forums/t/gke-autopilot-atlas-vpc-peering-intermittent-dns-resolution-failures-despite-successful-java-driver-shell-connection/326252) ## Problems, failure modes and limitations - Issue 606 records a July 2022 GCP Atlas autoscaling reconciliation failure: DeploymentNotUpdatedInAtlas with ATTRIBUTE_READ_ONLY for createDate. It gives a concrete reason to inspect AtlasDeployment conditions and Atlas API responses when reconciliation fails, but does not identify a GKE-specific cause or prove the bug persists. (low confidence) — [source](https://github.com/mongodb/mongodb-atlas-kubernetes/issues/606) ## Comparisons and alternatives - On GKE, the Atlas Kubernetes Operator is a Kubernetes control plane for MongoDB Atlas resources. It manages externally hosted Atlas databases through custom resources rather than running database members on GKE nodes. (high confidence) — [source](https://github.com/mongodb/mongodb-atlas-kubernetes) ## Facts and statements - Atlas projects, deployments, users, IP access lists, peering and private endpoints can be managed through the operator. Declarative configuration can therefore travel through the Kubernetes API alongside application configuration. (high confidence) — [source](https://github.com/mongodb/mongodb-atlas-kubernetes) - MongoDB publishes distinct Helm charts for the Atlas operator and for self-managed MongoDB controllers. The seeded GKE multi-cluster reference architecture creates GKE clusters for the latter architecture; its node layouts are not Atlas operator deployment requirements. (medium confidence) — [source](https://github.com/mongodb/helm-charts) - For Kubernetes-based platform workflows, the operator reconciles Atlas custom-resource specifications with Atlas state; MongoDB describes GitOps and Crossplane compositions as ways to expose this workflow to application teams. This is applicable to a GKE-hosted control plane without making the database a GKE workload. (medium confidence) — [source](https://github.com/mongodb/mongodb-atlas-kubernetes) - The documented Google Cloud private-endpoint workflow spans two control planes: declare provider GCP and region in AtlasProject, obtain Atlas service attachment names, create Google Cloud addresses and forwarding rules, then supply gcpProjectId, endpointGroupName and endpoints in the custom resource. An AtlasProject manifest alone does not perform all consumer-VPC provisioning in that procedure. (low confidence) — [source](https://www.mongodb.com/docs/atlas/operator/current/ak8so-private-link-dedicated.md) - After the documented Google Cloud endpoint setup, inspect AtlasProject status and use the operator-generated application Secret with connectionStringPrivate. A Kubernetes Secret reference lets the application consume that private connection string. (low confidence) — [source](https://www.mongodb.com/docs/atlas/operator/current/ak8so-private-link-dedicated.md) - A 2020 first-hand GKE integration tutorial demonstrates private nodes with Cloud NAT, a reserved public egress address and an Atlas IP access-list entry. This is historical evidence for the stable-egress pattern, not evidence that its old tier names, UI steps or AWS-only private-endpoint statement remain current. (low confidence) — [source](https://dchaykin.medium.com/connect-a-gke-cluster-with-mongodb-atlas-through-cloud-nat-b0ffb2683b7d) - Issue 261 records an August 2021 environment of GKE 1.19.10-gke.1700 and AKO 0.5.0, with a report of an Atlas project reappearing after deletion. This is historical GKE lifecycle evidence, not a statement about present deletion behavior. (low confidence) — [source](https://github.com/mongodb/mongodb-atlas-kubernetes/issues/261)