Summary

When an auth_oidc k8s task's setup fails early, its teardown step exits with an unbound-variable error before cleanup completes. Resources the setup already created are left behind, most notably an Atlas deployment in the shared drivers QA project.

Motivation

Who is the affected end user?

Driver maintainers and DevProd Runtime Environments who run and triage the test-auth-oidc-eks, test-auth-oidc-aks, and test-auth-oidc-gke tasks.

How does this affect the end user?

Leaked deployments accumulate in a shared Atlas project with a cluster cap. Enough leaks exhaust the quota, and unrelated drivers' tasks then fail with "Project cluster limit reached". The failed teardown step also adds noise to every triage of an already-failing task.

How likely is it that this problem or use case will occur?

Whenever setup fails early: quota errors, credential problems, or provider outages. Rare on a green mainline, recurring under quota pressure and during incidents.

If the problem does occur, what are the consequences and how severe are they?

Compounding leakage that fails unrelated tasks. Recovery needs manual Atlas deletions today.

Is this issue urgent?

No fixed timeline. Worth landing before the DRIVERS-3712 red/green verification merges, because its patch runs induce setup failures and consume the most quota.

Is this ticket required by a downstream team?

No.

Is this ticket only for tests?

Yes, CI scripts only. No product or functional impact.

Acceptance Criteria

• A task whose setup fails early completes teardown cleanly and states which cleanup steps it skipped.
• No resource created before the failure is left behind, or leftovers are logged for manual cleanup.
• Evidence: a task log with an early setup failure and a clean teardown finish.

Evidence

Observed in test-auth-oidc-eks: https://spruce.corp.mongodb.com/task/mongo_python_driver_auth_oidc_ubuntu_22_test_auth_oidc_eks_patch_439b3ff8e3541338d71e21d10171d918e3213076_6ae5412fc29bd70007004770_26_10_31_01_15_30/logs?execution=0

.evergreen/auth_oidc/k8s/teardown-pod.sh: line 13: K8S_VARIANT: unbound variable