[chrischeng-c4]

Summary

• Every infra/gke-test/pools.yaml row now sets pool. Terraform refuses a name that is not <repo>-<role>.
• The server pool is <repo>-server, so it pairs with <repo>-client.
• Renamed: keep → keep-server, lumen → lumen-server, lumen-perf → lumen-perf-server.
• run.sh maps pools by .pool only. The docs and the keep overlay example use the new names.

Plan

google_container_node_pool.repo["keep"]              destroy
google_container_node_pool.repo["lumen"]             destroy
google_container_node_pool.repo["lumen-perf"]        destroy
google_container_node_pool.repo["keep-server"]       create
google_container_node_pool.repo["lumen-server"]      create
google_container_node_pool.repo["lumen-perf-server"] create
Plan: 3 to add, 0 to change, 3 to destroy.

The pools are at 0 nodes when no test runs. Apply only when no test namespace is live (kubectl get ns -l faberline.dev/test-run).

Follow-up in other repositories

• keep (feat(gke-test): test keep on the shared faberline GKE cluster keep#6): overlay uses keep-server.
• lumen: overlays that select lumen or lumen-perf change to lumen-server and lumen-perf-server.

Merge and apply this before those overlays run, or their pods stay Pending.

Checks

• terraform fmt -check, terraform validate: pass.
• shellcheck -S warning scripts/gke-test/*.sh: pass.
• terraform plan: as above. Not applied.

🤖 Generated with Claude Code