A patch release that makes the v0.47.0 kind/k3d evaluation quickstart a fully working platform: make eval now stands up a runner listener that executes plans on Kubernetes Jobs, not just the API + UI.
Highlights
- Server-side runs work in the eval out of the box — the quickstart bootstraps an agent pool and runs a listener, so you can create a workspace, queue a plan in the UI, and watch it execute on a real Kubernetes Job. Previously the eval brought up only the API + UI, so queued runs sat
pendingwith nothing to execute them. The kind + k3d CI now boots the stack and runs a real plan through the listener end to end.
Bug Fixes
- Eval boots reliably on constrained nodes — the API no longer crash-loops waiting on Redis: the chart-managed (eval/dev) Redis now starts before the API, the same way the chart-managed Postgres already did.
- Server-side plans resolve OpenTofu versions reliably — the binary cache now resolves partial
tofuversions from OpenTofu's officialget.opentofu.orgindex instead of the rate-limited GitHub releases API (which returned504/502from CI and cloud IPs and could fail a plan). Terraform already used the equivalentreleases.hashicorp.comindex; this brings tofu to parity. - Eval
helm upgradeno longer deadlocks on the single-replica filesystem PVC handover.
These eval-profile changes (chart-managed Redis ordering, agent pool/listener) are gated on eval/dev settings and do not alter production deployments.
Status
Beta — actively developed and used in production by its maintainers. The evaluation quickstart now demonstrates the full plan-execution path end to end.
Full Changelog: v0.47.0...v0.47.1