Lataa kobe 0.53.0

Kobe sandboxes now include Nix, so projects can run additional tools and define development environments without root. This release also removes timer waits from lease cleanup, fixes stale pool status writes, and keeps k3s provisioning from occupying a reconciliation worker while a cluster boots.
cargo install kobectl --locked --version 0.53.0
# Or install the prebuilt CLI:
brew install kunobi-ninja/kunobi/kobe
Added
- Sandbox tools: Nix 2.35.2 is available to kobe exec and SSH sessions. The login banner suggests nix shell; use a project's nix develop environment for build dependencies. The store counts toward ephemeral storage, and Nix builds within a lease share its isolation boundary. #355 by @jleni.
- Pool recovery: Full pools report Exhausted. Operators can explicitly release quarantined capacity, with warning events and metrics recording completed overrides. The override also works with separate teardown authority. #350, #354 by @jleni.
Fixed
- Lease cleanup: Cluster and Sandbox controllers react to resource changes instead of waiting for short polling timers. Orphaned reservations retire, normal sandbox refills preserve readiness, and stalled execution cleanup reports its cause. These changes remove waits; release-time improvements have not yet been measured on the deployed version. #336, #340, #343, #347, #349 by @jleni.
- Cluster capacity: Pool reconciliation no longer overwrites newer instance state. Concurrent lease requests enforce principal quotas and replay keyed requests. k3s startup no longer blocks a controller worker, and transient bootstrap lookups keep booting clusters pending. #337, #338, #345, #352, #353 by @jleni.
- CLI and desktop: kobe init detects advertised OIDC authentication, doctor stops reporting false auth mismatches, and trust errors point to kobe login --retrust. The sandbox terminal renders UTF-8 correctly and the desktop keeps its background. #334, #341, #342 by @jleni.
Changed
- Lease API: Canonical lease routes accept caller-owned aliases for both lease kinds. Error bodies include consistent reasons; quota refusals and retryable throttling are distinct. Cluster extension refusals now return HTTP 409 instead of 400. #351, #353 by @jleni.
Upgrade notes
Apply the release's CRDs before upgrading the operator; Helm does not update crds/ on upgrades. See Deploy.
Pools with configuration fields their backend ignores now report ConfigSupported=False and stop creating members. Existing members continue serving leases. Remove the unsupported fields to resume provisioning. #348 by @jleni.
Before rolling back to an older operator, clear Exhausted pool phases as described in Troubleshooting.
Contributors
jleni
Full changelog: v0.52.0...v0.53.0