Both EventHorizon hostnames returned HTTP 502. The EventHorizon deployment itself remained healthy with two ready replicas and valid service endpoints.
The single-node cluster crossed the kubelet ephemeral-storage eviction threshold. Kubernetes evicted the NGINX ingress controller and applied node.kubernetes.io/disk-pressure:NoSchedule, leaving its replacement Pending.
Removed 2.0 GiB of reproducible Rust build artifacts with cargo clean from /home/ax_julian/temp/eventhorizon-roleplay-enforcement/target. No source, session, database, or persistent application data was removed.
Disk pressure and the taint cleared automatically. Kubernetes scheduled a fresh ingress controller, which became Ready with zero restarts.
eventhorizon.app.lan and event-horizon.app.lan returned HTTP 200./healthz returned {"status":"ok"}./readyz reported Postgres and Qdrant ready.sha-710348c2a77a, two replicas ready.DiskPressure=False; no disk-pressure taint remained.Review retention of temporary worktrees and build artifacts under /home/ax_julian/temp; they account for most reclaimable host storage. Do not bulk-delete them without checking worktree ownership and current tasks.