The replacement 1 TB NVMe was provisioned as the Kubernetes worker's runner storage, the preserved readable state was restored, and every configured runner lane was returned to service.
drg-k8s-worker-01.auth.home.lan/dev/nvme0n1, Intel SSDPEKNW010T8, serial BTNH942420E71P0B/dev/nvme0n1p1, GPT, one ext4 partition labeled runner-storage15532216-8eb0-4355-9f1f-3da233d6e377/mnt/runner-storage, rw,noatime/etc/fstab entry with nofail and a 10-second device timeout/etc/fstab.pre-runner-storage-replacement-20260923The administrator explicitly authorized erasing the replacement drive after checking for a wallet. A mistaken deleted-file recovery path was later corrected: the resulting wallet-recovery folder on /mnt/bigdata/personal/backups/ and its temporary NTFS scan files were deleted, and the temporary ntfs-3g packages were removed.
The preserved worker archive at /mnt/bigdata/recovery/drg-k8s-worker-01/2026-09-22-nvme-ext4/worker-readable-state.tar was verified against its recorded SHA-256 checksum before extraction. The archive restored the AutoCoder worktrees and administrator task data. Rebuildable build targets, Docker layers, caches, and temporary paths were intentionally not present in the archive.
The canonical /mnt/runner-storage/autocoder-engine ownership and setgid mode were reapplied, and runner data, cache, and Docker directories were provisioned for the general and Rust runner lanes.
e2fsck -f -n -v passed with exit status 0 and no bad blocks.findmnt --verify reports no errors.The old failing NVMe is no longer used by the worker. The verified archive on /mnt/bigdata/recovery/ remains the rollback and forensic preservation source; it was not deleted during this change.