Moving DeepSeek Harness from the local ~/.dsh into k3s, with a vendored flat chart, four RWO PVCs, manual scripts instead of initContainers, and two hard constraints imposed by Tailscale L7 Ingress and Cilium SNAT.
Badges
Badges come from GitHub once the author claims this feed.
Recent posts
every link opens the author’s post-
Deploying DeepSeek Harness into K8s
-
Atuin Threads and dsh Integration in Practice
Wiring atuin mcp into dsh with a single config entry
-
IPv6 Address Present but No Connectivity: A Complete Diagnosis of an iStoreOS ip6class Mismatch
Documenting a troubleshooting session where iStoreOS obtained an IPv6 address but could not reach the internet, traced to a mismatch among ip6class, delegate, and wan6.device, with the fix and verification steps.
-
Moving Certificate Renewal from crontab to a Kubernetes CronJob
The design, secret management, rollback path, and three-layer verification strategy for moving certificate renewal from crontab to a Kubernetes CronJob, with a focus on the switchover points and the go-live checklist.
-
Homelab Domain Certificate Automation: acme.sh + Tencent Cloud/Qiniu CDN Deployment Pipeline
Issue a single multi-SAN certificate with acme.sh, deploy it automatically to Tencent Cloud CDN and Qiniu CDN through the --reloadcmd hook, and have cron check every 6 hours.
-
Fully Removing Nix: Rebuilding the Local Toolchain with mise + uv, and Handing CI Over to a Single Self-Built Image
Remove Nix entirely from both local development and CI: Python goes to uv, binary tools go to mise, and on the CI side we build a single self-hosted image on the host machine while the pipeline only pulls it — plus a check-no-nix script to keep it from creeping back.
-
Mending the Fold After a NAS Ransomware Attack
AI has pulled vulnerability discovery down to hours, and PHP2Shell and C3RB3R tore through internet-exposed NAS devices in the same week.
-
Node IPv6 Prefix Drift: The Silent Cross-Node Pod Network Failure Behind a "Just an Alert"
Behind a single kubelet secondaryNodeIP alert, cross-node IPv6 pod communication had already silently failed.
-
No Application Could Reach the Database: An AppArmor Stacking Incident Misdiagnosed as a Network Failure
After a Ubuntu kernel upgrade, signal delivery inside containers was blocked by AppArmor stacking — gosu spun, database logs stayed empty, and applications reported no route to host.
-
K8s Best Practices Updated: Use Node Swap!
A record of changing memorySwap.swapBehavior from NoSwap to LimitedSwap in a k3s cluster via a kubelet drop-in, hardening it with Ansible, and rolling it out node by node with verification.