Fix NVIDIA DKMS build failures after upgrade
A kernel upgrade made the NVIDIA DKMS build fail. The log DKMS leaves under /var/lib/dkms names the real cause. Here is how to read it and fix it.
Filtering by topic #troubleshooting · clear
A kernel upgrade made the NVIDIA DKMS build fail. The log DKMS leaves under /var/lib/dkms names the real cause. Here is how to read it and fix it.
Read the SearXNG service log and /stats to tell a timeout from a suspended engine or a crash like KeyError, then fix each one in settings.yml.
Claude web checkout refuses while your App Store subscription is still inside its paid period or in billing retry. What to check, and how long to wait.
nvidia-smi fails with Driver/library version mismatch when an upgrade replaces the NVIDIA libraries under a loaded kernel module. Diagnose it, then fix it.
Update one container in a multi-service stack: docker compose up -d web, why depends_on still starts, --no-deps, start vs up, stop, and --scale.
nvidia-smi cannot talk to the driver. Walk the ladder: lsmod, dmesg, dkms status, then fix missing headers, nouveau, a kernel update or Secure Boot.
Your 24.04 to 26.04 upgrade stopped: reattach the screen session, repair dpkg, fix half-flipped sources, and know when to restore the snapshot instead.
Claude Code says invalid API key while you pay for a subscription? Find the stray ANTHROPIC_API_KEY on your VPS, and learn why /login will not fix it.
Omarchy update stops with invalid or corrupted package PGP signature? Refresh archlinux-keyring first, then repair a broken pacman keyring by hand.
An update broke something on Rocky or AlmaLinux 9. Use dnf history to find the transaction, undo it, and pin the package so it cannot come back.
Ollama context deadline exceeded means a timeout fired before the model answered. Find which layer set it: your client, the load, keep_alive, or nginx.
SearXNG engines answer a VPS with CAPTCHA pages more often than a home connection. Read the real error, then pick a fix that survives a restart.
Restart, up -d, rebuild or pull: which Docker Compose command picks up an edited compose file, a new .env, a changed Dockerfile or a new image.
A process vanished with no error of its own. Find the OOM kill in dmesg and journalctl -k, and read the disk and network messages a guest kernel sends.
Argument list too long is the kernel refusing an oversized exec, and the glob you typed is usually why. Reproduce it, then fix it with xargs or find.
Claude Code shows an overloaded message when Anthropic's API returns HTTP 529. Tell that apart from a 429 rate limit and from a plan usage limit.
Locked out because cloud-init never made your login user? Find the real username, read cloud-init status and the logs, and get back in from the console.
Slow WireGuard is usually an MTU problem. Find your path MTU by bisection, clamp TCP MSS, check steal time, and measure the path before blaming the tunnel.
Read systemctl status before anything else. What 203/EXEC and 226/NAMESPACE mean, and why a unit can start cleanly and exit one second later.
You paid for a higher Claude plan and still hit the same wall. Work through the checks in order: billing timing, account mismatch, org seats, stale CLI login.
Requires, Wants, After, Before, ExecStartPre and the Condition family do different jobs. What each one promises, and how to debug a unit that never runs.
A save that fails in nano has four usual causes. Check the file owner, the parent directory, a read only or full filesystem, then your uid in a container.
See what is really listening with ss, then test the port from outside with nc or nmap. Learn why a blocked port hangs while a closed one refuses at once.
Every published DeepSeek Harness build is a release candidate. Pin an exact dsh version, clear the npx cache, and check which npm your Node bundles.
Too many authentication failures means your SSH agent offered every key it holds and the server cut you off. Diagnose it with ssh -v and fix it for good.
FTP logs in fine and then the directory listing hangs. The data channel needs its own ports. Here is the passive range to declare, and the firewall rule.
Port 10250 is the kubelet API. Fix the address already in use error on kubeadm init, and the firewall drops that break kubectl logs and exec.
Your plan quota is a different control. This message means Claude used up the tool calls allowed inside a single turn. Here is how to finish the task.
Your VPS has three clocks and only one matters. Find the drift, read chronyc and timedatectl output, and fix the sync that broke your 2FA login.
Five reasons a cron job never runs: a minimal PATH, an unescaped percent sign, the wrong crontab, output lost to mail, and a script that assumes your shell.
Four different failures look like n8n going offline. Learn to tell the websocket banner, a restart loop, an out of memory kill and a dead schedule apart.
Your VPS disk reads full and a directory walk cannot find the space. Find the deleted file a process still holds open, and the other causes.
SearXNG 429 errors have two causes: your own limiter, or the engines blocking your server IP. Read the log, fix the right one, and stop guessing.
Your unit says active but the daemon is gone. Pick the right systemd Type= for simple, exec, forking, oneshot and notify, and find the real main PID.
Your Ubuntu server reports no new release found. Work through the Prompt setting, the LTS point release gate, third party repos and held packages.
Two SSH errors, opposite causes. Refused means the server answered. Timed out means nothing answered at all. Learn which test to run, and from where.
Most Tailscale install errors on Ubuntu are apt errors. Read the status code apt printed, then fix the release codename or the signing keyring.
Permission denied (publickey) is five different faults. Read the ssh -v output, find which one you have, and fix it without locking yourself out.
apt update says a target is configured multiple times. Find the legacy .list and deb822 .sources pair, keep one, and get a clean update run back.
Your WireGuard tunnel is up but names do not resolve, or the queries leak to the local router. Find which of the three DNS failures you have, and fix it.