Troubleshooting

Start with the object itself. nodus describe <kind>/<name> shows its phase, conditions, recent events, attempts and cost so far, and every error names a code with a fix and a docs link. When you contact support, include the requestId from the error.

Signing in

• The browser never opens. Run nodus login --device and approve the code from any device.
• A CI machine needs a key. Create an API key in the console (or a ServiceAccount key) and pipe it in: echo "$NODUS_API_KEY" | nodus login --with-token. NODUS_API_KEY alone also works.
• The wrong org or project. nodus whoami prints the principal, org, project and scopes in use. nodus config get-contexts lists every org you signed in to and nodus config use-context switches between them.

A launch is refused

Code	Meaning	What to do
InsufficientCredits (402)	The hold for this launch is larger than your available credit	Top up with nodus billing top-up <usd>, or lower --max-cost
BudgetExceeded (402)	A budget that covers this object would be exceeded	Raise the budget or launch in a project it does not cover
QuotaExceeded (429)	An org or project quota is reached	nodus get quotas; starter limits lift at your first purchase
CapacityUnavailable (503)	No offering matches the requirements right now	Allow more accelerators or regions, or wait for the ETA in the message
Invalid (422)	The spec fails validation	The message names each field; nodus explain job.spec documents them

Every code has its own page under Error codes.

A job does not start

• Pending for a long time. nodus describe job/<name> shows why under Conditions and Events, including the offerings considered and why others were rejected.
• The image fails to pull. Check the image reference and, for a private registry, the pull Secret it names.

A job fails or stops

• The command exits non-zero. nodus logs job/<name> shows your program’s output; the phase is Failed with the exit code. nodus run exits with the same code.
• It stopped when credit ran out. A running job stops gracefully inside its reserved amount and reports it in its conditions. Add credit and it resumes from its last checkpoint.
• The CLI itself failed. Exit code 125 is a Nodus or API error and 124 a timeout, never your program’s.

Outputs and checkpoints

• An output is missing. Only paths declared with --output NAME=/path (or spec.outputs) are collected. List them with nodus get job/<name> -o jsonpath='{.status.outputs}'.
• A resumed job started from scratch. Nodus restores the files your program saved in its checkpoint directory (NODUS_CHECKPOINT_DIR); your program must load them on start. Restoring files does not restore process memory.

Billing

• A charge looks higher than the run time. Rented machines bill from creation to confirmed deletion, including start-up and shutdown. nodus get usage --field-selector object.name=<name> --group-by segment itemizes it.
• Where did my credit go? nodus billing shows available credit, open holds and budgets; nodus get usage --group-by project breaks spending down.