> ## Documentation Index > Fetch the complete documentation index at: https://docs.tokenfactory.nebius.com/llms.txt > Use this file to discover all available pages before exploring further. # Troubleshooting > Diagnose access, execution, output, state, subprocess, network, and concurrency failures. Find your symptom below. The [support bundle](#support-bundle) lists the details to collect if you need help. If a request was interrupted, check its operation ID before submitting the command again. Accepted work may still be running; resubmitting can run it twice. If no operation ID was returned, resolve the uncertain submission before retrying. ## Access and images ### 401 or invalid credential Refresh or replace the credentials using [Set up access](/sandboxes/start/set-up-access). Check token expiry, configuration, and the current profile. ### 403 Select the intended project and confirm Sandboxes access. Inspect the project header and token permissions too: activation is not the only cause of a 403. ### Image or tag not found Resolve the tag again or use a recorded immutable UUID. Check the project, literal tag, and UUID in the request. ### No result image Check `disposable`, terminal status, and the result field. A disposable result cannot be reused. Retry from the input image only when execution is safe. ## Execution and output ### No output Flush buffered output, check stderr, and inspect the final process state. Confirm that the command started, the intended stream was selected, and output was not truncated. ### Nonzero exit code Correct the workload input or code using its exit code and output. A nonzero process exit is a workload result. ### Timed out or signaled Inspect `timed_out`, the signal, and exit code. Classify this separately from test failure and retry from known state. See [Timeouts and disconnections](#timeouts-and-disconnections). ### Stream ended Resume from the last processed event ID when supported, then query final operation status. Keep the event cursor and operation ID: ending the stream does not establish that execution ended. ## Subprocesses and input ### Stdin write timed out Reconcile delivery using the child ID and last application sequence. The input may already have arrived; do not blindly replay the write. ### Child missing Check parent status and the subprocess ID. A child cannot outlive a finalized parent; start a new operation. ## Service and networking ### Rate or service rejection Inspect the HTTP status, response, and token limits. Back off only as instructed. An undocumented 503 does not establish a capacity problem. ### Network unavailable Check the spawn request's `networking` field and guest interfaces. Enable networking when required, or keep verification offline. ## Timeouts and disconnections ### REST request or wait timeout Query a known operation ID before resubmitting: accepted remote work may continue. ### SDK 0.3.6 wait timeout Resolve the operation outcome. The client attempts best-effort cancellation, but that attempt does not establish that cancellation succeeded. ### Stream disconnect Resume from the last fully processed event ID and inspect final status. Execution is independent of observation. ### Execution timeout Inspect the terminal process result, including `timed_out`. Reuse a result image only when it is present and the workload validates it. ### Stdin write timeout Reconcile with application sequence IDs and avoid automatic replay: the input may have been delivered. ### Image expiry Recreate the image from pinned inputs and preparation steps. Its stored filesystem is no longer available. ## Support bundle Send the operation UUID, input and result image UUIDs when available, client version, timestamp, terminal status, a minimal reproduction, and redacted status/output to [contree@nebius.com](mailto:contree@nebius.com). Do not send authorization headers, API keys, secrets, customer files, or complete unredacted logs.