Skip to main content
Start with the current project and original operation:
Read the stable error code, suggested action and any retry interval. Keep source-admission failures, build failures and pending DNS/mail separate.

Reconcile only when requested

Use this when the operation reports reconciliation is required and no attempt is pending. Retain the same request key across uncertainty. A completed reconciliation attempt does not itself mean the original deployment succeeded.

Read diagnostic pages

Deployment diagnostics/logs are retained seven days. CLI deployment logs reads one page of at most100 items despite —follow; MCP deployment_logs takes cursor. Pass next_cursor as cursor for older records. DEPLOYMENT_FAILED carries no detail: follow the failed operation’s suggested_action. Operation logs is separate: CLI logs —follow has bounded reconnects/Last-Event-ID; operation_stream_ended means the original still runs,so poll operation get about every sixty seconds. No work was cancelled.

Error codes

Act on code,retryable andsuggested_action. Only CLI errors returned by the API add docs_url; local CLI/MCP and terminal operation failures use their own guidance.

Report a platform issue

If an error lacks a useful cause or the documented operation cannot continue, submit feedback with the project/operation IDs, expected behavior and a short redacted reproduction. Do not paste tokens, secret values, raw environment files or customer records. A provider observation failure means unavailable, not ready or zero usage. Keep customer source changes separate from a provider-capacity issue.