Errors

Errors use Connect codes (the same names as gRPC). Over the JSON protocol an error looks like:

{
  "code": "invalid_argument",
  "message": "page_size must be between 1 and 500",
  "details": [
    { "type": "google.rpc.BadRequest", "value": "..." }
  ]
}

The HTTP status follows the Connect mapping (for example invalid_argument is 400, permission_denied 403, not_found 404, resource_exhausted 429).

Codes you will meet

Code	When	Detail
invalid_argument	Bad input	google.rpc.BadRequest with field violations (field path and description)
unauthenticated	Missing, expired or revoked credentials
permission_denied	You lack a permission	The message names the missing permission, for example environments.deploy:production
not_found	Unknown id, or an object you cannot see	Cross-tenant ids look the same as missing ones
already_exists	Duplicate slug or name
failed_precondition	A state-machine violation	For example approving a deployment that is not waiting_approval, cancelling a finished run
resource_exhausted	A plan limit or rate limit	A slipway.v1.EntitlementViolation detail names the limit and the plan that raises it
aborted	A concurrent modification	Re-read and retry
unavailable	Transient server or dependency problem	Safe to retry with backoff
internal	A bug. Includes a request id in the message	Report it

Handling

resp, err := client.GetRun(ctx, connect.NewRequest(&slipwayv1.GetRunRequest{Id: id}))
if err != nil {
    switch connect.CodeOf(err) {
    case connect.CodeNotFound:
        // ...
    case connect.CodeResourceExhausted:
        for _, d := range connect.NewError(connect.CodeOf(err), err).Details() { _ = d }
    }
}

Retries

• Retry unavailable and, for rate limits, resource_exhausted, with exponential backoff and jitter. Honour Retry-After when present.
• Do not retry invalid_argument, permission_denied, failed_precondition or not_found without changing something.
• Create calls are not idempotent unless the method documents an idempotency key; when a create times out, list before re-creating.