BerriAI/litellm · error · RuntimeError

Failed to import file into RAG corpus: {e}

Error message

Failed to import file into RAG corpus: {e}

What it means

RuntimeError raised when the SDK call that imports an already-uploaded GCS file into the RAG corpus raises any exception. The original exception is logged and chained with `raise ... from e`, and its text is embedded in the message - so the underlying google-cloud SDK error is the real signal. It fires only after the GCS upload step succeeded.

Source

Thrown at litellm/llms/vertex_ai/rag_engine/ingestion.py:297

            Tuple of (corpus_id, gcs_uri)
        """
        if not file_content or not filename:
            verbose_logger.warning("No file content or filename provided for Vertex AI ingestion")
            return _get_str_or_none(self.corpus_id), None

        # Step 1: Upload file to GCS
        gcs_uri: Final = await self._upload_file_to_gcs(
            file_content=file_content,
            filename=filename,
            content_type=content_type or "application/octet-stream",
        )

        # Step 2: Import file into RAG corpus
        try:
            await self._import_file_to_corpus_via_sdk(gcs_uri=gcs_uri)
        except Exception as e:
            verbose_logger.error("Failed to import file into RAG corpus: %s", e)
            raise RuntimeError(f"Failed to import file into RAG corpus: {e}") from e

        return str(self.corpus_id), gcs_uri

View on GitHub (pinned to 77b7c6c40c)

Solutions

1. Read the SDK error text inside the message - it comes verbatim from google-cloud-aiplatform and names the real cause
2. Verify the corpus name: same project, same location, existing corpus ID
3. Enable the Vertex AI API in the project and grant the service account Vertex AI User (plus RAG permissions)
4. Note that the GCS upload already succeeded - focus on the corpus/API side, not the bucket

def corpus_name_ok(project: str, location: str, corpus_id: str) -> bool:
    return all([project, location, corpus_id]) and location == EXPECTED_CORPUS_LOCATION

assert corpus_name_ok(project, location, corpus_id), 'corpus name parts missing or location mismatch'

try:
    ...  # run the Vertex RAG ingestion (upload + import)
except RuntimeError as e:
    # e.__cause__ holds the original google-cloud SDK exception
    log.error('rag import failed: %s', e.__cause__ or e)
    raise

When it happens

Trigger: Corpus resource name wrong or in another region (projects/{project}/locations/{location}/ragCorpora/{id} does not resolve); the Vertex AI / RAG API not enabled; the service account lacks aiplatform or RAG permissions; corpus quota limits hit.

Common situations: vertex_location mismatch between corpus and config; SA granted storage rights but not Vertex AI User; corpus deleted between job creation and import; document in a format the importer rejects.

Related errors

• gcs_bucket is required for Vertex AI RAG ingestion. Set via
• vector_store_id (corpus ID) is required for Vertex AI RAG in
• vertex_project is required for Vertex AI RAG ingestion. Set
• vertexai.rag module not found. Vertex AI RAG requires google
• Source video has neither gcsUri nor bytesBase64Encoded. Cann

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/79799459058b835c. Report an issue: GitHub.